partisan fertility and presidential electionsgdahl/papers/partisan...us and provides detailed...
TRANSCRIPT
Partisan Fertility and Presidential Elections
Gordon Dahl Runjing Lu William Mullins
Abstract
This paper documents a new consequence of elections and a new determinant of fertility
A marked partisan change in economic optimism followed the unexpected election of Trump
in 2016 We find large downstream fertility effects with Republican-leaning counties experi-
encing a sharp increase in fertility relative to Democratic counties a 06 to 14 increase in
annual births depending on the intensity of partisanship Hispanics a group targeted by the
Trump campaign see a relative fertility decline of 16 percentage points compared to non-
Hispanics Bolstering the notion that political sentiment is driving these fertility changes we
find that following a Trump pre-election campaign visit relative Hispanic fertility declines
Keywords Fertility Partisanship
JEL Classification Numbers J13 D72
lowastGordon Dahl UC San Diego NBER Norwegian School of Economics CESifo CEPR IZA(gdahlucsdedu) Runjing Lu University of Alberta (runjing1ualbertaca) William Mullins UC SanDiego (wmullinsucsdedu)
April 6 2021
1 Introduction
Growing political polarization has magnified the salience of shifts in political power be-
tween Republicans and Democrats (Dimock andWike 2020) This partisanship is manifested
in economic optimism Voters become more optimistic about the direction of the nationrsquos
economy when they are politically aligned with the winning president and vice versa (see
Figure 1) These swings by partisan orientation are large immediate and persistent after the
2016 US Presidential election when President Trump was the unexpected winner Similarly
after the 2020 Presidential election Democratic and Republican optimism rapidly exchanged
positions These swings are of the same magnitude as the fall in economic optimism following
the onset of COVID-19 Do these sharp partisan changes in optimism result in meaningful
downstream effects Measures of optimism may not translate into changed behavior leav-
ing open the question of whether and how households incorporate partisanship into their
economic decisions
This paper examines effects on fertility of the surge in Republican and the collapse in
Democratic sentiment following the 2016 election Fertility is an irreversible long horizon
decision made by households with ensuing effects on labor force participation housing in-
vestment and consumption choices These effects may have distributional consequences with
groups targeted by political rhetoric displaying a stronger response to changes in political
leadership
The 2016 Presidential election is especially valuable for identifying the effect of exogenous
shifts in political power because the outcome was more unexpected than most elections1
We exploit the 2016 upset to cleanly identify a sharp and unexpected change in political
leadership This allows us to use event study designs comparing fertility across groups likely
to favor Republican or Democratic candidates
Using administrative data for the universe of births in the US our first approach com-
pares fertility across counties with low versus high democratic vote shares and before versus
after the November 2016 Presidential election We find Republican counties experience an
1Langer and Lemoine (2020) calculate a 12 probability of a Trump victory using options marketswhereas The New York Times and FiveThirtyEightrsquos polling-based forecasts were 15 and 29 respectively
1
immediate increase in births approximately nine months later Relative to Democrats there
are about 19000 excess births to Republican mothers in the year following the election
which translates to a 06 increase in annual births
Trumprsquos candidacy attracted a different set of voters relative to prior Republican coali-
tions (Confessore and Cohn 2016) To include these new Republican voters we use the
county-level change in the Republican vote shares between 2008 and 2016 Using this sec-
ond measure we estimate around 30000 excess births in the year after the election (09
of the annual average) in counties whose rightward shift was above the median relative to
those below
These two measures of partisanship capture different sources of variation the correlation
between them is only 016 We can combine both measures to define counties in the tails of
the partisan distribution To do so we take counties which had both a high ex ante level of
Republican support and shifted strongly towards Trump and compare them to those with
a high level of Democratic support and shifted less This contrast estimates excess births
of over 37000 (14 of the annual average) in the extreme Republican counties relative to
extreme Democratic counties
Our third approach compares Hispanics to other groups within the same counties His-
panics were specifically targeted by the Trump campaign and voted approximately two-to-
one for Hillary Clinton in 20162 Over the subsequent twelve months we estimate a relative
fertility decline of 16 percentage points for Hispanic mothers compared to non-Hispanics
When we contrast Hispanic births to those in predominantly rural counties to white moth-
ers we find even stronger effects We further find heterogeneous effects by the degree of
political polarization in a county (Autor et al 2020) the fertility gap between Hispanics
and non-Hispanics is twice as large in more polarized countiesTaken together the larger
effects we find as the likely intensity of partisanship increases mdash ie when we consider more
politically extreme counties compare Hispanics to rural non-Hispanic whites and contrast
2When launching his campaign Trump said ldquoWhen Mexico sends its people theyrsquore not sending theirbest Theyrsquore sending people that have lots of problems and theyrsquore bringing those problems with usTheyrsquore bringing drugs Theyrsquore bringing crime Theyrsquore rapists And some I assume are good peoplerdquoTheyrsquore sending us not the right people Itrsquos coming from more than Mexico Itrsquos coming from all overSouth and Latin America (Phillips 2017)
2
more to less-polarized counties mdash strongly point towards partisanship driving these effects
as opposed to alternative mechanisms3
Our estimated fertility effects which range from 06 to 14 across our three ap-
proaches are of comparable magnitude to the effects of unemployment and cash transfers
on fertility For example Dettling and Kearney (2014) and Schaller (2016) report that a 1
increase in the unemployment rate is associated with a decrease in birth rates between 14
and 22 Similarly Raute (2019) and Milligan (2005) find that a $1000 increase (in 2020
USD) in cash subsidies for a birth results in a 18 to 21 increase in fertility
To further bolster the idea that it is political sentiment which drives fertility changes
we examine the effects of Presidential campaign visits We examine how Hispanic versus
non-Hispanic fertility changes immediately following a visit by candidate Trump We use a
dynamic event study design which compares fertility over time in counties visited by Trump
using counties he will visit later as controls In the months following a campaign visit the
fertility rate falls for Hispanics relative to non-Hispanics by 12 of the mean
Our paper is related to a recent literature which documents the rising political polariza-
tion in the US (Gentzkow 2016 Autor et al 2020 Boxell et al 2020) broadening the
scope of activities affected by political sentiment Political polarization is associated with
fewer interparty marriages higher reporting of partisan affiliation on dating profiles and
fewer ideologically discordant Facebook friendships among other outcomes (Iyengar et al
2019)
The literature has also documented that survey-based economic optimism of partisans
changes around elections (Bartels 2002 Evans and Andersen 2006) but it has proven
challenging to causally link partisanship to real choices by economic agents A few papers
report a relationship between partisanship and spending on consumer goods (Gerber and
Huber 2009 Benhabib and Spiegel 2019 Gillitzer and Prasad 2018) while others have
challenged this link (McGrath 2017 Mian et al 2018) More recent papers have linked
partisanship with financial outcomes such as tax evasion stock market trading corporate
3All three of our national election-based results display parallel pre-trends and each measure of parti-sanship is robust to permutations In addition we provide supporting evidence using Google search data forthe term ldquopregnancy testrdquo in an event study using similar measures of party affiliation
3
credit ratings and retirement investing (Cullen et al 2020 Cookson et al 2020 Kempf
and Tsoutsoura 2020 Meeuwis et al 2018)4
Our contribution is to causally link sentiment to one of the most consequential household
decisions whether to have a child (Becker 1960) Unlike many consumption and investment
choices having a child is a long-term commitment requiring significant time and money
According to the USDA the estimated financial cost of raising a child from birth to age 17
is $233000 (Lino 2020)
Taken together this paper documents a new consequence of elections and a new de-
terminant of fertility Growing polarization makes understanding the downstream effects of
elections increasingly important The resulting shifts in fertility we document across political
groups have practical implications for regional public finance given stark partisan sorting
across residential geographies (Brown and Enos 2021) Moreover understanding the drivers
of fertility is increasingly important in light of below-replacement fertility and its structural
effects on economic growth (Jones 2020)
2 Evidence from the 2016 Presidential Election
We begin by examining how fertility responds to the unexpected election victory of
candidate Trump in 2016 We use two main strategies in a difference-in-differences (DID)
design comparing fertility in Republican-leaning versus Democratic counties and comparing
Hispanic fertility to that of non-Hispanics within the same county
21 Fertility Data
We use restricted-use US administrative natality data between 2012 and 2018 from Na-
tional Center for Health Statistics (NCHS) The data covers the universe of births in the
US and provides detailed information on each birth and the mother including the month
of birth (MOB) the month of the first day of the motherrsquos last menstrual period (MLMP)
4More broadly our paper relates to a literature on how economic factors influence fertility choicesincluding unemployment income housing prices coal busts fracking booms Medicaid eligibility COVID-19 cash transfers and child subsidies (Autor et al 2019 Cohen et al 2013 Schaller 2016 Dettling andKearney 2014 Kearney and Levine 2009 2020 Lindo 2010 Lovenheim and Mumford 2013 Black et al2013 Aizer et al 2020 McCrary and Royer 2011 Raute 2019 Duncan et al 2017)
4
and her age education raceethnicity and county of residence We restrict our attention to
singleton births to mothers who are US residents between the ages of 18 and 445
Our main outcome of interest is the number of births conceived in a month in a county
per 1000 females aged between 15 and 44 in the county6 Summary statistics for fertility
are in Table A1 the mean monthly fertility rate in a county is 45 births per 1000 females
To account for seasonality in fertility we de-seasonalize by subtracting its county-monthly
average using data starting from 2012 We refer to this variable as excess fertility
The natality data only records the month of the beginning of the motherrsquos last menstrual
period There is typically a seven-day lag between the first day of menses and the fertile
period which lasts approximately two weeks (NCHS 2005) Thus MLMP measures the
month of conception with noise Since the election occurred on November 8 2016 a mother
whose MLMP is in October could have conceived her child after the election Assuming a
uniform distribution of conception dates in a month about 30 of mothers whose MLMP
is in October are predicted to conceive after the election
22 Research Design
Our main research design is a DID event study using the 2016 Presidential election as
the event Our first approach compares fertility across Democratic and Republican counties
before versus after the election To measure county partisanship we obtain the county-level
vote share in Presidential elections from the MIT Election Data and Science Lab In our
first definition counties are categorized as Democratic if their Democrat vote share in 2012
is above the median and Republican otherwise
To include the new Republican voters who were drawn to Trump in 2016 our second
definition uses the county-level change in the Republican vote share between 2008 and 2016
5We use mothersrsquo reported MLMP as a proxy for conception date rather than nine months before MOBfollowing the literature (eg Dehejia and Lleras-Muney 2004) We remove misreported records by requiringthe difference between MOB and MLMP to be between five and 12 months
6We use as the denominator the number of fertile females aged between 15 and 44 - rather than 18 and44 - because US intercensal county population estimates are reported in five-year age bins The populationestimates are from Census Bureau We adjust excessive births due to the extra day in February of leap yearsby multiplying the fertility rate in that month by 2829 To ensure that the fertility rate is calculated basedon a reasonably-sized female pool we drop counties whose fertile female population is less than the 10thpercentile in 2012 (ie 769 women)
5
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
1 Introduction
Growing political polarization has magnified the salience of shifts in political power be-
tween Republicans and Democrats (Dimock andWike 2020) This partisanship is manifested
in economic optimism Voters become more optimistic about the direction of the nationrsquos
economy when they are politically aligned with the winning president and vice versa (see
Figure 1) These swings by partisan orientation are large immediate and persistent after the
2016 US Presidential election when President Trump was the unexpected winner Similarly
after the 2020 Presidential election Democratic and Republican optimism rapidly exchanged
positions These swings are of the same magnitude as the fall in economic optimism following
the onset of COVID-19 Do these sharp partisan changes in optimism result in meaningful
downstream effects Measures of optimism may not translate into changed behavior leav-
ing open the question of whether and how households incorporate partisanship into their
economic decisions
This paper examines effects on fertility of the surge in Republican and the collapse in
Democratic sentiment following the 2016 election Fertility is an irreversible long horizon
decision made by households with ensuing effects on labor force participation housing in-
vestment and consumption choices These effects may have distributional consequences with
groups targeted by political rhetoric displaying a stronger response to changes in political
leadership
The 2016 Presidential election is especially valuable for identifying the effect of exogenous
shifts in political power because the outcome was more unexpected than most elections1
We exploit the 2016 upset to cleanly identify a sharp and unexpected change in political
leadership This allows us to use event study designs comparing fertility across groups likely
to favor Republican or Democratic candidates
Using administrative data for the universe of births in the US our first approach com-
pares fertility across counties with low versus high democratic vote shares and before versus
after the November 2016 Presidential election We find Republican counties experience an
1Langer and Lemoine (2020) calculate a 12 probability of a Trump victory using options marketswhereas The New York Times and FiveThirtyEightrsquos polling-based forecasts were 15 and 29 respectively
1
immediate increase in births approximately nine months later Relative to Democrats there
are about 19000 excess births to Republican mothers in the year following the election
which translates to a 06 increase in annual births
Trumprsquos candidacy attracted a different set of voters relative to prior Republican coali-
tions (Confessore and Cohn 2016) To include these new Republican voters we use the
county-level change in the Republican vote shares between 2008 and 2016 Using this sec-
ond measure we estimate around 30000 excess births in the year after the election (09
of the annual average) in counties whose rightward shift was above the median relative to
those below
These two measures of partisanship capture different sources of variation the correlation
between them is only 016 We can combine both measures to define counties in the tails of
the partisan distribution To do so we take counties which had both a high ex ante level of
Republican support and shifted strongly towards Trump and compare them to those with
a high level of Democratic support and shifted less This contrast estimates excess births
of over 37000 (14 of the annual average) in the extreme Republican counties relative to
extreme Democratic counties
Our third approach compares Hispanics to other groups within the same counties His-
panics were specifically targeted by the Trump campaign and voted approximately two-to-
one for Hillary Clinton in 20162 Over the subsequent twelve months we estimate a relative
fertility decline of 16 percentage points for Hispanic mothers compared to non-Hispanics
When we contrast Hispanic births to those in predominantly rural counties to white moth-
ers we find even stronger effects We further find heterogeneous effects by the degree of
political polarization in a county (Autor et al 2020) the fertility gap between Hispanics
and non-Hispanics is twice as large in more polarized countiesTaken together the larger
effects we find as the likely intensity of partisanship increases mdash ie when we consider more
politically extreme counties compare Hispanics to rural non-Hispanic whites and contrast
2When launching his campaign Trump said ldquoWhen Mexico sends its people theyrsquore not sending theirbest Theyrsquore sending people that have lots of problems and theyrsquore bringing those problems with usTheyrsquore bringing drugs Theyrsquore bringing crime Theyrsquore rapists And some I assume are good peoplerdquoTheyrsquore sending us not the right people Itrsquos coming from more than Mexico Itrsquos coming from all overSouth and Latin America (Phillips 2017)
2
more to less-polarized counties mdash strongly point towards partisanship driving these effects
as opposed to alternative mechanisms3
Our estimated fertility effects which range from 06 to 14 across our three ap-
proaches are of comparable magnitude to the effects of unemployment and cash transfers
on fertility For example Dettling and Kearney (2014) and Schaller (2016) report that a 1
increase in the unemployment rate is associated with a decrease in birth rates between 14
and 22 Similarly Raute (2019) and Milligan (2005) find that a $1000 increase (in 2020
USD) in cash subsidies for a birth results in a 18 to 21 increase in fertility
To further bolster the idea that it is political sentiment which drives fertility changes
we examine the effects of Presidential campaign visits We examine how Hispanic versus
non-Hispanic fertility changes immediately following a visit by candidate Trump We use a
dynamic event study design which compares fertility over time in counties visited by Trump
using counties he will visit later as controls In the months following a campaign visit the
fertility rate falls for Hispanics relative to non-Hispanics by 12 of the mean
Our paper is related to a recent literature which documents the rising political polariza-
tion in the US (Gentzkow 2016 Autor et al 2020 Boxell et al 2020) broadening the
scope of activities affected by political sentiment Political polarization is associated with
fewer interparty marriages higher reporting of partisan affiliation on dating profiles and
fewer ideologically discordant Facebook friendships among other outcomes (Iyengar et al
2019)
The literature has also documented that survey-based economic optimism of partisans
changes around elections (Bartels 2002 Evans and Andersen 2006) but it has proven
challenging to causally link partisanship to real choices by economic agents A few papers
report a relationship between partisanship and spending on consumer goods (Gerber and
Huber 2009 Benhabib and Spiegel 2019 Gillitzer and Prasad 2018) while others have
challenged this link (McGrath 2017 Mian et al 2018) More recent papers have linked
partisanship with financial outcomes such as tax evasion stock market trading corporate
3All three of our national election-based results display parallel pre-trends and each measure of parti-sanship is robust to permutations In addition we provide supporting evidence using Google search data forthe term ldquopregnancy testrdquo in an event study using similar measures of party affiliation
3
credit ratings and retirement investing (Cullen et al 2020 Cookson et al 2020 Kempf
and Tsoutsoura 2020 Meeuwis et al 2018)4
Our contribution is to causally link sentiment to one of the most consequential household
decisions whether to have a child (Becker 1960) Unlike many consumption and investment
choices having a child is a long-term commitment requiring significant time and money
According to the USDA the estimated financial cost of raising a child from birth to age 17
is $233000 (Lino 2020)
Taken together this paper documents a new consequence of elections and a new de-
terminant of fertility Growing polarization makes understanding the downstream effects of
elections increasingly important The resulting shifts in fertility we document across political
groups have practical implications for regional public finance given stark partisan sorting
across residential geographies (Brown and Enos 2021) Moreover understanding the drivers
of fertility is increasingly important in light of below-replacement fertility and its structural
effects on economic growth (Jones 2020)
2 Evidence from the 2016 Presidential Election
We begin by examining how fertility responds to the unexpected election victory of
candidate Trump in 2016 We use two main strategies in a difference-in-differences (DID)
design comparing fertility in Republican-leaning versus Democratic counties and comparing
Hispanic fertility to that of non-Hispanics within the same county
21 Fertility Data
We use restricted-use US administrative natality data between 2012 and 2018 from Na-
tional Center for Health Statistics (NCHS) The data covers the universe of births in the
US and provides detailed information on each birth and the mother including the month
of birth (MOB) the month of the first day of the motherrsquos last menstrual period (MLMP)
4More broadly our paper relates to a literature on how economic factors influence fertility choicesincluding unemployment income housing prices coal busts fracking booms Medicaid eligibility COVID-19 cash transfers and child subsidies (Autor et al 2019 Cohen et al 2013 Schaller 2016 Dettling andKearney 2014 Kearney and Levine 2009 2020 Lindo 2010 Lovenheim and Mumford 2013 Black et al2013 Aizer et al 2020 McCrary and Royer 2011 Raute 2019 Duncan et al 2017)
4
and her age education raceethnicity and county of residence We restrict our attention to
singleton births to mothers who are US residents between the ages of 18 and 445
Our main outcome of interest is the number of births conceived in a month in a county
per 1000 females aged between 15 and 44 in the county6 Summary statistics for fertility
are in Table A1 the mean monthly fertility rate in a county is 45 births per 1000 females
To account for seasonality in fertility we de-seasonalize by subtracting its county-monthly
average using data starting from 2012 We refer to this variable as excess fertility
The natality data only records the month of the beginning of the motherrsquos last menstrual
period There is typically a seven-day lag between the first day of menses and the fertile
period which lasts approximately two weeks (NCHS 2005) Thus MLMP measures the
month of conception with noise Since the election occurred on November 8 2016 a mother
whose MLMP is in October could have conceived her child after the election Assuming a
uniform distribution of conception dates in a month about 30 of mothers whose MLMP
is in October are predicted to conceive after the election
22 Research Design
Our main research design is a DID event study using the 2016 Presidential election as
the event Our first approach compares fertility across Democratic and Republican counties
before versus after the election To measure county partisanship we obtain the county-level
vote share in Presidential elections from the MIT Election Data and Science Lab In our
first definition counties are categorized as Democratic if their Democrat vote share in 2012
is above the median and Republican otherwise
To include the new Republican voters who were drawn to Trump in 2016 our second
definition uses the county-level change in the Republican vote share between 2008 and 2016
5We use mothersrsquo reported MLMP as a proxy for conception date rather than nine months before MOBfollowing the literature (eg Dehejia and Lleras-Muney 2004) We remove misreported records by requiringthe difference between MOB and MLMP to be between five and 12 months
6We use as the denominator the number of fertile females aged between 15 and 44 - rather than 18 and44 - because US intercensal county population estimates are reported in five-year age bins The populationestimates are from Census Bureau We adjust excessive births due to the extra day in February of leap yearsby multiplying the fertility rate in that month by 2829 To ensure that the fertility rate is calculated basedon a reasonably-sized female pool we drop counties whose fertile female population is less than the 10thpercentile in 2012 (ie 769 women)
5
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
immediate increase in births approximately nine months later Relative to Democrats there
are about 19000 excess births to Republican mothers in the year following the election
which translates to a 06 increase in annual births
Trumprsquos candidacy attracted a different set of voters relative to prior Republican coali-
tions (Confessore and Cohn 2016) To include these new Republican voters we use the
county-level change in the Republican vote shares between 2008 and 2016 Using this sec-
ond measure we estimate around 30000 excess births in the year after the election (09
of the annual average) in counties whose rightward shift was above the median relative to
those below
These two measures of partisanship capture different sources of variation the correlation
between them is only 016 We can combine both measures to define counties in the tails of
the partisan distribution To do so we take counties which had both a high ex ante level of
Republican support and shifted strongly towards Trump and compare them to those with
a high level of Democratic support and shifted less This contrast estimates excess births
of over 37000 (14 of the annual average) in the extreme Republican counties relative to
extreme Democratic counties
Our third approach compares Hispanics to other groups within the same counties His-
panics were specifically targeted by the Trump campaign and voted approximately two-to-
one for Hillary Clinton in 20162 Over the subsequent twelve months we estimate a relative
fertility decline of 16 percentage points for Hispanic mothers compared to non-Hispanics
When we contrast Hispanic births to those in predominantly rural counties to white moth-
ers we find even stronger effects We further find heterogeneous effects by the degree of
political polarization in a county (Autor et al 2020) the fertility gap between Hispanics
and non-Hispanics is twice as large in more polarized countiesTaken together the larger
effects we find as the likely intensity of partisanship increases mdash ie when we consider more
politically extreme counties compare Hispanics to rural non-Hispanic whites and contrast
2When launching his campaign Trump said ldquoWhen Mexico sends its people theyrsquore not sending theirbest Theyrsquore sending people that have lots of problems and theyrsquore bringing those problems with usTheyrsquore bringing drugs Theyrsquore bringing crime Theyrsquore rapists And some I assume are good peoplerdquoTheyrsquore sending us not the right people Itrsquos coming from more than Mexico Itrsquos coming from all overSouth and Latin America (Phillips 2017)
2
more to less-polarized counties mdash strongly point towards partisanship driving these effects
as opposed to alternative mechanisms3
Our estimated fertility effects which range from 06 to 14 across our three ap-
proaches are of comparable magnitude to the effects of unemployment and cash transfers
on fertility For example Dettling and Kearney (2014) and Schaller (2016) report that a 1
increase in the unemployment rate is associated with a decrease in birth rates between 14
and 22 Similarly Raute (2019) and Milligan (2005) find that a $1000 increase (in 2020
USD) in cash subsidies for a birth results in a 18 to 21 increase in fertility
To further bolster the idea that it is political sentiment which drives fertility changes
we examine the effects of Presidential campaign visits We examine how Hispanic versus
non-Hispanic fertility changes immediately following a visit by candidate Trump We use a
dynamic event study design which compares fertility over time in counties visited by Trump
using counties he will visit later as controls In the months following a campaign visit the
fertility rate falls for Hispanics relative to non-Hispanics by 12 of the mean
Our paper is related to a recent literature which documents the rising political polariza-
tion in the US (Gentzkow 2016 Autor et al 2020 Boxell et al 2020) broadening the
scope of activities affected by political sentiment Political polarization is associated with
fewer interparty marriages higher reporting of partisan affiliation on dating profiles and
fewer ideologically discordant Facebook friendships among other outcomes (Iyengar et al
2019)
The literature has also documented that survey-based economic optimism of partisans
changes around elections (Bartels 2002 Evans and Andersen 2006) but it has proven
challenging to causally link partisanship to real choices by economic agents A few papers
report a relationship between partisanship and spending on consumer goods (Gerber and
Huber 2009 Benhabib and Spiegel 2019 Gillitzer and Prasad 2018) while others have
challenged this link (McGrath 2017 Mian et al 2018) More recent papers have linked
partisanship with financial outcomes such as tax evasion stock market trading corporate
3All three of our national election-based results display parallel pre-trends and each measure of parti-sanship is robust to permutations In addition we provide supporting evidence using Google search data forthe term ldquopregnancy testrdquo in an event study using similar measures of party affiliation
3
credit ratings and retirement investing (Cullen et al 2020 Cookson et al 2020 Kempf
and Tsoutsoura 2020 Meeuwis et al 2018)4
Our contribution is to causally link sentiment to one of the most consequential household
decisions whether to have a child (Becker 1960) Unlike many consumption and investment
choices having a child is a long-term commitment requiring significant time and money
According to the USDA the estimated financial cost of raising a child from birth to age 17
is $233000 (Lino 2020)
Taken together this paper documents a new consequence of elections and a new de-
terminant of fertility Growing polarization makes understanding the downstream effects of
elections increasingly important The resulting shifts in fertility we document across political
groups have practical implications for regional public finance given stark partisan sorting
across residential geographies (Brown and Enos 2021) Moreover understanding the drivers
of fertility is increasingly important in light of below-replacement fertility and its structural
effects on economic growth (Jones 2020)
2 Evidence from the 2016 Presidential Election
We begin by examining how fertility responds to the unexpected election victory of
candidate Trump in 2016 We use two main strategies in a difference-in-differences (DID)
design comparing fertility in Republican-leaning versus Democratic counties and comparing
Hispanic fertility to that of non-Hispanics within the same county
21 Fertility Data
We use restricted-use US administrative natality data between 2012 and 2018 from Na-
tional Center for Health Statistics (NCHS) The data covers the universe of births in the
US and provides detailed information on each birth and the mother including the month
of birth (MOB) the month of the first day of the motherrsquos last menstrual period (MLMP)
4More broadly our paper relates to a literature on how economic factors influence fertility choicesincluding unemployment income housing prices coal busts fracking booms Medicaid eligibility COVID-19 cash transfers and child subsidies (Autor et al 2019 Cohen et al 2013 Schaller 2016 Dettling andKearney 2014 Kearney and Levine 2009 2020 Lindo 2010 Lovenheim and Mumford 2013 Black et al2013 Aizer et al 2020 McCrary and Royer 2011 Raute 2019 Duncan et al 2017)
4
and her age education raceethnicity and county of residence We restrict our attention to
singleton births to mothers who are US residents between the ages of 18 and 445
Our main outcome of interest is the number of births conceived in a month in a county
per 1000 females aged between 15 and 44 in the county6 Summary statistics for fertility
are in Table A1 the mean monthly fertility rate in a county is 45 births per 1000 females
To account for seasonality in fertility we de-seasonalize by subtracting its county-monthly
average using data starting from 2012 We refer to this variable as excess fertility
The natality data only records the month of the beginning of the motherrsquos last menstrual
period There is typically a seven-day lag between the first day of menses and the fertile
period which lasts approximately two weeks (NCHS 2005) Thus MLMP measures the
month of conception with noise Since the election occurred on November 8 2016 a mother
whose MLMP is in October could have conceived her child after the election Assuming a
uniform distribution of conception dates in a month about 30 of mothers whose MLMP
is in October are predicted to conceive after the election
22 Research Design
Our main research design is a DID event study using the 2016 Presidential election as
the event Our first approach compares fertility across Democratic and Republican counties
before versus after the election To measure county partisanship we obtain the county-level
vote share in Presidential elections from the MIT Election Data and Science Lab In our
first definition counties are categorized as Democratic if their Democrat vote share in 2012
is above the median and Republican otherwise
To include the new Republican voters who were drawn to Trump in 2016 our second
definition uses the county-level change in the Republican vote share between 2008 and 2016
5We use mothersrsquo reported MLMP as a proxy for conception date rather than nine months before MOBfollowing the literature (eg Dehejia and Lleras-Muney 2004) We remove misreported records by requiringthe difference between MOB and MLMP to be between five and 12 months
6We use as the denominator the number of fertile females aged between 15 and 44 - rather than 18 and44 - because US intercensal county population estimates are reported in five-year age bins The populationestimates are from Census Bureau We adjust excessive births due to the extra day in February of leap yearsby multiplying the fertility rate in that month by 2829 To ensure that the fertility rate is calculated basedon a reasonably-sized female pool we drop counties whose fertile female population is less than the 10thpercentile in 2012 (ie 769 women)
5
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
more to less-polarized counties mdash strongly point towards partisanship driving these effects
as opposed to alternative mechanisms3
Our estimated fertility effects which range from 06 to 14 across our three ap-
proaches are of comparable magnitude to the effects of unemployment and cash transfers
on fertility For example Dettling and Kearney (2014) and Schaller (2016) report that a 1
increase in the unemployment rate is associated with a decrease in birth rates between 14
and 22 Similarly Raute (2019) and Milligan (2005) find that a $1000 increase (in 2020
USD) in cash subsidies for a birth results in a 18 to 21 increase in fertility
To further bolster the idea that it is political sentiment which drives fertility changes
we examine the effects of Presidential campaign visits We examine how Hispanic versus
non-Hispanic fertility changes immediately following a visit by candidate Trump We use a
dynamic event study design which compares fertility over time in counties visited by Trump
using counties he will visit later as controls In the months following a campaign visit the
fertility rate falls for Hispanics relative to non-Hispanics by 12 of the mean
Our paper is related to a recent literature which documents the rising political polariza-
tion in the US (Gentzkow 2016 Autor et al 2020 Boxell et al 2020) broadening the
scope of activities affected by political sentiment Political polarization is associated with
fewer interparty marriages higher reporting of partisan affiliation on dating profiles and
fewer ideologically discordant Facebook friendships among other outcomes (Iyengar et al
2019)
The literature has also documented that survey-based economic optimism of partisans
changes around elections (Bartels 2002 Evans and Andersen 2006) but it has proven
challenging to causally link partisanship to real choices by economic agents A few papers
report a relationship between partisanship and spending on consumer goods (Gerber and
Huber 2009 Benhabib and Spiegel 2019 Gillitzer and Prasad 2018) while others have
challenged this link (McGrath 2017 Mian et al 2018) More recent papers have linked
partisanship with financial outcomes such as tax evasion stock market trading corporate
3All three of our national election-based results display parallel pre-trends and each measure of parti-sanship is robust to permutations In addition we provide supporting evidence using Google search data forthe term ldquopregnancy testrdquo in an event study using similar measures of party affiliation
3
credit ratings and retirement investing (Cullen et al 2020 Cookson et al 2020 Kempf
and Tsoutsoura 2020 Meeuwis et al 2018)4
Our contribution is to causally link sentiment to one of the most consequential household
decisions whether to have a child (Becker 1960) Unlike many consumption and investment
choices having a child is a long-term commitment requiring significant time and money
According to the USDA the estimated financial cost of raising a child from birth to age 17
is $233000 (Lino 2020)
Taken together this paper documents a new consequence of elections and a new de-
terminant of fertility Growing polarization makes understanding the downstream effects of
elections increasingly important The resulting shifts in fertility we document across political
groups have practical implications for regional public finance given stark partisan sorting
across residential geographies (Brown and Enos 2021) Moreover understanding the drivers
of fertility is increasingly important in light of below-replacement fertility and its structural
effects on economic growth (Jones 2020)
2 Evidence from the 2016 Presidential Election
We begin by examining how fertility responds to the unexpected election victory of
candidate Trump in 2016 We use two main strategies in a difference-in-differences (DID)
design comparing fertility in Republican-leaning versus Democratic counties and comparing
Hispanic fertility to that of non-Hispanics within the same county
21 Fertility Data
We use restricted-use US administrative natality data between 2012 and 2018 from Na-
tional Center for Health Statistics (NCHS) The data covers the universe of births in the
US and provides detailed information on each birth and the mother including the month
of birth (MOB) the month of the first day of the motherrsquos last menstrual period (MLMP)
4More broadly our paper relates to a literature on how economic factors influence fertility choicesincluding unemployment income housing prices coal busts fracking booms Medicaid eligibility COVID-19 cash transfers and child subsidies (Autor et al 2019 Cohen et al 2013 Schaller 2016 Dettling andKearney 2014 Kearney and Levine 2009 2020 Lindo 2010 Lovenheim and Mumford 2013 Black et al2013 Aizer et al 2020 McCrary and Royer 2011 Raute 2019 Duncan et al 2017)
4
and her age education raceethnicity and county of residence We restrict our attention to
singleton births to mothers who are US residents between the ages of 18 and 445
Our main outcome of interest is the number of births conceived in a month in a county
per 1000 females aged between 15 and 44 in the county6 Summary statistics for fertility
are in Table A1 the mean monthly fertility rate in a county is 45 births per 1000 females
To account for seasonality in fertility we de-seasonalize by subtracting its county-monthly
average using data starting from 2012 We refer to this variable as excess fertility
The natality data only records the month of the beginning of the motherrsquos last menstrual
period There is typically a seven-day lag between the first day of menses and the fertile
period which lasts approximately two weeks (NCHS 2005) Thus MLMP measures the
month of conception with noise Since the election occurred on November 8 2016 a mother
whose MLMP is in October could have conceived her child after the election Assuming a
uniform distribution of conception dates in a month about 30 of mothers whose MLMP
is in October are predicted to conceive after the election
22 Research Design
Our main research design is a DID event study using the 2016 Presidential election as
the event Our first approach compares fertility across Democratic and Republican counties
before versus after the election To measure county partisanship we obtain the county-level
vote share in Presidential elections from the MIT Election Data and Science Lab In our
first definition counties are categorized as Democratic if their Democrat vote share in 2012
is above the median and Republican otherwise
To include the new Republican voters who were drawn to Trump in 2016 our second
definition uses the county-level change in the Republican vote share between 2008 and 2016
5We use mothersrsquo reported MLMP as a proxy for conception date rather than nine months before MOBfollowing the literature (eg Dehejia and Lleras-Muney 2004) We remove misreported records by requiringthe difference between MOB and MLMP to be between five and 12 months
6We use as the denominator the number of fertile females aged between 15 and 44 - rather than 18 and44 - because US intercensal county population estimates are reported in five-year age bins The populationestimates are from Census Bureau We adjust excessive births due to the extra day in February of leap yearsby multiplying the fertility rate in that month by 2829 To ensure that the fertility rate is calculated basedon a reasonably-sized female pool we drop counties whose fertile female population is less than the 10thpercentile in 2012 (ie 769 women)
5
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
credit ratings and retirement investing (Cullen et al 2020 Cookson et al 2020 Kempf
and Tsoutsoura 2020 Meeuwis et al 2018)4
Our contribution is to causally link sentiment to one of the most consequential household
decisions whether to have a child (Becker 1960) Unlike many consumption and investment
choices having a child is a long-term commitment requiring significant time and money
According to the USDA the estimated financial cost of raising a child from birth to age 17
is $233000 (Lino 2020)
Taken together this paper documents a new consequence of elections and a new de-
terminant of fertility Growing polarization makes understanding the downstream effects of
elections increasingly important The resulting shifts in fertility we document across political
groups have practical implications for regional public finance given stark partisan sorting
across residential geographies (Brown and Enos 2021) Moreover understanding the drivers
of fertility is increasingly important in light of below-replacement fertility and its structural
effects on economic growth (Jones 2020)
2 Evidence from the 2016 Presidential Election
We begin by examining how fertility responds to the unexpected election victory of
candidate Trump in 2016 We use two main strategies in a difference-in-differences (DID)
design comparing fertility in Republican-leaning versus Democratic counties and comparing
Hispanic fertility to that of non-Hispanics within the same county
21 Fertility Data
We use restricted-use US administrative natality data between 2012 and 2018 from Na-
tional Center for Health Statistics (NCHS) The data covers the universe of births in the
US and provides detailed information on each birth and the mother including the month
of birth (MOB) the month of the first day of the motherrsquos last menstrual period (MLMP)
4More broadly our paper relates to a literature on how economic factors influence fertility choicesincluding unemployment income housing prices coal busts fracking booms Medicaid eligibility COVID-19 cash transfers and child subsidies (Autor et al 2019 Cohen et al 2013 Schaller 2016 Dettling andKearney 2014 Kearney and Levine 2009 2020 Lindo 2010 Lovenheim and Mumford 2013 Black et al2013 Aizer et al 2020 McCrary and Royer 2011 Raute 2019 Duncan et al 2017)
4
and her age education raceethnicity and county of residence We restrict our attention to
singleton births to mothers who are US residents between the ages of 18 and 445
Our main outcome of interest is the number of births conceived in a month in a county
per 1000 females aged between 15 and 44 in the county6 Summary statistics for fertility
are in Table A1 the mean monthly fertility rate in a county is 45 births per 1000 females
To account for seasonality in fertility we de-seasonalize by subtracting its county-monthly
average using data starting from 2012 We refer to this variable as excess fertility
The natality data only records the month of the beginning of the motherrsquos last menstrual
period There is typically a seven-day lag between the first day of menses and the fertile
period which lasts approximately two weeks (NCHS 2005) Thus MLMP measures the
month of conception with noise Since the election occurred on November 8 2016 a mother
whose MLMP is in October could have conceived her child after the election Assuming a
uniform distribution of conception dates in a month about 30 of mothers whose MLMP
is in October are predicted to conceive after the election
22 Research Design
Our main research design is a DID event study using the 2016 Presidential election as
the event Our first approach compares fertility across Democratic and Republican counties
before versus after the election To measure county partisanship we obtain the county-level
vote share in Presidential elections from the MIT Election Data and Science Lab In our
first definition counties are categorized as Democratic if their Democrat vote share in 2012
is above the median and Republican otherwise
To include the new Republican voters who were drawn to Trump in 2016 our second
definition uses the county-level change in the Republican vote share between 2008 and 2016
5We use mothersrsquo reported MLMP as a proxy for conception date rather than nine months before MOBfollowing the literature (eg Dehejia and Lleras-Muney 2004) We remove misreported records by requiringthe difference between MOB and MLMP to be between five and 12 months
6We use as the denominator the number of fertile females aged between 15 and 44 - rather than 18 and44 - because US intercensal county population estimates are reported in five-year age bins The populationestimates are from Census Bureau We adjust excessive births due to the extra day in February of leap yearsby multiplying the fertility rate in that month by 2829 To ensure that the fertility rate is calculated basedon a reasonably-sized female pool we drop counties whose fertile female population is less than the 10thpercentile in 2012 (ie 769 women)
5
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
and her age education raceethnicity and county of residence We restrict our attention to
singleton births to mothers who are US residents between the ages of 18 and 445
Our main outcome of interest is the number of births conceived in a month in a county
per 1000 females aged between 15 and 44 in the county6 Summary statistics for fertility
are in Table A1 the mean monthly fertility rate in a county is 45 births per 1000 females
To account for seasonality in fertility we de-seasonalize by subtracting its county-monthly
average using data starting from 2012 We refer to this variable as excess fertility
The natality data only records the month of the beginning of the motherrsquos last menstrual
period There is typically a seven-day lag between the first day of menses and the fertile
period which lasts approximately two weeks (NCHS 2005) Thus MLMP measures the
month of conception with noise Since the election occurred on November 8 2016 a mother
whose MLMP is in October could have conceived her child after the election Assuming a
uniform distribution of conception dates in a month about 30 of mothers whose MLMP
is in October are predicted to conceive after the election
22 Research Design
Our main research design is a DID event study using the 2016 Presidential election as
the event Our first approach compares fertility across Democratic and Republican counties
before versus after the election To measure county partisanship we obtain the county-level
vote share in Presidential elections from the MIT Election Data and Science Lab In our
first definition counties are categorized as Democratic if their Democrat vote share in 2012
is above the median and Republican otherwise
To include the new Republican voters who were drawn to Trump in 2016 our second
definition uses the county-level change in the Republican vote share between 2008 and 2016
5We use mothersrsquo reported MLMP as a proxy for conception date rather than nine months before MOBfollowing the literature (eg Dehejia and Lleras-Muney 2004) We remove misreported records by requiringthe difference between MOB and MLMP to be between five and 12 months
6We use as the denominator the number of fertile females aged between 15 and 44 - rather than 18 and44 - because US intercensal county population estimates are reported in five-year age bins The populationestimates are from Census Bureau We adjust excessive births due to the extra day in February of leap yearsby multiplying the fertility rate in that month by 2829 To ensure that the fertility rate is calculated basedon a reasonably-sized female pool we drop counties whose fertile female population is less than the 10thpercentile in 2012 (ie 769 women)
5
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
and classifies counties with an above-median change (ie a shift more than 58 percentage
points) as Republican and Democratic otherwise
Since the two measures of partisanship capture different sources of variation (correlation
= 016) we can combine both measures to define counties in the tails of the partisan distri-
bution To do so we take counties which had both a below-median Democrat vote share in
2012 and an above-median Republican shift between 2008 and 2016 and compare them to
those with a higher Democrat vote share and shifted less
Our first regression model is
Yct =3sum
t=minus3
βt timesDemocratc + αc + αt + εct (1)
where Yct is the excess fertility rate in county c and t is the number of time periods relative
to the election in November 2016 We omit the indicator for t = minus1 and use t = minus1 as our
comparison period Our treatment variable is Democratc which equals one if county c is
classified as Democratic and zero otherwise We include county fixed effects αc and event
time fixed effects αt to control for the average fertility rate in a county and national fertility
trends We cluster standard errors by county
While the data is monthly for precision and ease of presentation we often collapse the
data by quarter In the appendix we additionally report estimated effects at the monthly
level For the quarterly estimates we define t = 0 for the months of October November and
December of 2016 For the reasons described in section 21 October is a partially-treated
month which is why we include it with November and December
Our second DID approach compares fertility by race and ethnic group within the same
county Our comparison focuses on Hispanic versus non-Hispanic fertility due to Trumprsquos
adversarial rhetoric towards this group Our second regression model is
Ykct =3sum
t=minus3
βt timesHispanick +Hispanick + αc + αt + εkct (2)
where Ykct is the excess fertility rate for females with ethnic group k in county c in time
t Hispanick is one if the ethnic group is Hispanic and zero otherwise The fixed effects are
6
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
the same as in equation 1
If the result of the 2016 presidential election was unanticipated and fertility trends across
counties and across different ethnic groups are parallel in the absence of the election the βt
vector identifies the impact of presidential election on fertility decisions before and after the
election As we will show both of these conditions appear to hold
23 Optimism and Elections
It is well established that partisans of the winning side in a Presidential election become
more optimistic about the direction of the economy Figure 1 plots the percentage of positive
minus negative responses (which we label net better) to the question ldquoDo you think the
nationrsquos economy is getting better or worserdquo among registered voters Republican voters
became immediately more optimistic following the November 2016 election with the net
better fraction rising from -63 to +63 over the course of four months In contrast the
net better fraction for Democrats falls from +52 to -4 over the same period after which
it continues to erode Similar swings in optimism but in the other direction occurred after
close election which candidate Biden won in 2020 To benchmark the large magnitude of
these drops the COVID-19 stock market crash caused the Republican net better percentage
to fall from +83 to -19 and the Democratic percentage from -42 to -87
An alternative measure of economic optimism is the Bloomberg Consumer Comfort Index
reported in Figure A1 This uses an entirely different sampling scheme but reveals very
similar patterns
24 Fertility Effects across Political Geographies
In Figure 2 we show the effects of 2016 election on fertility in Democratic and Republican
counties In panel A1 we start by comparing the raw trend of monthly excess fertility of
counties with above- or below-median Democrat vote share in the prior presidential election
The blue line captures the excess fertility in Democratic counties and the red in Republican
counties Both lines are normalized to be 0 in September 2016 The vertical shaded area
spanning the months of October and November indicates the period immediately surrounding
the election As described in section 22 October represents a partially-treated month due
to how conceptions are measured in our data
7
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
The first thing to note is that the blue and red lines lie on top of each other in the pre-
period These parallel trends diverge after the election with the red line rising rapidly in
November and December There remains a gap between the red and the blue lines through
the remainder of the sample period
Panel A2 plots βt coefficients corresponding to equation 1 at the quarterly level repre-
senting the effect of the Presidential election on fertility in Democratic versus Republican
counties Confirming the pattern in the raw data there are no pre-trends but large effects in
all four quarters after the election which are statistically different from zero Corresponding
monthly estimates are found in Figure A2
In 2016 Trump attracted a different voter coalition relative to prior Republican candi-
dates We use the county-level change in the Republican vote share between 2008 and 2016
to capture these new voters Panel B1 plots the monthly excess fertility for counties with
an above-median shift towards the Republicans (red line) and counties with a below-median
shift (blue line) Panel B2 plots the quarterly regression coefficients Again there is no evi-
dence of differential pre-trends either in the raw data or in the regression setting Further
the gaps in excess fertility rate are as large or larger compared to those in column A
The correlation between the aforementioned measures of a countyrsquos partisanship is only
016 reflecting different sources of variation We combine both measures to capture counties
in the extreme ends of the partisan distribution in column C of Figure 2 We define extreme
Republican counties (red line) as those with a below-median level of Democrat vote share
and an above-median vote shift towards Trump We similarly define extreme Democratic
counties (blue line)
Comparing extreme counties yields even larger effects the gaps in column C approxi-
mately doubles compared to either column A or B In panel C1 while the blue line shifts
downwards the jump in the red line is more dramatic These patterns suggest that it is Re-
publican fertility in the most extreme counties that drives majority of the effect assuming
the excess birth rates would have stayed at zero for both groups otherwise The stronger
fertility effects we find for more politically extreme counties adds credence to the idea that
partisanship drives our results
Note that in May 2016 there is a sizeable jump in the red line in panel C1 One potential
8
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
explanation consistent with this being attributable to partisan sentiment is that this was the
month that Trump became the presumptive nominee of the Republican party
Regression results corresponding to these figures are found in Table 1 columns (1)
through (3) For example consider the Treat0 coefficient in column (1) this represents
a drop of 0144 excess births per one thousand women in Democratic versus Republican
counties in the quarter of the election (quarter 0)
The sum of the treatment effects in quarters 0 through 3 in column (1) indicates on aver-
age there were 0597 fewer excess births per 1000 women in Democratic versus Republican
counties in the year following the election This translates to a fertility gap of 19000 births
between Republican and Democratic counties amounting to 06 of total annual births
Similarly column (2) indicates 0964 fewer excess births per 1000 women in counties which
experienced a below versus above median shift in Republican vote share This gap amounts
to a difference of 30000 annual births between Democratic and Republican counties or 09
of the annual average Finally column (3) reports that the extreme Democratic counties
experienced 1501 lower excess fertility race than extreme Republican ones equivalent to
37000 fewer births (14 of the annual average) The translated effects for extreme counties
are not directly comparable to those in columns (1) and (2) because there are fewer counties
in this estimation sample It is therefore all the more surprising that the implied birth gap
is larger supporting the idea that fertility effects are driven by partisan sentiment
25 Fertility Effects by Ethnicity
As a different measure of political partisanship we compare Hispanics to non-Hispanics
This split is motivated by Trumprsquos harsh rhetoric towards the Hispanic population begin-
ning with his first campaign speech where he compared Hispanic immigrants to rapists and
murderers Moreover Hispanics have historically backed Democratic candidates by a wide
margin voting two-to-one for Hillary Clinton in 2016
Figure 3 plots excess fertility for Hispanics and non-Hispanics surrounding the 2016
election Panel A1 shows similar pre-trends for Hispanic and non-Hispanic women Following
the election non-Hispanic fertility rises for two months while that of Hispanics falls markedly
over time the gap between the two persists in every month in the post-election period Panel
9
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
A2 plots the quarterly regression estimates for equation 2 There are no pre-trends but large
and consistent negative fertility effects
Given the strong support for Trump among whites in rural areas in panel B we replace
the control group of non-Hispanics with non-Hispanics whites in predominantly rural counties
as defined by the Census Panel B1 shows an immediate and dramatic rise in rural non-
Hispanic white fertility As Panel B2 shows there are no differential pre-trends and the
effect is almost twice as large as found in Panel A2 We also plot monthly event study
estimates in Figure A3
Corresponding quarterly estimates are reported in Table 1 columns (4) and (5) Over
the ensuing year we estimate a relative fertility decline of 16 percentage points for Hispanic
mothers compared to non-Hispanics
As a final exercise we explore whether more politically polarized counties engender larger
effects We take advantage of the arguably exogenous shock to local economic conditions
caused by the China trade shock Autor et al (2020) show that trade-exposed counties
become more polarized both on the left and right of the political spectrum we use their
proxy for county-level polarization
We create two interaction terms multiplying βt in equation 2 with whether a county is
above or below the median of the instrumented China trade shock of Autor et al (2020)
Results are plotted in Figure A4 and reported in Table A3 The post-election fertility gap
between Hispanics and non-Hispanics is twice as large in more polarized counties
Taken in combination the larger effects we find as the likely intensity of partisanship
increases mdash ie when we consider more politically extreme counties compare Hispanics to
rural non-Hispanic whites and contrast more to less-polarized counties mdash strongly point
towards partisanship driving these effects as opposed to alternative mechanisms
26 Robustness Checks
Table A2 provides the results for a variety of robustness tests Columns (1) to (5) repeat
our main regressions in Table 1 with the addition of controls for county income and income
squared to absorb confounding effects from changes in economic conditions Our estimates
are unaffected Column (6) changes the definition of county partisanship used in Table 1
10
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
column (1) and in Figure 2 column A Instead of using the 50th percentile of Democrat vote
share in a county as the partisan classification cutoff here we define Democratic counties as
those with Democrat vote shares above the 60th percentile and Republican counties as those
below the 40th percentile We drop the counties with vote shares in between these cutoffs
Using this classification the cumulative effect in the year after the election is slightly larger
than in our original estimate at 06 of annual births
Column (7) of Table A2 changes the measure we use in Table 1 column (2) and in Figure 2
column B Our main estimate uses countiesrsquo shift towards Republicans between 2008 and
2016 to define treatment and control groups here we use the shift between 2012 and 2016
The results are likewise unaffected
Columns (8) and (9) change the treatment group that we use in Table 1 column (4)
and in Figure 3 Column (8) replaces the treatment group (Hispanics) with Mexicans as
Trump often referred to them specifically in his anti-Hispanic rhetoric and they make up the
largest Hispanic group in the US The cumulative fertility effects measured in annual births
are 50 larger for Mexicans than for Hispanics To test whether the relative fertility losses
we identify are specific to Hispanics or whether they occurred among all minority groups
we drop Hispanics from column (9) We use non-Hispanic minorities as the treatment group
and non-Hispanic whites as controls While there is a relative fertility drop in quarter
one after the election the effect over the whole year is much smaller and not statistically
distinguishable from zero This suggests the fertility losses are concentrated among Hispanics
and not in other minority groups
As a final exercise we perform our main analysis with Google search index for ldquopregnancy
testrdquo (scaled by the highest local weekly search rate for the term) as the dependent variable
Consistent with our main results Table A4 shows that Democratic-leaning designated mar-
keting areas (DMAs) and those with a high proportion of Hispanics both see relative falls in
this search following the 2016 election
3 Evidence from Pre-Election Campaign Visits
To bolster the notion that it is political sentiment that drives fertility changes we examine
the effects of Trumprsquos pre-election campaign visits Local residents appear to pay attention
11
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
to these events Figure A5 shows that Google search index for Trump starts to rise about
two weeks before Trumprsquos first campaign visit to a DMA peaking in the week of the visit
We hypothesize that when Trump visits an area for the first time on the election trail it
excites his base and potentially reduces the optimism of his opponents which in turn affects
relative fertility
31 Data
We collect Trumprsquos campaign visits between January 2015 and November 2016 from the
National Journalrsquos Travel Tracker7 The data contains date time and city information for
each visit We map city names to county Trump visited a total of 230 counties during his
campaign the mean and variance of these countiesrsquo fertility are reported in Table A1
While we have the actual date of Trumprsquos campaign visit the natality data only records
the month of the first day of motherrsquos last menstrual period (MLMP) This means that
women whose menses occur in the second half of the month preceding a visit may be fertile
when a visit occurs Likewise women whose menses begin in the month of a visit may not
be fertile until the following month This means that women whose MLMP is in month -1
can also be treated by Trumprsquos campaign visit8 Similarly women whose MLMP is in month
0 may not have been treated
32 Research Design
We use a triple DID dynamic event study comparing fertility between Hispanic and non-
Hispanic fertile females in counties visited by Trump before and after the visit using counties
he will visit later as controls We focus on Trumprsquos first campaign visit to a county so as
not to contaminate our estimate with prior visits
We use counties that Trump will visit in the future rather than counties he will never
visit as controls because un-visited counties are considerably different We use a dynamic
7Travel Tracker compiles information from candidatesrsquo public campaign schedules and excludes eventsthat candidates hold in their home states
8The predicted percentage of fertile days in the month of a campaign visit (month 0) for women whoseMLMP occurs in month -1 is approximately two thirds that of women whose MLMP occurs in month 0 Thiscalculation assumes a uniform distribution of conception dates and visits throughout a month MoreoverGoogle search data in Figure A5 suggests some anticipation of the campaign visit Including anticipationwould increase the percentage of fertile days for women whose MLMP occurs in month -1 to over 90 thatof women whose MLMP occurs in month 0
12
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
event study because there may have been heterogeneous effects across counties visited at
different times as Trumprsquos campaign strategy and rhetoric evolved over time (Abraham and
Sun 2018) To implement the dynamic event study we stack our panel data as a series of
4times2 matrices (Hispanicnon-Hispanic in treatmentcontrol counties times prepost) and adapt
the R package from Novgorodsky and Setzler (2019)
The outcome is the fertility rate among women by Hispanic ethnicity in a county and
month We define the omitted period as month -3 because women whose MLMP occurs in
month -1 may also be treated and to allow for potential anticipation one month before the
actual visit We define counties visited in month g as cohort g and cohort-specific event
time in calendar month m as eg = mminus g
Under the identifying assumption that earlier visited and later visited counties share sim-
ilar fertility trends in the absence of Trumprsquos campaign visits we can identify the treatment
effect on the fertility of Hispanics versus others in treated cohort g in event time eg which
we label as βHeg Following Abraham and Sun (2018) and Novgorodsky and Setzler (2019)
we define the average treatment effect for event time e as
βHe =
sumgisinG
βHeg lowast wg (3)
where wg is the Hispanic fertile female population in counties belonging to cohort g We
calculate clustered standard errors for βHeg via the delta method
All counties in the regression sample experience their first Trump visit by November
2016 As a result restricting controls to be eventually-visited counties forces us to trade off
the number of post periods with the number of cohorts we can estimate treatment effects
for We estimate effects for event times from -7 to +5 months This implies that the last
cohort we can estimate effects for have their first Trump visits by April 20169
9Since control counties must be visited by November 2016 and there is one month of anticipationtreatment effects can only be estimated up to September 2016 Six (5+1) months before September 2016 isApril 2016
13
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
33 Fertility Response to Trump Campaign Visits
Figure 4 plots the average treatment effect from seven months before a visit to five
months after The Hispanic fertility rate relative to non-Hispanic in the same county starts
to decrease in month -1 which is when partial treatment of mothers begins This lower
fertility rate continues through month 2 However the positive treatment effect in month
5 indicates that the depressed fertility in months -1 to 2 may constitute harvesting In
the months following a campaign visit the fertility rate for Hispanics falls relative to non-
Hispanics by 12 of the mean Corresponding regression estimates are found in Table A5
column (1) As a robustness check we use the total fertile female population in counties
belonging to cohort g as wg in column (2) results are quantitatively similar
4 Conclusion
This paper causally links political partisanship to one of the most consequential household
decisions whether to have a child Unlike many consumption and investment choices this
is a long-term commitment requiring significant time and money We compare Republican-
leaning to Democratic counties before and after the 2016 Presidential election Using the
universe of US births we estimate a shift of 19000 to 37000 births across political geogra-
phies We likewise find a disparate and negative impact on Hispanic relative to non-Hispanic
fertility within the same county Reinforcing the idea that political sentiment drives these
changes relative Hispanic fertility declines after a Trump pre-election campaign visit
Across many nations growing political polarization and declining fertility are two funda-
mental challenges facing society We estimate effects at the intersection of these two forces
From a policy perspective shifts in fertility across political groups have practical implica-
tions for natality programs and for regional public finance given the stark partisan sorting
across residential geographies
This is the first paper to causally link political partisanship to fertility choices opening
up several avenues for future research It would be interesting to further disentangle whether
fertility is responding to pure sentiment different partisan information-processing models
or expectations of how future policy will benefit them While we estimate relative effects
14
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
how aggregate fertility responds including over longer time horizons is still unknown Fur-
ther questions include whether Democratic versus Republican Presidencies have differential
impacts and how increasing polarization over time amplifies fertility effects
References
S Abraham and L Sun Estimating dynamic treatment effects in event studies with hetero-
geneous treatment effects arXiv preprint arXiv180405785 2018
A Aizer S Eli and A Lleras-Muney The incentive effects of cash transfers to the poor
Technical report National Bureau of Economic Research 2020
D Autor D Dorn G Hanson et al When work disappears Manufacturing decline and
the falling marriage market value of young men American Economic Review Insights 1
(2)161ndash78 2019
D Autor D Dorn G Hanson K Majlesi et al Importing political polarization the
electoral consequences of rising trade exposure American Economic Review 110(10)
3139ndash83 2020
L M Bartels Beyond the running tally Partisan bias in political perceptions Political
behavior 24(2)117ndash150 2002
G S Becker An economic analysis of fertility In Demographic and economic change in
developed countries pages 209ndash240 Columbia University Press 1960
J Benhabib and M M Spiegel Sentiments and economic activity Evidence from us states
The Economic Journal 129(618)715ndash733 2019
D A Black N Kolesnikova S G Sanders and L J Taylor Are children ldquonormalrdquo The
Review of Economics and Statistics 95(1)21ndash33 2013
L Boxell M Gentzkow and J M Shapiro Cross-country trends in affective polarization
Technical report National Bureau of Economic Research 2020
J R Brown and R D Enos The measurement of partisan sorting for 180 million voters
Nature Human Behaviour pages 1ndash11 2021
A Cohen R Dehejia and D Romanov Financial incentives and fertility Review of Eco-
nomics and Statistics 95(1)1ndash20 2013
15
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
N Confessore and N Cohn Donald trumprsquos victory was built on unique coalition of white
voters New York Times 2016
J A Cookson J E Engelberg and W Mullins Does partisanship shape investor beliefs
evidence from the covid-19 pandemic The Review of Asset Pricing Studies 10(4)863ndash893
2020
J B Cullen N Turner and E L Washington Political alignment attitudes toward gov-
ernment and tax evasion Technical report National Bureau of Economic Research 2020
R Dehejia and A Lleras-Muney Booms busts and babiesrsquo health The Quarterly Journal
of Economics 119(3)1091ndash1130 2004
L J Dettling and M S Kearney House prices and birth rates The impact of the real estate
market on the decision to have a baby Journal of Public Economics 11082ndash100 2014
M Dimock and R Wike America is exceptional in the nature of its political divide Pew
Research Center 2020
B Duncan H Mansour and D I Rees Itrsquos just a game the super bowl and low birth
weight Journal of Human Resources 52(4)946ndash978 2017
G Evans and R Andersen The political conditioning of economic perceptions The Journal
of Politics 68(1)194ndash207 2006
M Gentzkow Polarization in 2016 Toulouse Network for Information Technology Whitepa-
per pages 1ndash23 2016
A S Gerber and G A Huber Partisanship and economic behavior Do partisan differences
in economic forecasts predict real economic behavior American Political Science Review
pages 407ndash426 2009
C Gillitzer and N Prasad The effect of consumer sentiment on consumption Cross-
sectional evidence from elections American Economic Journal Macroeconomics 10(4)
234ndash69 2018
S Iyengar Y Lelkes M Levendusky N Malhotra and S J Westwood The origins and
consequences of affective polarization in the united states Annual Review of Political
Science 22129ndash146 2019
C I Jones The end of economic growth unintended consequences of a declining population
Technical report National Bureau of Economic Research 2020
16
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
M Kearney and P Levine The coming covid-19 baby bust
2020 URL httpswwwbrookingsedublogup-front20201217
the-coming-covid-19-baby-bust-update
M S Kearney and P B Levine Subsidized contraception fertility and sexual behavior
The review of Economics and Statistics 91(1)137ndash151 2009
E Kempf and M Tsoutsoura Partisan professionals Evidence from credit rating analysts
Technical report National Bureau of Economic Research 2020
A Langer and D Lemoine What were the odds estimating the marketrsquos probability of
uncertain events Technical report National Bureau of Economic Research 2020
J M Lindo Are children really inferior goods evidence from displacement-driven income
shocks Journal of Human Resources 45(2)301ndash327 2010
M Lino The cost of raising a child 2020 URL httpswwwusdagovmediablog2017
0113cost-raising-child
M F Lovenheim and K J Mumford Do family wealth shocks affect fertility choices
evidence from the housing market Review of Economics and Statistics 95(2)464ndash475
2013
J McCrary and H Royer The effect of female education on fertility and infant health
evidence from school entry policies using exact date of birth American economic review
101(1)158ndash95 2011
M C McGrath Economic behavior and the partisan perceptual screen Quarterly Journal
of Political Science 11(4)363ndash83 2017
M Meeuwis J A Parker A Schoar and D I Simester Belief disagreement and portfolio
choice Technical report National Bureau of Economic Research 2018
A R Mian A Sufi and N Khoshkhou Partisan bias economic expectations and household
spending Fama-Miller working paper 2018
K Milligan Subsidizing the stork New evidence on tax incentives and fertility Review of
Economics and statistics 87(3)539ndash555 2005
D Novgorodsky and B Setzler Practical guide to event studies 2019
A Phillips Theyrsquore rapistsrsquopresident trumprsquos campaign launch speech two years later
annotated The Washington Post 16(6)2017 2017
17
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
A Raute Can financial incentives reduce the baby gap evidence from a reform in maternity
leave benefits Journal of Public Economics 169203ndash222 2019
J Schaller Booms busts and fertility testing the becker model using gender-specific labor
demand Journal of Human Resources 51(1)1ndash29 2016
18
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Figure 1 Economic Outlook by Party AffiliationNote This figure plots the percentage of positive minus negative responses (net better) to the questionldquoDo you think the nationrsquos economy is getting better or worserdquo among registered voters The survey isadministered by CIVIQS The blue line denotes responses among Democrats the red line denotes Republicanresponses CIVIQS uses a list-based sampling methodology to select panelists to receive daily online pollsThey use a statistical model (dynamic Bayesian multilevel regression with post-stratification weights) toadjust the demographics of the sample to those of the population and smooth out day-to-day samplingvariability
19
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
20
A1 Dem vs Rep B1 High vs low Rep shift C1 Vote share times shift
A2 Dem vs Rep B2 High vs low Rep shift C2 Vote share times shift
Figure 2 Effects of 2016 Election on Fertility in Democratic and Republican CountiesNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate in Democratic-leaning relative to Republican-leaning countiesaround the 2016 Presidential election Fertility rate in panels A1 through C1 are normalized by their levels in September 2016 The vertical shadedarea spanning the months of October and November indicates the period immediately surrounding the election As described in section 22 Octoberrepresents a partially-treated month due to how conceptions are measured in our data Panel A1 plots the excess fertility rate in counties with above-median (blue) versus below-median (red) Democrat vote shares in the 2012 Presidential election Panel B1 plots the excess fertility rate in countieswith above-median (red) versus below-median (blue) change in Republican vote shares between the 2008 and 2016 Presidential elections Panel C1plots the excess fertility rate in counties with both below-median Democrat vote shares and above-median shifts in Republican vote share (red) versuscounties where both measures are the opposite (blue) Panels A2 through C2 plot the estimates of coefficients on the interactions between a quarterlyevent vector and an indicator for Democratic-leaning counties from equation 1 The omitted quarter is -1 ie July August and September of 2016Specifications correspond to those in Table 1 columns (1) through (3)
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
A1 Hisp vs non-Hisp B1 Hisp vs rural white
A2 Hisp vs non-Hisp B2 Hisp vs rural white
Figure 3 Effects of 2016 Election on Hispanic and Non-Hispanic FertilityNote This figure plots effects (and 90 confidence intervals) for the excess fertility rate of Hispanics versusother groups around the 2016 Presidential election Fertility rate in panels A1 and B1 are normalized bytheir levels in September 2016 The vertical shaded area spanning the months of October and Novemberindicates the period immediately surrounding the election As described in section 22 October representsa partially-treated month due to how conceptions are measured in our data Panel A1 plots the excessfertility rate of Hispanics (solid) versus non-Hispanics (dash) Panel B1 plots the excess fertility rate ofHispanics (solid) versus non-Hispanic whites living in rural counties (dash) Census classifies a county asrural if more than 50 of the population is living in rural areas (ie census tracts or blocks where thereare fewer than 2500 people or fewer than 1500 people living outside institutional group quarters) PanelsA2 and B2 plot the interactions between a quarterly event vector and an indicator for being Hispanic fromequation 2 The omitted quarter is -1 ie July August and September of 2016 Specifications correspondto those in Table 1 columns (4) and (5)
21
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Figure 4 Effects of Trumprsquos Pre-Election Campaign Visits on RelativeHispanic Fertility
Note This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a county on the monthly fertility of Hispanics and non-Hispanics in the same county relative to thatof Hispanics and non-Hispanics in counties with later visits The omitted event period is the third monthbefore Trumprsquos visit Specifications correspond to those in Table A5 column (1)
22
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Table 1 Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5)Dem vs High vs low Vote share Hisp vs Hisp vs
Rep Rep shift times shift non-Hisp rural white
Treatminus3 -0031 -0055 -0082 -0050 -0007(0061) (0060) (0093) (0057) (0086)
Treatminus2 -0012 -0065 -0084 -0065 -0062(0053) (0057) (0088) (0050) (0082)
Treat0 -0144 -0168 -0264 -0196 -0245(0056) (0058) (0091) (0049) (0083)
Treat1 -0099 -0198 -0368 -0272 -0434(0059) (0062) (0087) (0056) (0089)
Treat2 -0179 -0289 -0421 -0314 -0513(0065) (0064) (0093) (0056) (0088)
Treat3 -0175 -0308 -0448 -0315 -0546(0075) (0073) (0104) (0058) (0096)
Sum Treat(0-3) -597 -964 -1501 -1097 -1739p value 005 0000 0000 0000 0000
Observations 19691 19691 11438 39620 30947R-squared 0424 0425 0446 0270 0260County FE Y Y Y Y YQuarterly event FE Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830
Note This table reports the effects of the 2016 Presidential election on excess fertility rates in Democratic-leaning relative to Republican-leaning counties and on excess fertility rate of Hispanics relative to otherracialethnic groups Columns (1) through (3) report coefficients on interactions between a quarterlyevent vector and a Democratic-leaning indicator from equation 1 Column (1) defines Democratic-leaning(Republican-leaning) counties as those with above-median (below-median) Democrat vote shares Column(2) defines Democratic-leaning counties as those with an above-median change in Republican vote sharesColumn (3) interacts these two measures to identify extreme Republican and extreme Democratic countiesColumns (4) and (5) report coefficients on interactions between a quarterly event vector and an indicatorfor being Hispanic from equation 2 The comparison group is non-Hispanics in column (4) and non-Hispanicwhites in rural counties in column (5) Census classifies a county as rural if more than 50 of the populationis living in rural areas (ie census tracts or blocks where there are fewer than 2500 people or fewer than1500 people living outside institutional group quarters) The omitted quarter is -1 ie July August andSeptember of 2016 Standard errors are clustered by county
23
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Online Appendix for
ldquoPartisan Fertility and Presidential Electionsrdquo1
This appendix provides additional empirical evidence to supplement the main text
1Citation format Gordon Dahl Runjing Lu and William Mullins Online Appendix for ldquoPartisan Fertilityand Presidential Electionsrdquo 2021 Working Paper
24
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
2016 Presidential Election 2020 Presidential Election
-20
020
40R
epub
lican
- D
emoc
rat (
p
ositi
ve re
spon
se)
2015
Jan
2015
Apr
2015
Jul
2015
Oct
2016
Jan
2016
Apr
2016
Jul
2016
Oct
2017
Jan
2017
Apr
2017
Jul
2017
Oct
2018
Jan
2018
Apr
2018
Jul
2018
Oct
2019
Jan
2019
Apr
2019
Jul
2019
Oct
2020
Jan
2020
Apr
2020
Jul
2020
Oct
2021
Jan
Figure A1 Optimism survey the Bloomberg Consumer Comfort Index byParty Affiliation
Note The Bloomberg Consumer Comfort Index is derived from 1000 landline and cellular telephone in-terviews (national random sample) 250 per week reported as a four-week rolling average Respondentsare asked to rate (i) the national economy (ii) their personal finances and (iii) the buying climate in thefollowing 4 categories Excellent Good Not so Good Poor The index produced weekly since December1985 is derived by averaging positive responses to each subindex question The scale range is 0-100 and theresults of each question have a 35-point error margin Interviews are conducted in English and SpanishData are weighted to adjust for probabilities of selection given household size and telephone usage as wellas US Census Bureau data for age sex race education metro status region and phone service
25
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
A Dem vs Rep
B High vs low Rep shift
C Vote share times shift
Figure A2 Effects of 2016 Election on Fertility in Democratic andRepublican Counties ndash Monthly Event Study
Note This figure parallels Figure 2 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
26
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
A Hisp vs non-Hisp
B Hisp vs rural white
Figure A3 Effects of 2016 Election on Hispanics and Non-HispanicFertility - Monthly Event Study
Note This figure parallels Figure 3 but is at the monthly rather than the quarterly level Plots include 90confidence intervals
27
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Figure A4 Effects of 2016 Election on Fertility in More and LessPolarized Counties
Note This figure plots heterogeneous effects of the 2016 Presidential election on the excess fertility rateof Hispanics versus non-Hispanics in counties that are more or less polarized A county is defined as more(less) polarized if the county experienced an above-median (below-median) level of the instrumented Chinatrade shock between 2000 and 2008 (Autor et al 2020) The plotted estimates are obtained by includinginteraction terms for being moreless polarized in equation 2 The omitted quarter is -1 ie July Augustand September of 2016 Specification corresponds to that in Table A3 Plots include 90 confidenceintervals
28
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Figure A5 Google Search Index for Trump around Campaign VisitsNote This figure plots dynamic treatment effects (and 90 confidence intervals) for Trumprsquos first campaignvisit to a DMA on the weekly Google search index for ldquoTrumprdquo in the same DMA relative to DMAs withlater visits The omitted event period is the third week before Trumprsquos visit
29
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Table A1 Summary Statistics
Panel A Election sampleFertility rate Excess fertility rate
Race Mean SD Mean SDTotal 4499 0855 -0113 0493Hispanic 5140 1461 -0279 1041African American 4612 1569 -0023 1211Mexican 4729 1802 -0239 1402Non-Hispanic 4287 0889 -0083 0537Non-AA 4147 0910 -0132 0520Non-Hisp minority 4731 1328 -0065 0994Non-Hisp white 4263 1028 -0086 0628Num county 2830 2830 2830 2830
Panel B Campaign sampleRace Mean SDHispanic 5159 1064Non-Hispanic 4299 0649Non-Hisp white 4204 0768Num county 230 230
Notes The fertility rate corresponds to monthly births per 1000 women who are between15 and 44 years old in a county The excess fertility rate is calculated by subtracting eachcountyrsquos monthly average fertility using data from 2012 onward
30
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
31
Table A2 Robustness - Effects of 2016 Election on Fertility
(1) (2) (3) (4) (5) (6) (7) (8) (9)Dem vs Rep H vs L Rep shift Vote sharetimesshift HP vs non-HP HP vs Rural white Dem vs Rep H vs L Rep Mexican MinorityIncome ctrl Income ctrl Income ctrl Income ctrl Income ctrl p40p60 shift 12-16 vs non-HP vs white
Treatminus3 -0031 -0055 -0082 -0050 -0007 0011 0031 0013 -0002(0061) (0060) (0093) (0057) (0086) (0071) (0061) (0089) (0075)
Treatminus2 -0012 -0065 -0084 -0065 -0062 0003 0000 -0060 0078(0053) (0057) (0088) (0050) (0082) (0061) (0055) (0068) (0056)
Treat0 -0144 -0168 -0264 -0196 -0245 -0130 -0177 -0244 0042(0056) (0058) (0091) (0049) (0083) (0062) (0057) (0055) (0053)
Treat1 -0101 -0202 -0366 -0272 -0458 -0121 -0227 -0403 -0168(0061) (0064) (0090) (0056) (0090) (0063) (0063) (0065) (0056)
Treat2 -0181 -0293 -0419 -0314 -0537 -0230 -0244 -0392 -0090(0067) (0065) (0096) (0057) (0089) (0069) (0068) (0064) (0061)
Treat3 -0177 -0312 -0446 -0315 -0570 -0183 -0314 -0414 -0040(0076) (0074) (0107) (0058) (0097) (0083) (0074) (0066) (0065)
Income (10k) 0031 0048 -0023 -0040 0201(0114) (0112) (0135) (0107) (0163)
Income2 (10k) -0001 -0001 0001 0001 -0004(0004) (0004) (0004) (0004) (0005)
Sum Treat(0-3) -602 -976 -1496 -1096 -1809 -664 -963 -1453 -256p value 006 0000 0000 0000 0000 003 0000 0000 188
Observations 19691 19691 11438 39620 30947 15750 19691 39620 39487R-squared 0424 0425 0446 0270 0260 0429 0425 0234 0238County FE Y Y Y Y Y Y Y Y YQuarterly event FE Y Y Y Y Y Y Y Y YN cluster (county) 2813 2813 1634 2830 2830 2250 2813 2830 2830
Notes Columns (1) through (5) parallel the specification in Table 1 but add controls for county income and income squared Column (6) categorizedRepublican and Democratic-leaning by whether a county is below the 40th percentile or above the 60th percentile in the Democrat vote share Column(7) uses the shift in Republican vote shares between 2012-2016 instead of 2008-2016 Column (8) replaces Hispanics (HP) as the treatment group withMexicans while column (9) compares non-Hispanic minorities versus whites The omitted quarter is -1 ie July August and September of 2016Standard errors are clustered by county
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Table A3 Effects of 2016 Election on Fertility in More and LessPolarized Counties
(1) (2)More polarized Less polarized
Treatminus3 -0063 -0040(0062) (0091)
Treatminus2 -0100 -0031(0059) (0079)
Treat0 -0260 -0144(0064) (0071)
Treat1 -0410 -0154(0076) (0073)
Treat2 -0428 -0226(0075) (0079)
Treat3 -0393 -0243(0071) (0088)
Sum Treat(0-3) col(1) - col(2) -0722p value 0031
Observations 117978R-squared 0116County FE YQuarterly event FE YN cluster (county) 2830
Note This table presents heterogeneous effects of the 2016 Presidential election onthe excess fertility rate of Hispanics versus non-Hispanics in counties that are more orless polarized A county is defined as more (less) polarized if the county experiencedan above-median (below-median) level of the instrumented China trade shock between2000 and 2008 (Autor et al 2020) The estimates are obtained by including interactionterms for being moreless polarized in equation 2 The omitted quarter is -1 ie JulyAugust and September of 2016 Standard errors are clustered by county
32
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Table A4 Effects of 2016 Election on Google Searches forldquoPregnancy Testrdquo
(1) (2) (3)Dem vs Rep High vs low Rep shift High vs low Hisp
Treatminus3 -3587 0628 3322(3483) (3114) (2851)
Treatminus2 -2514 -0226 -0155(2842) (2925) (2648)
Treat0 -5709 -7443 -6320(2899) (2818) (2813)
Treat1 -3589 -6193 -5425(3109) (2683) (2834)
Treat2 -1992 -2560 -2241(3506) (3253) (3235)
Treat3 -0424 -5333 -4427(4146) (3604) (3485)
Sum Treat (0-3) -11714 -21529 -18413p value 343 048 086
Observations 1435 1435 1456R-squared 0490 0497 0498DMA FE Y Y YQuarterly event FE Y Y YN cluster (DMA) 201 201 203
Note This table reports the effects of the 2016 presidential election on Google searches for ldquopregnancytestrdquo The dependent variable is the weekly percentage of Google searches taken from a random sample oftotal searches and scaled by the highest weekly search rate in the same DMA during the entire extractionperiod Since each extraction is based on a random sample we use the average Google search rate across15 extractions between November 2020 and January 2021 as our outcome Columns (1) through (3) reportcoefficients for the interactions between a quarterly event vector and an indicator for DMAs having above-median Democratic vote shares in the 2012 election an indicator for DMAs having an above-median changein Republican vote shares from the 2012 to the 2016 election and an indicator for having above-medianpercentage of the Hispanic population respectively The omitted quarter is -1 namely July August andSeptember of 2016 Standard errors are clustered by DMA
33
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-
Table A5 Effects of Trumprsquos Campaign Visits on Fertility by HispanicOrigin
HP vs non-HP ∆Agg wtTreat-7M -0068 -0021
(0062) (0055)Treat-6M -0070 -0018
(0062) (0053)Treat-5M 0047 0031
(0062) (0050)Treat-4M -0018 -0040
(0063) (0053)Treat-2M 0042 0040
(0058) (0051)Treat-1M -0108 -0127
(0059) (0048)Treat0M -0073 -0093
(0068) (0065)Treat1M -0154 -0166
(0069) (0056)Treat2M -0157 -0177
(0083) (0077)Treat3M -0029 -0060
(0091) (0077)Treat4M 0009 -0004
(0087) (0083)Treat5M 0123 0140
(0110) (0100)
Sum Treat (-1 to 5) -0486 -0486p value 0203 0203
Observations 129872 129872R-squared 0412 0412Outcome mean 4540 4540N cluster (county) 230 230
Note This table presents the dynamic treatment effects of Trumprsquos first campaign visit to a county onthe Hispanic fertility rate relative to the non-Hispanic rate using counties with later visits as controlsThe omitted quarter is the third month before Trumprsquos first visit to a county Column (1) uses Hispanicwomen population in the treatment cohort as the aggregation weight while column (2) uses the total womenpopulation in the treatment cohort Standard errors are clustered by county
34
- Introduction
- Evidence from the 2016 Presidential Election
-
- Fertility Data
- Research Design
- Optimism and Elections
- Fertility Effects across Political Geographies
- Fertility Effects by Ethnicity
- Robustness Checks
-
- Evidence from Pre-Election Campaign Visits
-
- Data
- Research Design
- Fertility Response to Trump Campaign Visits
-
- Conclusion
-