Primary Election Systems and Representation Author(s): Elisabeth R. Gerber and Rebecca B. Morton Source: Journal of Law, Economics, & Organization, Vol. 14, No. 2 (Oct., 1998), pp. 304-324 Published by: Oxford University Press Stable URL: http://www.jstor.org/stable/765107 . Accessed: 05/08/2013 20:55 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . 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 support@jstor.org. . Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to Journal of Law, Economics, &Organization. http://www.jstor.org This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 304 JLEO,V14N2 Primary Election Systems and Representation ElisabethR. Gerber San Diego Universityof California, Rebecca B. Morton Universityof Iowa Weexaminehowdifferencesinthe institutionsthatregulatecandidatenomination procedures,specificallydirectprimaryelection laws, affect the types of candidates elected innonpresidential Americanelections. Wehypothesizethatin more closed primarysystems, controlover candidatenominationsby ideologicalextremistswilltranslateinto a higherlikelihoodthat extremecandidateswin in the by a generalelection. We hypothesizethat in moreopen systems, participation widerspectrumof the electoratemeansthatcandidatesmustappealto moremoderatevoters,leadingto the electionof moremoderatecandidates. Usingpooled cross-section time-seriesregressionanalysis, we findthat U.S. representatives fromstates withclosed primariestake policypositionsthatare furthestfromtheir district'sestimatedmedianvoter'sideal positions. Representativesfromstates withsemi-closedprimariesarethe most moderate.Weconcludethatthe costs of strategicbehaviorcreatedby electoralinstitutionshave importantconsequences forelectoraloutcomes. 1. Introduction In recent years, several American states have changed or consideredchanging their proceduresfor nominatingcandidatesfor elected office.' In 1984, for example, the ConnecticutRepublican Party changed its bylaws to permit independentsto participatein its previously closed primaries. After the SupremeCourtruledin Tashjianv. RepublicanPartyof Connecticut[479 U.S. 208 (1986)] that the state's interest in requiringa closed primarywas insubstantial, the state legislaturepassed legislation allowing the partiesto permit participationby independentsin their primaries.2 More recently, California 1. In this researchwe consider nominationsto offices other than the presidency. Presidential nominationsare characterizedby a unique sequentialstate-by-statenominationprocess. Nominationsvia direct primaryto congressional,state legislative, state executive, and local offices are characterizedby common procedureswhich we describebelow. 2. Ironically,aftertheTashjiandecision, theRepublicanPartyremovedthequestionableprovision from its bylaws and at the time of this writing,none of the state's majorpartieshave optedto allow independentsto participatein theirprimaries. ) 1998 OxfordUniversityPress This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions PrimaryElection Systems and Representation 305 voterspassed the OpenPrimaryAct of 1996 by directvoterinitiative,replacing the state's restrictiveclosed primarywith a more open variety. All of these changes have attractedvigorous legal scrutinyand scholarlydebate,and at the time of this writing, the constitutionalityof California'smost recent changes (and by implication,Alaska's and Washington'sprimaryelection systems) is still unresolved.3 Proponentsand opponentsof these changes in primaryelection laws agree thattheirconsequences are likely to be profound.4Some of the consequences of these changes are easy to predict. For example, few doubt that opening nominationproceduresto previouslyexcluded nonpartisanswill increasemass participationin the nominationprocess. Other consequences, however, are more difficultto anticipate. For example, will more open nominationsinduce candidatesto compete for the electoralcenter,therebyproducingmore centrist elected representatives,or will they create opportunitiesfor partisanmischief, allowing partyadherentsto "spoil"the otherparties' nominationsand leading to the election of extremist representatives?Further,even when proponents and opponentsagreeon what the electoralconsequencesof changes in election laws will be, the normativeimplicationsof those changesareoften unclear.Are votersmade betteroff by, for example, increasedparticipation?Are the parties made betteroff? Are some voters or groupsmade betteroff thanothers? Much of the ongoing debate over the likely consequences of changing a state'sprimaryelection laws resultsfrom the virtualabsence of any systematic analyses of candidatenominationprocedures. The literaturethat does exist focuses on how partyorganizationinfluencesthe choice of nominationprocedures (Ranney, 1975; Eldersveld, 1982; Jewell, 1984; Epstein, 1986); on the roles of the media, voter information,and candidatecharacteristicsin presidentialprimariesandcaucuses (Aldrich, 1980;Bartels, 1988); on the effects of presidentialprimaryelection systems on voterbehavior,particularlycrossover voting (Wekkin, 1988; Southwell, 1991); and recently on the effects of primary systems on the election of U.S. senators (Grofmanand Brunell, 1997). Togetherthese studies informour understandingof some aspects of candidate nominations,especially presidentialnominations.However,none of these existing works provide a systematic analysis of the electoral consequences of nonpresidentialcandidatenominations. In this researchwe empiricallyexamine how differencesin candidatenominationproceduresin nonpresidentialelections affect one importantaspect of election outcomes: the relationshipbetween winning candidates'policy positions and their constituencies' preferences. In other words, we study how primaryelection systems affect the representationof citizen preferences. We 3. Californiaadopted a blanket primarysimilar to that used in Washingtonand Alaska. See below for a descriptionof the blanketprimary. 4. But see Riker (1983) who arguesthat since institutionsare themselves the productof social choices, theirindependenteffectson politicaloutcomes(especiallystability)arelikely to be minimal in the long run. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 306 TheJournalof Law,Economics,&Organization, V14N2 focus on "dyadic"representation-the extent thatthe positions of a single representativereflectthe preferencesof his or her district-because it reflectsthe primarylegislator-citizen relationshipin a system of single memberdistrict representation. Dyadic representationis in contrastto collective representation, which refers to the extent that an individual'sor group'spreferencesare representedby the entirepolicy-makingbody (see Weissberg,1978). We analyze the representationalconsequencesof the most common mechanism for nominatingnonpresidentialcandidatesin the United States: direct primaries.Primariescome in severalvarieties,dependinguponhow theyrestrict participationby the electorate. A primaryis consideredopen if participantseither do not need to declare party affiliationas a prerequisiteto participating in a primaryelection or may do so on election day. Two variantsof the open primaryare blanketprimaries,in which voters receive a single ballot listing all candidatesfrom all partiesand may participate,office by office, in all or some of the parties' primaries;and nonpartisanprimaries,in which voters choose among candidatesin a primaryregardlessof the partymembershipof the candidate or the voter.5 A primaryis defined as closed if participationis limited to voterswho declaretheiraffiliationto the partya specifiedperiodpriorto the election.6 Withinthe broadcategoryof closed primaries,statesexhibitvarying degreesof"closedness,"dependingupon the comprehensivenessof the preregistrationrequirement.A primaryis defined as semi-closed if new registrants are allowed to both registerandchoose theirpartyon the day of the primaryor if independentsare allowed to participate.7Table 1 reportsthe primarysystem used to nominatecandidatesfrom state legislative and executive and federal legislative offices for the 50 states. We expect institutionaldifferencesin primaryelection systems to influence the positions of winning candidates. In the following two sections we review a body of theoreticaland empiricalwork that forms the basis of our research. The analysesimply severalempiricallytestablehypothesesaboutthe effects of the degreeof closedness of primarysystems on the types of candidateselected. In particular,we hypothesizethatin closed primarysystems, the likelihoodof extreme general election winners is highest. In more open primarysystems, 5. A majordifference between blanket primariesand nonpartisanprimariesis that in blanket primaries,the top vote receiver from each party becomes the party's general election nominee, while in nonpartisanprimaries,the top vote receiverin the primarywins the seat outrightif he or she receives over 50% of the primaryvote. Otherwisethe top two vote receivers, regardlessof party,meet in a runoffelection. This raises the possibilitythattwo candidatesfrom the same party may meet in the runoff. In its recent decision in Fosterv. Love, 96-670, the SupremeCourtruled thatthe timing of these nonpartisanelections, but not theirstructureper se, is unconstitutional. 6. Unlike the Court-imposed30 day maximum registrationdeadline for voting in a general election [see Dunn v. Blumstein,405 U.S. 330 (1972)], partyregistrationdeadlinesvary in length from 10 days in NebraskaandNew Hampshireto a year in New York(Bott, 1990). 7. During the period understudy, 10 southernstates also used runoffs when no candidatereceived a majorityin the primary.See Gerber,Morton,and Rietz (1994) for an analysis of runoff requirements. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 307 Election andRepresentation Systems Primary Table1. Primary SystemType,U.S.States,1990 State Primary State Primary System System MT Open AL Open NE Closed Blanket AK NV Closed Closed AZ NH Closed AR Open NJ Semi(independent) CA Closed NM Closed CO Closed NY Closed CT Closed NC Closed DE Closed ND Closed FL Open OH Open GA Open OK Semi(newandindependent) HI Open OR Semi(new) ID Open Closed PA IL Open RI Semi(independent) IN Open SC IA Open Open Closed SD KS Closed TN KY Closed Open TX LA Open Nonpartisan ME Semi(independent) UT Open VT MD Semi(new) Open MA Semi(independent)VA Open WA Blanket Ml Open WV Closed MN Open WI MS Open Open WY Open MO Open Source: Bott. 1990. the likelihood of moderatewinners is higher. We then presentempiricaltests of our hypotheses. Analyzing data from'a series of recent U.S. congressional elections, we findthatrepresentativeselected underclosed primarysystems do not as accuratelyreflectour estimatesof the medianvoter'spreferencein their districtsas do legislatorselected undermore open primarysystems. We find thatrepresentativeselected undersemi-closed primarysystems most accurately reflect their district'smedian voter's preference. These results are robustto a varietyof alternativeempiricalspecifications. Ourresearchhas importanttheoretical,political, and practicalimplications. Froma theoreticalperspective,understandingthedynamicsof primaryelections will provide insight into how these and other multistage elections function. Froma politicalperspective,changinga state'selection laws meansshiftingthe balanceof politicalpower,giving an advantageto some interestsat the expense of others. In other words, it creates winners and losers. Understandingwho wins and who loses underdifferentelection laws allows us to betteranticipate these political dynamics. And from a practicalperspective,studying election laws will help the courtsand policymakersbetterdesign institutionsto achieve socially desirableoutcomes. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions ofLaw/, 308 TheJournal &Organization, V14N2 Economics, 2. Primary Voter Preferences and Candidate Positions 2.1 Closed Primaries Most Americanelections are three-stageelectoralprocesses. In the firststage, votersaffiliatethemselvesorregisterwithaparticularparty.In thesecondstage, voters choose a partynominee from the subset of candidatesrunningin their party'sprimary.In the thirdstage, voters choose between the partynominees. A number of models have considered how this multistage electoral process affects candidatepositions (Coleman, 1971, 1972; Aranson and Ordeshook, 1972; Wittman, 1977, 1983, 1991; Aldrich, 1983; Aldrich and McGinnis, 1989).8 In these models, there are typically two parties, each of which field two candidateswho compete in closed partyprimaries.Partymemberschoose betweenthe candidatesin theirprimary,andthe winnersof eachparty'sprimary meet in the general election. These parties are assumed to be dominatedby members or party elites who have preferencesover policy, that is, they seek electoral victory in order to enact their preferredpolicies. Accordingly, we referto this literatureas the "partyelite"literature.9This assumptionof policymotivatedpartymembersis in contrastto the Hotelling-Downsianassumption of party competition in which parties instrumentallychoose policy positions strictlyfor the purposeof achievingelectoralvictory (Hotelling, 1929; Downs, 1957). In the partyelite theory,partymembersare assumedto have policy preferences thatare distinctfrom the preferencesof nonmembers.Preferencesof the membersof each partyare also assumedto divergefrom those of membersof the otherparty,with the ideal policy position of each party'smedianmember located on opposite sides of the ideal point of the median voter in the electorate. Candidatesfor office choose policy positions in the primaries"as if" they aremaximizingthe expectedutilityof the medianvoterin theirpartyrather than the expected utility of the median voter in the electorate. When there is some uncertaintyabout the general election outcome, the expected utility of the party's median voter may be maximized by a position that diverges substantiallyfrom the ideal point of the median voter in the generalelectorate.0l? 8. Most of these works model electoralcompetitionin a unidimensionalpolicyspace. NWittman (1983) generalizes the basic multistage model to a multidimensionalpolicy space. To obtain divergenceresults in multipledimensions,however,additionalassumptionsabout the underlying utility functionsand candidates'election probabilitiesare required. 9. By "partyelites"we meancommittedmembersof theparty.This may includecaucusactivists, partyleaders,and regularmemberswho affiliatewith the party.Partyelites, by our definition,are in contrastto primari voters who participatein partyactivities (includingprimaries)not because of a strongattachmentto the partybut ratherbecause of transientor strategicinterests. 10. Note that in thes- models, candidateswill only diverge from the generalelectoratemedian voter's ideal point if there is uncertaintyabout the general election outcome. If, conversely, candidatesknowall aspectsof the game forcertain,theycan anticipatethegeneralelection outcome and maximize the expected utility of the party's median voter by maximizing its probabilityof winningin the generalelection, thatis, by convergingto the generalelectoratemedianvoter'sideal point. These models typically assumethatthe sourceof uncertaintyis the ideal pointof the general electorate median voter due to variationsin turnoutor some nonpolicy candidatecharacteristics that are revealed between the candidatenominationstage and the general election (see Aranson This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions Election 309 Primary SystemsandRepresentation Thus,as a consequenceof divergentpartymemberpolicy preferencescombined with uncertaintyaboutthe genei al election outcome,candidatepositions in the multistagemodel do not converge'. the ideal point of the Hotelling-Downsian medianvoter in the generalelectorate. ' Theirpositions are insteada function of the ideal point of the median voter ill their party's primary. Thus the further the ideal point of the median voter in a party'sprimaryfrom the general election medianvoter,the more extremethe position of thatparty'snominated candidate.12 A key assumptionin these models of multistageelections is thatthe position of the median voter in a closed primaryis likely to diverge substantiallyfrom the position of the median voter in the general electorate. Several studies together provide empiricaljustificationfor this assumption. One set of studies suggests thatvoter participationin closed primariesis lower than in open primaries,ceteris paribus.For instance,Jewell (1984) shows thatvoterturnoutin gubernatorialprimariesfrom 1952 to 1982, as a percentageof partyvote in the subsequentgeneralelection, is lowerin closed primariesthanin open primaries, even after controlling for other institutionaland election-specific factors that can affect turnout.This result suggests that of the set of voters who may vote for a party's nominee in the general election, the subset that participatesin choosing thatnominee in the party'sprimaryis smallerin closed thanin open primaries. A second set of studies shows thatvoters with strongpartisanties are much more likely to participatein political activities than are other voters. For example, Beck and Sorauf (1992) note that in 1988 total voter turnoutwas approximately50%, but for strong partisansit was over 80%.13 A thirdbody of work establishes that one group of party elites-convention delegates and caucus participants-have more extremeissue positions thanthe generalelectorate. Buel and Jackson (1991) review studies of national convention delegates thatconsistentlyshow thatthese delegateshold moreextremeviews, with Democratssubstantiallymore liberaland Republicansmore conservative,than the averagevoter. Abramowitz,Rapoport,and Stone (1991) study participants at the 1988 Iowa caucusesandthe 1984 Iowa, Michigan,andVirginiacaucuses andshow thattheseparticipantsaresignificantlymoreextremethanthe general and Ordeshook, 1972; Wittman 1977, 1983, 1991; Aldrich, 1983; Calvert, 1985; Aldrich and McGinnis, 1989; Londreganand Romer, 1993). 11. Morton(1993) shows in laboratoryelections thatwhen such uncertaintyexists, policy divergence does indeedoccur. 12. These models use the partyelite approachto explain the observed policy divergenceof the two parties. It should be noted that thereare also a numberof theoreticalvoting models in which partiesor candidates are primarilymotivatedby electoral considerationsand policy divergence occurs. In those models, equilibriamay exist with partiesor candidateschoosing divergentpolicy positionsundera varietyof assumptionsandconditions(see, for example,Palfrey,1984;Bernhardt andIngberman,1985;Austen-Smith,1987;Morton,1987;CameronandEnelow, 1992;Feddersen, 1992; Ingbermanand Villani 1993). 13. See Beck and Jennings(1979, 1984) andAbramson,Aldrich,andRhode(1991) foradditional studies of partisanpolitical participation. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 310 TheJournal ofLaw.Economics. &Organization. V14N2 electorate.Togetherthesestudiessuggestthatvoterswhoparticipate in closed primariesarelikely to be a subsetof the electoratewith strongpartisanties andextremepolicypreferences.Thisimpliesthattheidealpointof theclosed fromtheideal primaryelectionmedianvoteris likelyto divergesubstantially pointof the generalelectionmedianvoter. By competingfor theirparties' nominations, then,candidatesnominatedin closedprimariesareexpectedto be moreextremethancandidates nominated in otherprimarysystems. 2.2 MoreOpenPrimaries Whilemost existingtheoreticalandempiricalanalysesof primaryelections assumeclosedprimaries, we canapplythelogic of thoseanalysesto generate about voter preferencesandcandidatepositionsin otherprimary hypotheses as well. The key insightof the extanttheoryis thatprimariesaffect systems theidentityof themedianvoterin theparties'primaries voter by constraining When members with extreme onlyparty participation. policypositionsparticipatein thenomination process,as in thetheoretical analysesdescribedabove, thepreferencesof the medianvoterin a party'sprimarycandivergesubstanof themedianvoterin thegeneralelectorate.When tiallyfromthepreferences and/ornew votersparticipate in the nomination nonmembers, independents, the the in of median voter a process, preferences party'sprimarymaybe closer of themedianvoterin thegeneralelectorate. to thepreferences Primarysystemrulesaffectthe identityof the medianvoterin theparties' formsof primariesby affectingthe cost to votersof engagingin particular strategicbehavior. Specificallythey affectthe ease with whichvoterscan engagein crossovervoting(i.e., votingin the primaryof a partywithwhich thevoterdoesnotnormallyidentify).Crossover votingcaneitherbe "sincere," in whichvotersvotefor theirmostpreferred candidatesin theotherparty,or candidates inwhichvotersvoteforlesspreferred whosenomination "strategic," in theirown to a morepreferred wouldprovidea strategicadvantage candidate party.14Whenvotersengagein sincerecrossovervoting,moderatesfromthe otherpartycrossoverandmovetheprimaryelectoratemedianvoter'sposition closerto thatof the generalelectoratemedianvoter.Whenvotersengagein strategiccrossovervoting,extremistsfromthe otherpartycross over. The effectof strategiccrossoveris moredifficultto anticipate.If a relativelysmall numberof votersengagein strategiccrossover,the idealpointof the median voterin thatparty'sprimarybecomesmoremoderate, butif strategiccrossover electoratebecomesmore themedianvoterin theparty'sprimary is substantial, extreme. Weexpecttheclosednessof a primarysystemto affectvoters'costsof en14. Ourdefinitionof sincere crossovervoting allows for voting for the most preferredcandidate whose ideal point is closest to the voter's and voting for a candidatewhose ideal point is further from the voter's but whose electoralprospectsarebetter. Both of these forms of sincerecrossover voting are distinguishedfrom strategiccrossover voting for a candidatethe voter hopes will lose in the generalelection. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions ElectionSystemsand Representation 311 Primary gaging in crossovervoting. In closed primaries,both types of crossovervoting are costly since voters must resolve to participatein a given party'sprimarya fixed period before the election and register as a memberof that party. The "cost"of crossover voting for closed primaryvoters, then, involves the difficulty of anticipatinghis or her strategicopportunitieslong before the election and reregisteringin the otherparty.We thereforeexpect closed primaryvoters to participateon the basis of theirunderlyingpartisanaffiliationsratherthanthe immediateelectoral circumstances. To the extent that membersof the parties are ideologically distinct, we thereforeexpect the ideal point of the primary electoratemedianvoter in closed primariesto reflect the ideological positions of the party'selite and to divergesubstantiallyfrom the ideal point of the general electoratemedianvoter. In semi-closed primaries,crossovervoting is less costly for voters who affiliate with one party but are either not registeredor are registeredas independents. Those voters can participatein either party's primarywithoutprecommittingto thatparty,therebymaking the cost of both forms of crossovervoting quite low.15 Which form of crossovervoting dominates in semi-closed primariesthereforedependson the ideological positions of independentsand new voters. To the extent that these potentialcrossover voters are close to the ideological center, we expect them to engage largely in sincere crossover voting. To the extent that independentsand new voters are ideological extremists,we expect them to engage largely in strategiccrossover voting. Althoughsurveyevidence suggests thatthereis substantialheterogeneity among independentvoters, especially between "PureIndependents"and partisanleaners, most pure independentsare, by a variety of measures, ideologically moderate(Keith et al., 1992). We thereforeexpect these potential crossovervotersto engage mostoften in sincerecrossovervotingin semi-closed primaries,and for their behaviorto move the ideal point of the primaryelectoratemedian voter closer to the ideal point of the general electoratemedian voter. In open primaries,voters can engage in crossover voting by choosing a party on election day without incurringthe costs of reregisteringwith the other party. However, crossover voters are constrainedto participateonly in that party'sprimary.This constraintmay reduce the attractiveness(i.e., raise the opportunitycosts) of both sincere and strategiccrossover voting for one race since voters cannot then vote for candidatesof their own party in other races. We thereforenote the possibility for both sincereand strategiccrossover voting in open primariesbutcannotanticipate,on the basis of ourtheory,which will be more important.Finally,in blanketandnonpartisanprimaries,the costs of both types of crossovervoting are lowest since voters can choose in which primaryto participatein the voting booth (i.e., they need not incurthe costs of reregistration)and can switch partiesrace by race (i.e., the opportunitycosts of crossovervoting are low). Again, however,we cannotanticipatewhich type of crossovervoting will dominatevoter behavior. 15. As in closed primaries,however,voters who are registeredwith a party must change their registrationpriorto the election. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions ofLavw. V14N2 &Organization. 312 TheJournal Econormcs, A numberof studies have attemptedto measure crossovervoting directly. Ranney(1975), Adamany(1976), Hedlund,Watts,andHedge (1982), Hedlund and Watts(1986), and Wekkin(1988) all examine voting behaviorin Wisconsin's presidentialprimaries,with differingconclusionsabouttheextentto which such crossovervoting occurs.16These studies,however,focus on only one state and are thereforeunable to demonstratethat crossovervoting is substantially greaterin open primarystatesthanin closed primarystates. Jewell (1984) compares limited survey data on voter behaviorin open and closed primarystates and concludes that while voters seem more likely to identify with a partyin a state with closed primaries,"[t]he availableevidence does not supportthe assumptionthatmany votersfrequentlyshift betweenprimariesin open-primary states.'17 Finally, Southwell (1991) finds the prevalenceof "strategicvoting" to be aboutthe same in closed, semi-closed, and open primarystates.18 3. Hypotheses The theoreticalresults summarizedabove can be restatedas a series of empirically testable hypotheses about the consequencesof primarysystem type on election outcomes. * Ho: Primaryelection system has no relationshipto the policy positions of generalelection winners. * HI: Closedprimarysystemswill producemoreextremegeneralelection winners,relativeto theirconstituencies'generalelection medianvoter, thanmore open primarysystems. The mainhypothesisis thatclosed primarieswill producegeneralelection winners whose policy positions diverge substantiallyfrom their district'sgeneral election median voter. We hypothesize that in the most closed systems, winning candidatepositions will be most extreme,and that in more open systems they will be more moderate. However,given the differentcosts of crossover voting createdby the varioussemi-closed, open, blanket,and nonpartisanprimaryelection systems, the relationshipbetweenclosedness andextremitymay not be linear. Since we expect sincere crossovervoting to dominatein semiclosed primaries,we expect the primaryelectoratemedian voter to be closer to the general electoratemedian voter in those primariescomparedto closed primariesandfor those votersto vote for the most moderatecandidatesin their 16. As Wekkin(1988) notes, some of the differencesin results are a consequenceof variations in the methodsused to measurecrossovervoting. He arguesthat independentswho have partisan leanings are often not counted correctly. He also points out that measuringcrossover vote as a percentof the entireprimaryvote understatesits impacton a particularprimarysince such voting tends to be one-sided. 17. Jewell notes that there is evidence of voter shifting in Alaska and Washington,which have blanketprimarysystems. 18. Southwell differentiatesbetween "positive strategic voting" in which a voter votes for a less preferredcandidatebecause her most preferredcandidate'schances of winning are low, and "negativestrategicvoting,"in which a voter votes for a less preferredcandidatein the primaryto increasethe chance of her most preferredcandidatefacing the weakest possible opponent. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions ElectonSystems andRepresentaton313 Primary primaries. Hence we expect the most moderatewinners (relative to the district'sgeneral election median voter) to be elected from semi-closed systems. Since the costs of both sincere and strategiccrossovervoting in open, blanket, and nonpartisanprimariesare low, it is possible thatstrategiccrossovervoting may negate some of the moderatingeffects of sincere crossover voting. We thereforeexpect winnerselected from those systems to be more moderatethan winnerselected from closed primarysystems, but perhapsnot as moderateas semi-closed winners. * H2: Semi-closed primarysystems will producethe most moderategeneral election winners. * H3: Open,blanket,andnonpartisanprimarywinnerswill be more moderate thanclosed primarywinners. 4. Data Analysis We test hypothesesHi, H2, andH3 with datafromU.S. congressionalelections from 1982 to 1990. The variableof interestis thepolicy position of the winning congressionalcandidate,Winneri,,from each congressionaldistricti, i = 1 to 435, at time t, t = 1 to 5. For each of the i districts at each election, we estimate Winneri,as the winning candidate'sADA score averagedover the 2 yearsimmediatelyfollowing theelection, correctedfor abstentions.19We chose this time period because the districtlines were constantthroughoutas a result of the 1982 redistrictingand no states significantlychangedtheirelection laws duringthis period. 19. Therearenumerousways to measurerepresentatives'policy positions. Perhapsthe two most widely used measuresareADA scores (votingindicesconstructedby the AmericansforDemocratic Action) and NOMINATEscores [constructedwith Poole and Rosenthal's(1985) Nominal Threestep EstimationProcedure]. Severalscholars have arguedthatNOMINATEscores are preferable to ADA scores as measures of legislators' policy positions because they are based on a formal spatial model (Poole and Rosenthal, 1985), account for a great deal of variancein roll call votes (Cox and McCubbins,1993), are based on a large numberof roll call votes (Cox and McCubbins, 1993), and are comparableacross years (Groseclose, Levitt, and Snyder, 1998). While we agree that NOMINATEscores may be preferablein some applications,we believe ADA scores are at least as appropriatefor ourcurrentpurposes,for severalreasons. First,ADA scores aredeliberately constructedto reflect a single liberal-conservativedimension. The purposeof the ADA scores is to rateeach representative's"liberalness":' Hence the ADA bases its scores on a set of votes thatare selected specifically because they deal with left-right issues. NOMINATEscores, by contrast,are constructedfromthe full set of (nonunanimous)roll call votes andso, to the extent thatotherpolicy dimensionsunderliea legislature'sagenda,the resultingscores naturallypick up moredimensionality. Furthermore,to the extent thatthese additionaldimensionsare not strictlyorthogonalto the primaryleft-right dimension,our estimatesof the firstdimensionwill be biased. Second, many of the criticismscommonly made of ADA scores do not apply to our application.Most importantly, as Jackson and Kingdon (1989) argue, ADA scores may be inappropriatemeasuresof legislator ideology in statisticalmodels explainingroll call voting behavior.In the analysis that follows, we use the ADA scores as our dependentvariable, so we need not be concerned about introducing bias by explainingvotes with votes. Third,while the distributionalpropertiesof ADA scores may be problematic,particularlyarbitraryscaling, such issues are relevantfor all other voting indices, includingNOMINATEscores. Fourth,we are able to treatabstentionsagnostically by removing them from the ADA's calculations(see Franciset al., 1994). This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions &Organization, 314 TheJournal ofLaw, V14N2 Economics, We employ a multivariatemodel that provides the frameworkfor testing Hi, H2, and H3 against Ho while controllingfor otherfactors that may affect candidatepositions. Conceptuallythe model allows us to directlyestimatethe proximity of winning candidates'policy positions to estimated voter preferences in their districts and to compare this relationshipfor general election winnerselected underdifferentprimarysystems. We hypothesizethatwinning candidatepositionsin elections with closed primarieswill be extremerelativeto the district's(generalelection) medianvoter. We hypothesizethatthe positions of winnersfrom open, nonpartisan,andblanketsystems will be moremoderate relativeto the generalelection medianvoter, and that the positions of winners from semi-closed primarieswill be the most moderate.20 Formally,we describethe base empiricalmodel as follows: = ao + a Semii+ ct2Openi IWinnerldeolitDistrictIdeolil + a3 NP/Blanketi + u i (1) Winnerldeoli,is thewinningcongressionalcandidate'sADA score,corrected for abstentions,as describedabove. Districtldeoli is an estimateof districti's medianvoter'spolicy position. We operationalizeDistrictldeol as the average of the percent in the district voting for Mondale in 1984 and for Dukakis in 1988. Districtldeol thereforemeasuresthe district'sideological composition as the percentof liberalsin the district.21 20. Our cross-sectionalapproachto estimatingthe effects of primaryelection systems on the extremityof legislatorpolicy positionshas severaladvantagesoveralternativeapproaches.Perhaps most importantly,none of the states institutedsignificantchanges in their primaryelection laws during this period (except Connecticut,but as explained in footnote 2, the partieschose not to use semi-closed primariesafter the state allowed them). Thus we are able to treatthe effects of those institutionsas strictlyexogenous. The main disadvantageof our approachis that theremay be other sources of cross-sectional variationin the extremity of legislator policy positions that we inadvertentlyomit from our analysis, leading to potentialomitted variablesbias. Of course, however, such omissions will affect our inferences about the effects of primaryelection laws only if the omitted variables are also correlatedwith the state's electoral institutions(Gujarati, 1995:204-7). An alternativeapproachto estimatingthe effects of primaryelection systems is to comparethe relationshipbetweenlegislatorpositions andconstituencypreferencesbeforeandafter states changedtheir primaryelection laws. While this time-seriesapproachhas the advantageof eliminatingthe possibilityof cross-sectionalvariationin legislatorpolicy positionsacrossstates, it has two importantdisadvantages.From a theoreticalperspective,we can no longer treatelectoral institutionsas exogenous. Infact, since manyof thesame factorsarelikely to affectbothinstitutional change and the natureof representationin a state, unbiasedestimationwouldrequiremodelingthe process of institutionalchange explicitly. And from a practicalperspective,few states actually changed their nominationproceduresin the way requiredto test our theory (i.e., from closed to open, open to closed, etc.). Rather,most of the changes in candidatenominationprocedures that states have implementedsince their initial adoptionof direct primarieshave involved other restrictionson participation(especiallyverbalpledges of allegiance),changesin ballotaccess laws, and uses of conventionsand endorsements. 21. By averagingthe percentin each districtvoting for Mondale in 1984 and Dukakisin 1988, our measureof districtideology is less sensitive to election-specificfluctuationsin districtvoting behaviorthana single election's percentage.As a proxy for the district'smedianvoter'sideology, our operationalizationrequiresthatthe distributionof voterpreferencesin each districtis roughly symmetricandsingle peaked. Ourapproachis comparableto the approachestakenby Eriksonand This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions Election 315 Primay SystemsandRepresentation The independentvariablesin the base model include threedummyvariables designed to capturethe effects of the primarysystem. The dummy variables indicatewhetherthe state uses semi-closed primaries,open primaries,or nonpartisanor blanketprimaries.22The excluded category is the closed primary. Thereforethe dummy variablesare interpretedas the change in the dependent variablefor states with each of the threeprimarysystems, relativeto the effect of having a closed primary. Thus if more open systems lead to more moderate winners, the distance between the winner and the median voter will be less (i.e., the signs will be negative)on Semi, Open, and NP/Blanket. If semiclosed systems produce the most moderaterepresentatives,as hypothesized, the magnitudeof the negativecoefficient will be largeston Semi.23 4.1 EstimationProcedures Several factors complicateestimationof Equation(1). Most significantly,although Winnerldeoland Districtldeol are both boundedby the range [0, 100], it is readily evident that the two variablesfollow very differentdistributions. Winnerldeol,measuredas a winning candidate'sADA score in the following Wright(1993) and Kennyand Morton(1993) to estimatingstate voterpreferencesfrom state vote returns.Otherapproachesto estimatingaggregatevoterpreferencesin the politicalscience literature include employing aggregatestate demographicsas proxies of state voter ideology (Matsusaka, 1995) or estimatingstate voter preferencesdirectly from survey data (Gerber,in press). While estimatingdistrictvoter preferencesfrom survey data would providea more directapproachthan proxying them from district vote returns,existing surveys lack an adequatesampling framework for drawinginferencesaboutaggregatevoter preferencesin differentdistricts. 22. Since the numberofcongressmenelected undernonpartisan,blanket,andthethreevarietiesof semi-closedprimariesaresmall, we combinetheobservationsfromnonpartisanandblanketprimary states,and fromthe threetypes of semi-closed primarystates,respectively.Note thatby comparing the relationshipbetween the ideology of winning candidates and their districts from different primarysystems, our implicit dependentvariableis whethercandidateswith moderatepositions relative to their district's median voter are more likely to prevail in districts with open versus closed primaries. Our data are not sufficient for testing hypotheses about intrapartycompetition underdifferentnominationrules. In other words, we are unable to test whethercandidateswith more moderatepositions relative to the other candidatesfrom their own partyare more likely to prevailin districtswith open versusclosed primaries.This second questionrequiresdataon losers' ideologicalpositions,which areavailableonly on an ad hoc basis, such as when a loser is eventually elected in a subsequentelection. As interestingas the questionof intrapartycompetitionis, it does not addressthe immediateconcern of this article. 23. Of course, it is possible that some of the effects picked up by the primarysystem dummy variablesmay be due not to the independenteffects of the state's primarysystem on representative's policy positions, but ratherto the fact that states with moderatevoters may elect moderate representativesand adopt more open primaryelection systems. In other words, the estimatedrelationshipbetween institutionsand policy positions may be spurious. While election laws may be partiallyendogenous, we note two factorsthatjustify our treatmentof electoral institutionsas largely exogenous. First, citizen preferencesvary a great deal among states with similar primary systems. In other words, it is not just states with moderatevoters, for example, that adopt open primaries-open primarystates may have moderateor extremevoters. Second, we are interestedin the relationshipbetween representatives'positions and voters' positions. Therefore,even if there were some relationshipbetween voter preferencesand primarysystems, this relationshipwould not necessarilytranslateinto a closer (or weaker)correspondencebetween voter preferencesand the representativesthey elect. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions V14N2 ofLaw. 316 TheJournal &Organization, Economics, withonemodeatthelowerendof term,followsa stronglybimodaldistribution therangeandonemodeatthehigherendof therange.Districtldeol,measured as the averageof the district'sMondaleandDukakisvotes,exhibitsa single, betweenthetwo variablesis therefore highlyskewedmode. Therelationship to be and so standard aboutnormality nonlinear, inherently likely assumptions of the errorsareclearlyviolated. The possibilityalso existsthatone of the at differentpoints.24Further,the or botharetruncated variablesis truncated areat differentpoints.25 meansof thetwodistributions Toreducethepossibilitiesof biasorincorrectinferencesdueto violatonsof we generalizeourspecification to allow the standard regressionassumptions, Tocapturethispotential andDistrictldeol toberelatednonlinearly. Winnerldeol nonlinearity,we remove the absolutevalue from the dependentvariable,move totheright-hand sideof theequation Districtldeol byaddingit tobothsides,and This addsecond-andthird-order terms. polynomial polynomialspecification betweenWinnerldeol andDistrictldeol, allowsus to estimatethe relationship for the nonlinearities introduced distributions of the by accounting potential betweenWinnerldeol andDistrictldeol two variables.Thusif therelationship is characterized by a second-order polynomial,we will estimatea significant effecton thesecond-order term.If it is capturedby a third-order polynomial, termas well. we will estimatea significanteffecton thethird-order Table2 reportspreliminary OLS regressionestimatesfor the relationship between Winnerldeoland Districtldeol, Districtldeol2, and Districtldeol3on one thefull dataset.Thedataarestackedsuchthateachobservation represents districtin oneelection,witha maximumof 435 * 5 = 2175observations.26 In eachestimation,the dependentvariableis WinnerIdeolit.The threecolumns of therelationship betweenWinnerldeol specifications reportthreealternative and Districtldeol. betweenWinnerldeolandDistrictldeolreported The bivariaterelationship in column1 is strong,positive,andsignificant.Districtldeolplustheconstant togetheraccountfor 44%of the variancein Winnerldeol. The second-order whiletheaddition effectsestimatedin column2 arealsostrongandsignificant, effectsdoeslittleto improvethemodel.Onthebasisof this of thethird-order of therelationship analysis,we limitourpolynomialspecification preliminary betweenWinnerldeolandDistrictldeolto a second-order polynomial. betweenWinnerldeol andDistricWhilesolvingtheproblemof nonlinearity tIdeol, moving Districtldeol (and Districtldeol2)to the right-handside of the Recallthatthe originaldeequationcreatesanotherproblem:directionality. pendentvariable,IWinnerIdeol- Districtldeoll, was expressedas an absolute value.Thisallowedus to measuredeviationsfromthedistricts'medianvoter's 24. Many legislatorsare clusteredat the endpointswith ADA scores of 0 (very conservative)or 100 (very liberal). However, if there is diversity in ideology among legislators at the endpoints (thatis, some legislatorswith scores of 100 are actually more liberalthan others), then assigning all the same score may attenuatethe variancein distancemeasures. 25. The mean of Winnerldeolis 51.34; the mean of Districtldeolis 43.61. 26. The actualnumberof observationsis fewer (2170) because of missing data. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions PdmaryElectionSystemsand Representation 317 Table2. RegressionCoefficientsRelatingWinner'sADAScore and District'sAverage VoteforMondale1984 and Dukakis1988, U.S. CongressionalDistricts,1982-1990 IndependentVariable DV= Winnerldeol DV= Winnerldeol DV= Winnerldeol -92.12** -133.36** -35.34** Constant (17.18) (5.86) (2.14) 1.99** (.05) Districtldeol Districtldeol2 6.06** (.23) 3.46** (1.05) -.04** .01 (.00) (.02) -.0003** Districtldeol3 (.0001) .45 2171 R2 N .52 2171 .52 2171 ** p < .05, two-taiedtest. StandarderrorsInparentheses. ideal points in either directionwith a single distance measure. By removing the absolute values to allow for the polynomial specification, however, it is now possible that deviations in either directionwill cancel out, leading to the appearanceof no relationship.Therefore,to capturethese directionaleffects, we estimatethe model two differentways. First,we note thatmost Democrats' ADA scores aregreaterthantheirdistricts'estimatedmedianvoter'sidealpoint, and that most Republicans'ADA scores are less than their district'sideology scores. This allows us to capturedirectionalityby simply adding a variableto indicate the winner's party (scored 1 for Democrats, -1 for Republicans)and to interactthe independentvariablesof interestwith this party variable,as in Equation(2). + B3Party,i Winnerldeolit= P0 + f,1Districtldeoli+ jB2Districtldeol2 + p4Semii* Partyi, + ftsOpeni* Partyi, + fBNP/Blanketi * Party,, + u2it (2) Interpretationof the coefficients in Equation(2) is as follows. The coefficients on Districtldeol,Districtldeole,andPartyestablishthe baselinerelationship between the averagerepresentative'sADA score from each partyand his or her district'sideology in a closed primarystate. The coefficients on Semi, Open,andNP/Blanketreflectthe averagedeviationfromthis baselinerelationship for representativesin each primarysystem. Negative coefficientson Semi, Open, and NP/Blanket are interpretedas indicatingless policy divergencein those systems from a member'sdistrictideology-in a negative directionfor Democratsand in a positive directionfor Republicans.Positive coefficientson Semi, Open, and NP/Blanketindicatemore policy divergence. While Equation(2) solves the problems of nonlinearityand directionality, thereremainseveralproblemswith the estimatesit produces. Most importantly, the specificationof Equation(2) masksdifferencesin the effects of primarysys- This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 318 TheJournalof Law,Economics,&Organization. V14N2 tems on Democraticand Republicanmembersof Congress. These differences may arise due to organizationaldifferences in the parties, national partisan forces, and idiosyncraticfeaturesof each party'scandidates. To account for potential partisandifferences, we next estimate Equation (3) separatelyfor Democraticand Republicanwinners: Winnerldeolit= Yo+ yl Districtldeoli + y2 DistrictIdeolI + Y3Semii + Y4Open, + ys NP/Blanketi + u3it (3) of theDistrictldeolcoefficientsis comparableto thatfor EquaInterpretation tion(2). Weinterprettheprimarysystemcoefficientsas follows. ForDemocrats, negative coefficients on Semi, Open, and NP/Blanket mean that Democrats elected underthose systems are more moderate(conservative)thanDemocrats elected underclosed primaries.ForRepublicans,positive coefficientson Semi, Open, andNP/Blanketmean thatRepublicanselected underthose systems are more moderate(liberal)thantheirclosed primarypartisancounterparts. Giventhe panelnatureof the data,with multipletime observationsfromeach cross-sectionalunit(congressionaldistrict),we employanestimationprocedure thatallows us to test and correctfor likely featuresof the errorstructurein the data. In particular,we estimate Equations(2) and (3) with the pooled crosssection time-seriesestimationprocedurein Stata5.0 thatallows for bothserially correlatedandcross-sectionallyheteroscedastic(butindependent)errorsacross observations(i.e., we allow both between and within districtrandomeffects). 4.2 Results Table 3 reportsthe pooled cross-section time-series regressionestimates for Equation(2). All of the regressionanalyses in Tables3 and4 use Winnerldeol as the dependentvariable.The firstcolumnof Table3 reportspooled regression estimates of a baseline model using Districtldeol, Districtldeol2,Party,Semi, Open, and NP/Blanketas the independentvariables. Subsequentcolumns add independentvariablesto controlfor otherpotentiallyimportantfactors. In the base modelin the firstcolumn,thepositivesignificanteffect on Districtldeol indicates that membersof Congress elected from more liberal districts have higher ADA scores and those elected from more conservativedistricts have lower ADA scores. The negativecoefficient on Districtldeol2shows that these effects taperoff at higherlevels of Districtldeol. The positive coefficient on Party shows that Democratshave much higher ADA scores than Republicans. Finally,the significantnegativecoefficientson Semi*Party,Open*Party, and NP/Blanket*Partyare consistent with hypothesis 1, indicatingthatrepresentativesfrom semi-closed, open, nonpartisan,and blanketprimarysystems are more moderatethanrepresentativesfrom closed primarysystems. Columns2 and 3 add independentvariablesto controlfor otherfactorsthat may also affect a winner's ideology. Column 2 adds a dummy variableindicating whetherthe state uses runoffsin additionto theirotherprimaryelection institutions. While the extanttheory describedin this articledoes not address the effects of runoffsdirectly,there are several reasons to believe that runoffs This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions ElectionSystemsand Representation 319 Primary Table3. PooledCross-SectionTime-SeriesRegressionCoefficientsRelatingWinner's ADAScore and Primary SystemType,U.S. CongressionalDistricts,1982-1990 DV = Winnerldeol DV= Winnerldeol DV= Winnerldeol Variable Independent Constant -31.90* (8.45) -28.43** (8.10) -28.82** (8.09) Districtldeol 2.46** (.33) 2.39** (.32) 2.41** (.32) Districtldeol2 -.01** (.00) -.01** -.01** (.00) (.00) Party 26.62** (.75) 29.32** (.80) 30.19** (.84) Semi*Party -7.66** (1.36) -9.34** (1.35) -9.53** (1.43) Open*Party -2.68** (.99) -2.63** (.97) -2.44** (.98) NP/Blank*Party -9.07** (2.60) -5.00* (2.58) -4.81* (2.57) -8.95** (1.01) -3.96* (2.10) Runoff*Party President*Party -1.55** (.40) OpenSeat*Party 2.10 (.81) -5.87** (2.11) South*Party -2.63 (1.79) NewEngl*Party OverallR2 N .75 2170 .77 2170 .77 2170 ** p < .05, two-tailed test;*p < .10, two-taiedtest Standarderrorsin parentheses. may furtheraffect the types of representativeselected. First, runoffs are used primarilyin the South, where both voters and theirrepresentativeshave historically been more conservative.To the extent thatDistrictldeol fails to capture some of the distinctivenessof the AmericanSouth, Runoff may compensate. Second, runoffs are widely believed to create advantagesfor majoritycandidates, who may be more conservativethantheir minoritychallengers.27 The significantnegativecoefficienton Runoff*Partyshows thatmembersof Congresselected in primarysystems with runoffsare, indeed, more moderate. 27. See BullockandJohnson(1992)fora reviewof theempiricalliterature on runoffsandan opposingviewpoint. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 320 TheJournalof Lavi,Economics,&Organization. V14N2 In addition,includingRunoff*Partydecreases the size of the Open*Partyand NP/Blanket*Partycoefficientsandincreasesthe relativesize of the Semi*Party coefficient, consistentwith hypotheses2 and 3. Column 3 adds variablesto capturefeaturesof the individualelections and otherregionalfactors. We see thatrepresentativeselected in presidentialelection years are more moderatewhile those elected from open seats are more extreme. Representativeselected fromthe Southaremoremoderate,even after we controlfor whetherthe stateuses runoffs. The additionof these independent variablesleaves the primarysystem coefficients largely intact, slightly reducing the (negative) effects of Open*Partyand NP/Blanket*Partyand slightly increasingthe (negative)effect of Semi*Party. Table4 reportsthe pooled cross-sectiontime-seriesregressionestimatesseparatelyfor Democratsand Republicans. The estimates in column 1 show that Democraticmembersof Congress elected undersemi-closed, open, nonpartisan, and blanketprimariesare more moderate(conservative)than Democrats elected underclosed primaries. The effects of Open and NP/Blanketare substantivelymeaningfuland statisticallysignificant.However,the negativeeffect of Semi is not significant. When we add controls for whetherthe state uses runoffs,whetherthe memberwas elected in a presidentialelection yearor from an open seat, and whetherthe districtis in the South or New England,the effect on Open and NP/Blanketremainnegativeand the effect on Semi becomes significant.28For the most part,then, the resultsfor Democratsare consistent with hypotheses 1 through3, althoughthose memberselected fromnonpartisan and blanketprimariesare more moderatethanexpected. Columns 3 and 4 report comparableestimates for Republican members of Congress. The estimates in column 3 show that Republican members of Congress elected from semi-closed primarysystems have more moderate (liberal) ADA scores than their closed primarycounterparts.However,those elected from open, nonpartisan,and blanketprimarystates have moreextreme (conservative)ADA scores, on average,contraryto expectations.The negative effect on Openpersistseven once we controlfor runoffs,election-specificvariables, andregion,while theeffect on NP/Blanketbecomespositive(as expected) and significant.Finally,the effect on Semi is large,positive, and significant,as hypothesized,but smallerthanthe effect on NP/Blanket.29 28. In the full model for Democrats(column 2), we also include a dummy variablefor Washington state. Exclusion of this variableleaves the coefficients largelyunchanged,but the effect of NP/Blanketis not significant. Similarly,in the full model for Republicans,we include a dummy variablefor Louisiana. 29. To test the robustnessof our results,we reranthe analysis using the winningcandidate'sWNOMINATEscore (firstdimension)as the dependentvariable.In both the combinedand separate partyanalyses, the primarysystem variableswere all the same sign and relativemagnitudeas in the analysis presentedin Tables 3 and 4. Only one primarysystem was statisticallysignificantby traditionalstandards,however,suggesting thatthe NOMINATEscores containmore variabilityin individuallegislatorvoting behaviorthan the ADA scores. Complete results of the supplemental analyses are availableby requestfrom the authors. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions Primary ElectionSystemsand Representation 321 Table4. PooledCross-SectionTime-SeriesRegressionCoefficientsRelatingWinner'sADA Score and Primary SystemType,Democratsand Republicans,U.S. Congress, 1982-1990 Democrats Republicans Independent DV= WinnerldeolDV= WinnerideolDV= WinnerldeolDV= Winnerldeol Variable 16.29 23.45 -2.14 -36.57** Constant (11.11) (16.94) (18.49) (12.79) -1.13 (.94) -1.35 (.86) -.02** .03** .03** (.00) (.01) (.01) -.39 (3.00) -6.72** (2.79) 13.12** (3.18) 9.05** (2.98) Open -7.10** (2.11) -3.03* (1.82) -4.18** (1.89) -3.03* (1.77) NP/Blank -13.81** (5.17) -13.69** (5.98) -.83 (4.63) 12.13** (5.74) Districtldeol 3.75** (.47) 2.62** (.40) Distinctldeol2 -.02** (.00) Semi Runoff -8.58** (3.62) -3.97 (3.85) President -3.53** (.51) -1.40** (.51) OpenSeat 1.97* (1.16) -1.38 (.96) -12.80** (3.68) -.54 (3.70) NewEngl 10.19** (3.66) 21.55** (3.74) WA/LA 15.57* (8.46) -20.70** (8.28) South R2 N .39 1302 .55 1302 .37 868 .48 868 test. **p < .05, ht;otailedtest;*p < .10, two-tailed Standarderrorsin parentheses. 5. Conclusions We find strongsupportfor the hypothesisthatU.S. representativesfrom states with closed primariestake more extremepolicy positions, relativeto theirdistrict'smedianvoter,thanrepresentativesfrom stateswith moreopen primaries. We take this as evidence that primaryelection laws systematicallyaffect the types of candidateselected and the choices they make once in office. Not all closed primarysystems are created equal, however. Our results indicatethatsemi-closedprimarysystemsthatallow new votersorindependents to participatein the candidatenominationprocessproduceRepublicanwinners This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions 322 TheJoural of Law,Economics,&Organization, V14N2 whose positions are more liberaland Democraticwinnerswhose positions are moreconservativethantheirclosed primarycounterparts.This patternsuggests thatincentivesmay exist for semi-closedprimaryvotersto registerstrategically as independentsin orderto participatein whicheverparty'sprimaryaffordsa betteropportunityto affect the general election outcome. This possibility has real practical implications. Several states are now considering or have just adoptedchanges in their primaryelection laws. Our researchillustratesthat the importantfeatureof a system is not just whetherit is nominally open or closed; rather,the specific institutionaldetails of the system may be critically important. Not all open primariesare created equal, either. Our results indicate that nonpartisanand blanketprimariesproducethe most moderategeneralelection winnersfrom both parties. We interpretthis result as compelling indirectevidence thatwhile bothstrategicandsincerecrossovervotingarepossible in these very open primarysystems, sincere crossoverby moderatevoters dominates and leads to the election of moderatecandidatesfrom both parties. This conclusion is consistentwith preliminaryanalyses by Alvarez and Nagler (1997) which show minimallevels of strategiccrossovervoting. Our results also have implications for subsequenttheoreticalanalyses of multistageelectoral processes. They underscorethe importanceof modeling the institutionaldetailsof electionsystemsandthe effects thoseinstitutionshave on voter and candidateincentivesin both the registrationand voting stages. Finally,while ourresearchdemonstratesthatdifferencesin primaryelection laws can have significanteffects on the types of candidateselected, it is unclear which primarysystem is desirablefrom a normativepoint of view. Thatis, the empiricalresults supportthe conclusion that candidatescloser to the median voter are more likely to be elected in districts with semi-closed, open, nonpartisan,and blanketprimarysystems. Thus if closeness to the median voter is viewed as desirable,then open primariesare preferableto closed primaries. However,in districtsin which importantgroupshavepreferencesdifferentfrom the medianvoter,more open primariesmay lessen the probabilityof these voters' preferencesreceivingrepresentation. References Abramowitz,Alan I., Ronald B. Rapoport,and WalterJ. Stone. 1991. "Up Close and Personal: The 1988 Iowa Caucuses and PresidentialPolitics,"in EmmettH. Buel Jr. and Lee Sigelman, eds. Nominatingthe President.Knoxville: Universityof TennesseePress, pp. 42-71. Abramson,Paul R., John H. Aldrich, and David W. Rhode. 1991. Change and Continuityin the 1988 Elections, rev. ed. Washington,D.C.: CQ Press. Adamany,David. 1976. "CrossoverVotingandthe DemocraticParty'sReformRules,"70American Political Science Review536-41. Aldrich,John H. 1980. Beforethe Convention.Chicago: Universityof Chicago Press. . 1983. "A Downsian SpatialModel with PartyActivism,"77 AmericanPolitical Science Review974-90. , and Michael D. McGinnis. 1989. "A Model of PartyConstraintson OptimalCandidate Positions,"12 Mathematicaland ComputerModelling437-50. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions Election andRepresentation 323 Primary Systems Alvarez, R. Michael, and JonathanNagler. 1997. "Analysisof Crossoverand StrategicVoting." Social Science WorkingPaper 1019, CaliforniaInstituteof Technology. Aranson, Peter H., and Peter C. Ordeshook. 1972. "SpatialStrategiesfor SequentialElections" in R. G. Niemi and H. F. Weisberg,eds. ProbabilityModels of Collective Decision Making. Columbus,Ohio: CharlesE. Merrill. Austen-Smith,David. 1987. "InterestGroups,CampaignContributions,and ProbabilisticVoting," 54 Public Choice 123-39. Bartels,LarryM. 1988. PresidentialPrimaries.Princeton,N.J.: PrincetonUniversityPress. Beck, Paul Allen, and M. Kent Jennings. 1979. "PoliticalPeriods and Political Participation,"73 AmericanPolitical Science Review737-50. and M. KentJennings. 1984. "UpdatingPolitical Periods and Political Participation,"78 AmericanPolitical Science Reviewv198-201. , and FrankJ. Sorauf. 1992. PartyPolitics in America,7th ed. New York:HarperCollins. Bemhardt,M. D., and Daniel E. Ingberman.1985. "CandidateReputationsand the 'Incumbency Effect';' 27 Journalof Public Economics47-68. Bott, AlexanderJ. 1990. Handbookof United States Election Laws and Practices. New York: GreenwoodPress. Buel, EmmettH., Jr.,and John S. JacksonIII. 1991. "TheNational Conventions:Diminishedbut Still Importantin a Primary-DominatedProcess;"in EmmettH. Buel Jr.and Lee Sigelman, eds. Nominatingthe President.Knoxville: Universityof TennesseePress, pp. 213-49. Bullock, CharlesS., III, and Loch K. Johnson. 1992. RunoffElections in the UnitedStates. Chapel Hill: Universityof NorthCarolinaPress. Calvert, Randall. 1985. "Robustness of the Multidimensional Voting Model: Candidates Motivations,Uncertainty,and Convergence,"29 AmericanJournalof Political Science 69-95. Cameron, Charles M., and James M. Enelow. 1992. "AsymmetricPolicy Effects, Campaign Contributions,and the SpatialTheoryof Elections" 16 Mathematicaland ComputerModelling 117-32. Coleman,James S. 1971. "InternalProcesses GoverningPartyPositions in Elections;"11 Public Choice 35-60. . 1972. "ThePositions of Political Partiesin Elections;' in R. Niemi and H. Weisberg,eds. ProbabilityModels of CollectiveDecision Making.Columbus,Ohio: CharlesE. Merrill. Cox, GaryW., and MathewD. McCubbins. 1993. LegislativeLeviathan.Berkeley: Universityof CaliforniaPress. Downs, Anthony. 1957. An EconomicTheoryof Democracy.New York:Harper& Row. Eldersveld,SamuelJ. 1982. Political Parties in AmericanSociety. New York:Basic Books. Epstein,Leon D. 1986. Political Partiesin the AmericanMold.Madison: Universityof Wisconsin Press. Erikson,RobertS., and GeraldC. Wright. 1993. "Voters,Candidates,and Issues in Congressional Elections;"in LawrenceC. Dodd and Bruce I. Oppenheimer,eds. CongressReconsidered,5th ed. Washington,DC: CQ Press, pp. 91-115. Feddersen,TimothyJ. 1992. "AVotingModel ImplyingDuverger'sLaw andPositiveTurnout,"36 AmericanJournalof Political Science 938-62. Francis, Wayne, Lawrence W. Kenny, Rebecca B. Morton, and Amy B. Schmidt. 1994. "RetrospectiveVoting and Political Mobility,"38 AmericanJournal of Political Science 9991024. Gerber,ElisabethR. In press. The PopulistParadox: InterestGroupInfluenceand the Promise of Direct Legislation.Princeton,N.J.: PrincetonUniversityPress. , Rebecca B. Morton,and Thomas A. Rietz. 1994. "MajorityRequirementsand Minority Presentedat the AnnualPoliticalMethodologyMeetings,Madison,Wisconsin. Representation;" Grofman,Bernard,andThomasL. Brunell. 1997."Explainingthe IdeologicalDifferencesBetween the Two SenatorsElected from the Same State: An InstitutionalEffects Model."Researchnote, Universityof California,Irvine. Groseclose, Tim, Steven D. Levitt, and James M. Snyder,Jr. 1998. "AnInflationIndex for ADA Scores,"workingpaper,Ohio State University. Gujarati,DamodarN. 1995. Basic Econometrics.3rd ed. New York:McGrawHill. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions V14N2 324 TheJoural of Law.Economics,&Organization, Hedlund,RonaldD., MeredithW. Watts,andDavidM. Hedge. 1982. "Votingin an OpenPrimary," 10 AmericanPolitics Quarterly197-218. , andMeredithW. Watts.1986. "TheWisconsinOpenPrimary,1968to 1984:"14American Politics Quarterly55-73. Hotelling, H. 1929. "TheStabilityof Competition,"39 EconomicJournal41-57. Ingberman,Daniel E., and John Villani. 1993. "An InstitutionalTheory of Divided Government and PartyPolarization,"37 AmericanJournalof Political Science 429-71. Jackson, John E., and John Kingdon. 1992. "Ideology, Interest Group Scores, and Legislative Votes,"36 AmericanJournalof Political Science 805-23. Jewell, Malcolm. 1984. Partiesand Primaries: NominatingState Governors.New York:Praeger. Keith, Bruce E., David B. Magleby, Candice J. Nelson, Elizabeth Orr, Mark C. Westlye, and Raymond E. Wolfinger. 1992. The Myth of the IndependentVoter.Berkeley: University of CaliforniaPress. Kenny, Lawrence W., and Rebecca B. Morton. 1993. "Representation,Parties, and Policy Divergencein the U.S. Senate,"workingPaper,Universityof Iowa. Londregan,John,andThomasRomer.1993. "Polarization,Incumbency,andthe PersonalVote,"in W. Barnett,M. J. Hinich,andN. J. Schofield,eds. Political Economy:Institutions,Competition, and Representation.Cambridge:CambridgeUniversityPress. Matsusaka,JohnG. 1995. "FiscalEffects of the VoterInitiative:Evidencefromthe Last30 Years," 103 Journalof Political Economy587-623. Morton,Rebecca B. 1987. "A GroupMajorityVotingModel of Public Good Provision,"4 Social Choice and Welfare117-31. . 1993. "IncompleteInformationand Ideological Explanationsof PlatformDivergence," 87 AmericanPolitical Science Review382-92. Palfrey,ThomasR. 1984. "SpatialEquilibriumwith Entry,"51 Reviewof EconomicStudies 139-57. Poole, KeithT., andHowardRosenthal.1985. "ASpatialModel forLegislativeRoll Call Analysis," 29 AmericanJournalof Political Science 357-84. Ranney,Austin. 1975."TurnoutandRepresentationin PresidentialPrimaryElections'"66American Political Science Reviewv21-37. Riker,William H. 1983. LiberalismAgainst Populism.San Francisco: V.H. Freeman. Southwell, PriscillaL. 1991. "Openversus Closed Primaries:The Effect on StrategicVotingand CandidateFortunes,"72 Social Science Quarterly789-96. Weissberg, Robert. 1978. "Collective vs. Dyadic Representationin Congress," 72 American Political Science Review53547. Wekkin,GaryD. 1988.'The ConceptualizationandMeasurementof CrossoverVoting,"41 Vestent Political Quarterly105-14. Wittman,Donald A. 1977. "Candidateswith Policy Preferences:A DynamicModel," 14 Journal of EconomicTheory 180-89. . 1983. "CandidateMotivations: A Synthesis of Alternatives,"77 American Political Science Review 142-57. __ . 1991. "SpatialStrategiesWhen CandidatesHave Policy Preferences,"in J. Enelow and M. Hinich, eds. Advances in the Spatial Theoryof Voting.Cambridge:CambridgeUniversity Press, pp. 66-98. This content downloaded from 169.226.11.193 on Mon, 5 Aug 2013 20:55:47 PM All use subject to JSTOR Terms and Conditions
© Copyright 2025 Paperzz