the geographic distribution of political...

ANRV412-PL13-16 ARI 15 January 2010 13:55

R

E V I E W

S

IN

AD V A

NC

E

The Geographic Distributionof Political PreferencesJonathan RoddenDepartment of Political Science, Stanford University, Stanford, California 94305;email: [email protected]

Annu. Rev. Polit. Sci. 2010. 13:321–40

The Annual Review of Political Science is online atpolisci.annualreviews.org

This article’s doi:10.1146/annurev.polisci.12.031607.092945

Copyright c© 2010 by Annual Reviews.All rights reserved

1094-2939/10/0615-0321$20.00

Key Words

economic geography, political geography, party platforms, electoralbias, electoral rules

AbstractIn order to address classic questions about democratic representationin countries with winner-take-all electoral districts, it is necessary tounderstand the distribution of political preferences across districts. Re-cent formal theory literature has contributed new insights into how par-ties choose platforms in countries with a continuum of heterogeneousdistricts. Meanwhile, increases in survey sample sizes and advances inempirical techniques have made it possible to characterize the distri-bution of preferences within and across electoral districts. This reviewaddresses an emerging literature that builds on these new tools to ex-plore the ways in which the geography of political preferences can helpexplain the parties that compete, the platforms and policies they choose,and even the rules under which they compete. Building on insights fromeconomic and political geography, it pays special attention to electoraland policy biases that can emerge when there is an asymmetric distri-bution of preferences across districts.

321

Review in Advance first posted online on January 26, 2010. (Changes may still occur before final publication online and in print.)

Changes may still occur before final publication online and in print.

Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


INTRODUCTION

Voters cannot be regarded as scattered at ran-dom over the various constituencies. (Kendall& Stuart 1950, p. 188).

This article considers the implications ofKendall & Stuart’s important but relatively ob-vious insight about British elections in the firsthalf of the twentieth century. Through someprocess involving homophily, segregation, andsocialization, voters are clustered into neigh-borhoods of individuals with similar politicalpreferences and party orientations. In soci-eties with a high degree of residential mobil-ity, individuals sort themselves into neighbor-hoods with similar demographic, occupational,income, and ultimately political profiles. Forthe poor, residential choices in metropolitan ar-eas are severely circumscribed by housing andtransportation options. Even in traditional soci-eties without much residential mobility, votersare often clustered into villages or neighbor-hoods where strong social ties reinforce sim-ilar political preferences and voting behavioramong neighbors, and these often endure forgenerations.

In other words, Tobler’s (1970) “first law ofgeography” is true of political preferences: “Ev-erything is related to everything else, but nearthings are more related than distant things.”Modern spatial statistics allows the quantifica-tion of this relationship, and it is relatively easyto show that voting behavior is spatially depen-dent: the probability that two randomly drawnindividuals (or precincts) exhibit similar votingbehavior is a function of the distance betweentheir locations (Chen & Rodden 2009).

This is old news dressed up with new tech-niques. Political geographers and historianshave long attempted to describe and analyzethe historical roots of such clusters as the “redbelt” in Italy (Bagnasco 1977) or the “blackbelt” in the U.S. South (Key 1949). More re-cently, empirical scholars have attempted togain greater insights into the mechanism be-hind so-called neighborhood effects or con-textual effects, whereby political socializationis thought to transmit values or behaviors to

newcomers and offspring (Johnston 1992, Choet al. 2006).

Yet these observations have done surpris-ingly little to inform modern positive politi-cal economy. Scholars have learned to quan-tify the extent to which political behavior isgeographically clustered and, with mixed suc-cess, attempted to explain why this is the case,but implications for elections and representa-tion have not been fully explored. For the mostpart, when developing basic models addressingsuch crucial topics as platform choice, party sys-tems, representation, and the transformation ofpreferences to policies, geography has been ablind spot for political scientists. This is an es-pecially noteworthy oversight in former Britishcolonies such as the United States, where leg-islative representation takes place via winner-take-all districts.

Beginning with Downs (1957), the spatialtheory of elections provides the analytical foun-dation for most contemporary theorizing aboutrepresentative politics. Policies are understoodas points along a single issue dimension, whereeach voter can be characterized by an idealpoint. Two candidates offer platforms designedto maximize their chance of winning the seat,and in the workhorse model, electoral com-petition forces them to converge to the idealpoint of the median voter. The vast majority oftheoretical work making use of this frameworktreats competition between two candidates ina single district the same way as competitionbetween two parties in a national election. Inorder to sweep geography under the rug, thisliterature often makes one of two simplifyingassumptions: Either each district contains anidentical distribution of voter ideal points, or,perhaps more realistically, the overall distribu-tion of individuals mirrors the distribution ofdistrict medians.

Perhaps these simplifying assumptions havebeen reinforced by challenges related to sur-vey research. Scholars work hard to obtain rep-resentative national samples, but until very re-cently, they have not had the resources to aimfor representative samples at lower levels suchas provinces or electoral districts.

322 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


While students of parties and elections haveapplied a workhorse model with a blind spot forgeography, political economists have done thesame thing with their workhorse model of in-come and redistribution. Building on some ofAristotle’s observations about democracy andthe income distribution in the Politics, Romer(1975) and especially Meltzer & Richard (1981)lay the foundation for much modern politicaleconomy with a simple median-voter model inwhich the level of redistribution in a polity isdriven by the difference between the income ofthe median voter and the average income acrossall voters. As in the Downsian literature, ap-plications of this model generally ignore thedetails of representation through geographicdistricts, implicitly assuming that the distribu-tion of income across individuals resembles thatacross electoral districts.

Such work has led to important insights,but it has left some nagging concerns. Aboveall, since we know that individuals are clus-tered into relatively homogeneous neighbor-hoods through some social process that is onlydimly understood, the distribution of district-level ideal points or income might be differ-ent from the distribution across individuals.Though avoiding the notions of ideal points orideology and focusing instead on partisanship,this is the central insight of some of the classicsof British electoral geography (e.g., Taylor &Johnston 1979, Gudgin & Taylor 1979). Theseworks viewed elections as two interacting spa-tial distributions—the pattern of party votersand the pattern of constituency boundaries—and demonstrated the many ways in which thesuperimposition of boundaries on the geogra-phy of party voters might generate distortionsin the transformation of votes to seats.

Their insights have had surprisingly littleimpact on the work of contemporary politicalscientists and economists. Since the golden eraof British electoral geography in the 1970s, ge-ographers have largely turned away from math-ematics and formal modeling, while politicalscientists have embraced them. Beginning withHinich & Ordeshook’s (1974) work on the elec-toral college, a handful of formal theorists have

approached these same questions, but in a dif-ferent spirit: They are attempting to rectifythe geographic blind spot of the Downsianmodel by developing theories of platformchoice in which party leaders adopt a singlenational platform in a context where the rele-vant distribution of preferences or partisanshipis across a continuum of districts rather thanindividuals.

While formal theorists have begun to de-velop some insights about how the geographicdistribution of preferences affects the logicof platform choice, they have not asked eventhe most basic questions about what might bedriving these distributions. Meanwhile, eco-nomic geographers and urban economists haveplanted the seeds for systematic theories ofgeography and political preferences by takingTobler’s Law very seriously. They have dis-covered that economic activity and residen-tial choices are most certainly not randombut follow distinctive patterns that are drivenby agglomeration and urbanization economies,real estate markets, the political economyof zoning, and the logic of transportationinfrastructure.

This article argues that by extending thenascent political science theory literature onplatform choice in a way that highlights theinsights of classic British electoral geography,adding some basic lessons from economic ge-ography and urban economics, and taking ad-vantage of new advances in data quality andempirical technique, it is possible to derivesome surprising new insights into old ques-tions about representation and the (imperfect)transformation of preferences into public pol-icy. After we establish some basic facts aboutthe distribution of political preferences in ge-ographic space, and then explore the processthrough which plurality districts are drawn, itbecomes clear that because of the way individ-uals cluster together in space, the overall distri-bution of individual preferences can have a dif-ferent shape, and a different median, than theinterdistrict distribution. Once we add detailsabout electoral and legislative institutions, wecan gain fresh insights into key questions about

www.annualreviews.org • Geographic Distribution of Political Preferences 323

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


party systems, platform choice, and the trans-formation of votes to seats and preferences topolicies, as well as questions about the rela-tionship between electoral rules and long-termcross-country differences in policies.

This article pays special attention to onesuch argument with roots in classic British po-litical geography. There are good reasons to be-lieve that in many industrialized democracies,the early postwar period featured a left skew inthe distribution of partisanship across districts.Leftists were highly concentrated in industrial-ized urban districts and mining regions. Thisinefficient distribution across districts causedthe parties of the left to suffer in the transforma-tion of votes to seats, and perhaps even causedleftist ideologues to suffer in the transformationof preferences to policies. Moreover, the highlyskewed distribution of socialists across districtsaround the turn of the century might help ex-plain the strategies of the key players duringthe period of franchise expansion and institu-tional choice in Europe in the early twentiethcentury.

Although it is possible to use electoral datato show a long-term pattern whereby the sup-port for labor and socialist parties is more con-centrated than that of right-wing parties, it isimportant not to confuse the distribution ofvoting behavior across districts with the distri-bution of political preferences on some salientissue dimensions. The latter has been particu-larly difficult to measure because, in years past,surveys have not contained enough observa-tions to support reliable inferences about pref-erences at the district level. Survey researchershave made slow, steady progress on this issue,however, and at least in the United States, itis now possible to get a relatively believablesnapshot of district-level preferences on mul-tiple issue dimensions. As with district-levelvote shares for Republican presidential candi-dates, the distribution of median preferencesacross districts indeed has a pronounced leftskew owing to the concentration of leftists indensely populated cities. This observation hassome important implications for the transla-tion of votes to seats, and possibly some more

controversial implications for the translation ofpreferences into policies. Moreover, the inter-district distribution of preferences might helpresolve some other puzzles in the study ofCongress.

These observations might also have im-plications for the more abstract literature inthe Meltzer-Richard (1981) tradition that as-sumes political preferences are derived exclu-sively from one’s place on the income spectrum.In many societies the poor live in higher den-sity than the rich, such that the median voterin the median district is wealthier than the me-dian voter in the society as a whole. Moreover,the distribution of income across districts willalways be far less right-skewed than the dis-tribution across individuals, and if the districtsare large enough, the distribution will not beskewed at all. These simple observations haveclear, as-yet-untested implications for redistri-bution. Putting together the arguments aboutthe influence of leftist platforms and poor vot-ers, political geography might be an importantpart of the explanation for the long-term differ-ences between the policy profiles of countrieswith single-member plurality districts and thosewith proportional representation.

This article proceeds as follows. The firstsection reviews the theory literature on plat-form choice with heterogeneous plurality dis-tricts, highlighting recent progress as well asremaining holes, while paying special attentionto the possibility of a skewed distribution ofpreferences across districts. The second sec-tion explains why the interdistrict distributionmight indeed be unavoidably skewed in somesocieties, and explores implications for electoraland policy bias. The third section explains whyit is important to distinguish between the in-terdistrict distributions of partisanship and po-litical preferences, and explores efforts to mea-sure political preferences and characterize theirdistribution within and across districts, payingspecial attention to the United States. Sectionsfour and five are more speculative, focusing onimplications and questions for further researchin the American and comparative politics liter-atures, respectively.

324 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


POLITICAL COMPETITIONWITH HETEROGENEOUSPLURALITY DISTRICTS

Symmetric Distributionsof District-Level Preferences

Beginning with Hinich & Ordeshook (1974),a handful of theory papers have attempted tomove beyond the single-district framework ofHotelling (1929) in order to address the factthat, at least in Britain and its former colonies,parties must set their platforms in a context ofmultiple, heterogeneous plurality districts.

Figure 1a displays the spatial distribution ofpreferences in a hypothetical society with fivedistricts, where there is a symmetric, unimodaldistribution of preferences within each districtand a symmetric distribution of district medi-ans. The medians of each district are markedwith ticks on the horizontal axis. One of thefirst multi-district models was that of Hinich& Ordeshook (1974), which proves the analogof the famous single-district result: two com-peting parties converge to the ideal point of themedian voter in the median district. In the sym-metric example in Figure 1a, the median voterin the median district is identical to the medianvoter in the society.

Yet this type of model is somewhat unsatis-fying, above all because we observe in practicethat platforms do not converge, either at thedistrict level or the national level (Ansolabehereet al. 2001). The parties might have to worryabout entry by third parties in the extreme dis-tricts (Palfrey 1984). Moreover, given the het-erogeneity portrayed in Figure 1a, it seemslikely that parties will face internal tension be-tween different constituencies, and this tensionmight not always be resolved in a way thatleads to the adoption of the seat-maximizingplatform.

The first problem is taken up by Callander(2005), who considers a uniform distributionof districts like that displayed in Figure 1a.Callander focuses on the competing needs ofthe two major national parties to appeal tomoderates and win districts in the middle ofthe spectrum while deterring entry by third

M1 M2

M3

M4 M5

MVM1 M2

M3

M4 M5

M1 M2 M4 M5

M3

(a)

(b)

(c)

Figure 1Hypothetical distributions of voters and districts.


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


parties in the extreme districts. In this model, aparty enters only if it can win a district, so theentry-deterring equilibrium platforms for thetwo parties in Figure 1a would be at the me-dians of the two extreme districts. This wouldstave off entry in the extreme districts whilealso barely avoiding the entry of a centrist partythat adopts the median preference in the me-dian district. Thus, in contrast to the Hinich &Ordeshook (1974) model, the platforms of theparties are quite far apart.

In Callander’s model, party leaders are ableto choose the most efficient platform for theparty as a whole. This is not the case in Austen-Smith’s (1984) account, where national partyplatforms are aggregations of the policy po-sitions of individual candidates who care pri-marily about securing their own reelection,and where the mechanism through which partymembers’ platforms are aggregated into partyplatforms might allow some individuals to bemore influential than others. In a similar vein,Snyder (1994), Ansolabehere et al. (2005), andLeblanc (2007) model national party platformsas emerging from a process of collective choiceamong the party’s legislative incumbents.

A key feature of the model developed byAnsolabehere, Leblanc, and Snyder is that itinvolves two periods. The platforms in the sec-ond period are determined by the median of thelegislative incumbents elected in the first pe-riod, but the outcome of the initial election isaffected by some exogenous valence shock (e.g.,economic crisis or war) that favors one partyor the other. To understand the logic, considerFigure 1a once again. In the initial election,simply apply the logic of Hinich & Ordeshook(1974) and assume that the parties converge tothe preference of the median voter in the me-dian district (that of district 3), such that nei-ther party has an advantage on the ideologicaldimension. But imagine that the party of theright benefits from a valence shock, allowingit to capture the normally indifferent voters atthe national median, which allows the party ofthe right to win districts 3, 4, and 5. In thesecond period, the party’s platforms are chosen

by majority rule among the incumbents, whoseinduced platform preferences are their districtmedians. Thus, the platform of the right party,R, is the median of district 4, and the platform ofthe left party, L, is the average of the medians ofdistricts 1 and 2. Thus, as the parties approachthe next election, the R party is slightly closerto the national median than the L party, whoseself-interested incumbents set the platform in away that undermines the party’s chances in thenext election. Because they are uncertain aboutfuture valence shocks, they cannot afford to al-low the party platform to wander too far fromtheir district medians.

This intuition seems to match up quitenicely with reality. When times are good fora party, its platform is influenced by moderatevoices that it has been able to bring into thefold. When times are bad, it becomes extremeand experiences a time in the wilderness, andit must wait for some exogenous good fortune,such as an unpopular war or a recession, to bringmoderates back into the party.

Asymmetric Distributionsof District-Level Preferences

Ever since Kendall & Stuart (1950), who pre-sented a normal distribution of district-levelpartisanship as a kind of natural law of politi-cal geography, much of the literature consider-ing the distribution of partisans across districtsbegins and ends with a symmetric, unimodaldistribution.

However, consider the distribution of Re-publican presidential vote shares across U.S.congressional districts, displayed in Figure 2for each election since 1952. Largely becauseDemocrats are highly concentrated in cities,the distribution demonstrates a pronouncedleft skew. In fact, a similar observation wasmade in the classic electoral geography liter-ature about several first-past-the post countriesin the decades immediately after World War II:Australia (Rydon 1957), Great Britain (Gudgin& Taylor 1979), New Zealand (Johnston 1976),and the United States (Erikson 1972).

326 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


012345

0.2

.4.6

.81

1952

012345

0.2

.4.6

.81

1956

012345

0.2

.4.6

.81

1960

012345

0.2

.4.6

.81

1964

012345

0.2

.4.6

.81

1968

012345

0.2

.4.6

.81

1972

012345

0.2

.4.6

.81

1976

012345

0.2

.4.6

.81

1980

012345

0.2

.4.6

.81

1984

012345

0.2

.4.6

.81

1988

012345

0.2

.4.6

.81

1992

012345

0.2

.4.6

.81

1996

012345

0.2

.4.6

.81

2000

012345

0.2

.4.6

.81

2004

Ke

rne

l den

sitie

s of

dis

trict

-leve

lRe

publ

ican

vot

e s

hare

in p

resi

den

tial e

lect

ions

.

The

top

row

of d

ots

indi

cate

s di

stric

ts w

on b

yRep

ublic

an H

ouse

can

dida

tes.

Th

e m

iddl

e ro

w o

f dot

s in

dica

tes

dist

ricts

won

by

Sou

ther

n D

emoc

rats

. Th

e b

otto

m ro

wof

dot

s in

dica

tes

dist

ricts

won

by

non-

Sou

ther

n D

emoc

rats

.

The

sol

id li

ne m

arks

50

perc

ent,

and

the

brok

en li

ne m

arks

the

Rep

ublic

an v

ote

Figu

re2

Dis

trib

utio

nof

supp

ortf

orR

epub

lican

sin

pres

iden

tiale

lect

ions

acro

ssU

.S.H

ouse

dist

rict

s.(F

rom

Rod

den

War

shaw

2009

.)


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


Gudgin & Taylor (1979) demonstrate thatwhen the distribution of partisanship across dis-tricts is skewed, the party competing on the sidewith the long tail is disfavored in the transfor-mation of votes to seats, for the simple reasonthat it receives too many “surplus” votes in thedistricts it wins. Yet Gudgin & Taylor do notconsider a spatial model in which voters havepreferences on some issue dimension and setplatforms. Rather, the partisanship of each in-dividual is binary and exogenous.

By considering asymmetric distributions ofcardinal district-level preferences (as opposedto binary partisanship) in spatial models withendogenous platform choice, it is possible toderive normative implications beyond electoralbias. Consider the distribution of preferencesdepicted in Figure 1b. Districts 3 and 4 are un-changed, but districts 2 and 5 have been pulledaway from the median district, and district 1 hasbeen pulled even further. Although each districtcontains a symmetric distribution of voters, thedistribution of district medians now demon-strates a left skew.

Immediately, one can draw a striking impli-cation from Hinich & Ordeshook (1974). If theparties converge on the ideal point of the me-dian voter in the median district and transformit directly into policy, the policy profile willveer substantially to the right of that preferredby the national median voter (see also Leblanc2007). By creating winner-take-all districts witha sufficiently skewed distribution of district me-dians, a society can create policy bias, suchthat plurality elections with a single nationaldistrict or elections using proportional rep-resentation would yield different equilibriumpolicies.

Though not explicitly addressed in the pa-per, the logic of Callander (2005) also has inter-esting implications for countries with skeweddistributions of district medians. In Figure 1b,if the party of the left attempted to move to themedian of district 1 in order to stave off entryof a far-left party, it would not only have dimhopes of winning a two-party contest with theparty of the right, it would be open to entry bya center-left party. The optimal strategy of the

party of the left is thus to cede district 1 to anentrant and focus on competing in districts 2,3, and 4. The long-term policy implications areless clear, given that one would need a theory ofcoalition formation, but this might be a promis-ing framework for understanding Westminstersystems such as the United Kingdom and espe-cially Canada. District-level competition therelargely abides by Duverger’s Law and focusesprimarily on two parties, but a different mix ofparties competes in different districts, and at thenational level the party system features three ormore parties.

Ansolabehere et al. (2005) and Leblanc(2007) focus more directly on the possibilityof an asymmetric distribution of district-levelmedians than others in the literature. Consideronce again a starting point in the first electionwhere the parties converge on the median ofdistrict 3. In the case of a small valence shockfavoring the right, the platform of the R partyin the next period would be the rather moder-ate median of district 4, but the L party wouldbe stuck with the very extreme midpoint be-tween the medians of districts 1 and 2. In thisway, a party can suffer from a structural disad-vantage such that it easily falls into a long-termelectoral slump because its platforms are too ex-treme for voters in the pivotal districts, threat-ening to make it a “permanent minority.” In thiscase, the structural advantage of the R party willintroduce policy bias if it faithfully implementsthe median of district 4, which is even furtherfrom the national median voter than the district3 median that is favored in the simpler Hinich& Ordeshook model.

In short, an asymmetric distribution ofdistrict-level ideal points can bring systematicbias not only in the transformation of votes toseats but also in the transformation of prefer-ences to policies.1

1A recent literature on optimal districting considers re-lated questions about ideology, the vote-seat curve, and rep-resentation, but largely ignores the geography of prefer-ences. Examples include Besley & Preston (2007), Coate &Knight (2007), Friedman & Holden (2008), and Gilligan &Matsusaka (2006).

328 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


Variations in Party Discipline

Is it really conceivable that political geogra-phy can cause a party to get stuck with an ex-treme position and lose well over 50% of elec-tions over a relatively long period, as positedby Ansolabehere et al. (2005)? Do parties reallyhave to worry about entry? Americans mighthave good reasons for skepticism, but the “long-term minority” implication of Ansolabehereet al.’s model might provide some insight intothe persistent difficulties of Labor parties inGreat Britain, Australia, and New Zealand inthe early postwar period, when their support-ers were highly concentrated in urban and min-ing districts. In fact, Iversen & Soskice (2006)document that left governments have beenformed far more often under proportional rep-resentation than under plurality systems amongOECD countries in the postwar period. If thereis a left-skewed distribution of district-levelpreferences in Ontario, the Callander (2005)model might indeed help explain why the Lib-erals maintain a centrist platform in order tocompete against the right in the pivotal sub-urban districts while allowing the New Demo-cratic Party (NDP) a foothold in the left-wingindustrial and extraction-oriented districts. Andas “New Labor” has moved to the right in re-cent years in Great Britain in order to capturethe pivotal suburban districts, it finds itself chal-lenged from the left in some leftist districts byLiberal Democrats. In the past, perhaps Laborprevented entry in the leftist districts while al-lowing the Liberals a foothold in the moderatedistricts.

Although these models yield interesting in-sights in parliamentary systems, they seem tofall flat in the United States. For most ofthe past century, third-party entry has notbeen a major concern of congressional candi-dates. And though the distribution of partisansacross districts displayed a pronounced left skew(Figure 2), it was the Republicans who spentmost of the postwar period in the wilderness.Indeed, the models above implicitly assumesomething like a Westminster-style parliamen-tary democracy. All of these models assume that

the party imposes a single national platform thatcannot be disavowed on the campaign trail byits members, and they implicitly assume strictparty-line voting in the legislature, which inturn assumes that party leaders have at theirdisposal some effective carrots and sticks, likethe threat of a no-confidence vote (Diermeier& Feddersen 1998) or the ability to nominatecandidates in the districts and dole out cam-paign funds (Mayhew 1974).

In the United States, the executive does notrely on the maintenance of a partisan major-ity in order to stay in office, and majority partyleaders do not have the threat of no-confidencevotes at their disposal. Moreover, since the riseof primaries, they have not been able to controlnominations in the districts. This allows con-siderable latitude to candidates in the districtsto break with the party leadership and bringtheir platforms closer to the district median.For example, Southern Democrats can crediblyoffer progun and antiabortion platforms, sincetheir voters know that the Democratic leader-ship has no way of forcing them to vote for guncontrol or relaxed abortion restrictions. In turn,Democratic party leaders are unlikely to bringsuch legislation to the floor in the first place.Because of parliamentary institutions and disci-plined parties, a candidate of a center-left partycannot credibly make such promises in placeslike Alberta.

As a result of this slack in the party labelsin the United States, Democrats can competequite effectively in “Republican” districts andvice versa. By indicating the district won byDemocratic and Republican candidates at thebottom, Figure 2 demonstrates that Democratsand Republicans win quite frequently on theideological “turf” of the other party. YetFigure 2 also suggests that the party label mustmean something in congressional races, andconvergence to the ideal point of the medianvoter in each district is clearly not complete,since Democratic and Republican wins are notrandomly distributed across districts. Indeed,the empirical literature on candidate position-ing (Ansolabehere et al. 2001, Burden 2004)


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


shows that there is always some space betweenthe Democratic and Republican candidates inthe districts, but there is also considerable lee-way for both candidates to adapt their platformsto the ideology of the district.

Thus, the next step in the theory literatureshould be a refinement that allows for the weak(but not completely absent) party discipline andcorresponding platform flexibility that emergein a presidential democracy such as the UnitedStates. For example, it might be more real-istic to model party platforms as somethingmore like reputations that emerge from actualpolicies, which are determined through legisla-tive bargaining (e.g., Baron & Ferejohn 1989)rather than the simple intraparty majority ruleof Ansolabehere et al. (2005) or Leblanc (2007).Another possibility is provided by Eyster &Kittsteiner (2007), who consider a model inwhich parties set platforms, but candidates inindividual districts can deviate from the plat-form at some cost.

Alternatively, a fruitful approach to theUnited States might be to dispense altogetherwith the notion of a national platform and focuson party primaries as the mechanism throughwhich candidate platforms are chosen. Return-ing to Figures 1a and 1b, consider the Demo-cratic (Republican) primary constituency in anydistrict to be all individuals with preferences tothe left (right) of the district 3 median, and allowthe median of each party’s primary constituencyto set the platform in each district. In the ex-treme districts, one of the party’s platforms willbe driven by a small number of marginalizedvoters (e.g., conservatives in Detroit, liberalsin rural Alabama).2 But in the districts rightaround the median district (e.g., districts 2 and4 in Figure 1a), the primary constituenciesare closer to symmetric, such that one party isonly slightly closer to the district median thanthe other. This gives one party a slight ideo-logical advantage, but one that could easily be

2Note that in the highly skewed distribution of district me-dians in Figure 1b, there is simply no Republican primaryconstituency in the leftmost district. This might not be farfrom reality in the most urban U.S. congressional districts.

overcome with a modest valence shock. Such amodel would correspond well with the findingsof candidate positioning studies: platform di-vergence in each district with imperfect district-level tailoring. Such a model would also squarenicely with Figure 2, in which Democrats andRepublicans can win in districts on the oppositeside of the national median, but their likelihoodof doing so declines as the distance from themedian increases.

GEOGRAPHY AND THE ROOTSOF ELECTORAL BIAS

The discussion above indicated that in Britainand its industrialized former colonies, support-ers of the left have been more concentrated inspace than supporters of the right, leading toa skew in the distribution of partisans acrosselectoral districts. Before looking more closelyat possible implications, it is useful to ask whythis might be the case.

The Geography of Industrializationand Urban Form

Urban economics and economic geographyhave developed rich insights into the evolutionof urban form and industrial location that haverelatively clear implications for electoral poli-tics. In the late nineteenth and early twentiethcentury, many countries experienced a dramatictransformation, as peasants, small farmers, andformer slaves moved from the countryside towork in industrial jobs. Because of agglomer-ation and urbanization economies (Krugman1991, Rosenthal & Strange 2004), these jobswere often highly concentrated in geographicspace. In the places where the industrial revo-lution preceded the development of mass publictransportation and automobile ownership, theerection of smelters, steel mills, and factorieswas accompanied by the construction of denseworking-class housing that has proven to beextremely durable. A similar phenomenon oc-curred in mining areas. These dense working-class neighborhoods provided an opportu-nity for political entrepreneurs on the left,who worked with trade unionists to mobilize

330 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


workers around an agenda of social insurance,public housing, workplace regulation, and thelike. Even after the factories closed and theservice economy emerged, the durable, afford-able housing from the industrial revolution re-mained (Glaeser et al. 2005), along with some ofthe public transportation networks that even-tually came into place, allowing the poor toreach service and other low-skill jobs withoutthe expense of automobile ownership (Glaeseret al. 2007). Thus, even after deindustrializa-tion, these neighborhoods have continued toattract poor migrants.

The industrial revolution cast a long shadowthrough the legacy it left on the built envi-ronment. Despite fascinating differences acrossand within countries (and across U.S. states andCanadian provinces), those that industrializedaround the turn of the century generally stillhave dense urban neighborhoods that provideattractive targets to political entrepreneurs onthe left. In the United States, factory construc-tion often coincided with the initial growth ofthe city, so these neighborhoods are close to ordirectly in the city center; in Europe, the centersof the great cities predate the industrial revolu-tion, so their industrial neighborhoods sprangup on the outskirts (Hohenberg 2004).

Since the industrial revolution, changes intransportation technology, from horse-drawntrams to the streetcar to automobiles and high-ways, have also allowed the rich and eventuallythe middle class to segregate themselves fromthe poor (Nas et al. 1998, Mieszkowski & Mills1993), which often resulted in a metropolitanpattern where the poor live in higher densitythan the rich.

Cosmopolitan Versus TraditionalSocial Values

All of this suggests a straightforward po-litical economy explanation for the positivecorrelation between population density andleftist voting observed in so many coun-tries. Yet income, occupation, and social classmight only be a part of the story. In citieswhere the nineteenth-century industrialists left

behind a sufficiently attractive legacy of ameni-ties and cultural and consumption opportuni-ties, Victorian-era working-class housing hasbeen taken over not by poor immigrants seek-ing cheap housing, but by educated young pro-fessionals willing to pay a premium for con-sumption opportunities (Brueckner et al. 1999,Glaeser et al. 2001). In many countries, thesevoters are traitors to their class, and vote as re-liably for the left as did their working-class pre-decessors (see Gelman et al. 2008 on the UnitedStates).

Even before the expansion of the franchisearound the turn of the century in Europe, acontrast had emerged between the “orthodox-fundamentalist beliefs of the peasantry and thesmall-town citizens and the secularism fosteredin the larger cities and the metropolis” (Lipset& Rokkan 1967, p. 12). Indeed, a noneconomicdimension of electoral conflict related to reli-gious versus secular values has always been atleast as powerful as class or income in predict-ing voting behavior in industrialized countries(Dalton 2006, de la O & Rodden 2008). If any-thing, the importance of this issue dimensionseems to have increased in the United States inrecent years (Gelman et al. 2008, Ansolabehereet al. 2006). Lipset & Rokkan’s observationstill applies today in many societies: Secularorientation, progressive social values, and leftvoting are all highly correlated with urbanresidence.

Urbanization, Electoral Bias,and Tobler’s Law

Although there are many reasons to expecta correlation between population density andleftist preferences, this correlation does not ex-plain why the distribution of voters across dis-tricts would be asymmetric. In the Americanliterature, the preoccupation has been with the“clever political cartographer” (Ansolabehereet al. 2005, p. 23), such as Elbridge Gerry, PhilBurton, or Tom de Lay, who achieves a skeweddistribution of partisans across districts by ma-nipulating maps, and the assumption is that par-tisan bias emerges from attempts to “crack” and


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


“pack” one’s opponents, or as an outgrowth ofminority-majority districts.

But by returning once again to the insightof Tobler’s Law, we see that in the presence ofa tight correlation between population densityand left voting, a left-skewed distribution ofpartisans across districts can emerge quitenaturally without any partisan manipulation.Using geo-coded registration data fromFlorida, Chen & Rodden (2009) show that theprobability that two randomly drawn individu-als exhibit the same partisanship is a function ofthe distance between their residential locations.This implies that any districting scheme basedon geographic contiguity and compactness,and requiring that each district be of equalpopulation size, will tend to create small,homogeneous urban districts and sprawling,relatively heterogeneous rural districts. Whenurban districts are drawn, they bring togetherproximate individuals with highly correlatedpreferences, and in many societies, as discussedabove, these are leftist preferences. Rural orexurban districts draw together individuals thatlive quite far from one another, and hence dis-play more heterogeneous preferences. Thoughconservative on average, these districts mightinclude liberal pockets of unionized schoolteachers, college towns, or small-scale miningor manufacturing operations.

Chen & Rodden (2009) use automateddistricting algorithms to draw compact, con-tiguous districts from the building blocks ofprecinct-level results of the notorious tied 2000Florida presidential election, along with severalother close statewide elections, and show thatsubstantial pro-Republican electoral bias owingto excessive spatial concentration of Democratsis virtually impossible to avoid when carvingFlorida up into plurality districts. This helpsexplain why Florida statewide elections are ex-tremely close, but the Republicans enjoy hugemajorities in both chambers of the state legisla-ture and the congressional delegation. Prelimi-nary application of automated districting algo-rithms to other urbanized states demonstratesa strikingly similar pattern. The next steps inthis research agenda are to see if the simulated

pro-Republican bias in different states links upwith observed levels of bias in state legislativeelections, and to establish the aspects of eco-nomic geography and urban form that explaincross-state variations in electoral bias.

Highly disaggregated, geo-coded electiondata are harder to come by in other countries,but in unpublished work, I have used district-level data going back to the turn of the centuryto show that votes for parties of the left havebeen more concentrated within districts thanthose of parties of the right—generally becauseof high concentrations of leftists in dense neigh-borhoods built to house manual workers—inAustralia, Canada, Great Britain, France, andNew Zealand. In each country, this has beenassociated with systematic electoral bias in fa-vor of the right over the course of the twentiethcentury, but in most countries this trend hasbeen declining throughout the century.

MEASURING DISTRICT-LEVELPREFERENCES

Even if the distribution of district-level voteshares is skewed and one party can expect toreceive 50% of the seats with less than 50%of the votes, this need not translate into pol-icy bias. This situation might be troubling tothe leaders of the afflicted party but does notnecessarily imply a normative problem for rep-resentative democracy. To see this, examineFigure 1c. In this example, districts 2, 3, and4 are identical to the symmetric case displayedin Figure 1a. Moreover, districts 1 and 5 havethe same medians as in Figure 1a. Yet the left-wing (let us now call it urban) district now has atighter, more leptokurtic distribution, while theright-wing (rural) district has a more platykurticdistribution. This hypothetical example flowsdirectly from the discussion of urbanization andTobler’s Law above. It could be the case that theideological preferences of the median voter inurban districts are not especially far from thenational median, but that voters in urbandistricts are merely more homogeneous intheir preferences than those in rural districts.Whether one applies the platform-setting logic

332 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


of Hinich & Ordeshook (1974), Callander(2005), or Ansolabehere et al. (2005), the partyof the left would win district 1 with a largermajority than that with which the party of theright would win district 5, which would cre-ate electoral bias as normally defined. Yet sincethe district medians are symmetrically arrangedaround the median district, there is no reasonto expect policy bias.

Confronted with a distribution of district-level partisanship like that displayed inFigure 2, a crucial empirical question iswhether the underlying distribution of prefer-ences resembles Figure 1b or 1c, or perhapssome combination of the two. Yet ever sinceMiller & Stokes (1963), survey researchers havesuffered from a lack of sufficient observationswithin individual electoral districts to reliablycharacterize district preferences. To get aroundthis problem, some researchers have used de-mographic variables in order to generate prox-ies for district preferences (e.g., Pool et al. 1965)or simulate them (Ardoin & Garand 2003).Other scholars have used electoral returns (e.g.,Kernell 2009). Levendusky et al. (2008) usea Bayesian approach to estimate district-levelpartisanship that builds on the strengths of bothapproaches by combining election returns withdistrict demographics and a variety of otherfactors.

Recently, scholars have been able to re-turn once again to the survey-based approachof Miller & Stokes, taking advantage of sur-veys with much larger sample sizes obtainedthrough random-digit dialing such that thereis a reasonable number of observations ineach district. They use self-identified ideology(Clinton 2008) or create scales out of mul-tiple questions (Park et al. 2004, Bafumi &Herron 2007, Peress 2008; Gelman et al. 2008)in order to characterize state- or district-levelpreferences.

Warshaw & Rodden (2009) attempt to buildon the strengths of these previous studies. Us-ing the rich set of policy questions and reason-ably large samples within districts afforded bythe 2004 Annenberg National Election Study,they employ a Bayesian item response theory

model to estimate individuals’ latent prefer-ences on two issue dimensions—one relatedto economics and another related to moralvalues—and estimate a median for each district,using a Bayesian hierarchical model to addressthe problem of small sample sizes in some dis-tricts. This approach is a marriage of the surveyand demographic approaches and borrows fromthe strengths of each, so that information canbe drawn from the entire distribution of dis-trict preferences to make inferences regardingthe median of each district. Warshaw & Roddenfind that the distributions of both economic andmoral-values medians across districts demon-strate a pronounced left skew, much likethe distribution of presidential vote shares inFigure 2. A similar result is obtained for theone-dimensional partisanship estimates gener-ated by Levendusky et al. (2008). As in theexample of Figure 1b, Warshaw & Rodden’sanalysis suggests that the median voter on theeconomic dimension is slightly to the left of themedian voter in the median district.

IMPLICATIONS FORAMERICAN POLITICS

As empirical researchers provide a clearer mapof political preferences across congressionaldistricts in multiple dimensions, it is possibleto approach some old questions from a newperspective.

Electoral Bias and Responsiveness

If a party can expect to win 50% of the votes butless than 50% of the seats, the gap is defined aselectoral bias. This concept is not easy to mea-sure, however, since tied elections are rarely ob-served, and as Kendall & Stuart (1950) pointedout, a party that wins even slightly more than50% of the votes can expect a much larger ma-jority of the seats.3 Estimation of bias is even

3To understand this “winner’s bonus,” return to Figure 1b

and consider the case where the platforms are set at the me-dians of the second and fourth districts, but a small valenceshock allows one of the parties to win the votes right at thenational median. The winning party would receive a slimmajority of 53% of the votes, but 3/5 (60%) of the seats.


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


more difficult in the United States, where dis-tricts are frequently uncontested and incum-bents rarely face serious challengers. To dealwith these problems, Gelman & King (1991,1994) have developed a Bayesian technique thatanalytically removes the impact of incumbencyadvantage and the existing configuration of in-cumbents and challengers. They demonstratethat although the dramatic pro-Republican biasthat existed in the 1940s and 1950s decreasedsubstantially after the courts became more in-volved in the districting process in the 1960s,much of the apparent disappearance of pro-Republican bias was driven by the fact thatmany safe Democratic incumbents were win-ning in seats that were not being contested byserious Republican challengers (see also Cox& Katz 2002). Yet underneath the surface,Figure 2 underscores the argument ofErikson (1972, 2002) and Rodden & Warshaw(2009) that the asymmetric distribution of par-tisanship across districts generated a structuralpro-Republican bias that manifests itself mostclearly in close elections where neither partyis benefiting from any clear valence shock orasymmetric incumbency advantage.

Yet such an argument sits uncomfortablywith the fact that the Democrats have con-trolled the legislature for roughly three quartersof the sessions since World War II. Rodden &Warshaw (2009) argue that the skewed distri-bution of preferences across districts has a sil-ver lining for Democrats. When the Democratsenjoy a valence shock in their favor, there are alarge number of center-right districts for themto capture in the vicinity of the national me-dian district, whereas a pro-Republican valenceshock (such as Reagan’s popularity) nets fewerseats because too many of the districts on theleft of the national median are in the left tail ofthe distribution that is untouchable for Repub-licans. This allows Democrats to compete onthe opposite side of the national median districtmore effectively than Republicans. A similarlogic might also explain why the vast majority of“split” congressional districts in the postwar pe-riod, even outside the South, and even exclud-ing seats contested by incumbents, have been

won by Republican presidential candidates butDemocratic congressional candidates.

If Chen & Rodden (2009) are correct in as-serting that the asymmetric distribution of par-tisanship across districts is unavoidable whencompact, contiguous districts are drawn in somelarge, industrialized states because of the con-centration of Democrats in cities, this has im-portant implications for debates about redis-tricting. There is a strong push by advocacygroups for apolitical districting procedures thatemphasize compactness, contiguity, and respectfor municipal boundaries, and to the extent thatthe Supreme Court has signaled that it mayeventually be willing to strike down egregiouspartisan gerrymanders, recent decisions betraya fondness among the pivotal justices for com-pact districts. Yet it may be the case that the onlyway to achieve nonbiased districting schemes instates such as Florida, Ohio, and Michigan is todraw relatively noncompact wedge-shaped dis-tricts that combine sections of cities like Miami,Cleveland, and Detroit with their surroundingsuburbs.

Candidate Positioning

The emerging snapshot of the geography ofpreferences across U.S. congressional districtshelps explain some observations from stud-ies examining candidate positioning in con-gressional elections (Ansolabehere et al. 2001,Burden 2004). First, the left skew in the dis-tribution of preferences helps explain whyDemocrats’ party positions are more heteroge-neous than those of Republicans. In order towin, Democrats must cover a wider ideologi-cal range. Moreover, these studies have found agreater correlation between district-level presi-dential vote and candidate positions for Demo-cratic than for Republican candidates. Thismakes sense in light of the theory of primaryconstituencies laid out above: Republicans willfind it difficult to field candidates with compet-itive platforms in leftist districts, whereas this iseasier for Democrats in the right-wing districtsthat are relatively close to the national median.This logic also helps explain the observation

334 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


that the gap between candidate platforms ap-pears to be larger in left-wing districts than inright-wing districts.

Multiple Issue Dimensions

One of the main advantages of survey-basedapproaches to measuring district preferencesis that they make it possible to examine mul-tiple policy dimensions. Warshaw & Rodden(2009), having looked at both an economics-related and a values-related dimension, showthat more than one third of the districts areto the right of the national median on one di-mension but to the left on the other. Moreover,though both distributions are unimodal, thedistribution of moral-values medians across dis-tricts is much wider and more platykurtic thanthat of economic medians. These stylized factsmight help provide a fresh perspective on sev-eral literatures. First, moral-values issues seemto have a greater impact on presidential thancongressional voting patterns. One possibilityis that given the much greater heterogeneity ofpreferences across districts on the moral-valuesdimension, strategic agenda control by partyleaders keeps these potentially divisive issuesoff the congressional agenda. In turn, the lackof roll-call votes on these issues might allowcandidates from both parties greater latitude intailoring their platforms to the district medi-ans. Differences between district-level prefer-ences on two distinct issue dimensions mightalso help explain why districts split their presi-dential and congressional votes.

IMPLICATIONS FORCOMPARATIVE POLITICS

Electoral Regimes and Policy

As mentioned above, it appears that the relativeconcentration of left-wing votes owing to ur-ban concentration was, and in some cases stillis, a fairly widespread phenomenon in indus-trialized countries. At least in the early post-war period, this might help explain why left-wing parties found it difficult to win majorities

in parliamentary countries with single-memberdistricts. It is tempting to take this logic furtherand argue that a bias against socialist and laborparties is responsible for the well-known differ-ence in the extent of redistribution (Iversen &Soskice 2006) and social transfers (Persson &Tabellini 2003) that emerged over the courseof the second half of the twentieth centurybetween countries using proportional represen-tation and those using plurality electoral sys-tems. Yet, as demonstrated in the discussionabove, an asymmetric distribution of partisansacross districts does not necessarily move equi-librium policy away from the preference of themedian voter. An examination of this argu-ment would require data on the distribution ofpreferences across districts. A worthy goal forfuture research is to use demographic and sur-vey data to estimate political preferences acrosselectoral districts in other countries, although adifficult impediment is the frequent use of sam-pling techniques in surveys on political attitudesthat leave large numbers of districts with zeroobservations.4

Perhaps a simpler and more data-friendlyapproach to explaining cross-country differ-ences in redistribution would be to leave be-hind the notion of ideology and focus exclu-sively on the distribution of income, addingsome geography to workhorse models of po-litical economy in the tradition of Meltzer &Richard (1981). In these models, the right skewof the interpersonal income distribution shapesthe incentives of the median voter to adopt a re-distributive tax-transfer scheme. Yet an impor-tant implication of the discussion in this reviewis that individuals are clustered in space suchthat the interpersonal distribution of income islikely quite different from the distribution ofdistrict medians, and the latter is more rele-vant in countries with winner-take-all electoraldistricts.

4A successful attempt to analyze the impact of electoral biason policy choice is provided by Besley & Preston (2007), whouse data from English local governments to show that elec-toral bias allows parties to implement more extreme policiesin favor of their supporters.


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


Specifically, one implication of the urbaneconomics literature on metropolitan real es-tate markets and the political economy of trans-portation cited above is that, within metropoli-tan areas, the poor are likely to live in higherdensity and in conditions of greater incomehomogeneity than the rich. Thus, if pluralityelectoral districts are drawn according to tra-ditional criteria of compactness and contiguity,the poorest voters might end up being highlyclustered within homogeneous low-income dis-tricts, such that the median income of the me-dian district is higher than the overall median.This logic seems most plausible in places wherethe poor live primarily in metropolitan ratherthan rural areas.

Striking evidence to this effect is provided byBradbury & Crain (2005, table 1) for U.S. statelegislatures. Using the 2000 census, they pointout that for both legislative chambers in 43 outof 50 states, the median income of the medianlegislative district is substantially higher thanthe statewide median. The only exceptions aresparsely populated states in the West that lackconcentrated pockets of urban poverty. Fromthese facts, one might hypothesize that by fo-cusing political competition on the median dis-trict rather than the median voter, the imposi-tion of plurality electoral districts in urbanizedsocieties tends to favor the interests of higher-income voters.

If the relevant distribution of income in plu-rality systems is that of the district medians, notonly might the politically relevant median in-come be altered, but also the overall skew ofthe distribution that ultimately drives the tax-transfer system in the Meltzer-Richard model.The long right tail of the typical interpersonalincome distribution cannot help but be shornoff in the distribution of district medians. Thevery rich are few in number, and as a resultthey are likely to end up in relatively hetero-geneous upper-income districts where they aremuch wealthier than the median voter. As dis-tricts become larger and more heterogeneous,e.g., states and provinces used as plurality dis-tricts in upper chambers of federations, the

right skew of district medians can disappear al-together, as the rich and poor alike find them-selves in very heterogeneous districts where theinternal income distribution mirrors the over-all national distribution. Rodden (2009) usescensus data from the United States, Canada,and Australia to show that the right skew ofdistrict median income across lower-chamberdistricts is far less pronounced than that ofthe interpersonal distribution, and the skew ofstate or provincial medians disappears almostcompletely.

It is plausible that by artificially creating amore symmetric income distribution, the im-position of plurality districts undermines thepolitical logic of redistribution, especially whenthe districts are quite large, as in senates. Thesepossibilities merit further theoretical and em-pirical exploration.

In related work, Jusko (2009) argues thatantipoverty policy is a reflection of legislators’incentives to be responsive to low-income vot-ers, which are in turn shaped by the geographicdistribution of the poor across electoral dis-tricts. She defines low-income voters as thosewho comprise the lowest third of the nationalmarket income distribution, and discovers thatthe geography of poverty and the map of districtboundaries coincide such that there is substan-tial cross-national variation in the number ofdistricts that might plausibly be won by a low-income voting bloc.

Using a set of OECD countries, she findsa correlation between the extent of poverty re-lief and the proportion of seats that might bewon by a low-income coalition. One of the at-tractive features of this account is that it ex-plains variations among countries that use plu-rality districts rather than focusing on the bluntdistinction between plurality and proportionalsystems. For instance, Jusko estimates that thepoor are distributed across the massive U.S.House districts such that they can be pivotal inless than one quarter of them. In contrast, thepoor are distributed across France’s relativelysmall districts such that they can be pivotal inalmost half of them.

336 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


Endogenous Party Systemsand Electoral Rules

A central theme of this article is that voters arearrayed in space such that, relative to propor-tional legislative elections or nationwide presi-dential elections, districted plurality legislativeelections can lead to systematic differences inelectoral and policy outcomes. If these im-pacts are sufficiently predictable, party leadersand ideologues should have clear preferencesover institutions. This raises the possibilitythat electoral regimes do not directly causecross-national differences in policies, since theyare themselves outcomes of partisan battlesand bargains. More specifically, it may be thecase that the difference between proportionaland plurality electoral rules at the heart ofsome empirical studies (e.g., Iversen & Sos-kice 2006, Persson & Tabellini 2003) is endoge-nous to battles between parties of the left andright at the moment of constitutional choice.This difference may reflect antecedent struc-tural conditions favoring one group over theother.

Focusing in particular on the era of franchiseexpansion in industrializing countries of theearly twentieth century, Boix (1999) applies theDownsian model, assuming that electoral com-petition under the old plurality systems tookplace within a single national district. In thisstory, the old bourgeois parties (conservativesand liberals) feared that the expansion of thefranchise would allow entry by socialists whowould quickly dominate the left end of the po-litical spectrum, forcing the existing parties tosplit the votes on the right. Faced with a coor-dination problem, the best solution for the par-ties of the right was to advocate proportionalrepresentation.

One of the difficulties of this argument isthat most socialist parties in Europe favoredproportional representation, and there is littledebate about the fact that they and perhaps theirvoters benefited from its introduction. Alesina& Glaeser (2004) go so far as to argue that pro-portional representation is a direct manifesta-tion of revolutionary agitation by leftists.

Once again, it is useful to examine politicalcompetition as taking place in a continuum ofdistricts, and again, it is crucial to understandthe geographic concentration of leftists inmanufacturing and mining areas. Socialistsadvocated proportional representation in largepart because their geographic concentration ina small number of high-population districts ledto dramatic bias in the translation of votes toseats. The socialists did not threaten to cross themajority threshold throughout the country butonly in the proletarian districts. Rodden (2008)applies Callander’s (2005) model of party entryand argues that in most countries (e.g., Belgiumor the Netherlands) this created a coordinationproblem on the left. Socialists entered in theformer strongholds of the liberals, thus raisingthe possibility of socialist-liberal splits thatwould hand districts to minority conservatives.This might help explain the oddity that conser-vatives sometimes pressed for franchise expan-sions over the objections of liberals. Moreover,it explains why in these countries, proportionalrepresentation was largely a response to theself-preservation efforts of liberals.

In other countries, such as Denmark, thecities had been the strongholds of the con-servatives, and entry by the socialists threat-ened to squeeze them out of existence. In thissmaller group of countries with a history ofright-dominated cities, the case for propor-tional representation was made by an odd cross-class coalition of conservatives and socialistswho stood to fare badly in the translation ofvotes to seats under plurality districts. (For arelated argument about electoral bias and in-centives for electoral reform, see Calvo 2009.)

The impact of geography on the origins andevolution of party and electoral systems is a richarea for further research. The anticipated ge-ographic distribution of supporters is often acrucial factor in the strategies of parties at mo-ments of constitutional deliberation over elec-toral institutions (see, e.g., Brady & Mo 1992on Korea). Moreover, combined with models(e.g., Callander 2005) that allow for entry ofnew parties, an understanding of the geography


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


of preferences might help explain basic thingsabout the nature of party systems, such as thedifferent strategic decisions and long-term fatesof liberal parties in New Zealand, Canada, andthe United Kingdom over the course of thetwentieth century. Alternatively, Jusko’s (2009)focus on the geographic distribution of the poormight help explain the conditions under whichleft-wing parties develop platforms that focuson poverty relief.

CONCLUSION

After a period in which economic theory paidlittle attention to distance and space, eco-nomic geography experienced a revival in the1990s, using novel theoretical and empiricaltools to breathe new life into topics covered byMarshall (1920) and Myrdal (1957). Economicactivity is not randomly distributed in space,and economists have developed theories andempirical techniques for understanding variouspatterns of spatial dependence. By building onthese findings and taking advantage of rapidincreases in the availability of geo-coded dataand large surveys, political scientists have theopportunity to breathe new life into the issuesaddressed by Kendall & Stuart (1950), Gudgin& Taylor (1979), and Miller & Stokes (1963).

By developing maps of preferences and partisanproclivities across individuals, precincts, wards,and districts, it might be possible to gain bet-ter answers to old questions about democraticrepresentation: Which parties compete, underwhat rules, with what platforms and what im-plications for policy?

This article has paid special attention tothe impact of urbanization and the industrialrevolution on the distribution of preferencesand partisanship, focusing especially on Britainand its most industrialized former colonies.Attempts to obtain better measurements of po-litical preferences in space are worthwhile indeveloping countries as well. Moreover, someof the issues raised in this review might be quiteimportant in countries where taxation and eco-nomic policy are not the most salient issues.For instance, in much of the Middle East, thereare large differences in political preferences be-tween secular residents of major cities and re-ligious voters in sparse rural areas. In muchof Africa and in the Caribbean, some ethnicgroups are more geographically concentratedthan others, and parties are organized aroundethnicity. As in early-twentieth-century Europeand the contemporary United States, battlesover electoral rules and districting proceduresin such settings are highly consequential.

DISCLOSURE STATEMENT

The author is not aware of any affiliations, memberships, funding, or financial holdings that mightbe perceived as affecting the objectivity of this review.

LITERATURE CITED

Alesina A, Glaeser EL. 2004. Fighting Poverty in the U.S. and Europe: A World of Difference. Oxford, UK: OxfordUniv. Press

Ansolabehere S, Leblanc W, Snyder JM. 2005. When parties are not teams: platform choice in single member districtand proportional representation systems. Work. pap., Dep. Polit. Sci., Mass. Inst. Technol.

Ansolabehere S, Rodden J, Snyder JM. 2006. Purple America. J. Econ. Persp. 20(2):97–118Ansolabehere S Snyder JM, Stewart C. 2001. Candidate positioning in U.S. House elections. Am. J. Polit. Sci.

45(1):136–59Ardoin P, Garand J. 2003. Measuring constituency ideology in U.S. House districts: a top-down simulation

approach. J. Polit. 65(4):1165–89Austen-Smith D. 1984. Two-party competition with many constituencies. Math. Soc. Sci. 7:177–98Bafumi J, Herron M. 2007. Preference aggregation, representation, and elected American political institutions. Work.

pap., Dep. Polit. Sci., Dartmouth Univ.

338 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


Bagnasco A. 1977. Tre Italie: La Problematica Territorial Dello Sviluppo. Bologna: MulinoBaron D, Ferejohn J. 1989. Bargaining in legislatures. Am. Polit. Sci. Rev. 83(4):1181–1206Besley T, Preston I. 2007. Electoral bias and policy choice: theory and evidence. Q. J. Econ. 122(4):1473–1510Boix C. 1999. Setting the rules of the game: the choice of electoral systems in advanced democracies.

Am. Polit. Sci. Rev. 93(3):609–24Bradbury JC, Crain WM. 2005. Legislative district configurations and fiscal policy in American states. Public

Choice 125:385–407Brady D, Mo J. 1992. Electoral systems and institutional choice: a case study of the 1988 Korean elections.

Comp. Polit. Stud. 24(4):405–29Brueckner J, Thisse JF, Zenou Y. 1999. Why is central Paris rich and downtown Detroit poor? An amenity-

based theory. Eur. Econ. Rev. 43(1):91–107Burden B. 2004. Candidate positioning in U.S. congressional elections. Brit. J. Polit. Sci. 34:211–27Callander S. 2005. Electoral competition in heterogeneous districts. J. Polit. Econ. 113(5):1116–45Calvo E. 2009. The competitive road to proportional representation: partisan biases and electoral regime

change under increasing party competition. World Polit. 61(2):254–95Chen J, Rodden J. 2009. Tobler’s law, urbanization, and electoral bias: why compact, contiguous districts are bad for

the Democrats. Work. pap., Dep. Polit. Sci., Stanford Univ.Cho WKT, Gimpel JG, Dyck JJ. 2006. Residential concentration, political socialization, and voter turnout.

J. Polit. 68(1):156–67Clinton J. 2008. Representation in Congress: constituents and roll calls in the 106th House. J. Polit. 68(2):397–

409Coate S, Knight B. 2007. Socially optimal districting: a theoretical and empirical exploration. Q. J. Econ.

122(4):1409–71Cox GW, Katz JN. 2002. Elbridge Gerry’s Salamander: The Electoral Consequences of the Reapportionment Revolu-

tion. New York and Cambridge: Cambridge Univ. PressDalton RJ. 2006. Citizens Politics: Public Opinion and Political Parties in Advanced Industrial Democracies.

Washington, DC: CQ PressDe la O A, Rodden J. 2008. Does religion distract the poor? Income and issue voting around the world. Comp.

Polit. Stud. 41(4–5):437–76Diermeier D, Feddersen TJ. 1998. Cohesion in legislatures and the vote of confidence procedure. Am. Polit.

Sci. Rev. 92(3):611–21Downs A. 1957. An economic theory of political action in a democracy. J. Polit. Econ. 65(2):135–50Erikson R. 1972. Malapportionment, gerrymandering, and party fortunes in Congressional elections.

Am. Polit. Sci. Rev. 66(4):1234–45Erikson R. 2002. Sources of partisan bias in U.S. Congressional elections: an update stimulated by Ron

Johnston’s essay. Polit. Geog. 21:49–54Eyster E, Kittsteiner T. 2007. Party platforms in electoral competition with heterogeneous constituencies.

Theor. Econ. 2:41–70Friedman J, Holden R. 2008. Optimal gerrymandering: sometimes pack, but never crack. Am. Econ. Rev.

98(1):113–44Gelman A, King G. 1991. Systemic consequences of incumbency advantage in U.S. House elections. Am. J.

Polit. Sci. 35(1):110–38Gelman A, King G. 1994. A unified method of evaluating electoral systems and redistricting plans. Am. J.

Polit. Sci. 38(2):514–54Gelman A, Park D, Shor B, Bafumi J, Cortina J. 2008. Red State, Blue State, Rich State, Poor State: Why Americans

Vote the Way they Do. Princeton and Oxford: Princeton Univ. PressGilligan T, Matsusaka J. 2006. Public choice principles of redistricting. Public Choice 129:381–98Glaeser EL, Gyourko J, Saks R. 2005. Urban decline and durable housing. J. Polit. Econ. 113(2):345–75Glaeser EL, Kahn M, Rappaport J. 2007. Why do the poor live in cities? The role of public transportation.

J. Urban Econ. 63(1):1–24Glaeser EL, Kolko J, Saiz A. 2001. Consumer city. J. Econ. Geog. 1:27–50Gudgin G, Taylor PJ. 1979. Seats, Votes, and the Spatial Organisation of Elections. London: Pion


R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.


Hinich MJ, Ordeshook P. 1974. The electoral college: a spatial analysis. Polit. Methodol. 1:1–30Hohenberg P. 2004. The historical geography of European cities: an interpretive essay. In Handbook of Regional

and Urban Economics, 4: 150–89. Amsterdam: ElsevierHotelling H. 1929. Stability in competition. Econ. J. 39(153):41–57Iversen T, Soskice D. 2006. Electoral institutions and the politics of coalitions: why some democracies redis-

tribute more than others. Am. Polit. Sci. Rev. 100(2):165–81Johnston RJ. 1976. Spatial structure, plurality systems and electoral bias. Can. Geog. 20:310–28Johnston RJ. 1992. A Question of Place: Exploring the Practice of Human Geography. Oxford, UK: BlackwellJusko KL. 2009. The electoral foundations of poverty relief in contemporary democratic societies. Work. pap., Dep.

Polit. Sci., Stanford Univ.Kendall MG, Stuart A. 1950. The law of the cubic proportion in election results. Brit. J. Sociol. 1(3):183–96Kernell G. 2009. Giving order to districts: estimating voter distributions with national election returns.

Polit. Anal. 17(3):215–35Key VO. 1949. Southern Politics in State and Nation. Knoxville: Univ. Tenn. PressKrugman P. 1991. Geography and Trade. Cambridge, MA: MIT PressLeblanc WM. 2007. Party positions and the seats/votes relationship with ideological voters. PhD thesis, Dep. Polit.

Sci., Mass. Inst. Technol. 137 pp.Levendusky MS, Pope JC, Jackman SD. 2008. Measuring district-level partisanship with implications for the

analysis of U.S. elections. J. Polit. 70(3):736–53Lipset SM, Rokkan S. 1967. Party Systems and Voter Alignments: Cross-National Perspectives. New York:

Free PressMarshall A. 1920. Principles of Economics. London: MacmillanMayhew D. 1974. Congress: The Electoral Connection. New Haven, CT: Yale Univ. PressMeltzer A, Richard S. 1981. A rational theory of the size of government. J. Polit. Econ. 89(5):914–27Mieszkowski P, Mills E. 1993. The causes of metropolitan suburbanization. J. Econ. Perspect. 7(3):135–47Miller WE, Stokes DE. 1963. Constituency influence in Congress. Am. Polit. Sci. Rev. 57(1):45–56Myrdal G. 1957. Economic Theory and Underdeveloped Regions. London: DuckworthNas A, Arnott R, Small K. 1998. Urban spatial structure. J. Econ. Lit. 36:1426–64Palfrey TR. 1984. Spatial equilibrium with entry. Rev. Econ. Stud. 51(1):139–56Park D, Gelman A, Bafumi J. 2004. Bayesian multilevel estimation with poststratification: state-level estimates

from national polls. Polit. Anal. 12:375–85Peress M. 2008. Candidate positioning and responsiveness to constituent opinion in the U.S. House of Representatives.

Work. pap., Dep. Polit. Sci., Univ. RochesterPersson T, Tabellini G. 2003. The Economic Effects of Constitutions. Cambridge, MA: MIT PressPool IS, Abelson RP, Popkin SL. 1965. Candidates, Issues, and Strategies: A Computer Simulation of the 1960 and

1964 Presidential Elections. Cambridge, MA: MIT PressRodden J. 2008. Why did Western Europe adopt proportional representation? A political geography explanation.

Presented at Annu. Meet. Am. Polit. Sci. Assoc., 104th, BostonRodden J. 2009. Federalism and inter-regional redistribution. Work. Pap. 2009/3, Barcelona Inst. Econ.Rodden J, Warshaw C. 2009. Why the Democrats need boll weevils and blue dogs: the distribution of political preferences

across U.S. House districts. Work. Pap., Dep. Polit. Sci., Stanford Univ.Romer T. 1975. Individual welfare, majority voting, and the properties of a linear income tax. J. Publ. Econ.

4:163–85Rosenthal S, Strange W. 2004. Evidence on the nature and sources of agglomeration economies. In Handbook

of Urban and Regional Economics, 4:312–32. Amsterdam: ElsevierRydon J. 1957. The relation of votes to seats in elections for the Australian House of Representatives 1949–54.

Polit. Sci. 9:49–65Snyder JM. 1994. Safe seats, marginal seats, and party platforms: the logic of platform differentiation. Econ.

Polit. 6(3):201–24Taylor PJ, Johnston RJ. 1979. Geography of Elections. London: Croom HelmTobler W. 1970. A computer movie simulating urban growth in the Detroit region. Econ. Geog. 46:234–40Warshaw C, Rodden J. 2009. Measuring district-level economic and moral ideology. Work. pap., Dep. Polit. Sci.,

Stanford Univ.

340 Rodden

R

E V I EW

S

IN

AD V A N

CE


Ann

u. R

ev. P

olit.

Sci

. 201

0.13

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

Sta

nfor

d U

nive

rsity

- M

ain

Cam

pus

- R

ober

t Cro

wn

Law

Lib

rary

on

04/2

1/10

. For

per

sona

l use

onl

y.

the geographic distribution of political...

Documents