07 boix - setting the rules of the game the choice of electoral systems in advanced democracies

8/7/2019 07 Boix - Setting the Rules of the Game The Choice of Electoral Systems in Advanced Democracies

1/17

Setting the Rules of the Game: The Choice of Electoral Systems in Advanced Democracies

Author(s): Carles BoixSource: The American Political Science Review, Vol. 93, No. 3 (Sep., 1999), pp. 609-624Published by: American Political Science AssociationStable URL: http://www.jstor.org/stable/2585577

Accessed: 29/10/2009 16:47

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/action/showPublisher?publisherCode=apsa.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

American Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to

The American Political Science Review.

http://www.jstor.org
http://www.jstor.org/stable/2585577?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/2585577?origin=JSTOR-pdf


2/17

American Political Science Review Vol. 93, No. 3 September 1999

SettingheRules ftheGame: heChoice f ElectoralystemsnAdvanced emocraciesCARLES BOIX University of Chicago

ookingat the history f democraciesn the developedworld, showthatelectoral ystems eriveromthe decisions the ruling parties make to maximize their representationaccording to the followingconditions. As long as the electoral arena does not change and the currentelectoral regime benefitsthe rulingparties, the electoral systemis not altered.As the electoralarenachanges (due to the entryof newvoters or a change in voters'preferences), the rulingparties modify the electoralsystem, dependingon theemergenceof new parties and the coordinatingcapacities of the old parties. Whenthe new partiesarestrong,the old parties shift from plurality/majority o proportional representation f no old party enjoys a dominantposition, but they do not do this if there is a dominant old party. Hhen new entrants are weak, a system ofnonproportionalrepresentations maintained, regardlessof the structureof the old party system.

T he literature on the interaction between elec-toral rules and the performance of the politicalsystem is imbalanced. Political scientists havefocused on determining the effects of electoral laws onboth political stability (Hermens 1941; Lijphart 1994)and voting behavior and party systems (Duverger 1954;Rae 1971; Taagapera and Shugart 1989). By contrast,little effort has been devoted to what causes the highdegree of cross-national variation in electoral laws,with the exception of two seminal (yet still theoreticallyunderdeveloped) contributions by Rokkan (1970) andRogowski (1987).1 To fill the gap, this article maps theconditions under which the ruling parties, anticipatingthe effects of different electoral regimes on voters andcandidates, choose different sets of electoral rules to

maximize their chances of securing parliamentary rep-resentation as well as cabinet posts.Electoral rules are formal institutions that encour-age the strategic behavior of both elites and voters andhence force their coordination around a set of viablecandidates. To avoid wasting their ballot on hopelesscandidates, voters may choose politicians who areranked second or lower in their preference ordering.Similarly, elites tend to pay attention to and concen-trate resources on candidates who are expected to win.The extent of strategic behavior among voters andelites varies with the constraining effects of electoralrules. Generally speaking, the higher the entry barrier(or threshold) set by the electoral law, the moreextensive strategic behavior will be.Anticipating the coordinating consequences of elec-Cares Boix is Assistant Professor of Political Science, University ofChicago, Chicago, IL 60637.This research was supported by grants from the Centre de Recercaen Economia Internacional (Universitat Pompeu Fabra, Barcelona)and The Ohio State University. For helpful comments, I am gratefulto Alicia Adsera, John Carey,Stathis Kalyvas, Herbert Kitschelt, BillLiddle, Jos6 Maria Maravall, Jos6 Ram6n Montero, GuillermoO'Donnell, Ronald Rogowski, Ignacio Sdnchez-Cuenca, David Sos-kice, Susan Stokes, seminar participants at the 1997 annual meetingof the American Political Science Association, Fundaci6n JuanMarch, Universitat Autbnoma de Barcelona and University ofChicago, and several anonymous reviewers.1 For a review of previous attempts to determine the causal mecha-nisms of the origins of electoral rules, see Lijphart (1985, 1992) and,to some extent, Cox (1997, 15-6).

toral rules, any current government (provided it has themonopoly over electoral rulemaking) shapes the elec-toral rules to its advantage. Two results follow. As longas the electoral situation does not change substantiallyand the current rules serve the ruling parties well, thegovernment has no incentives to modify the electoralregime. As soon as the electoral arena changes, how-ever, the government considers altering the electoralsystem. If it calculates that the strategic behavior ofvoters will not upset its dominant position, it willmaintain (or introduce) high entry barriers (i.e., aplurality rule). On the contrary, if it foresees that, byinducing any strategic behavior (among voters andelites), the current rules will erode its parliamentarypower substantially, it will change them (loweringthresholds or entry barriers) to increase the degree ofproportionality.As shown in this article, these sets of calculationsexplain the development of electoral systems in thedeveloped world since the turn of the century. Theelectoral system (structured around pluralityor major-ity rules) remained unchanged during the era of limitedsuffrage. As soon as universal suffrage was adopted,which led to the massive entry of mostly left-wingvoters and, hence, to a radically new electoral arena,the ruling elites followed different solutions. The plu-rality/majority system survived under two circum-stances. First, it remained in place in those countries inwhich the new entrant (a socialist party) was weak and,itself the victim of strategic voting, could not challengeany of the established parties. Second, it was main-tained in those countries in which, although the newentrant became strong, one of the established ornonsocialist parties retained a dominant position in thenonsocialist camp. Since it could easily attract thestrategic vote of all nonsocialist voters (mostly worriedabout blocking the victory of social democracy), thedominant party acted rationally in maintaining a highlyconstraining electoral rule. By contrast, proportionalrepresentation was adopted in those countries in whichthe socialist party was strong and nonsocialist partiescontrolled roughly similar shares of the electorate.Failure to reduce the electoral threshold would haveled to an overwhelming victory of the socialist party.Assoon as the-electoral arena became stable and the party

609


3/17

The Choice of Electoral Systems in Advanced Democracies September 1999system froze along certain cleavages, policymakers lostinterest in modifying the electoral regime. Abruptchanges in electoral laws have been rare in the lasteight decades, with the exception of those nations inwhich party systems have remained unsettled.In addition to shedding light on the strategic calcu-lations of political elites, this article tests the validity ofseveral competing explanations. A high degree ofethnic or religious fragmentation is shown to encour-age, under certain conditions, the adoption of propor-tional representation. The claim that states chooseproportional representation to maximize social wel-fare, because it generates efficiency gains by preventingrent-seeking and securing free trade, is found to beunconvincing.

A THEORY OF THE SELECTION OFELECTORAL SYSTEMSThe Rokkan HypothesisIn his discussion of the adoption of proportionalrepresentation (PR) in European countries at the turnof the century, Rokkan offers a first interpretation ofthe causal forces that determine the selection of dif-ferent electoral regimes. PR rules were introduced"through a convergence of pressures from below andfrom above. The rising working class wanted to gainaccess to the legislatures, and the most threatened ofthe old-established parties demanded PR to protecttheir position against the new waves of mobilizedvoters created by universal suffrage" (Rokkan 1970,- 157).2Although Rokkan rightly points to the key roleplayed in the design of electoral rules by the calcula-tions that political elites make about their futureelectoral. strength, his argument is underspecified.Other than stating that political elites adopt PR toprotect their political stakes, he does not indicate theconditions under which policymakers will feel suffi-ciently threatened to change the current electoralsystem. As a result, Rokkan's hypothesis runs intoseveral problems. First, if PR is adopted to lessen thechances that an increasingly stronger socialist partymay win an absolute parliamentary majority, then whydid Great Britain (as well as Australia and NewZealand) embrace universal suffrage without shifting toPR (as did Sweden and Denmark)?3 Second, if elec-toral rules were changed as a result of the extension ofuniversal suffrage in the late nineteenth and earlytwentieth century, then why have certain countries(France, Greece, and to some extent Spain) shiftedback and forth between plurality (or majority) rule andPR systems over the last century? Third, and more2 According to Lijphart (1992, 208-9), the extension of universalsuffrage forced both the ruling elites and their challengers tointroduce PR "to protect their [respective] interests"; the formertried to minimize their (predictable) losses and the latter wished to"guarantee that they would gain at least a substantial share ofrepresentation and political power."3A similar point is made by Lijphart (1992, 209).

610

generally, Rokkan's explanation is too historicallybounded. The rise of socialism and the correspondingcalculations of conservative elites did affect the formthat electoral rules took under fully democratic re-gimes. Still, we need a more encompassing theory toexplain the selection of electoral rules in EasternEurope, Latin America, and the newly democratizingcountries today.4An Analytical Generalization of the RokkanHypothesisTo understand why the ruling parties shifted (or not) toPR at the turn of the century in the developed world, Iproceed to develop an argument organized aroundthree sequential steps.1. The consequences of electoral rules.These rules areconstraining devices that, by encouraging strategic be-havior among voters and elites, force the coordinationof resources and ballots on a reduced set of candidates.Instrumentally rational voters eschew voting for candi-dates they expect to do poorly in the next election, evenif this means supporting second-ranked candidates intheir preference orderings. Similarly, political elitesavoid wasting their time and resources on hopelesscandidates. As a result of these two processes, and aslong as everyone agrees on each candidate'sprobabilityof winning, votes and resources flow to "stronger"candidates. In equilibrium, through either the instan-taneous coordination of all political actors or thewinnowing out of "weak" candidates over repeatedelections, only a certain number of viable candidatescompete at the ballot box.The extent of strategic behavior, as well as thenumber of candidates willing and able to stand inelections, varies with the electoral rule in place. Insingle-member plurality systems voters coordinate,given a set of conditions discussed in detail below(when I define the mechanisms that lead to the choiceof the electoral law), around two candidates (Cox 1997,69-79; Duverger 1954, 217). Strategic voting declines,however, as the proportionality of the electoral systemincreases. Because seats can be gained with only afraction of the total vote, voters have fewer incentivesto abandon their most preferred candidates. Accord-ingly, the number of viable candidates increases withPR.54 To explain the constitutional choices (primarily regarding theelectoral system) made in Czechoslovakia, Hungary, and Poland,Lijphart (1992) refines Rokkan's hypothesis by adding three factors:(1) the degree to which "the 'old-established parties' [are] retainingsufficient power and legitimacy to negotiate a relatively favorablecompromise" (p. 213); (2) the expectations that ruling parties haveabout their future electoral chances; and (3) the extent to whichvoters distrust partylists and would rather vote for candidates. It isunclear how Lijphart's first and third conditions can apply to theadoption of PR in either Europe at the beginning of this century ornon-European nations in general. The second condition comes closerto the model developed below, but it does not specifywhat structuredthe politicians' expectations and how those expectations led to theselection of different electoral systems.5 Cox (1997, 103-21) provides empirical evidence that strategicvoting diminishes substantiallyin large (four or more seats) districts.


4/17

American Political Science Review Vol. 93, No. 32. The calculations of rulers and the stability of theelectoral arena. Since electoral laws are determined bypolicymakers, we should expect that the ruling politicalparties, anticipating the (varying) effects of differentelectoral regimes, choose the regime that maximizestheir chances of staying in power.6Two points follow.

As long as the electoral arena does not change sub-stantially and the electoral rules serve them well, thegoverning parties have no incentives to change theelectoral system.7 As soon as change takes place andthe previous structureof partisan competition starts tounravel, the ruling parties consider modifying theelectoral system to maintain their political advantage.Electoral systems were relatively stable throughoutthe nineteenth century. With the introduction of uni-versal suffrageat the turn of the century, the conditionsunder which political competition had developedchanged dramatically in the developed world. In ashort period, the size of the electorate increased sev-eral times. In countries such as Belgium, Sweden, andItaly, it rose from about one-tenth of the male popu-lation at the end of the nineteenth century to universalmale suffrage after World War I. Even in countries inwhich universal or nearly universal suffrage had beenintroduced earlier, such as Denmark, France, andNorway, urbanization and industrialization led to asubstantialpolitical realignment, that is, to a shift froma rural-urban conflict to a deepening capital-laborcleavage. The massive entry of new voters as well as atransformation in the preferences of already enfran-chised citizens threatened the electoral strength of theold parties substantially. Accordingly, the old parties'elites had a strong incentive to reshape the electoralrules of the game.83. The reformof the electoralsystem as a function ofthe viabilityof the oldparty system.The shift to PR is notan automatic response to a changing electoral arena.The extent to which the ruling parties embrace PRdepends on the interaction of two main conditions,which determine the electoral viability of the rulingparties in the future. First, the strength of the newentering parties (the socialist party at the turn of thecentury); second, the coordinating capacity of the6 Throughout this article I assume that the parties in government(i.e., those with a parliamentary majority) have a monopoly overelectoral norms. As should become apparent from the discussionbelow, if changing electoral norms require a supramajority, theincentives to adopt PR are likely to increase.7 If two main parties are contending for government, then they willbe satisfied with the current electoral arrangements as long as theirchances of attaining power are even. The longer one of the partiesstays out of office (i.e., its probabilityof winning declines below 0.5),the greater is the likelihood that it will press for electoral reform.This would explain the growing popularity of PR within BritishLabour by the late 1980s and early 1990s.8 The introduction of universal suffrage is taken as given in thisarticle, but the ruling parties could block any change in the electoralarena. Whether they choose to do so (a decision not modeled in thisarticle) depends on how feasible it is to maintain the status quo.Whenever the ruling policymakers calculated that an authoritarianstrategywas not rational (i.e., if it should lead to a bloody uprisingoreven civil war, then generating losses greater than full participation),their only solution lay in reshaping the rules of the game (includingthe electoral system) to maintain a certain political or institutionaladvantage.

ruling parties, that is, whether they are tied in votes(which prevents coordination) or one party is dominant(and thus can become the focal point around whichnonsocialist voters coordinate). If there are no newparties and/or one of the old parties leads in thenonsocialist camp, then a non-PR system will surviveany changes in the electoral market (such as theintroduction of universal suffrage). Yet, as soon as anew party draws substantial support and the rulingparties are tied in votes, the incentives to embrace PRbecome irresistible.For this third step in the argument, there are twopoints of departure: the single-member plurality sys-tem and the single-member dual-ballot system. Beforethe adoption of universal suffrage, all elections wereconducted under non-PR rules, mainly using the sin-gle-member plurality system but also applying thesingle-member dual-ballot systems in some countries.9Employing the single-member plurality system as thebenchmarkcase, I will examine first how the conditionsspecified above shaped the decisions of the governingparty (parties) at the beginning of the twentieth cen-tury. I will then extend my discussion to the two-roundsystem.Consider, to startwith, the situation in which the oldparties face a strong new party. Two scenarios maydevelop. In the first, the old parties coexist in anon-Duvergerian equilibrium. Either their electoralstrength is balanced, and hence voters cannot deter-mine around which one they should coordinate todefeat the socialist party, or voters have such intensepreferences that they deliberately eschew strategicvoting.10 Panels A and B in Figure 1 represent thissituation. The electorate is uniformly distributed on asingle policy dimension, from Left (0) to Right (1).Before the introduction of universal suffrage,only halfthe population has the right to vote-voters withpreferences from 0.5 to 1. There are two parties,Liberal (L1) and Conservative (C1), symmetricallypositioned around the old median voter (mi = 0.75)at, say, 0.65 and 0.85, respectively, and thus they win50% of the vote each (Figure 1A). After universalsuffrage is introduced, the median voter is m' = 0.5.The Socialist Party (S) enters the electoral processannouncing a position 0.35 + , (Figure 1B). Under asingle-member plurality system, this is enough tosnatch the district from the Conservatives becausevoters, unable to determine which nonsocialist partyhas a better chance to defeat S, cannot coordinate oneither L1 or C1. Anticipating a crushing victory of S,9 The single-member plurality system was in place in Australia,Canada, Denmark, Iceland, Ireland (before independence), Japan,New Zealand, Spain, Sweden, the United Kingdom, and the UnitedStates. Austria (1907-19), France, Germany, Italy, the Netherlands(in part), Norway (since 1905), and Switzerland had a single-membersystem with two rounds (three rounds in Switzerland until 1900).Belgium, Luxembourg, and the urban districts of the Netherlandsoperated under multimember districts and two rounds.10See Cox (1997, 76-80, 96-8) for the full set of (rather strict)assumptions under which a single-member plurality system leads toonly two viable candidates. I focus on the conditions that shed lighton the calculations the ruling parties make about which electoral ruleshould be adopted.

611


5/17

The Choice of Electoral Systems in Advanced Democracies September 1999

FIGURE 1. The Emergence of a StrongSocialist PartyA. The Old Party System

Li C,l Il Il I0 0.5 MO 1

{ enfranchised voters }

B. A Strong Socialist Party ConfrontingEqually Balanced Nonsocialist PartiesS Li C,

0 1new electorate } { old electorate }

C. A Strong Socialist Party Confrontinga Dominant Nonsocialist Party

S L2 C2l ~ ~~~~Il ~ ~~~~Il ~ ~~~~I0 1

new electorate } { old electorate }

the old ruling parties introduce PR.1" This was the casewith the Danish and Swedish nonsocialist parties.12Even if S is strong, the incentives to change theelectoral system may remain low. In this second sce-nario, represented in panel C of Figure 1, one of theold ruling parties has a dominant position in theelectoral arena. Whereas the Conservative Party (C2)is still at 0.85, the Liberal Party (L2) positions itselfdifferently (0.75, that is, the old median voter, mi)from before. As a result, L2 has a dominant positionunder the old system, polling about 55% of the votes.Even if, after the whole electorate is enfranchised, Semerges and adopts a position equal to 0.35, whichshould lead to an easy victory under plurality rule, theincentives of L2 to shift to PR remain low. Sincesingle-member plurality systems encourage strategicbehavior, L2 rationally expects all former Conservativevoters to coordinate around L2 to defeat S. Once C211Moreover, since by shifting to PR they make easier the entry ofany party to the left of S, they force S to slow any movement towardthe median voter.12 This situation also fits the case of several nations, such as Belgiumand the Netherlands, in which nonsocialist coordination could notoccur (even if one party dominated, as did the Belgian CatholicParty) in the face of intensely committed voters.

612

FIGURE 2. The Emergence of a WeakSocialist PartyA. Confronting Equally Balanced Nonsocialist Parties

S LI C,

I ~~I II0 mn m0new electorate } { old electorate }

B. Confronting a Dominant Nonsocialist Party

S L2 C2

0 mn m0


disappears, L2 can move toward the new median voter(m') to regain control of the parliament. Although inthe reverse way, with the Tories the dominant force,this is what happened in Britain. After the Liberalssplit during World War I, the Conservative Party hadno incentives to shift to PR to minimize the chances ofa Labour victory.13Consider also the case in which the new electors donot vote for new parties and, in the developed world atthe turn of the century, socialist parties are weak. Theruling old parties have no incentives to abandon theplurality system, regardless of whether they are lockedinto a non-Duvergerian equilibrium. In Figure 2A, L1and C1 have a roughly similar share of votes. In Figure2B, C2 enjoys a dominant position. Yet, in neither casewill voters desert either party to stop S. In fact, it is Sthat is most damaged by strategic voting. This patternfits the cases of Canada and the United States.14Now let us examine the single-member dual-ballotsystem as the point of departure. In a two-roundsystem, in which the coordination around two candi-dates is postponed to the second ballot, the incentives13 Labour made similar calculations once it became a dominant partyin the left-hand side of the policy space. Until 1921 a majority ofLabour MPs favored PR. By 1923-24, once Labour had solidified itslead over the Liberals in the polls, an overwhelming majority ofLabour parliamentarians were against PR (Butler 1963).14 The pace at which universal suffrage was extended also may haveaffected the selection of electoral rules. In those countries in whichsuffrage was extended incrementally overthe nineteenth century, theold parties were likely to control most of the electoral arena by thetime universal suffrage was introduced, and therefore their incentiveto shift to PR must have been low. By contrast, in countries in whichthe electorate increased sharplyin a short period, the old parties hadto compensate their rapidly weakened position with the introductionof a very low threshold. Yet, this explanation does not fare well forGermanyand Switzerland,where PR came much later than universalmale suffrage.


6/17

American Political Science Review Vol. 93, No. 3

FIGURE 3. A Three-Party System in theNonsocialist Camp under a Two-RoundSystemS L C Ml l l Il l l I~~~I I I I0 mn m0 1


to shift to PR are rather similar to those under thesingle-member plurality system.If the new party is weak, the ruling elite will not shiftto PR. Regardless of whether all the candidates in thefirst round or only a limited number go into the runoffelection, in the second round the first-round socialistvoters will be very likely to vote for their second-rankedcandidate.If the Socialist Party is strong and there is onedominant nonsocialist party, then the latter has noincentives to abandon the dual-ballot arrangement. Infact, the incentives to shift to PR are even lower thanunder a plurality system. Either because only the twotop parties are legally entitled to advance to the runoffor because voters now know with absolute certainty forwhich nonsocialist party they should rationally vote, thedominant nonsocialist force can become the only non-socialist party in parliament without having to risklosing an election. By contrast, in a plurality system,unless all the nonsocialist voters shift automatically tothe party with higher electoral chances, the nonsocialistdominant party may have to endure one electoraldefeat before getting all the right-wing votes to beat theSocialist Party.Finally, if the Socialist Party is strong and the partiesof the Right are similar in strength, then they havestrong incentives to shift to PR. Consider the examplein Figure 1B. In the first round, both L 1 and C1 will get25% of the vote each, and S will get 50% (i.e., one voteshort of the absolute majority). Although S may still bebeaten in the second round, one of the old parties willlose all parliamentary representation.15 More precisely,under a dual-ballot system and fully balanced nonso-cialist forces, the odds that the old governing party willdisappear are one-half. Under these conditions, it takesonly a very slightly risk-averse party to opt for PR.The decision to shift to PR is even stronger if thenonsocialist camp is more fragmented. Figure 3 repre-sents a case in which three nonsocialist parties sharethe old electorate, which historically is a rather realistic15 This outcome will take place either because legally only the twotop candidates can advance to the runoff or because all the nonso-cialist voters coalesce around one party. Notice that a legal limit onthe number of candidates allowed into the second round makes itmuch easier to stop S. The legal constraint eases the coordinationprocess of all nonsocialist voters around a single candidate. If thereis no legal limit, then neither L1 nor C1 has any incentive to leave therace, and voters will continue to have a difficult time knowing whatparty they should choose.

situation.16M, the Monarchist Party, is positioned at0.95. C and L are positioned at 0.8 and 0.6, respec-tively. Before the extension of suffrage to all thepopulation, M polls 25% of the votes, C gets 35%, andL has the remaining 40%. Universal suffrage halveseach old party's share of the vote. More important, itthreatens to obliterate both M and C, since L auto-matically becomes the only credible alternative to thenonsocialist party in the second round. Given thateither C or M or both are in government (in coalitionwith L or forming a Conservative-Monarchistcabinet),any deal to introduce universal suffrage will be linkedto a shift to PR.Trade and Proportional RepresentationTaking a rather different approach, Rogowski (1987,204) concludes that "the more an economically ad-vanced state relies on external trade, the more it will bedrawn to the use of PR, a parliamentary system, andlarge districts."The adoption of PR by a trade-depen-dent economy is desirable for two reasons. On the onehand, by inducing the formation of a strong partysystem that integrates and restrainsparticularinterests,PR insulates the state from protectionist interests andenhances its autonomy from rent-seeking groups. Onthe other hand, in interaction with societal forces, PRinduces political and policy stability.As a result, openeconomies are drawn o PR, eitherbytheirown consciouschoice (PR is seen a priori as advantageous)or by thefunctionalrequirements mposedby trade (countriesthatdo not adopt PR will collapse or perish in the long run).Electoral Rules as Generators of PoliticalStabilityIn response to research insisting on the beneficialeffects of plurality rule for governmental responsive-ness and political stability (Hermens 1941; see alsoDowns 1957), a more recent literature emphasizes thatPR constitutes the most adequate system to govern(and therefore will be adopted by) any society with ahigh degree of political segmentation (Lijphart 1977).As noted by Rokkan (1970, 157), "it was no accidentthat the earliest moves toward proportional represen-tation came in the ethnically most heterogeneous Eu-ropean countries.... In linguistically and religiouslydivided societies majority elections could clearlythreaten the continued existence of the political sys-tem. The introduction of some element of minorityrepresentation came to be seen as an essential step ina strategy of territorial consolidation" (1970, 157). Inhomogeneous polities, by contrast, plurality rule canremain in place safely. Since its coordinating effects donot entail the suppression of minority representation, it16 At the turn of the century, the average number of nonsocialistparties in dual-ballot systems was slightly above three. Cox (1997,chap. 6) formally shows that, in equilibrium, the number of viableparties in the single-member dual-ballot system is M + 1, where Mis the number of parties allowed to advance to the runoff. MostEuropean dual-ballot systems allowed only the two top candidates tocompete in the second round.

613


7/17

The Choice of ElectoralSystems n AdvancedDemocracies September1999does not jeopardize the basis of civil peace. Althoughthis explanation carries some weight, later I show thatthe presence of minorities only leads to PR conditionalon both their geographical distribution and the extentto which other mechanisms of representation (such asfederalism) are (not) employed.

THE MEASUREMENT OF ELECTORALSYSTEMSI examine the evolution of electoral laws in the uni-verse of developed countries over the period 1875-1990: Australia, Austria, Belgium, Canada, Denmark,Finland, France, Germany, Greece, Iceland, Ireland,Italy, Japan, Luxembourg, the Netherlands, New Zea-land, Norway, Portugal, Spain, Sweden, Switzerland,the United Kingdom and the United States.

The Dependent Variable: The EffectiveElectoral ThresholdTo explain what caused many (but not all) countries toshift awayfrom pluralityor majorityrule at the turn ofthe century, that is, to determine the extent to whichthe structure of electoral competition, the presence ofminorities and the openness of the economy shapedthe electoral system, we need an appropriate measureof proportional representation. For this purpose, I usethe concept of EffectiveElectoral Threshold-the pro-portion of votes that, for each electoral system, securesparliamentaryrepresentation to any party with a prob-ability of at least 50% (Lijphart 1994; Taagapera andShugart 1989).17The effective electoral threshold hastwo advantages. In abstractingfrom the complexity ofelectoral systems, which are the composite of differentrules regulating the access of citizens to suffrage, thenumber and use of votes by voters, the number and sizeof electoral districts,the introduction of thresholds andbonuses, and the allocation mechanisms used to trans-form votes into seats, it provides us with an empiricallytractable variable. Most important, it is a good predic-tor of the degree to which the electoral law distorts theproportional representation of voters' preferences. Asthe effective electoral thresholds increases, the propor-tionality of the electoral system declines and the like-lihood of strategic behavior among voters and elitesrises.The definition and measurement of the effectiveelectoral threshold is based on the idea that thepercentage of votes a party needs to gain representa-tion is not a specific number but a range of possibilities.In each electoral system, this range is a function of thestrength and fragmentation of the remaining parties. Itextends from the threshold of inclusion, that is, theminimum percentage of the vote that gives a party aseat under the most favorable circumstances (the rest17 Amorim Neto and Cox (1997), Ordeshook and Shvetsova (1994),and Taagapera and Shugart (1989) employ instead the measure ofeffective magnitude. Since the effective threshold is the reverse of theeffective magnitude, both should be seen as "two sides of the samecoin" (Lijphart 1994, 12).

614

of the parties are extremely fragmented), to the thresh-old of exclusion, that is, the maximum percentage ofthe vote that, under the most unfavorable conditions(an opposition party gathers all the remaining vote), isstill insufficient for a party to obtain representation. Asingle-member plurality system provides a straightfor-ward illustration of these different thresholds. Withfour candidates competing for the seat, the thresholdof inclusion is 25%. Any candidate who gets more thanthis percentage wins the seat if the other three candi-dates split the other votes in equal parts. The thresholdof exclusion, however, will be 50% minus one vote, thecase in which the rest of the vote is concentrated on asingle alternative candidate. The literature calculatesthe inclusion and exclusion thresholds on the basis ofthe (average) district magnitude (which includes thepossibility of compensatory seats in secondary districts)and the presence of a legal threshold.18The effectiveelectoral threshold is then calculated as an average ofthe inclusion and exclusion thresholds. The effectivethreshold of single-member plurality and dual-ballotdistricts is 35%.19In a system with four-seat districtsand no legal threshold (the average case in Ireland),the effective threshold is 17%. In a 100-seats district,the effective threshold becomes a mere 0.75%.20 Ap-pendix A lists the cases under study and the variablesemployed in the article.21The Historical Evolution of ElectoralRegimesBased on yearly estimations of the effective electoralthreshold in individual countries in the sample,22 Fig-18 The threshold of exclusion (Texcl) is calculated as Texcl = VIM +1, where V is the total percentage of votes, M is the number of seatsin the district, and 1 stands for the fact that only one party is runningagainst the candidate in the district. The threshold of inclusion(Tincl), which depends on the average number of parties that areassumed to run in each district, is equal to the highest of thefollowing figures: (1) the legal threshold (i.e., the legally stipulatedminimum percentage parties must obtain to be. entitled to get seats);(2) Tinclusion = 100/2M (Lijphart 1994, 26ff.). In a differentapproximation, Taagapera and Shugart (1989, 274-7) calculate thethreshold as Tinclusion = NIMp, wherep is parties and is assumedto be M + 1. Using the latter formula to calculate the dependentvariable does not alter the results reported in this article.19This is the value estimated by Lijphart (1994). Taagapera andShugart (1989), however, put the effective threshold of a single-member district at 50%. See a discussion in Lijphart (1994, 28).20 An alternative operationalization of the dependent variable wouldspecify that it is dichotomous (a value of 1 if the countryshifts to PR,0 otherwise). This has certain disadvantages over the use of acontinuous variable (such as the effective electoral threshold). The(rather extended) classification of electoral systems according toallocation formulas is highly arbitrary and excludes fundamentaldeterminants (district size and legal thresholds) of proportionality.Also, a dichotomous classification based on electoral formulasas wellas district magnitude and legal thresholds suppresses variance anddoes not capture the differences that occur within PR systems(Lijphart 1995, 107-10). Still, if a dummy variable is used to measurethe dependent variable, the statistical results generally confirm theimportance of the threat variable (although confidence levels areslightly weaker). Results are available from the author.21 The full year-by-yeardata set is available from the author and onthe APSR web site.22 For the period after 1945, the values of the effective thresholds aretaken from Lijphart (1994). For the period before World War II, the


8/17


FIGURE 4. Evolution of the Average Effective Electoral Threshold in the Advanced World50 -

40 - , . . -'-:

0 H~ :0~~~~~~~~~~~~~~~~~~~~~~a~ '0 F~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ . X-- . . ., ,, . . . . . . . . . . .CD... . .

10. 20-'

10 - :

Yearl ~ ~ ~ ~~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ..... ......

ure 4 presents the average effective threshold for eachyear from 1875 to 1990.23The plurality and dual-ballotsystems were the only ones in use in the developedworld in the last quarter of the nineteenth century,24regardless of whether (male) suffrage was universal orlimited. By the turn of the century, and as soon assuffrage was extended and modern mass parties werefounded, electoral rules were modified. PR was intro-duced in Belgium in 1899, Finland in 1906, and Swedenin 1907. The turning point, however, was World War I.By 1919 all the small European states as well asGermany and Italy had embraced PR. The averageeffective threshold had fallen to around 18% by 1919,and has moved downward only slightly, to around 14%,since then.As the plurality/majorityrule was abandoned, varia-tion in electoral regimes became substantial. Figures 5through 7 present three sets of cases, grouped accord-effective thresholds are my own estimations, based on Mackie andRose 1991, Nohlen 1981, and Taagapera and Shugart 1989.23 The average effective threshold is the unweighted (for eitherpopulation, geographical size, or GDP) mean of all thresholds acrossthe countries in the sample for any given year.24 The same threshold is attributed to both the plurality and thedual-ballot systems by Taagapera and Shugart (1989) as well asLijphart (1994).

ing to regime evolution and overall stability. Figure 5,which plots stable democracies with stable electoralsystems (after the 1910s), includes two sets of nations.One group (Australia, Canada, New Zealand, theUnited Kingdom, and the United States), despite theintroduction of universal suffrage, did not shift to PR.The other group (exemplified by Austria, the Nether-lands, and Switzerland), once universal suffrage hadbeen introduced, shifted permanently to PR. Similarcases not shown in Figure 5 are Denmark, Finland,Iceland (since World War II), Luxembourg, Norway,and Sweden.Figures 6 and 7 graph the evolution of the effectivethreshold in unstable democracies. Figure 6 showsthose countries that, once they moved to universalsuffrage,hardly changed their electoral system. Despitean episode of democratic breakdown in the interwarperiod, which some authors have associated with ex-treme PR rules, Italy and Germany have maintainedlow thresholds after World War I. After World War II,Germany raised it moderately; Italy followed the re-verse path. In both cases, however, the effective thresh-old has not exceeded 7.5% after 1919. With the excep-tion of the relatively proportional system imposed bythe Allies in the 1946 elections, Japan has employed a

615


9/17


FIGURE 5. Evolution of the Effective Electoral Threshold in Countries with Both StableDemocracy and Stable Electoral Systems50

40

0-c2 30

20

10 ~~~~............. ............... .........

so ~ ~ ~ ~~ ~~~~~~~~~~~~~~~ L- - - - -.. -.. - .... ..- - - - - - - - - - - - - - - - - - -.. -... - .. ..... .. - - - -... ----------

0

YearI:~Austria-- Netherlands - Switzerland * Australia,Canada, New Zealand, UnitedKingdom,United States

semi-PR system since 1925 (with an effective thresholdslightly above 16% before and after World War II).Figure 7 represents the very few cases in which bothdemocracy and the electoral regime have been unsta-ble. From 1875 to 1990 the French effective electoralthreshold changed seven times, and electoral ruleshave been modified many more times than that. Since1926, Greece has changed its electoral threshold eleventimes. Spain has moved from a relatively high thresholdduring the 1930s to a moderate PR system since 1977.EXPLAININGVARIATIONIN EFFECTIVEELECTORAL THRESHOLDSampleGiven the evolution of the effective electoral thresholdover time, I build two samples to explain variation inthe selection of electoral rules.The first sample includes the average effectivethreshold for all those countries that experienced aperiod of democratic government in the interwaryears(22 observations): Australia 1919-39 Austria 1919-34,Belgium 1919-39, Canada 1919-39, Denmark 1919-39, Finland 1919-39, France 1919-39, Germany 1919-33, Greece 1923-36, Iceland 1934-39, Ireland 1922-39,Italy 1919-23, Japan 1925-40, Luxembourg 1919-39,

616

the Netherlands 1919-39, New Zealand 1919-39, Nor-way 1919-39, Spain 1931-36, Sweden 1919-39, Swit-zerland 1919-39, the United Kingdom 1919-39, andthe United States 1919-39.The second sample adds those countries in whichdemocracy was restored after 1945 (31 observations):Austria 1949-90, France 1945-90, Germany 1949-90,Greece 1946-67 and 1975-90, Italy 1946-90, Japan1946-90, Portugal 1975-90, and Spain 1977-90.Explanatory VariablesTwo sets of theories are explored in this article. Thetwo groups of variables relevant to each are describedbelow.First, to test the effect of ruling parties' calculationson electoral structure, I use the following variables. (1)The proportion of socialist votes, or Strengthof Social-ism, ranges from 1% in Japan, to 44% in Sweden, to53% in Portugal. (2) the EffectiveNumberof Old Parties(nonsocialist), or N,25 ranges in the sample fromaround 2 in Italy (1913) and the United States to 6 inGermany (1913). (3) Threat is the interactive term of25 N = 1/>p2i, where pi is the fractional share of every nonsocialistparty i. For a definition and discussion of N, see Taagapera andShugart (1989, 79-80).


10/17


FIGURE 6. Evolution of the Effective Electoral Threshold in Unstable Democracies with StableElectoral Systems50 -

40

2 30

20

> 0

. ... ... ..... .. .... .. .. .. .. .. .. .. .......

0 I I I III I I I I

YearGermany - - - Italy Japan

the two previous variables. If the model suggested inthis article is correct, then the higher the threat vari-able, the more. likely any country will be to shift to PR.The components of the interactive terms also couldexplain on their own why PR was introduced by the oldparties in power. First, at a very high level of socialistvote, even if conservatives were able to merge into asingle party, the winner-take-all nature of the pluralitysystem could still deliver a socialist absolute majorityinparliament. Second, an extraordinary degree of frac-tionalization within the nonsocialist camp could allowan essentially weak socialist party to gather a parlia-mentary majority under non-PR rules. These two ex-treme conditions were exceptional, however. At thebeginning of the interwar period, the average level ofsocialist vote was 22.5%, and the average effectivenumber of nonsocialist parties hovered around three.Thus, it is reasonable to expect that the transformationof electoral rules was mainly drivenby the joint effect ofsocialist strength and conservative fragmentation.26(4)26 I have operationalized the effective number of parties as N forsystems with plurality rule and N - 1 for countries with runoffelections, since the fractionalization of the nonsocialist camp mayhave to be higher in the dual-ballot system than under pluralityruleto secure the victoryof a socialist partyand therefore to push the oldparties to shift to PR. This alternative operationalization of fraction-alization (not reported in Table 1) leads to results very similarto theuse of N.

The year in which male universal suffrage was intro-duced reflects the expectation that the earlier theintroduction, the more capable are old parties incontrolling the electoral arena and the less likely theyare to shift to PR.The percentage of socialist votes and the number ofeffective nonsocialist parties for the interwar sampleare calculated on the following basis: (1) results of thefirst elections under male universal suffrage (in thetwentieth century) in those countries that did notchange to PR,27 (2) results of the last elections heldunder plurality/majorityrule in those countries thatshifted to PR, provided they were contested undermale universal suffrage,28and, (3) when male universalsuffrage was jointly introduced with PR, the results ofthose elections.29 Socialist strength and the number of27 Australia 1902, Canada 1921, Ireland 1922, Japan 1928, NewZealand 1919, Spain 1931, United Kingdom 1918, and United States1904. For France the elections are 1914, since an electoral reform in1919 lowered the threshold from 35% to 29%.28 Austria 1911 (returns in German-speaking provinces), Belgium1894, Germany 1912, Italy 1913, and Switzerland 1917.29 Denmark 1918, Finland 1907, Luxembourg 1919, the Netherlands1919, Norway 1918, and Sweden 1921. Employing the results of theprior election in each country does not change the results (Finland isnot used in this case because its elections before 1906 encompassedonly 4% of the adult population). In Iceland the first elections withcomplete universal suffrage (i.e., after the exclusion of people on

617


11/17


FIGURE 7. Evolution of the Effective Electoral Threshold in Countries with Both UnstableDemocracy and Unstable Electoral Rules50

40

30

20'' '''i

10 ...........0

00 L, Ib I ~ IC~ I~ I II IYear

|: -i:France - - - Spain - -- - -- Greece |

nonsocialist effective parties for the new postwar de-mocracies are estimated for the country's first demo-cratic election.The second set of variables is used to test otherpossible explanations (discussed at the end of the firstsection). I consider three aspects. (1) TradeOpennessismeasured by the log value of the sum of exports andimports as a proportion of GDP during the first yearsof the interwar period and at the time the new demo-cratic regimes were established after 1945.30 (2) Thesize of the country is measured as the log value of thePopulation and the log value of GeographicalArea (inthousands of kM2). (3) The presence of minorities ismeasured through Ethnic and Linguistic Fractionaliza-tion, an average of two indexes. The Russell Indexreported in Taylor and Hudson (1972, 271-4), rangesfrom 0.02 in Japan, to 0.70 in Belgium, to 0.75 inCanada. For religious fractionalization, I also madeestimates following the data reported in Taylor andpublic assistance was abolished) were held in 1934. A semi-PRsystem had been introduced there in 1920 (leading to an effectivethreshold of 11.3%), and more proportional measures were enactedin 1942. Employing the results of 1942 does not change the results inTable 1.30 The measure is taken from Mitchell 1975, 1982, 1983.

618

Hudson (1972, 275-8), and this index ranges from 0 inSpain and Sweden to 0.68 in Canada.31Empirical ResultsTable 1 displays the results for the sample of interwarcountries. When the whole sample is used (includingthe postwar cases), the results hardly vary.32

Column 1 reports the results of the central regres-sors: threat, (log of) geographical area, (log of) trade-openness, (log of) population, and ethnic and linguisticfragmentation.33The level of variance explained is 61%31 Both indices measure the likelihood that two randomly selectedmembers of a given country will belong to different ethnic orlinguistic (or religious in the second index) groups and are calculatedusing the Rae index of fractionalization.32 Results using the whole sample are reported in Appendix B.33 The year in which universal suffrage was introduced has nostatistical significance and a coefficient opposite in sign to mytheoretical expectations. This result may be in line with Przeworski(1975), who shows that most voters have a well-defined politicalidentity before their actual electoral mobilization takes place. Thatwould reduce the effect of the timing of universal suffrage onelectoral rulemaking and would explain why there is no apparentrelationship between male universal suffrage and the likelihood ofadopting PR. This variable was dropped from the results.


12/17


TABLE 1. The Choice of Electoral Rules inthe Interwar PeriodDependent Variable:Average EffectiveThreshold n Model Model Model1919-39 1 2 3

Constant -2.48 -22.74 31.21*(43.12) (17.14) (4.70)Threata -1 1.58* -40.17* 1 3.55*(4.39) (18.92) (4.94)Strengthof socialismb 78.96(58.71)Effectivenumber(N) 8.92of old parties (5.18)Geographicalarea 10.19* 9.67*

(log 000 kM2) (5.24) (2.17)Trade openness (log)d 1.15(8.36)Population log) 0.04(1.54)Ethnicand religious -4.32

fragmentation (13.34)Fragmentation area -32.14*dummyf (14.61)R 0.61 0.67 0.39CorrectedR2 0.49 0.59 0.33SEE 9.22 8.24 10.56Numberofobservations 22 22 22Note: Estimation is by ordinary least squares. Standard errors are inparentheses. *p < 0.05.aThreat is the interactive term of the strength of socialism and effectivenumber of old parties variables.bStrength of socialism is measured by the proportion of votes (from 0 to1) for socialist and communist parties in the first elections contestedunder universal male suffrage (Mackie and Rose 1991).CN is the effective number of nonsocialist parties in the first electionscontested under universal male suffrage. It is calculated as N = 1I/>p2j,where pi is the fractional share of every nonsocialist party i. For adefinition and discussion of N, see Taagapera and Shugart (1989,79-80). Estimations are based on data from Mackie and Rose 1991.dTrade openness is the log value of sum of exports and imports as apercentage of GDP in 1920-25 (Mitchell 1978, 1982, 1983.eEthnic and religious fragmentation is the average of Russell indices ofethnic and linguistic fractionalization and of religious fragmentation(Taylor and Hudson 1972, 271-7).fFragmentation x Area Dummy is the interactive term of the combinedindex of ethnic and religious fragmentation and geographical area,which is 0 for countries bigger than 450,000 kM2, 1 otherwise.

(see Table 1). Both threat (the interaction of socialistvote and conservative fragmentation) and geographicalarea are statistically significant, but population andtrade are not.34Although ethnic and religious diversityhas no statistical significance either, its coefficient hasthe right sign: The more ethnically and religiouslyfragmented a country, the lower the electoral thresh-34 Given that trade and geographical area are correlated, a jointF-test is in order. It showsthat openness is not statisticallysignificant.

TABLE 2. Simulating the Choice of theEffective Electoral Threshold: The InteractiveEffects of Socialist Strength and the EffectiveNumber of Nonsocialist PartiesEffectiveNumber Percentage of Socialist Votesof NonsocialistParties 10 20 30 40

2 29.2 26.9 24.6 22.3(3.7) (3.3) (3.2) (3.2)4 26.9 22.3 17.6 13.0(3.3) (3.2) (4.0) (5.3)6 24.6 17.6 10.7 3.8(3.2) (4.0) (6.0) (8.3)

Note: Results are derived from Table 1, column 1. The log of geograph-ical area is set to 3, and trade openness, population, and ethnicfragmentation are set at their mean values. Standard errors are inparentheses.

old.35 When trade, population, and fragmentation aredropped, threat and geographical size alone still ex-plain 61% of the variance-the corrected r2 is 0.58 (notreported in any of the tables).The Efect of Threat. As predicted, the higher thethreat faced by the old ruling parties, the lower is thenew electoral threshold. Based on the results in column1 of Table 1, Table 2 simulates the level of the effectivethreshold, once universal suffrage is introduced, fordifferent combinations of socialist strength and conser-vative fragmentation.36 The effective threshold staysclose to 30% (a value equivalent to having single-member districts) when there are two main nonsocialistparties and the socialist alternative is weak (the U.S.case). For low levels of socialist vote, growing nonso-cialist fragmentation leads to a mild reduction ineffective thresholds: from around 29% to 24% (theequivalent of an average district magnitude of 2.5seats). In turn, holding constant the effective number ofconservative parties, socialist vote drives the thresholddown from 29% to around 22%. Although socialism isa threat, the ruling parties are still unified enough totake advantage of a relatively disproportional system.Above all, the simulation shows the powerful interac-tive effect of both variables. With a socialist partycapturing just 20% of the vote and four nonsocialistparties, the threshold falls by 10 points, from 31.5%(the level with no socialist party) to 22%. With veryhigh levels of conservative fragmentation and a strongsocialist party, the ruling parties move decisively to apure PR system. The German case fits quite well. In1913 the Socialists polled 34% of the vote, and the35An important strand of the literature, dating back to Almond(1956), distinguishes between homogeneous and heterogeneous po-litical cultures. To test its effects, I built a dummy variable that,following Lijphart 1977, takes the value of 1 for Anglo-American andScandinavian countries as well as West Germany.It is not statisticallysignificant.36 For the simulation, the log of geographical area was set to 3(around 600,000 km2), and trade openness, population, and ethnicand religious fragmentation were set at their mean values.

619


13/17


FIGURE 8. Geographical Size and Level of Ethnic and Religious Fragmentation0.8 -

0.7 - 13Canada

o 0.6 -.2C 3 SwitzerlandE0)C 0.5 -

iLc).o 0Belgium 0USAa) 0.4

03Netherlands03 0 Iceland New Zealand-c 0.3 0 0 G Australiail 10Ireland 1 Germany

o 3 FranceX Finland0 o Spainc0.20 Greece

0 Italy0. 1 Luxembourg a Austria 0 3 Norway.1 - a~~~~~~~~enmark 13UK03Portugal 03Sweden

0 Japan0

0.4 0.9 1.4 1.9 2.4 2.9 3.4 3.9 4.4Log of Geographical Size ('000 kM2)

effective number of conservative parties was six. The1919 German effective threshold was 1.84%.Column 2 (in Table 1) explores the robustness of thethreat variable.Trade, population, and ethnic fragmen-tation (which were not statistically significant in col-umn 1) are dropped, and the two components of theinteractive threat term are added as separate variables.Threat is still highly significant. Its coefficient increases,which should be expected, since the interactive term iswell correlated with both of its components.37Noticethat, against theoretical expectations, the coefficients ofsocialism and the effective number of old parties arepositive. Yet, a joint F-test shows that neither socialiststrength nor conservative fragmentation is statisticallysignificant. Thus, the overall results of column 2 inTable 1 and particularly the statistical strength of theinteractive term show that, in line with my mainargument, threat is the fundamental factor in deter-mining the choice of electoral threshold.Size, Trade,and InternalFragmentation. Geographicalarea explains part of the variation in electoral thresh-olds.38 The larger the country, the higher is the elec-toral threshold. It is possible that size itself determinesthe choice of electoral regime, and it has been ad-vanced as an explanatory variable in politics (Dahl andTufte 1973). The coordinating consequences of plural-37 The correlation coefficient is .67 between threat and socialiststrength and .65 between threat and conservative fragmentation.38 The same result is obtained by Blais and Massicotte (1997).

620

ity rule may be particularlyvaluable for large countries.It is more likely, however, that size is a proxy for othervariables.Size may capture the effect of trade openness. Geo-graphical area is highly correlated with trade: ThePearson's coefficient is -.85 for the interwar sampleand -.68 for the whole sample. Still, the claim isproblematic, given the poor statistical performance oftrade once we enter size (see Table 1, column 1).39Size is, above all, a proxyfor the way in which ethnicand linguistic fragmentation affects each country andthe means elites devise to deal with it. Consider Figure8, which displays the relationship between ethnic andreligious fragmentation and geographical size. Thegraph shows a concave relationship (particularly if weexclude Luxembourg). Medium-sized nations such asSweden and the United Kingdom have low fractional-ization scores. Fragmentation turns out to be especiallyhigh in either very small (Switzerland,Belgium, and theNetherlands) or very large countries (Australia, Can-ada, and the United States).The institutional response to fragmentation varies,however. In small countries, political elites have intro-duced PR to accommodate ethnic and religious minor-ities. By contrast, in extremely large countries, such asAustralia, Canada, and the United States, PR has notbeen adopted. Why? It is unconvincing to argue that39 For a full discussion of Rogowski's theory, its empirical test, andthe possible reasons for its weaknesses, see Boix 1992.


14/17

American Political Science Review Vol. 93, No. 3there was no pressure to embrace PR to secure minor-ity representation because ethnic and linguistic frag-mentation in these countries (a by-product of large-scale migrations in the nineteenth century) did notcrystallize into specific political cleavages, as it did inEurope. This explanation may be valid for Australiaand even the United States, but it does not fit theCanadian case.A more satisfactory explanation (which accepts, in aqualifiedmanner, the importance of minorityrepresen-tation) must consider, instead, both the geographicaldistribution of minorities and the adoption of alterna-tive mechanisms to represent them. In many smallEuropean nations, where religious (and sometimesethnic) cleavages tend to be distributed uniformlyacross the country, the maintenance of a single-mem-ber single-ballot system clearly would have benefittedthe strongest minority.Accordingly, either for the sakeof civil peace or mere survival, the old ruling partiesavoided a (plurality rule) that would have suppressedminority representation. By contrast, in very largecountries,where ethnic and religious minorities tend tobe concentrated in specific regions, PR is not necessaryto secure the political participation of any significantethnic or linguistic group. Canada is a case in point. Ithas a plurality system, but Quebecois interests arerepresented in Ottawa.40Even if a country is extremelyheterogeneous at the national level, if its regions andlocal districts are rather homogeneous, a different setof mechanisms-such as federalism and a strict sepa-ration of powers-can secure the representation ofpolitical minorities and hence make PR superfluous.41In short, under certain conditions, federalism operatesas a (quasi-perfect) substitute for PR and minimizesany potential pressures to abandon a plurality/majoritysystem.42To measure the effect of fragmentation while con-trolling for size, I developed the variable Fragmenta-tion X Area Dummy, where the dummy is 0 forcountries larger than 450,000 km2, 1 otherwise. Col-umn 3 (in Table 1) reports a regression for threat andthis new interactive term. Both variables are statisti-cally significant.For small and medium-sized countries,40 The evolution of electoral law in Switzerlandfits this model nicely.The introduction of PR rules at the national level took place late(when the emergence of the Socialist Partythreatened the hegemonyof the Radicals), mostly because majority rule did not harm anyminorities significantly. Since they were relatively concentrated geo-graphically, they could attain a fair share of seats in the federalparliament. By contrast, within some cantons, such as Geneva andTicino, different social groups were more likely to overlap, andpolitical clashes over the electoral system emerged there as early asthe mid-nineteenth century (Carstairs 1980, 135-46).41 The creation of personal constituencies (the step taken by NewZealand with the establishment of several Maori-only constituencies)or gerrymandering can also be seen as ways to secure the represen-tation of certain minorities without shifting to PR.42 From this discussion, it follows that the index of ethnic andreligious fragmentation employed here is partly flawed for mypurposes. Since it only captures fragmentation at the national level,it cannot measure properly the degree to which plurality systems,bythwarting the representation of minorities, may generate instabilityin the national political system. It will only be possible to examinethis problem if we develop measures of ethnic and linguistic frag-mentation at the district level.

fragmentation increases the chances of adopting a PRsystem. Other things being equal, a highly fragmentedcountry (such as Switzerland) should have a threshold17 points lower than a homogeneous country (such asJapan).CONCLUSIONThree historical periods can be distinguished in theevolution of electoral regimes in the developed worldover the last century.During the era of limited suffragein the nineteenth century, plurality/majorityrule wasconsistently used across all nations. At the turn of thecentury, the old consensus around the single-membersystem broke down. Anglo-Saxon countries preservedthe plurality rule, but most nations in continentalEurope embraced PR. Finally, the 1920s ushered in anew era of remarkable stability in the structure ofelectoral regimes. In the last eight decades, majorchanges in electoral rules have been limited to France,Greece, and (to some extent) Spain.43The selection (and preservation) of different elec-toral rules can be traced to the strategic decisions madeby the current ruling parties, foreseeing the coordinat-ing consequences of different electoral systems, tomaximize their representation in parliament. As longas the electoral arena in which they compete is stableand the electoral system serves them well, the rulingparties have no incentives to modify any electoralnorm. A sudden transformation of the electoral marketand a corresponding increase in the degree of uncer-tainty are likely to trigger a change in the electoralregime. Four different phenomena may lead to trans-formation of the political arena: the extension ofuniversalsuffrage (Western Europe in the 1910s or newdemocratic nations in the postwar period); the intro-duction of competitive elections (Eastern Europe andseveral African nations in the 1990s); a massive polit-ical realignment among voters (the rise of socialism atthe turn of the century or today's rise of protectionistparties, which would partly explain why France tempo-rarilyshifted to PR in 1986-88); and a high turnover inparty organizations (France and Greece in this centu-ry).44The degree to which the ruling parties decide tomodify the current electoral rules depends on theextent to which the latter undermine the former'spolitical viability in the new electoral arena. This is, inturn, a function of two main conditions: the strength ofthe new parties and the capacity of the old rulingparties to coordinate among themselves to block the43 Italy and New Zealand also have introduced changes very recently.44 The instability of the electoral systems in those cases is stronglyrelated to the characteristics of the party system. Between 1945 and1970, the instability of voter support for French parties was two andone-half times greater than the average in other advanced democra-cies (Rose and Urwin 1970). More generally, the Pearson's r betweentotal volatility in vote shares and the standard variation of theeffective threshold in the sample of OECD nations in 1900-80 is0.74. Total volatility is computed [=1' (piK pi, +,/2)], where n isthe number of parties in the system, andpi is the electoral support inpercentage for party i at time t and t + 1. See Bartolini and Mair(1990) for a detailed discussion of this measure.

621


15/17

The Choice of Electoral Systems in Advanced Democracies September 1999growth of new parties. On the one hand, if the new-comers are strong, the old parties shift to PR wheneverthey are locked into a non-Duvergerian equilibrium,that is, when they are either tied in votes or aresupported by strongly committed voters. Since elite orvoter coordination around one of the old parties isextremely unlikely to happen, the current governmentabandons plurality/majorityrule to avoid the extraor-dinary costs it imposes on weaker candidates of the oldparties. This matches the decision to embrace PR inmost small European countries. On the other hand,even if the new parties command high levels of support,there are no incentives to shift to PR if one of the oldparties enjoys a dominant position among the elector-ate. Under this circumstance, which fits the Britishcase, the dominant party rationally expects to becomethe rallying point for voters who want to block thevictory of the new party.Finally,when the new entrantsare weak, non-PR rule will remain in place, regardlessof the structure of the old party system.My analysis of the origins of different (electoral)institutional equilibria in democratic countries opensup at least two broad research questions. First, giventhat the selection of different electoral rules hinges onthe political conditions under which the ruling partiesoperate, in other words, because electoral rules ulti-mately are endogenous to the political system, we arepushed again to examine what shapes those conditions.What determined the number of nonsocialist parties inthe 1910s and 1920s across countries? Why was thecoordinating capacity of elites different across coun-tries? Why did it take so long to establish stable partysystems and electoral rules in countries such as Franceor Greece? Responses to these questions (as well as anexamination of the bargainingrounds among elites thatled to different institutional solutions) require a histor-ical analysis beyond the scope of this article. The modelpresented here, however, provides the theoreticalfoundations for such work.Second, I have examined the choice of electoral rulesin developed democracies. In these countries, thegovernment could predict with some certainty thefuture structure of electoral competition; parties had,in most cases, become national; and parliamentaryandsemipresidential systems were (excluding the UnitedStates) the norm. Since the choice of an electoralsystem in a developing country that embraces democ-racy for the first time or after a very long period of

622

authoritarianismmay not meet these conditions, pre-dictions about that choice ought to change correspond-ingly. First, under conditions of very high uncertaintyabout the structure of the electoral arena, the rulingelite will select the system most likely to minimize risks,so it will lean toward a mixed or pure PR system.Second, if parties are collections of local notables, thenthere may be an incentive to embrace single-memberdistricts or multimember districts in which voters haveas many votes as seats, since these structuresstrengthen local ties and patronage politics.45 Third,the choice of electoral systems is likely to be affected bythe type of transition to democracy. PR (and a weakpresidency) will be more likely in countries wheredemocracy is imposed from below. By contrast, inthose places where the old elite has liberalized theregime while controlling significant resources, a highelectoral threshold may be more common.46 Finally,given that newly democratizing countries vote a newconstitution ex nihilo, the electoral law will be partic-ularly shaped by the broad constitutional framework-the powers of executive and assembly as well as thelevel of decentralization-finally chosen. As revealedby my examination of the influence of ethnic andreligious diversity, the degree of proportionality in theelectoral system will depend in part on the use ofalternative mechanisms, such as federalism, to manageconflict and fulfill the goal of fair political representa-tion.45 Being a former British colony has been found to be a strongpredictor of whether a country chooses a single-member pluralitysystem (Blais and Massicotte 1997). Accordingly, the adoption ofelectoral rules in developing countries is mostly presented as avalidating example of the influence of ideas and the diffusion ofcultural models on political and constitutional choices. (For recentworks on the role of ideas, see Goldstein and Keohane 1993 and Hall1989.) The strategic model proposed here questions this traditionalmodel in the following way. Assume that Britain shaped, throughboth the political institutions it established and the way negotiationstoward independence were carried out, a rather stable set of rulingparties or elites in its colonies. As long as these parties (perhaps acoalition of local patrons) were viable under the first-past-the-postsystem, no changes should be expected after the declaration ofindependence (i.e., India and the dominant position of the CongressParty).46 The introduction of a constrainingelectoral system could take twoforms: a parliamentaryregime and plurality rule, or a presidentialsystem with strong powers for the executive and concurrent presi-dential and legislative elections. See Shugart and Carey (1992) forevidence suggesting that this latter system depresses the number ofparties significantly.


16/17


APPENDIX AEffective IndexofNumberof Ethnicand% of Nonsocialist Size (in000 LinguisticCountry Threshold LeftistVotes Parties km2) Fragmentation

InterwarPeriodAustralia 35.0 31.0 2.29 7686.8 0.293Austria 8.9 25.4 2.57 83.8 0.106Belgium 4.8 14.9 2.75 30.5 0.416Canada 35.0 2.3 3.12 9976.2 0.715Denmark 2.9 28.7 3.12 43 0.089Finland 5.7 37.0 3.50 337 0.214France 32.9 20.7 4.79 547 0.245Germany 1.8 34.8 6.02 456.8 0.279Greece 25.8 10.9 4.42 131.9 0.163Iceland 11.3 27.7 2.26 103 0.318Ireland 17.2 21.3 2.91 70.3 0.277Italy 7.5 17.6 2.03 301.2 0.127Japan 24.1 1.2 2.65 369.7 0.020Luxembourg 5.2 15.6 2.31 2.6 0.117Netherlands 0.8 24.3 4.97 33.6 0.344New Zealand 35.0 23.8 2.38 268.7 0.289Norway 9.2 31.6 2.66 324.2 0.105Spain 21.5 21.7 3.65 504.7 0.210Sweden 8.8 44.0 2.75 449.8 0.060Switzerland 8.5 30.8 2.42 41.3 0.552UnitedKingdom 35.0 22.5 2.46 244 0.072UnitedStates 35.0 3.6 2.10 9363.4 0.423

Postwar PeriodAustria 4.1 38.7 2.03 83.8 0.106France 28.0 41.1 2.73 547 0.245West Germany 7.1 29.2 4.70 248.5 0.283Greece - 1946-67 20.9 19.3 1.97 131.9 0.163Greece - 1975 14.7 23.1 1.75 131.9 0.161Italy 2.0 39.6 2.70 301 0.127Japan 16.2 21.8 3.09 369.7 0.020Portugal 5.8 53.5 2.35 92 0.057Spain 10.2 34.8 4.26 504.7 0.210

APPENDIX B. The Choice of Electoral Rules in the Interwar Period and New Democracies after 1945Dependent Variable:AverageEffectiveThreshold Model1 Model 2 Model3Constant -13.03 (22.51) -18.13 (14.40) 30.02* (4.70)Threat - 11.27* (3.39) -34.07* (15.82) -13.24* (3.89)Strength of Socialismb 57.70 (45.86)Effectivenumber N) of old parties 7.31 (4.50)Geographical area (log 000 kM2) 11.33* (3.45) 9.89* (1.99)Trade openness (log)d 2.96 (4.43)Population log) 0.08 (1.32)Ethnicand religious ragmentation' -2.26 (11.18)Fragmentation dummyarea -31.45* (12.50)R 0.60 0.42 0.37CorrectedR2 0.52 0.34 0.33SEE 8.27 7.76 9.71Numberof observations 31 31 31Note: Variable efinitions rethe same as in Table 1. Estimations by ordinaryeast squares. Standard rrors n parenthesis.kp?.O5.

623


17/17

The Choice of Electoral Systems in Advanced Democracies September 1999REFERENCESAlmond, Gabriel A. 1956. "Comparative Political Systems."Journalof Politics 18 (August): 391-409.Amorim Neto, Octavio, and Gary W. Cox. 1997. "Electoral Institu-tions, Cleavage Structures, and the Number of Parties."AmericanJournal of Political Science 41 (January): 149-74.Bartolini, Stefano, and Peter Mair. 1990. Identity,Competition,and

Electoral Availability. The Stabilisation of European Electorates1885-1985. Cambridge:Cambridge University Press.Blais, Andr6, and Louis Massicotte. 1997. "Electoral Formulas: AMacroscopic Perspective."European Journal of Political Research32 (August): 107-29.Boix, Carles. 1992. "Trade, Politics and Proportional Representa-tion. A Reinvestigation of Rogowski's 'Trade and the Variety ofDemocratic Institutions.' " Department of Government, HarvardUniversity. Typescript.Butler, D. E. 1963. TheElectoralSystemin Britain since 1918. 2d ed.Oxford: Clarendon Press.Carstairs, Andrew McLaren. 1980. A Short History of ElectoralSystems in WesternEurope. London: George Allen & Unwin.Cox, GaryW. 1997.Making Votes Count:StrategicCoordination n theWorld'sElectoralSystems. New York: CambridgeUniversity Press.Dahl, Robert A., and Edward R. Tufte. 1973. Size and Democracy.Stanford: Stanford University Press.Downs, Anthony. 1957. An Economic Theory of Democracy. NewYork: Harper and Row.Duverger, Maurice. 1954. Political Parties. New York: Wiley.Goldstein, Judith, and Robert 0. Keohane, eds. 1993. Ideas andForeignPolicy. Beliefs, Institutionsand Political Change.Ithaca, NY,Cornell University Press.Hall, Peter A., ed. 1989. The Political Power of Economic Ideas:Keynesianism across Nations. Princeton: Princeton UniversityPress.Hermens, Ferdinand A. 1941. Democracy or Anarchy?A Study ofProportional Representation.Notre Dame: University of NotreDame.Lijphart, Arend. 1977. Democracy in Plural Societies. New Haven,CT.: Yale University Press.Lijphart,Arend. 1985. "The Field of Electoral Systems Research: ACriticalSurvey."ElectoralStudies 4 (March): 3-14.Lijphart,Arend. 1992. "Democratization and Constitutional Choices

in Czecho-slovakia, Hungary and Poland, 1989-1991." Journal ofTheoreticalPolitics 4 (April): 207-23.Lijphart, Arend. 1994. ElectoralSystems and Party Systems.A Study ofTwenty-SevenDemocracies,1945-1990. New York: Oxford Univer-sity Press.Mackie, Thomas T., and Richard Rose. 1991. The InternationalAlmanac of Electoral History. 3d ed. Washington, DC: Congres-sional Quarterly.Mitchell, Brian R. 1975. European Historical Statistics, 1750-1970.New York: Columbia University Press.Mitchell, Brian R. 1982. InternationalHistorical Statistics:Africa andAsia. New York: New York University Press.Mitchell, Brian R. 1983. InternationalHistorical Statistics:The Amer-icas and Australasia. London: MacMillan.Nohlen, Dieter. 1981. Sistemaselectoratesdel mundo. Madrid:Centrode Estudios Constitucionales.Ordeshook, Peter, and Olga Shvetsova. 1994. "EthnicHeterogeneity,District Magnitude, and the Number of Parties."AmericanJournalof Political Science 38 (February): 100-23.Przeworski, Adam. 1975. "Institutionalization of Voting Patterns, orIs Mobilization the Source of Decay." American Political ScienceReview 69 (March): 49-67.Rae, Douglas W. 1971. The Political Consequences of Electoral Laws.New Haven, CT: Yale University Press.Rogowski, Ronald. 1987. "Trade and the Variety of DemocraticInstitutions." International Organization41 (Spring): 203-24.Rokkan, Stein. 1970. Citizens, Elections, Parties:Approaches to theComparative Study of the Process of Development. Oslo: Univer-sitetsforlaget.Rose, Richard, and Derek Urwin. 1970. "Persistence and Change inWestern Party Systems since 1945." Political Studies 18 (Septem-ber): 287-319.Shugart, Matthew S., and John Carey. 1992. Presidents and Assem-blies. Constitutional Design and Electoral Dynamics. New York:Cambridge University Press.Taagapera, Rein, and Matthew S. Shugart. 1989. Seats and Votes: TheEffects and Determinants of Electoral Systems. New Haven, CT.:Yale University Press.Taylor, Charles L., and Michael C. Hudson. 1972. WorldHandbookof Political and Social Indicators.2d ed. New Haven: Yale Univer-sity Press.

624

07 boix - setting the rules of the game the choice of electoral systems in advanced democracies

Documents