CREST Social Cleavages, Attitudes and Voting Patterns: A Comparison of Canada and

CREST
CENTRE FOR RESEARCH INTO
ELECTIONS AND SOCIAL TRENDS
Working Paper
Number 81
September 2000
Social Cleavages, Attitudes and Voting
Patterns: A Comparison of Canada and
Great Britain
By Robert Andersen and Anthony Heath
The Centre for Research into Elections and Social Trends is an ESRC Research Centre based jointly at
the National Centre for Social Research (formerly SCPR) and the Department of Sociology, University
of Oxford
http://www.crest.ox.ac.uk
ABSTRACT
This paper develops a novel method for comparative research on social cleavages that
integrates the three major approaches to voting behaviour—the sociological approach,
rational choice theory and the party identification model—under a single theoretical
paradigm. We apply this integrated theory to the major regions of the USA, Canada and
Great Britain. We find striking national and regional similarities in the effects of social
group memberships on attitudes, but considerable diversity in the effects of social group
membership on vote. Race is a notable exception—it deviates in the absence of
uniformity of its effects of the cleavage on both attitudes and vote. That we find
significant regional differences within countries suggests the importance of using region,
rather than country, as the unit of analysis. These findings also underline the importance
of paying close attention to political context when assessing the effects of social groups
on voting.
INTRODUCTION
Most striking about the voting behaviour literature is that it groups neatly according to
three paradigms—the sociological approach, party identification models and rational
choice theory. Although there have been some attempts to integrate party identification
within a rational choice framework (see, for example, Fiorina, 1981; Achen, 1992;
Gerber and Green, 1998), for the most part researchers have tried to explain voting
behaviour using only one of the approaches and have decidedly ignored the others. In
fact, these approaches are almost universally treated as incompatible. We suggest that all
of the approaches have merit and limitations, and that they should be seen as
complementary rather than as opposing. We argue that each approach is applicable under
different conditions of political context (e.g., different party competition structures and
strengths of social cleavages). The best way to understand general patterns of voting is to
integrate these approaches, and apply them in comparative analysis, paying careful
attention to varying political contexts.
Aside from these theoretical considerations, research on voting behaviour
typically falls ill to other problems. First, if there are regional differences within a
country, the common practice of using country as the unit of analysis in comparative
research can average out the effects of social cleavages on attitudes and vote, masking
their true impact. Secondly, single studies in political sociology often examine the impact
of only one cleavage despite that including other groups is important if there are political
parties representing their interests. Finally, the common practice of specifying social
group variables and attitudinal measures as explanatory variables in a single model of
voting behaviour ignores that attitudes are largely influenced by social group membership
(see Bartle 1998, for a good discussion of this problem), and thus weakens the grouprelated coefficients since they account only for direct effects.
The primary aim of this paper is to integrate the major approaches to voting
behaviour in a comparative context. To accomplish this goal we develop a novel
approach and method for comparing attitudes and voting in different societies. We make
a number of distinctive moves. First, we focus on regional differences within countries,
as well as on the usual national differences. Secondly, we break from convention by
examining the relative impact of many social cleavages rather than exploring the effects
of a single cleavage. Thirdly, rather than include attitudes as mediating variables within a
single model, we use separate models to explicitly compare the effects of group identities
on attitudes and vote. Finally, we develop a graphical technique for comparing the pattern
and magnitude of relative social group effects on both attitudes and voting in different
societies. Our method is an advancement over previous research because it allows
sensible comparison of the relationship between social group variables and attitudes and
vote across societies, even if datasets with non-identical measures are used.
THEORETICAL BACKGROUND
Sociological Approach to Voting
The sociological approach to voting behaviour emphasizes the impact of social structure,
suggesting that social group memberships influence voting choices (see for example,
1
Lazarsfeld et al.,1944; Alford 1967, Rose and Urwin 1969, 1970, Lijphart 1979, 1980).
Voters are considered to be instrumental—i.e., they vote for parties that best reflect the
interests of their groups. The origins of this approach can be traced to the Columbia
school which carried out the first systematic surveys of the American electorate (see
Lazarsfeld et al.,1944). These early studies stated the conditions for persistent group
voting as the following:
In sum, the conditions underlying persistent voting cleavages seem to be (1)
initial social differentiation such that the consequences of political policy are
materially or symbolically different for different groups; (2) conditions of
transmitability from generation to generation; and (3) conditions of physical and
social proximity for continued in-group contact in succeeding generations
(Berelson, Lazarsfeld and McPhee 1954:75).
Another influential work fitting under the sociological paradigm is that of Lipset and
Rokkan (1967) who argued that not only do group identities influence voting behaviour,
but that cleavage structures determine the number of political parties in a given polity. In
other words, political parties evolve in response to the interests of social cleavages.
The sociological approach, then, holds that group identities affect attitudes and
interests. These attitudes in turn affect how people vote. By implication in any given
society the effects of group membership should be the same on attitudes as they are on
vote. It is difficult to deny the existence of social cleavages and their potential effects on
attitudes and voting. However, this approach is unable to explain why cleavages such as
social class have stronger relationships to attitudes and vote in some countries than others
(see an edited volume by Evans, 1999).
Party Identification Model
Unlike the sociological model, the party identification model assumes voters to be
expressive rather than instrumental, and attitudes and issue preferences are considered to
be endogenous to vote. This approach holds that voters have long-standing psychological
ties to specific political parties, and seldom waver from voting for them (Belknap and
Campbell 1952, Campbell et al., 1960, Converse 1964). These party attachments are
largely due to early socialization, reflecting mostly family influences. Simply put, people
are influenced by the partisanship of their parents.
The party identification model can be seen as similar to the sociological approach
in that partisans “come to see themselves as members of social groups (e.g., Democrats,
Republicans), in much the same way that certain people incorporate religious, regional,
or ethnic groups into their self-conceptions” (Gerber and Green, 1998:794). On the other
hand, unlike the sociological model, it holds that causation runs in both directions
between attitudes and vote. As Campbell et al. (1960:128) state, “In the competition of
voices reaching the individual the political party is an opinion-forming agency of great
importance. The strength of relationship between party identification and the dimensions
of partisan attitude suggests that responses to each element of national politics are deeply
affected by the individual’s enduring party attachments.” This implies that the
relationship between group membership and attitudes should be similar to that between
group membership and vote. We could expect, however, that different political contexts
2
would induce different relationships between groups and attitudes than expected by the
sociological approach.
The party identification approach has been quite successful when applied to the
two-party system of the US for which it was developed and where recent research has
reaffirmed that partisanship is highly stable over time for the adult population (Green and
Palmquist, 1990, 1994). It has been less successful elsewhere, in particular in Britain,
where voters are less likely to make distinctions between their vote choice and partisan
dispositions (Butler and Stokes, 1974).
Rational Choice
Although instrumental like the sociological model, the rational choice approach is much
more individualistic, suggesting that voting decisions are based on cost-benefit analyses
where voters match their individual issue preferences with party platforms. As Olson
(1965:51) states, “only a separate and ‘selective’ incentive will stimulate a rational
individual in a latent group to act in a group-oriented way.” According to Downs
(1958:39), if the voter “is rational, he knows that no party will be able to do everything
that is says it will do. Hence he cannot merely compare platforms; instead he must
estimate in his own mind what the parties would actually do were they in power.”
According to the rational choice approach, then, policy preferences are exogenous, but
vote choices depend on the interplay between voters’ preferences and parties’ policy
positions.
Despite its individualist emphasis, the rational choice approach is not necessarily
incompatible with either the sociological and party identification approaches. Although
not explicit, rational choice theory allows for the possibility that social identities play a
role in voting decisions since individual preferences can be determined by one’s position
in society. Moreover, not all rational choice theorists discount party attachments. Rather
than see them as influencing attitudes, however, these attachments are considered to
represent ongoing tallies of assessments of party performances (see Fiorina, 1981).
Although social groups may affect attitudes, this does not mean that voting decisions are
made solely on the basis of these group-determined attitudes. Accordingly, the rational
choice model implies that the relationship between preferences and vote will vary across
different political contexts: if voters are given different political options from which to
choose, then the relationship between attitudes and vote may also vary.
Towards an Integrated Theory
The three major approaches to voting behaviour can be integrated in a single theoretical
framework. We start with the basic premise that voters are rational actors. We then adopt
the sociological approach by arguing that attitudes are influenced by the groups to which
one belongs. As we suggest earlier, this is not incompatible with the notion that voters are
rational actors since their individual preferences at least partly reflect their position in
social structure. We are not suggesting that social groups are homogenous in terms of
attitudes and interests, only that social identities influence how people see the world (see
Jenkins, 1996; Reid and Deaux, 1996).
3
If voters are rational actors we should expect that their electoral choices will
reflect the concerns of the groups to which they belong. If these groups have a strong
political presence—i.e., a viable political party represents their concerns—then voters
will be inclined to vote for them. On the other hand, if the social group is not clearly
represented by a particular party, voters will have no satisfying vote choice. Voters are
then forced to choose a party that does not represent the interests of at least one of their
group identities. Here the party identification model becomes relevant. Over time these
voters, and their offspring, may identify with the party and its platform, taking on
attitudes that seemingly do not reflect their group’s position in social structure. In such
cases, vote and attitudes would still be highly related—e.g., right-wing voting will be
associated with right-wing attitudes—since attitudes are affected by the platform of the
party voters support.
Let us now consider the situation where a particular party does represent the
interests of a social group identity, but it also represents the interests of another opposing
social identity. By way of an example, we briefly discuss a conflicting religious identity
and social class identity. Assume a polity with a political party that is both secular and
pro-working class. Suppose that no other parties represent the working class, but that
other parties represent religious interests. In this situation, highly religious working class
people may decide against voting for the working class party since it goes against their
religious concerns. Likewise, those who do not belong to a religion, and hence are
opposed to policies reflecting religious interests, may vote for the party regardless of their
social class. In such cases, the attitudes of the working class may still be generally more
left than those of other classes, but they may not necessarily be more likely to vote for the
left-wing party. Simply put, vote is most likely to deviate from attitudes when a social
identity is represented by a party, but that party also represents a conflicting identity.
In summary, we are suggesting that voters are rational actors who try to maximize
their utility. This utility is largely based on their position in social structure. Position in
social structure is determined by membership in many, often competing, social groups
(for a discussion of multiple social identities, see Stagnor, Duan and Glass, 1992;
Deschamps and Doise, 1978). If there are no parties representing the interests of a
particular social group, voters are forced to choose from other parties that represent other
social group interests. If there is only one acceptable party though not a perfectly
matching party, voters may adopt attitudes and policy preferences that reflect that party’s
platform in the long term. If, on the other hand, the social group’s interests are reflected
in a party platform, but a conflicting social group interest is also met, voters may be
inclined not to vote for that party. In this case, attitudes will still reflect social group
interests, despite that vote apparently does not. The present paper applies this integrated
theory to voting behaviour in the major regions of Canada, Great Britain and the USA.
Hypotheses
Although our analysis controls for the impact of gender and education, we do not discuss
these variables in detail because of space constraints. Our focus is on the relative impact
of age, race, religiosity and social class. We evaluate four hypotheses based on our
integrated theory:
4
1) In all regions, compared with the young (those under 30 years old), the elderly
(those over 65 years old) are more authoritarian and more likely to vote for rightwing parties.
A significant body of research shows political attitudes to be more resistance to change as
people age (Alwin and Kosnick, 1991; Glenn, 1980; Jennings and Markus, 1984; Markus,
1979), with people be coming more socially conservative (see, for example, Park, 2000).
In none of the regions we examined is there a party that uniquely represents the old, but
the party structures are all similar in that there are distinct right-wing (socially
conservative) parties with similar authoritarian platforms. We expect, then, that these
authoritarian attitudes will be reflected in a greater propensity for seniors to vote for
right-wing parties.
2) In all regions, compared to whites, visible racial minorities are more left in terms
of attitudes and vote.
The basis for this hypothesis is that ethnic minorities are more likely to be discriminated
against in the job market and elsewhere than others. It follows that left-leaning policies,
such as laws ensuring equal opportunity and income redistribution, are in the interest of
racial minorities. Since it is rational to want to improve one’s position, one would expect
racial minorities to be generally more left-leaning in terms of attitudes and vote than the
majority population.
3) Those who regularly attend religious services are more authoritarian and more
likely to vot e for right-wing parties than those who seldom or never attend
religious services.
Considering that religions are usually socially conservative to at least some degree (e.g.,
they are usually anti-abortion, anti-gay rights, pro-censorship etc.), it is sensible to expect
that only those who hold similar attitudes would regularly attend religious services.
Although none of the regions under study have parties specifically representing religious
interests, in every region the right parties are more likely to cater to the religious than any
other parties.
4) Unskilled labourers are more left in terms of attitudes than managers. If there are
viable left parties, we would expect these attitudes to be reflected in left -voting.
Because of the lack of a strong labour party in the US, we expect US regions to
deviate from this pattern, perhaps exhibiting a pattern akin to the party
identification model.
It is clear that the material interests of the working class and managers are quite different
(e.g., the working class is certainly more likely to depend on social programs). A great
deal of recent research has shown that, while the relationship may be weakening, class is
still significantly related to voting (see an edited volume by Evans, 1999; Lambert and
Curtis, 1993).
5
DATA AND METHODS
Data
We use data from the 1997 Canadian Election Study1 (CES), the 1997 British Election
Study (BES), the 1999 Welsh Assembly Election Study (WAES) and the 1996 American
National Election Study (ANES).
The CES is a representative rolling cross-sectional sample of all Canadians over
the age of 18 during the 36-day official election campaign. Respondents were selected
using random-digit-dialing and interviewed over the telephone. The total sample size is
3949, of which 3170 respondents were re-interviewed in a post-election survey2. Since
most of the questions of interest to this study were asked only in the post-election survey,
our results are based on this portion of the sample. We divide the CES data according to
five regions (sample sizes after removing missing cases are in parentheses): West (890),
Ontario (596), Quebec (681); East (290) and North (117).
The BES is a representative sample of all British adults aged 18 or over living in
private households. Respondents were selected using the Postcode Address File and
interviewed face-to-face in their homes shortly after the May 1997 election. The total
sample size is 3615, of which 3093 completed and returned a self-completion survey3.
Since the items necessary to construct attitude scales were included only in the selfcompletion questionnaire, our analysis is based on this portion of the sample. We
supplement the BES data with the WAES because the BES interviewed only 182 people
in Wales. The WAES is similar in structure to the BES and all items used in this study
were measured identically in both studies. The WAES has a sample size of 686, thus
giving substantially more power to our statistical models of Wales. Respondents were
interviewed in-home in May and June of 1999. The British data are divided into five
regions (sample sizes omitting missing cases are in parentheses): Scotland (752), Wales
(1040), North England (611); Midlands England (595) and South England (976).
The ANES is a nationally representative sample of the American electorate. The
ANES has a total sample size of 1714 respondents who were interviewed before the 1996
Presidential election; 1534 of these were re-interviewed in a post-election survey during
November and December 1996.4 We restrict our analysis to the post-election survey.
Respondents to the post-election survey were randomly allocated to face-to-face and
1
Data from the 1997 Canadian Election Survey were provided by the Institute for Social Research, York
University. The survey was funded by the Social Sciences and Humanities Research Council of Canada
(SSHRC), grant number 412-96-0007 and was completed for the 1997 Canadian Election Team of Andre
Blais (Université de Montreal), Elisabeth Gidengil (McGill University), Richard Nadeau (Université de
Montreal) and Neil Nevitte (University of Toronto). Neither the Institute for Social Research, the SSHRC,
nor the Canadian Election Survey Team are responsible for the analyses and interpretations presented here.
2
The response rate for the campaign sample of the CES was 59%; The response rate for the post-election
re-interview was 80% (Northrup 1998).
3
The total response rate for the in-home personal interviews of the BES was 62%; the response rate for the
self-completion portion of the study was 53% (Thomson, Park and Brook, 1999).
4
Of these 1534 respondents, 1197 were initially interviewed in the 1994 election study; 337 were freshly
sampled. The response rate to the pre-election survey was 71%. The response rate to the post-election reinterview was 90%.
6
telephone interviews. Limitations in sample size dictated that we used all respondents
regardless of whether they were administered a personal or telephone interview. The
ANES data are divided into the US Census’s five major regions (samples sizes omitting
missing cases are in parentheses): Northeast (233), North Central (415), South (572) and
West (312).
Response Variables
There are two dependent variables for our attitudes models—left-right attitudes and
libertarian-authoritarian attitudes—both of which are operationalised as Likert-scales
(Details of the scales are in Appendix A).5 The left-right scale measures positions on
government intervention in the economy, specifically where one stands between
socialism and laissez faire. Those who score low on the left-right scale favour
government intervention, social spending and greater wealth redistribution; those who
score high favour a freer economy. Libertarian-authoritarian attitudes are related to
personal freedom of thought, association and lifestyles. Those who score high on the
libertarian-authoritarian scale are more authoritarian in their views, while those who
score low are more tolerant of alternative lifestyles and attitudes.
The vote models have two dichotomous dependent variables: left vote versus
other, and right vote versus other. In all Canadian regions the right parties consist of the
Progressive Conservative Party and the Reform Party. The New Democratic Party
(NDP) is the only party coded as left in all Canadian regions except Quebec, where the
left-nationalist Bloc Quebecois is also included in the analysis of left vote.6 Although it
received more votes than any other party in 1997, the Liberal Party is included as part of
the “other” categories for our analysis because it can be considered a catchall party that
occupies the centre of Canadian politics.
In Great Britain, the left is occupied by the Labour Party and the right party is the
Conservative Party. The national parties in Scotland and Wales, the Scottish National
Party and Plaid Cymru respectively, were included in the other vote categories. In 1997
the Liberal Democrats were still considered the centralist party and thus are included in
the “other” party category.
Although there is arguably no left party in the US, we followed convention by
treating the Democrats as the left party and the republicans as the right party (see Hout,
Manza and Brooks, 1999). This practice is justified on the grounds that the Democrats are
considerably more left and libertarian in terms of platform than the Republicans.
Explanatory Variables
In all regions gender is treated as a dummy variable with men coded as the reference
category. Age is divided into four categories: less than 30 years old, 30-45, 46-64 and
over 65 which is the reference category. With the exception of Scotland, for all Canadian
and British regions race is operationalised simply a visible ethnic minority or white (the
5
Research on Britain (Heath, Evans and Martin, 1993) and the USA (Fleishman, 1988) have found these
two attitudes are the major attitudinal dimensions.
6
Only 9 respondents in Quebec stated that they voted for the NDP, meaning that the non-nationalist left
could not be analysed separately from the Bloc Quebecois.
7
reference category). The small number of nonwhites in Scotland prohibited including this
variable for models of that region. For all US regions, race is divided into three
categories: black, Hispanic and white (reference category). Religiosity is divided into two
categories: those who attend church at least once a month (coded 1), and others (coded
0).7
Education is treated as a categorical variable with those without qualifications (or
unrecognised foreign qualifications) being the reference category. In Canada and the US,
the dummy variables for education are: (1) high school, (2) some post-secondary, and (3)
university degree. In Great Britain the dummy variables represent (1) CSE or O-level, (2)
A-level, (3) some post-secondary, and (4) university degree.
Following from Hout, Manza and Brooks (1999) social class is divided into the
same five categories for all regions: managers, professionals, clerical workers, skilled
manual labourers and, semi- and unskilled manual labourers. Due to a significant portion
of missing cases in most regions, a six category representing all those whose occupations
are unknown is included in the models.
STATISTICAL MODELS
Attitudes are regressed on social group variables using ordinary least squares (OLS)
regression. Probit models are used to regress vote on the same social group variables. We
adjust the coefficients from the OLS models of attitudes in order to standardized the
regression equations, making them comparable with the corresponding coefficients from
the probit models of vote. We then explore the relationship between social groups,
attitudes and votes across societies using scatterplots, plotting the coefficients from the
attitudes model on the horizontal axis and the coefficients from the vote model on the
vertical axis. We provide greater details of the method below.
We start by discussing the probit models of left and right vote.8 The general probit
model assumes that a binary dependent variable, y (1,0), represents a continuous latent
variable, Z. For each combination of predictors there is a different mean ì , which
represents the cut-off point where Z > 0 (the point where the observed discrete response
variable changes from 0 to 1). In our case, Z represents the “propensity” to vote for left
(or right) and y represents left vote (or right vote). The probit model assumes that the
errors are independent and normally distributed and that the residual variance of the
model equals one (Fox, 1997: 447). As we shall see later, that the residual variance is
equal to one is a desirable property for our purpose of comparing different regression
models. The basic probit model is as follows:
7
It would be substantively more interesting to split religiosity according to religion as well. Unfortunately,
however, there were too few cases to make any such analysis sensible.
8
To determine whether there were significant differences in social group voting, we initially fit
multinomial logit models (see Fox, 1997, Sections 15.2 and 15.3). It makes little sense to continue without
evidence that social groups vote as blocks to at least some degree. Likelihood ratio tests showed that there
were indeed group differences, suggesting that we could proceed to examine the effects of social cleavages
on vote for the different types of left and right parties using probit models.
8
π i = Φ (α + β 1x i 1 + β 2 x i 2 + ... + β k x ik
)
Where π i represents the probability that y =1 (in our case, vote for the party), Φ is the
normal cumulative distribution function and (α + β 1 x i 1 + β 2 x i 2 + ...β k x ik ) represents
the mean of Z given x 1 , x 2 + ... + x k . The â’s represent the slopes of the social group
variables. â1 can be interpreted as the increase in Z for a one unit increase in x1, holding
all other x’s constant.
We now turn to the standardization of OLS models of attitudes. A simple way of
standardizing regression equations so their coefficients can be compared is to constrain
the mean squared error (MSE) for the two models to be equal. Recall that an assumption
of the probit model is that the residual variance, and hence the MSE, of Z |x1, x2… xk is
equal to 1. After fitting the OLS models, we can adjust the coefficients and standard
errors so that the MSE of the model equals 1 using the following formula:
β
Adjusted β =
MSE
SE (β )
Adjusted SE (β ) =
MSE
While this adjustment is arbitrary from a substantive point of view, it is sensible
considering that our dependent variables, the left-right scale and libertarian-authoritarian
scale, have artificial metrics. Although we have different measures of attitudes in each of
the data sets, meaning that we can not determine differences in attitudes across societies,
this method allows us to directly compare the relative strength of social cleavages on
attitudes and vote across societies. By way of clarification, let us use gender as an
example. Our method cannot determine in which society women are most libertarian, but
it can show in which societies the impact of gender on attitudes and vote is strongest.
Rather than simply examine the tables of the coefficients from these models,
which is not conducive to determining patterns of association when there are many
regions under analysis, we can graph them on scatterplots, allowing us to clearly
visualize the impact of social groups on attitudes and vote across regions. Figure 1
explains how these graphs are to be interpreted. The horizontal axis represents the
attitude of the adjusted coefficient from the attitudes model; the vertical axis represents
the attitude of the corresponding coefficient from the probit model of vote. If a social
cleavage affects attitudes and vote similarly across regions (as implied by the sociological
approach), the data will be in the top right quadrant (labelled I). If the impact of social
cleavages on attitudes is similar across regions but varies on vote according to the social
and political context (rational choice perspective), we might expect some of the data
points to be in the bottom right quadrant of the graph (labelled II). Finally, if the impact
of social cleavages on attitudes and vote varies in tandem according to the political
context (the party identification model) we would expect to find data points in the bottom
left quadrant of the graph (labelled III).
[Figure 1 about here]
9
This model is a null hypothesis model. That is, it cannot validate the sociological
approach, but it allows us to determine if regions deviate from the pattern implied by the
sociological approach. Likewise, it does not disprove the party identification or rational
choice models, but it can suggest support for them.
[Figure 1 about here]
RESULTS
Tables 1 and 2 display the unadjusted coefficients for the models of attitudes regressed on
social group variables. Notice that the residual standard error for each model is also
reported so that the adjusted coefficients used in the scatterplots can be calculated. Tables
3 and 4 display the probit models fit to left and right vote respectively. A glance at the
tables does indeed show that there are significant group differences in attitudes and
voting in all regions. As said earlier, however, we confine our discussion to those
variables related to our hypotheses.
Figure 2 evaluates the hypothesis that, compared to the young, the old are more
authoritarian and more likely to vote for right-wing parties. The adjusted coefficients
representing the old (a dummy variable contrasting over 65 years of age with those under
30 years of age) from the models of libertarian-authoritarian attitudes are plotted along
the horizontal axis. The corresponding coefficients from the probit models of left vote
and right vote are plotted along the vertical axes. The lines through the data are lowess
smooths of the trends in the relationships of the impact of social groups on attitudes and
vote across regions (see Fox, 2000 for a good description of smoothing techniques).
Figure 2 provides a very clear picture that, as predicted, age has a similar effect on
attitudes in all regions—the young are more libertarian than the old.
[Figure 2 about here]
As expected, the old are also typically more likely to vote for right-wing parties,
which is what was expected. The findings are somewhat surprising with respect to left
vote, however. Quebec deviates significantly from the general pattern—in all other
regions there is no significant effect, but old Quebecers are less likely to vote left. Here
the party competition structure of the region plays a role. Although the New Democratic
Party does exist in Quebec, it is a very minor player, with the left being dominated by the
Bloc Quebecois, which is a strong nationalist party. Since the old are less likely to be
nationalist in Quebec (Nadeau and Fleury, 1995), they avoid the Bloc Quebecois because
of its nationalist rather than left-wing policies. Age also seems to have a stronger effect
on vote in the east and west of Canada than in other regions, though the coefficients are
statistically insignificant.
We now turn to Figure 3 which allows us to examine the hypotheses that
nonwhites are more likely than whites to hold left attitudes and to vote for left parties.
Recall that Scotland is omitted from this analysis because it had too few non-white
respondents. Also in all other non-US regions race is simply treated as non-white versus
10
white, while in the US regions the nonwhites are further divided into blacks and
Hispanics. The coefficients in Figure 3 contrasts visible minorities with whites. Striking
about this graph is how strongly related are the impact of race on attitudes and vote—i.e.,
the coefficients from the left-right attitudes and left and right vote models vary in tandem.
Even more striking, however, are the significant regional differences within the USA,
especially with respect to the contrasts between blacks and whites. The Democratic
Party’s greater emphasis on the concerns of blacks is clearly reflected in higher support
from blacks than experienced by the Republicans. Despite this, it is difficult to explain
why the impact of being black versus white has a positive effect on authoritarian attitudes
in the US south while it is a negative impact elsewhere.
[Figure 3 about here]
Figure 4 displays the coefficients contrasting the religiously devoted (those who
attend religious services at least once a month) with the less religious. As expected,
religiosity is positively associated with authoritarian attitudes in all regions—in no region
is the coefficient negative. These attitudes are not always reflected in voting patterns,
however, and two things are noteworthy. First, religiosity has its strongest impact on vote
in the US regions. It seems sensible to suggest that this is largely due to the fact that the
Republican Party has closer ties to the religious right than do the right-wing parties in
Canada and Britain. Secondly, in Quebec—the region with the strongest nationalist
movement—religiosity has a much stronger impact on vote than it does elsewhere. This
finding has less to do with religiosity than the relationship between church (its quite
possible that religiosity and religion are closely tied together) and party competition. The
Reform Party is a non-factor in Quebec because of it’s perceived anti-Quebec sentiment.
As a result, the much less socially conservative Progressive Conservative Party is the
only choice for those on the right. The Liberal Party, on the other hand, has historically
been the party of Quebec Catholics, and may benefit from the Reform Party’s lack of
presence. Although Scotland and Northern Canada appear to differ from the general
pattern of the data, the impact of religiosity on vote is statistically insignificant in both
regions.
[Figure 4 about here]
Figure 5 allows us to test the hypothesis that social class affects left-right attitudes
and vote similarly. We expect to find that unskilled manual labourers are more left in
terms of attitudes and voting. Not surprisingly, considering previous research on class
voting, we find significant support for this hypothesis in Britain (see Weakliem and
Heath, 1999). Here the sociological model works perfectly well. We find little support for
the hypothesis that class affects attitudes and voting in a similar manner in the regions of
Canada and the USA, however.
[Figure 5 about here]
Ontario is exceptional compared to other North American regions in that the
attitudes of the unskilled are significantly more left than those of managers, and these
11
attitudes are also reflected in a greater likelihood for working class people to vote left
compared to managers. Ontario’s apparent greater level of class awareness may reflect
the strength of the manufacturing sector and the strong connection between the labour
unions in these sectors and the NDP.
The situation is more striking in the US. Although unskilled workers are, on
average, equally or more left than managers in all regions, they are not significantly more
likely to vote left in any of the regions. Most interesting is the case of the northeast,
where although the difference between unskilled workers and managers in terms of right
voting is just as big as it is in any of the British regions, this is not reflected in a greater
propensity to vote left. This seemingly disjointed finding probably reflects that a much
higher proportion of unskilled labourers compared to managers abstain from voting in the
US northeast.
DISCUSSION
This paper set out to integrate the major approaches to voting behaviour in one theoretical
framework. We applied our theory to the major regions of Canada, Great Britain and the
USA. Our analysis is unique in that it models attitudes and voting separately rather than
includes attitudes as a predictor of vote. We argue that the conventional method ignores
that attitudes are, at least partly, determined by social group membership. We were also
unconventional in our analysis of the relative effects of many different social group
variables. Political sociologists usually examine only one major social group (e.g., social
class) without controlling for, or analysing the effects of, other major social groups. By
including attitudes in models of vote one deflates the full effects of social group
membership on voting, ignoring indirect effects through attitudes. By not controlling for
other social group variables we cannot have confidence that the social group under
analysis actually has the observed effect.
Using a novel method we showed how the effects of social cleavages on vote and
attitudes in different regions can be compared even when survey questions are not
identical. Our results indicate that countries should not necessarily been treated as
homogenous regions; that even in apparently similar regions there can be differences in
group attitudes and voting.
The results of our analyses suggest that the sociological explanation of voting
behaviour holds most often when there are strong parties representing the interests of the
relevant social groups. In particular, we found remarkable uniformity in the social basis
of attitudes, with race being an obvious exception. Despite the near uniformity in the
relationship between social groups and attitudes across regions, there were significant
differences in voting patterns that seem to reflect differences in party structures. For
example, the impact of social class on voting was highly variable despites very similar
affects on attitudes.
There are some limitations to the study, most of which are related to the data
employed. The data were not collected with the intent of being divided into many
regions, meaning that the samples sizes employed here were somewhat smaller than
ideal. Moreover, some of our regions could have been further divided had the data
permitted it. For example, party competition is quite different in British Columbia and
Alberta, but because of small numbers we group together these provinces in the West of
12
Canada. Larger samples sizes would allow greater confidence in our estimates, making it
practical to plot confidence regions around them in the scatterplots. (For the present data,
confidence regions are large enough that inclusion of them in the scatterplots would
muddle the picture). Despite this limitation, we find significant differences between
regions within countries. This point cries out for further research and the collection of
new data for this purpose.
Further research could also address other attitude dimensions not tackled in this
paper. With the intensification of global trade and nationalist movements, national
sentiment is an increasingly important attitudinal dimension to consider. Research has
shown that British national sentiment cannot be reduced to the left-right and libertarian
attitude dimensions, but is related to attitudes towards European integration and Scottish
and Welsh devolution (Heath et al 1999, Heath and Kellas, 1998). Heath et al (1999)
conclude that, although not as important as left-right attitudes, national sentiment plays as
strong a role as libertarian attitudes in voting behaviour in Britain. No similar work has
been done in Canada and the US, however, and we did not have appropriate questions to
be able to measure the concept of national sentiment for either country. Even more
obvious, national identity would have a strong impact on voting in Quebec, Scotland and
Wales, where there are independent parties representing national interests.
Despite that we examined 14 regions, that we had only three countries also limits
the scope of our findings. While the three countries are a good comparison because of
their many similarities and differences, the inclusion of regions from other countries with
different party structures and different social cleavages would provide greater insight.
There is no party representing specific religious interests in any of the regions used in our
study. Including countries like the Netherlands and Germany, which have strong religious
cleavages (see DeGraff et al., 2001), and parties competing for their vote, would allow us
to better test the impact of religiosity on vote.
13
REFERENCES
Achen, C. (1992) “Social Psychology, Demographic Variables, and Linear Regression:
Breaking the Iron Triangle in Voting Research,” Political Behavior, 14:195-211.
Alford, R. (1967) “Class Voting in the Anglo-American Political Systems” In S. M.
Lipset and S. Rokkan (eds), Party Systems and Voter Alignments: Cross National
Perspectives. New York: The Free Press.
Alwin, D.F. and J.A. Krosnick (1991) “Aging, Cohorts, and the Stability of Sociopolitical
Orientations over the Life Span,” American Journal of Sociology, 97:169-95.
Bartle, J. (1998) “Left-Right Position Matters, But Does Social Class? Causal Models of
the 1992 British General Election,” British Journal of Political Science, 28:501-529.
Belknap, G. and A. Campbell (1952) “Political party identification and attitudes toward
foreign policy,” Public Opinion Quarterly, 15:601-23.
Berelson, B., P. Lazarsfeld and W. McPhee (1954)Voting: A Study of Opinion Formation
in a Presidential Campaign.Chicago: The University of Chicago Press.
Blais, A., P. Martin and R. Nadeau (1995) “Attentes econmiques et linguistiques et appui
a la souverainete du Quebec: une analyse prospective et comparative,” Canadian Journal
of Political Science, 28: 637-657.
Butler, D. and D. Stokes (1974) Political Change in Britain: The Evolution of Electoral
Choice. London: Macmillan.
Campbell, A., P.E. Converse, W.E. Miller, and D.E. Stokes (1960)The American Voter.
New York: John Wiley & Sons.
Converse, P. (1964) “The nature of belief systems in mass publics,” In D. Apter (ed)
Ideology and Discontent. New York: Free Press.
DeBoeck, P. and S. Rosenberg (1988) “Hierarchical classes: model and data analysis,”
Psychometrika, 53:361-81.
DeGraaf, N.D., A. Heath, and A. Need (2001) “Declining cleavages and political choices:
the interplay of social and political factors in the Netherlands,” Electoral Studies, 20.
Deschamps, J.-C. and Q. Doise (1978) “Crossed category memberships in intergroup
relations.” In H.Tajfel (ed.), Differentiation between Social Groups: Studies in the Social
Psychology of Intergroup Relations, pp141-158. London: Academic Press.
Downs, A. (1957) An Economic Theory of Democracy . New York: Harper and Row.
14
Evans, G., A.F. Heath and M.G. Lalljee (1996) “Measuring left-right and libertarianauthoritarian attitudes in the British electorate,” British Journal of Sociology, 47:93-112.
Evans, G. (ed.) (1999) The End of Class Politics? Oxford: Oxford University Press.
Fleishman, J.A. (1988) “Attitude organization in the general public: the role of core
beliefs and attitudes,” American Journal of Political Science, 32:416-440.
Fiorina, M. (1981) Retrospective Voting in American National Elections . New Haven:
Yale University Press.
Fox, J. (1997) Applied Regression Analysis, Linear Models, and Related Methods .
London: Sage Publications.
Fox, J. (2000) Nonparametric simple regression. Smoothing Scatterplots. London: Sage.
Gerber, A. and D.P. Green (1998) “Rational Learning and Partisan Attitudes,” American
Journal of Political Science, 42:794-818.
Glenn, N.D. (1980) “Attitudes, Attitudes, and Beliefs,” pp.596-640 in O.G. Brim and J.
Kegan (eds.) Constancy and Change in Human Development. Cambridge, Mass.:
Harvard University Press.
Green, D.P. and B. Palmquist (1990) “Of Artifacts and Partisan Instability,” American
Journal of Political Science, 34:872-902.
Green, D.P. and B. Palmquist (1994) “How stable is Party Identification?” Political
Behavior, 43:437-66.
Heath, A. and J. Kellas (1998) “Nationalisms and Constitutional Questions,” Scottish
Affairs , Special Issue:110-128.
Heath, A.F., G.A., Evans and J. Martin (1993) “The measurement of core beliefs and
attitudes: the development of balanced socialist/laissez-faire and libertarian/authoritarian
scales,” British Journal of Political Science, 24:115-32.
Heath, A., B. Taylor, L. Brook and A. Park (1999) “British National Sentiment,” British
Journal of Political Science, 29:155-175.
Hout, M., J. Manza, and C. Brooks (1999) “Classes, Unions, and the Realignment of US
Presidential Voting, 1952-1992.” In G. Evans, The End of Class Politics: Class Voting in
Comparative Context. Oxford: Oxford University Press.
Jenkins, R. (1996) Social Identity. London: Routledge.
15
Jennings, M.K. and R.G. Niemi (1981) Generations and Politics: A Panel Study of Young
Adults and Their Parents. Princeton, N.J.: Princeton University Press.
Lambert, R.D. and J.E. Curtis (1993) “Perceived Party Choice and Class Voting,”
Canadian Journal of Political Science, 26:273-85.
Lazarsfeld, P., B. Berelson, and H. Gaudet (1944) The People’s Choice. New York:
Columbia University Press.
Lijphart, A. (1979) “Religion vs. Linguistic vs. Class Voting,” American Political
Science Review 65:686.
Lijphart, A. (1980) “Language, Religion, Class, and Party Choice: Belgium, Canada,
Switzerland and South Africa Compared,” In R. Rose (ed.), Electoral Participation: A
Comparative Analysis. Beverly Hills: Sage.
Lipset, S. M. and S. Rokkan (1967) “Cleavage Structures, Party Systems and Voter
Alignments: An Introduction,” In S.M. Lipset and S. Rokkan (eds.), Party Systems and
Voter Alignments: Cross -National Perspectives. New York: The Free Press.
Markus, G.B. (1979) “The Political Environment and the Dynamics of Public Attitudes:
A Panel Study,” American Journal of Political Science Review, 23:338-59.
Nadeau, R. and C.J. Fleury (1995) “Gains linguistiques anticipes et appui a la
souverainete du Quebec,” Canadian Journal of Political Science, 28:35-50.
Neto, O. A. And G.W. Cox (1997) “Electoral Institutions, Cleavage Structures, and the
Number of Parties,” American Journal of Political Science, 41:149-74.
Northrup, D. (1998) The 1997 Canadian Election Survey: Technical Documentation.
Toronto: Institute for Social Research, York University.
Olson, M. (1965) The Logic of Collective Action . Cambridge: Harvard University
Press.
Park, A. (2000) “The generation game,” in R. Jowell et al. (eds.) British Social Attitudes,
The 17 th Report. Focusing on Diversity. London: Sage Publications.
Reid, A. and K. Deaux (1996) “Relationship between social and personal identities:
segregation or integration?” Journal of Personality and Social Psychology, 71:1082-91.
Rose, R. and D.W.Urwin (1969) “Social Cohesion, Political Parties and Strains in
Regime,” Comparative Political Studies 2:7-67.
Rose, R. and D.W. Urwin (1970) “Persistence and Change in Western Party Systems
Since 1945,” Political Studies 18:287-319.
16
Stagnor, C., L. Lynch, C. Duan and B. Glass (1992) “Categorization of individuals on the
basis of multiple social features,” Journal of Personality and Social Psychology, 62:207218.
Thomson, K., A. Park, and L. Brook (1999) British General Election Study 1997:
Technical Report. London: National Centre for Social Research.
Weakliem, D.L. and A.F. Heath (1999) “The Secret Life of Class Voting: Britain, France,
and the United States since the 1930s,” In G. Evans (ed.), The End of C lass Politics?
Oxford: Oxford University Press.
17
APPENDIX A: Attitudes scales construction
The BES attitude scales were constructed in the form of additive scales following
from Heath et al (1993) and Evans et al, (1996), who have found these scales to be highly
reliable in the British context. The CES data were not designed with the specific purpose
of creating these indices but they do include a large number of appropriate questions. Our
initial strategy was to limit the number of items to be used through exploratory factor
analysis. Once the number of items was reduced, we created additive scales similar to
those employed from the BES data. All scales have good internal consistency as shown
by the Cronbach’s alpha (British libertarian-authoritarian scale a=.590; British left-right
scale a=.667; Canadian libertarian-authoritarian scale a=.563; Canadian left-right scale
a=.655; American libertarian-authoritarian scale=.637; American left-right scale=.743).
The items used for each of the scales are listed below.
BES Left-right scale9
1. Ordinary people get their fair share of the nation’s wealth
2. There is one law for the rich and one for the poor
3. There is no need for strong trade unions to protect employee’s working conditions
and wages
4. Private enterprise is the best way to solve Britain’s economic problems
5. Major public services and industries ought to be in state ownership
6. It is the government’s responsibility to provide a job for everyone who wants one
BES Libertarian-Authoritarian Scale
1. Young people today don’t have enough respect for traditional British attitudes
2. Censorship of films and magazines is necessary to uphold moral standards
3. People should be allowed to organise public meetings to protest against
government
4. Homosexual relations are always wrong
5. People in Britain should be more tolerant of those who lead unconventional lives
6. Political parties which wish to overthrow democracy should be allowed to stand
in general elections
CES Left-Right scale
10
1. There's not much ANY government can do these days to solve the unemployment
problem
9
All of the items used in both BES scales are Likert items with the following response format: “agree
strongly”, “agree”, neither agree nor disagree”, “disagree” or “disagree strongly”. Both scales are also
balanced so that there are equal numbers of questions worded in favour of both directions of the scales,
limiting response-set bias.
10
The questions used from the CES to construct the Left-Right scale did not all have similar response
formats. The first four questions used five point Likert scales with the following responses: “agree
strongly”, “agree”, neither agree nor disagree”, “disagree” or “disagree strongly”. As can be seen from the
wording of the questions, the response format for questions 4 thru 9 was the following: “a lot”, “some”, not
at all” “not at all”. Don’t know responses were coded so that they were situated between “some” and “not
at all”.
18
2.
3.
4.
5.
6.
7.
8.
To maintain our social programmes we must eliminate the deficit
The government should leave it ENTIRELY to the private sector to create jobs
When businesses make a lot of money, everyone benefits, including the poor
Would you cut WELFARE spending A LOT, SOME, or NOT AT ALL?
Would you cut PENSIONS A LOT, SOME, or NOT AT ALL?
Would you cut HEALTH CARE A LOT, SOME, or NOT AT ALL?
Would you cut UNEMPLOYMENT INSURANCE A LOT, SOME, or NOT AT
ALL?
9. Would you cut EDUCATION A LOT, SOME, or NOT AT ALL?
CES Libertarian-Authoritarian Scale11
1.
2.
3.
4.
5.
6.
Only people who are married should be having children
Society would be better off if more women stayed home with their children.
How do you feel about Aboriginal peoples?
How do you feel about people on welfare?
How do you feel about gays and lesbians?
How do you feel about racial minorities
ANES Left-Right Scale12
1. If people were treated more equally in this country we would have many fewer
problems.
2. Our society should do whatever is necessary to make sure that everyone has an
equal opportunity to succeed.
3. One of the big problems in this country is that we don't give everyone an equal
chance.
4. ONE, the less government, the better; or TWO, there are more things that
government should be doing?
5. ONE, we need a strong government to handle today's complex economic
problems; or TWO, the free market can handle these problems without
government being involved.
6. ONE, the main reason government has become bigger over the years is because it
has gotten involved in things that people should do for themselves; or TWO,
government has become bigger because the problems we face have become
bigger.
11
As with the Left-Right scale, the questions used from the CES to construct the Libertarian-Authoritarian
scale did not all have similar 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”. For questions 3 thru 6 respondents were asked to give a score on a 0 to 100 scale, where 0 means
they “really dislike” the group and 100 means they “really like” the group. In order to make them similar
to the Likert questions so that they could be added together, these responses to these questions were
collapsed into 5 categories. For the most part this had little effect on the response distributions of the
variables since the data were essentially in five clusters.
12
The first three questions used five point Likert scales with the following responses: “agree strongly”,
“agree”, neither agree nor disagree”, “disagree” or “disagree strongly”. Questions 4 thru 6 respondents had
three category responses, with a middle option accepted when declared.
19
ANES Libertarian-Authoritarian Scale13
1. The newer lifestyles are contributing to the breakdown of our society.
2. This country would have many fewer problems if there were more emphasis on
traditional family ties
3. Mothers should remain at home with young children and not work.
4. The world is always changing and we should adjust our view of moral behavior to
those changes.
5. Do you feel STRONGLY or NOT STRONGLY that homosexuals should (or
should not) be allowed to serve in the United States Armed Forces?
6. Do you favor (or oppose) laws to protect homosexuals against job discrimination
STRONGLY or NOT STRONGLY?
13
Each of the six questions used for the ANES libertarian-authoritarian scale were Likert-type items with
five response categories.
20
Table 1. Ordinary Least Squares regression of left-right attitudes on social group variables (standard errors in parentheses)
Canada
Great Britain
Explanatory
West
Ontario
Quebec
East
North
South
Mid.
Variable
England
England
Men
2.154**
1.639**
1.751**
.854
1.008
.419
.015
(.446)
(.538)
(.460)
(.671)
(1.137)
(.259)
(.288)
Age
(reference=18-29)
65 +
1.779*
2.189*
1.697
.248
-1.687
.396
.842
(.919)
(1.121)
(.959)
(1.517)
(3.524)
(.453)
(.485)
46-64
1.232
1.127
.013
-2.058
-1.013
-.275
.682
(.730)
(.864)
(.693)
(1.057)
(2.120)
(.408)
(.422)
30-45
.927
1.087
.341
-.599
-.692
-.352
-.447
(.705)
(.798)
(.628)
(.989)
(1.928)
(.390)
(.394)
Race
(reference=white)
Black
1.939*
1.550
-.162
-4.250*
-1.870
1.236*
.607
(.882)
(.900)
(.861)
(2.076)
(1.550)
(.594)
(.810)
Hispanic
__
__
__
__
__
__
__
Religious
Education
(reference=low education)
University
Some post-secondary
A-level
High School/
O-level or CSE
Social Class
(reference=managers)
Professionals
Self-employed
Clerical
Skilled manual
Semi- or
unskilled manual
Unclassified
Constant
Residual standard error
R2
N
*p-value<.05 **p-value<.01
North
England
-.203
(.351)
Scotland
Wales
.434
(.299)
.445
(.255)
.813
(.589)
.288
(.506)
-.626
(.489)
.680
(.477)
-.175
(.416)
-.254
(.402)
-1.490
(.988)
__
__
United States
West
North-east
South
.243
(2.555)
North
Central
-1.482
(1.868)
.657
(2.017)
1.170
(1.754)
1.231**
(.423)
-.274
(.382)
.226
(.368)
-.624
(.887)
-.144
(.795)
.505
(.762)
.526
(.767)
.457
(.714)
.957
(.700)
1.874
(1.031)
1.679
(.921)
1.698
(.878)
1.269*
(.573)
1.078*
(.512)
.437
(.491)
__
-.738*
(.314)
__
-3.445**
(1.280)
-3.210**
(.950)
.333
(.251)
-5.509**
(1.201)
-.685
(1.142)
.495*
(.214)
-3.790**
(.803)
-.006
(1.164)
-.113
(.273)
3.252**
(.628)
1.960**
(.544)
1.265*
(.513)
.201
(.237)
.138
(.277)
.610*
(.287)
-.219
(.329)
.851
(.696)
.165
(.151)
.196
(.175)
-.042
(.208)
.116
(.158)
.060
(.148)
-.451
(.703)
.785
(.644)
__
.355
(.825)
.152
(.782)
__
-.890
(.743)
-.302
(.674)
__
.750
(1.046)
.572
(.929)
__
1.202
(1.691)
.911
(1.656)
__
1.540
(1.857)
.756
(.450)
.712
(.382)
.540
(.474)
1.044**
(.300)
3.252**
(.628)
1.960**
(.544)
__
-.564
(.954)
1.683**
(.562)
.058
(.470)
-.436
(.420)
.253
(.359)
.770
(1.018)
.002
(.951)
__
.193
(.708)
.022
(.616)
.335
(.567)
.782
(.599)
.121
(.405)
1.190
(.869)
.552
(.763)
__
.233
(.778)
-.044
(.590)
.017
(.477)
.298
(.497)
-.210
(.354)
1.747
(1.179)
1.474
(1.125)
__
.728
(.670)
.733
(.453)
1.014
(.327)
1.030*
(.456)
1.343**
(.327)
.742
(1.081)
.795
(.693)
-.575
(.924)
1.265*
(.513)
-.365
(.989)
-.224
(.993)
-.592
(1.291)
-2.009
(1.159)
.581
(.955)
-1.544
(.999)
19.81**
6.27
.052
890
-.983
(1.137)
.012
(1.230)
-1.858
(1.431)
-2.277
(1.438)
-.445
(1.141)
-.508
(1.163)
20.68**
6.11
.050
596
.501
(.975)
1.939
(1.067)
.909
(1.154)
.054
(1.271)
.148
(.942)
.171
(.942)
20.85**
5.81
.052
681
-2.408
(1.398)
1.556
(1.529)
-4.583**
(1.769)
-1.509
(1.633)
-1.668
(1.317)
-1.556
(1.402)
21.21**
5.28
.099
290
-.913
(1.924)
1.084
(1.961)
-1.598
(2.756)
2.659
(2.599)
.308
(2.129)
.497
(2.032)
19.88**
5.63
.067
117
-1.046**
(.371)
-.357
(.509)
-1.037
(.401)
-2.527**
(.437)
-1.293**
(.419)
.229
(.920)
15.12**
3.64
.075
976
-1.422**
(.478)
-.444
(.654)
-.736
(.498)
-2.144**
(.517)
-1.974**
(.485)
-.017
(.712)
16.28**
3.14
.089
595
-1.051*
(.533)
-1.064
(.885)
-2.002**
(.569)
-2.654**
(.555)
-2.430**
(.540)
-.001
(.999)
18.08**
3.68
.092
611
-.622
(.495)
.003
(.821)
-1.310*
(.514)
-2.702**
(.523)
-2.484**
(.489)
-.363
(.738)
15.67**
3.57
.123
752
-.967
(.536)
-.969
(.525)
-1.463**
(.379)
-1.532**
(.439)
-1.861**
(.399)
-.548
(.497)
15.98**
3.51
.073
1040
-2.039*
(.867)
-.396
(1.041)
-1.741
(.903)
-1.818*
(.865)
-3.167**
(1.319)
-2.763**
(.972)
13.72**
4.19
.140
312
-.027
(.844)
2.024*
(.937)
-.408
(.777)
.333
(.770)
.255
(.866)
.054
(.900)
12.15**
4.09
.102
415
-.807
(.930)
-.024
(1.135)
-1.919
(.993)
-1.222
(.964)
-2.561*
(1.138)
-.662
(1.032)
12.29**
3.83
.206
233
-1.207
(.619)
-.013
(.768)
-.645
(.677)
-.133
(.654)
-.853
(.699)
-.877
(.720)
11.16**
3.83
.224
572
Table 2. Ordinary Least Squares regression of libertarian-authoritarian attitudes on social group variables (standard errors in parentheses)
Canada
Great Britain
Explanatory
West
Ontario
Quebec
East
North
South
Mid.
North
Variable
England
England
England
Men
1.075**
.719
.867**
.100
.448
.222
-.352
.566
(.323)
(.393)
(.290)
(.552)
(.832)
(.217)
(.282)
(.280)
Age
(reference=18-29)
65 +
3.749**
3.663**
3.216**
2.442*
2.491
2.500**
3.085**
3.572**
(.665)
(.818)
(.603)
(1.247)
(2.579)
(.380)
(.473)
(.472)
46-65
2.591**
1.869**
1.688**
2.214
2.895
1.887**
2.508**
2.292**
(.528)
(.630)
(.436)
(.868)
(1.551)
(.342)
(.411)
(.405)
30-45
1.712**
1.317*
.652
1.308
1.956
.647*
1.377**
1.029**
(.510)
(.582)
(.395)
(.813)
(1.410)
(.328)
(.384)
(.392)
Race
(reference=white)
Black
.228
1.418*
-.626
-2.698
-.353
-.564
-1.264
-.045
(.638)
(.657)
(.541)
(1.706)
(1.131)
(.497)
(.791)
(.791)
Hispanic
__
__
__
__
__
__
__
__
Religious
Education
(reference=low education)
University
Some post-secondary
A-level
High School/
O-level or CSE
Social Class
(reference=managers)
Professionals
Self-employed
Clerical
Skilled manual
Semi- or
unskilled manual
Unclassified
Constant
Residual standard error
R2
N
*p-value<.05 **p-value<.01
Scotland
Wales
-.252
(.253)
.512*
(.205)
2.579**
(.404)
1.741**
(.352)
.435
(.340)
__
United States
West
North-east
South
3.439
(2.703)
North
Central
-4.787*
(1.947)
-3.921
(2.365)
3.117
(2.088)
3.419**
(.341)
2.514**
(.308)
1.250**
(.296)
4.370**
(.938)
3.557**
(.842)
2.745**
(.806)
2.202**
(.799)
1.679*
(.744)
1.381
(.729)
3.290**
(1.208)
1.848
(1.080)
.969
(1.030)
3.740**
(.682)
2.410**
(.613)
2.075**
(.584)
__
.767**
(.253)
__
-3.061*
(1.354)
-1.173
(1.005)
1.797**
(.265)
-2.868*
(1.256)
-1.146
(1.190)
1.358**
(.223)
.654
(.942)
-.010
(1.365)
.527
(.320)
-2.397**
(.514)
-.877
(.755)
.988**
(.197)
.927**
(.172)
.800**
(.202)
.519**
(.181)
.603**
(.271)
1.600**
(.509)
.845**
(.127)
.749**
(.171)
.393*
(.167)
.599**
(.134)
.342**
(.119)
-2.900**
(.509)
-1.743**
(.467)
__
-2.964**
(.601)
-2.114**
(.570)
__
-2.847**
(.467)
-1.522**
(.424)
__
-3.578**
(.860)
-2.090**
(.764)
__
-1.135
(1.238)
1.305
(1.211)
__
-.192
(1.359)
-1.003**
(.353)
-.348
(.307)
-.657
(.382)
.269
(.242)
-.635
(.747)
.433
(.648)
__
-1.311
(.784)
-1.532**
(.476)
-.773
(.397)
-1.238**
(.355)
-.095
(.304)
-1.855
(1.193)
-1.666
(1.115)
__
-1.009*
(.445)
-3.102**
(.493)
-.235
(.454)
-.846
(.479)
-.372
(.324)
-.827
(.906)
-.146
(.795)
__
-.648
(.568)
-2.661**
(.576)
-.500
(.465)
-.751
(.489)
.355
(.342)
-1.882
(1.247)
-1.253
(1.190)
__
-1.009*
(.485)
-2.548**
(.379)
-.667*
(.324)
-1.177**
(.383)
-.158
(.274)
-1.136
(1.143)
-.508
(.723)
-1.031
(1.083)
.091
(.611)
-.340
(.716)
.764
(.718)
.795
(.718)
.880
(.839)
.039
(.691)
-.456
(.723)
17.03**
4.53
.157
890
-1.221
(.829)
-.677
(.897)
-1.847
(1.043)
.144
(1.048)
-.218
(.832)
-1.036
(.848)
17.84**
4.57
.176
596
.668
(.613)
1.767
(.671)
1.234
(.725)
1.687
(.799)
.980
(.592)
.771
(.592)
16.30**
3.65
.205
681
.867
(1.149)
1.870
(1.256)
.508
(1.454)
1.061
(1.341)
1.581
(1.082)
1.800
(1.152)
16.45**
4.34
.219
290
-1.380
(1.408)
-.588
(1.435)
.320
(2.017)
3.719
(1.902)
.604
(1.558)
.946
(1.487)
14.48**
4.12
.302
117
-.267
(.311)
.847*
(.427)
.370
(.335)
.310
(.366)
.505
(.351)
.130
(.771)
18.86**
3.05
.234
976
-.903
(.466)
-.025
(.632)
.918
(.486)
.324
(.504)
-.193
(.473)
-.608
(.695)
19.55**
3.06
.256
595
-.558
(.426)
-.551
(.709)
-.278
(.456)
-.025
(.445)
-.180
(.432)
.285
(.800)
17.95**
2.95
.262
611
-.442
(.419)
-.008
(.435)
.539
(.695)
-.092
(.443)
-.521
(.414)
-1.106
(.625)
18.57**
3.02
.187
752
-.096
(.432)
.601
(.422)
.016
(.305)
.761*
(.354)
.447
(.322)
.025
(.401)
16.44**
2.83
.201
1040
-.783
(.917)
-.225
(1.100)
-1.111
(.955)
-.978
(.915)
-1.149
(1.395)
.671
(1.029)
17.48**
4.44
.259
312
-.614
(.880)
2.608**
(.976)
.287
(.810)
1.205
(.803)
1.447
(.902)
3.024
(.939)
18.09**
4.27
.172
415
-.427
(1.090)
1.248
(1.331)
1.240
(1.165)
-.786
(1.131)
.343
(1.335)
.342
(1.210)
18.21**
4.49
.134
233
.558
(.737)
2.047*
(.914)
.999
(.805)
1.645*
(.779)
1.615
(.833)
1.276
(.857)
18.81**
4.56
.156
572
1
Table 3. Probit models of left vote regressed on social group variables (standard errors in parentheses)
Canada
Great Britain
Explanatory
West
Ontario
Quebec
East
North
South
Variable
England
Men
-.482**
-.429**
.131
-.152
-.525
.030
(.119)
(.159)
(.217)
(.188)
(.349)
(.098)
Age
(reference=18-29)
65 +
-.324
.165
-.405
-.391
1.303
.076
(.234)
(.309)
(.529)
(.452)
(.979)
(.175)
46-64
-.310
.145
-.208
.110
.530
.264
(.183)
(.241)
(.350)
(.280)
(.670)
(.151)
30-45
-.195
-.095
.239
.101
-.089
0.067
(.173)
(.228)
(.276)
(.262)
(.631)
(.058)
Race
(reference=white)
Black
-.187
-.253
.181
.591
-.464
-1.191**
(.243)
(.277)
(.357)
(.512)
(.588)
(.221)
Hispanic
__
__
__
__
__
__
Religious
Education
(reference=low education)
University
Some post-secondary
A-level
High School/
O-level or CSE
Social Class
(reference=managers)
Professionals
Self-employed
Clerical
Skilled manual
Semi- or
unskilled manual
Unclassified
Constant
X 2 Model Improvement
df
N
*p-value<.05 **p-value<.01
Mid.
England
-.062
(.119)
North
England
.006
(.120)
Scotland
Wales
-.006
(.108)
-.069
(.092)
.262
(.205)
.336
(.178)
.511*
(.166)
.038
(.204)
.254
(.176)
.302
(.170)
-.054
(.172)
.012
(.150)
.142
(.144)
-.941**
(.346)
__
-.593
(.347)
__
__
United States
West
North-east
South
-2.989
(3.109)
North
Central
-2.356
(2.396)
.145
(.663)
.267
(.591)
.133
(.153)
.336*
(.139)
.234
(.135)
.385
(.275)
.383
(.248)
.246
(.239)
.222
(.244)
.143
(.229)
-.383
(.228)
.404
(.346)
.247
(.310)
.296
(.297)
.474
(.198)
.276
(.180)
.253
(.173)
__
-.011
(.114)
__
.712
(.403)
-.017
(.297)
-.058
(.076)
1.240**
(.386)
.122
(.370)
-.157*
(.069)
.785
(.281)
.183
(.386)
-.057
(.091)
.758**
(.146)
.296
(.213)
.026
(.056)
-.179**
(.065)
-.124
(.081)
-.022
(.154)
-.174
(.091)
-.232
(.212)
-.067
(.058)
-.102
(.073)
.037
(.072)
-.015
(.057)
-.087
(.053)
.142
(.184)
-.090
(.174)
__
.588*
(.244)
.417
(.237)
__
.991**
(.439)
.496
(.438)
__
.166
(.295)
.067
(.267)
__
1.297*
(.605)
.817
(.609)
__
.838
(.666)
-.082
(.162)
-.096
(.138)
-.244
(.171)
-.551**
(.109)
.065
(.212)
-.202
(.185)
__
.354
(.261)
-.285
(.209)
-.293
(.172)
-.018
(.149)
-.064
(.128)
-.024
(.339)
.284
(.318)
__
.312
(.486)
-.081
(.213)
-.282
(.197)
-.180
(.207)
.043
(.139)
.198
(.287)
.193
(.253)
__
.128
(.249)
.414
(.242)
.055
(.197)
-.031
(.208)
-.033
(.147)
-.182
(.367)
.092
(.350)
__
-.087
(.183)
-.080
(.167)
-.645**
(.154)
-.294
(.171)
-.277*
(.120)
-.134
(.335)
.313
(.229)
-.080
(.307)
-.254
(.174)
.496
(.272)
.175
(.287)
.152
(.355)
.527
(.318)
.166
(.273)
.322
(.279)
-1.054**
35.51
15
890
.473
(.370)
.464
(.400)
.674
(.428)
.322
(.485)
.355
(.381)
.534
(.379)
-1.921**
25.41
15
596
.024
(.395)
-2.190
(3.731)
.231
(.482)
-2.209
(4.969)
.139
(.416)
.324
(.404)
-2.170**
20.65
15
681
.087
(.373)
.471
(.396)
.363
(.458)
-.093
(.450)
-.007
(.354)
-.009
(.381)
-1.043*
19.52
15
290
.828
(.551)
.064
(.592)
.738
(.753)
-2.208
(4.382)
-.300
(.743)
.312
(.593)
-2.021**
25.13
15
117
.299*
(.146)
.388*
(.191)
.367*
(.157)
.491**
(.166)
.427**
(.160)
-1.005
(.528)
.229
81.33
16
976
.318
(.199)
.120
(.276)
.142
(.210)
.546
(.216)
.466
(.203)
-.043
(.317)
-.090
34.97
16
595
-.120
(.183)
-.865**
(.335)
-.030
(.195)
.223
(.190)
.098
(.185)
-.300
(.353)
.362
25.40
16
611
-.041
(.186)
.236
(.295)
.199
(.188)
.531
(.190)
.500
(.179)
.413
(.264)
-.469
40.65
15
752
.109
(.193)
.246
(.189)
.245
(.137)
.543**
(.161)
.481**
(.145)
.236
(.180)
-.250
65.52
16
1040
.001
(.261)
-.316
(.276)
-.124
(.317)
-.151
(.262)
-.486
(.411)
-.454
(.300)
-.291
15.43
16
312
.856**
(.275)
.204
(.311)
.294
(.256)
.259
(.251)
-.117
(.285)
-.223
(.303)
-.792*
42.33
16
415
.337
(.311)
-.104
(.381)
.165
(.331)
.127
(.323)
.024
(.382)
-.126
(.345)
-.571
18.08
16
233
-.094
(.206)
-.385
(.267)
-.268
(.230)
-.195
(.221)
-.109
(.237)
-.466
(.247)
-.474
45.48
16
572
2
Table 4. Probit models of right vote regressed on social group variables (standard errors in parentheses)
Canada
Great Britain
Explanatory
West
Ontario
Quebec
East
North
South
Variable
England
Men
-.165
.206
.114
.383*
-.285
-.072
(.122)
(.137)
(.111)
(.171)
(.322)
(.096)
Age
(reference=18-29)
65 +
.315
-.139
.550*
-.169
1.329
.293
(.269)
(.271)
(.249)
(.378)
(.102)
(.170)
46-64
.467*
-.194
.344
-.550*
.661
.212
(.223)
(.215)
(.181)
(.267)
(.650)
(.154)
30-45
.254
-.201
.132
-.434
.205
-.051
(.222)
(.200)
(.167)
(.246)
(.620)
(.151)
Race
(reference=white)
Black
-.347
-.171
.003
.332
-.526
.774**
(.290)
(.238)
(.227)
(.501)
(.553)
(.274)
Hispanic
__
__
__
__
__
__
Religious
Education
(reference=low education)
University
Some post-secondary
A-level
High School/
O-level or CSE
Social Class
(reference=managers)
Professionals
Self-employed
Clerical
Skilled manual
Semi- or
unskilled manual
Unclassified
Constant
X 2 Model Improvement
df
N
*p-value<.05 **p-value<.01
Mid.
England
-.204
(.131)
North
England
-.149
(.143)
Scotland
Wales
-.014
(.151)
.049
(.114)
.730**
(.223)
.579**
(.198)
.205
(.192)
.910
(.259)
.601
(.233)
.120
(.236)
1.304**
(.284)
.842**
(.266)
.446
(.268)
.751
(.505)
__
-.082
(.430)
__
__
United States
West
North-east
South
.666
(.928)
North
Central
.466
(.665)
.381
(.770)
.899
(.651)
.862**
(.204)
.705**
(.188)
.290
(.186)
1.130**
(.337)
.866**
(.315)
.816**
(.307)
.591
(.278)
.368
(.260)
.480
(.257)
.770
(.447)
.452
(.416)
.456
(.400)
.557*
(.235)
.397
(.209)
.159
(.204)
__
-.296
(.155)
__
-3.591
(3.582)
-.367
(.324)
.224**
(.081)
-3.334
(2.194)
-.454
(.421)
.346**
(.073|)
-1.214*
(.497)
-.361
(.483)
.306**
(.110)
-3.498*
(1.750)
-.322
(.257)
.247**
(.067)
.066
(.063)
.098
(.069)
-.221**
(.078)
.071
(.084)
-.256
(.211)
.111**
(.055)
.066
(.076)
.042
(.079)
-.073
(.078)
.173**
(.060)
.506**
(.206)
.440*
(.195)
__
.396
(.213)
.359
(.205)
__
-.214
(.189)
-.134
(.170)
__
-.279
(.263)
-.562
(.239)
__
.145
(.475)
-.185
(.493)
__
.244
(.525)
.422*
(.192)
.162
(.172)
.758**
(.205)
.621**
(.134)
1.974**
(.353)
1.694**
(.331)
__
-.237
(.238)
.382
(.248)
.074
(.233)
.160
(.222)
.211
(.190)
.534
(.418)
.268
(.410)
__
-.076
(.178)
.103
(.236)
.128
(.227)
.457
(.239)
-.092
(.176)
1.147**
(.318)
.781**
(.288)
__
.173
(.209)
-.227
(.280)
.210
(.208)
.351
(.221)
.222
(.158)
.722
(.415)
.526
(.395)
__
.446*
(.200)
-.272
(.173)
.256
(.141)
.085
(.172)
.235
(.121)
.348
(.385)
.434
(.272)
.555
(.409)
1.282**
(.323)
-.222
(.249)
-.365
(.259)
-.329
(.341)
-.410
(.327)
-.299
(.245)
-.101
(.250)
-1.633**
22.53
15
890
-.091
(.289)
.190
(.303)
.107
(.357)
-.093
(.366)
-.248
(.299)
.181
(.290)
-1.205**
17.68
15
596
-.362
(.237)
-.204
(.254)
-.110
(.275)
-.226
(.306)
-.313
(.227)
-.514*
(.231)
-.881**
20.81
15
681
-.033
(.367)
-.270
(.404)
.394
(.451)
-.026
(.423)
.104
(.340)
.057
(.364)
-.097
18.76
15
290
-.351
(.509)
-.678
(.544)
-.630
(.760)
-.333
(.738)
-.352
(.568)
-.791
(.581)
-.931
9.01
15
117
-.138
(.135)
-.182
(.144)
.048
(.181)
-.678**
(.171)
-.463**
(.157)
-.158
(.343)
-1.196**
66.92
16
976
-.565**
(.203)
-.675*
(.287)
-.474*
(.210)
-.732**
(.226)
-.764**
(.210)
-.606
(.325)
-1.269*
48.11
16
595
-.208
(.197)
-.118
(.339)
-.363
(.218)
-.811**
(.237)
-.742**
(.220)
-.480
(.416)
-.830
62.10
16
611
.157
(.210)
-.100
(.231)
.132
(.343)
-.897**
(.286)
-.897**
(.260)
-.190
(.354)
-1.889**
77.11
15
752
-.293
(.229)
-131
(.151)
-.334
(.225)
-.432*
(.189)
-.489**
(.170)
-.557
(.227)
-1.444**
94.18
16
1040
.143
(.286)
.410
(.345)
.119
(.310)
.277
(.290)
.352
(.457)
.267
(.326)
-1.986**
36.29
16
312
-.536
(.281)
.198
(.305)
-.331
(.261)
-.170
(.259)
-.321
(.312)
-.195
(.308)
-1.388**
78.24
16
415
.071
(.333)
-.234
(.377)
.069
(.396)
-.855*
(.413)
-1.148*
(.582)
-.030
(.375)
-1.332**
43.19
16
233
-.041
(.216)
.101
(.275)
-.304
(.267)
.197
(.244)
.151
(.283)
.066
(.290)
-2.298**
166.98
16
572
3
Figure 1
Graph showing the expected relationships between the adjusted coefficients from OLS
regression models of attitudes and coefficients from probit models of vote for the three
models of voting behaviour. The sociological explanation is plausible when the data are
in quadrant I (i.e, social group affects both attitudes and voting in the same manner) .
Data in quadrant II support the rational choice model (i.e., social group affects attitudes
in one way, but voting does not reflect attitudes). Support for the party identification
model (i.e., when attitudes do not reflect social group interests, but voting reflects
attitudes) is found when the data are in quadrant III.
1.0
Vote coefficient
.5
0.0
-.5
-1.0
-1.0
-.5
0.0
.5
Attitides coefficient
4
1.0
Figure 2. Old versus young
2.0
cano
Left vote
1.0
usne
usnc
usso
caon
gbse
gbsc
0.0
uswe
gbme
Country
United States
cawe
caqu
caea
Great Britain
Canada
-1.0
.4
.8
1.2
Libertarian-authoritarian attitudes
Figure 2. Old versus young
2.0
cano
gbsc
uswe
1.0
Right vote
usne
usnc
gbme
usso caqu
cawe
gbse
Country
0.0
caea
caon
United States
Great Britain
Canada
-1.0
.4
.8
1.2
Libertarian-authoritarian attitudes
5
Figure 3. Nonwhites versus whites
2.0
usnc_b
usne_b
usw_b
caea
usso_b
caqu
usne_h
usnc_h
gbwa
usw_h
0.0
Left vote
cano
gbne
usso_h
cawe
caon
gbme
gbse
Country
-2.0
United States
Great Britain
Canada
Total Population
-4.0
-1.5
-1.0
-.5
0.0
.5
1.0
1.5
Left-right attitudes
Figure 3. Nonwhites versus whites
2
gbmegbse
caea
caqu
caon
gbwausne_h caweusso_h
usnc_h
cano
gbne
0
Right vote
usw_h
usne_b
Country
-2
United States
Great Britain
usnc_b
usso_b
usw_b
Canada
Total Population
-4
-1.5
-1.0
-.5
0.0
.5
1.0
Left-right attitudes
6
1.5
Figure 4. Religious versus less religious
.5
.4
.3
Left vote
.2
.1
gbne
caqu
usne
gbwa
.0
-.1
usso
gbsc
gbme
caon
caea
uswe
gbse
Country
usnc
cawe
cano
-.2
United States
-.3
Great Britain
-.4
Canada
-.5
0.0
.1
.2
.3
.4
.5
Libertarian-authoritarian scale
Figure 4. Religious versus less religious
.5
.4
usnc
usne
.3
Right vote
.2
usso
uswe
gbwa
caea
gbne
.1
.0
caon
gbse
cawe gbme
gbsc
Country
-.1
caqu
-.2
United States
cano
-.3
Great Britain
-.4
Canada
-.5
0.0
.1
.2
.3
.4
Libertarian/authoritarian scale
7
.5
Figure 5. Unskilled labour versus managers
.8
gbsc gbme gbwa
gbse
caon
.4
caqu cawe
Left vote
gbne
usne
caea
-.0
usso
usnc
cano
Country
-.4 uswe
United States
-.8
Great Britain
Canada
-1.2
-.8
-.6
-.4
-.2
0.0
.2
Left-right attitudes
Figure 5. Unskilled labour versus managers
.8
.4 uswe
Right vote
caea
usso
-.0
caon
gbwa
-.4
cawe
caqu
usnc
cano
gbse
Country
United States
gbne
gbme
-.8
gbsc
Great Britain
Canada
usne
-1.2
-.8
-.6
-.4
-.2
.0
Left-right attitudes
8
.2
`