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

Economics, Issues and the Perot Candidacy: Voter Choice in the 1992 Presidential Election Author(s): R. Michael Alvarez and Jonathan Nagler Reviewed w...
Author: Ashlie White
0 downloads 0 Views 2MB Size
Economics, Issues and the Perot Candidacy: Voter Choice in the 1992 Presidential Election Author(s): R. Michael Alvarez and Jonathan Nagler Reviewed work(s): Source: American Journal of Political Science, Vol. 39, No. 3 (Aug., 1995), pp. 714-744 Published by: Midwest Political Science Association Stable URL: http://www.jstor.org/stable/2111651 . Accessed: 06/11/2012 19:54 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 [email protected].

.

Midwest Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to American Journal of Political Science.

http://www.jstor.org

Economics,Issues and thePerot Candidacy: VoterChoice in the 1992 PresidentialElection* R. Michael Alvarez, CaliforniaInstituteof Technology JonathanNagler,University of California,Riverside Theory:Theoriesofpresidential elections(economicvotingandspatialissueand ideologymodels),combined withthepopularexplanation of "angryvoting,"are usedto accountforvoterchoicein the1992presidential election. Hypotheses: Voterchoicein thisthree-candidate raceis a function of economic issueandideological perceptions, ofvoters positions andcandidates, orvoteranger. Methods:Multinomial probitanalysisof 1992NationalElectionStudiesdataincludingindividual-specific andalternative-specific variables. Simulations basedon counterfactual scenariosof ideologicalpositionsof thecandidatesand of voter of theeconomy perceptions Results:The economywas thedominant factorin accounting forvoterdecisions in 1992,andClinton, notPerot,wasthebeneficiary ofeconomicdiscontent. While issues(mainly abortion) andideology didplaysomerole,Clinton wasnotperceived bytheelectorate as a New Democrat. We findlittlesupport forthehypothesis of angryvoting.Last,PerottookmorevotesfromBushthanfromClinton.

Introduction The influenceof different factorson presidentialelectionshas been a researchtopic in politicalscience forthe greaterpartof thiscentury,and fascinatesthe media and the public,especiallyduringelectionyears.The leadingcandidatesfordetermining electionoutcomesare the stateof the economy,thepositionsof thecandidatesand voterson theissues, and the effectiveness of thecandidates'campaigns.The widespreadeconomicdis*Earlierversions ofpartofthisresearch werepresented atthePoliticalMethodology SummerConference, Tallahassee, Florida, July1993,andattheannualmeetings oftheSouthern PoliticalScienceAssociation, Savannah, Georgia,November 1993.We thank LarryBartels, Beavis,NealBeck,MichaelBerkman, Jeff Billandal, Dubin,JohnFerejohn, CharlesFranklin,Dave Grether, WendyHansen,SimonJackman, Jonathan Katz,RodKiewiet, GaryKing, Dave Lanoue,JanLeighley, BrianLoynd,DougRivers, BrianRoberts, Dave Romero, Larry Annette Rothenberg, andparticipants in seminars Steinacker, at UCSD, Caltech,andPenn Statefortheircontributions, andAbbyDelmanforherassistance. Alvarezthanks theOlin Foundation forsupport of his research. Replication Note:The data usedin thisanalysis weregathered bytheNationalElectionStudiesandmadeavailablebytheInter-University Consortium forPoliticalandSocial Research. Therecodedsubsample usedinouranalysis, documentation abouttherecodingprocedures, and theGAUSS code forthemultinomial probitmodelcan be obtainedfromtheICPSR. AmericanJournalofPoliticalScience,Vol. 39, No. 3, August1995,Pp. 714-744 (?) 1995bytheBoardof RegentsoftheUniversity of Wisconsin System

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

715

tressin theearly1990sled manyto believethatthenationaleconomy was ofprimeimportance inaccounting forGeorgeBush'sdefeatin 1992. Alternatively, somearguethatthistransition of poweroccurred onlybecause theDemocratswereable to present a New Democratticketwhich was moderate enoughto appealto Democratswhohad previously supported tothesetwointerpretations ReaganandBush.Inaddition concentratingon majorpartycandidates, thestrong ofRoss Perothas been showing interpreted as evidencethatmanyvoterswereangryat Washington and eagerto breakfromthestatusquo. In thispaperwe explorethreeprimary explanations ofthe1992election.First,we considertheeffectof theeconomy.Second,we consider theimpactof issuesand ideology.We examineboththegeneralliberalconservative issue dimension and severalspecificissuesexpectedto be important tovoters. Third,weexaminewhether ornotPerot'sstrong showingwasduelargely tothelevelofangerintheelectorate. Andwe introduce a methodological techniquenew to analysesof elections-multinomial of a three-candidate probit-tohandlethecomplexity racein a waythat is consistent withthesubstantive questions beingexamined. The Economy

The first andmostpopularofthethreeprevailing wisdomsaboutthe 1992 electionis the "it's theeconomy,stupid"schoolof thought. The electoral oftherecent stateoftheeconomy significance hasbeenwelldocumentedin politicalscienceresearchon elections, whichhas shownthat votersevaluatethepriorperformance oftheeconomyundertheopposing partiesand choosethepartywhichhas thebesteconomicrecord,albeit therecentpastmoreheavilythanthedistant weighing past(Fiorina1981; Kiewiet1983;Markus1988;Rosenstone 1984;Tufte1978). Thisretrospective economicvotingmodelsuggeststhatthiselection was a referendum on thelackluster of theeconomyunder performance attheendoftheworstfour-year GeorgeBush.The 1992electionoccurred stretch ofeconomic inmostvoters'memories, withdisposable performance a nettotalofonly1% duringBush'sterm.In percapitaincomegrowing thetwopreviousReagantermshad generated contrast, netincreasesof 8.5% and6.6%,andCarter'stermhadseenan increaseof7.3% (U.S. Census Bureau1993).Buttheretrospective as a votingmodelis inadequate simpledecisionruleforvotersin the1992election.First,votershadtwo choicesif theywereto voteno on Bush:Clintonor Perot.And second, votershadvirtually no wayto tellhowtheeconomywould retrospective haveperformed hadPerotbeentheincumbent, sincehehadnopriorrecord ofmacro-economic tocampaign performance upon.Thus,theretrospective modelof votingoffers no guidanceas to howvotersdissatisfied withthe

716

Nagler R. MichaelAlvarezand Jonathan

economywouldchoosebetweenClintonand Perot.Ouranalysisenables whobasedtheirdecisionona negative ustoexaminehowvoters evaluation of theeconomychosebetweentheremaining twonon-incumbent candidates. Ideology and Issues

A secondpopularaccountof theelectionis thatBushwas unableto smearClintonwiththe"L" word(liberal)as he didto Dukakisin 1988, toBush'sloss.Proponents ofthisaccountinterpret andthatthiscontributed theelectionas vindication ofClinton'sacumenintakingcorrect positions on keyissues,andofhiscampaign'sskillin avoidingbeingcharacterized theelectorate did notwant.This accountimplies by Bushas something That thatideologyandissuesplayedsubstantial rolesinthe1992election. electionsis notin ideologyandissueshaveimportant rolesinpresidential andStimson1980;Jackson1975;Key 1966;Page and dispute(Carmines thecontempoBrody1972;Page andJones1979;Pomper1972).Rather, andon raryissue-voting literature hasfocusedonhowmuchissuesmatter, whichissuesmatter in different elections(Abramson, Aldrich, andRohde 1983,1987,1990,1994). Theusualformulation ofissuevoting followsthespatialmodelofvotclosesttothemselves ontheissues ing,inwhichvoters choosethecandidate thatvotersarepre(Downs1956;EnelowandHinich1984).Thisrequires ontheirpositions on several sentedwithcandidates clearlydistinguishable thatvotersare majorissues(Shepsle1972;Page 1978).Yet,to theextent notcertainof theseissuepositions of thethreecandidates, theymaynot be able to employtheseissuesin theirdecisions(Alvarez1992).Clearly, ofthismodelis problematic withthepresenceof Ross Perot. application Foritrequires thatvoters theissueplacement ofa third candidate determine whopromised severeandidentifiable changesinfiscalpolicy,butwas unon manyotherissues.Therewereofcourseotherfacclearandunknown to andtheirattempts tors-suchas themediaexposureofthecandidates information tothevoters-whichworkedto disseminate aboutthemselves makesucha votingrulemoreplausible.The 1992electionis one of the in whichtwoof thecandidates fewin recenthistory had theircampaign In theanalysis andpledgespublished andforsale atbookstores. platforms thevoters'choicesofcandidates, belowwe showwhichissuesinfluenced andmeasuretheimpactof severalissues. The Perot Influence

the1992presidential election Thethird pieceoffolk-wisdom regarding a horde of alienated turned off Washis thatitwasinfluenced voters by by with and with as gridlock, ington, fed-up politics usual,disgusted partisan

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

717

thestatusquo.Theangry-voter hypothesis seemstobe seekingtooverturn Forexample, Germond coverageoftheelection. a favored oneinanecdotal andWitcover titledtheir1993accountoftheelectionMad as Hell: Revolt has also beenfueledby inat theBallotBox, 1992. This interpretation in 1992.Allegedly tovote voterswereinspired creasedturnout theseangry the quo choice,andmayhaveprovided bytheavailability ofan anti-status We believeit is important to distinguish angry basisforPerot'ssupport. alternative policychoices.We think votersfromissuevoterswhoprefer theseangryvotersis as anti-status thata moreprecisewayto characterize or anti-incumbent voters. quo voters, inthepresexplanations between thesethreealternative Distinguishing bothmethodological ence of a seemingly viablethirdcandidatepresents A three-candidate raceis forpoliticalscientists. andtheoretical challenges thana twotheoretically andempirically tounderstand muchmoredifficult oftheusualmodelswe apply-from candidate race,sincetheassumptions two-candidate to probitand logit thestandard spatialmodelof elections, andinferences drawnfromthem econometric models-maybe incorrect, usually Spatialmodelsof electionsin one dimension maybe erroneous. withthreecandidates the reaching equilibrium; dependon twocandidates is muchless tractable. And economicvotingmodelsare based situation betweentwocandidates. Finally,theusual logitand uponcomparisons techniques arebasedon a binarychoicefacingthevoter. probit estimation Therehavebeenseveralimportant worksinpoliticalsciencethatdealwith three-candidate elections(Converseet al. 1969; Rosenstone1984). We withtheeconometric breaknewground, approachwe use here. however, witha We simultaneously examineeach of the above hypotheses interestmodeloftheelection. We demonstrate several multinomial probit Voter'sopinionsaboutthestateof ingfindings. First,itwas theeconomy. different thantheywere thenationaleconomyin 1992weredramatically on theirvotein 1988,andourestimates showthatthishada largeeffect ofthe choice.Second,we demonstrate thatwhiletheideologicalpositions to thevoters,voters'perceptions of thetwo candidateswereimportant werevirtually candidates ontheliberal-conservative dimension major-party that no matter which from And we demonstrate 1988. plausible unchanged we examinetheresultoftheelecofthecandidates ideologicalplacement tionwouldnothavechanged.However,we showthatissuesdid matter. themost:the Perotvoterswereinfluenced bytheissuePerotemphasized hada surprisingly Andwe showthatBush'sposturing on abortion deficit. thatwhilePerotmayhavebeenespewe demonstrate largeimpact. Finally, ciallyappealingto angryvoters,it remainsforsomeoneto demonstrate in whatthesevoterswereangryabout.Forwe showthatvotersinterested ofPerotthan reform (termlimits)werenomoresupportive antigovernment

718

Nagler R. MichaelAlvarezand Jonathan

andthatvoterswhowereupsetabouttheeconomywereno othervoters, sincesuchvoterswentoverPerotthanothervoters, morelikelytosupport in Clinton'sdirection. whelmingly A FirstLook at the 1992 Election examinethe modelwefirst Beforeproceeding toanalyzea multivariate evaluation of their downbyseveralfactors: broken choicesmadebyvoters, andof thenational thechangein thepastyearof theirpersonalfinances in theirgender,theirvote-choice economy,theirpartisanidentification, Thedata andabortion. thedeficit, on termlimits, 1988,andtheirpositions we use arefromtheNationalElectionStudy(Miller,Kinder,andRosenstone1993). economicevaluations and It is therelationship betweenrespondents' inTable 1. The morelikelyrespontheirvote-choice thatis moststriking ownpersonal finances, dentsweretonegatively evaluatethechangeintheir theless likelytheywereto voteforBush;whiletheoppositewas truefor whenexamining Thissamerelationship is evenmorepronounced Clinton. of evaluations of thenationaleconomyandtheirlikelihood respondents' aboutthis votingforBush.Herewe also see hintsof one robustfinding withthenational economydidnotturnto election:thosemostdissatisfied voterswhofelt Perot.ClintonandPerotsplitalmostequallythenon-Bush better thenationaleconomyhadgotten (15.8% forClintonand 13.2%for Perot).But amongnon-Bushvoterswho feltthenationaleconomyhad choice relativeto Perot gottenworse,Clintonwas the overwhelming (55.4% forClintonand 18.6%forPerot). for identification hadtheexpectedimplications Notsurprisingly, party more bothwerethechoiceofsignificantly thetwomajor-party candidates: higher than50% of theirownpartisans, thoughBushhad a significantly ratethanClinton(29.5% versus21.1%). The Perotresultsare defection from thanfrom partialsoas expected: hedrewmorestrongly independents sans;and we see-in linewithBush's weakholdon voters-thatPerot thanDemocrats. But did betterin an absolutesenseamongRepublicans thesameshare Perotpickedup approximately forbothsetsof partisans of defectors: 68% of Republicandefectors (135 outof 198) and 64% of defectors Democratic (114 outof 178). A significant factorin recentpresidential votinghas beenthegender TheDemofrom1992aresomewhat results surprising. gap.Yet,thegender 19 gendergap is clearlyvisible;withClintonrunning cratic-Republican pointsahead of Bush amongwomencomparedto only8 pointsahead amongmen.Yet Bush's shareof the voteis constantacrossmenand women:it is theClinton-Perot splitthatchangesacrossgenders.Clinton runssignificantly better amongwomenthanmen;and Perotrunssignifi-

719

VOTER CHOICE IN THE 1992 PRESIDENTIAL ELECTION

Table 1. Vote-ChoiceBy Economics,Ideologyand Anger in the 1992Election

Personalfinances: Better Same Worse Nationaleconomy: Better Same Worse Partyidentification: Republican Independent Democrat Votechoicein 1988: Bush Dukakis Did notvote Gender: Men Women Termlimits: Favor Oppose Deficit: Not important Important Abortion: Pro-Life Onlyrape Whenneeded Pro-Choice EntireSamplea

Perot

Clinton

Bush %

N

%

N

%

N

42.7 38.6 21.5

217 225 121

38.2 46.3 58.1

194 270 327

19.1 15.1 20.4

97 88 115

71.1 51.3 26.0

54 196 309

15.8 30.9 55.4

12 118 659

13.2 17.8 18.6

10 68 221

70.2 22.9 7.6

466 33 64

9.5 41.0 78.9

63 59 667

20.3 36.1 13.5

135 52 114

55.0 5.2 29.5

410 25 80

23.6 83.1 50.6

176 403 137

21.4 11.8 19.9

160 57 54

34.8 33.3

271 293

42.5 52.5

331 462

22.6 14.2

176 125

35.5 24.2

435 64

45.6 59.2

559 157

18.9 16.6

232 44

32.9 36.9

342 176

51.3 40.3

534 192

15.9 22.9

165 109

47.5 46.3 42.2 22.3 34.2

66 202 100 180 564

40.3 37.8 39.2 57.4 47.8

56 165 93 464 793

12.2 15.8 18.6 20.3 18.2

17 69 44 164 301

listedarerowpercentages. Note:Percentages eachquestionlistedincolumn whoanswered arebasedon all respondents aThesenumbers theirvotechoice. one,as wellas reporting

better amongmenthanamongwomen.SinceonewouldexpectBush cantly betteramongmenthanamongwomen,thissuggeststhat to be running ofPerotamongmencameat Bush'sexstrong performance therelatively pense. ofvotersfrom1988to 1992reveals thechangeinbehavior Examining 45% First, originated. factsaboutwherePerot'ssupport tworather striking

720

Nagler R. MichaelAlvarezand Jonathan

himin 1992.AndsecBushin 1988deserted ofthevoterswhosupported of to another Bushin 1988butdefected ond,ofthevoterswhosupported in 1992,almosthalfvotedforPerot.Thisfinding can be thecandidates in two-mutuallyexclusive-ways:Perotwas takingvoters interpreted fromBushanywaywere fromBush;or,voterswhowouldhavedefected analysisbelowgives Ourmultivariate thantoClinton. goingtoPerotrather whichoftheseis closerto thetruth. us themeansto determine oftermlimits a testoftheangry-voter hypothoffers Next,thequestion ifvotersareangrytheywillfavortheforcedretirements esis.Presumably, in however, difference, ofthetargets oftheiranger.We findno noticeable termlimitsandthoseopposedto betweenthosefavoring Perot'ssupport Perot as angry, a groupofvoterswe canidentify termlimits. Thusamongst does no better amongtheangrythanamongthesatisfied. theangryvoterchords,butit The deficit is another issuethatstrikes undertypicalnotionsof issuevoting.One moregenerally also resonates deficit-inthefaceofbothmajorparties' viewis thatthesizeofthefederal lackofresponsibility. it-is a symbolofgovernment's promiseto shrink viewis thatthesize of thedeficitis simplyan issueforwhich Another werenotasked Respondents different preferences. peoplehavedifferent Whenaskedin an opendirectly. theiropinionsofthesize of thedeficit facingthecountry, problems endedquestionto namethemostimportant thedeficit as oneoftheirtopthree offered 28.7%ofrespondents however, thedeficit is an imclaiming We cannotinferfromrespondents problems. wouldbestsolveit. thattheythinka particular candidate portant problem stanceon theimporButPerotwas thecandidatewhotookthestrongest themost-evennew andwas willingtopromise tanceofdeficit reduction, wholistedthesize of the taxes-to solveit.And22.9% of respondents voted as one ofthethreemostimportant issuesfacingthecountry deficit thedeficit votedforPerot. forPerot,whileonly15.9%ofthosenotlisting with themostappearedto resonate ThustheissuethatPerotemphasized thevoters. stoodapartfromtheothertwowas Another issuewhereonecandidate he was opposed fromhistwoopponents: strongly abortion. Bushdiffered and to abortion, a stanceunderscored duringtheRepublicanconvention intheiropPro-choice forceswereunequivocal thecampaign. throughout their fourchoicesto identify to Bush.TheNES gaverespondents position from"abortion shouldneverbe permitted" on abortion, ranging positions as a to "by law, a womanshouldalwaysbe able to obtainan abortion 1 of personalchoice."As Table shows,respondents choosingthe matter twice as alternatives werealmost likelyto vote threemoreanti-abortion The themselves forBushas respondents pro-choice. factthata declaring

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

721

Table 2. Distribution of Respondent Placements of Self and Candidates on Ideology Self-Placement Respondent Bush Ideologya Bush: Ideological-Distance' DukakisIdeologya Dukakis:Ideological-Distance' ClintonIdeologya Clinton:Ideological-Distance' PerotIdeologya Perot:Ideological-Distance'

1988

1992

4.37 5.11 1.24 3.24 1.50

4.21 5.05 1.31 3.19 1.46 4.31 1.15

ofcandidate byrespondents. aMeanplacement meanplacement byall reandcandidates' bMeanideologicaldistancebetweenrespondent spondents. respondent's position on abortion is so discriminatinga factorin determining whom he or she would vote for suggests that abortion was a major issue in the 1992 election.

New Democrat? Last, we offerevidence bearing on the "Clinton as New Democrat" hypothesis. A central tenet of this hypothesis is that voters perceived Clinton differentlythan they perceived Dukakis. Table 2 shows respondents' self-placement; placement of Bush and Dukakis on the NES seven-point liberal-conservative scale in 1988; placement of Bush, Clinton, and Perot on the liberal-conservative scale in 1992; and the mean ideological distance between respondentsand each of the candidates forboth elections. Ideological distance between the respondent and the candidate was computed as the absolute value of the differencebetween the respondent's self-placement and the mean of all respondents' placement of the candidate. Using the mean placement for the candidate, ratherthan the respondent's own placement of the candidate, reduces problems of projection.' Table 2 reveals somethingvery startling:the electorate did not perceive Clinton to be a moderate Democrat. The mean placement of Clinton in 1992 (3.19) was actually .05 to the leftof the mean placement of Dukakis in 1988 (3.24) on the NES seven-point liberal-conservative scale. Alterna'Projectionrefersto thephenomenaof respondents'projecting'theirpositiononto the candidateof theirchoice; thiswould cause respondentsto appearcloser to theirchoice than theyreallyare (Brody and Page 1972).

722

R. Michael Alvarez and JonathanNagler

tively,Clinton'splacement was 1.02to theleftoftheoverallrespondent meanin 1992 (3.19 versus4.21), whereasDukakis'placement was 1.13 totheleftoftheoverallrespondent meanin 1988(3.24versus4.37).Hence, to theextentthatClintonmovedcloserto thecenterof theideological spectrum thanDukakis,he didso bybarely10% ofthedistanceDukakis neededto moveto reachthecenter. Andfinally, comparing themeandistancebetweenClintonandeachvoterandDukakisandeachvoter,we see respective scoresof 1.46and 1.50.Again,anywaywe examinethedata, Clintonsimplydidnotconvincetheelectorate he was different fromDukakisonthebroadliberal-conservative ideological continuum. Itis possible thatdistinctions betweenClintonand Dukakisled respondents to weigh different issuesin evaluating relativeplacements on theliberal-conservativedimension. ClintondidnotshareDukakis'aversionto thedeathpenalty,nordidClintoncarryan ACLU card.AndClintonadvocateda brand ofwelfare reform thatwas notperceived as veryliberal.Butrespondents' evaluations on theliberal-conservative dimension suggestthatClinton's statusas a NewDemocrat couldnothaveaccounted formuchtowards his becausevoterssawhimas an oldDemocrat-they victory, perceived Clintonto be as liberalas Dukakis.2 Thesesimpleanalysesprovidesomeinsight intothe1992election. The datainTables 1 and2 suggestthattheangryvoterhypothesis maynotbe thatClinton'sclaimto be a New Democratcouldnothaveafsupported, fectedtheelection, thatabortion was a majorissuein 1992,andthatthe economywas dominant and Clinton-notPerot-wonthebattleforthe economic discontents. To takeintoconsideration thethree-candidate choice process,andto further developtheabovefindings andshowthattheyare notthespurious artifacts oftwo-by-two tables,we usemultivariate analysis todistentangle andtoestimate theeffects ofdifferent factors onindividuals' vote-choices. We also use multivariate analysistodetermine theimpactof Peroton theelection. MultivariateAnalysisof the 1992 Election In orderto answerthequestions posedaboveanddistinguish between therelativecorrectness of thedifferent folk-wisdoms and theoriesabout theelection,a methodology is required thatallowsus to simultaneously evaluatetheeffects ofindividual characteristics andcandidate characteristicsina three-candidate Themultinomial setting. probitmodelwe employ 2Bushmovedonlya slightbitto theleftfrom1988 to 1992 in respondents'perceptions: from5.11 to 5.05. Thus the voterssaw the two major-party candidatesin roughlythe same positionsin 1992 as theydid in 1988.

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

723

allowsus to do thesethings(Bolduc 1992;Bunch1991;Daganzo 1979; coefficients separate Dansie 1985;HausmanandWise 1978).We estimate for totheirpreference oftherespondents characteristics individual relating trait: fora candidate-specific as wellas a singlecoefficient eachcandidate, fromtherespondent. ideological-distance techniques a modelofthe1992electionusingtraditional To estimate andestimate we couldproceedinthreeways:1) ignorethePerotcandidacy an ormodelsofbinomialchoicesbetweenClintonandBush;2) estimate Perot logitmodelsincluding multinomial model;or3) estimate deredprobit as a choice. arebadlyflawed,andthat twoofthesetechniques We thinkthefirst Thefirst techinthisapplication. problems logithaspotential multinomial Moreimporofalmost20% oftheelectorate. thepreferences niqueignores modbinary-choice andestimating candidate throwing outthethird tantly, onthedependent is a clearcase ofselecting candidates els ontheremaining estimates inconsistent (Manskiand Lerman variable,whichwillgenerate easiestto see in a case whereone choice is probably 1977).The problem Forinstance, to treatWallacevotersin fromtheothers. is clearlydistinct 1968 as missingdataand thenassumethattheywouldhavebehavedas behavedstatusand issue-preferences othersof similarsocioeconomic ofwhichwe havemeasures-isto ignorean on thefewissue-preferences factaboutthesevoters:theydidnotbehaveas theNixon/Humimportant phreyvotersbehavedsincetheyvotedforWallace.Thattheyvotedfor fromthevoterswho to us thattheyweredifferent Wallaceshouldsuggest chosenotto voteforWallace.The samelogicappliesto thePerotvoters: thantheBushandClintonvoters. theymaybe different probitmodel(McKelveyand The secondapproach, usingan ordered modelassumesthat Theordered probit Zavoina1975)is also problematic. continuum. Sincewe are on a unidimensional thechoicescan be ordered dimensions-issues thatvoters multiple mayperceive considering explictly andtheeconomy-thismodelwouldbe inappropriate. logit,assumesthattherandomdisThe thirdtechnique, multinomial termsassociatedwitheach candidateforeach voterare indeturbance to makingthestrongbehavioralassumption This is equivalent pendent. Alternatives" (IIA) withregardto the of Irrelevant of "Independence impliesthatthe in the model.This assumption randomdisturbances to candidate theprobability of choosingthefirst ratioof theprobability of the bytheavailability of choosingthesecondcandidateis unchanged the relabeliefs about prior thirdcandidate.Sincewe do nothavestrong to avoid candidates we for the prefer tionshipbetweenthedisturbances aboutthosedisturbances. assumption usinga modelmakingsucha strong

724

R. Michael Alvarez and JonathanNagler

The multinomial probitmodelwe use allows us to avoid thisassumption.3 Model Specification Modelsofbinary choicesgenerally deal withcharacteristics thatvary by individual, notthatvaryby choice.The choicesin suchmodelscan be described simply as "box number 1" and"box number 2." Inpolychotomouschoicemodelsitmaybe moredesireable to measurecharacteristics ofthealternatives. Anadvantage ofmeasuring characteristics ofthealternativesis thatwe can determine theeffectof addinga choicewithgiven characteristics. It also allowsus to makeuse ofobservable distinguishing features amongthealternatives. Following HausmanandWise(1978)we beginbydefining therandom ofeachvoterovereachofthethreecandidates utility inthe1992election: (1)

whereai is a vectorof characteristics uniqueto thevoteri, Xijis a vector ofcharacteristics uniqueto candidate to voteri, j (j = 1,2,3)withrespect ofparameters to be estimated, andE is a disturbance '4j and , arevectors term.We assumethatthethreeerrorterms(El, E2, ?3) havea multivariate normaldistribution, and we allow theerrorsto be correlated acrossthe candidates. We alsoassumetheerror variances arehomoskedastic. So, with themultinomial probitmodelwe can accountfortheerrorcorrelations acrossthethreeutility functions without a specificstructure assuming for thechoiceprocess.As usual,we assumetheindividual votesforthecandidateoffering thehighest utility. Usingthemultinomial probitmodelonlyone coefficient is estimated percharacteristic ofthealternatives. Forcharacteristics thatvarybyindividualswe mustestimate (M-1) coefficients percharacteristic, whereM is thenumber ofchoices.Thusinthepresent caseweestimate twocoefficients perindividual characteristic. One coefficient givestheeffect of a change in thevariableon therespondent's of votingforBushrelativeto utility theother Perot, coefficient givestheeffect ofa changeinthevariableonthe respondent's ofvoting utility forClinton relative toPerot.Theestimation of thismodelis discussedin Appendix1. The datawe use toestimate themodelcomefromthe1992American NationalElectionStudy(Miller,Kinder,and Rosenstone 1993).Because 3Generalized Extreme-Value modelsalso allowforIIA to be violated;butsincethey stillimposea stronger on thedisturbances assumption thandoes MNP we use theMNP model.

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

725

of Perotwe voters'impressions verylittledatawas availableregarding variablein themodel:theideologicaldishaveonlyone choice-specific As discussedabove,we estitancebetweenthevoterandeachcandidate. of ideologicaldistanceacrossthe fortheeffects matea singlecoefficient usedinthemultivariofideological distance Themeasure threecandidates.4 self-placetherespondent's between ateanalysisis thesquareddifference ideologyscale and thecandidate'smean menton theNES seven-point on thesamescalebyall respondents. placement ofthechangein theirpersonal assessments We usedtherespondent's ofthechange assessments overthepastyear,andtherespondents' finances of in thenationaleconomyoverthepastyearto measuretheinfluences votechoices.Forbothvariables on therespondents' economicevaluations werecodedhigher. responses pessimistic opinrespondents' To measure theimpactofpolicyissueswe included roleto provide:jobs forcitizens; it is thegovernment's ionsof whether wascoded Eachoftheseresponses tominorities. healthcare;andassistance Furthercodedhigher. responses on a seven-point scale,withconservative witha questionofopinionson abortion respondents' more,we measured government feringfourchoicesto describetheirviews of appropriate codedhigher. responses in theissue,withpro-choice involvement of termlimitsfor approvalor dissaproval We includedrespondents' We to testtheangry-voter (approvalcodedhigher). hypothesis politicians the degrees,mayilluminate havetwoothermeasureswhich,to varying of reporting First,we have the respondents' hypothesis. angry-voter didnotvotein 1988,butdid theyvotedin 1988.Iftherespondent whether wouldsuggestthatthiswas hypothesis votein 1992,thentheangry-voter anger,andhenceshould a votercomingoutin 1992solelyto demonstrate theresponvotefortheanti-status quo choice:Perot.Second,we included mostimportant wasoneofthethree ofwhether thedeficit dent'sassessment issuewas whether Perot'sprimary to determine issuesfacingthecountry to theextent thatattaching importance forhim.As notedearlier, effective as an angryvoter, thenvoterswhoviewedthedeficit tothedeficit defines to as voters. shouldhavegone Perot angry important exofrespondents' characteristics We also includedseveralmeasures We included the education, vote-choice. respondents' pectedto influence wasmeasured Education andparty-affiliation. byrespondent's age,gender, formoreeducatedvoters ceterisparibus, We expected, yearsofschooling. tohavea prioriexpectaIt difficult to or Perot. was toprefer Bush Clinton Rather thanassumea linearrelationtionsregarding PerotversusClinton. In thosemodels,thethree modelsinwhichwe relaxedthisassumption. 4Weestimated eachother. distinct from distance werenotstatistically coefficients forideological estimated

726

R. MichaelAlvarezand Jonathan Nagler

shipbetween age andcandidate-preference we included threedummy variablesforage: 18-29, 30-44, and45-59; withpersons60 or overbeing theexcludedgroup.Ifvotersdevelopparty loyalty overtime,thenyounger votersshouldhavebeenmoreperceptive to Perot'sappeal.We included dummyvariablesforDemocratic and Republicanpartisans (we included thoseleaningtowards oneoftheparties as partisans), leavingIndependents as theomitted To allow forregionaleffects, category. we specified the modelwiththreedummy variables, withtheMidwestbeingtheexcluded category. Withthisspecification, wecandetermine thevalidity ofthethreemajor accountsof the 1992electionwe discussedearlier.To examinetheeconomicvotingaccount, we looktothecoefficients on theeconomicassessmentsvariables.For theissuesand ideologyexplanations, theestimated effects ofbothcandidate ideologyandrespondent attitudes onthefourpolicyissueswillbe important. Last,fortheangry-voter hypothesis, we are interested intheeffects ofrespondents' evaluations oftermlimitsandthe deficit on voterchoice,as wellas thebehavior ofthosevoterswhodidnot votein 1988,butdidvotein 1992. The MultinomialProbit Results

The estimates ofthemultinomial probitmodelarepresented in Table

3.5Thecolumnonthefarleftgivestheindependent variables, andtheother

columns givethecoefficients forBushrelative toPerot, andClinton relative to Perot,respectively. The coefficients can be usedto generate predicted foreachindividual probabilities votingforanyofthethreecandidates. Usingthemeanprobability as thepredicted vote-share foreach candidate, ourmodelpredicts theproportions ofvotersforeachcandidate quitewell. In thesampleusedtogenerate theseestimates, 34.1% ofrespondents voted forBush,45.8% forClinton, and 20.0% forPerot.6 Ourmodelpredicts a three-candidate voteoutcomeof 34.0% forBush,46.0% forClinton, and 20.1% forPerot.'We generated predicted vote-choices foreachindividual usingthealgorithm thata person'svoteis assignedto thecandidatefor whomthatpersonhasthehighest estimated ofvoting for.Using probability 'We estimated a modelwithan identicalsystemic component usingbothGEV and independent probit(theindependent probitmodelis a specialcase of MNP whereall offdiagonalelements of theerror-covariance matrix areconstrained to be zero).Thereareno appreciable differences in theestimated coefficients ineithertheGEV northeindependent probit estimates. In theGEV results, theestimates oftheinclusive valueswerenotinformativeaboutthegroupings of thecandidates. 6IntheNES subsample of909 respondents we use,thereis a slight(3%) positivebias forClinton, whichourmultinomial probitmodelreproduces. 7Thesepredictions aretheaveragevalueofPi overthe909 respondents inoursample.

Model Table 3. MultinomialProbitEstimatesfora Three-Candidate Zero) to Normalized (PerotCoefficients Coefficientsfor IndependentVariables Constant Feltpersonalfinanceswereworse Feltnationaleconomywas worse jobs Oppose government healthcare Oppose government assistance minority Oppose government Abortion

Region(East) Region(South) Region(West) voter New or returning Termlimits Feltdeficitwas a majorproblem Democrat Republican Gender(Female) education Respondent's Age: 18-29 Age: 30-44 Age: 45-59 (aBC

GBP

aCP

Clinton

Bush

IdeologicalDistance .50 .44 -.04 .05 -.14** .08 .07 .05 .10* .05 .01 .05

-.09* .02

.01

-.35*

.14 -.15 .17 .25 .18 -.11 .18 .28** .15 .06 .13 -.58* .22 -.19 .17 1.00* .43 .38* .19 .14* .07 -.86* .41 -.64* .30 -.51* .24

-.44 .58 .02 .05 .21* .10 -.01 .05 .06 .04 -.17* .05

-.08 .28 .27 .54 -.07

.26

errors below. withtheirestimated standard estimates Note:Maximum-likelihood LL = -568.18; % correct= 70.6; numberof observations= 909. at thep = .05 level. *indicates an estimate significant at thep = .10 level. **indicates an estimate significant

.11 .32 .21 .50* .19 -.03 .21 -.23* .17 .08 .11 -.003 .18 1.34* .28 -.74 .46 .21 .14 .004 .06 -.57* .26 -.54* .19 -.10 .21

728

R. MichaelAlvarezandJonathan Nagler

thismethodtheoverallcorrect prediction rateforthemultinomial probit modelacrossthethreecandidates is 74.0%. Ourinitialdiscussion ofourestimates is brief, describing onlywhich coefficients reachstatistical significance anddiscussing theimplications of theestimates forsomeexisting ofpresidential theories election.Following thiswe present moredetailedanalysesoftheeffects oftheimportant independent variablesbasedon first-differences whichwe report in Table4. The economy,ideologyand issues. In line withrecentworkon eco-

nomicvoting(e.g.,KinderandKiewiet1981),we findthatvoters'assessmentoftheirpersonal financial condition do nothavea statistically significantinfluence on whichcandidate in 1992.We see a very theysupported strong effect ofthevoter'sassessment, however, ofthenationaleconomy overthepastyear. The estimated effectof ideologicaldistancebetweentherespondent andthecandidate is negative andstatistically as expected. significant, The closera voterwasto oneofthethreecandidates, themorelikelyhe orshe was to support thatcandidate, ceterisparibus.Of thefivespecificissues we examinein thismodel,we see thatonlytwohadsignificant influences on thechoicebetweenBushandPerot:government-sponsored healthcare andabortion rights. Theseresults showthatthosewhosupported a government-run healthcaresystem weremorelikelyto support Perotrelative to Bush.Additionally, thosewhowerepro-life supported Bushandthosewho werepro-choice supported Perot,whichis consistent withtheclearstands takenbybothBushandPeroton thisissue. FortheClintonversusPerotcomparison, we see one significant issue effect: thosewhosupported ofgovernment continuation assistance forminorities weremuchmorelikelyto support ClintonthanPerot.Notethat thisissue matters in additionto generalliberal-conservative positions. Giventhatthiseffect is neither largenorcompelling, issueswereprobably notwhatseparated ClintonfromPerotin themindsof thevoters.In an electionyearwitha slumping theissueofgovernment economy, provision ofjobs appearedtohavehadno effect on voterchoice.Despiteall ofClinton'scampaignrhetoric abouttheimportance ofa government rolein the thequestionof government economy, ofjobs did notseemto provision matter as muchas assessments ofthestateofthenationaleconomy. ThePerotinfluence. One of thecommoninterpretations of the 1992 electionis thatPerotwasabletomobilizehordesofdisaffected ornormally uninterested voters.However,we findthatpeoplewhoreported votingin 1992butdidnotparticipate in the1988electionweremorelikelyto vote forBushthanPerot,thoughmorelikelyto voteforPerotthanClinton, whencompared to peoplewhodidreport votingin 1988.In otherwords,

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

729

toparticiwhoweremobilized ofthosewhodidnotvotein 1988,thevoters Bush,theincumbent, patein the1992electionweremorelikelyto prefer forvoters'opinion thecoefficients Furthermore, thanthetwoalternatives. those Andifanything, werenotsignificant. of termlimitsforpoliticians Perotrelalesslikelytosupport termlimitsweremarginally whosupported Thisleadsus tobelievethattermlimits ofthetwocandidates. tivetoeither northatitleadthemtosupport voters, focusoftheangry wasnotthecritical bythemedia. issuewas overblown Perot.Thustheangry-voter Respondents'characteristics.Partisanshipproducedresultsas antici-

While resultsproducedsome surprises. pated.But, otherdemographic lesslikelythanmentovoteforPerot,controlling womenweresignificantly forthegender forall of theothervariablesin themodel,thecoefficient forBushas forClinton.Thus,all other variableis twicethemagnitude Bushthan thingsbeingequal,femalevotersweremorelikelyto support resultgiventhegendergap beweremalevoters.Thisis an unexpected did10pointsbetter amongwomen andgiventhatClinton tweentheparties, thatPerot withthehypothesis thisis consistent thanamongmen.However, fromBush,makingBushappearweakerrelativeto tookmalesupporters Clintonamongmen.8 thatbothBushandClintondidbetter demonstrate Theagecoefficients voters. Younger whilePerotappealedmoretoyounger amongoldervoters, allegiancesand hencebeenmore votersmayhavehad less firmpartisan thanPerot to Perot'sappeal.BothBushandClintondidbetter susceptible reachesstatistical only significance buttheeffect amongeducatedvoters, observedis Perot'spoorershowing forBush.The clearestregionaleffect whencomto therestofthecountry relative in theSouth,ceterisparibus, paredto bothBushandClinton.9 sizeablecorrelaChoiceprocess.Last,noticethatwe do notestimate norarethe model, terms in multinomial probit the the tionsbetween error attraditional levels.Notethat significant estimated correlations statistically is notviolatedin theelection; thisdoes notmeanthattheIIA condition does ofvoters'utility component thissimply meansthatthenon-systematic We showbelowinourideolaccountfora largeviolation. notnecessarily in amongwomenpersisted Democrats 8AsTable 2 showed,thegendergap favoring thiselection.If in 1996 Perotdoes notrun,we expectto againobservetheRepublican running better amongmenthanamongwomen. candidate in American 9Thereaderusedto seeingraceas a variablein modelsof vote-choice nationalelectionswillnotfindithere.So fewblacksin thesamplevotedforPerotthatit We ranthemodelon a sampleof only thecoefficient. forus to estimate was impossible identical results. whitesandgotessentially

730

R. Michael Alvarez and JonathanNagler

Table 4. Effectsof Economics,Issues,and Angerin the 1992 Presidential Election Probabilityof Votingfor Bush PersonalFinances Better Worse Difference NationalEconomy Better Worse Difference VoterIdeologya Near Far Difference Minorities Assist. No Assist. Difference Abortion Pro-Life Pro-Choice Difference TermLimits For Against Difference Deficit Not Important Important Difference

Clinton

Perot

0.42 0.35 0.07

0.31 0.35 -0.05

0.27 0.29 -0.02

0.54 0.24 0.29

0.19 0.49 -0.30

0.27 0.27 0.00

0.46 0.32 0.14

0.39 0.28 0.11

0.31 0.20 0.12

0.30 0.46 -0.17

0.48 0.20 0.27

0.22 0.33 -0.11

0.62 0.28 0.34

0.22 0.38 -0.16

0.16 0.34 -0.18

0.39 0.38 0.01

0.33 0.32 0.01

0.28 0.30 -0.02

0.39 0.22 0.17

0.33 0.40 -0.07

0.28 0.38 -0.09

Note:Table entriesare thepredicted of a hypothetical probabilities individual votingfor valuesof therow-variable. The profile of this Clinton,Bushor Perotbasedon different voteris discussedin thetext. hypothetical inthevoter-ideology aProbabilitiesforeachofthecandidates rowarebasedontheideological distancebetweenthevoterandtheparticular candidate.

thatthepresenceof a thirdchoice,Perot,hadan impact ogysimulations on therelativevotesharesoftheothertwocandidates. TheMagnitudeofEffects Variables of theIndependent in Table 3 aretranslated in a Sincethecoefficients intoprobabilities inTable4 (King1989).This complexway,we present "first-differences" showsthechangein estimated of choosingeach of thethree probability

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

731

Forinstance, variables. independent basedonchangesinspecific candidates choosing of an individual probabilities rowshowstheestimated thefirst in had improved Bush,Clinton,or Perotif shefeltherpersonalfinances forthe probabilities thepredicted thepastyear.The secondrowindicates worseinthe hadgotten hadshefeltherpersonalfinances samerespondent in thetwo probabilities betweenthepredicted pastyear.The difference herviewofherperchanging oftherespondent theeffect rowsindicates from from toworse.Sincethechangesinprobabilities better sonalfinances variabledependuponthevaluesof theother changingone independent on a hypothetiall thesecalculations we performed variables, independent each ofsupporting probabilities whowouldhavehadsimilar cal individual voterwas female,was ofaverageeducation, Thishypothetical candidate. overthe wereunchanged believedtheeconomyandherpersonalfinances fromthesouth,had votedin 1988,was pastyear,was an independent was at (30 to 44), approvedof termlimitsforpoliticians, middle-aged ideologicaldison theissues,andwas at sample-average modalpositions tancesfromeachcandidate. revealsveryclearlyhowmuchmorerespondents' difference The first opinions mattered thandidrespondents' economy opinionsofthenational If thehypothetical feltherpersonal respondent oftheirpersonalfinances. shewas 7% less likelyto hadgotten worseas opposedto better, finances feltthenationaleconomy respondent voteforBush.If thishypothetical she was then29% less however, had gottenworseas opposedto better, differences also revealthatClinton,not likelyto voteforBush.The first The hypothetical of theeconomicdiscontent. Perot,was thebeneficiary was 30% more thanbetter voterbelievingtheeconomygotworserather likelyto voteforClintonbutwas no morelikelyto voteforPerot.Thus withthe Perotwas unableto capitalizeon the voters'dissatisfaction economy. as follows.The row ofchangesin ideologyaremeasured The effects of themodelvoterchoosingeach marked"near" givestheprobability ontheideological ifshewereoneunitawayfromthatcandidate candidate of the voter "far"givestheprobability hypothetical scale.Therowmarked from if 2.4 units the candidate candidate shewere choosingtheindicated movement away on theideologicalscale.Thusforeachofthecandidates, of forthatcandivoting froma voterwouldreducethevoter'sprobability to a moredetailedanalysisof this .12. We return dateby approximately below. had a verylargeimpact.If ourhypothetical Amongissues,abortion of voting havehadovera 62% likelihood she would voterwerepro-life, insteadof pro-life forBush.However,ifthesamevoterwerepro-choice ofvotingforBushwould herprobability to ourestimates, then,according

732

R. MichaelAlvarezandJonathan Nagler

have droppedfrom.62 to .28 and herprobability of votingforClinton wouldhaverisenfrom.22 to .38. Hencea 40% advantage forBushover Clintonwouldswingto a 10% deficit. Thusvoterswereaffected strongly on abortion. bythecandidates'positions Theestimated impactofrespondents' viewson government provision showsthatitwasa veryimportant ofminority assistance issue.Ifourhypothetical voterbelievedthegovernment shouldsupport minorities, shehad ofsupporting whichdropsto .20 ifshe a .48 estimated probability Clinton, shouldnotprovideassistance to minorities. believedthatthegovernment ofvoting The oppositeeffects areobservedin Table4 fortheprobability forBush and Perot,sincetheprobability thatthesamevotersupported ofthemwasgreater ifthevoterbelievedinnogovernment either assistance (insteadof government help)forminorities. Notethatthiseffectis estiforgeneralliberal-conservative matedcontrolling ideology.Thus while Clintonwas notDukakis,he was moreappealing thaneither PerotorBush topeoplewhobelievedinadditional assistance forminorities; government andhewaslessappealing thanPerotorBushtopeoplewhodidnotbelieve in additional government assistanceforminorities. which Lastly, justas seeninTable3,thevoter'sopinionontermlimits, shouldcharacterize angryvoting,had littleeffectupontheprobabilities of votingforanyofthethreecandidates. Whether ourhypothetical voter or opposedtermlimitsresultsin less thana 2% changein the supported of supporting had likelihood anyofthecandidates. The deficit potentially votershifted herposition on theimpora largeimpact.Ifourhypothetical itcauseda 10% increaseinthelikelihood ofvotingfor tanceofthedeficit ofvotingforBush.ThusBush Perot,anda 17% decreaseinthelikelihood forthedeficit, was heldaccountable and Perotdid well on theissuehe emphasized. Effectsof Candidate IdeologicalMovement season candidatesmay tailortheirideological Duringthe primary stancetofittheelectorate athand(Aldrich andAlvarez1994).Ourevidence onClinton'sperceived toDukakis'perceived relative position position sugcouldmoveinthe geststhatthismaynotbe easy.But,evenifcandidates primary, it mightnothave a majorimpactin thegeneralelection.It is truethata shiftin a voter'sideologicalpositioncausesa largeswingin ofthatvoterchoosingdifferent candidates. Butitwouldbe a probabilities mistaketo thinkthatby choosinga different ideologicalposition(or by hisopponents toappeartorepresent a different forcing ideological position) a candidate couldhavesignificantly raisedhisvote-share. For,byimprovrelative to somevoters, a candidate mustsiinghisideologicalproximity worsenhisideologicalproximity toothervoters.To testthe multaneously

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

733

Movement Figure1A. CandidateVote-Shares:Hypothetical in Bush's IdeologicalPosition

g~~ T

o

~

~~~~ ,, S48'

'''

0

5B

3

7

Bush's Positionon 7-PointScale * PerotI JABush 0 Clinton

withregardto behavioron thepartof thecandidates effectof strategic we simulated ontheliberal-conservative themselves dimension, positioning of eachcandidatemovingacrosstheideologicalspace,holding theeffect this We performed thepositionof theothertwocandidatesunchanged. for each ofeachrespondent theprobability voting simulation bycomputing as we heldevThe probabilities wererecomputed ofthethreecandidates. whose of fixedexcepttheideologicalposition a singlecandidate, erything of from to in increments .02. 0.1 7.0 ideologicalpositionwas adjusted 1 of candidate A show vote-share each IC thepredicted Figures through weremoved and Perot,respectively, (theverticalaxis) as Bush,Clinton, horizontal dimension (the axis). The most alongtheliberal-conservative too far to therightin the is Bush did not move observation that striking hismaximum would have received to our he election. According estimates, electorate at Since the had he been 4.60. vote-share positioned (34.6%) his andhis close to was at he was optimal position; he 5.05, very thought Thus at best Bush his was 34.3%. vote-share at perceived position predicted towards hisvote-share couldhaveimproved by0.3% bymovingslightly theideologicalcenter.Had Bushmovedtoofarto thelefthe wouldhave lostvotes,mostofwhichwouldhavegoneto Perot.Clinton'svote-share nomatter whereBushwasperceived almostconstant wouldhaveremained to be. Clintonwas also close to his optimalideologicalposition. Similarly,

Figure1B. CandidateVote-Shares:Hypothetical Movement in Clinton'sIdeologicalPosition

0

3

1

C

5

7

Cllnton's Position on 7-Polnt Scale A Bush

* Clinton O Perot

Movement Figure1C. CandidateVote-Shares:Hypothetical in Perot'sIdeologicalPosition

-

AA

*~~~~~~~~~~~~~~~~~~~~~~~~*,

P

3 Perot's

Position on 7-Point Scale

JABush

* Clinton O Perot

5

7

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

735

(47.3%)hadhebeenperceived wouldhavebeenmaximized His vote-share was3.19,which position tobe at4.0 ontheideologicalscale.Hisperceived Bushwouldhave of46.4%. Interestingly vote-share gavehima predicted tooneextreme ofClintonsignificantly hadtopushthepublic'sperception belowBush's tohavedropped ortheotherinorderforClinton'svote-share In fact,sincethelikelyplaceforBushtotrytopushthepercepvote-share. belowBush wouldnothavedropped wastotheleft,Clinton tionofClinton untilhe hittheveryedgeof thescale.Thisis a positionon thescale at are simplynotlikelyto be perceived. whichcandidates hadhebeenperceived Perot'svotesharewouldhavebeenmaximized would vote-share tobe at4.24 ontheideologicalscale,wherehispredicted fromhisactualmean indistinguishable havebeen20.2%.Thisis virtually fromPerot'sperceived perceivedpositionof 4.31. Thusanymovement positionwouldhavecosthimvotes.Andtheoverall middle-of-the-road by movement ideological areclear:perceived thissimulation findings from way. theelectionin a significant wouldnothaveaffected thecandidates Effectsof Changes in the Distributionof EconomicViews or preferences ofa singlevoterchanging theeffects Table4 indicated IC showedtheeffects opinionson theissues.And FiguresIA through movingtheirideologicalposition.The otherinteresting ofthecandidates underdiffereffect toexamineis whatwouldhavehappened counterfactual stupid?To see Was it reallytheeconomy, enteconomiccircumstances. this,we wantto knowwhatwouldhavehappenedifvotershadthesame opinionsabouttheeconomyin 1992 thattheydid in 1988.To simulate on thetwo ofvoterpreferences thedistribution thisoutcomewe examined andtheirview personalfinances to theeconomy-their relating questions reasof thenationaleconomy-for1988 and 1992.Thenwe randomly so thatthe signedopinionsabouttheeconomyto the 1992 respondents distribution the1988aggregate distribution ofopinionsmatched aggregate theprobability ofvotingforeach Thisallowedus tocompute ofopinion.10 varivaluesfortheeconomicperception candidate usingthesehypothetical actualvaluesforall oftheothervariables. ables,andtherespondents' on thenational of preferences the 1992 distribution '?Thiswas doneby comparing ourmodelto the1988 of 909 voterson whichwe estimated economyforoursubsample reasas 'worse'in 1992we randomly whoratedtheeconomy Forrespondents distribution. 64% oftherespondents reassigned We randomly signed58% ofthemtothe'same'category. Thisgaveus a distribution category. as 'same'in 1992tothe'better' whoratedtheeconomy personal of respondent's forratings was performed 1988.A similarprocedure matching finances.

736

R. Michael Alvarez and JonathanNagler

Table 5A. Distribution of RespondentPerceptionson theEconomy 1988 Respondentviews of Nationaleconomy Personalfinances

1992

Better

Same

Worse

Better

Same

Worse

18.8 42.4

50.0 32.8

31.2 24.8

4.6 30.3

23.1 35.0

72.3 34.7

aTable entries indicatepercentage ofrespondents in theappropriate category.

Table5A showsthedistributions ofopinionabouttheeconomy in 1988 and 1992,anddemonstrates thehugeshiftin respondent opinionson the stateof thenationaleconomybetweenthetwo elections.In 1988 only 31.2% of respondents felttheeconomyhad gottenworsein thepast 12 In 1992thispercentage months. had morethandoubled o over72% of The shift inopinionregarding respondents. respondents' pePonalfinances is notnearlyas severe,butis stillsignificant. Thus,ifvoterschosetobase theirdecisionson theeconomy, thisshiftwouldhavebeendevastating for Bush. Table 5B presents theresultsofthesimulation described above.The first rowof Table 5B showsthepredicted vote-share foreach candidate giventheactualvaluesfor1992.The secondrowgivesthepredicted vote sharewhenrespondents' abouttheir areadjusted opinions personal finances to matchthe 1988 distribution. The thirdrowshowsthepredicted vote sharewhenrespondents' opinionsaboutthenational areadjusted economy to matchthe1988distribution. Andfinally thefourth rowgivesthepredictedvote-shares whenrespondents' opinionsaboutboththeirpersonal finances andthenational areadjustedtomatchthe1988distribueconomy tions.Thetableoffers a striking result:ifvoters'beliefsabouttheeconomy in 1992wereidentical totheirbeliefsin 1988,thenitwouldhavechanged Table 5B. Effectsof Changesin Distribution of Respondent Perceptions:1988 EconomySimulation Predicted VoteShare for Distribution of Respondents' of Economy Bush Clinton Perot Perceptions 1992 Sample(N = 909) 1988 respondent finances 1988 nationaleconomy 1988 respondent financesand 1988 nationaleconomy

34.0 34.2 37.6 38.0

46.0 45.7 41.5 41.4

20.1 20.0 20.9 20.6

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

737

themarginbetweenClintonandBushby 8.5%. Underthecounterfactual thegap inthesamplebetweenBushandClintonmovesfroman scenario, race.Andsincethesamplewe usedto 11.9%routto a 3.4% contestable 3% towardsClintonand is biasedapproximately makethesepredictions wouldhavegiven 3% againstBush,an 8.5%changeintheactualelectorate overClinton.Thus,simplyput,itwas theeconomy. Busha 2.9% victory indeturnsouttobe essentially Finally,Perot'sshareofthevote-total he goes from of economiccircumstances: of voters'perceptions pendent withobservaconsistent Thisis generally a 20.1% to a 20.6% vote-share. ofTable3 andTable4. Therewe tionswe madeearlierin ourdiscussions we findlittlewhich indicate that,acrosstheelectorate, sawthatourresults ofPerot.Fromdisafthecandidacy drovepeopletosupport systematically toideologyandissues,andeventogeneraldisafwiththeeconomy, fection we haveshownthatnoneof these fectionwiththenationalgovernment, factorsaccountsmuchforthereasonspeoplevotedforPerot.The only ofthe viewson theimportance tothisappearsto be respondent exception deficit. Effectof Ross Perot toask wherePerotvoterswouldhavegonein Evenso, itis important The manner. hisabsence?We areableto answerthatin a straightforward toeachvoterfor givesus an expectedutility technique multinomial probit beoftherelationship as wellas an estimate eachofthethreecandidates, Thus,foreachPerot termsforthethreecandidates. tweenthedisturbance forBushto his or her voter,we can comparehis or herexpectedutility of ofthedistribution andutilizeourestimate forClinton, expectedutility termsto predictforwhomtheywouldvote:" 49.5% of thedisturbance thePerotvoterswouldhavevotedforBush;50.5% wouldhavevotedfor ThusthePerotvoterswouldhavebeensplitalmostevenlybeClinton.12 ThismeansthatthePerotvoterswerenotvoters tweenthetwocandidates. Perot'spresence;in fact Bushwithor without whowouldhavedeserted theyweremore likelyto be Bushvotersthanwas therestof thevoting Amongthesetof Clintonand Bushvotersthetwo-waysplit population. toa 57.5% and42.5% shareofthe was 46% versus34%,whichtranslates ThusthePerotvotersweremorepro-Bush vote,respectively. two-party a 55.5%Clinton IntheabsenceofPerotwe wouldpredict thanothervoters. in moredetail. 2 we describethiscalculation "In Appendix among ofvotingforBushandforClinton aretheaverageprobability 12Thesenumbers in Appendix 2. as described thePerotvoters,

738

R. MichaelAlvarezand Jonathan Nagler

victoryin our sample,ratherthantheabove-mentioned 57.5% Clinton victory. Thus Perot'spresenceinflated Clinton'smarginover Bush by 4%. Conclusionand Discussion BetweenPerot'scandidacy, thestateof theeconomy, and theoverwhelming Democratic victory, thereweremanyunusualcharacteristics of the1992election. Ourresults disprove someinterpretations ofthiselection. The first commoninterpretation of the1992electionwe haverebukedis theangry-voter hypothesis. Voterswereangry, buttheywereangryabout thestateof theeconomy-notthestateof thegovernment. The second we discount interpretation is thattheoutcomeoccurred becausetheDemocratssucceededin nominating a moremoderate candidate. On one hand, ourmodelshowsthatcandidate tovotersin 1992,conideologydidmatter trolling formanyotherpoliticalissues.The electorate perceivedClinton tobejustas liberalas theyhadperceived Dukakistobe in 1988,however, andso itcannotbe thatClinton wontheelection moremoderbyappearing ate thanhad Dukakis.In fact,we haveshownthattheelectionoutcome wouldhavebeenthesameno matter howClintonwas perceivedon the ideologicalspectrum. However,we shouldnotlose sightof thetreesfor theforest. Someissuesraisedduring thecampaign did matter, sinceBush's to therightcosthimdearlyamongpro-choice pandering voters. This leaves standingthe commoninterpretation thatthe economy playeda keyrolein Clinton'ssuccess.Ouranalysisdemonstrates thatthe nationaleconomywas thedominant factorin the1992election.Votersin 1992 wereoverwhelmingly convincedthatthenationaleconomywas in bad shape.The effect of voters'perceptions ofthenationaleconomywas A respondent whofeltthenational staggering. hadimproved was economy 35% morelikelytovoteforBushthanClinton; butifthatsamerespondent felttheeconomyhaddeteriorated he or shewas 25% morelikelyto vote forClintonthanBush!Nowhere else do we see sucha largeshift invoting anditsuggests themagnitude oftheimpactofeconomicevalprobabilities, uations.Ouraggregate estimates indicate thateconomicchangefrom1988 to 1992 cost Bush 8.5% relativeto Clinton.And we have been able to demonstrate thatit was Clinton,notPerot,whoobtaineda greatdeal of his support fromvotersdissatisfied withtheeconomy.Thisis one of our mostimportant thatClintonwonthebattlefortheeconomically findings: ThusdespitetheDemocrats'poorshowingin the1994elecdissatisfied. in 1996appeartohingenoton the tion,Clinton'sprospects forre-election fateoftheRepublican's withAmerica, Contract butprimarily onthevoters' oftheeconomy. perceptions

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

739

Butevenifvotershadperceived thenationaleconomyto be in good shape,Bushstillwouldhavehad a muchtougher electionthanhe did in 1988.Notethatwe predicted a 38.0% to 41.4% splitbetweenBushand Clintonunder1988 economicconditions. Whenwe correctthisforthe slightClintonbias in oursample,andtranslate itintoa percentage ofthe two-party vote,we geta projected two-party splitof 51.8% to 48.2% for Bush over Clintonin 1992 under1988 economicconditions.This is much

closerthanthe53.8%to46.2% Bush-Dukakis race.Thus,therewereobviouslyotherfactorsat workin Bush's loss. Apparently voterswerenot happywithBushin 1992,beyondtheeffect oftheeconomy. Thuswe think a coherent story ofthe1992electionis as follows.First, theoverwhelming dissatisfaction withtheeconomywas a largenail in Bush'scoffin. Butitwasn'tnecessarily fatal.Perceptions oftheeconomy hadBushstarting therace8.5% behindwherehe wouldhavebeenunder 1988economiccircumstances. Thisis not,inandofitself, enoughtohave himfromtherace.If one considersthatBush wonthe 1988 eliminated election by6.4% thenthe8.5%handicap doesnotseemimpossible tooverif Bushhad beenable to retainsomeelectoralbenefit come,particularly fromDesertStormand if one considersClinton'sconsiderable personal baggage.And despitea challengefromtherightin theprimary, Bush's movement on theideologicalliberal-conservative dimension didnotcause himsignificant harm.However,giventhepowerful influence of abortion in determining respondents' votechoicesit appearsthatBush's pro-life stancewasquitecostly.Thefamily valuesnightattheRepublican conventionmayhaveimposedsubstantial politicalcostson Bush. Thereare threesystematic we can drawaboutthePerot conclusions candidacy. First,theissuethatworkedforhimwas thedeficit. Second,he tookmorevotersfromBushthanfromClinton. Third,thosevotershetook werefroma groupexpectedtofavorBushheavilyoverClinton:men.Beyondthis,Perot'sappealseemedto havelittlesystematic He component. did notgrabthevotesof peoplemostdissatisfied witheconomicperformanceormostdesiring changeinWashington. Againwe havefoundlittle the for support angry-voter hypothesis, forthe especiallyas an explanation votesPerotreceivedin 1992.RossPerotillustrates alsowhathasbeenlong knownaboutAmerican elections: moneybuysvotes,andmoneyevenbuys votesin presidential electionsamidstthedinand stormof thecampaign (Jacobson1978;NaglerandLeighley1992). Last,thiselectionhasledus toemploya newmethodology toanalyze presidential electionvoting:multinomial probit.Most electionsin the UnitedStatesinvolveonlytwocandidates, andpoliticalsciencehas wellknowntheoretical and methodological toolsto studythoseelections.As

740

R. MichaelAlvarezand Jonathan Nagler

we have argued in this paper,electionsinvolvingmore thantwo candidatespresentparticularproblemsforthesemodels.These problemsare not intractable, butrequiretheuse ofappropriate methodologicaltoolsto insure thatthe resultswe obtainare not influencedby improperassumptions. Manuscriptsubmitted6 September1994. Final manuscriptreceived8 December 1994.

APPENDIX 1 DerivationoftheMultinomialProbitModel TheBasicsoftheModel The multinomial probitmodelallowsus to estimate thecoefficients ofthemodel whileassuming theerrors arecorrelated, andtoestimate theseerror correlations. thedetailsofthemultinomial Here,we present probitmodel,whichfollowsa framework originally proposed byHausmanandWise(1978);though we deviatefromthoseauthors inthespecification ofthecovariance matrix oftheerror terms.First,we developthebasicsof a multinomial probitmodelfora threecandidate election.We thendescribeourmodelling oftheerrorvariances. We definea randomutility function forvoteri overeachcandidate j, wherej = 1,2,3: Uij= U(Xij,ai) + s(Xij, ai) = Xij3 + ai;

+ Ei;,

(2)

whereXi0is a vectorofcharacteristics choicej relative uniqueto thecandidate to voteri, ai is a vectorofcharacteristics uniqueto theindividual decisionmaker and U defines thesystematic variable, oftheutility component i, ? is a random function of a voter.U is assumedto havethefollowing functional form: U = U(Xij,a,) = Xij + aiVj.

(3)

Notethatwe are assuming thatU is a linearfunction of boththecharacteristics specificto thechoice(Xij)andtheindividual parameters (ai), withrespective characteristics i forthechoice-specific andvj fortheindividual-specific characteristics. The lattercoefficient is subscripted byj to indicatethattheeffects of theindividual-specific characteristics varyacrosschoices.NotethatAV3is normalized to zerohere. We assumethattherandom elements oftheutility functions, -ij,havea multivariate normaldistribution withmeanzeroandcovariance matrix: [1

.

2

13 C23

C03

(4)

VOTER CHOICE IN THE 1992PRESIDENTIALELECTION

741

whowillbringhimor her Now we assumethatthevoterchoosesthecandidate that fortheprobability expression Thisgivesthefollowing utility. thegreatest ofthethreealternatives: wouldchoosethefirst theindividual & (Ui1 > Ui3)]

Pi, = Pr[(Uil > Ui2) jil >

Pi, = Pr[(Uil + Pi, = Pr[(?i2

-

UJi2+ ?i2) & (UC, + ?il > Ui3 + ?i3)] Uil -

?il

Suggest Documents