Corruption, Political Allegiances, and Attitudes Toward Government in Contemporary Democracies

19 Pages·2003·0.22 MB·English

Corruption, Political Allegiances, and Attitudes
Toward Government in Contemporary Democracies
Christopher J. Anderson
Binghamton University, SUNY
Yuliya V. Tverdova
Binghamton University, SUNY
Usingsurveysconductedinsixteenmatureandnewlyestablisheddemocraciesaroundtheglobe,thisstudyexaminesthe
effectofcorruptiononpeople’sattitudestowardgovernment.Theanalysisdemonstratesthatcitizensincountrieswithhigher
levelsofcorruptionexpressmorenegativeevaluationsoftheperformanceofthepoliticalsystemandexhibitlowerlevelsof
trustincivilservants.However,theresultsalsoshowthatthenegativeeffectofcorruptiononevaluationsofthepolitical
systemissignificantlyattenuatedamongsupportersoftheincumbentpoliticalauthorities.Thesefindingsprovidestrong
andsystematicevidencethatinformalpoliticalpractices,especiallythosethatcompromiseimportantdemocraticprinciples,
shouldbeconsideredimportantindicatorsofpoliticalsystemperformance.Moreover,theyimplythat,whilecorruption
isapowerfuldeterminantofpoliticalsupportacrosswidelyvaryingpolitical,cultural,andeconomiccontexts,itdoesnot
uniformlydiminishsupportforpoliticalinstitutionsacrossallsegmentsoftheelectorate.
T
heprinciplesunderlyingdemocraticpoliticalsys- thatthenegativeeffectof corruptiononpeople’sbeliefs
tems presume that governments are accountable aboutgovernmentisfilteredthroughvoters’politicalal-
to their citizens, that they administer laws equi- legiance. Building on the insight that those who voted
tably and fairly, that their actions are transparent, and for the incumbent government are more likely to eval-
that all citizens have access to the political process. As uate the performance of political institutions positively,
a result, political scientists have long assumed that po- wefindthatcorruptionhaslessofacorrodingeffecton
litical systems that fail to live up to these promises are people’sevaluationsofthepoliticalsystem’sperformance
likely be plagued with low levels of legitimacy. This amongsupportersofthegovernmentthanamongthose
study investigates one feature of modern democracy— whoopposeit.
corruption—thatsystematicallyunderminesdemocratic Thisarticleseekstomakeseveralcontributions.First,
principles and, as a result, diminishes people’s faith in it aims to enhance our knowledge of the distortion ef-
thepoliticalprocess.Wearguethatcorruptionisanim- fectsproducedbycorruption.Eventhoughcorruptionis
portant indicator of the performance of a political sys- widelyassumedtohavenegativeconsequencesforacoun-
tem and show that high levels of corruption reduce cit- try’ssocial,economic,andpoliticallife,politicalscientists
izen support for democratic political institutions across havenotsystematicallyexaminedhowitaffectspeople’s
mature and newly established democracies around the views of the political system and institutions of govern-
globe. Moreover, we hypothesize and show empirically ment. Instead, political researchers have focused mainly
ChristopherJ.AndersonisProfessorandChair,DepartmentofPoliticalScience,Fellow,CenteronDemocraticPerformance,Bingham-
tonUniversity,SUNY,Binghamton,NewYork13902-6000([email protected]).YuliyaV.TverdovaisaPh.D.Candidateinthe
DepartmentofPoliticalScience,BinghamtonUniversity,SUNY,Binghamton,NewYork13902-6000([email protected]).
Earlierversionsofthisarticlewerepresentedatthe2000AnnualMeetingoftheMidwestPoliticalScienceAssociationandtheAnnual
MeetingoftheAmericanPoliticalScienceAssociation,Washington,D.C.WearedeeplyindebtedtoMichaelD.McDonaldforallhishelpin
revisingthisarticle.WealsowouldliketothankseminarparticipantsattheUniversityofOxford(NuffieldCollege)andCornellUniversity
(DepartmentofGovernmentandJohnsonGraduateSchoolofManagement)fortheirinsightfulcomments.ThanksalsotoRayDuch,
Hans-DieterKlingemann,andtheanonymousreviewersfortheirthoughtfulcommentsonearlierdrafts,SilviaMendesforhelpwith
thedata,andMarcoSteenbergen,HarveyPalmer,andChrisZornforadviceregardingtheanalyses.Asalways,manythankstoKathleen
O’Connorforallherhelp.ThisresearchwassupportedbyNSFgrantSES-9818525toChrisAnderson.Thesurveydataareavailableas
ICPSRStudyNo.2808.Theoriginalcollectorofthedata,ICPSR,andtherelevantfundingagencybearnoresponsibilityforusesofthis
collectionorforinterpretationsorinferencesbaseduponsuchuses.ThedatawereanalyzedwiththeMLwiN,Version1.10,statistical
software.
AmericanJournalofPoliticalScience,Vol.47,No.1,January2003,Pp.91–109
(cid:2)C2003bytheMidwestPoliticalScienceAssociation ISSN0092-5853
91
92 CHRISTOPHERJ.ANDERSONANDYULIYAV.TVERDOVA
ontheeffectsofcorruptionondemocraticprinciplessuch systemperformancethatshapesthereputationofpoliti-
as accountability, equality, and fairness. We go beyond calinstitutions.Incontrast,thequestionofhowpolitical
theseearlierstudiesbyestablishingaconnectionbetween performance affects system support has received much
corruptionandmassperceptionsofhowwellthepolitical morelimitedattentionbysocialscientists.1Thefewstud-
systemworksandwhetherpublicofficialscanbetrusted ies that do exist are important, however, because they
across widely different contemporary democracies. Sec- show that political performance and the functioning of
ond,ourresearchbreaksnewgroundinthestudyofpo- formalpoliticalinstitutionsmattersforhowpeopleview
liticallegitimacybyfocusingoncorruption,anindicator thefunctioningofthepoliticalsystem.Below,weseekto
ofpoliticalperformanceandanexplanatoryvariablenot add to our understanding of the effects of political per-
usually examined in studies of political support. Thus, formance on political support by investigating whether
we seek to develop a more complete understanding of andhowinformalinstitutionsandpracticesintheform
thelegitimacyofpoliticalsystemsbyexaminingwhether ofcorruptioninfluencepeople’sattitudestowardtheex-
andhowcorruptionaffectsattitudesaboutgovernment. istingpoliticalorder.
Third, our research examines the contingent nature of Examining the effects of corruption, which we de-
therelationshipsamongcorruption,politicalallegiances, fine as “the misuse of public office for private gain”
andpoliticalsupport.Inparticular,weseektodetermine (SandholtzandKoetzle2000,32),meansinvestigatinga
whetherindividual-levelfactors,suchashavingallegiance phenomenonwhoseexistenceismoredifficulttoconcep-
to those in power, can serve to neutralize the negative tualizeandmeasurethanthatofeconomicperformance
impact of corruption on people’s views of the political orformalpoliticalinstitutions,andwhoseconsequences
system. If they do, inferior performance may not nec- arenotalwaysobvious.Avoluminousliteraturehasdoc-
essarily be recognized by all citizens, and we may not umentedthenegativeeffectsofcorruptiononanation’s
see a decrease in the level of public support for politi- socialandeconomiclife(forreviews,seeMontinolaand
calauthorityasaresultamongimportantsegmentsofthe Jackman2002;Rose-Ackerman1999).Moreimportantly
electorate. for the purposes of this study, however, corruption also
Thenextsectiondescribessomeofthegapsinwhat hasbeenfoundtofundamentallyunderminetheprinci-
weknowabouttherelationshipbetweencorruptionand plesofdemocraticaccountability,equality,andopenness
mass support for the political system. Subsequently, we (Dahl 1971).2 When corruption is present, democracy’s
discussthehypothesizedeffectof politicalallegianceon tenets of procedural and distributive fairness become a
attitudesaboutgovernment,aswellasourcontentionof myth; this, in turn, is likely to diminish the legitimacy
why the effect of corruption on political support may
hinge, in part, on membership in the political majority
1Fewcomparativestudieshaveexaminedhowformalpoliticalin-
and minority. We then turn to issues of measurement
stitutionsandtheiroutputsaffectsupportforthepoliticalsystem.
anddataanalysis.Afterpresentingtheresults,wediscuss Amongthem,MillerandListhaug(1990,1999),forexample,have
the importance of the findings for the study of political foundthatopportunitiestoexpressdiscontentandpositiveper-
ceptionsofproceduralandoutcomefairnessarerelatedtopositive
supportindemocraciesandspelloutavenuesforfurther
attitudesaboutgovernment(seealsoHofferbertandKlingemann
research.
1999).Similarly,studieshaveshownthatmoreproportionalelec-
toralsystemsareassociatedwithhigherlevelsof regimesupport
(Anderson1998).Asidefrominstitutionalelementssuchasoppor-
CorruptionandAttitudes tunitiesfordissent,itappearsthatgovernmentstabilitymattersfor
howpeopleviewthepoliticalsystem.Specifically,peopleinsystems
TowardGovernment
withmoredurablegovernmentsaremoresupportiveoftheexist-
ingpoliticalarrangements(HarmelandRobertson1986).Finally,
studiesofsystemsupportinnewdemocracieshavepointedtothe
Political trust or system support is an important indi-
importanceofpoliticalperformanceasdeterminantsofsupportfor
catorof ahealthycivicanddemocraticpoliticalculture. democraticinstitutionsmoregenerally(EvansandWhitefield1995;
Scholarscommonlyassumethatdisenchantedcitizensare MishlerandRose1997,2001a),andthattheinstitutionalqualityof
domesticinstitutionscanaffectsupportforsupranationalinstitu-
morelikelytopushforradicalchangesinthesystem,and
tions(Rohrschneider2002).
that distrust of government may be detrimental to the
2Forexample,corruptionunderminesdemocraticrulewhenpublic
establishmentandsurvivalofdemocraticlifeinthelong
goodsareavailableonlyforthosewhohaveeitherconnectionsor
run.Italsoiswidelyacknowledgedthatsystemoutputs—
money(orboth)(Treisman2000).Andalthoughcorruptiondoes
also commonly referred to as system performance—are notnecessarilypreventagovernmentfromaccomplishingsociety’s
key to understanding why public support for the politi- ends,itwilldosoinefficiently.Asaresult,corruptionviolatesim-
portantprinciplesofmodernbureaucracy,includingtheideathat
calsystemfluctuates(Easton1965).Curiously,economic
publicagenciesshouldoperateinanimpartialandrule-basedfash-
performance has been the most widely studied facet of ion(SandholtzandKoetzle2000).
CORRUPTIONANDATTITUDESTOWARDGOVERNMENT 93
ofdemocraticpoliticalinstitutions(GibsonandCaldeira levelsofsupportfortheregimeandadecreasedlikelihood
1995;Tyler1990). thatpeoplewouldrejectundemocraticalternatives.Sim-
The idea that corruption has a negative impact on ilarly,MishlerandRose’s(2001a)studyofpoliticaltrust
people’sviewsoftheirgovernmentisopentoboththeo- across10East-CentralEuropeanstatesfoundthathigher
reticalandempiricalchallenge,however.Onatheoretical levelsofcorruptionwererelatedtolowerlevelsofpolitical
level,somewouldarguethatacountry’spoliticalculture, trust.However,bothstudiesalsoreportedthat,onceal-
which shapes people’s perception of corruption, can di- ternativeexplanationsofsystemsupportweretakeninto
minishoreliminatetherelationshipwehypothesizehere. account, the effect of corruption as a predictor of sys-
Culture provides a lens for how people view the world, tem support was substantially attenuated or reduced to
motives for human behavior, criteria for evaluating ac- insignificance (Rose, Mishler, and Haerpfer 1998, 189–
tions, and, more generally, orientations to action, all of 94;MishlerandRose2001a).Anindividual-levelstudyof
whicharelearnedduringculturalsocialization(Eckstein Bolivia,Nicaragua,Paraguay,andElSalvadorshowedthat
1988).If acountry’sculturalcontextpredisposespeople people’sexperienceswithcorruptionarenegativelycorre-
to view corruption as acceptable practice and therefore latedwithdiffuseregimesupport(Seligson2002).Given
relatively benign, measures of corruption may not co- thepotentialforendogeneityamongtheprimaryvariables
incide with how people in different cultural settings re- of interest (trust and reports of corruption) inherent in
spondtocorruptpoliticalpractices(cf.Nye1967).Under the study’s cross-sectional individual-level research de-
theseconditions,higherlevelsof corruptionshouldnot signandtheselect,smallnumberofcountriesexamined,
be associated with more negative attitudes toward gov- these results are suggestive of a correlation among the
ernment. This view also is consistent with a strand of relevantindividual-levelconstructsbutfarfromgeneral-
research labeled “the functional theory” after Merton’s izableacrossawiderrangeofcontemporarydemocracies.
(1968)seminaldiscussionof thelatentfunctionsof po- Atthistime,theempiricalrecordthusisinsufficient
liticalmachines,whichhighlightsthepositiveaspectsas- and open to crucial challenges when it comes to deter-
sociated with the buying and selling of political favors mining conclusively whether there is a relationship be-
(Leff 1964; Huntington 1968; see also Montinola and tween corruption and system support at the individual
Jackman2002).Viewedfromthisperspective,corruption leveloncealternativeexplanationsofsystemsupportare
mayincreasetheefficiencyofgovernmentbecausepublic accountedfor.Forone,whenpresent,theeffectsofcor-
servants may become more helpful and effective if paid ruption on system support have not been found to be
directly. Thus, bribes may help overcome bureaucratic overwhelminglystrong.Moreover,becausethesampleof
obstacles and “add to a nation’s economic efficiency” countriesconsideredinthesestudieshasbeenlimitedto
(Goldsmith1999,869). newdemocraciesinEasternEuropeorLatinAmerica—
Empirically,thehypothesisthatthereisarelationship thatis,countriesthatscorerelativelyhighoncorruption
betweencorruptionandattitudestowardgovernmentis and related factors, such as level of development—it is
also open to challenge, given the limited evidence the uncertainwhethercorruptioniscorrelatedwithpolitical
nascentliteratureonthetopichasproduced.Simplyput, supportincountriesrepresentingvaryinglevelsofwealth
todateresearchershavepaidlittlesystematicattentionto anddifferentpoliticalcultures.
theeffectscorruptionmighthaveontheattitudesofor- Moreover,itisyettobedeterminedwhethercitizens
dinarypeopletowardpoliticalinstitutionsintheircoun- associatecorruptionwiththepoliticalsystemgenerallyor
try.Thisissurprising,giventhatscholarshavepointedto specificpoliticalactors,suchasgovernmentbureaucrats.
attitudesaboutgovernmentresponsiveness(externalef- Below,wethereforecomparetheeffectofcorruptionon
ficacy)andproceduralfairnessasimportantindicatorsof onegeneralattitudetowardgovernment—evaluationsof
supportforthepoliticalsystem(Iyengar1980;Millerand the performance of the political system generally—and
Listhaug1999).Andwhilethereareafewcasestudiesthat onespecificattitudeparticularlyrelevanttothequestion
discuss the relationship between corruption and system of corruption—trustincivilservants.Attheoutset,itis
legitimacy(e.g.,dellaPorta2000;Pharr2000),fewschol- not obvious whether both attitudes should be affected,
arshaveexaminedtherelationshipsystematicallyfroma orwhethertheyshouldbeaffectedsimilarlybythepres-
cross-nationalperspective. ence of corruption. Given that much of everyday cor-
Among the few studies that systematically exam- ruptionisperpetratedbygovernmentbureaucrats,while
ine the impact of corruption on system support, Rose, the political system is a more abstract and distant ob-
Mishler, and Haerpfer’s (1998) cross-national study of ject of consideration, attitudes toward civil servants are
nine Central and East European countries found that themoreproximatemeasureinthecontextofthisstudy.
higher levels of corruption were associated with lower Followingthislogic,wewouldexpectstrongereffectsof
94 CHRISTOPHERJ.ANDERSONANDYULIYAV.TVERDOVA
corruptionontrustincivilservantsthanonevaluations ruptionlessnegativelythanthoseintheminority.Ifthis
of system performance. At the same time, it is plausi- is the case, the effect of corruption on attitudes toward
ble that people will be more willing to identify corrup- government should be smaller for the political majority
tion as a systemic problem than to lay the blame for thanthepoliticalminority.
corruptiononthedoorstepofspecificactorstheyhaveen- Ourexpectationsregardingthisinteractionarebased
countered.Afterall,itispossiblethatcitizensthemselves onresearchshowingthatpeoplearepronetoviewthepo-
either are civil servants or know someone who is and, liticalworldinwaysthatareconsistentwiththeirpoliti-
therefore,maybemorewillingtoblameananonymous calpredispositions(Zaller1992).Publicopinionscholars
set of institutions than specific individuals they know. theorizethatindividualsuseheuristicstofilter,andtosim-
Moreover,becausethepoliticalprocessinvolvesmultiple plify analysis of, the information they choose to receive
governmental agents, citizens may have more evidence (Sniderman,Brody,andTetlock1991;Zaller1992).Wear-
availableaboutthesystemandhowwellitfunctionsthan guethat,inthecontextofanalyzingcorruptionandsup-
about specific actors (Weatherford 1987). Whether cor- portforthepoliticalsystem,therelevantheuristicislikely
ruption is viewed as a systemic phenomenon, as result- tobesupportfortheincumbentpoliticalauthorities.Em-
ing from the actions of a specific group of political ac- pirically,thisexpectationis,generallyspeaking,compati-
tors, or both, is therefore a question we seek to answer blewiththenotionthatindividualswithparticularviews
below. interpretnewinformationsothatitreinforcespreviously
held attitudes, thereby augmenting rather than temper-
ingthedifferencesbetweentheirbeliefsandthoseofin-
TheContingentEffectsofCorruption
dividualswithopposingpredispositions(Zaller1992).It
onPoliticalSupport:TheRole
is also consistent with the finding that people exagger-
ofPoliticalAllegiances
ate the performance of the macroeconomy in line with
their partisan leanings, for example (Duch, Palmer, and
We also posit that the effects of corruption on attitudes
Anderson2000).Finally,itisconsistentwithMishlerand
toward government are strengthened or weakened, de-
Rose’sconjecturethattheeffectsofcorruptionarelikely
pending on whether people have a stake in the existing
tobemediatedbymicro-levelattitudes(MishlerandRose
political regime and the maintenance of the status quo.
2001a,50).
Thisexpectationemergesfromagrowingnumberofstud-
Aside from the psychological mechanisms that may
ies,whichshowthatcitizenswhoidentifywithorvoted
condition people’s views of corruption, there also may
foragoverningpartyarepredisposedtoevaluatethegov-
be direct pay-offs associated with corruption that are
ernment’s performance positively and to be more sup-
more likely to accrue to one part of the population
portive of the political system (Anderson and Guillory
thananother.Inparticular,wehypothesizethatgovern-
1997; Anderson and LoTempio 2002; Ginsberg and
ment supporters are more likely to be the beneficiaries
Weissberg 1978; Nadeau and Blais 1993; Norris 1999).3
of the goods distributed by corrupt public officials in
Wetesttwokeyhypothesesassociatedwiththisliterature.
the form of attention by officials, favorable treatment
First,wehypothesizethatindividualswhobelongtothe
in the awarding of government contracts, or patron-
political majority are more likely to exhibit positive at-
age jobs, for example (Fiorina and Noll 1978; Golden
titudes toward government than those in the minority.
n.d.;Warner1997).Asaresult,thosewithanallegiance
Second,andmoreimportant,wearguethatpeople’spo-
to the incumbent government also are likely to take a
litical allegiances motivate them to connect corruption
morebenignviewofcorruptionbecauseitbenefitsthem
withtheirviewsofthepoliticalsystemindifferentways.
personally.
Specifically,wepositthatpoliticalmajorityandminority
In sum, we argue that those who elected the in-
status acts as a screen for how people view government
cumbent government are less likely to seek out infor-
corruption and the extent to which they use corruption
mation about corruption and less likely to interpret
to judge the performance of the political system. Those
such information negatively. Moreover, we posit that,
whoelectedthecurrentgovernmentarelikelytoviewcor-
whencorruptionoccurs,itismorelikelytobenefitsup-
porters of the government than supporters of the op-
3The so-called “winner effect” or “home-team effect” has been position. As a result, corruption should produce less
documentedwithregardtoanumberofdifferentpoliticalattitudes of a negative impact on political support among those
such as feelings of government responsiveness, satisfaction with who elected the incumbent government (the majority)
democracy, as well as people’s willingness to engage in political
than among those who supported the opposition (the
activism(cf.AndersonandTverdova2001;ClarkeandAcock1989;
GinsbergandWeissberg1978;WhiteleyandSeyd1998). minority).
CORRUPTIONANDATTITUDESTOWARDGOVERNMENT 95
DataandMeasures Togaugewhetherpeoplehadtrustincivilservants,
theywerepresentedwiththefollowingstatement:“Most
civilservantscanbetrustedtodowhatisbestforthecoun-
Our individual-level data come from surveys collected
try.”Respondentscouldanswer“stronglyagree,”“agree,”
aspartoftheInternationalSocialSurveyProgram(ISSP)
“neitheragreenordisagree,”“disagree,”and“stronglydis-
in 1996 as part of a study called Role of Government
agree.”Thisquestiongaugespeople’strustinafairlyspe-
III. Countries that provided the most important sur-
cific set of actors—more specific, certainly, than asking
vey items and that had a sufficient number of cases
aboutthepoliticalsystemasawhole.Generally,thisques-
for multivariate analysis included Australia, Canada,
tionandsimilaronesaimedatparliament,thecourts,the
the Czech Republic, Germany, Great Britain, Hungary,
armed services, etc., have been considered indicators of
Ireland, Italy, Japan, Latvia, New Zealand, Norway,
support for regime institutions (cf. Klingemann 1999).
Russia,Slovenia,Sweden,andtheUnitedStates.Thus,we
The variable was coded as a five-category measure that
wereabletoemploysurveysfromadiversesetof coun-
rankedtrustfrom1to5,with5denotingthemosttrust-
trieswithwidelyvaryingpoliticalcultures,structures,and
ingresponse.
histories.
Figure1showsthedistributionofpositiveresponses
tothetwosupportvariables(percentagessayingthatthe
DependentVariables politicalsystemworkswellandthatcivilservantscanbe
trusted)acrossthecountriesincludedinthisstudy.First,
Ourdependentvariableindicatorstapintodifferentdi- on average, 60 percent of citizens have a positive opin-
mensions of support for the political regime, with one ionabouttheirpoliticalsystem,butonly23percenthave
geared toward general evaluations of the system’s per- trustincivilservants.Second,thegraphshowsthatthere
formance and the other asking respondents to evaluate is considerable cross-national variation in people’s atti-
particular institutional actors—civil servants.4 To mea- tudes toward government. Specifically, positive perfor-
sure performance evaluations, respondents were asked: manceevaluationsrangefromaround18percentsaying
“Allinall,howwellorbadlydoyouthinkthesystemof thatthesystemworkswellinRussiaallthewaytomore
democracy in (country) works these days?” The answer than88percentsayingthesystemworkswellinNorway,
categories were: “It works well and needs no changes; it withmostcountries’levelsof positiveevaluationsrang-
works well, but needs some changes; it does not work ingsomewherebetween60and70percent.Thegraphalso
wellandneedsalotofchanges;itdoesnotworkwelland showsthatthereisconsiderablevariationacrosscontem-
needstobecompletelychanged.”Theseanswercategories porary democracies with regard to whether people feel
rangedfrom1to4,with4denotingthemostpositiveand theycantrustcivilservantstodowhatisbestforthecoun-
1themostnegativeevaluation. try.WhiletheIrishandtheNorwegiansarequitetrusting
This measure does not capture citizen attitudes to- oftheircivilservantsat52and37.5percent,respectively,
ward democracy as an ideal; instead, it focuses on peo- Italians, Swedes, Hungarians, Czechs, and Russians are
ple’s responses to the actual process of democratic gov- considerablylesstrusting,withonlyabout10–15percent
ernance and their attitudes toward a country’s “consti- ofthemexpressingtrustincivilservants.
tutionalreality”(Fuchs,Guidorossi,andSvensson1995,
328). Using Easton’s categories, this indicator has been
IndependentVariables
identifiedasameasureofsupportfortheperformanceof
thepoliticalregime(cf.Klingemann1999;Norris1999). Corruption. Recall that our primary goal is to test the
Becauseitallowsustoconnectcorruption,whichwear- following relationships: (1) between corruption and at-
gue to be an indicator of system performance, with an titudes toward government, (2) between political alle-
evaluation of the performance of the system in the eyes giances and attitudes toward government, and (3) the
of itscitizens,thismeasureisparticularlyusefulforour interactive effects of corruption and political allegiance
purposes. onattitudestowardgovernment.Toestimatetheimpact
of corruption, we need to connect attitudes about gov-
ernmentwithinformationaboutcorruptpracticesinthe
4Using a single item indicator, as opposed to a multi-item con-
respondent’scountry.Todoso,wecombinedacountry-
struct,maycompromisethereliabilityofourdependentvariables.
Note,however,thataddingmoreitemstoanindicatorhasnodirect level measure of corruption with the individual-level
effectonitsvalidity(Guilford1954;KerlingerandLee2000).More- surveys.
over,anunreliabledependentvariabledoesnotbiasregressionesti-
OurmeasureofcorruptionisbasedontheCorrup-
mates,butitmakesithardertoachievestatisticalsignificance(King,
Keohane,andVerba1994). tionPerceptionIndex(CPI)developedbyTransparency
96 CHRISTOPHERJ.ANDERSONANDYULIYAV.TVERDOVA
FIGURE1 EvaluationsofPoliticalSystemPerformanceandTrustinCivil
Servants
n 100
o
ni 90
pi
O 80
e
v
siti 70
o
P 60
a
g 50
n
ssi 40
e
xpr 30
E
e 20
g
a
nt 10
e
c
er 0
P Australia Britain CaCnzaedcahRepublicGermany Hungary Ireland Italy Japan LatNvieawZealand Norway Russia Slovenia SwedUennitedStates
Evaluation of System Trust in Civil Servants
International, the most frequently used and most com- fortotallycleanfromcorruptionand10indicatesahighly
prehensivemeasureofcorruptionaroundtheworld.The corruptcountry.7
CPIisacompositemeasurebasedontheassessmentsof Figure 2 shows the distribution of the corruption
country experts as well as the views of many different index across the countries included in this study. As in
individuals in each country who encounter corruption thecaseofthepoliticalsupportquestions,thedatashow
indifferingwaysandinavarietyof contexts.Itisbased considerable cross-national variation in corruption. On
onthescoresofratingspublishedbyinternationalorga- average,thecountriesincludedintheanalysiswereonly
nizations and constitutes the average of the scores from moderatelycorruptat3.1.Thehighestlevelsofcorruption
thosesurveys(seetheappendixfordetails).5Weusedthe werefoundinRussiaandLatvia(atover7onthe10-point
corruption index measured for 1996 to match with our scale), whereas the cleanest countries turned out to be
survey data.6 The original measure runs from 0 (highly
corrupt)to10(absolutelycleanfromcorruption).Tofa- 7Ourresearchdesignrequiresthatourmeasureofcorruptionbe
cilitatetheinterpretationofthestatisticalresultsinlight exogenoustoattitudestowardthesystem.Giventhatourcorrup-
tion index is a measure of public and elite perceptions of cor-
ofourhypotheses,wereversedthescalesuchthat0stands
ruption(albeitaggregatedatthesystemlevel),itispossiblethat
ourmeasureof corruption,atleastinsmallpart,isendogenous
to the model. To test whether this was the case, we conducted a
Hausman two-step test of exogeneity (Hausman 1978; see also
PindyckandRubinfeld,1991,303–05).Inthefirststage,corrup-
tionwasregressedonallindependent(exogenous)variablesofthe
5Althoughresearchershaveevaluatedthevalidity,reliability,and
originalmodelplussomeinstrumentalvariablesthataredirectly
robustnessoftheindexwithextremelypositiveresults(Lancaster
relatedtothesuspectedendogenousvariablebutnotthedependent
andMontinola1997),itisnotwithoutlimitationsbecauseitas-
variableofinterest.Followingpreviousresearch(cf.Sandholtzand
sumes that corruption is a one-dimensional phenomenon that
Koetzle2000;Treisman2000),weusedtrade(%GDP)andamea-
can be measured along a single continuum, and that those sur-
sureofeconomicfreedomasexogenouspredictorsofcorruption
veyed operate with a similar definition of corruption. However,
inthefirststageregression(yieldinganR2of.87).Theestimated
given that there usually are few witnesses to corruption and be-
residualsfromthisregressionwerethendefinedtoforma“residual
cause those who have knowledge of it typically have an interest
variable”thatwasaddedtothemodelexplainingsystemsupport
inkeepingitsecret,people’sperceptionsarelikelytobemoreac-
inasecondstageregression.Iftheresidualvariableisstatistically
curatethananygovernment-sanctionedcountofcorruptactions
significantinthesecondstageregression,evenwhencontrolling
(Treisman2000).
forthepresumedexogenousvariable,thentheregressoris,infact,
6Because the CPI for 1996 did not include data for Latvia and endogenous. Results from the second stage regression show that
Slovenia,weusedthedatafor1998instead,assumingthatcorrup- theresidualvariablefailedtoachievestatisticalsignificance(sys-
tionlevelsareunlikelytochangeradicallyovertime.Inspectionof temperformancemodel:(cid:2)=.011(s.e.:.011);trustincivilservants
trendsintheCPIdataforothercountriesconfirmsthatthiswasa model:(cid:2)=.030(s.e.:.019).Wethereforeconcludedthatthecor-
reasonableassumption(seealsoTreisman2000). ruptionvariablewasnotendogenoustosystemsupport.
CORRUPTIONANDATTITUDESTOWARDGOVERNMENT 97
FIGURE2 Cross-NationalLevelsofCorruption
10
h)
g 9
hi
=
0 8
1
w; 7
o
=l
0 6
n (
o 5
pti
ru 4
r
o
C 3
of
el 2
v
e
L 1
0
Australia Britain CaCzneacdhaRepublicGermany Hungary Ireland Italy Japan LatNvieaw Zealand Norway RussiaSlovenia SwedUennitedStates
NewZealandandSwedenatlessthan1.Oftheremaining voted for the current government, he was coded as be-
countries,sixscoredbetween1and2(Australia,Britain, longingtothemajority(coded1);allotherswerecoded
Canada,Germany,Ireland,andNorway),twoscoredbe- 0.Toexaminetheinteractionbetweenlevelsofcorruption
tween 2 and 3 (Japan and the United States), and four andpoliticalallegianceonattitudestowardthesystem,we
(theCzechRepublic,Hungary,Italy,andSlovenia)scored alsoaddedaninteractionterm(majority-minoritystatus
∗
higherthanfourbutlowerthanRussiaandLatviaatthe corruption)totheequation.
highend.
ControlVariables
Politicalallegiance:Majorityandminoritystatus. The
political majority and minority status variable was cre- We also sought to control for a variety of factors that
ated with the help of a question asking which party or have been found to predict support for the political
presidentialcandidatetheindividualhadvotedforinthe system in previous analyses. Including these variables
last national election. Relying on Keesing’s Archives, the avoidsdrawingfaultyinferencesduetospuriousnessthat
Political Handbook of the World, and the European Jour- can result from omitting relevant variables. The control
nal of Political Research, the next step was to determine variables fall into two categories: system-level variables
whowontheelectionineachcountry.Wethencombined and individual-level variables. At the system level, these
theinformationabouttheperson’spastvotewiththein-
formation about the party in power.8 If the respondent
comesandrecalledelectionoutcomes.Inoursample,49.1percent
of respondents reported voting for the majority parties in their
countries;thisisslightlyhigherthanthe46.6percentofactualvot-
8Themeasureofmajority-minoritystatusmaysufferfromthepo- ersforwinningparties.Asthesefiguressuggest,overreportingis
tentialproblemofbiasedrecalloroverreportsfavoringthevictori- thusslightlymorepervasivethanunderreporting.Thismeansthat
ousparty(cf.Wright1993).Notethat,dependingontheelectoral ourresultsarelikelytounderstatethetrueeffectsofpoliticalma-
cycle,over-andunderreportingbotharepossible,giventhatgov- jority and minority status. Another source of bias could exist if
ernments’honeymoonisusuallyfollowedbyadeclineinpopular voterswhoaremore(less)trustingofthepoliticalsystemwould
supportduringtheelectoralterm.Whenpresent,overreportingis over(under)reporttheirvotefortheincumbentgovernment.Ifsuch
likely to be less problematic because it implies that some of the abiasexists,thereshouldbearelationshipbetweenlevelsoftrust
respondentswhoclaimtohavevotedforthevictorinfactdidnot. andover(under)reportofthevote.Wethereforecalculatedthedif-
Inthiscase,someoftherespondentsclassifiedasbelongingtothe ferencebetweenthereportedandtheactualvoteinacountryand
majority are, in fact, members of the minority. This would lead correlatedthisfigure(ameasureofover/underreportofthevote)
tounderestimatingthetrueeffectofhavingvotedforthewinning withthelevelsofsystemsupport.Thesecorrelationswerestatisti-
partyorparties.Asacheckontheaccuracyofthereportedvote, callyinsignificant(trustincivilservants:.05(p=.9);democracy
we compared the aggregate distributions of actual election out- satisfaction:−.2(p=.9)).
98 CHRISTOPHERJ.ANDERSONANDYULIYAV.TVERDOVA
include current macroeconomic performance,9 eco- ahigherprobabilityofTypeIerrors(seealsoZorn2001).
nomic development,10 democratic longevity, and level Toestimateourmodels,wethereforereliedonstatistical
of democracy.11 At the level of individual respondents, techniquesdevelopedspecificallyformodelingmultilevel
we controlled for such factors as political interest, elec- datastructures(SteenbergenandJones2002).13
toralparticipation,12socioeconomicstatus,employment Several specific issues may pose problems for infer-
situation, and a standard set of demographic variables. ence because of the multilevel nature of our data. First,
Codingproceduresanddescriptivestatisticsforallvari- theinterceptsmaybevariableacrosscountries;failureto
ablesarelistedintheappendix. controlforthismayresultinbiasedestimates.Specifically,
ifinterceptsarevariable,wemaybeoverestimatingtheef-
fect of corruption on system support, as the corruption
AnalysisandResults
coefficientcouldbecapturingboththetrueeffectsofcor-
ruptionaswellasothercountry-specificeffects.14 Asec-
Our research design requires that we combine infor- ondaryconcernisthattheindividuallevelvariablesmay
mation at the level of respondents (micro-level) and haveunequalslopesacrossnations.Inthiscaseapooled
countries (macro-level). This means that our data have estimator may be biased for each particular country. A
a multilevel structure where one unit of analysis (vot- thirdconcernrelatestotherobustnessof ourinferences
ers) is nested within the other (countries) (Bryk and basedonpotentiallyinefficientstandarderrorsresulting
Raudenbush1992).Thistypeofdatastructurecangener- from potential clustering (cf. Zorn 2001). To deal with
ateanumberofstatisticalproblems,suchasnonconstant theseissues,multilevelmodelingtechniquesallowforesti-
variance and clustering. Failure to recognize the hierar- matingvaryinginterceptsandslopes,produceasymptoti-
chicalnatureofthedatacanleadtounderestimatingstan- callyefficientstandarderrors,andprovideforadirectesti-
darderrors—particularlyatthemacro-level—and,thus, mationofvariancecomponentsateachlevelofthemodel.
Below,weshowthecoefficientsofinterest(constants
andindependentvariables),aswellasthevariancecom-
9TheISSPsurveysusedheredidnotfurnishvariablesmeasuring
ponentsateachlevelofourdata(individualandcountry-
people’s perceptions of economic performance. Thus, to the ex-
tentthateconomicperceptionsdeviatefromobjectivereality,our level).Theseestimationsallowustoestablish(a)whether
estimatesof the“true”economiceffectsmaybebiased.Togeta corruption is a significant determinant of system sup-
roughsenseofhowwellobjectiveeconomicconditionsmatchup
portonceweallowtheinterceptstovaryacrosscountries
witheconomicevaluations,weconstructedadatasetthatincluded
theWestEuropeancountriesincludedinoursample(comparable and obtain better estimates of standard errors; and (b)
datawerenotavailablefortheothercountries).Usingsurveydata whether our macro-variables explain a substantial pro-
drawnfromtheEurobarometerandobjectiveeconomicindicators
portionofthecountry-levelvarianceinorderforustobe
for1991–95,thecorrelationofgrowthandretrospectiveevaluations
abletoclaimthatwehaveminimizedapotentialomitted
ofnationaleconomywasarobustandstatisticallysignificant.68
(p<.01);thecorrelationofgrowthandretrospectiveevaluations variablebias.
ofpersonalfinancialsituationwasasimilarlystrong.67(p<.01).
10A number of scholars have argued that levels of economic de-
AnalysisofVariance
velopmentareassociatedwithlevelsofdemocracy.However,most
analystshavenotdrawnadirectlinkbetweeneconomicdevelop-
mentanddemocraticlegitimacy;insteadtheyhavefocusedona Todetermine,first,whetherthereissignificantvariation
moreindirectconnectionofthetwoviademocraticstability.We insystemsupportattheindividualandcountrylevels,we
alsoincludedameasureofdevelopmentbecausecorruptionismore
estimated an ANOVA model that decomposes the vari-
prevalentinlessdevelopedcountries,anditisconceivablethatany
anceinthedependentvariables,where
effectforcorruptiononsystemsupportmightpickupboththeac-
tualeffectofcorruptionaswellastheeffectofacountry’swealth. Support =(cid:3) +(cid:4) +ε
ij 00 0j ij
11Democraticlongevitywasmeasuredbythenumberofyearsof
In this model, (cid:3) is the grand mean of support. The
continuousdemocracysince1920.Weusedtheloggedvaluesofthis 00
variabletoreducetheeffectsofextremecases,thoughtheresultsdo sourcesofcross-nationalvariation,whichcauseparticu-
notchangenoticeablywhenthesimplescaleisusedintheanalyses. larcountriestodeviatefromthismean,arecontainedin
12Followingtheargumentthatparticipationenhancesfeelingsof
trust and external efficacy (and vice versa) (cf. Finkel 1987), we
13WeusedMLwiN1.10.0006(2000)toestimatethesemodels(see
includedavariabledistinguishingvotersandnonvoterstocontrol
alsoRasbashetal.1999).
fordifferencesattributabletohavingparticipatedintheelection.
Includingthisvariablealsoensuresthatweproperlyattributeany 14Awaytocapturevariableinterceptswouldbetointroduceaseries
effectsforbeingamongthemajorityandminority.Weexpectthat ofdummyvariablesforeachcountryinthedataset,butone.In
thosewhodidnotparticipateintheelection(coded1)willhave ourcase,however,thissolutionisimpossible,becauseof perfect
morenegativeattitudestowardthepoliticalsystemthanthosewho multicollinearity among the macro-level variables and the set of
did(coded0). countrydummies.
CORRUPTIONANDATTITUDESTOWARDGOVERNMENT 99
TABLE1 ANOVA minoritystatus,aswellastheinteractiveeffectsof these
variables,onevaluationsofpoliticalsystemperformance.
Estimate We show four models: Models 1 and 2 are random in-
tercept multilevel maximum likelihood IGLS (Iterative
EvaluationofSystem Trustin
GeneralizedLeastSquares)modelswithandwithoutthe
Parameter Performance CivilServants
interaction terms of corruption and majority-minority
FixedEffects
status. In addition, we report the results of parametric
∗∗∗ ∗∗∗
Constant 2.555 2.561
andnonparametricbootstrappingestimations(Models3
(.082) (.079) and4).16 Theresultsfromthebootstrappedestimations
Variance
servetoexaminetherobustnessoftheIGLSestimates,in
∗∗∗ ∗∗
Components .107 .098
particulartheresultsinvolvingthemacro-levelvariables
Country-level (.038) (.035)
(suchascorruption),giventhatourlevel2variablesare
∗∗∗ ∗∗∗
.417 1.069
measured at the level of countries and thus for a fairly
Individual-level (.004) (.010)
smallnumberofcases.
−2loglikelihood 41,886.88 62,023.61 Regardlessofestimationmethod,theresultsprovide
unambiguousevidenceinsupportofourmainhypothe-
Notes:Entriesaremaximumlikelihood(IGLS)estimates;standard
sis:individualsincountrieswithhigherlevelsofcorrup-
errorsinparentheses.
∗p<0.05;∗∗p<0.01;∗∗∗p<.001. tionevaluatetheperformanceofthepoliticalsystemmore
negatively.Inshort,corruptionbreedsdiscontentwiththe
(cid:4) ,andε containssourcesofinter-individualvariation. performanceofthepoliticalsystem.Theresultsobtained
0j ij
withtherandominterceptmodels(Models1and2)show
Theargumentthatbothlevelsof analysisareimportant
thatthecorruptioncoefficientwasstatisticallyandsub-
for understanding attitudes toward the political system
stantively significant. As importantly, the bootstrapped
issupportedifbothvariancecomponentsarestatistically
results shown in Models 3 and 4, which reexamine the
significant(cf.SteenbergenandJones2002).
resultsofthefullyinteractiveModel2,indicatethatthese
Table 1 shows the ML estimates of the grand mean
resultswereextremelyrobust.Visualinspectionofthedis-
andthevariancecomponents.Bothvariancecomponents
tributionofthecoefficientsforcorruptionshowedthem
arestatisticallysignificant,suggestingthatthereissignif-
tobenormallydistributed.
icantvarianceinsystemsupportatbothlevelsof analy-
Ourresultsalsoshowthatthosewhovotedforparties
sis.ResultsoftheANOVAmodelshowthatcountry-level
in power had more positive attitudes toward the politi-
varianceisproportionallymuchsmallerthanindividual-
calsystem.Asexpected,citizensinthepoliticalmajority
level variance. Given that the data are measured at the
thoughtthepoliticalsystemworkedbetterthandidthose
individual level, this is not surprising (Steenbergen and
intheminority.Again,allfourmodelsusingdifferentesti-
Jones 2002, 231). Specifically, individual-level variance
mationtechniquesshowedthatthisresultwassignificant
constitutes79.6percentofthetotalvarianceinthegeneral
andextremelyrobust.
systemsupportmodel,whereasthispartofthevarianceis
To better understand the estimated substantive im-
evengreaterinthetrustofcivilservantsmodel(91.6per-
pactof thevariablesof interest,wecalculatedhowaver-
cent).15TheresultsoftheANOVAmodelindicateclearly
agerespondents’evaluationsofthepoliticalsystemvaried
that there is significant variation in system support for
both dependent variables with regard to both levels of
analysis. Thus, we now turn to the question of whether 16Bootstrappingproceduresestimatethesamplingdistributionof
themodelwehavespecifiedcanaccountforthisvariance. astatistic,treatingthesampleasapopulation.Wepresentresults
estimatedbybothparametricandnonparametricbootstrapping.
Intheparametricbootstrapprocedure,MLwiNresamplesfromthe
TheEffectsofCorruptionandPolitical residualdistributionspecifiedbythemodelateachlevel,whichin
ourparticularmodelisassumedtobenormal.Themodelisthenre-
AllegianceonEvaluations
fittedusingtheseresponses.Intheparametricbootstrapprocedure
ofSystemPerformance
confidenceintervalsareestimatedinaccordancewiththenormal
distributionsamplingtheoryandareeasytointerpret.However,
Directeffects. Table2showstheresultsofseveralmodels sincewedidnothaveanystrongreasonstobelievethatourresiduals
estimatingthedirecteffectsofcorruptionandmajority- aredistributednormally,wealsoperformednonparametricboot-
strapping,whichdoesnotrequireanydistributionalassumptions.
However,notethatthemorecomplexcomputationofconfidence
15Thesecalculationsarebasedontheratiosofeachvariancecom- intervalsinthenonparametricbootstrapprocedurecallsformore
ponentrelativetothetotalvarianceinsystemsupport(cf.Brykand cautionwhendrawinginferencesonthebasisofresultsobtained
Raudenbush1992;SnijdersandBosker1999). withthisprocedure(Mooney1996).
100 CHRISTOPHERJ.ANDERSONANDYULIYAV.TVERDOVA
TABLE2 EffectsofCorruptionandPoliticalAllegianceonEvaluationsofPoliticalSystem
Performancein16Democracies
EvaluationsofSystemPerformance
Model1 Model2 Model3 Model4
Random RandomIntercept Parametric Non-Parametric
IndependentVariable InterceptModel ModelwithInteraction Bootstrapping Bootstrapping
FixedEffects
Constant 2.423∗∗∗ 2.542∗∗∗ 2.619∗∗∗ 2.586∗∗∗
(.475) (.451) (.522) (.572)
Corruption −.078∗∗∗ −.092∗∗∗ −.101∗∗∗ −.092∗∗∗
(.023) (.022) (.023) (.024)
Majority-minoritystatus .149∗∗∗ .042∗∗ .042∗∗ .043∗∗
(1=majority,0=minority) (.011) (.018) (.017) (.018)
Corruption∗majority-minority .037∗∗∗ .037∗∗∗ .037∗∗∗
status (.005) (.005) (.005)
Nonvoter −.021 −.022∗ −.022∗ −.022∗
(1=didnotvote,0=voted) (.013) (.012) (.012) (.012)
Socioeconomicstatus .059∗∗∗ .058∗∗∗ .058∗∗∗ .057∗∗∗
(high=highsocialclass) (.005) (.005) (.005) (.005)
Interestinpolitics .020∗∗∗ .021∗∗∗ .020∗∗∗ .021∗∗∗
(high=greatinterest) (.005) (.005) (.005) (.005)
Employmentstatus .080∗∗ .079∗∗ .080∗∗∗ .079∗∗
(high=employed) (.026) (.026) (.023) (.026)
Gender −.065∗∗∗ −.065∗∗∗ −.065∗∗∗ −.065∗∗∗
(1=female) (.011) (.011) (.011) (.010)
Age .000 .000 .000 .000
(high=old) (.000) (.000) (.000) (.000)
Education .012∗∗ .011∗∗ .011∗ .011∗∗
(high=higheducation) (.004) (.004) (.005) (.004)
GNPpercapita .013∗∗ .013∗∗ .011∗ .013∗
(.005) (.005) (.006) (.006)
Economicgrowth .032∗ .033∗ .030 .032∗
(percentchangeinGDP) (.017) (.016) (.019) (.019)
Democraticage −.032 −.011 .004 −.010
(inyearssince1920,logged) (.054) (.052) (.057) (.052)
Democracyscore −.037 −.057 −.062 −.065
(FreedomHouseindex) (.078) (.074) (.072) (.117)
VarianceComponents
Country-level .017∗∗ .016∗∗ .023∗∗ .024∗
(.006) (.006) (.010) (.012)
Individual-level .384∗∗∗ .383∗∗∗ .382∗∗∗ .382∗∗∗
(.005) (.005) (.005) (.004)
N 13,867 13,867 13,867 13,867
−2loglikelihood 26142.230 26086.050 27537.440 27597.070
Notes: Estimates are maximum likelihood estimates (IGLS) and bias corrected bootstrap estimates; standard errors in parentheses.
Bootstrap estimates and standard errors are computed using resampling of the residuals with 5 sets of 300 replicates in parametric
bootstrappingand5setsof500replicatesinnon-parametricbootstrapping.
∗p<.05;∗∗p<.01;∗∗∗p<.001.