democracy in america? partisanship, polarization, and the ...€¦ · violate core democratic...
TRANSCRIPT
American Political Science Review (2020) 114, 2, 392–409
doi:10.1017/S0003055420000052 © American Political Science Association 2020
Democracy in America? Partisanship, Polarization, and theRobustness of Support for Democracy in the United StatesMATTHEW H. GRAHAM Yale University
MILAN W. SVOLIK Yale University
I s support for democracy in the United States robust enough to deter undemocratic behavior by electedpoliticians?Wedevelop amodel of the public as ademocratic check and evaluate it using two empiricalstrategies: anoriginal, nationally representative candidate-choice experiment inwhich somepoliticians
take positions that violate key democratic principles, and a natural experiment that occurred duringMontana’s 2017 special election for the U.S. House. Our research design allows us to infer Americans’willingness to trade-off democratic principles for other valid but potentially conflicting considerations suchas political ideology, partisan loyalty, and policy preferences. We find the U.S. public’s viability asa democratic check to be strikingly limited: only a small fraction of Americans prioritize democraticprinciples in their electoral choices, and their tendency to do so is decreasing in several measures of po-larization, including the strength of partisanship, policy extremism, and candidate platform divergence.Our findings echo classic arguments about the importance of political moderation and cross-cuttingcleavages for democratic stability and highlight the dangers that polarization represents for democracy.
“It is the function of public opinion to check the use offorce in a crisis, so that men, driven to make terms, maylive and let live.”
Walter Lippmann, The Phantom Public (1925, 64)
INTRODUCTION
“It is nearly impossible to find an American whosays that he is opposed to democracy or favorssome alternative…On the contrary, nearly
everyone professes to believe that democracy is the bestform of government.” This is how Robert A. Dahl,writing in 1966, summarized contemporary evidence forthe support for democracy in the United States (Dahl1966, 40). It remains conventional wisdom to this day.Research that traces its intellectual origins to Tocque-ville’s Democracy in America finds that the UnitedStates consistently exhibits some of the highest levels ofsupport for democracy in theworld (AlmondandVerba1963; Inglehart and Welzel 2010; Norris 2011).
We show that this conventional wisdom rests onfragile foundations. Rather than asking about support
for democracy directly, we adopt an approach thatinfersAmericans’ commitment todemocratic principlesfrom their choices of candidates in hypothetical electionscenarios. Each candidate is experimentally assignedattributes and platforms that approximate real-worldelections and, crucially, may endorse positions thatviolate core democratic principles, including free andfair elections, civil liberties, and checks and balances. Inthis framework, voters “support democracy” not whenthey say so, but rather when their choices reveala preference for democratic principles over other validbut potentially conflicting considerations such as po-litical ideology, partisan loyalty, or policy preferences.
This research design builds on the observation thatelections represent a fundamental instrument of dem-ocratic self-defense: Especially in advanced de-mocracies, voters have the opportunity to stoppoliticians whose positions violate democratic princi-ples by defeating them at the polls. We argue that a keyobstacle to the viability of such a democratic check ispartisan, ideological, and policy-based polarization.Electoral competition often confronts voters witha choice between two valid but potentially conflictingconsiderations: partisan interests and democraticprinciples. Polarization raises the stakes of elections andin turn the price of prioritizing democratic principlesover partisan interests. When faced with the choicebetween a co-partisan candidate whose positions vio-late democratic principles and a candidate who com-plies with democratic principles but is otherwiseunappealing, a significant fraction of voters may sacri-fice democratic principles to elect a candidate whochampions their party or interests. In a sharply polar-ized electorate, even pro-democratically minded votersmay act as partisans first and democrats only second.
In the next section, we formalize these intuitions anddevelop a model of the public as a democratic check.Building on Svolik (2020), we extend the classic, spatialframework for electoral competition to account forcandidates who may hold positions that underminedemocratic principles. The latter are conceptualized as
Matthew H. Graham , Ph.D. Candidate, Department of PoliticalScience, Yale University, [email protected].
Milan W. Svolik , Professor, Department of Political Science,Yale University, [email protected].
Wewould like to thank JacobHacker,DavidHendry, GregHuber,Steven Levitsky, DavidMayhew, NolanMcCarty, AdamPrzeworski,Arturas Rozenas, Daniel Treisman, Bonnie Weir, and audiences atAPSA, EPSA, Boston University, Caltech, Columbia, Georgetown,Harvard, IAS Toulouse, Peking University, Technical University ofMunich,TexasA&M,UCBerkeley,UTAustin,Vanderbilt, andYalefor their helpful comments and conversations and Isaiah Affron andLilyEngbith for researchassistance.Thisprojectwas supportedby theInstitution for Social and Policy Studies and the Whitney and BettyMacMillan Center for International and Area Studies, both at YaleUniversity. Replication files are available at the American PoliticalScience Review Dataverse: https://doi.org/10.7910/DVN/EEARKA.
Received: March 20, 2019; revised: September 27, 2019; accepted:January 28, 2020.
392
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
negativevalence attributes:while votersmaydiffer overpolicy, ideology, or partisanship, they agree that elec-toral competition should be democratic and prefercandidates who comply with key democratic principles.This framework yields a number of predictions aboutthe consequences of polarization for an electorate’sresilience to undemocratic candidates: (i) voters whohold extreme or intense policy preferences will bewilling to sacrifice democratic principles at higher ratesthan centrist and moderate voters, (ii) electorates thatare polarized or lack cross-cutting cleavages will be lesspunishing of candidates that undermine democraticprinciples, and (iii) candidate platform polarization willbe detrimental to democracy independent of voterpolarization. Our model thus provides micro-founda-tions for classic arguments about the importance ofpolitical moderation and cross-cutting cleavages fordemocratic stability (Dahl 1956; Lipset 1960).
This framework guides the design and analysis of ourcandidate-choice experiment as well as a natural ex-periment that occurred during Montana’s 2017 specialelection for theU.S.House. The following is a summaryof our experimental findings:
1. Americans value democracy, but not much: A candi-date who considers adopting an undemocratic positioncan expect to be punished by losing only about 11.7%ofhis overall vote share. When we restrict attention tocandidate-choice scenarios with combinations of par-tisanship and policies that we typically see in real-worldelections, this punishment drops to 3.5%.
2. Support for democracy is highly elastic: When the priceof voting for a more democratic candidate is that can-didate’s greater distance from the voter in terms of herpreferred policies, even the most centrist voters arewilling to tolerate at most a 10–15% increase in sucha distance.
3. Centrists are a pro-democratic force: “Centrist” voterswho see small policy differences between candidatespunish undemocratic behavior at four times the rate of“extremist” voters who strongly favor one of thecandidates.
4. Most voters are partisans first and democrats onlysecond: Only about 13.1% of our respondents arewilling to defect from a co-partisan candidate for vio-latingdemocraticprincipleswhen thepriceofdoing so isvoting against their own party. Only independents andpartisan “leaners” supportmore democratic candidatesenough to defeat undemocratic ones regardless of theirpartisan affiliation.
5. Supporters of both parties employ a partisan “doublestandard”:Respondentswho identify asRepublicanaremore willing to punish undemocratic behavior byDemocratic Party than Republican Party candidatesand vice versa. These effects are about equal amongboth Democrat and Republican respondents.
6. Platform polarization is bad for democracy: The largerthedifferencebetween thecandidates’policyplatforms,the weaker the punishment for their undemocraticbehavior.
7. Sensitivity to the menu of manipulation varies: Votersare most punishing of undemocratic positions that
undermine the free press and the rule of law; they areleast punishing of restrictions on the freedom of as-sembly and executive aggrandizement. Nonetheless,when we benchmark these against extramarital affairsand underpayment of taxes—two negative valenceattributes unrelated to democracy—we find that voterspunish the latter more severely than they punish vio-lations of democratic principles.
8. Americans have a solid understanding of what de-mocracy is and what it is not: The vast majority of ourrespondents correctly distinguish real-world un-democratic practices—including those endorsed by ourexperimental candidates—from those that are consis-tent with democratic principles.
After examining these findings nonparametrically,we take advantage of the close connection between thedesign of our candidate-choice experiment and ourtheoretical framework and adopt a structural approach.This approach allows us to parsimoniously summarizethe variation in our respondents’ candidate choicesusing a few theoretically grounded parameters, in-cluding the weight that voters place on democracyrelative to party and policy. We estimate this weight at18% and find that candidates’ democracy positions,economic and social policy platforms, and partisan af-filiation jointly account for 80% of the systematic var-iation in voters’ candidate choices. We also “price”democracy in terms of the candidate characteristicsvoters are willing to forgo to punish candidates whoviolate democratic principles. We find that this pricecorresponds to about half the difference between ourrespondents’ favorite and least favoritepolicy positions.Wethenextendour theoreticalmodel toaccount forourexperimental findings about a partisan double standardin the punishment of candidateswho violate democraticprinciples. We estimate a 48.4% co-partisan bias, im-plying that Americans are neither fully principled norpurely instrumental in their support for democracy.
The last portion of our analysis moves from hypo-thetical election scenarios to a real-world election:a natural experiment that occurred during the 2017specialelectionforMontana’sonly seat in theU.S.HouseofRepresentatives.On the eve of the election, one of thetwo major candidates assaulted a journalist, which weinterpret as a negative public signal about his respect fora free press, or at a minimum, an undesirable valenceattribute. Crucially, only election day voters saw thissignal before they voted; absentee voters, who in Mon-tanamakeupamajorityof registeredvoters, had alreadycast their ballots. This allows us to adopt a difference-in-differences empirical strategy that compares precinct-level vote shifts between absentee and election dayvoters to infer theirwillingness to punish the assault.Ourfindings are consistent with both our theoreticalexpectations and experimental results.We find that onlymoderateprecincts punished theassault on the journalistby voting across party lines. In heavily partisan precincts,partisan loyalty trumped valence considerations.
Our findings about the robustness of support for de-mocracy in the United States contribute to a number ofdebates in comparative and American politics. Most
Democracy in America?
393
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
immediately, our article joins a growing bodyofwork thathas, especially in the wake of the 2016 presidential elec-tion, begun to reassess our knowledge about democraticstability in the United States1 and other advanced de-mocracies.2 Theoretically, our arguments parallel Levit-sky andZiblatt (2018),whoalso highlight thedangers thatpolarization represents for democratic stability. Whereastheir analysis focuses primarily on how polarizationweakens the often informal norms that regulate inter-actions among political elites, our emphasis is on howpolarization undermines the public’s capacity to punishpoliticians for subverting the democratic process.
Our empirical methodology and substantive focus areclosest to Carey et al. (2018), who also employ a candi-date-choice experiment to study the commitment todemocratic principles among theAmerican public. Theytoo report that although voters do punish candidateswhose positions violate democratic principles, themagnitude of that punishment may be overshadowed byother political considerations—most notably partisan-ship. As we discuss in the conclusion, these findingssuggest that conventional, direct-questioning measuresof support fordemocracymaybeflawedbecause they failto account for voters’willingness to trade off democraticprinciples for partisan interests. In turn, our existingknowledge about the support for democracy in theUnited States and other advanced democracies is oflimitedutilitywhen it comes to answering akeyquestion:When can we realistically expect the public to check theauthoritarian temptations of elected politicians?
Our theoretical framework helps us to address thisquestion by proposing a new perspective on democraticstability: democracy is “self-enforcing” when politiciansanticipate that, were they to behave undemocratically,their own supporters would punish them by voting fora competitor in large enough numbers to bring abouttheir defeat.Weexplainwhy this electoral checkmay failin polarized societies and even among voters who valuedemocracy for its own sake: in sharply divided societies,voters put partisan ends above democratic principles.3
The microfoundations that we develop thus combineinsights from two lines of classic democratization re-search. The first views intense political cleavages asa threat to democratic stability (Dahl 1956; Lipset 1960);the second asks that explanations of democratic stabilitybe explicit about the incentives of key actors to complywith the rules of democratic politics (Przeworski 1991;Weingast 1997). More broadly, we contribute to com-parative politics research on democratic backsliding(Gandhi and Ong 2019; Haggard and Kaufman 2016;
Luo and Przeworski 2018; Nalepa, Vanberg, andChiopris 2018; Waldner and Lust 2018).
Jointly, our theoretical framework and empiricalfindings suggest an explanation for the puzzling per-sistence in theUnited States of a number of deficienciesin the democratic process, especially at the state andlocal levels. Most frequent among these aregerrymandering (Chen and Rodden 2013; Cho and Liu2016) and voter suppression (Grimmer et al. 2018). Ourmodel implies that electedofficials’ incentives tocomplywith key democratic principles critically depend on thepublic’s willingness to sanction those who violate themor neglect their enforcement.4 Yet, our empiricalanalysis reveals that this check is at best limited inmagnitudeand subject toapartisandouble standard.Aswe discuss in the conclusion, our findings imply thatpublic officials may be effectively insulated from elec-toral sanction in states and districts where one partyenjoys a significant electoral advantage.
Analytically as well as empirically, we examine thepernicious consequences of a number of distinct con-ceptions of polarization. At the level of individualvoters, polarization may characterize voters who holdeither extremeor intensepreferences.At the levelof theelectorate, polarization may correspond to either a U-shaped distribution of voter preferences or one witha high correlation of preferences across issues or be-tween issue areas and partisan affiliation.At the level ofthe candidates, polarization can be conceived of ascorresponding to a large distance between candidateplatforms. These distinct conceptions of polarizationare rarely examined within a single theoretical frame-work. Our analysis demonstrates that each kind ofpolarization independently undermines the electorate’sresilience to undemocratic candidates.
This comprehensive examination of the relationshipbetween polarization and democratic stability con-tributes to a large research agenda that studies elite andmass polarization in the United States. Whereas mostresearch onmass polarization focuses on characterizingits nature and origins (Abramowitz and Saunders 2008;Fiorina, Abrams, and Pope 2008; Iyengar et al. 2018;McCarty, Poole, and Rosenthal 2008), our focus is ona political consequence that this literature has yet toexamine.5 To our knowledge, our study is the first toexamine what is possibly the most concerning conse-quence of increasing polarization in the United States:its potential to undermine the public’s ability to checkthe undemocratic temptations of elected politicians.
A MODEL OF THE PUBLIC AS ADEMOCRATIC CHECK
Consider amodel in which voters’ preferences over twokinds of candidate attributes determine their electoral
1 See especially Carey et al. (2019), Huq and Ginsburg (2018),Kaufman and Haggard (2019), Lieberman et al. (2019), Miller, Sza-konyi, and Morgenbesser (2017), and Przeworski (2019).2 SeeBecherandBrouard(2019), FoaandMounk(2017), andVoeten(2017).3 The trade-off between partisan interests and democratic principlescan be seen as a special case of a more general trade-off betweenpartisan and valence considerations (Ashworth and Bueno de Mes-quita, 2009). Such an interpretation of our findings echoes Eggers(2014), who finds that voters punished politicians implicated in the2009 U.K. expenses scandal less severely when the electoral stakes intheir district were higher.
4 Clark (2009) sees a similar role for public opinion in his analysis ofcourt curbing.5 Because our primary focus is onmass polarization, we do not discussresearch on polarization in Congress. For a review of this literature,see Barber and McCarty (2015).
Matthew H. Graham and Milan W. Svolik
394
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
choices: (i) positional issues, including candidates’policy positions, ideology, and partisan affiliation and(ii) candidates’ compliance with key democratic prin-ciples. Building on Svolik (2020), we conceive of thelatter as a valence issue: while voters may differ overpolicy, ideology, or partisanship, they all prefer candi-dates whose positions comply with key tenets of dem-ocratic electoral competition.
Formally, voter i’s payoff from candidate j is
ui Xj;Mj� � ¼ ��
K
k¼1ak xik � xjk
� �2 � dMj; (1)
where for eachofK issues,xik is voter i’s favoritepositionon issuek,xjk is candidate j’splatformon issuek, andak isthe weight that i attaches to that issue. Meanwhile,Mj iscandidate j’sdemocracypositionwhereM is increasing inhow undemocratic j’s platform is (M stands for “ma-nipulation”).Thetermd is theweight thatvotersattachtofair democratic competition—in effect, the intensity oftheir support for democracy.
Thismodel yields severalpredictions thatweevaluatethroughout the article. First, voters who hold intense orextreme policy preferences are willing to tolerate un-democratic behavior by their favored candidate. To seethe intuition behind these predictions, suppose there isonly a single policy issue k. Then, i votes for candidate 1as long as
xik $x1k þ x2k
2þ d M1 �M2ð Þ2ak x1k � x2kð Þ for x1k > x2k; (2)
where we are assuming that candidate 1’s policy plat-form is to the right of candidate 2’s platform.
Call the voter whose ideal policy xik barely satisfiesthe inequality in (2) the swing voter xjs.Note that thefirstterm on the right-hand side of this inequality is themidpoint between the two candidates’ policy platforms,separating the electorate into thosewho are policy-wisecloser to candidate 1 and those who are closer to can-didate 2. The swing voter xjs is located either to the rightor the left of this midpoint, depending on which of thetwo candidates adopts an undemocratic position,
x1s ; x2s ¼
x1k þ x2k2
6d
2ak x1k � x2kð Þ : (3)
When candidate 1 adopts an undemocratic position(M15 1,M25 0), the swing voter x1s is located to the rightof themidpoint x1kþx2k
2 . Voters to the right of themidpointbut to the left of x1s favor candidate 1 basedon their policypreferences, yet are sufficiently put off by his un-democraticposition that theyvote forcandidate2 instead.By contrast, voters whose policy preferences are extreme(largexik)or intense (largeak) enoughtobe to the rightofx1s are willing to tolerate candidate 1’s undemocraticposition as their concern for democracy is outweighed bytheir proximity to his policies. The converse holds whencandidate 2 adopts an undemocratic platform.
The segment between the two swing voters is relatedto another set of empirical predictions. It delineates theset of voterswhoalways vote for themoredemocratic ofthe two candidates. We may therefore refer to this
portion of the electorate as “democracy first” voters, asopposed to“policyfirst”voters.The fractionof voters ineach category depends on both the length of this seg-ment and the distribution of voters’ ideal points. Itslength d
ak x1k�x2kð Þ is increasing in the support for de-mocracyd; it is decreasing in thepolicyweightakand thedistance between the candidates’ platforms (x1k2 x2k).
Meanwhile, the more polarized (U-shaped) the dis-tributionofvoters’ idealpolicies, thesmaller thefractionofdemocracy first voters. The consequences of voter po-larization are amplifiedwhen voters’ preferences over theK policy issues correlate: when a voter’s extreme positionon one issue correlates with her extreme position onanother, herpreference for themoreproximate candidatewillcompoundacross thetwoissuesratherthancancelout.
Theprimarypurposeof this framework is to guide thedesign of our candidate-choice experiment and theanalysis of our data. This has shaped our theoreticalanalysis in twoways. First, wehave intentionally treatedthe candidates’ policy platforms and democracy posi-tions asexogenous, as thesewill be randomlyassigned inthe candidate-choice experiment that we introduce inthe next section.6
Second, to explicitly characterize the model’s impli-cations for our analysis of the candidate-choice ex-periment, we introduce a stochastic structure into thedeterministic formulationof thevoters’payoff in (1)andoutline its consequences. We follow the classic randomutility model of discrete choice and add to voter i’spayoff from candidate j an error term «ij,
ui Xj;Mj� � ¼ ��
K
k¼1ak xik � xjk
� �2 � dMj þ «ij: (4)
Assuming that the error terms «ij are (independently)drawn from type I extreme value distribution impliesthat the effect of candidate positions on voter i’sprobability of voting for a candidate can be estimatedusing the logistic regression.7We take advantage of thiscorrespondence between our theoretical frameworkand the randomutilitymodel of discrete choicewhenweestimate the model’s key parameters, including civicvirtue d and policy weights ak.
Here, we highlight the aggregate empirical patternsimplied by this stochastic formulation. First, in contrastto the deterministic case, the two swing voters no longersharply separate “democracy first” from “policy first”voters. Rather, the segment between the two swingvoters now delineates voters who, regardless of theirpolicy proximity to the candidates, vote for the moredemocratic candidate with a probability greater thanone-half. The left panel in Figure 1 illustrates this byplotting a voter’s probability of voting for candidate 1 asa function of ideal policies xik over a single issue k.
8 Thesolid line plots the case when both candidates adopt
6 For a model with endogenous policy and democracy positions, seeSvolik (2020).7 This is only one of several plausible stochastic structures; seeCameron and Trivedi (2005, 476–8, 486–7).8 Figure1 is basedonparameter valuesak58, d52, andx1k52x2k51/5.
Democracy in America?
395
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
a democratic position; the dashed line plots the casewhen candidate 1 adopts an undemocratic position.
An alternative interpretation of the vertical axis in theleft panel of Figure 1 that we employ when analyzing thecandidate-choice experiment is that it corresponds tothe fraction of voters in subgroups along the horizontalaxis that support candidate 1. Furthermore, it will often bemorepractical touseas thehorizontalaxis thedifference invoters’ policy proximity to the two candidates rather thanvoters’ ideal points alone. In that case, the intersection ofthedashedcurvewith the0.5horizontal linewill, insteadofthe swing voter x1s , mark the quantity d
ak, which we later
interpretasvoters’value fordemocracy in termsof issuek.A key quantity that we estimate throughout the ar-
ticle is the overall fraction of voters who defect fromacandidate that adoptsanundemocraticposition,whichwedenote byD. In effect,Dmeasures the public’s abilityto serve as a democratic check. It results from a com-bination of two factors: the rate at which voters defectfromtheundemocratic candidate and thedistributionofvoters’ idealpoints.9The right panel inFigure1plots theformer. It shows that while centrists will in generaldefect from undemocratic candidates at higher ratesthan extremists, the maximum defection rate (markedby x1D) will be located to the right of the midpoint x1kþx2k
2but to the left of the swingvoterx1s . Intuitively, it is voterswho would otherwise narrowly break in favor of can-didate 1 that abandon him at the highest rate once headopts an undemocratic position.
To recapitulate, this theoretical frameworkyieldsfivepredictions about the relationship between a number ofdistinct conceptions of polarization and a decline in thepublic’s ability to serve as a democratic check, whichwedenote by D↓:10
1. Centrists are a pro-democratic force: D↓ for voterswhose policy preferences are more extreme relative tothe candidates’ platforms (xik far from the midpointx1kþx2k
2 );2. Moderation is good for democracy: D↓ for voters who
hold more intense policy preferences (a large ak);3. Voter polarization is bad for democracy: D↓ as the
distribution of voters’ ideal policies becomes more U-shaped (polarized) as opposed to inverse-U-shaped(centrist);
4. Candidate polarization is bad for democracy: D↓ as thedistance between candidate platforms increases (large|x1k 2 x2k|);
5. Cross-cutting cleavages are good for democracy: D↓when voter preferences over the K distinct issues arealigned as opposed to cross-cutting with respect tocandidates platforms.
THE CANDIDATE-CHOICE EXPERIMENT
Our candidate-choice experiment investigates a keymechanismunderlying theabovepredictions:evenvoterswho value democratic principles may trade off thoseprinciples for partisan ends when confronted witha choice between the two. By asking respondents tochoose between two candidates in an election, the ex-periment examines this mechanism at the same level atwhich it ishypothesized tooperate—thatof the individualvoter. Moreover, because it entails one of the most es-sential and familiar actions that democratic citizensperform—voting—the experiment allows us to study ourrespondents’ willingness to trade off democratic princi-ples for partisan interests without alerting them to it.
In the candidate-choice experiment, respondentsmade a series of 16 choices, eachbetween two candidatesfor a state legislature. The candidates were described byexperimentally manipulated attributes typically seen inreal-world elections: age, gender, race, profession, yearsof experience, partisan affiliation, two policy platforms,and a “democracy” position. This last attribute is the
FIGURE 1. Model Predictions: The Fraction Voting for Candidate 1 (Left) and the Decline in the lessDemocratic Candidate’s Vote Share (Right)
9 That is, D ¼ R ‘�‘
11þe�D � 1
1þe� D�dM1ð Þ� �
dF xið Þ, where F(xi) is the distribu-
tion of voters andD ¼ ��K
k¼1ak xik � x2kð Þ2 ��
K
k¼1ak xik � x1kð Þ2.
10 See the appendix for a formal statement and proofs of thesepredictions.
Matthew H. Graham and Milan W. Svolik
396
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
focus of our analysis; we thereforedescribe its designandassignment below. We introduce most of the remainingattributes throughout thearticle.Theappendixdescribesour design in detail and presents an example of a can-didate-choice scenario as seen by our respondents.
Each candidate was assigned a democracy positionthat was either “undemocratic”—an action or statementby the candidate that violates a key democratic princi-ple—orademocraticallyneutral,“generic”position.Theundemocratic positions are listed in Table 1 and labeledD2. There, [own party] refers to a candidate’s randomlyassigned political party (Democrat or Republican);[opposite party] denotes the complement. For instance,one possible realization of item 4 reads “Said the Re-publican governor should ignore unfavorable courtrulings by Democrat-appointed judges.”
In designing these undemocratic positions, weemployed the following criteria:
Conceptual validity: The undemocratic positionscapture violations of key democratic principles. Follow-ing classic scholarship on democratization (Dahl 1971),this includes measures that undermine electoral fairness(items1a,1b,and2inTable1), checksandbalances(items3 and 4), and civil liberties (items 5, 6a, and 6b).
We also verify that our respondents indeed see thepositions in Table 1 as undemocratic. One to three weeksbefore the candidate-choice experiment, each respondentwas asked to evaluate a battery of both democratic andundemocraticpractices, including thepractices thatwouldlater appear as ourD2 treatments.Aswedocument in theappendix, the vast majority of our respondents correctlyrated the D2 treatments as “undemocratic.”
Contextual realism and partisan balance: The un-democratic positions approximate practices that havebeen used by politicians to subvert the democratic pro-cess in theUnitedStates and can be plausibly adoptedbyboth major parties. Accordingly, the undemocraticpositions are situated at the state level, where mostattempts to subvert the democratic process for partisangain in theUnitedStatesoccur andhavehistoricallybeenattempted by bothmajor parties. The appendix providesreal-world examples of each undemocratic position.
Incremental violations: A key feature of attempts tosubvert thedemocratic process, both in theUnitedStatesand around the world, is the use of ostensibly legal, in-cremental, and complementary measures (Levitsky andZiblatt 2018; Waldner and Lust 2018). This has severalconsequences. First, such measures must often beimplemented by or in conjunction with the executive.This is why some of our undemocratic positions refer toactions that the candidate suggests a co-partisan gover-nor take. Second, because such measures are typicallyadopted through a constitutionally mandated process,they may undermine democratic principles without vi-olating the law.Thisapplies to items1a,1b, and3.Finally,any single measure may allow for a partisan in-terpretation according to which it is consistent withsome—often more majoritarian—conception of de-mocracy or corrects an existing deficiency in the dem-ocratic process. For instance, proponents of stricter voterID lawsmay respond to accusations of voter suppressionby claiming that such measures are needed to preventvoter fraud, and proponents of gerrymandering mayclaim to be correcting an unfair status quo. Jointly and intheir political context, however, such measures result inan uneven playing field that favors their proponents(Levitsky and Way 2010; Schedler 2002).
Neutral presentation: The undemocratic positionswere presented in amanner that avoids conspicuousnessor normatively leading language. To prevent candidatesnot assigned to hold an undemocratic position fromappearingvisually conspicuous, eachwasassignedoneofseven democratically neutral, “generic” positions. Forinstance, one read: “Served on a committee that estab-lishes the state legislature’s schedule for each session”;we list the remaining six in theappendix.Thecontent andlengthof suchgeneric positionsweredesigned tobalancethe cognitive effort required to distinguish a candidatewho endorsed an undemocratic position from one thatdid not.11 The wording of our undemocratic positions
TABLE 1. Undemocratic Positions Endorsed by Candidates Assigned to the D2 Treatment Condition
D2 Undemocratic position Democratic principle
1a Supported a redistricting plan that gives [own party]s 2 extra seats despitea decline in the polls
Electoral fairness
1b Supported a redistricting plan that gives [own party]s 10 extra seats despitea decline in the polls
Electoral fairness
2 Supportedaproposal to reduce thenumberofpolling stations inareas that support[opposite party]s
Electoral fairness
3 Said the [own party] governor should rule by executive order if [opposite party]legislators don't cooperate
Checks and balances
4 Said the [ownparty] governor should ignore unfavorable court rulings by [oppositeparty]-appointed judges
Checks and balances
5 Said the [own party] governor should prosecute journalists who accuse him ofmisconduct without revealing sources
Civil liberties
6a Said the [own party] governor should ban far-left group rallies in the state capital Civil liberties6b Said the [own party] governor should ban far-right group rallies in the state capital Civil liberties
11 For the same reason,we randomized the order inwhich candidates’democracy and policy positions were listed.
Democracy in America?
397
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
alsoavoidsnegative connotationsornormatively leadinglanguage. For instance, positions 1a, 1b, and 2 describegerrymandering and voter suppression, respectively, butwe intentionally avoided employing those terms. Putsimply, we want respondents to decide for themselveswhether or not a position violates a democratic principle.
In the experiment, each respondent made 16 distinctcandidate choices, of which 11 were based on the fol-lowing experimental design.12 In four randomly chosenscenarios, both candidates adopted one of the demo-cratically neutral, “generic” positions. Throughout, wetreat these as our control scenarios and label themD1 vs.D1. In seven randomly chosen scenarios, one of thecandidates adopted an undemocratic position, whereasthe other held a neutral position.We refer to these as ourtreatment scenarios and label themD2 vs.D1. Whetherthe undemocratic position was held by the candidatevisually presented on the left or right was random. Tosimplify the presentation of our findings, we reshape ourdata so that candidate 2 always holds a neutral position(D1) and, depending on the experimental condition,candidate 1 varies between D1 and D2.
The candidate-choice experiment was embedded ina nationally representative survey of American votersthat took place inAugust–September 2018.13 The 1,691respondents made a total of 21,604 candidate choices.
Democratic Principles vs. Policy Preferences
We begin our analysis of the candidate-choice experi-ment by examining Americans’ willingness to trade offdemocratic principles for their preferred policies. Eachcandidate proposed aplatform inoneeconomic andonesocial policy area. Economic policies concerned eitherstate income taxes or state funding for local education;social policies concerned either immigration or mar-ijuana’s legal status. These policy areas were randomlyassigned but identical across the two candidates.Withineach policy area, candidates could adopt one of fourplatforms, ranging from extreme liberal to extremeconservative. The appendix lists all 16 policy platformsand justifies their selection.
One to three weeks before being presented with thecandidate-choice scenarios, respondents rated each ofthe 16 policy platforms on a 0–100 proximity scale. Thisallows us to identify each respondent’s ideal policyacross the four areas and, following our theoreticalframework, to compute the squared distance betweena respondent’s ideal policy and the two candidates’platforms for each of the candidate-choice scenarios
that the respondents faced.14 The results that follow arerobust to a range of alternative measures of policyproximity between respondents and candidates andaccount for the possibility that voters’ policy prefer-encesmaybe ideologically incoherent (Converse 1964),multidimensional (Treier and Hillygus 2009), or non-separable from partisanship. We present these alter-native measures in the appendix.
The left panel in Figure 2 plots the fraction ofrespondents voting for candidate 1 as a function of thedifference in policy proximity to the respondent be-tween candidate 1 and candidate 2.15 On the horizontalaxis, a value of 0 refers to scenarios when the twocandidates are equally proximate to the respondent anda value of 1 (21) to scenarios when candidate 1 is a fullscale closer to (further away from) the respondent thancandidate 2onbothpolicy areas.We treat theD1vs.D1
scenario (solid line), when both candidates adoptneutral democracy positions but differ across otherattributes, as our control condition; we treat theD2 vs.D1 scenario (dashed line) as our treatment condition.Vertical bars denote 95%confidence intervals.16 Figure2 is thus the policy analog of Figure 1.
The D1 vs. D1 control scenario provides an initialplausibility check of our design. Consistent with ourspatial framework, the closer candidate 1 is toa respondent’s ideal policies relative to candidate 2, themore likely the respondent votes for candidate 1. Spe-cifically, the fractionof respondentsvoting for candidate1increases from 13.0%when candidate 1 is a full scale lessproximate to the respondent than candidate2 to 87.0% inthe opposite case. Furthermore, when the two candidatesare equally proximate to respondents, the vote splitsabout evenly. Put simply, a respondent’s proximity toeach candidate’s policy platform is a strong predictor ofher candidate choices.
Figure 2 also provides an initial estimate of whetherandhowmuchAmericans valuedemocracy.Becauseallcandidate attributes were randomly assigned, the onlysystematic difference between our control and treat-ment conditions is candidate 1’s democracy position.We can therefore interpret a decline in the fraction of
12 The remaining five scenarios were designed to provide extensionsand robustness checks of our core design. We discuss them in theappendix.13 The first wave, which asked questions about partisanship, policypreferences, and support for democracy, took place onAugust 28–29,2018.Theprimary focusof the secondwave,which tookplacebetweenSeptember 4 and 25, 2018, was the candidate-choice experiment.Lucid fielded the survey. A pilot survey was fielded on AmazonMechanical Turk in March 2018. The appendix benchmarks oursample against demographic data from the US Census Bureau andpartisan and attitudinal questions from the 2016 American NationalElection Study.
14 That is, the 0–100 proximity scale is ameasure of xik2 xjk. For eachcandidate-choice scenario, this allows us to compute the squared
distance�2
k¼1ak xik � xjk
� �2 between each respondent’s ideal policies
and both candidates’ platforms, and in turn, candidate 1’s policy
proximity advantage D ¼�2
k¼1ak xik � x2kð Þ2��
2
k¼1ak xik � x1kð Þ2.
15“Voting” here refers to the respondents’ stated preference for one
of the two candidates.Wedistinguishbetween respondents’ candidatepreferences and their stated intent to turn out to vote in the appendixand in Graham and Svolik (2019). We find that the strongera respondent’s preference for aD2 candidate, themore likely she is toabstain.16 Because each respondent made multiple choices, estimates thattreat all observations as independent may understate statistical un-certainty.Unless otherwise noted, we compute all standard errors andconfidence intervals using the block bootstrap, which accounts fordependence by resampling observations at the level of the respondent(see, e.g., Bertrand, Duflo, and Mullainathan 2004).
Matthew H. Graham and Milan W. Svolik
398
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
voters who support candidate 1 in the D2 vs. D1 asopposed to theD1 vs.D1 condition as a measure of thepublic’s ability to serve as a democratic check. Consid-ering all candidate-choice scenarios in our experiment,the shift from theD1 vs.D1 to theD2 vs.D1 conditiondecreases candidate 1’s vote share from 50.0% to38.3%—an 11.7% decline (CI: 10.0, 13.5). Put differ-ently, just over 10% of Americans value democracyenough to punish an otherwise favored candidate forviolating a democratic principle by voting against him.
Are Americans willing to trade off democratic princi-ples in exchange for more appealing policies? Figure 2allows us to address this question by partitioning ourexperimental electorate into two politically consequentialsubsets of voters anticipated by our theory: “democracyfirst” and “policy first” voters. A majority of the formervote for themore democratic candidate even when doingsogoesagainst theirpolicy interests.Theserespondents liein the interval at the center of Figure 2 between the in-tersectionof theD2vs.D1 linewith the0.5horizontal axisand its mirror image along that axis. This interval corre-sponds to the values (20.3, 0.3) on thehorizontal axis, andits limits are the empirical counterpart of the two swingvoters inFigure1.Bycontrast,voters tothe leftandrightofthis interval are “policy first” voters: a majority of thesesupports the more policy-wise proximate candidate, evenif that candidate adopts an undemocratic position.
We gain additional insights into the robustness of sup-port for democracy by examining differences in the se-verity with which respondents punish candidate 1 foradoptinganundemocraticplatform.Themagnitudeof thispunishment is a combination of two factors: the baselinelevel of support for candidate 1 in each of the policyproximity subgroups and the rate at which respondents ina subgroup defect from candidate 1 after he adopts anundemocratic position.17 The right panel in Figure 2 plots
the defection rate. Consistent with our theory, we see thatthe defection rate is highest among “bare suppor-ters”—respondents who narrowly break in favor of can-didate 1 in the D1 vs. D1 scenario—and declines as wemove toward “policy extremists” on either side.
These policy-based differences in respondents’ will-ingness to punish undemocratic behavior are consistentwith our arguments about the pernicious consequencesof polarization for democracy. Our representativesample allows us to simulate counterfactual electorateswith increasing levels of policy polarization by varyingthe ratio of “policy centrists” to “policy extremists.”Assuggested by Figure 2, an electorate consisting entirelyof “policy centrists”—the middle two bins—would re-sult in a resounding defeat of a candidate who wouldadopt an undemocratic platform (12.9%,CI: 11.1, 15.1).In an electorate consisting entirely of the two mostextreme subgroups, undemocratic candidates wouldonly lose 3.0 percentage points of the vote share (CI:24.1, 11.1), about one-fourth of the penalty amongcentrists.18
Does Partisanship Trump Civic Virtue?
Are voters willing to put democratic principles ahead oftheir partisan loyalties? To address this question, con-sider first only contests between candidates from dif-ferent parties. Denote by Rep D1 vs. Dem D1 the
FIGURE 2. Fraction Voting for Candidate 1 (Left) and Fraction Defecting from the Less DemocraticCandidate (Right) by the Difference between Candidates’ in Policy Proximity to the Respondent
17 That is, the defection rate is r5%1(D1 vs.D1)2%1(D
2 vs.D1)where%1(T) refers to the fractionof respondents voting for candidate1 in treatment condition T.
18 Because the candidates’ two policies were independently andrandomly assigned, respondents of all types—those who hold mod-erate as well as extreme policy preferences—faced some candidate-choice scenarios in which they were in the position of “centrists” andsome in which they were “extremists.” This results in an overalldistribution of differences in policy proximity in Figure 2 that is heavyat the center: about 55%of scenarios are locatedon the interval (20.1,0.1) and about 78% on the interval (20.3, 0.3). To probe the ro-bustness of our claims, the appendix displays the distribution of policyproximity separated by respondents’ strength of partisanship andreplicates Figure 2 using a rank-based measure of respondents’proximity to candidates that, by construction, relies only on within-respondent differences in policy preferences.
Democracy in America?
399
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
control condition when both candidates adopt a neutraldemocracy position and by Rep D1 vs. Dem D2 andRepD2 vs. DemD1 the treatment conditions in whichthe Democrat or Republican, respectively, adopts anundemocratic position. Overall, undemocratic candi-dates are penalized by a loss of 9.9% (CI: 6.7, 13.0) and10.9% (CI: 7.6, 14.0) of voters in the Rep D1 vs. DemD2 and Rep D2 vs. Dem D1 scenarios, respectively.19
Both effects are statistically different from zero but notstatistically different from each other (difference: 1.0%,CI:25.3, 3.3). Voters punish undemocratic behavior byboth parties and they do so fairly evenly.
Are Americans willing to vote across party lines topunish a candidate for adopting an undemocratic po-sition? 63.1% of our respondents support their ownparty in the control condition; this number declines to54.8% when a respondent’s co-partisan adopts an un-democratic position (28.3%, CI: 5.0, 11.6). Put differ-ently, only 13.1% of our respondents are willing topunish a co-partisan for violating democratic principlesby voting against their own party. A majority ofAmericans are willing to give undemocraticco-partisans a pass.20
Figure 3 yields further insight into how ourrespondents’ willingness to punish undemocratic can-didates varies by the strength of their partisanship. Itplots the fraction of respondents voting for a Re-publican Party candidate as a function of ourrespondents’ party identification on the conventional 7-point scale,with the two treatment conditionsplottedbydashed and dotted lines. As expected, stronger parti-sans vote for their party in greater proportion in thecontrol condition, with independents breaking aboutevenly for the two parties. Crucially, only independentswho lean toward a party defect from an undemocraticco-partisan in large enough numbers to defeat thatcandidate. By contrast, respondents who identify asa Democrat or Republican only break even. And a ma-jority of strongpartisanswould rather elect a candidatewho violates democratic principles than cross partylines.
Strong partisans’ failure to punish undemocraticcandidates fromtheirownparty is theresultof twoforces.First, strong partisans need to defect from undemocraticcandidates at higher rates thanmoderates to compensatefor theirhighbaseline support for co-partisancandidates.Second, they do exactly the opposite: compared withother partisan subgroups, strong partisans are more le-nientonviolationsofdemocraticprinciplesbycandidatesfrom their party. Figure 4 shows this by plotting thefraction of respondents that defect from the D2 candi-date, conditioning on both the respondent’s and theD2
candidate’spartisanship.Wesee that strongpartisansare
less thanhalf as likely topunishundemocratic candidatesfrom their ownpartyasare respondentswho lean towardthe opposite party. Put differently, strong partisans arewilling to punish their own for violating democraticprinciples, but they hold candidates from the other partyto a much higher standard.
This raises the question of whether contests betweencandidates fromthesameparty—that is, primaries ratherthan general elections—are the more viable check oncandidates who undermine democratic principles.Consistent with our theory, our respondents are morewilling to punish undemocratic candidates in same partycontests than in cross-party contests: a candidate whoadopts an undemocratic position is penalized by a loss of13.3% and 10.4% of voters, respectively (difference:2.9%, CI: 0.0%, 6.1%). Figure 5, however, raises doubtsabout the promise of primaries as a democratic check. Itplots the fractionof respondents voting for candidate 1 incontests between either two Democrats or two Repub-licans.Whenboth candidates adopt a neutral democracyplatform (either DemD1 vs. DemD1 or RepD1 vs. RepD1), that fraction is (by construction) 0.5. Now consider
FIGURE 3. Different Party Contests: FractionVoting for a Republican Candidate
FIGURE 4. Different Party Contests: FractionDefecting from the Less Democratic Candidateby the Respondent’s Partisanship
19 Both penalties imply that aD2 candidate is opposed by a majorityrespondents, regardless of that candidate’s party.20 That is, among respondents who support their own party in thecontrol condition, the treatment effect was 13.1% of the baseline,(63.052 54.76)/63.05. Among respondents who crossed party lines inthe control condition, defection is higher: an undemocratic out-par-tisan’s vote share declines from 37.0% to 24.4% (212.6%, CI: 9.4,15.9), which is 34.1% of the baseline, (36.95 2 24.36)/36.95.
Matthew H. Graham and Milan W. Svolik
400
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
the punishment for a Democrat who adopts an un-democratic position (Dem D2 vs. Dem D1 in a dashedline): its severity is increasingaswemoverightwardalongthe horizontal axis, away from partisan supporters andtoward partisan opponents. The reverse holds when aRepublican candidate adopts an undemocratic position(Rep D2 vs. Rep D1 in a dotted line). The promise ofprimaries as a democratic check is thus undermined bya partisan double standard: inmost states, voters are notallowed toparticipate in theprimary inwhich theywouldpunish undemocratic candidates most severely.
In sum, both Democrats and Republicans employapartisandouble standardand theydo soevenwhen thepartisanship of thewinning candidate is preordained.Apartisan double standard amplifies the consequences ofpolarization for the public’s resilience to undemocraticcandidates: an electorate consisting of strong partisansalonewouldprovideonlya tenuous checkoncandidateswho violate democratic principles.
The Consequences of Candidate Polarization
Adistinct advantageofourresearchdesign is that itallowsus to explore the consequences of a number of distinctconceptions of polarization. In our empirical analysis sofar, we have examined polarization as an individual orelectorate-level phenomenon. We now shift to the can-didate level, at which an increase in the distance betweencandidate platforms can be interpreted as an increase ina conceptually and empirically distinct kind of polar-ization—candidate polarization. Because candidates’policy platforms and partisanship were independentlyassigned in our experiment, we can examine how candi-date polarization affects voters’ ability to serve asa democratic check. Our theoretical model implies thatgreater candidate polarization results in a greater share ofvoters who are willing to tolerate undemocratic behavior.Crucially, these consequences of candidate polarizationarise independently of voter polarization—they obtaineven if we hold voter polarization fixed.
We now assess this prediction by examining theconsequences of polarization in candidates’ policyplatforms. InFigure6, thehorizontal axis plots themeanabsolute distance between the candidates’ two policyplatforms; the vertical axis plots the fraction ofrespondentsdefecting fromaD2 candidate.Wesee that15.9%of respondents defect from aD2 candidatewhenthe two candidates take the same policy or have policydifferences that cancel out (CI: 11.9%, 20.1%); thedefection rate declines substantially when both policiesare as far apart as possible.21 Consistent with our the-oretical framework, voters become more reluctant topunish candidates who violate democratic principles ascandidates’ policy platforms move apart.22
Resisting the Menu of Manipulation
Whenexaminingour respondents’willingness topunishcandidates that violate democratic principles, we haveso far pooled all democracy positions into two groups,neutral and undemocratic. We now examine Ameri-cans’ willingness to tolerate the distinct undemocraticpositions inTable 1 and interpret them in light of severalbenchmarks.
We estimate the following linear model:
Pr i votes for candidate 1ð Þ ¼ aþ�k
bk Xi1k �Xi2kð Þ þ «i:
(5)
In(5),Xi1kandXi2karedummyvariables forallpossiblevaluesofexperimentallymanipulatedattributek inchoicei between candidates 1 and 2, respectively. Figure 7 plotsthe estimates of bk. Bars represent the associated 95%confidence intervals; estimates without confidenceintervals correspond tobaseline categories. In the spirit ofHainmueller, Hopkins, and Yamamoto (2015), we in-terpret these coefficients as attribute k’s average causal
FIGURE 5. Same Party Contests: FractionVoting for Candidate 1 by the Respondent’s andthe D2 Candidate’s Partisanship
FIGURE 6. The Effect of the Mean AbsoluteDistance between the Candidates’ Two PolicyPlatforms on the Fraction of RespondentsDefecting from a D2 Candidate
21 See the appendix for a regression-based test of this relationship.22 Our earlier analysis of contests between candidates from the sameversus different parties in effect found support for our predictionabout the consequences of candidate polarization in terms of parti-sanship: we saw that respondents were less willing to punish un-democratic candidates in different-party than same-party contests.
Democracy in America?
401
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
effect on a candidate’s vote share, relative to the baselineattribute and averaging over the distribution of all otherattributes.
Consider first the coefficients associated with thedemocratically neutral, “generic” positions. All sevenare individually (and jointly) statistically indistinguish-able from0, implying that theydonot affect a candidate’svote share. This validates our design and interpretationof these attributes as not only democratically neutral butalso more generically inconsequential.
Figure 7 also demonstrates considerable variation inthe effect of the individual undemocratic positions ona candidate’s vote share. While all undemocratic posi-tions impact a candidate’s vote share negatively, themagnitude of that effect ranges from 10.2% to 16.1%.23
Respondents most severely punish candidates whowant to prosecute journalists (16.1%) and ignore courtrulings (14.1%). Respondents are least sensitive to
candidates who endorse gerrymandering (by 2 seat,10.6%) and suggest that the governor ban protests orrule by executive order (10.2 and 10.5%, respectively).Figure 8 differentiates these estimates by respondents’partisanship.Consistentwithourearlierfindings,we seefew differences between supporters of the two majorparties.24
To put the magnitude of these estimates in context,compare the effect of these undemocratic positionswiththat of other positional and valence candidate attrib-utes. Consistent with our earlier analysis, the two mainpositional attributes—a candidate’s party and policyplatforms—have an impact on a respondent’s candidatechoice that is either comparable or greater inmagnitudethan individual undemocratic positions. The effects ofthe remaining attributes are an order of magnitudesmaller than that of any undemocratic position.
FIGURE7. EffectsofCandidateAttributesandDemocracyPositionsonaCandidate’sVoteShare.DotsRepresentCoefficient EstimatesBasedonaLinearRegressionModel; BarsRepresent 95%ConfidenceIntervals
23 An F-test rejects the joint hypothesis that all coefficients associatedwith D2 positions in Figures 7 and 8 are equal at a p-value , 0.001.That is, there are statistically significant differences in how severelyvoters punish the distinct D2 positions.
24 Figure 8 plots marginal vote shares (note the different scale of thehorizontal axis than in Figure 7), ensuring that partisan differences inpreferences toward the baseline attribute do not distort the com-parisons (Leeper, Hobolt, and Tilley forthcoming).
Matthew H. Graham and Milan W. Svolik
402
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
To help us further interpret the magnitude of un-democratic positions, two of each respondent’s 16choices included two negative valence attributes in-tentionally unrelated to democracy. According to thefirst, the candidate “was convicted of underpaying fed-eral income taxes”; according to the second, the candi-date “was reported to have had multiple extramaritalaffairs.” Estimates associated with these two attributesappear at the bottom of Figures 7 and 8 and are labeledV2.Weseethatvoterspunishcandidatesforextramaritalaffairs and underpaying taxes more severely than theypunish them for undermining democratic principles.
STRUCTURAL ESTIMATES OF SUPPORTFOR DEMOCRACY
Throughout our analysis, we have systematically founda significant, if modest decline in the vote share of can-didates whose positions violate democratic principles.Thisfinding is consistentwithour theoretical framework,which conceives of such positions as a negative valencecharacteristicand suggests that the civic virtueparameterd in our model is positive.25 We now shift from thenonparametric approach employed above to a structuralapproach that directly estimates our model’s funda-mentals: civicvirtuedand theweightsak thatvotersplaceon other candidate attributes. This approach allows us tointerpret d as the relative weight that voters place ondemocracy in their voting decisions, to express voters’value fordemocracy in termsofotherdesirablecandidateattributes that voters are willing to forgo to punishcandidates who violate democratic principles, and toderive other theoretically informed quantities that yieldinsights about the robustnessof support fordemocracy inthe United States. In effect, the variation in our
respondents’ candidate choices that we have seen so farcan be parsimoniously summarized by a few funda-mental, theoretically grounded parameters.
The estimation of these parameters is aided by a closecorrespondence between our theoretical model, the de-sign of our candidate-choice experiment, and the randomutilitymodelofdiscretechoice.Recall that theassumptionof type I extreme value distributed error terms «ij in (4)implies that the effect of candidate attributes on voter i’sprobability of voting for candidate 1 can be estimatedusing logistic regression.26 For instance, our earlier dis-cussion of voters’ willingness to trade off democraticprinciples for their preferred policies and party suggeststhe estimation equation
Pr i votes for candidate 1ð Þ¼ logit�1 b0 þ b1M þ b2econþ b3social þ b4party½ �;
(6)
whereM, econ, social, andpartyaredifferencesbetweencandidate 1 and 2’s democracy positions, economic andsocial policy platforms, and partisan affiliation. Esti-mates from (6) allow us to express the value that votersplace on democracy vs. other candidate attributes k asnormalized weights d and ak,
d ¼ b1j j
b1j j þ�L
l¼2blj j
and ak ¼ bkþ1
�� ��
b1j j þ�L
l¼2blj j
;
where l50, 1,…,L indexes the regression coefficients in(6).
Table 2 presents estimates of the weights d and akbased on six alternative specifications. Column (1) esti-mates the logit model in (6) after the four input variableshavebeendividedby twice their standarddeviation.Thisstandardization puts candidates’ democracy positions,economic and social policy platforms, and partisan af-filiation on a common, dispersion-based scale, allowingus to compare their importance (Gelman 2008). We seethat the weight that voters place on democracy in theirvotingdecisions isabout18.1%comparedwith32.8%forsocial policy, 27.1% for economic policy, and 22.0% forpartisanship.
Column (2) in Table 2 extends the model in (6) toinclude all experimentally manipulated candidateattributes.27 The presented weights for candidates’ de-mocracy positions, economic and social policy platforms,and party imply that these four factors jointly account for79.6% of the systematic variation in voters’ candidatechoices. These results are consistent with our earlier,nonparametric analysis:Americans valuedemocracy—dis significantly greater than zero—but they value it lessthanmajor competing political considerations, especiallyeconomic and social policies and partisanship.
Estimates of the weights d and ak allow us to sum-marizeanelectorate’s support fordemocracy in termsofvoters’ value for democracy. This concept expresses the
FIGURE 8. Effects of Neutral, Valence, andUndemocratic Positions on a Candidate’s VoteShare by Respondents’ Partisanship. DotsRepresent the Mean Vote Share amongCandidates Adopting Each UndemocraticPosition; Bars Represent 95% ConfidenceIntervals
25 This interpretation is bolstered by the fact that the vote share ofD2
candidatesdeclinesacrossall respondent subgroups inFigures 2 and3.
26 See the appendix for evidence of goodness of fit.27 We present a complete set of results in the appendix.
Democracy in America?
403
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
trade-offs that voters are willing tomake in terms of theimplied marginal rate of substitution—the amount ofother desirable candidate attributes that voters areready to forgo to punish candidates who violate dem-ocratic principles. Specifically, the model in (6) impliesthat avoter’spayoff remainsunchangedas longasa shiftin a candidate’s democracy position M is accompaniedby a shift of magnitude d
akin some other attribute k. Put
simply, voters’ value for democracy summarizes therobustness of support for democracy parsimoniously, interms of the “price” voters are willing to pay to electa more democratic candidate.
Estimates of value for democracy are most easilyinterpreted when we reestimate the model in (6) withcandidate attributes entering in theirmost “natural”units.Wedosoby lettingMandpartyenterascandidate-specificdummies and expressing economic and social policies ona 0–1 proximity scale. Columns (3) and (4) display theresulting estimates of theweights d andak, paralleling thesimpleandextendedmodels in(1)and(2).Estimates in(3)yield voters’ value for democracy of 0.495 (CI: 0.424,0.570) in economic policy, 0.454 (CI: 0.486, 0.523) in socialpolicy, and 0.815 (CI: 0.687, 0.958) in co-partisanship.These quantities imply that to punish a candidate whoadopts an undemocratic position, the voter is willing totolerate a 0.5 and 0.45 increase in the distance from herideal economic or social policies, respectively—about halfthe scale. The value for democracy in terms of co-parti-sanship is smaller than one, implying that the typical voteris not willing to vote across party lines to punish a co-partisan for adopting an undemocratic position.
Columns (5) and (6) probe the robustness of theestimates in columns (1)–(4) by letting M enter as eachrespondent’s own pretreatment rating of “how un-democratic” the specific D2 treatment appearing ina candidate-choice scenario was. This approach allowsfor thepossibility thatonly somerespondentsdisapproveof the D2 positions or that they disapprove of the D2
positionsatdifferent rates.Our results remainessentiallyunchanged when we account for the respondents’ ownperception of the severity of each D2 position.
In our earlier, nonparametric analysis of partisanship,we found that our respondents punish platforms that
violate democratic principles less severely when adoptedby co-partisan candidates. The present approach allows usto derive and estimate this tendency in terms of a single,theoretically informed quantity: the partisan biasp.Specifically,weextendouroriginalmodelbymultiplyingcandidate j’s democracy platform Mj in (1) by the termd[1 2 pIj(party)] instead of d alone, where Ij(party) is anindicator of co-partisanship between the respondent andcandidate j. In turn,wemaysay that support fordemocracyis principledwhen p 5 0 and voters punish candidates forviolating democratic principles equally, regardless of theirparty; support for democracy is instrumental when p 5 1,and voters only punish undemocratic positions whenadopted by candidates from the other party.
This formulation implies that the degree of partisanbiasp in theelectorate canbeestimatedbyadding to thelogitmodel in (6) an interaction termbetweenparty andM. Denoting the coefficient on that interaction term bybint, we havep ¼ 1� bint
b1. The estimate ofp is 0.484 (CI:
0.358, 0.605), implying that Americans are neither fullyprincipled nor purely instrumental in their support fordemocracy. Rather, voters are about 50%more lenienttoward violations of democratic principles by candi-dates from their own party.
THE 2017MONTANANATURALEXPERIMENT
Depending on whether they voted on the election day orby absentee ballot, voters in the 2017 special election forMontana’s single U.S. House seat saw two differentraces.28 Both races pitted Republican Greg Gianforteagainst Democrat Rob Quist. Absentee voters, who inMontana make up about 70% of registered voters, sawa small-governmentRepublicanwith business credentialscompete against a former musician Democrat who sup-ported mainstream liberal positions. All three major
TABLE 2. Structural Estimates of Our Model of the Public as a Democratic Check
Standardized Natural Rating-based
(1) (2) (3) (4) (5) (6)
d democracy 0.181*** 0.144*** 0.155*** 0.126*** 0.211*** 0.169***(0.161, 0.201) (0.118, 0.164) (0.137, 0.173) (0.105, 0.143) (0.191, 0.229) (0.144, 0.188)
a1 economic policy 0.271*** 0.216*** 0.312*** 0.254*** 0.292*** 0.235***(0.254, 0.290) (0.183, 0.236) (0.291, 0.333) (0.218, 0.276) (0.273, 0.313) (0.202, 0.257)
a2 social policy 0.328*** 0.261*** 0.342*** 0.279*** 0.320*** 0.256***(0.310, 0.346) (0.222, 0.285) (0.323, 0.363) (0.239, 0.303) (0.301, 0.339) (0.222, 0.282)
a3 party 0.220*** 0.176*** 0.190*** 0.155*** 0.177*** 0.142***(0.202, 0.240) (0.149, 0.197) (0.173, 0.208) (0.133, 0.173) (0.161, 0.194) (0.123, 0.161)
N 18,195 18,195 18,195 18,195 17,371 17,371
Note: Bootstrapped 95% confidence intervals in parentheses.* p , 0.05; ** p , 0.01; *** p , 0.001.
28 The special election was held to fill the U.S. House seat vacated byRyan Zinke, who became the Secretary of the Interior in the Trumpadministration.
Matthew H. Graham and Milan W. Svolik
404
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
newspapers in Montana initially endorsed RepublicanGreg Gianforte.
Electiondayvoters saw the sameracewithonecrucialdifference: on the eve of the election, GianforteassaultedTheGuardian reporterBenJacobsafter beingrepeatedly questioned about his position on health carereform.29 The attack dominated the news coverage thatevening and led the threemajornewspapers inMontanato rescind their endorsements of Gianforte on themorning of the election day. Gianforte nonethelessdefeated Quist 50.2% to 44.1%, with 5.7% going toLibertarian Mark Wicks.
The timing of Gianforte’s assault offers a real-world,quasi-experimental opportunity to evaluate our theo-retical framework and corroborate our experimentalresults. We adopt a difference-in-differences empiricalstrategy that compares shifts in absentee and electionday voters’ support for the Republican U.S. Housecandidate between theNovember 2016 general electionand the May 2017 special election.30 Among absenteevoters, shifts from 2016 to 2017 act as a control thatreflects what would have happened if no voters ob-served the assault.31 Vote shifts among election dayvoters, by contrast, reflect the causal effect of the as-sault. Even though absentee voters may be differentfrom election day voters, such differences cancel out ina difference-in-differences framework where eachgroup serves as its own, pretreatment baseline.
We interpret the assault as a public signal aboutGianforte’s respect for the freepressor—atminimum—asa negative valence signal about his fitness for office. Ourtheorypredicts thatvoters’willingness topunishGianfortefor the attack will be decreasing in the strength of theelectorate’s partisanship. In the context of Montana’spartisan makeup, this implies that the precinct-level de-cline in Gianforte’s vote share should be largest in mod-erate precincts and decreasing as Republican strengthacross precincts grows. This obtains because voters’ will-ingness to tolerate a co-partisan who violates democraticprinciples increases in the intensity of their partisanship.32
To investigate thesepredictions,wecontactedelectionadministrators in all of Montana’s 56 counties andidentified five counties that tallied absentee and electiondayballots separately in 2016and2017. Figure9presentsdata based on the 87 precincts in these five counties.Separately for each voting method, it plots the precinct-level differences in Republican vote shifts betweenNovember 2016 and May 2017 as a function of the 2016Republican two-party vote share in the entire precinct.Given the absence of extreme Democratic precincts inour sample, we treat the latter as an indicator for theintensity of a precinct’s partisanship.33 Absentee voteshifts are shown as circles and election day vote shifts asdiamonds; positive differences in vote shifts are high-lighted by upward-facing arrows and negative differ-ences by downward-facing arrows. Consistent with ourexpectations, differences in vote shifts are negative andlargest in moderate precincts; they decrease in magni-tude and some even become positive as we move rightalong the horizontal axis, suggesting that punishment forthe assault was smaller in more Republican precincts.34
To investigate this pattern formally, we estimate thefollowing linear models:
Rit ¼ aþ b1Y17it þ b2Eit þ b3Y17itEit þ gXit þ «it; (7)
Rit ¼ aþ b1Y17it þ b2Eit þ b3Y17itEit þ b4R16
þ b5Y17itRi16 þ b6EitRi16 þ b7Y17itEitRi16
þ gXit þ «it: (8)
Above,Rit is theRepublican candidate’s vote share inprecinct i in year t,Y17it is a dummy for the year 2017 (asopposed to 2016), Eit is a dummy for voting on elec-tionday (asopposed toabsentee),Ri16 is theRepublicanvote share for the entire precinct i in 2016, and Xit isa vector of control variables. The latter includes the
FIGURE 9. Differences in Precinct-Level VoteShifts for the Republican U.S. House Candidatein Montana between November 2016 and May2017
29 See, for example, Martin, Jonathan. “Montana Republican GregGianforte, Charged With Assault, Awaits Fate in Vote.” The NewYork Times, May 24, 2017.30 On difference-in-differences estimation, see Bertrand, Duflo, andMullainathan (2004).31 According to the Montana voter file, election officials had alreadyreceived 92.5% of all absentee ballots by the time of Gianforte’sassault. The correspondingfigure for our sample of precincts is 94.1%.32 That is, we think of more Republican precincts as containingstronger partisans—a large xik in the voter’s payoff function in (1)—and hence a greater fraction of “party first” as opposed to “democracyfirst” voters. Although this is only one of several possible voter-levelinterpretations of our precinct-level data, it is supported by actualassociationsbetween individual-level political attitudes andaggregatevoting outcomes: in the appendix, we merge the 2016 county-levelpresidential vote data with the 2016 Cooperative CongressionalElection Study to show that in more-Republican (Democratic)counties, Republicans (Democrats) tend to be stronger partisans,more conservative (liberal) on the liberal-conservative scale, moreconservative (liberal) on issue position-based measures of ideology,and more disapproving (approving) of former President BarackObama.
33 The 2016Republican two-party vote share ranges from31%to 91%,with just nine precincts below 50% and three precincts below 40%.34 The tophorizontal axis lists themeandifferences betweenabsenteeand election day vote shifts for the corresponding subset of precincts.
Democracy in America?
405
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
percentage of absentee voters, percentage living withinthe city limits, and mean age.35
Table 3 displays regression estimates for models inequations (7) and (8). Our main coefficients of interestare b3 and b7. In equation (7), b3 refers to the overalleffect of Gianforte’s assault on his 2017 election dayvote share and is estimated to be 20.036 of the two-party vote share (column 1). Thus, overall, Gianfortewas punished by the loss of 3.6% of election day voters.
Our main interest is in b7, which captures how theassault’s effect varied with the 2016 Republican voteshare across precincts.36 A positive b7 implies thatGianforte’s 2017 vote loss was less severe in precinctswith ahigher 2016Republican vote share. This is indeedwhat we observe (column 2): the more Republicana precinct was in 2016, the more forgiving election dayvoters were of Gianforte’s assault in 2017. Columns 3and 4 probe the robustness of these findings byrestricting the sample to precincts most consistent withthe parallel trend assumption.37
These findings are consistent with our theoreticalframework and experimental findings: only moderateprecincts punished Gianforte for assaulting the jour-nalist by either abstaining or voting for a Democrat. Inheavily partisan precincts, partisan loyalty trumpedvalence considerations.
CONCLUSION: IT CAN’T HAPPEN HERE?
This study investigates a conspicuous contradictionbetween the ideals and the practice of American de-mocracy. At least since Tocqueville’s Democracy inAmerica, theUnitedStateshas servedasanarchetypeofdemocratic political development and an aspirationalmodel for the rest of the world. Within the UnitedStates, a shared devotion to democratic ideals is a corepart of the country’s national identity. “To reject thedemocratic creed,” to quote Dahl (1961, 317), “is ineffect to refuse to be an American.”
Yet theUnitedStates stands out as exceptional amongadvanced democracies in the perplexing recurrence ofa number of democratic defects, especially at the stateand local levels. Distortions of the democratic processlong abandoned by most democracies—especiallygerrymandering and the partisan administration and ad-judication of elections—remain constitutional in theUnited States.38 And even those that are indisputablyunconstitutional—like voter disenfranchisement—resurface with unsettling regularity.
Our research addresses these contradictions by ex-amininga fundamentalquestionabout the robustnessofAmerican democracy: Is allegiance to democraticprinciples in the United States strong enough for theelectorate to serve as a check onundemocratic behaviorby elected politicians? The relevance of this questionrests on the key role that elections play as an instrumentof democratic stability: the public can stop candidateswith authoritarian tendencies by defeating them at thepolls.
Our analysis of a candidate-choice experiment aswellas a natural experiment consistently found that onlya small fraction of Americans prioritize democraticprinciples in their electoral choices when doing so goesagainst their partisan identification or favorite policies.
TABLE 3. Difference-in-Differences Estimates
Dependent variable: Republican two-party vote share
Full sample Restricted sample
(1) (2) (3) (4)
b1 year 2017 20.048*** 20.024 20.063*** 20.111***(0.008) (0.022) (0.012) (0.027)
b2 election day 0.087*** 0.171*** 0.128*** 0.301***(0.023) (0.038) (0.032) (0.061)
b3 2017 3 election day 20.036* 20.237*** 20.042* 20.174**(0.014) (0.041) (0.021) (0.055)
b7 2017 3 %Ri16 3 election day 0.313*** 0.220**(0.066) (0.085)
N 348 348 164 164Adjusted R2 0.315 0.904 0.432 0.923
Note: Standard errors clustered by precinct.* p , 0.05; ** p , 0.01; *** p , 0.001.
35 These controls are based on the voter file and account for time-varying factors that may differentially affect absentee and election dayvoters in the same precinct. In particular, there is a trend toward ab-sentee voting in Montana: absentee ballots constituted 42.6% of allballots cast in 2008, 47.2% in 2010, 58.9% in 2012, 60.2% in 2014, 65.4%in 2016, and 73.1% in 2017. Source: “AbsenteeTurnout 2,000-Present,”Montana Secretary of State, accessed on November 16, 2018.36 In equation (8), b3 estimates Gianforte’s election day vote share ina precinct with the 2016 Republican vote share of 0 (i.e., Ri16 5 0).37 These are precincts for which we can verify that the 2014 to 2016difference-in-differences was less than 5%. This was the case for 41 of68 precincts for which we have data from the year 2014. See the ap-pendix for details and further plausibility checks for key assumptionsbehind difference-in-differences estimation.
38 For a historical perspective on democratic development in theUnited States, see Bateman (2018) and Mickey (2015).
Matthew H. Graham and Milan W. Svolik
406
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
We proposed that this is the consequence of twomechanisms: (i) voters are willing to trade off demo-cratic principles for partisan ends and (ii) voters employa partisan “double standard” when punishing candi-dates who violate democratic principles. These ten-dencies were exacerbated by several types ofpolarization, including intense partisanship, extremepolicy preferences, and divergence in candidate plat-forms. Put simply, polarization undermines the public’sability to serve as a democratic check.
We conclude by discussing the implications of thesefindings for our understanding of democratic stability intheUnited States and the rest of the world.We saw thatroughly 10–13%of our respondents—depending on thetype of contests considered—value democracy enoughto punish otherwise favored candidates for violatingdemocratic principles by voting against them. Wehave interpreted this result in light of several bench-marks, including the punishment that voters meted outfor two undesirable valence attributes unrelated todemocracy—underpayment of taxes and extramaritalaffairs—as well as the trade-offs that voters made be-tween democratic principles and their favorite party orsocial and economic policies. While our respondentsvalued democracy almost as much as some of theseattributes, such comparisons potentially disguise ourmost concerning finding: because of the greater com-bined importance of policies and partisanship, mostAmericans will not cross party lines to punish a co-partisan for violating democratic principles, especially ifthe latter proposes appealing policies.
This conclusion should be interpreted in light ofanother benchmark: our candidate-choice experimentprioritized realistic, if incremental violations of demo-craticprinciplesover those signalinga sharp turn towarddictatorship. This was a deliberate design choice,intended to capture key features of the process knownasdemocratic backsliding, erosion, or degradation.Thisprocess starts in a democratic status quo, is typicallyinitiated by or in conjunction with the executive, andentails violations of democratic principles that tilt theplaying field in favor of the incumbent without beingoutright unconstitutional—often claiming to correct anostensible deficiency in thedemocratic process (HuqandGinsburg 2018; Levitsky and Ziblatt 2018; Svolik 2019).Accordingly, our experiment featured undemocraticpractices that have been attempted or employed bypoliticians from both major parties, avoided normative
connotations in their presentation, and focused ondepartures from a democratic status quo (as opposed toimprovements to democratic practice). In sum, we haveattempted to mirror as realistically as possible the keyfeatures of democratic backsliding in the context ofcontemporary American politics.
We now consider what is arguably the most relevantbenchmark: whether the American electorate valuesdemocracy enough to punish a real-world candidate forundermining democratic principles by defeating him atthe polls. To address this question, recall that we basedmost of our analysis on a candidate-choice experiment inwhich all candidate attributes were independently andrandomly assigned.Akey advantageof this design is thatit allows us to identify each attribute’s causal effect. Itspotential downside is limited external validity, whichmanifests in two ways: first, many of our candidatesfeatured combinationsof policies andpartisanship rarelyobserved in the real world; second, the overall distri-bution of candidate policies was more centrist than istypically seen in contemporary U.S. elections.
To characterize the most plausible real-worldimplications of our analysis, we now progressively trimthe least realistic scenarios from the more than 21,000candidate choices made by our respondents. Table 4lists our criteria and their implications. Condition 1restricts attention to candidate-choice scenarios that pita Democrat against a Republican—the most frequenttype of a general election contest. Condition 2 requiresat least some platform divergence by discarding anycontests in which the two candidates adopted the samepolicy platform (a rare occurrence in real-world elec-tions). Condition 3 generates a moderate party–policyalignmentbyasking that theRepublican candidatebe tothe right of the Democrat on both economic and socialpolicies.39
Aswegradually restrict attention to candidate-choicescenarios that better approximate real-world elections,the punishment for candidates who violate democraticprinciples declines from 11.7% to 3.5%. This is becausethe progressive application of the three conditions inTable 4 explores the consequences of a conception ofcandidate polarization that we have anticipated theo-retically but have not addressed empirically so far:a positive correlation between candidates’ policies andpartisanship. As each condition induces an increasingalignment between each candidate’s policies and par-tisanship, the differences that respondents see betweenDemocrats and Republicans compound, reducing vot-ers’ willingness to prioritize democratic principles overpartisan interests.When candidate policy platforms andpartisanship align—just as they do in the realworld—the viability of the public as a democratic checkdeclines.
In sum, our initial analysis yielded a conservativeestimate of the U.S. electorate’s vulnerability to can-didates who violate democratic principles. Jointly withour findings from the 2017 Montana natural
TABLE 4. The U.S. Electorate’s Resilience toUndemocratic Candidates: A SensitivityAnalysis
Condition % defecting from D2
Overall 11.74 (9.99, 13.47)
1. Cross-party contests 10.39 (8.12, 12.72)2. Platform divergence 7.27 (4.25, 10.36)3. Moderate party-policy
alignment3.46 (22.64, 9.54)
39 Note that condition 3 precludes Republican candidates fromadopting the leftmost position on any issue and vice versa forDemocrats.
Democracy in America?
407
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
experiment, this exploration of the most plausible real-world implications of our candidate-choice experimentsuggests an explanation for the recurrence of demo-cratic defects across the United States: in states anddistricts where one party enjoys a significant electoraladvantage, politicians from the majority party may beeffectively insulated from an electoral punishment forviolating democratic principles. To get a sense of thereal-world relevance of this implication, consider that in2016 only 5.1% of US House districts were won bya margin of less than 6.9%—the smallest margin thatTable 4 implies is necessary for violations of democraticprinciples to be electorally self-defeating. That share ofdistricts was still only 15.2% in 2018. Put bluntly, ourestimates suggest that in the vastmajority ofU.S.Housedistricts, a majority-party candidate could openly vio-late one of the democratic principles we examined andnonetheless get away with it.
Our findings contrast sharply with political scientists’conventional measures of support for democracy. Thelatter arebasedonquestions like“Democracymayhaveproblems, but it is better than any other form of gov-ernment. To what extent do you agree or disagree withthis statement?” Just like citizens of other advanceddemocracies, Americans express by this measure anoverwhelming support for democratic principles: Morethan 85% either “agree” or “strongly agree.” In ourframework, by contrast, respondents “support de-mocracy” not when they profess a commitment to ab-stract democratic ideals but when their choices reveala willingness to act on those ideals. We saw that, onceconfronted with candidate-choice scenarios that mirrorthe process of democratic backsliding by putting a priceon democracy—as when a co-partisan candidate pro-poses a measure that violates democratic princi-ples—only a fraction of Americans choose democracyover partisan loyalty. Our findings thus suggest thatconventional measures of support for democracy havea fundamental blind spot: they fail to capture voters’willingness to act on their commitment to democracyprecisely when democracy is at stake.
Our focus on the United States brings back intodemocratization research a case that featured promi-nently in the early, influential research on democraticculture (Almond and Verba 1963; Dahl 1966) but hassince been conspicuously absent. Our findings raisequestionsabout the reliability ofourexistingknowledgeabout support for democracy in both the United Statesand the rest of the world. On the one hand, our analysisreveals that the American voter is not an outlier:American democracy may be just as vulnerable to thepernicious consequences of polarization as are elec-torates throughout the rest ofworld (McCoy andSomer2019; Svolik 2020).40 On the other hand, because con-ventional measures of support for democracy aroundthe world suffer from the same flaws that our studyhighlights, we may have been unduly confident aboutsupport for democracy worldwide. Put differently, if
only 3.5% of voters realistically punish violations ofdemocratic principles in one of the world’s oldest de-mocracies, we should not be surprised by the public’sfailure to stop aspiring autocrats in new democracies(Ahlquist et al. 2018; Nalepa, Vanberg, and Chiopris2018). Our findings suggest a sobering upper bound onwhat can reasonably be expected from ordinary peoplein defense of democracy.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, pleasevisit https://doi.org/10.1017/S0003055420000052.
Replication materials can be found on Dataverse at:https://doi.org/10.7910/DVN/EEARKA.
REFERENCES
Abramowitz,Alan I., andKyleL. Saunders. 2008.“IsPolarization andMyth?” The Journal of Politics 70 (2): 542–55.
Ahlquist, John S., Nahomi Ichino, Jason Wittenberg, and DanielZiblatt. 2018.“HowDoVotersPerceiveChanges to theRulesof theGame? Evidence from the 2014 Hungarian Elections.” Journal ofComparative Economics 46 (4): 906–19.
Almond, Gabriel A., and Sidney Verba. 1963. The Civic Culture:Political Attitudes and Democracy in Five Nations. Newbury Park,CA: Sage.
Ashworth,Scott, andEthanBuenodeMesquita. 2009.“ElectionswithPlatform and Valence Competition.” Games and Economic Be-havior 67: 191–216.
Barber, Michael J., and Nolan McCarty. 2015. Causes and Con-sequences of Polarization. In Solutions to Political Polarization inAmerica, ed. Nathaniel Persily. New York: Cambridge UniversityPress, 15–58.
Bateman, David A. 2018.Disenfranchising Democracy: Constructingthe Electorate in the United States, the United Kingdom, and France.New York, NY: Cambridge University Press.
Becher, Michael, and Sylvain Brouard. 2019. “Executive Account-ability beyond Outcomes: Experimental Evidence on PublicEvaluations of Powerful Prime Ministers.” Unpublished Manu-script, IAST Toulouse.
Bermeo, Nancy. 2003. Ordinary People in Extraordinary Times: TheCitizenry and the Breakdown of Democracy. Princeton, N.J.:Princeton University Press.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004.“How Much Should We Trust Differences-in-Differences Esti-mates?” Quarterly Journal of Economics 119 (1): 249–75.
Cameron, A. Colin, and Pravin K. Trivedi. 2005.Microeconometrics:Methods andApplications. NewYork:CambridgeUniversityPress.
Carey, John M., Gretchen Helmke, Brendan Nyhan, MitchellSanders, and Susan Stokes. 2019. “Searching for Bright Lines in theTrump Presidency.” Perspectives on Politics 17 (3): 699–718.
Carey, John M., Katherine P. Clayton, Gretchen Helmke, BrendanNyhan,Mitchell Sanders, and SusanC. Stokes. 2018. “Party, Policy,Democracy and Candidate Choice in U.S. Elections.” Report,Bright Line Watch.
Chen, Jowei, and Jonathan Rodden. 2013. “UnintentionalGerrymandering: Political Geography and Electoral Bias in Leg-islatures.” Quarterly Journal of Political Science 8 (3): 239–69.
Cho, Wendy K. Tam, and Yan Y. Liu. 2016. “Toward a TalismanicRedistricting Tool: A Computational Method for Identifying Ex-treme Redistricting Plans.” Election Law Journal 15 (4): 351–66.
Clark, Tom S. 2009. “The Separation of Powers, Court Curbing, andJudicial Legitimacy.” American Journal of Political Science 53 (4):971–89.
Converse, Philip E. 1964. “The Nature of Belief Systems in MassPublics.” Critical Review 18 (1–3): 1–74.
Dahl, Robert A. 1956. A Preface to Democratic Theory. Chicago, IL:University of Chicago Press.
40 For a classic analysis of the relationship between polarization anddemocratic stability, see Bermeo (2003) and Sartori (1976).
Matthew H. Graham and Milan W. Svolik
408
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052
Dahl, Robert A. 1961. Who Governs? Democracy and Power in anAmerican City. New Haven: Yale University Press.
Dahl, Robert A. 1971. Polyarchy: Participation and Opposition. NewHaven: Yale University Press.
Dahl, Robert A., ed. 1966. Political Oppositions in Western De-mocracies. New Haven: Yal.
Eggers, AndrewC. 2014. “Partisanship and Electoral Accountability:Evidence from the UK Expenses Scandal.” Quarterly Journal ofPolitical Science 9 (4): 441–72.
Fiorina, Morris P., Samuel A. Abrams, and Jeremy C. Pope. 2008.“Polarization in the American Public: Misconceptions and Mis-readings.” The Journal of Politics 70 (2): 556–60.
Foa, Roberto Stefan, and Yascha Mounk. 2017. “The Danger ofDeconsolidation: The Democratic Disconnect.” The Danger ofDeconsolidation: The Democratic Disconnect 27 (3): 5–17.
Gandhi, Jennifer, and Elvin Ong. 2019. “Committed or ConditionalDemocrats? Opposition Dynamics in Electoral Autocracies.”American Journal of Political Science 63 (4): 948–63.
Gelman, Andrew. 2008. “Scaling Regression Inputs by Dividing byTwo Standard Deviations.” Statistics in Medicine 27 (15): 2865–73.
Graham, Matthew, and Milan W. Svolik. 2019. “Turnout as a Dem-ocratic Check?” Unpublished Manuscript, Yale University.
Grimmer, Justin, Eitan Hersh, Marc Meredith, Jonathan Mummolo,and Clayton Nal. 2018. “Obstacles to Estimating Voter ID LawsEffect on Turnout.” The Journal of Politics 80 (3): 1045–51.
Haggard, Stephan, and Robert R. Kaufman. 2016. Dictators andDemocrats: Masses, Elites, and Regime Change. Princeton:Princeton University Press.
Hainmueller, Jens, Daniel J. Hopkins, and Teppei Yamamoto. 2015.“Causal Inference in Conjoint Analysis: Understanding Multi-Dimensional Choices via Stated Preference Experiments.”PolticalAnalysis 22 (1): 1–30.
Huq,AzizZ., andTomGinsburg. 2018.“HowtoLoseaConstitutionalDemocracy.” UCLA Law Review 65 (1): 78–169.
Inglehart, Ronald, and Christian Welzel. 2010. “Changing MassPriorities: The Link between Modernization and Democracy.”Perspectives on Politics 8 (2): 551–67.
Iyengar, Shanto, Yphtach Lelkes, Matthew Levendusky, Neil Mal-hotra, and Sean J. Westwood. 2018. “The Origins and Con-sequences of Affective Polarization in the United States AffectivePolarization: An Outgrowth of Partisan Social Identity.” AnnualReview of Political Science 22: 129–46.
Kaufman, Robert R., and Stephan Haggard. 2019. “DemocraticDecline in the United States: What Can We Learn from Middle-Income Backsliding?” Perspectives on Politics 17 (2): 417–32.
Leeper, Thomas J., Sara B. Hobolt, and James Tilley. Forthcoming.“Measuring Subgroup Preferences in Conjoint Experiments.”Political Analysis.
Levitsky, Steven, and Daniel Ziblatt. 2018. How Democracies Die.New York: Crown.
Levitsky, Steven, and Lucan A. Way. 2010. Competitive Authoritar-ianism: Hybrid Regimes after the ColdWar. New York: CambridgeUniversity Press.
Lieberman, Robert C., Suzanne Mettler, Thomas B. Pepinsky,Kenneth M. Roberts, and Richard Valelly. 2019. “The TrumpPresidency and American Democracy: A Historical and Compar-ative Analysis.” Perspectives on Politics 17 (2): 470–9.
Lipset, Seymour M. 1960. Political Man: The Social Bases of Politics.Garden City, NY: Doubleday.
Luo, Zhaotian, and Adam Przeworski. 2018. “Subversion by Stealth:Dynamics of Democratic Backsliding.” Unpublished Manuscript,New York University.
McCarty, Nolan, Keith T. Poole, and Howard Rosenthal. 2008. Po-larized America: The Dance of Ideology and Unequal Riches.Cambridge, MA: MIT Press.
McCoy, Jennifer, and Murat Somer. 2019. “Polarizing Polities: AGlobal Threat to Democracy.” The Annals of the AmericanAcademy of Political and Social Science 681: 1–273.
Mickey, Robert. 2015. Paths Out of Dixie: The Democratization ofAuthoritarian Enclaves in America’s Deep South, 1944–1972.Princeton, NJ: Princeton University Press.
Miller, Michael K., David Szakonyi, and Lee Morgenbesser. 2017.Authoritarian Warning Survey.
Nalepa, Monika, Georg Vanberg, and Caterina Chiopris. 2018.“Authoritarian Backsliding.”UnpublishedManuscript, Universityof Chicago and Duke University.
Norris, Pippa. 2011. Democratic Deficit: Critical Citizens Revisited.New York: Cambridge University Press.
Przeworski, Adam. 1991. Democracy and the Market: Political andEconomic Reforms in Eastern Europe and Latin America. Cam-bridge: Cambridge University Press.
Przeworski,Adam.2019.Crises ofDemocracy.NewYork:CambridgeUniversity Press.
Sartori, Giovanni. 1976. Parties and Party Systems: A Framework forAnalysis. Cambridge, U.K.: Cambridge University Press.
Schedler, Andreas. 2002. “The Menu of Manipulation.” Journal ofDemocracy 13 (2): 36–50.
Svolik, Milan. 2019. “Polarization vs. Democracy.” Journal of De-mocracy 30 (3): 20–32.
Svolik, Milan W. 2020. “When Polarization Trumps Civic Virtue:Partisan Conflict and the Subversion of Democracy by Incum-bents.” Quarterly Journal of Political Science 15 (1): 3–31.
Treier, Shawn, and D. Sunshine Hillygus. 2009. “The Nature of Po-litical Ideology in the Contemporary Electorate.” Public OpinionQuarterly 73 (4): 679–703.
Voeten, Erik. 2017. “Are People Really Turning Away from De-mocracy?” Journal of Democracy Web Exchange: 1–16.
Waldner,David, andEllenLust. 2018. “UnwelcomeChange: ComingtoTermswithDemocraticBacksliding.”AnnualReviewofPoliticalScience 21: 93–113.
Weingast, Barry R. 1997. “The Political Foundations of Democracyand the Rule of Law.” American Political Science Review 91 (2):245–63.
Democracy in America?
409
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. Y
ale
Uni
vers
ity L
ibra
ry, o
n 23
Apr
202
0 at
10:
35:1
5, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
2000
0052