Social Identities and Political Cleavages: The Role of Political Context

Author(s): Robert Andersen and Anthony Heath

Reviewed work(s):
Source: Journal of the Royal Statistical Society. Series A (Statistics in Society), Vol. 166, No. 3
(2003), pp. 301-327
Published by: Blackwell Publishing for the Royal Statistical Society
Stable URL: http://www.jstor.org/stable/3559745 .
Accessed: 09/05/2012 12:08

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing and Royal Statistical Society are collaborating with JSTOR to digitize, preserve and
extend access to Journal of the Royal Statistical Society. Series A (Statistics in Society).

http://www.jstor.org
J. R. Statist.Soc. A (2003)
166, Part3, pp.301-327

Social identities and political cleavages: the role

of political context

Robert Andersen
Universityof WesternOntario,London,Canada, Universityof Oxford,UK,and Centre
forResearch into Electionsand Social Trends,Oxford,UK

and AnthonyHeath
Universityof Oxfordand CentreforResearch into Electionsand Social Trends,
Oxford,UK

[ReceivedApril2002. RevisedOctober2002]

Summary.Usinga novelmethod,the paperinvestigatesthe influenceof social groupidenti-

ties on attitudesandon votingin a varietyof politicalcontexts.Examiningthe majorregionsof
Britain,Canadaandthe USA,we findconsiderablenationaland regionaldiversityinthe nature
of socialcleavages.Forexample,socialclass and racehad widelydifferenteffectsacross so-
cieties,butwithinsocietiestheireffectson attitudesandon votingwereverysimilar.However,
despitethat,age and religionhad a similareffecton attitudesacross societies;the effectson
votingvariedconsiderably. The significantwithin-countrydifferencesunderlinethe importance
of using region,ratherthan country,as the unitof analysis.Moreimportantly, these results
highlightthe roleof politicalcontext,especiallycompetingcleavagesandthe structureof party
competition, relevantsocialcleavages.
inthe establishmentof politically
Keywords:Comparative methods;Politicalsociology;Socialcleavages;Socialand political
attitudes;Votingbehaviour

1. Introduction
The study of the relationshipbetweensocial structureand supportfor politicalpartieshas a
strongtraditionin politicalsociology(see Lazarsfeldet al. (1944),Berelsonet al. (1954),Alford
(1967),Lipsetand Rokkan(1967),Butlerand Stokes(1974)and Hout et al. (1999)).To vary-
ing degrees,social class, gender,age, race and religionhaveall been found to affect voting in
Westerndemocracies.The underlyingmechanismfor this relationship-i.e. that differencesin
socialpositionareassociatedwithdifferentconcerns,thatin turnencouragesupportforparties
that representtheseconcerns-is assumedto be mediatedby social and politicalattitudes(e.g.
Vanneman(1980),Schwartzand Huismans(1995),Brint(1984),Kelleyand Evans(1995)and
Weakliem(1991, 1993)).In the event that a particularsocial groupis not a relevantpolitical
cleavagecharacterizedby uniquevotingpatterns,it is often attributedto a lack of attitudinal
differencesbetweenthat groupand otherrelevantsocial groups.This type of argumentis also
usedto explainan apparentdeclinein cleavagevotingand an increasein moreindividualvoting
based on 'post-materialist' values (Inglehart, 1987, 1990, 1997).

Addressfor correspondence:Robert Andersen, Department of Sociology, University of Western Ontario,

London, Ontario,N6A 5C2, Canada.
E-mail:robert.andersen@uwo.ca

? 2003 RoyalStatisticalSociety 0964-1998/03/166301

302 R.AndersenandA.Heath
An alternative, although not necessarily opposing, view sees a lack of cleavage voting as
reflecting inadequate political party representation. Growing empirical evidence suggests that
there is indeed a reciprocal relationship between the nature of a polity and cleavage struc-
tures (Amorim Neto and Cox, 1997). There is significant evidence, for example, that the type
of electoral system helps to determine the number of viable political parties in a given polity
(Duverger, 1954; Sartori, 1976; Riker, 1982; Lijphart, 1990, 1994). If there are insufficient par-
ties, all major cleavages may not be uniquely represented.This could be detrimental to a particu-
lar social cleavage because it cannot become politically relevant unless there is a party explicitly
drawing on intergroup conflicts to mobilize support (Bartolini and Mair (1990); see also Manza
and Brooks (1999)). Moreover, as Kiltschelt and Hellemans (1990), page 20, argued, although
parties emerge from social and political cleavages, they
'areusuallyableto influencethe extentto whichcertaincleavagesarearticulatedin the populationand
contributeto the mobilizationor demobilizationof groupdivisions'.
A whole array of factors associated with the political context of a society could affect whether
political parties represent the interests of a social group-the size and social importance of the
cleavage, the number of competing cleavages and the type of electoral system, all could have an
effect.
Few empirical studies have simultaneously assessed the effect of political context on the ef-
fects of social structure on attitudes and voting (an exception is Weakliem (1993)). This paper
fills some of this void by exploring the effects of social identities on attitudes and voting across
a variety of political contexts. Examining the major regions of Britain, Canada and the USA,
we make some distinctive moves. First, we focus on regional differences within countries, as
well as on the usual national differences since political context often varies in important ways
between regions. Secondly, we examine the influence of many social identities, exploring the
effects of each individually but controlling for the others, ratherthan simply explore the effects
of a single identity. Finally, we develop a novel method that facilitates the explicit comparison
of the effects of social group identities on attitudes with their effects on voting, both within and
across societies.

2. Theoreticalbackground
2. 1. Social and politicalattitudesin Westerndemocracies
Social and political attitudes in Western democracies can generally be divided along two di-
mensions: left-right attitudes and liberal-conservative attitudes (see Berelson et al. (1954), Fle-
ishman (1986), Feldman (1988), Evans et al. (1996) and Baer et al. (1995)). Left-right issues
pertain to the state's role in the economy, with those on the left being more favourable towards
Government intervention, social spending and redistribution policies, and those on the right
being more favourable towards a free-market economy. Liberal-conservative issues are related
to personal freedom of thought, association and life styles. Those who are liberal in attitude
are tolerant of alternative life styles, whereas those on the conservative end are more likely to
favour traditional life styles and values, and to dislike social change.
By no means is there a consensus that left-right issues and liberal-conservative issues remain
of paramount importance. A leading proponent of the new issues argument is Inglehart (1987,
1990, 1997) who argued that post-materialist issues have begun to replace the traditional ideo-
logical dimensions in recent decades. Economic security has allegedly decreased the importance
of left-right issues, freeing voters to focus on other social and cultural issues. We could also
interpret the rise in prominence of nationalist issues as indicative of post-industrial society.
 andPoliticalCleavages
SocialIdentities 303
In fact, somearguethatnationalistissuesoccupya distinctideologicaldimensionof theirown
(Heath et al., 1999).For the most part, however,'post-materialist'issues can be incorporat-
ed withinthe traditionaltwo-dimensionframework,especiallywithinthe liberal-conservative
dimension(Weakliem(1991),page 1330),and even national issues generallycoexist with the
mainissuedimensions.The latteris evidentin Quebecwherethe Bloc Quebecoiscampaignsfor
independenceundera socialdemocraticplatform.

2.2. Thesociological approachto cleavages

Politicalsociology generallysees these fundamentalideologicaldimensionsas reflectionsof
social identitiesformedon the basis of social structureor social groupmemberships(see for
exampleLazarsfeldet al. (1944),Alford(1967),Rose and Urwin (1970)and Lijphart(1979)).
The left-rightdimensionis rootedin social class, with the working-classtendingto favourin-
terventionistand incomeredistributionpolicies.The liberal-conservative dimensionis closely
relatedto religiosityand education,with activemembersof religiousgroupsexhibitingmore
conservativesocial values,and highlyeducatedpeople beingmoreliberalminded.
The sociologicalapproachoriginateswith Lazarsfeldet al. (1944) (see also Berelsonet al.
(1954))who statedthe conditionsfor persistentgroupvoting as the following:
(a) initialsocial differentiationsuchthat the consequencesof politicalpolicy are materially
or symbolicallydifferentfor differentgroups;
(b) conditions of transmittabilityfromgenerationto generation;
(c) conditions of and
physical social proximityfor continuedin-groupcontactin succeeding
generations
(Berelsonet al. (1954),page75). Otherinfluentialresearchby Lipsetand Rokkan(1967)argued
that social cleavagesare so importantto a polity that they determinethe structureof party
competition.This explanationsuggests,then, that social groupidentitiesinducegroupnorms
supportinga particularpoliticalparty.Althoughit is not explicitfromLazarsfeldand his col-
leagues,this impliesthat these groupprocesseswill also generatean intragroupconsensuson
social and politicalattitudes.If this holds, societieswith similarsocialgroupstructuresshould
displaysimilarsociopoliticalattitudesand partysupport.
Recentempiricalresearchcastssomedoubt on this argument.Forexample,social classdoes
not havethe same influenceon voting acrosssocieties(see Evans(1999)).One possibleexpla-
nationforthis variabilityis thatthe groupsthemselvesmaybe less developedin some countries
comparedwith others,and that the secondand thirdof Berelson'sconditionsdescribedabove
may applyless strongly.This is the classic explanationfor the absenceof class voting in the
USA: as Sombart(1976)argued,highratesof intergenerational socialmobilityimplythatthere
will be little intergenerationaltransmissionof values and relativelylittle social proximityor
continuedin-groupcontact in succeedinggenerations.It is rarelypossible to test this expla-
nation sincecross-nationaldata on within-groupprocessesare difficultto come by. A testable
inferencefrom this argument,however,is that social groupidentitieswill haverathersimilar
effectson attitudesandvotingwithina givencountry.Thusin North Americawe mightexpect
thatthe absenceof strongandenduringwithin-classsocialrelationshipshas led both to a weak
relationship between class and left-right attitudes and to a similarly weak relationship between
class and party.

2.3. Considering political context

The simple sociological model may work well in a country such as Britain where, at least his-
torically, a cohesive social group has a strong political presence-i.e. a viable political party
304 R.AndersenandA.Heath
representsthe group'sconcerns.However,if a socialgroup'sconcernsarenot clearlyrepresent-
ed by a viablepoliticalparty,membersof thatgroupmayvote on the basisof some othersocial
groupidentity.Considera polity withouta strongworking-classparty (e.g. the USA), where
working-classvoterscan choose only betweenpartiesthat do not explicitlyaddressclass con-
cerns.In such a polity,working-classvotersmay be inclinedto vote accordingto the concerns
of anothergroupidentityand so may vote on religiousratherthan classlines.
Now contemplatea polity with a politicalparty that is both secularand pro working-class.
Assume that no other partiesrepresentthe working-classbut all other partiesrepresentreli-
gious concerns.Thosewho are both religiousand working-classmay decideagainstvotingfor
the working-classpartysinceit rejectstheirreligiousinterests.Likewise,thosewho areopposed
to policiesfavouringreligiousinterestsmayvote for the working-classpartyregardlessof their
socialclass.If non-religiousvotersconstitutea significantportionof the population,theremay
be no differencebetweenclassesin theirpropensityto vote left. Althoughwe are suggesting
herethat the structureof partycompetitioncan influencethe associationbetweensocialgroup
identitiesandpartysupport,it does not followthattherewillnecessarilybe a similarassociation
betweensocial groupidentitiesand sociopoliticalattitudes.In other words,just becauseclass
votingis low, it does not meanthatthe classpolarizationof attitudeswill be anydifferentfrom
that in a societythat is characterizedby a high level of classvoting.
Broadlyspeaking,if politicalcontextis irrelevant,each social groupidentityshouldhavea
similareffecton votingas it does on attitudeswithinsocieties.If politicalcontextdoes matter,
the effects of social groupidentitieson attitudesand voting should vary independently,and
shoulddifferacrosssocieties.It is widelyknownthatvotingdoes not alwaysreflectsocialstruc-
ture in a similarmanneracrosscountries.Less is known aboutwhetherattitudesare affected
in the samemannerandunderwhatconditionsthe two effectsmovein tandemor areindepen-
dent. To shedlight on theseissues,we applythis theoryto an analysisof Britain,Canadaand
the USA. In manywaysthese countriesare very similar:they are modernindustrialsocieties
with similarsocialclassstructures,have'firstpast the post' electoralsystems,havea largeEng-
lish-speakingmajorityand havesignificantminoritypopulations.They are also characterized
by regionalcleavages,enablingus to dividethe countriesaccordingly.More importantly,the
political context of these countriesis known to differ in ways--e.g. specificallyin terms of
partystructures-thatmay affectcleavagevoting.

3. Dataand methods
3. 1. Data and samples
Our analysisis restrictedto the period between 1992 and 2000, and is based on seven high
quality surveysemployingnationallyrepresentativesamples.Our desireto examinepatterns
at the regionalratherthan nationallevelrequiresus to combinedata sets for each countryto
increasethe samplesizesand statisticalpower.In all cases,modelswereinitiallyfittedfor each
data set separatelyto ensurecompatibility.
For the US analysiswe use a combineddata set of the 1996and 2000 AmericanNational
ElectionStudies(ANESs).TheCanadiananalysisis basedon a dataset combiningthe 1993and
1997 Canadian Election Studies (CESs). Finally, the British analyses are based on a combina-
tion of the 1992 and 1997 British Election Studies (BESs) and the 1999 Welsh Assembly Election
Study (WAES). These studies are ideally suitable because they contain extensive information
about respondents' social background, attitudes and voting behaviour. We restrict our analysis
to only those who reported voting in the most recent election. More details of the data, including
response rates, can be seen in Appendix A.
 SocialIdentitiesandPoliticalCleavages 305
WeanalysefourCanadianregions(samplesizes,afteromittingmissingcases,arein parenthe-
ses): the western provinces (N = 1939), Ontario (N = 1308), Quebec (N = 1206) and the eastern
provinces(N = 594).The Britishdata aredividedinto fiveregions:Scotland(N = 1642),Wales
(N = 1169), northernEngland(N = 1304), the midlands of England (N = 1168) and south-
ern England(N = 2006). The US data are divided accordingto the US census'sfour major
regions: the Northeast (N = 503), Midwest (N = 808), South (N = 1131) and West (N = 646).
The criteriafor distinguishingregionswerea combinationof theoreticalrationalesand prac-
tical considerations.Quebec,Scotlandand Walesall haveuniquepoliticalcontextsthat might
be expectedto inducedifferentsocial identityeffects-i.e. nationalcleavagesand partiesrep-
resentingthem-and werethus separatedaccordingly.Otherregionaldivisionswerebased on
geographicaldivisionsand samplesizes.

3.2. Dependent variables

Ourfour dependentvariablesareleft-rightattitudes,liberal-conservativeattitudes,rightvote
and left vote. The attitudevariablesare operationalizedusing two separatescales.High scores
on the liberal-conservativescaleindicatesociallyconservativeattitudes;highscoreson the left-
right scale indicatelaissez-faireattitudes.Since few questionsare common to all the surveys
employed, we could only constructscalesthat are functionallyequivalent.Functionalequiva-
lenceis sufficient,however,sinceour interestis in comparingsocietaldifferencesin the relative
effect of social variableson these attitudesratherthan absolute societal differences.Details
of the scales, includingthe specificitems that were used for each countryand the reliability
coefficients,arein AppendixB.
Each of the regionshas partiesthat can be placed into a left or liberalgroup and a right
or conservativegroup.For simplicity,we referto these groups as 'left vote' and 'rightvote'
respectivelythroughoutthe paper.Wecontrastvotingfor thesepartieswithvoting for all other
parties,includingcentrepartiesand nationalistparties,but excludingnon-voters.For all Can-
adian regionsthe ProgressiveConservativePartyand the ReformPartyare groupedtogether
as the rightparties,and the New DemocraticPartyis coded as the left party.For the British
analyses,the LabourParty is coded as the left party, and the ConservativePartyis treated
as the right party.We follow conventionby treatingthe Democratsas the left party and the
Republicansas the rightpartyin the USA (see Hout et al. (1999)).

3.3. Explanatoryvariables
The modelsincludesevensocialgroupexplanatoryvariables:age, gender,religiosity,race,na-
tionalidentity,educationand socialclass.All theseconceptsarecoded as categoricalvariables
and, exceptfor raceand nationalidentity,arecoded identicallyfor each region.
Age is dividedinto threecategories,18-34 years, 35-64 years and 65 years and over, and
operationalizedas two dummyregressorswith the youngestage categoryas the reference.Gen-
der is treatedas a dummyvariablewith womencoded as the referencecategory,and education
is coded as a dummyvariablefor those who havea universitydegree.
Religiosityis dividedinto fourcategories:practisingProtestants,practisingCatholics,those
who regularlypractise another religion and those who are less religious (the reference category).
The USA and British data included a question on the frequency of attendance at church so
that is employed here. Respondents who attended church at least once a month were classified
as 'practising'. The Canadian data did not contain a similar question on religious practices,
but rather a question asking respondents how important religion was to them. Only those who
claimed that it was 'very important' were coded as 'practising'.
306 R. Andersenand A. Heath
FollowingfromHout et al. (1999)social class is dividedinto five categories:managers,pro-
fessionals,clericalworkers(routinenon-manuallabour),self-employedand working-class(or
manuallabour).Becausetherewas a significantportionof missingcases in most regions,we
also includea sixthcategoryfor those whose occupationwas unknown.Social class is treated
as a set of five dummyregressorswith working-classas the referencecategory.
For all the US regions,race is dividedinto threecategories:black, other visibleminorities
and white(the referencecategory).The numbersof non-whitesweretoo smallin Canadaand
Britainto allow dividingthem into two categories,so raceis operationalizedsimplyas visible
minoritiesversuswhite(the referencecategory).
National identityis measuredonly for Quebec,Scotlandand Wales,whereit is treatedas a
dummyvariablefor the minoritynation versusothers.Those for whom Frenchwas theirfirst
languageand it was still spoken in the home were classifiedas Quebecoisnationalidentity.
For Scotlandand Wales,nationalidentitywas measuredsimplyby respondents'self-reportsof
Scottishor Welshidentity.

3.4. Modelling
procedure
Left-rightandliberal-conservative attitudeswereregressedseparatelyon the socialgroupvari-
ablesby using ordinaryleast squaresregression.Separateprobitmodelswereused to regress
left vote andrightvote on the samesocial groupvariables.Weadjustedthe coefficientsfromthe
ordinaryleastsquaresmodelsof attitudes-in effectstandardizingtheregressionequations-so
thatthey arecomparablewith the correspondingcoefficientsfromthe probitmodelsof voting.
Cross-regionalsimilaritiesand differencesin the relativeeffectsof socialgroupvariableson at-
titudesandvotingarethenexaminedby usinggraphs.Moredetailsof thismethodaredescribed
below.
We start by discussingthe probitmodels of left and right vote. The generalprobitmodel
assumesthat a binarydependentvariabley(1,0) representsa continuouslatentvariableZ. For
each combinationof predictorsthereis a differentmeanp, whichrepresentsthe cut-offpoint
whereZ > 0 (the point wherethe observeddiscreteresponsevariablechangesfrom0 to 1). In
our case,Z representsthe 'propensity'to vote left (or right)andy representsthe actualleft vote
(or rightvote). This model assumesthat the errorsare independentand normallydistributed
and thatthe residualvarianceequals 1 (Fox (1997),page447). The basicvotingmodelis

7i
= 4(a + )P•l.xii + 02Xi2 + ... - + kXik) (1)
where7rirepresentsthe probabilitythat y = 1 (in our case, vote for the party),4 is the normal
cumulative distribution function and a + Plxil + 02xi2 + ... + /kxik represents the mean of
Z given the social identityexplanatoryvariables,and the ps representthe slopes of the social
groupidentityvariables.Pl can be interpretedas the increasein Z for a 1-unitincreasein xl,
holdingall otherexplanatoryvariablesconstant.
We now turn to the attitudemodels.One way of standardizingregressionequationsis to
constrainthe mean-squarederrorMSE of the two models to be equal. Recallthat the probit
model assumesthatthe residualvariance,andhenceMSE, of Zlxl, x2,..., xk is equalto 1. We
can adjust the coefficients and standard errors from ordinary least squares regression models
so that MSE = 1 by using the formulae

/' = f/PJMSE, (2)

SE(/)' = SE(,3)/VMSE. (3)

 SocialIdentities
andPoliticalCleavages 307
Althoughthis adjustmentis arbitraryfrom a substantivepoint of view, it is sensiblecon-
sideringthat our dependentvariables,left-right attitudesand liberal-conservativeattitudes,
haveartificialmetrics.Sincewe havedifferentmeasuresof attitudesin each of the data sets, we
cannot determinedifferencesin attitudesacross societies.However,this method allows us to
comparethe relativeeffectsof socialidentitieson attitudesand votingacrosssocietiessensibly.
For example,our methodcannotdeterminein whichsocietywomenaremost liberal,but it can
show wherethe influenceof genderon attitudesand voting is strongest.In other words,this
methodallowsus to comparelevelsof grouppolarizationin termsof attitudesand voting,and
differencesand similaritiesin thesetwo types of polarization,acrosssocieties.
Ratherthan only examinethe coefficientsin tables,whichis not conduciveto determining
patternsof associationwhen thereare many regionsunderanalysis,we plot them on graphs.
The horizontalaxis on thesegraphsrepresentsthe adjustedcoefficientfromthe attitudesmod-
els; the verticalaxis representsthe correspondingcoefficientfromthe probitmodels of voting.
Simplyput, the horizontalaxis tells us the level of polarizationin termsof attitudes,whereas
the verticalaxis displaysthe polarizationin termsof voting.Regionsaredepictedby the codes
in Table1. Each graphincludesa LOESSnonparametricregressionline of the cross-societal
trend(see Fox (2000)).

4. Results
The unadjustedcoefficients,and theirstandarderrors,fromthe models of left-right attitudes
and liberal-conservative attitudesare in Tables2 and 3. The residualstandarderrorfor each
modelis also reportedso thatthe adjustedcoefficientsusedin the plotscan be calculated.Tables
4 and 5 displaythe probitmodelsfitted to the rightand left vote respectively.A glance at the
tablesdoes indeedshowthatthereare significantgroupdifferencesin attitudesand voting.We
now turnto the graphsfor a detaileddiscussionof the effectsof age,race,religiosityand social
class on attitudesand votingin the variousregionsthat we havedistinguished.

4.1. Effects of age

It is well knownthatpeoplegenerallybecomemoreresistantto changeand more sociallycon-
servativeas they age (Alwinand Krosnick,1991;Glenn, 1980;Markus,1979).In none of the
regionsthat we examineis therea party that uniquelyrepresentsthe old, but the party struc-
tures are similarin that they havedistinctright-wingpartieswith conservativeplatforms.All
else beingequal,we shouldexpectthatthe old (those65 yearsof ageand over)will be morecon-
servativethan the young(thoseunder35 yearsold), and thatthis conservatismwill be reflected
in a higherpropensityfor the old to vote for right-wingparties.

Table1. Regionalcodes

US codes Canadiancodes GreatBritaincodes

Northeast:USNE East: CAEA Scotland:GBSC

North Central:USNC West:CAWE Wales:GBWA
South:USSO Ontario:CAON North England:GBNE
West:USWE Quebec:CAQU Midlands,England:GBME
South England:GBSE
Table2. Coefficientsfromthe regressionof left-rightattitudeson socialgroupvariablest

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Res

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South Mid
Central England

Age (years)
35-64 -0.34 -0.43 0.07 0.36 -0.10 -0.07 -0.40 -0.51 -0.63t 0
(0.41) (0.38) (0.47) (0.38) (0.22) (0.26) (0.29) (0.35) (0.27) (0
65 and over -0.87 -0.39 0.29 0.56 -0.46 -0.35 0.10 -0.27 -0.06 0
(0.54) (0.45) (0.56) (0.29) (0.35) (0.41) (0.51) (0.56) (0.34) (0
Men 2.82? 1.85? 1.96? 2.39? 0.34 0.49 1.04 0.46 0.34 0
(0.40) (0.37) (0.44) (0.30) (0.20) (0.24) (0.27) (0.32) (0.18) (0
Race
Black -4.21? -3.95? -3.65? -4.03?
(0.88) (0.76) (0.64) (0.33)
Otherminority -1.06t -0.65 -0.41 -1.72? 0.27 -0.13 0.21 0.17 -0.82?? -1.
(0.51) (0.71) (0.77) (0.47) (0.39) (0.43) (0.57) (1.03) (0.25) (0
Religion
Protestant 2.31? 1.76? 0.92 0.81?? 0.36 1.02? -1.01 -0.11 0.72?? 1
(0.47) (0.36) (0.54) (0.28) (0.25) (0.31) (0.88) (0.38) (0.26) (0
Catholic -0.54 1.06?? 0.43 0.33 0.07 -0.86?? 0.80t 0.24 -0.99t -0.
(0.55) (0.39) (0.45) (0.28) (0.32) (0.32) (0.32) (0.38) (0.43) (0
Other -1.06 -0.65 -0.41 0.09 -0.45 -0.42 -1.59 2.24 0.161 1
(0.51) (0.71) (0.77) (0.42) (0.51) (0.24) (1.00) (1.40) (0.80) (1
National identity - - - - - -0.13 - --
(0.38)
Universitydegree 0.58 0.33 0.93t 0.99?? 0.40 0.54 0.28 1.53? -0.01 -0.
(0.41) (0.41) (0.46) (0.32) (0.26) (0.30) (0.35) (0.43) (0.29) (0
Table2 (continued)

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Re

variable of the USA. of Canada:

West North Northeast South West Ontario Quebec East South M

Central England

Social class
Professionals 0.51 0.20 -0.14 0.52 0.56 0.30 0.83 -0.11 1.58?
(0.59) (0.54) (0.65) (0.43) (0.34) (0.38) (0.46) (0.53) (0.31)
Managers 1.80% 0.59 0.84 0.85 0.49 0.84 -0.57 0.09 2.89?
(0.65) (0.56) (0.67) (0.47) (0.40) (0.46) (0.54) (0.66) (0.27)
Self-employed 1.461 1.76? 0.62 1.95? 0.84? 1.04% 1.26% 1.55% 1.99?
(0.65) (0.61) (0.73) (0.47) (0.29) (0.38) (0.46) (0.51) (0.35)
Non-manual 0.64 0.11 -0.32 0.77 0.07 0.56 1.191 0.13 1.35?
(0.50) (0.41) (0.53) (0.35) (0.36) (0.43) (0.47) (0.53) (0.24)
Missing -0.15 0.24 -0.63 0.64 -0.96? -0.03 -0.15 -0.84 1.85??
(0.67) (0.54) (0.72) (0.43) (0.31) (0.35) (0.38) (0.46) (0.54)
Constant 10.69? 10.96? 10.46? 10.22? 10.75? 10.64? 9.87? 9.15? 15.56
(0.52) (0.44) (0.56) (0.36) (0.27) (0.31) (0.50) (0.41) (0.22)
Residualstandard 18.29 17.43 16.29 16.48 17.66 16.72 20.17 13.23 14.86
error
R2 0.15 0.11 0.16 0.23 0.03 0.04 0.04 0.07 0.09
Numberof cases 643 803 643 1127 1922 1294 1202 590 2002

tNumbers in parenthesesare standarderrors.

Sp < 0.05.
?p < 0.001 (two-tailedtests).
?p < 0.01.
Table3. Coefficientsfromthe regressionof Liberal-Conservative
attitudeson socialgroupvariablest

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Res

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South Mi

Central England

Age (years)
35-64 0.941 0.53? 0.54 1.06$ 1.05t 1.73$ 1.19$ 2.29t 1.19t
(0.23) (0.21) (0.28) (0.18) (0.24) (0.29) (0.26) (0.40) (0.23) (
65 and over 1.571 0.65?? 0.97?? 1.381 3.071 3.734 3.88t 4.464 2.59T
(0.30) (0.25) (0.33) (0.23) (0.39) (0.47) (0.45) (0.65) (0.28) (
Men -0.12 -0.51? 0.28 -0.14 0.794 0.88?? 1.101 0.06 0.07 -0
(0.22) (0.21) (0.26) (0.18) (0.22) (0.27) (0.24) (0.37) (0.15) (
Race
Black -0.66 -1.23? -0.18 -1.09
(0.49) (0.42) (0.38) (0.20)
Otherminority -0.12 -0.55 -0.38 -0.37 -0.02 0.53 0.07 -2.20 -0.86$ -0
(0.28) (0.39) (0.46) (0.28) (0.44) (0.49) (0.51) (1.22) (0.21) (
Religion
Protestant 2.71t 1.844 1.304 1.721 3.391 2.814 2.621 2.85$ 1.92$
(0.26) (0.20) (0.32) (0.17) (0.28) (0.36) (0.80) (0.45) (0.22) (
Catholic 1.574 1.244 1.111 0.934 2.674 2.39t 1.671 2.074 0.97%
(0.31) (0.22) (0.26) (0.27) (0.35) (0.37) (0.28) (0.44) (0.35) (
Other 2.384 1.764 1.544 1.78$ 3.084 3.424 1.49 4.474 1.08
(0.31) (0.35) (0.44) (0.26) (0.57) (0.76) (0.90) (1.72) (0.66) (
National identity 1.05?? - -
(0.34)
Universitydegree -1.094 -0.34 -0.69? -0.734 -1.854 -1.644 -1.824 -2.544 -2.574 -2
(0.23) (0.23) (0.27) (0.20) (0.28) (0.35) (0.31) (0.51) (0.24) (
Table 3 (continued)

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Res

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South Mid
Central England

Social class
Professionals 0.55 -0.22 0.21 0.09 -0.94? -1.35$ -0.51 -0.69 -0.92$ -0
(0.33) (0.30) (0.39) (0.26) (0.37) (0.44) (0.41) (0.63) (0.25) (
Managers 0.76? -0.51 0.15 -0.12 -0.93? -0.49 -1.62% -1.49 -0.50 -0
(0.36) (0.31) (0.40) (0.29) (0.44) (0.53) (0.48) (0.77) (0.22) (
Self-employed 0.54 0.83? 0.15 0.09 0.26 -0.99? -0.13 -0.24 0.20
(0.36) (0.34) (0.43) (0.28) (0.32) (0.44) (0.40) (0.59) (0.29) (
Non-manual 0.38 -0.13 0.28 -0.11 -0.95? -1.66% -1.12% -2.45$ -0.60% -0
(0.28) (0.23) (0.32) (0.21) (0.40) (0.49) (0.41) (0.61) (0.20) (
Missing 0.46 0.15 -0.42 0.21 -0.39 -0.71 -0.03 -0.36 -1.37? -
(0.37) (0.30) (0.43) (0.26) (0.34) (0.40) (0.33) (0.54) (0.45) (
Constant 9.22$ 10.49$ 10.10$ 9.96$ 14.40$ 13.87$ 13.06$ 14.10$ 18.93$ 1
(0.29) (0.25) (0.33) (0.22) (0.29) (0.35) (0.44) (0.48) (0.18) (
Residualstandard 5.65 5.36 5.74 6.12 20.59 20.95 15.39 17.30 10.18 1
error
R2 0.26 0.16 0.11 0.16 0.19 0.19 0.21 0.29 0.16
Number of cases 643 803 643 1127 1922 1294 1202 590 2002 1

tNumbers in parenthesesare standarderrors.

$p < 0.001 (two-tailedtests).
?p < 0.05.
??p < 0.01.
Table4. Coefficientsfromthe probitmodelsof rightvotingregressedon socialgroupvariablest

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Res

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South Mid
Central England

Age (years)
35-64 0.06 -0.004 -0.13 0.07 0.06 0.002 0.09 -0.16 -0.14 -0
(0.16) (0.14) (0.19) (0.13) (0.08) (0.09) (0.11) (0.15) (0.11) (
65 and over 0.008 -0.17 0.27 -0.14 0.14 -0.17 0.26 -0.04 -0.009 0
(0.20) (0.17) (0.22) (0.16) (0.12) (0.15) (0.19) (0.16) (0.13) (
Men 0.31? 0.35t 0.46$ 0.55? 0.17?? 0.32$ 0.08 0.16 -0.09 -0
(0.14) (0.13) (0.17) (0.12) (0.07) (0.09) (0.10) (0.15) (0.07) (
Race
Black -1.75? -1.75? -1.17? -2.05? -
(0.55) (0.47) (0.33) (0.23)
Otherminority -0.45? -0.24 -0.42 -0.24 -0.38?? -0.54$ 0.05 -0.19 -0.321 -0
(0.20) (0.27) (0.33) (0.18) (0.15) (0.18) (0.22) (0.45) (0.10) (
Religion
Protestant 0.90? 0.79? 0.71? 0.41? 0.12 0.34$ -0.17 0.35% 0.22% 0
(0.16) (0.13) (0.20) (0.11) (0.09) (0.11) (0.41) (0.16) (0.09) (
Catholic 0.14 0.88$ 0.41?? 0.22 -0.12 -0.44? -0.28? -0.16 -0.49$ -0
(0.20) (0.14) (0.16) (0.16) (0.11) (0.12) (0.12) (0.16) (0.16) (0
Other 0.14 0.01 -0.04 0.09 0.01 0.24 0.06 0.21 -0.32? -6
(0.20) (0.25) (0.33) (0.19) (0.18) (0.28) (0.38) (0.58) (0.10) (0
National identity - - 0.33?
(0.15)
Universitydegree -0.08 0.23 0.06 0.003 -0.17 -0.09 -0.17 -0.02 -0.41$ -0
(0.14) (0.14) (0.17) (0.12) (0.09) (0.11) (0.13) (0.17) (0.11) (0
Table 4 (continued)

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Res

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South Mi

Central England

Social class
Professionals 0.13 -0.14 0.28 0.22 -0.16 0.05 -0.10 -0.02 0.39?
(0.20) (0.19) (0.25) (0.17) (0.11) (0.14) (0.18) (0.22) (0.11) (
Managers 0.23 0.24 0.43 0.04 -0.23 0.14 0.23 0.008 0.65?
(0.23) (0.20) (0.26) (0.17) (0.14) (0.16) (0.19) (0.29) (0.10) (
Self-employed -0.07 0.75? 0.61? 0.41? 0.18 0.23 0.10 0.06 0.57?
(0.23) (0.22) (0.29) (0.18) (0.10) (0.14) (0.17) (0.21) (0.13) (
Non-manual 0.05 0.27 0.11 0.18 -0.14 0.38? -0.03 -0.11 0.39?
(0.19) (0.15) (0.22) (0.15) (0.13) (0.16) (0.18) (0.22) (0.09) (
Missing 0.11 0.31 0.24 0.48? -0.06 0.25 -0.12 0.15 0.55$
(0.26) (0.20) (0.29) (0.20) (0.11) (0.13) (0.15) (0.20) (0.21) (
Constant -0.61$ -0.67? -0.86? -0.40? 0.04 -0.63? -1.21? -0.41?? -0.20? -0
(0.21) (0.18) (0.24) (0.16) (0.10) (0.12) (0.19) (0.17) (0.08) (
X2 54.9? 100.4? 60.6? 192.5? 45.8? 61.6? 18.19 14.14 103.2? 9
Degreesof freedom 14 14 14 14 13 13 14 13 13
Numberof cases 643 803 643 1127 1922 1294 1202 590 2002 1

tNumbers in parenthesesare standarderrors.

Tp < 0.01.
?p < 0.001 (two-tailedtests).
?p < 0.05.
Table5. Coefficientsfromthe probitmodelsof leftvotingregressedon socialgroupvariablest

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Re

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South M

Central England

Age (years)
35-64 0.11 0.05 0.19 0.03 0.02 -0.08 0.41 0.351 0.21
(0.16) (0.14) (0.19) (0.12) (0.10) (0.13) (0.24) (0.17) (0.11)
65 and over 0.29 0.26 -0.17 0.29 0.02 -0.0009 -0.22 -0.47 -0.07
(0.20) (0.17) (0.24) (0.16) (0.15) (0.20) (0.50) (0.32) (0.13)
Men -0.13 -0.17 -0.25 -0.41? -0.31? -0.31?? 0.07 -0.23 0.02
(0.42) (0.13) (0.17) (0.12) (0.09) (0.12) (0.20) (0.16) (0.07)
Race
Black 1.39? 1.62? 1.34? 2.02?
(0.42) (0.38) (0.33) (0.21)
Otherminority 0.56?? 0.26 0.56 0.31 -0.08 -0.07 0.14 0.82 0.30??
(0.20) (0.27) (0.33) (0.18) (0.19) (0.23) (0.39) (0.44) (0.10)
Religion
Protestant -0.78? -0.78? -0.65? -0.37? -0.04 -0.391 -5.19? -0.43t -0.30??
(0.16) (0.13) (0.19) (0.11) (0.11) (0.17) (0.00) (0.20) (0.10)
Catholic -0.21 -0.291 -0.38T -0.17 -0.34t -0.04 -0.23 -0.52? 0.31$
(0.19) (0.14) (0.16) (0.16) (0.15) (0.15) (0.28) (0.19) (0.15)
Other -0.36 -0.11 0.18 0.01 -0.19 -0.07 1.05$ -0.24 0.79%
(0.20) (0.24) (0.33) (0.19) (0.25) (0.23) (0.49) (0.66) (0.29)
National identity - - - - - - -0.14
(0.26)
Universitydegree -0.002 -0.06 -0.16 0.07 0.12 0.28t 0.61?? 0.02 0.10
(0.14) (0.14) (0.17) (0.12) (0.11) (0.14) (0.23) (0.20) (0.11)
Table 5 (continued)

Independent Resultsfor thefollowingregions Resultsfor thefollowingregions Resu

variable of the USA: of Canada:

West North Northeast South West Ontario Quebec East South Mid
Central England

Social class
Professionals 0.04 0.28 0.04 -0.12 -0.03 0.19 0.10 0.08 -0.47? -0.
(0.20) (0.19) (0.25) (0.16) (0.14) (0.19) (0.32) (0.25) (0.12) (0
Managers 0.0004 -0.20 -0.25 0.06 -0.07 -0.20 -0.04 0.23 -0.78? -0.
(0.22) (0.20) (0.26) (0.17) (0.17) (0.25) (0.41) (0.32) (0.11) (0
Self-employed 0.14 -0.55$ -0.64t -0.33 -0.38? 0.08 - 0.03 -0.48? -0.
(0.22) (0.22) (0.28) (0.18) (0.14) (0.20) (0.24) (0.13) (0
Non-manual 0.11 -0.05 0.09 0.006 -0.421 0.06 0.11 -0.004 -0.37? -0.
(0.19) (0.15) (0.22) (0.12) (0.18) (0.23) (0.38) (0.26) (0.09) (0
Missing 0.04 -0.14 -0.07 -0.29 0.05 0.30 0.27 0.17 -0.81?? -0.
(0.25) (0.20) (0.28) (0.20) (0.14) (0.18) (0.30) (0.24) (0.25) (0
Constant 0.14 0.27 0.471 0.006 -0.88? -1.27? -2.45? -0.85? -0.40? -0.
(0.20) (0.17) (0.24) (0.15) (0.12) (0.16) (0.39) (0.20) (0.08) (0
X2 45.6? 84.9? 63.7? 193.7? 35.1?? 24.71 27.51 27.21 125.1? 123
Degreesof freedom 14 14 14 14 13 13 13 13 13 1
Numberof cases 643 803 643 1127 1922 1294 1202 590 2002 11

tNumbers in parenthesesare standarderrors.

tp < 0.05.
?p < 0.001 (two-tailedtests).
??p< 0.01.
316 R.AndersenandA.Heath
1.2 .4
USNC

GBSC GBWA
CA
GBTE
USNE RWoN
S CAQU
41
GBSC
S
.4.4 CAEA
USNE CAQU
CAWE
USWE
0.0 C -.8
USSO
U , -A-
USNC CAON

-.4 -1.2
-.4 0.0 .4 .8 1.2 -.4 0.0 .4 .8 1.2

Liberal-Conservative
attitudes Liberal-Conservative
attitudes
Fig. 1. Effectsof age (65 years and over-under35 years contrast)on liberal-conservative
attitudesand
voting:0, USA;E, Canada;A, GreatBritain

Fig. 1 displays the adjusted coefficients for the old-young contrast from the models of liber-
al-conservative attitudes (plotted along the horizontal axes) compared with left and right vote
(on the vertical axes). As expected, age has a similar influence on attitudes in all regions-the
coefficients measured on the horizontal axis are all positive, and statistically significant, indicat-
ing that the old are more conservative than the young. However, controlling for the other major
social identities, age does not generally affect voting patterns. Only in Scotland and the north
of England is there a statistically significant relationship between age and right vote. Moreover,
there is no apparent cross-societal pattern. The findings are similar for the left vote. Once again,
aside from Scotland and the north of England, none of the coefficients for left voting are statis-
tically significant. The general lack of age effects on voting despite strong influences on attitudes
cries out for a political context interpretation-the obvious supposition is that it at least partly
reflects the absence of specifically age-based political parties.
The deviation for Scotland is easily explained by using the political context argument. In
this respect, the higher propensity for the old to vote right in this region is better suited to a
generational rather than a life-cycle explanation. Older Scots are typically more favourable to a
united Britain largely because their formative years took place during a time of common British
projects such as the British Empire and the World Wars (Heath and Kellas, 1998). Because of
this identification with the British state, older Scots are more likely to favour the Conservative
Party, the most unionist of the British parties. In contrast, young Scots have less affinity with the
British state and thus are more likely to avoid the Conservative Party, perhaps giving support
to the Scottish National Party instead.

4.2. Effects of race

Research suggests that visible racial minorities are more likely than others to be discriminated
against in the job market and elsewhere (Modood et al., 1997; Keith and Herring, 1991; Telles
and Murguia, 1990; Massey and Denton, 1993; Elliot and Fleras, 1992). It follows that left lean-
ing policies, such as laws ensuring equal opportunity and the redistribution of income, are in
the interest of racial minorities and we would therefore expect ethnic minorities to be generally
more left leaning, both in terms of attitudes and voting, than the majority population.
 Social Identitiesand PoliticalCleavages 317
.5 2.5

USSO
-b
0.0 2.0
0
USSO
a C EA USNC
b
G IE
ECN 1.5 U b
-.5

-1.0 GBSC
USNE
b GBSC

-1.5 us

b
USSO 0.0 CA
-2.0

-2.5
-2.5 -2.0 -1.5 -1.0 -.5 0.0 .5 -2.5 -2.0 -1.5 -1.0 -.5 0.0 .5

Left-Rightattitudes Left-Rightattitudes
Fig.2. Effectsof race(visibleminority-white
contrast)on left-rightattitudesandvoting(forthe US regions,
blacksareseparatedanddenotedbythe letterb at the end of the regioncode):e, USA,M,Canada;A,Great
Britain

Fig. 2 displays the relationships between racial minorities and whites in terms of left-right
attitudes and voting. Recall that, in all other regions outside the USA, race is coded as non-white
versuswhite, whereas in the US regions the non-whites are further divided into blacks and other
visible minorities. (The coefficients representingblacks in Fig. 2 are denoted by 'b' at the end of
the region code.) The effect of race on attitudes is most apparent for the black-white contrast
in the USA-in all regions blacks are significantly more left in attitudes (shown by the negative
coefficients on the horizontal axis). Other visible minorities in the USA are also typically more
left than whites, though to a lesser degree (the coefficients are statistically significant for the
South and West only). Excluding Wales, for which the relationship between race and attitudes
is not statistically significant, visible minorities in Britain also have significantly more left atti-
tudes than whites have. The Canadian regions are markedly different: in none of them is there
a significant race effect on left-right attitudes.
It is evident from Fig. 2 that racial differences in voting generally reflect attitudinal differences
between races. Reflecting its effect on attitudes, racial minorities are less likely to vote right and
more likely to vote left than are whites in most regions of the USA and Britain (witness the nega-
tive coefficients on the vertical axis). As with the left-right attitudes model, the effects on voting
in the USA are much larger for the black-white contrast than for the other minority-white
contrast. That US blacks stand out again is not surprising if we consider party platforms. The
Democratic Party is both more left and has campaigned for the rights of blacks to a far greater
degree than the Republican Party has. Once again, as with attitudes, the relationships between
race and vote are almost non-existent in Canada and do not reach statistical significance. The
Canadian exceptions perhaps indicate the success of Canada's official multicultural policy.

4.3. Effects of religiosity

Considering that religions usually uphold conservative values (see Schwartz and Huismans
(1995)), we should expect that those who are religious would generally hold socially conserva-
tive views. Although none of the regions that we examine have religious-specificparties, the right
parties in each region cater more to religious concerns than do any other parties. According to
the sociological model, we would expect the conservative views of the religious to be reflected
318 R. Andersenand A. Heath
in a higher propensity to vote for right parties than those who are not religious. If we consider
political context, we might also expect the voting patterns of practising Catholics and prac-
tising Protestants to differ across regions, especially given the dominance of Protestantism in
most of the regions. The following analysis recognizes this by distinguishing between practising
Catholics and practising Protestants, comparing each of them with the less religious.
We can see in Fig. 3 that, as expected, Protestantism is positively associated with conservative
attitudes in all regions (the coefficient is positive and statistically significant in all regions).
Protestants are also generally more likely to vote for right-wing parties, and less likely to vote
for left-wing parties. None-the-less, the US regions and Quebec are significant outliers. First,
Protestantism has a much stronger effect on the right vote in the US regions than elsewhere,
probably reflecting that the Republican Party has closer ties to the religious right than do the
right-wing parties in Canada and Britain. Secondly, contrary to all other regions, there is a
significant negative effect on both the left and the right vote in Quebec. This anomaly is also
easily explained by political context. The nature of party competition and the emphasis on na-
tionalism in Quebec politics limit party choices for Protestants. Both right parties and the New
Democratic Party are generally considered to represent the interests of the western provinces
rather than the interests of Quebec. This leaves the centralist Liberal Party as by far the most
viable party for Quebec Protestants to support.
Fig. 4 shows the Catholic-less religious contrasts to be very similar with respect to attitudes
but vastly different with respect to the vote across regions. Except in Scotland, Catholics hold
more Conservative attitudes on average than do the less religious. However, aside from a few
significant deviations, Catholicism generally has little influence on voting. The most notable
exception is Scotland, where Catholicism can be seen as a marker for Irish ethnicity, a histor-
ically disadvantaged minority that has opposed the dominant institutions of the British state
(see Clark (1998)). It is not surprising, then, that we find Catholics to be significantly less likely
to vote for the right-wing but unionist Conservative Party, and far more likely to vote for the
left Labour Party. Considering the large Irish minority in the south of England, particu-
larly in London, the effects of Catholicism in this region represent a paler version of the same
phenomenon.

1.0
USWE
CAWE
USNC 0
CAWE
USNE GBfs? c CA~O•N
.8,
-1
.6

4 GBE AON
0 GBWA8
A A
-3A
GBSE
.2GBNE WE
A -4

0.0
0-5 CAQU

CAQU
:
-.2 -6
-.2 0.0 .2 .4 .6 .8 1.0 .2 .4 .6 .8 1.0

attitudes
Liberal-Conservative Liberal-Conservative
attitudes
Fig. 3. Effectsof Protestantism Protestant-lessreligiouscontrast)on liberal-conservative
(practising atti-
tudes and voting;*, USA; i , Canada; A, Great Britain
 Social Identities and Political Cleavages 319

USNE
* .6
USNC GB~WA
USSO
.2 USWE .4GB BE
SGBE
o 0.0 --2 GBWA
G0 22GBAWA
.) CAWE
GBG BGE )S
-.
CAECAQU
-.4
USSO
USN
-.4 B CAN -.2 USNC
A BCAAWE
USNE

CAEA

.8 4 - -
-.6 -.4 -.2 -.0 .2 .6 .8 -.6 -.4 -.2 -.0 .2 .4 .6 .8
.4 _
Liberal-Conservative
attitudes Liberal-Conservative
attitudes
Fig.4. Effectsof Catholicism Catholics-lessreligiouscontrast)on liberal-conservative
(practising attitudes
andvoting;9, USA;F, Canada;A, GreatBritain

Two Canadian regions-Ontario and Quebec-also stand out because Catholics are signifi-
cantly less likely than the less religious to vote for right parties. In both cases it is the centre
Liberal Party that benefits, but for different reasons. The relatively low propensity for Quebec
Catholics to vote right reflects the strong historical ties between Quebec and the Liberal Party,
partly reflectedin the large number of Liberal Prime Ministers from Quebec. Ontario'sdeviation
can be explained by the large number of southern European immigrants, largely Catholic, who
show allegiance to the Liberal Party because they were in power when most of them arrived in
Canada.

4.4. Effectsof socialclass

Cross-national research suggests that social class generally affects attitudes, with working-class
individuals typically more left than those in other classes (see for example Kelley and Evans
(1995) and Vanneman (1980)). Although the relationship may be weakening, class is also signifi-
cantly related to voting in most Western societies (see Evans (1999)). Given their adversarial
relationship in the workplace, the largest differences in class attitudes and voting might be ex-
pected between the working-class and the managerial class, so we focus primarilyon the contrast
between these two classes.
Fig. 5 plots the contrasts between managers and the working-class for left-right attitudes
and voting. Although the relationships are generally in the expected direction, there are vast
differences between regions. Like previous research on class voting (see Weakliem and Heath
(1999) and Andersen and Heath (2002)), we find strong class effects in all British regions, with
managers being more right both in terms of attitudes and voting (correspondingly they are also
less likely to vote for the left). Similarly to previous research on the USA (Clark and Lipset,
1991) and Canada (Lambert and Curtis, 1993) the effects on attitudes and voting are generally
positive, but small and not statistically significant. (The coefficient for Quebec is negative but
not significantly different from 0.)
Once again, political context explains these cross-societal differences. The labour movement
and local working-class communities have traditionally been much stronger in Britain than in
Canada and the USA (Parkin, 1967). This greater class awareness has led to higher levels of
320 R.AndersenandA.Heath
1.0 .4

EA
'

.8
E.2
USWE
0.0 US
.6 CAWEUS

> A 3 -.2 *UE

USNC USWE
nF .2 CA
QQU -I -.4

(AEA USSGBC
0.0 -.6 G
B

.2 CAWE -.8 vB? GB

-.4 _ . ._-1.0 .
-.4 -.2 0.0 .2 .4 .6 .8 1.0 -.4 -.2 0.0 .2 .4 .6 .8 1.0

Left-Rightattitudes Left-Rightattitudes
Fig. 5. Effectsof socialclass (manager-working-class
contrast)on left-rightattitudesandvoting;9, USA;
•, Canada;A, GreatBritain

class voting, both from the working-class, who vote left, and from managers, who may have a
greaterpropensity to vote right as a reaction against the working-class (Przeworski and Soares,
1971). Despite these differences, the general pattern in Fig. 5 suggests that the relationships of
class with attitudes and voting tend to go in tandem.

4.5. Effects of nationalidentity

The small number of regions-Quebec, Wales and Scotland-for which minority national iden-
tities are relevant makes graphing the coefficients unnecessary. None-the-less, there are some
interesting findings that deserve to be reported. Most importantly, minority identities did not
have identical effects in each of the regions. For example, national differences in terms of left-
right attitudes were much more polarized in the British regions than in Quebec, with those
holding Welsh or Scottish attitudes being typically more left-wing. In none of these regions,
however, were these attitudinal differences reflectedin the left vote. Of course, this is because we
chose to exclude nationalist parties from the left block-other analyses including the nationalist
parties in the left block did show the expected relationship. None-the-less, context also plays
a role in Quebec in that those with Quebec national identity were more likely to vote for the
Conservative Party than others were (see Table 4).

5. Discussion and conclusions

This paper has explored the social bases of attitudes and of voting in a comparative perspective.
Our analysis was unusual in that it modelled attitudes and voting separately rather than includ-
ing attitudes as predictors of voting. We developed a novel methodology that allows an easy
comparison of regression coefficients from different types of models, specifically from linear
regression and probit models. In this way we could compare the effects of explanatory variables
on categorical dependent variables with the effects on continuous dependent variables.
Our method clearly demonstrates when the effects of social identities on voting are not con-
gruent with their effects on attitudes. We found significant regional differences in the effects of
social identities on voting and attitudes, indicating that countries should not always be treated
 SocialIdentitiesandPoliticalCleavages 321
as homogeneousregions;even in apparentlysimilarregionstherecan be differencesin group
attitudesand voting.Most importantly,this methodhighlightsthe importanceof considering
the role of partycompetitionwhenassessingsocial cleavages.
Wefoundremarkableuniformityin the socialbasisof attitudes,especiallywithrespectto age
and religiosity.In everyregion,olderrespondentsweremore likelythan youngerones to have
sociallyconservativeattitudesand this associationtendedto be quitepowerfuland invariably
significant.Therewaslittlevariationeitherwithinor betweencountries.In everyregionwe also
foundthat religiositywas associatedwith sociallyconservativeattitudes(evenaftercontrolling
for age and the othervariablesin the model).
Still, therewerelargedifferencesin votingpatternsaccordingto age and religion,especially
withrespectto rightvoting.Forexample,therewas no generalpatternfor the contrastbetween
Catholicsand the less religiousacrosscountries,perhapsreflectingthatnone of the regionshas
a partythatcatersspecificallyto Catholicconcerns.However,therewas a generaltendencyfor
Protestantsto be morelikelythanthe less religiousto vote for right-wingparties.For both the
Protestantand the Catholiccontrasts,Quebecstood out, suggestingthatthe structureof party
competitionand the politicaldominanceof nationalistissuesplayeda role.The strongestparty
in Quebec,the BlocQuebecois,is a separatistpartythatrepresentsthe interestsof secularQueb-
ecers,thus encouragingmarkedlydifferentreligiosity-votepatternsin Quebecthan elsewhere
(see Blais et al. (1995)and Nadeauand Fleury(1995)for a descriptionof the nationalvote in
Quebec).
Wealso foundevidenceof politicalcontextaffectingthe effectof age-relatedidentities.In no
countryis therea partyrepresentingspecificage groups,so, despitethat thereweresignificant
age effects on attitudes,these did not generallyinduce differencesin voting. Scotlandwas a
noteworthyexception,however,but age differenceshereare betterinterpretedas generational
effectsratherthan life courseeffects.OlderScottishvotersare typicallymore likelyto vote for
the ConservativeParty on the groundsthat it representsa united Britain.In other words, a
disparityin age effectson voting most likelyreflectsdifferencesin partyplatformsratherthan
differencesin the natureof the socialcleavage.
Age and religiosity,then, showa fairdegreeof uniformityin theirassociationwith attitudes
but high variabilityin theirassociationwith voting. In contrast,for raceand social class there
is considerablevariabilityin the associationwith left-rightattitudesandin the associationwith
voting, but the two tend to move in tandem.In Britainworking-classrespondentswere sub-
stantiallymore left in theirattitudesthan managers,and substantiallymore likelyto vote for
the LabourParty.In Canadaand the USA, in contrast,therewereonly weak class differences
in left-right attitudesand no significanteffects on left voting. We also found that race had a
significanteffect on attitudesand voting in all Britishand US regions.Therewas no notable
influenceon attitudesor votingin Canada.
Sincethe raceandclasseffectson attitudesand votingvariedin tandemacrossregions,these
findingscould in principlebe explainedby the sociologicalapproach.Still, this does not ex-
plain why the effects of race are so much strongerin the USA, especiallywith respectto the
votingpatternsof blacks.Similarly,it couldbe arguedthatclassvotingis weakin the USA and
Canadapreciselybecauseclass divisionsin left-right attitudesare weak. This still leavesthe
lack of attitudinalpolarizationunexplained,althoughthereare variouspotentialsociological
explanations,such as Sombart's(1976)classicexplanationfor the absenceof socialismin the
USA, whichcould apply.
In any event,the significantcross-regionaldifferencesin social identityeffects suggestthat
politicalexplanationsarenecessary.Consideringclass,Przeworskiand Sprague(1986)empha-
sized the way in which politicalpartiesact to promoteclass awareness.The British Labour
322 R.AndersenandA. Heath
Party, for example, went to great efforts to promote class awareness before its reincarnation
as New Labour. No parties behaved similarly in any of the other regions. Similarly, we can
see the strength of the race cleavage in the USA as reflecting that the Democrats take a clear
stand on racial issues, especially for blacks, whereas Canada and Britain do not have any major
parties that stand out on similar issues. The evidence cannot be conclusive, but the fact that the
associations of both attitudes and voting with social class and race tend to run in tandem is
suggestive that parties can influence social divisions.
This raises the further question why we have such different patterns for class and religiosity.
The answer to this may lie in the fact that none of the parties in our selection of regions is
primarily a religious party. We might expect to see stronger associations of both attitudes and
voting with religiosity in political contexts where there are primarily religious parties (as in the
Netherlands; see DeGraaf et al. (2001)). Following from this, we could predict that in Britain,
where New Labour has moved to become a catch-all party of the centre, the association of class
with left-right attitudes, as well as the association with voting, might decline in time. We leave
it to other researchersto address these issues in greater detail.

Acknowledgements
We wish to acknowledge our debt to the Economic and Social Research Council for the funding
of the Centre for Research into Elections and Social Trends, making this research possible. We
also thank the Data Archive at the University of Essex and the Institute for Social Research at
York University for providing the data. Finally, we express our thanks to David Firth, David
Cox and John Fox for their helpful suggestions.

AppendixA: Descriptionof the data sets

A. 1. Americandata:the 1996 and 2000 AmericanNationalElectionStudies
The 1996and2000ANESsemployednationallyrepresentative samplesof theAmericanelectorateduring
and followingthe 1996and 2000 US presidentialelections.We restrictour analysisto the post-election
samplesfromboth studies.In both studiesrespondentswererandomlyallocatedto face-to-faceand tele-
phoneinterviews.Preliminaryanalysesshowedno substantivedifferencesbetweenthe two samples,so the
analysesreportedhereuse all the respondents.
The 1996 ANES has a total samplesize of 1714 respondentsinterviewedbeforethe 1996presiden-
in a post-electionsurveyduringNovemberand December
tial election;1534of these werereinterviewed
1996.Of these 1534,1197wereinitiallyinterviewedin the 1994electionstudy;337 werefreshlysampled.
The responserates to the 1996 ANES were 71% for the pre-electionsurveyand 90% for the post-
electionreinterview.The 2000 ANES interviewed1803respondentsbeforethe 2000presidentialelection.
The post-electionreinterview,which took place immediatelyfollowingthe election in Novemberand
December2000, achieveda sample size of 1556. The overall responserate for the 2000 ANES was
61%;the post-electionreinterviewhad a responserateof 86%.
Both the 1996andthe 2000ANES areavailablefromthe Inter-university Consortiumfor Politicaland
Social ResearchData Archive,archivenumber6896for the 1996ANES (seeRosenstoneet al. (1997)for
moredetails)and 3131for the 2000ANES (see Burnset al. (2001)).

A.2. Canadian data: the 1993 and 1997 Canadian Election Studies
The 1993and 1997CESscontaindatafrompre-andpost-campaigninterviewssurroundingthe 1993and
1997 Canadianfederalelectionsrespectively.Both studiesconductedtelephoneinterviewsto a sample
selectedby usingrandom-digitdiallingand weredesignedto be representative of the Canadianpopula-
tion overthe age of 18 years.Sincemanyof the questionsof interestto this studywereaskedonly in the
post-electionsurveys,our resultswerebasedon this portionof the samples.
 and PoliticalCleavages
SocialIdentities 323
The total samplesize of the 1993CESwas 3775,of which3340werereinterviewedin the post-election
survey.Thetotal samplesizeof the 1997CESwas3949,of which3170werereinterviewed aftertheelection.
The responserateswere63%and 59%for the 1993and 1997CES campaigninterviewsrespectively. The
responseratesfor the post-electionreinterviewswere88%in 1993and 80%in 1997(formoredetails,see
Northrup(1998)for the 1997CES and Northrupand Oram(1994)for the 1993CES).
Data wereprovidedby the Institutefor SocialResearch,YorkUniversity.The 1993surveywas funded
by the SocialSciencesandHumanitiesResearchCouncilof Canada,grants411-92-0019and421-92-0026,
andwascompletedfor the 1992-1993Canadianelectionteamof RichardJohnston(Universityof British
Columbia),Andre Blais (Universit6de Montreal),HenryBrady(Universityof Californiaat Berkeley),
ElisabethGidengil(McGillUniversity)and Neil Nevitte (Universityof Calgary).The 1997surveywas
funded by the Social Sciencesand HumanitiesResearchCouncil of Canada, grant 412-96-0007,and
was completedfor the 1997Canadianelectionteam of AndreBlais (Universit6de Montreal),Elisabeth
Gidengil(McGill University),RichardNadeau (Universit6de Montreal)and Neil Nevitte (University
of Toronto).Neitherthe Institutefor SocialResearch,nor the SocialSciencesand HumanitiesResearch
Councilnor the Canadianelectionteamsare responsiblefor the analysesand interpretationsthat were
presentedhere.

A.3. Britishdata:the 1992 and 1997 BritishElectionStudies and the 1999 WelshAssembly
ElectionStudy
Both the 1992and the 1997BESwererepresentative samplesof all Britishadultsaged 18 yearsor over
livingin privatehouseholds.Respondentswereselectedby usingthe PostcodeAddressFileandinterviewed
face to face in theirhomesshortlyafterthe 1992and 1997generalelections.Both studiesalso included
a self-completionsupplementthat respondentswereto returnlater.Sincethe itemsthat are necessaryto
constructattitudescaleswereincludedonly in the self-completionquestionnaire,our analysisis basedon
thisportionof the samples.
The 1992BES has a completesamplesize of 3534,of which3304returnedthe self-completionsupple-
ment.The responseratefor the face-to-faceinterviewswas 73%;the responseratefor the self-completion
portionwas 68%.The total samplesizeof the 1997BESwas 3615,of which3093completedand returned
a self-completionsurvey.The responserateforthe in-homepersonalinterviewsof the 1997BESwas 62%;
the responseratefor the self-completionportionof the studywas 53%(seeThomsonet al. (1999)formore
details).
Sincethe BESinterviewedrelativelyfew respondentsfromWales,we supplementthe Britishdatawith
the 1999WAES,whichwas designedto yield a randomsampleof the populationof Walesaged 18 years
andover.A combinationof telephoneandface-to-faceinterviewswascarriedout in MayandJuneof 1999.
The sampleachievedwas 1255respondents;729wereinterviewedoverthe telephoneand 522 faceto face.
The responserateswereapproximately 67%for the face-to-facesurveyand 50%for the telephonesurvey.
Althoughspecificallyaboutthe 1999WelshAssemblyelections,the studycontainedrelevantquestions
aboutvotingduringthe 1997Britishelection.The WAESis similarin structureto the BES and all items
usedin this studyweremeasuredidenticallyin both studies.For moredetailsaboutthe WAESstudysee
Thomsonet al. (2001).
The 1992and 1997BES wereconductedunderthe auspicesof the Centrefor Researchinto Elections
and SocialTrends.The Directorsof the CentreareAnthonyHeath,RogerJowelland JohnCurtice.The
WAESwas conductedjointlyby the Institutefor WelshPolitics,Universityof Wales,Aberystwyth,and
the Centrefor Researchinto Electionsand SocialTrends.

AppendixB: Attitudescales
TheBritishattitudescaleswereconstructedusingquestionsthatweredesignedspecificallyforthe creation
attitudesandleft-rightattitudes(seeEvanset al. (1996)).
of additivescalesmeasuringliberal-conservative
Althoughthe CES andANES datawerenot designedwith the specificpurposeof creatingscalesthat are
identicalwith those fromthe BES, severalitemsin both studiescould be put togetherto createfunction-
allyequivalentscales.Sincewe combineddata fromtwo ANESs and two CESs,the scalewas necessarily
createdusingitemsthatwereusedin bothyearsforeachcountry.Weinitiallyexploreddimensionalityand
limitedthe numberof itemsby usingexploratoryfactoranalysis.After reducingthe numberof itemsto
measurethe two attitudes,we constructedscalesthat weresimilarto those fromthe BES.Details of the
scalesareprovidedbelow.
324 R.AndersenandA.Heath
B. 1. Britishscales
All the itemsusedin both BESscalesareLikertitemswith thefollowingresponseformat:'agreestrongly',
'agree','neitheragreenor disagree','disagree'or 'disagreestrongly'.Eachitemwascodedfrom1 to 5. The
specificitemsusedfor the left-rightscaleareas follows.
1. Ordinarypeopleget theirfairshareof the nation'swealth.
2. Thereis one lawfor the richand one for the poor.
3. Thereis no needfor strongtradeunionsto protectemployees'workingconditionsandwages.
4. Privateenterpriseis the bestwayto solveBritain'seconomicproblems.
5. Majorpublicservicesand industriesoughtto be in stateownership.
6. It is the government'sresponsibilityto providea job for everyonewho wantsone.
The itemsusedfor the Britishliberal-conservative scale areas follows.
1. Youngpeopletodaydon'thaveenoughrespectfortraditionalBritishattitudes.
2. Censorshipof filmsandmagazinesis necessaryto upholdmoralstandards.
3. Peopleshouldbe allowedto organizepublicmeetingsto protestagainstgovernment.
4. Homosexualrelationsarealwayswrong.
5. Peoplein Britainshouldbe moretolerantof thosewho leadunconventionallives.
6. Politicalpartieswhichwishto overthrowdemocracyshouldbe allowedto standin generalelections.
scale had a reliabilityof a = 0.56;for the Britishleft-rightscale the
The Britishliberal-conservative
reliabilitywas a = 0.67.

B.2. US scales
FiveitemswereusedfortheAmericanleft-rightscale.Thefirstthreequestionsusedfive-pointLikertscales
with the followingresponses:'agreestrongly','agree','neitheragreenor disagree','disagree'or 'disagree
strongly'(againcodedfrom 1 to 5). Items4 and 5 presentedrespondentswith two responsecategoriesbut
acceptedno opinionsand 'don'tknow'responsesif theyweredeclared,as follows.
1. If peopleweretreatedmoreequallyin this countrywe wouldhavemanyfewerproblems.
2. Oursocietyshoulddo whateveris necessaryto makesurethateveryonehas an equalopportunityto
succeed.
3. One of the big problemsin this countryis thatwe don'tgiveeveryonean equalchance.
4. Next, I am going to ask you to choose whichof two statementsI readcomes closerto your own
opinion.Youmightagreeto someextentwithboth,but we wantto knowwhichone is closerto your
own views.ONE, the less government,the better;or TWO,therearemorethingsthat government
shouldbe doing?
5. ONE, themainreasongovernmenthasbecomebiggerovertheyearsis becauseit hasgotteninvolved
in thingsthatpeople shoulddo for themselves;OR TWO,governmenthas becomebiggerbecause
the problemswe face havebecomebigger.
Each of the items used for the Americanliberal-conservative scale are five-pointLikertitemswith re-
sponses rangingfrom stronglyagreeto stronglydisagree(againcoded from 1 to 5). The items are as
follows.
1. The newerlifestylesarecontributingto the breakdownof our society.
2. This countrywould havemanyfewerproblemsif thereweremoreemphasison traditionalfamily
ties.
3. The worldis alwayschangingand we shouldadjustour viewof moralbehaviourto those changes.
The Americanleft-rightscaleachieveda Cronbach'sa of 0.72. Theliberal-conservative scale'sreliability
was reasonablegiventhatonly threeindicatorswereusedin its construction(a = 0.56).

B.3. Canadian scales

As can be seenfromthe wordingof the questions,the responseformatfor each of the questionsusedto
constructthe Canadianleft-rightscalewas the following:'a lot' (coded3), 'some'(coded2) or 'not at all'
(coded 1). The specificitemsof the scaleareas follows.
 Social Identitiesand PoliticalCleavages 325
1. Would you cut WELFARE spending A LOT, SOME, or NOT AT ALL?
2. Would you cut PENSIONS A LOT, SOME, or NOT AT ALL?
3. Would you cut HEALTH CARE A LOT, SOME, or NOT AT ALL?
4. Would you cut UNEMPLOYMENT INSURANCE A LOT, SOME, or NOT AT ALL?
5. Would you cut EDUCATION A LOT, SOME, or NOT AT ALL?
The questions used to construct the Canadian liberal-conservative scale had varying response formats.
The first two questions used five-point Likert scales with the following responses: 'agree strongly', 'agree',
'neither agree nor disagree', 'disagree'or 'disagree strongly' (coded from 1 to 5). For questions 3-5 respon-
dents were asked to give a score on a scale of 0-100, where 0 means they 'really dislike' the group and 100
means they 'reallylike' the group. To make these comparable with the Likert questions, responses to these
questions were collapsed into five categories and assigned scores from 1 to 5. This had only a minimal
effect on the response distributions since the data were essentially in five clusters. The last item, which has
three categories, was scored I for the most liberal response, 5 for the most conservative response and 3 for
the middle response. The items used for the scale are as follows.
1. Only people who are legally married should be having children.
2. Society would be better off if more women stayed home with their children.
3. How do you feel about Aboriginal peoples?
4. How do you feel about homosexuals?
5. How do you feel about racial minorities.
6. Of the following three positions, which is closest to your own opinion: One: abortion should NEVER
be permitted, Two: should be permitted only after NEED has been established by a doctor, OR Three:
should be a matter of the woman's PERSONAL CHOICE?
Cronbach's a for the left-right scale was 0.59, indicating that the scale was reasonably reliable especially
considering that there were only three response categories for all the items. The liberal-conservative scale
had good reliability (a = 0.66).

References
Alford, R. (1967) Classvoting in the Anglo-Americanpolitical systems.In PartySystemsand VoterAlignments.
CrossNationalPerspectives(eds S. M. Lipsetand S. Rokkan),pp. 67-94. New York:Free Press.
Alwin, D. E and Krosnick,J. A. (1991) Aging, cohorts, and the stability of sociopoliticalorientationsover the
life span. Am.J Sociol., 97, 169-195.
Amorim Neto, 0. and Cox, G. W. (1997) Electoralinstitutions,cleavagestructures,and the numberof parties.
Am.J Polit. Sci., 41, 149-174.
Andersen,R. and Heath, A. (2002) Class matters:the persistingeffects of contextualsocial class on individual
voting behaviourin Britain, 1964-97.Eur.Sociol. Rev.,18, 125-138.
Baer,D., Curtis,J., Grabb,E. and Johnston,W.(1995) Respectfor authorityin Canada,the United States, Great
Britainand Australia.Sociol. Focus,28, 177-195.
Bartolini, S. and Mair, P. (1990) Identity,Competitionand ElectoralAvailability:the Stabilisationof European
Electorates,1885-1985.Cambridge:CambridgeUniversityPress.
Berelson,B., Lazarsfeld,P. and McPhee,W.(1954) Voting:a Studyof OpinionFormationin a PresidentialCam-
paign. Chicago:Universityof ChicagoPress.
Blais, A., Martin,P and Nadeau, R. (1995) Attentes economiqueset linguistiqueset appuia la souverain6t6du
Quebec:une analyseprospectiveet comparative.Can.J Polit. Sci., 28, 637-657.
Brint, S. (1984) 'New-class'and cumulativetrendexplanationsof the Liberalpolitical attitudesof professionals.
Am. J Sociol., 90, 30-71.
Bums, N., Kinder,D. R., Rosenstone,S. J., Sapiro,V.and the National ElectionStudies(2001)NationalElection
Studies2000: Pre-/PostElectionStudy.Ann Arbor:Universityof Michigan.
Butler,D. and Stokes, D. E. (1974) PoliticalChangein Britain:the Evolutionof ElectoralChoice.London: Mac-
millan.
Clark,S. (1998)Internationalcompetitionand the treatmentof minorities:seventeenth-century cases and general
propositions.Am. J Sociol., 103, 1267-1308.
Clark,T. N. and Lipset,S. M. (1991)Are social classes dying?Int. Sociol., 6, 397-410.
DeGraaf, N. D., Heath,A. and Need, A. (2001) Decliningcleavagesand politicalchoices:the interplayof social
and politicalfactors in the Netherlands.Elect. Stud., 20, 1-16.
Duverger,M. (1954) PoliticalParties.New York:Wiley.
Elliot, J. L. and Fleras, A. (1992) UnequalRelations:an Introductionto Race and EthnicDynamicsin Canada.
Scarborough:PrenticeHall.
326 R. Andersenand A. Heath
Evans, G. (ed.) (1999) TheEndof ClassPolitics?Oxford:OxfordUniversityPress.
Evans, G., Heath, A. F. and Lalljee,M. G. (1996) Measuringleft-rightand libertarian-conservative attitudesin
the Britishelectorate.Br.J Socio., 47, 93-112.
Feldman,S. (1988)Structureand consistencyin publicopinion:the roleof core beliefsand attitudes.Am.J Polit.
Sci., 32, 416-440.
Fleishman, J. A. (1986) Types of political attitude structure:results of a cluster analysis.Publ. Opin. Q., 50,
371-386.
Fox, J. (1997) AppliedRegressionAnalysis,LinearModels,andRelatedMethods.London:Sage.
Fox, J. (2000)NonparametricSimpleRegression.ThousandOaks:Sage.
Glenn, N. D. (1980)Attitudes,attitudes,and beliefs.In Constancyand Changein HumanDevelopment(eds O. G.
Brimand J. Kegan),pp. 596-640. Cambridge:HarvardUniversityPress.
Heath, A. and Kellas,J. (1998)Nationalismsand constitutionalquestions.Scot. Aff, specialissue, 110-128.
Heath, A., Taylor,B., Brook, L. and Park,A. (1999) Britishnational sentiment.Br.J Polit. Sci., 29, 155-175.
Hout, M., Manza, J.and Brooks, C. (1999) Classes,unions, and the realignmentof US Presidentialvoting 1952-
1992. In The End of Class Politics: Class Votingin ComparativeContext(ed. G. Evans), pp. 83-96. Oxford:
OxfordUniversityPress.
Inglehart,R. (1987)Valuechange in industrialsocieties.Am. Polit. Sci. Rev.,81, 1289-1303.
Inglehart,R. (1990) CultureShift in AdvancedIndustrialSociety. Princeton:PrincetonUniversityPress.
Inglehart,R. (1997) Modernizationand Postmodernization: Cultural,Economic,andPolitical Changein 43 Soci-
eties. Princeton:PrincetonUniversityPress.
Keith, V.and Herring,C. (1991) Skin tone and stratificationin the blackcommunity.Am. Sociol., 97, 760-778.
Kelley, J. and Evans, M. D. R. (1995) Class and class conflict in six Westerndemocracies.J. Am. Sociol., 83,
386-402. J.
Kitschelt,H. and Hellemans,S. (1990) Beyondthe EuropeanLeft. London:Duke UniversityPress.
Lambert,R. D. and Curtis,J. E. (1993) Perceivedparty choice and class voting. Can.J Polit. Sci., 26, 273-285.
Lazarsfeld,P., Berelson,B. and Gaudet, H. (1944) ThePeople'sChoice.New York:ColumbiaUniversityPress.
Lijphart,A. (1979) Religionvs. linguisticvs. class voting. Am. Polit. Sci. Rev.,65, 686.
Lijphart,A. (1990)The politicalconsequencesof electorallaws, 1945-85. Am.Polit. Sci. Rev.,84, 481-496.
Lijphart,A. (1994)ElectoralSystemsandPartySystems:a Studyof Twenty-seven Democracies,1945-1990.Oxford:
OxfordUniversityPress.
Lipset, S. M. and Rokkan,S. (1967) Cleavagestructures,partysystemsand voteralignments:an introduction.In
Party Systemsand VoterAlignments:Cross-nationalPerspectives(eds S. M. Lipset and S. Rokkan),pp. 1-64.
New York:FreePress.
Manza,J.and Brooks,C. (1999)Social CleavagesandPoliticalChange:VoterAlignmentsand US PartyCoalitions.
New York:OxfordUniversityPress.
Markus,G. B. (1979)The politicalenvironmentand the dynamicsof public attitudes:a panel study.Am. Polit.
Sci., 23, 338-359. J.
Massey, D. S. and Denton, N. (1993) AmericanApartheid:Segregationand the Makingof the Underclass.Cam-
bridge:HarvardUniversityPress.
Modood, T., Berthoud,R., Lakey,J., Nazroo, J., Smith,P.,Virdee,S. and Beishon,S. (1997)EthnicMinoritiesin
Britain:DiversityandDisadvantage.London:Policy StudiesInstitute.
'
Nadeau, R. and Fleury, C. (1995) Gains linguistiquesanticipeset appui la souverainet6du Quebec. Can. J
Polit. Sci., 28, 35-50.
Northrup, D. (1998) The1997 CanadianElectionSurvey.TechnicalDocumentation.Toronto:YorkUniversity.
Northrup, D. and Oram, A. E. (1994) The 1993 CanadianElectionStudy.Incorporatingthe 1992 Referendum
Surveyon the Charlottetown Accord:TechnicalDocumentation.Toronto:YorkUniversity.
Parkin,F. (1967) WorkingClass Conservatives:a theory of politicaldeviance.Br. Sociol., 18, 278-290.
Przeworski,A. and Soares,G. A. D. (1971) Theoriesin searchof a curve:a contextualinterpretationof the left
vote. Am. Polit. Sci. Rev.,65, 51-68.
Przeworski,A. and Sprague,J. (1986) Paper Stones: a History of ElectoralSocialism.Chicago: Universityof
Chicago Press.
Riker, W. H. (1982) The two-partysystem and Duverger'slaw: an essay on the historyof political science.Am.
Polit. Sci. Rev.,76, 753-766.
Rose, R. and Urwin, D. (1970) Persistenceand change in Westernparty systems since 1945. Polit. Stud., 18,
287-319.
Rosenstone,S. J.,Kinder,D. R., Miller,W.E. and the National ElectionStudies (1997)NationalElectionStudies
1996: Pre-/PostElectionStudy.Ann Arbor:Universityof Michigan.
Satori,G. (1976) Partiesand PartySystems:a Framework for Analysis.New York:CambridgeUniversityPress.
Schwartz,S. and Huismans,S. (1995) Valueprioritiesand religiosityin four westernreligions.Socl Psychol. Q.,
58, 88-107.
Sombart,W (1976) Whyis thereNo Socialismin the UnitedStates?New York:Sharpe.
Telles,E. E. and Murguia,E. (1990) Phenotypicdiscriminationand income differencesamong MexicanAmeri-
cans. Socl Sci. Q, 71, 682-696.
 SocialIdentitiesandPoliticalCleavages 327
Thomson, K., Nicolaas, G., Bromley,C. and Park, A. (2001) WelshAssemblyElection Study 1999: Technical
Report.London:National Centrefor Social Research.
Thomson, K., Park, A. and Brook, L. (1999) British GeneralElection Study 1997. TechnicalReport.London:
National Centrefor Social Research.
Vanneman,R. (1980) U.S. and Britishperceptionsof class.Am. J Sociol., 85, 769-790.
Weakliem,D. L. (1991) The two lefts?:occupation and party choice in France,Italy, and the Netherlands.Am.
J Sociol., 96, 1327-1361.
Weakliem,D. L. (1993) Class consciousnessand political change:voting and political attitudesin the British
workingclass 1964 to 1970.Am. Sociol. Rev.,58, 382-397.
Weakliem,D. L. and Heath, A. F. (1999) The secretlife of class voting: Britain, France,and the United States
since the 1930s.In TheEndof ClassPolitics?(ed. G. Evans),pp. 97-136. Oxford:OxfordUniversityPress.