Social cleavages and political dealignment in contemporary Chile, 1995–2009

Abstract

There is abundant research on how social cleavages shape political preferences in developed countries with uninterrupted democracies, but we know less about this topic for middle income countries with recently restored democracies. In this analysis of the Chilean case, we examine with Latinobarometer survey data from 1995 to 2009 the evolution of social cleavages as shapers of political preferences (measured with a left–right self-placement scale). We find a general process of dealignment across time, indicated by the decreasing association between political preferences on the one hand, and class, religion and regime preferences on the other. We tentatively link dealignment at the mass level to the strategies pursued by political parties operating in a political and economic context that encourages ideological moderation and convergence to the centre. These strategies weaken the differentiated signals needed for sustaining an aligned citizenry.

Full text

Article
Social cleavages and political dealignment
in contemporary Chile, 1995–2009
Matı
´as A Bargsted
Pontificia Universidad Cato
´lica de Chile, Chile
Nicola
´s M Somma
Pontificia Universidad Cato
´lica de Chile, Chile
Abstract
There is abundant research on how social cleavages shape political preferences in developed countries with uninterrupted
democracies, but we know less about this topic for middle income countries with recently restored democracies. In this
analysis of the Chilean case, we examine with Latinobarometer survey data from 1995 to 2009 the evolution of social
cleavages as shapers of political preferences (measured with a left–right self-placement scale). We find a general process
of dealignment across time, indicated by the decreasing association between political preferences on the one hand, and
class, religion and regime preferences on the other. We tentatively link dealignment at the mass level to the strategies
pursued by political parties operating in a political and economic context that encourages ideological moderation and con-
vergence to the centre. These strategies weaken the differentiated signals needed for sustaining an aligned citizenry.
Keywords
Dealignment, Chile, social cleavages, vote choice/preference
Introduction
In 1989, Chileans elected their national political authorities
by popular vote, ending the 17-year-long military dictator-
ship of General Augusto Pinochet and honouring the coun-
try’s pedigree as one of the most robust democracies in
Latin America. As Chilean democracy consolidated in the
following years, scholars began addressing pressing ques-
tions. To what extent did social cleavages shape the polit-
ical preferences of Chileans in the new democratic
context? How did cleavages evolve as democracy consoli-
dated and socio-economic modernization ensued? What
might explain the observed changes?
Past research provided two answers to these questions.
One was developed by Valenzuela and Scully (Scully,
1992; Valenzuela and Scully, 1997; Valenzuela et al.,
2007). Inspired by Lipset and Rokkan’s (1967) sociological
model of party systems, they argued that political prefer-
ences in post-authoritarian Chile – including voting choices
and ideological positions in a left–right scale – were essen-
tially shaped by traditional religious and class cleavages.
These cleavages were not new – they had structured polit-
ical conflict in Chile since the mid-19th century (religion)
and early 20th century (class).
Scholars such as Tironi and Agu
¨ero (Tironi and Agu
¨ero,
1999; Tironi et al., 2001) and Torcal and Mainwaring
(2003), however, claimed that a new ‘regime preferences’
division between those who supported the Pinochet regime
and those who opposed it had displaced traditional clea-
vages like class or religion. This division, which was epito-
mized by the 1988 plebiscite (in which Pinochet was voted
out and the path for re-democratization opened), shaped the
new political landscape and had an enduring impact on
electoral preferences. Those supporting Pinochet favoured
the centre-right coalition (Alianza por Chile) and those
opposing him favoured the centre-left (Concertacio
´nde
Partidos por la Democracia). Several studies based on
cross-sectional survey data supported this claim (Agu
¨ero
Paper submitted 11 January 2013; accepted for publication 25 October
2013
Corresponding author:
Matı
´as A Bargsted, Instituto de Sociologı
´a, Pontificia Universidad Cato
´lica
de Chile, Av. Vicun
˜a Mackenna 4860, Macul (Campus San Joaquı
´n), Casilla
306, Correo 22, Santiago, Chile.
Email: mbargsted@suc.cl
Party Politics
2016, Vol. 22(1) 105–124
ªThe Author(s) 2014
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1354068813514865
ppq.sagepub.com
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

et al., 1998; Alvarez and Katz, 2009; Bonilla et al., 2011;
Ortega Frei, 2003; Tironi et al., 2001).
In this article we examine the evolution of social clea-
vages in shaping Chileans’ political preferences (measured
as self-identification in a left–right political scale) between
1995 and 2009. We find that, for the entire period, both
sociological and regime preference variables shape loca-
tions on the left–right scale. However, we also find a sys-
tematic reduction in the size of the marginal effects of
multiple cleavage variables, namely education, religious
denomination and regime preferences of respondents. We
relate this trend to a general process of ideological conver-
gence among Chilean parties, for which we provide some
exploratory evidence. This convergence process can be
understood as a progressive political learning whereby
political parties adapted to the institutional and economic
environment inherited from the Pinochet regime.
We contribute to the debate in four ways. First, by
exploring a case of relatively recent democratic consolida-
tion, we expand social cleavage research, which mostly
focuses on cases with uninterrupted democratic rule since
World War II (i.e. Western Europe and North America), but
barely so in middle-income nations with recent democratic
transitions. This allows us to consider a new division – that
between supporters and opponents of a previous authoritar-
ian regime, which for obvious reasons has not been explored
in uninterrupted democracies. We also look at the more tra-
ditional class and religion cleavages.
Second, we test the widespread hypothesis about a gen-
eralized decline of social cleavages (Dalton, 2002; Ingle-
hart, 1990) by looking at cleavage strength year after
year – therefore not assuming linear patterns of evolution.
This is consistent with an issue evolution perspective (Car-
mines and Stimson, 1989), which claims that the strength of
cleavages varies in different directions according to the
issues opportunistically activated by political elites.
Third, by using yearly data for a 14-year period (1995–
2009) we present a truly longitudinal study of the evolution
of Chilean cleavages. This is an improvement over past
studies about Chile, which typically use cross-sectional sur-
veys and therefore cannot assess whether the strength of
cleavages increases or decreases across time (two partial
exceptions are Torcal and Mainwaring [2003] and Ray-
mond and Feltch [2012]). We use a single dataset (the Lati-
nobarometer survey) containing comparable questionnaires
and the same model specification across time.
Finally, we take into account the fact that not all Chi-
leans express a preference on the dependent variable of our
analysis (the left–right ideological self-placement scale).
Consequently, we employ a Heckman selection model
(Heckman, 1979), which allows us to simultaneously esti-
mate individuals’ propensity to express any ideological
preference as well as their position on the scale.
We first review the literature on social cleavages and
political agency and then describe the particularities of the
Chilean political system. We then present our data, meth-
ods and results. In the discussion section we try to make
sense of the results, considering party strategies and the
institutional arrangements inherited from the Pinochet
regime. We conclude and present pending research tasks.
Social cleavages and political agency
One of the most enduring debates in political sociology and
political science revolves around social cleavages, i.e. the
extent to which structural traits like gender, class, religion
or ethnicity shape electoral choices and political ideologies.
The basic idea is that people from different social groups
systematically develop political preferences and make elec-
toral choices that seem to advance their group interests, val-
ues and identities.
While few would discuss that the association between
social categories and preferences exists under certain cir-
cumstances, a more pressing question is why cleavage
strength varies across time or place. One answer comes
from a sociological approach, which suggests that varia-
tions arise from changes in social-structural factors, such
as inequality between and within groups, socio-economic
modernization, value change and changes in group size
(Manza and Brooks, 1999; Inglehart, 1977; Lipset and
Rokkan, 1967). But because social structures typically
change slowly, this approach alone cannot explain
changes in cleavage strength that take place over short
time periods – such as those we diagnose below for Chile.
Thus, we emphasize a more dynamic ‘political agency’
approach, one which focuses on how political actors react
to the political, economic and social setting in which
they are embedded (Chhibber and Torcal, 1997; Evans
and Tilley, 2012; Przeworski and Sprague, 1986).
According to this approach the associations between
social categories and political preferences result from the
choices of political actors embedded in particular contexts.
Parties and politicians develop strategies for gaining the
support of certain social groups. Groups respond to these
appeals and increase their support towards the party, yet
at the same time other groups feel alienated and therefore
support disproportionately another party. This creates or
deepens certain political conflicts and identities – though
not necessarily intentionally (Hetherington, 2001; Main-
waring et al., 2013; Posner, 2004; Torcal and Mainwaring,
2003). Conversely, conflicts may be deactivated when par-
ties develop catch-all strategies or when they move to the
centre of the relevant axis of competition, because citizens
may stop perceiving substantial differences in political sup-
ply (Enyedi, 2005; Kriesi, 1998).
One key aspect for determining party strategies is the
institutional context. For instance, electoral rules may
encourage parties to seek the median voter (Cox, 1997;
Downs, 1957), and this may require downplaying some
conflicts and identities while activating latent others.
106 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

The impact of institutional arrangements may not be imme-
diate but require time, as actors learn how to operate within
certain rules. Political learning helps in explaining why
cleavage strength may change across time due to institu-
tional factors, even if rules do not change. We suggest that
this is the case regarding the impact of the binomial system
in post-transitional Chile. Before presenting the data on
Chile we review research showing how the political agency
approach accounts for changes in cleavage strength regard-
ing the three cleavages we consider in our analysis (class,
religion and regime preferences).
Class and political preferences
Classic post-war electoral studies have shown that in most
industrial societies the working class tended to support left-
ist parties while the middle and upper classes supported lib-
eral and conservative parties (G. Evans, 1999; J. Evans,
2004; Knutsen, 2007; Lazarsfeld et al., 1948; Lipset,
1963; Manza et al., 1995). Social-structural explanations
of variations in the class cleavage include the embourgeo-
isement thesis, the decline of labour unions, increasing
occupational mobility (Dalton, 2002: 153; Evans, 2004:
56; Manza et al., 2005: 215) and cultural change (Inglehart,
1990).
Recent studies have emphasized the political agency
model. Evans and Tilley (2012) found that the decline of
class voting in Britain resulted from the Labour Party’s
move to the political centre in the 1990s and 2000s rather
than from an increase in class heterogeneity. Chhibber and
Torcal (1997) argue that the resurgence of class voting in
Spain during the 1990s should be attributed to the policies
of the governing PSOE. With the raising of unemployment
benefits and increasing social expenditures, these policies
increased workers’ support for the PSOE and alienated the
upper classes. For contemporary Latin America, Mainwar-
ing et al. (2013) have found that class voting in Latin Amer-
ica is higher in countries with a strong and viable leftist
candidate. This is because these candidates emphasize
themes such as land reform, redistribution and social jus-
tice, which polarize the electorate along class lines.
Religion and political preferences
Religious identities are powerful in shaping political pre-
ferences (Manza and Brooks, 1999). Individuals with cer-
tain religious identities may perceive that a given
political party furthers their interests, values or beliefs to
a greater extent than others, thus supporting them dispro-
portionately. These alignments may vary across countries
and regions, and some scholars focus on social-structural
factors in explaining them (Dalton, 2002: 161; Esmer and
Pattersson, 2007: 499; Manza and Wright, 2003).
Other scholars consider the strategies and choices of
parties and candidates. For instance, conservative or
confessional parties may choose to emphasize moral issues
with strong religious overtones when they feel threatened
by liberal governments (Kalyvas, 1998), therefore mobiliz-
ing religious and alienating secular supporters (Mohseni
and Wilcox, 2008: 211). As a reaction to religious embat-
tlements, self-confident irreligious parties may become
more openly secular, as the American Democrats did in the
1980s to face the arousal of the Republican-aligned Chris-
tian Right (Mohseni and Wilcox, 2008: 211).
Parties and politicians may also make strategic choices
that weaken religious cleavages. European Christian Demo-
cratic parties originally mobilized voters on the basis of
religious identities, but as they became catch-all centre or
centre-right parties – as in Italy and Germany – they soft-
ened religious issues and attempted to attract non-Christian
groups and younger voters (Manza and Wright, 2003: 299;
Mohseni and Wilcox, 2008: 218).
Attitudes toward the authoritarian regime: A new
divide?
Finally, we focus on the authoritarian–democratic division.
This is absent in cleavage studies of consolidated democra-
cies simply because they do not have a recent authoritarian
past. Authoritarian regimes may leave a powerful legacy in
their societies – a legacy that colours citizens’ views of
most political issues once democracy is restored. Specifi-
cally, we argue that citizens’ political preferences may be
shaped by their positions toward the previous authoritarian
regime (be it in favour or against it). If they favoured the
authoritarian regime and the latter positioned itself as right-
ist, they should see themselves as rightist and favour right-
ist parties – and vice-versa. The political agency approach
suggests that parties will activate or downplay this division
as a means of obtaining votes or other kinds of political
advantage (e.g. internal cohesion). For instance, parties that
adhere to the regime will activate the division when the
population holds a positive image about it, yet will try to
deactivate it when the regime loses legitimacy (Kitschelt
et al., 2010: ch. 8; Moreno, 1999).
The Chilean case
Within Latin America, Chile is an interesting case because
its social cleavages are supposed to be comparatively
strong (Dix, 1989; Mainwaring and Scully, 1995), partially
as a result of a party system similar to multiparty continen-
tal systems. Since re-democratization in 1990, the Chilean
party system has revolved around two multiparty coali-
tions: the centre-left Concertacio
´n por la Democracia,
composed of the Partido Socialista (PS), Partido Por la
Democracia (PPD), Democracia Cristiana (DC) and Par-
tido Radical Socialdemo
´crata (PRSD) and the centre-
right Alianza por el Cambio, composed of Renovacio
´n
Nacional (RN) and the Unio
´n Demo
´crata Independiente
Bargsted and Somma 107
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

(UDI). The Concertacio
´ngoverned between 1990 and
2010, when it was defeated by the Alianza – currently in
power. The peculiar binomial electoral system – with only
two members being elected in each district – granted a sim-
ilar number of legislators to both coalitions.
Historically, the Chilean class cleavage stemmed from
the early development of a strong labour movement tied
to leftist parties – a tie that was reinforced during the
Socialist government of Salvador Allende (1970–1973) –
and the coalescence of the industrial and land-owning
classes around rightist parties for protecting their privi-
leges. While some studies suggest that class should con-
tinue shaping political positions in the post-authoritarian
period (Valenzuela et al., 2007), others claim the opposite
(Torcal and Mainwaring, 2003). Yet the issue has not been
settled because we lack a comprehensive exploration of the
evolution of such cleavages across time.
The religious cleavage shaped the origins of Chile’s first
party system in the 1850s. The conflict stemmed from
divergences within the political class regarding the influ-
ence of the Church on state and society (Scully, 1992;
Valenzuela, 1995). But there is less agreement about the
current strength of religious cleavages or their evolution
across time. Some believe it is very influential (Valenzuela
and Scully, 1997; Valenzuela et al., 2007), others claim the
opposite (Torcal and Mainwaring, 2003), and a recent study
suggests an increasing salience of religion (Raymond and
Feltch, 2012).
The regime division pits those who favour the authori-
tarian regime of Pinochet against those who prefer democ-
racy. Prior studies have found a strong association between
attitudes toward Pinochet’s regime and political prefer-
ences, with Pinochet supporters favouring rightist parties
and self-identifications and opponents favouring the left
(Tironi and Agu
¨ero, 1999; Tironi et al., 2001; Torcal and
Mainwaring, 2003). These associations stem from the
heavy consequences of the regime for Chileans. The
regime was enduring (it lasted 17 years) and highly
repressive (thousands of its opponents were tortured or
assassinated). Moreover, it engaged in multiple market
reforms that reduced the regulatory role of the state, priva-
tized social services such as health, education and pen-
sions, and increased the flexibility of the labour market.
Additionally, Pinochet sent clear clues that his regime was
a rightist one, e.g. he presented himself as saving the
country from the Marxist left.
Data and methods
We use Chilean survey data from the Latinobarometer proj-
ect between 1995 and 2009, which employs a probability
sample of voting age citizens conducted every year with the
exception of 1999. For reasons detailed at length in the
online supplement, this is the best dataset for a longitudinal
analysis of political preferences in Chile. We also provide
in the supplement a brief description of the methodological
details of the Latinobarometer surveys.
Dependent variable
We measure the political preferences of Chileans using
respondents’ self-placement on a scale ranging from 0 (left)
to 10 (right). This scale is widely used in political behaviour
research because it is shorthand to people’s orientation
‘toward a society’s political leaders, ideologies and parties’
(Mair, 2007: 207). Prior research shows that Chileans consis-
tently order their political parties along the scale and that it
represents a meaningful construct for a majority of the pop-
ulation (Fontaine, 1995; Harbers et al., 2012; Kitschelt et al.,
2010: ch. 5). Moreover, the political scale has been previ-
ously used in research on social cleavages and is strongly
correlated with voting choices not only internationally
(Barone et al., 2007; Norris and Inglehart, 2004: 448; Mair,
2007: 218 f.), but also in Chile (Tironi et al., 2001; Torcal
and Mainwaring 2003).
1
Measures of party preference or
vote recount might capture electoral preferences more
directly, but have high levels of non-response that hinder
multivariate analysis (Morales, 2010). Also, voting beha-
viour has limitations for the study of social cleavages and
political preferences (Barone et al., 2007). Because electoral
choices are strongly influenced by contingencies of political
supply (Alvarez and Nagler, 2000; Cox, 1997), many indi-
viduals vote for parties that do not reflect their true prefer-
ences. In this respect, the left–right scale presumably
better reflects long-term ideological orientations.
Still, the left–right scale in Chile has the problem that a
sizeable and increasing proportion of the population
(around 30 percent in recent years) refuses to locate on it.
Figure 1 shows the proportion of the population that men-
tions any position on the ideological scale between 1995
and 2009. After 1997, when this proportion peaked at 84
percent (according to the weighted data), it decreased to a
record low of 63 percent in 2003. Thereafter the proportion
of identifiers has remained around 70 percent. This propor-
tion is too high to ignore and, as we show below, there are
important differences between the population that mentions
any position on the scale and the population that does not.
This illustrates a classical selection problem where the
observed sample is not a random subset of the entire sample
(Achen, 1986). To address this problem we employ a Heck-
man selection model (Heckman, 1979) and simultaneously
estimate the propensity of individuals to express any ideo-
logical preference as well as their position on the scale.
Independent variables
We measure the democracy–authoritarian divide with the
following question: ‘Which of the following statements
do you agree most with? (a) Democracy is preferable to any
other kind of government; (b) In certain situations, an
108 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

authoritarian government can be preferable; or (c) For
people like me, it does not matter whether we have a
democratic or a non-democratic regime’. Hunneus and
Maldonado (2003) and Valenzuela et al. (2007) have con-
vincingly argued that in Chile answers to this question
reflect people’s attitudes (in favour or against) the Pinochet
regime rather than abstract appraisals about regime types.
Therefore, those who prefer democracy favour the demo-
cratic regime inaugurated in 1990, while those who prefer
authoritarianism favour the previous authoritarian regime.
For our regression model we create two dummy variables
that indicate the authoritarian (alternative b) and indiffer-
ence choice (alternative c), with the latter also including the
‘don’t know’ responses. Alternative (a) (full democrats) is
the reference category.
We measure the class cleavage with two indicators of
an individual’s socio-economic position: the level of edu-
cation (with eight categories ranging from illiterate to
complete university degree, and treated as a continuous
predictor) and a household goods index, which is an addi-
tional index that counts the number of goods each survey
respondent reports possessing or has in his/her house-
hold.
2
We would have preferred a measure of household
income, but this is not available in the Latinobarometer
surveys. Nonetheless, this index is highly correlated with
household income (in surveys of the Centro de Estudios
Pu
´blicos, correlations range between 0.65 to 0.7 during
different years; see section 4 online supplement for
details).
The religious cleavage is captured through a religious
denomination question. We introduce this variable in the
statistical models with three dummies (Catholics, Evange-
licals and a residual ‘others’ category, with people with no
religion as the reference category). All estimates are calcu-
lated controlling for respondent’s gender (dummy for male)
and age (which we divided into five age-group dummies).
Lastly, to avoid identification problems in our Heckman
regression model, we include a four-point interest in poli-
tics variable as an exclusive predictor of the selection equa-
tion. This variable is assumed to affect a respondent’s
propensity to locate on any position on the left–right scale
but not the position they prefer. Exploratory analyses con-
firmed that the correlation between interest in politics and
respondent’s left–right position is significant (0.12), but
also much smaller than that between interest in politics and
respondent’s propensity to locate on the scale (0.32).
3
Results
To evaluate the effect of cleavages on ideological prefer-
ences we conduct two sets of analyses. First, we model
respondent’s propensity to mention a left–right position
and their preferred position using the entire Latinobarom-
eter pooled dataset. Given some missing variables during
Figure 1. Proportion of the population that identifies with the left–right scale.
Bargsted and Somma 109
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

a few applications, this accounts for a total of 11 annual
surveys conducted between 1995 and 2009.
4
In this model
we incorporate all the independent variables mentioned in
the last section, as well as year-specific dummy variables
for absorbing the impact of specific events and circum-
stances that took place during application of the surveys.
The parameters from this model indicate the (partial) asso-
ciation of each independent variable and the left–right scale
for the entire period between 1995 and 2009. Second, we
apply this same model specification, with the exception
of the survey-year dummy variables, to each survey sepa-
rately. Then we calculate the ‘marginal effects’ for each
of the socio-economic, religious and regime preference
variables, and plot the estimates in order to capture the evo-
lution of the association between the different cleavage
variables and respondents’ ideological preferences.
Results from the pooled model are given in Table 1.
Given space constraints we cannot review the results from
the selection equation, but we notice that with the exception
of the ‘others’ religion dummy variable, all coefficients are
statistically significant at a 99 percent level of confidence
or higher. This reinforces the importance of accounting for
the dependency between mentioning a position and loca-
tion on the ideological scale.
The outcome equation also contains many significant
estimates. With the exception of the gender dummy vari-
able, all coefficients are positive, which indicates an
increase towards a more right-wing position. Older respon-
dents locate themselves more to the right than younger
ones, and the coefficients grow monotonically as the
groups grow older. Respondents with more formal educa-
tion and household goods also locate to the right of those
with less of each type of resources. Catholics, Evangelicals
and those identified with another religion are more rightist
than irreligious respondents. The estimates for Evangeli-
cals and Catholics are particularly stark, leading a positive
change equivalent to around three-quarters of a point on the
scale.
Lastly, the two dummy variables reflecting regime pre-
ference also show a positive and significant association
with the left–right scale. For instance, those who mentioned
that an authoritarian government can be preferable are
almost two points more rightist (in the 1–10 scale) than
those who always prefer democracy.
In sum, several social and political divisions have signif-
icant effects on Chileans’ ideological preferences over the
entire 1995–2009 period. We claim that these results pro-
vide simultaneous support for both the classical sociologi-
cal notion that emphasizes the role of traditional cleavages,
particularly social class and religious denomination, and
for the approach that emphasizes the relevance of the divi-
sion between supporters and opponents of the military dic-
tatorship. Moreover, the Chilean case is consistent with the
issue evolution perspective (Carmines and Stimson, 1989),
which is that alternative social and political divisions do
not necessarily substitute, but can complement each other
(see also Raymond and Feltch, 2012).
While the pooled model provides valuable information,
it also hides important levels of heterogeneity in the predic-
tive strength of the cleavage divisions across time. Thus,
we estimated separately for each survey the same model
specification given in Table 1 (though excluding the survey
dummy variables), and calculated the ‘marginal effects’ of
each cleavage variable. We show the results in Figures 2, 3
and 4. Each figure plots the marginal effect of the specified
independent variables along with their 95 percent confi-
dence intervals (calculated via non-parametric bootstrap)
and adds a local fit curve that makes the temporal patterns
more interpretable.
5
We provide the full details of all esti-
mated models, plus a comparison with OLS estimates, in
the online supplement.
Table 1. Heckman selection model for left–right ideological scale
(pooled data).
Selection Eq.
Outcome
Eq.
Intercept –0.748*** 2.139***
(0.081) (0.17)
Male 0.132*** –0.162***
(0.027) (0.047)
26–35 years 0.15*** 0.151**
(0.041) (0.07)
36–45 years 0.131*** 0.222***
(0.041) (0.072)
46–55 years 0.204*** 0.253***
(0.046) (0.079)
56–65 years 0.238*** 0.354***
(0.051) (0.087)
66 years or more 0.136*** 0.438***
(0.052) (0.094)
Education 0.08*** 0.123***
(0.01) (0.018)
Household goods index 0.022*** 0.101***
(0.008) (0.015)
Catholic 0.154*** 0.765***
(0.041) (0.071)
Evangelical 0.129** 0.702***
(0.053) (0.096)
Other religion 0.037 0.331***
(0.064) (0.111)
Don’t care about gov. type / Dk –0.23*** 0.621***
(0.03) (0.061)
Authoritarian gov. can be preferable 0.108*** 1.926***
(0.039) (0.062)
Interest in politics 0.54***
(0.018)
Inverse Mills ratio 1.299***
(0.134)
Rho 0.566
Sigma 2.297
N obs / N censored 12900 / 3047
***p< 0.01, **p< 0.05, *p< 0.1.
Note: Model includes year effects for each survey.
110 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

The evolution of the socio-economic indicators is shown
in Figure 2 showing that during the last few years a simul-
taneous modest increase in the magnitude of the marginal
effect of the household goods index and a sharp decrease
in the marginal effect of education. The decrease of this last
variable has been relatively systematic since 1998 up to the
most recent survey, but sharpens after the 2004 survey,
though the 2007 survey registers a small recovery. Note
that in the last survey the marginal effect of education is
non-significant and the point estimate is very close to zero.
This sharp decline is preceded by four years of a relatively
stable and significant estimate (2000 to 2004), which in
turn is preceded by a more unstable period registering a
very sharp increase in the association between education
and self-location on the political scale. In contrast to this
movement, the marginal effect of the household good index
registers a weak but stable decrease during the first eight
years of available data (and significant only on some occa-
sions). After this the trend reverses, and from 2004 onwards
the marginal effect becomes larger year after year. Its mar-
ginal effect becomes significant in the 2009 survey.
Figure 3 shows the results for the religious denomina-
tion variables. We can see again a reduction in the marginal
effect of a cleavage variable. Being Catholic is consistently
associated across the period with a more right-wing posi-
tion on the left–right scale, but this divergence has been
Figure 2. Evolution of socio-economic marginal effects applied to each available year of Latinobarometer data.
Figure 3. Evolution of religious group marginal effects applied to each available year of Latinobarometer data.
Bargsted and Somma 111
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

decreasing monotonically since 2005. In fact, the marginal
effect during the 2009 survey is non-significant, something
that only occurs one more time in 1995. Among Evan-
gelicals, one can also observe a continued decrease in the
magnitude of the positive marginal effect for the period
studied, though most of the changes occur between the
years 1996 and 2001. Thereafter, identifying with an Evan-
gelical denomination is associated with a positive, rela-
tively stable and sometimes significant marginal effect.
Results among the residual ‘others’ religion category also
show a reduction in the size of the coefficients, and even
becomes negative during the 2009 survey.
Finally, the marginal effects of respondents’ attitudes
to democracy appear in Figure 4. We can see in both cases
a spectacular and constant decrease (which is monotonic
in the case of the authoritarian response) in the size of the
marginal effects starting around the period 2000–2001.
Although the ‘authoritarian government can be prefer-
able’ option remains highly significant during all the
available years, the size of the marginal effect in the
2009 survey is just about half of its size in the 2001 sur-
vey. The ‘don’t care about government’ option remains
positive, but is only marginally significant during the last
survey. Compared to the estimates around the years 2000–
2003, the marginal effect of the 2009 survey is less than
half the size.
In sum, with the only exception of the household goods
index showing a modest increase in its marginal effect
during the last surveys, there is a systematic reduction
in the size of the marginal effects of multiple cleavage
variables, namely education, religious denomination and
regime preferences of respondents. While it is too early
to make definitive claims, it seems that during the period
we cover Chilean society experienced a generalized de-
alignment of the social and political basis of ideological
preferences.
Discussion
How can we make sense of these trends? A thorough
answer is beyond the limits of this article. However, we
suggest – and provide some admittedly non-conclusive evi-
dence – that the observed cleavage decline can be partially
traced to a process of ideological convergence and modera-
tion that has taken place among Chilean parties since re-
democratization on issues related to class, religion and
political regime. This convergence at the elite level wea-
kened the differentiated signals needed to sustain strong
cleavages among the masses (for a similar argument, see
Evans and Tilley, 2012). We argue that this process, in turn,
relates to the political and economic legacy of the Pinochet
regime and how political actors reacted to it. Specifically,
and consistent with a political agency approach, we claim
that several elements of the legacy of the Pinochet regime,
which we identify below, encouraged Chilean political par-
ties, through a learning process, to adopt centrist political
positions and strategies.
Evidence of ideological convergence among Chilean
political parties comes from the Political Elites in Latin
America (PELA) parliamentary survey (visit http://ameri-
co.usal.es/oir/elites/index.htm). This project surveyed Chi-
lean deputies on four occasions during the period under
study (1993, 1998, 2002 and 2006). In tapping the socio-
economic cleavage we consider a question about the
Figure 4. Evolution of attitudes towards regime marginal effects applied to each available year of Latinobarometer data.
112 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

socio-economic model that should prevail in society. Each
interviewed congressman was asked to locate their opinion
on a 5-point scale on which 1 equals ‘would privatize all
public services’ while 5 was ‘would not privatize any pub-
lic service’. The results for this variable are shown on plot
A of Figure 5, which indicates for each wave with available
data the standard deviation of the average position of each
party on this question. As indicated, there is a sizeable
reduction in the size of the standard deviation of about 50
percent between 1994 and 2002 (the question was not asked
in the 2006 survey). This, we claim, indicates an increasing
level of homogeneity, or ideological convergence, in the
views of congressmen on this matter. We found increasing
moderation particularly among leftist legislators. The mean
scores of PS legislators changed from 4,42 in 1993 to 2,66
in 2002, indicating a movement towards privatization. For
PPD legislators the trend was similar: 4,36 in 1993, 2,58 in
1998 and 2,93 in 2002.
The religious attitudes of legislators also show con-
vergence patterns. While a majority of deputies identify
themselves as Catholics across all PELA waves, the corre-
sponding percentage decreasedfrom98percentin1993to
84 percent in 2006. Conversely, in the 2000s there was a
sharp increase in those identifying themselves as ‘Chris-
tians’ – from 0 percent in 1993 to 17 percent in 2006. This
trend is most visible among Christian Democrats and to
a lesser extent among RN legislators and might indicate
a reluctance of congressmen to identify with the insti-
tutional structure of the Catholic Church while still
maintaining a diffuse link with religion. There is also evi-
dence of convergence on religiosity. According to the
PELA survey, the standard deviation of party means in a
question about personal religiosity (where 1 ¼minimum
and 10 ¼maximum) decreased steadily across time, indi-
cating lower religious polarization, as shown on plot C in
Figure 5.
Lastly, changes in attitudes toward democracy among
legislators are also consistent with the decline of the regime
divide at the mass level. The percentage of Chilean legisla-
tors agreeing with the sentence ‘Democracy is better than
Figure 5. Evolution of congressmen attitudes to cleavage related issues.
Bargsted and Somma 113
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

any other form of government’ increased across time from
84 percent in 1997 to 96 percent in 2006 (question not
asked in 1993). This change resulted from an increasing
valuation of democracy by RN and UDI legislators – all
Concertacio
´nlegislators chose the democratic option in the
three survey waves. As legislators achieved near consensus
regarding democracy, there was little room for activating
the political regime divide; elite politics was less capable
of ‘feeding’ this divide at the mass level.
The Pinochet legacy and political learning
How can we explain this apparent process of ideological
convergence and moderation among political elites? No
doubt many factors matter, including social-structural
trends. Increasing secularization and religious pluralism,
as well as rising levels of economic wealth, have perhaps
undermined the politicization of identities along religious
and socio-economic lines. However, we believe there is a
more immediate factor; namely, the political and economic
legacy of the Pinochet regime, and, particularly, the set of
incentives around which political competition was struc-
tured. Importantly, the adaptation to this new setting did not
happen overnight but through a process of political learning
across the 1990s and 2000s. Much of the timing of the
results shown above is relatively consistent with this
perspective.
From the political point of view, the most relevant fea-
ture of the inherited institutional setting refers to the bino-
mial electoral system. This system – in which only two
representatives are elected in each district – makes it dif-
ficult for a single party to gain representation by running
alone. Thus, parties have incentives for building enduring
electoral pacts sufficiently attractive at the local level to
obtain at least one representative. This results in large
coalitions composed of several parties which exclude
extremist forces and therefore reduce polarization in
the electoral supply (Alema´n, 2009; Siavelis, 2002).
Moreover, due to the electoral thresholds provided by the
system, in most cases each coalition obtains one represen-
tative per district (rarely one of them obtains both), which
grants predictability of results and limits real competition
(Alema´n, 2009). Arguably, these factors hinder the clear
and differentiated party signals that promote strong clea-
vages (Evans and Tilley, 2012).
A second relevant institutional element refers to the
‘authoritarian enclaves’ (Garreto´n, 2003) of which the ‘des-
ignated senators’ were particularly emblematic. Once a
president’s period was over, and Pinochet being the first
one, he had the constitutional power to assign a number
of senators to the upper chamber. During the 1990s this
granted veto power to the political right over many areas
of legislation, forcing the Concertacio
´ngovernments to
abandon deep social reforms with no chance of being
approved in Congress (Navia, 2009; Roberts, 2011). Facing
a moderate, non-revolutionary left restricted by ‘authoritar-
ian enclaves’, the right was much more receptive to playing
by the rules of democracy.
Though anecdotal, several empirical patterns in the
behaviour of parties can be connected with the incentives
derived from this general institutional setting, and from the
binomial system in particular. First, internal religious het-
erogeneity within both large coalitions conspired against
any attempt to politicize religious identities. The Concerta-
cio
´nencompasses the religious and (in moral issues) con-
servative Christian Democracy along with the secular and
liberal PS, PPD and PRSD. The Alianza includes the mildly
liberal RN and the very conservative UDI. While both coa-
litions resist some internal diversity, any party that dispro-
portionately favours some religious identities to the
detriment of others may create strains within its coalition
and indirectly favour the rival coalition (Luna, 2008).
Moreover, because parties may need time to learn this
logic, religious cleavages may be initially strong and
decline after a certain time. This partially explains why the
leftist parties of the Concertacio
´ndecided to moderate their
originally innovative bills on paternity, divorce and abor-
tion – all issues with strong religious overtones – and frame
them in ways that emphasized ‘family values’, therefore
being palatable to the Christian Democrats (Haas and
Blofield, 2005: 47). Along the same line, Alema´n and
Saiegh (2007) argue that Concertacio
´nleaders strategi-
cally removed certain moral issues from the legislative
agenda in order to avoid confrontation among coalition
parties.
The UDI provides a second example of how outreach
party strategies weakened social cleavages. Faced with a
large centre-left coalition and an internal coalition partner
(RN) also seeking the right-wing electorate, UDI carried
out a successful strategy in capturing the support of the
popular sectors. This involved developing personal con-
tacts and grassroots mobilization in poor communities,
distributing particularistic benefits and publicly down-
playing the elite character of UDI’s core constituencies
while highlighting its ‘popular’ side (Luna, 2010). Joa-
quı´n Lavı´n, a former UDI mayor and party leader who
almost wins the 2000 presidential election, emphasized
the need for better social protection for the poor and the
unemployed, and implemented high-impact targeted mea-
sures – such as building a beach for the popular classes
that remained in Santiago during the summer. This neo-
populist style, which was widely replicated in UDI’s
municipal governments, marked a sharp contrast with the
traditional upper-class bent of the Chilean right and pro-
moted rightist political identifications among the popular
classes. Perhaps not coincidentally, the class cleavage
(measured with education) declined dramatically after
Lavı´n’s arousal as the leader of the right (Figure 2).
Lastly, in the face of a series of incidents during the
1990s and early 2000s, a consensus emerged among
114 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

political parties regarding the political legacy of Pino-
chet’s regime itself. The political right had good reason
to remain attached to the Pinochet regime when democra-
tization ensued. Different from other authoritarian experi-
ences in the region – Argentina in sharp contrast – when
Chilean democracy was restored Pinochet was supported
by wide sectors of the population and his institutional
legacy was protected by the ‘authoritarian enclaves’.
However, several processes soon motivated part of the
Chilean right to distance itself from Pinochet’s legacy and
endorse democracy. First, as it became clear that the Con-
certacio
´ngovernments did not destabilize social order
through grassroots mobilization, and that they success-
fully promoted economic growth and decreased poverty
within the socio-economic model inherited from Pinochet,
there were few reasons for resenting democracy (Roberts,
2011). Second, several official reports commissioned by
the Concertacio
´ngovernments (the last one being the
Valech report published in 2004) confirmed, beyond
doubt, massive human rights violations by the military
regime. Third, high impact media scandals such as the
international arrest of Pinochet in London in 1998 and the
‘Riggs affair’, which revealed that Pinochet had commit-
ted millionaire fraud with public funds, inflicted severe
damage on Pinochet’s reputation among the citizenry.
These factors encouraged right-wing parties to develop a
strategy that emphasized distance between them and the
authoritarian regime to avoid losing their more moderate
supporters. Once again, this was best incarnated by UDI
leader Joaquı´n Lavı´n, who praised Pinochet’s economic
model but energetically deplored the abuses of the mili-
tary regime and predicated his absolute support for dem-
ocratic rule (Luna, 2010: 346; Navia, 2009). The
emergence of this newer and more moderate right relaxed
the links between rightist self-positioning and support for
the authoritarian regime, opening the way for a decline in
thecleavageasobservedinFigure4.
The economic legacy of the Pinochet regime, in which
multiple areas of public spending – such as education, pen-
sions and health insurance – were partly transferred to the pri-
vate system,also encouraged ideological convergence among
political actors. Although Pinochet’s economic policies were
controversial by the time the first Concertacio
´ngovernment
took office,the country had experienced an average economic
growth rate of 6 percent during the last six years of the dicta-
torship, and in the following decade this continued. During
these years the Concertacio
´nadministrations certainly
expanded social programmes (such as Chile Solidario,the
health reform AUGE and a social security reform), but they
also accepted private property, promoted growth and savings,
attracted domestic and foreign investment, and promoted
international trade. Also, the Concertacio
´ngovernments did
not replace the labour code or the privatized educational, pen-
sion and health systems. Moreover, the first Concertacio
´n
government with a socialist president, Ricardo Lagos, eased
regulations on private companies, increased private participa-
tion in mining and infrastructure projects and signed free trade
agreements with the US and the European Union.
These examples clearly indicate that the Chilean left,
particularly the Socialist Party, abandoned their historical
preferences towards radical socio-economic reform and
accepted the market-centred model of society imposed by
Pinochet (Siavelis, 2002), though, of course, with impor-
tant corrections. This process of political learning ulti-
mately undermined the fear of the upper classes of the
left and weakened the link between the left and the popular
classes. Moreover, this helps us in understanding why, in a
national survey carried out in late 2005, almost half of
those respondents self-identified with the right or centre-
right approved Lagos’s performance (Navia, 2009). In fact,
as with Lavı´n, it was during the Lagos administration that
the class cleavage declined most.
Conclusions
Past research on social cleavages and politics in Chile
revolved around the debate between the role of social cate-
gories (e.g. class and religion) versus that of political divi-
sions (such as attitudes in favour of or against the Pinochet
regime). Yet because this research was based on cross-
sectional data for one or at most two years (Raymond and
Feltch, 2012 for a notable exception), we do not know
whether the effect of cleavages changed across time. Using
yearly surveys for the 1995–2009 period, we found a gen-
eralized process of dealignment. We show that the effect of
the division between those favouring and those opposing
democracy dwindled too. Interestingly, our household
goods index goes against this pattern and becomes more
important as a predictor of political preferences from
2004 onwards.
We provide some preliminary evidence indicating that
cleavage decline is paralleled by a process of ideological
convergence and moderation among Chilean parties that
weakened the differentiated signals needed to sustain
strong cleavages among the masses. Also, we argue that
this process can be understood as a progressive political
learning dynamic whereby political parties adapted to the
institutional and economic environment inherited from the
Pinochet regime.
Of course, more research must be done before we can
reach firmer conclusions. For instance, a comprehensive
assessment of the class cleavage would require more
refined measures, particularly empirical operationaliza-
tions of class position based on occupational categories
such as those used in the social stratification literature (for
examples, see Nieuwbeerta (1995), Manza and Brooks
(1999) and Evans and Tilley (2012)). Also, the religious
factor merits more research that includes indicators of
public and private religiosity that cut across religious
denominations.
Bargsted and Somma 115
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

Appendix Article: ‘Social cleavages and
political dealignment in contemporary
Chile, 1995–2009’
In the following document we provide further methodolo-
gical and statistical information related to the empirical
analysis presented in the above article. Section 1 gives sam-
ple information of the survey data we employ and discusses
the reasons for not using other well-known data sources.
Section 2 provides detailed empirical estimates of the
Heckman selection model for each survey separately. Sec-
tion 3 gives empirical estimates showing the strong associ-
ation between respondents’ positions on the left–right scale
and their electoral preferences. And, finally, section 4 pro-
vides some empirical results validating our household
goods index as a reasonable socio-economic measure.
1. Latinobarometer survey data
According to documentation of the Latinobarometrer Cor-
poration, the Chilean surveys between 1995 and 1998
employed probability with quota surveys that included the
male and female population aged 18 years and older living
in the 29 cities over 40,000 inhabitants between regions I to
X region of the country (and thereby excluded regions XI
and XII). This equates to 70 percent coverage of the adult
population of the country. Respondents within households
were chosen using age and gender quotas, though house-
holds and census districts were chosen randomly. The
surveys conducted between 2000 and 2004 have the same
coverage as the earlier ones, but employed multi-stage
probability samples including respondent selection within
households. Thereafter, Latinobarometer surveys employed
multi-stage probability samples and covered the entire
adult population of the country. The data and abundant
methodological information can be downloaded at www.
latinobarometro.org.
It is worth mentioning that the analysis we carry out
cannot be done using the well-known Centro de Estudios
Pu
´blicos (CEP) surveys, which have employed multi-
stage probability samples since 1994. Several reasons jus-
tify this claim. First, the CEP failed to regularly include
in their surveys a measure capable of capturing the regime
preference dimension, and therefore cannot be used as a
measure of the evolution of the influence of this fundamen-
tal political cleavage. According to our search there is one
question that was asked on several occasions. This was:
‘Considering both the good and bad things of the govern-
ments I’m going to name, what grade from 1 to 7, where
1 is bad and 7 is excellent, you would put the government
of Augusto Pinochet?’ The question was applied during 7
years between 1994 and 2003. Not only is the time span
shorter, but during the surveys where this measure was
included some other key social cleavage variables were
excluded. For example, between 1994 and 1999 not a single
survey included both the Pinochet government evaluation
and a religious affiliation question. Considering this kind
of data limitation, we could potentially reproduce the type
of analysis done in the article (with simultaneous controls
for all relevant cleavage variables) for the years 1999,
2000, 2001 and 2003. This short time period clearly repre-
sents an unsatisfactory option.
In second place, the CEP surveys did not ask the left–
right self-identification 10-point scale up to the year 2004,
6
and even after that the question wording has one or two sig-
nificant changes that undermine the temporal comparabil-
ity of the data. The CEP surveys do include, since 1994
up to this date, an ideological preference question asking
respondents to mention the ideological position that best
describes themselves (with nominal response categories:
right, centre right, centre, centre left and left). Unfortu-
nately this last question has a much higher non-response
rate than the more abstract left–right 10-point scale. Indeed,
CEP surveys from 2004 to 2009 included both questions in
the same questionnaires, which allowed comparisons
between both measures. For this entire period (and using
un-weighted data) the response rate of the nominal ques-
tions was, on average, 18 percentage points lower. Conse-
quently, we favoured using the left–right scale because it
maximizes the number of survey respondents that provide
substantive information.
2. Heckman model selection and OLS results
We begin by providing the full results from each of the
Heckman equations employed to calculate Figures 2, 3 and
4 of the article, and, for comparison, include the OLS esti-
mates of the outcome equation. The results are given in
Tables A, B, C and D.
It is interesting to note that OLS estimates either
understate or overstate the magnitude of the associations
between some of the independent variables and the left–
right scale. Probably the most dramatic case refers to the
coefficients of education. In this case, the OLS estimates
are on several occasions much smaller than the respective
coefficients of the outcome equation of the Heckman
model. Consider the surveys from 1995, 2004, 2005,
2007 and 2009. In these cases the coefficients of the Heck-
man model at least double the size of the OLS estimate.
These results, however, should be of no surprise since edu-
cation is a strong positive predictor of whether each respon-
dent mentioned a position on the left–right scale. The OLS
estimates of the household goods index are also smaller
than the Heckman estimates in several surveys, the most
dramatic cases being observed in the surveys of 1995,
2000 and 2009.
The opposite can be observed among estimates of the
indifference option of the regime preference question (‘For
people like me, it does not matter whether we have a dem-
ocratic or a non-democratic regime.’). In this case, the OLS
116 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

Table A. Heckman selection model for left–right ideological scale in 1995–1997.
1995 1996 1997
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Intercept –0.958*** 1.862*** 3.286*** –1.161*** 3.206*** 3.637*** –0.29 2.159*** 2.721***
(0.257) (0.557) (0.426) (0.27) (0.519) (0.431) (0.287) (0.527) (0.372)
Male 0.147 –0.035 –0.186 0.104 –0.382** –0.44*** 0.109 –0.382*** –0.429***
(0.09) (0.166) (0.152) (0.093) (0.151) (0.147) (0.096) (0.134) (0.127)
26–35 years 0.087 0.294 0.269 –0.167 0.084 0.106 –0.043 0.07 0.094
(0.127) (0.221) (0.206) (0.131) (0.204) (0.204) (0.13) (0.183) (0.177)
36–45 years –0.098 0.24 0.366 –0.322** 0.165 0.201 0.15 0.162 0.132
(0.137) (0.25) (0.232) (0.138) (0.228) (0.224) (0.145) (0.195) (0.187)
46–55 years 0.143 0.47* 0.424* 0.06 0.379 0.368 0.02 –0.071 –0.068
(0.156) (0.275) (0.256) (0.163) (0.244) (0.243) (0.165) (0.228) (0.222)
56–65 years 0.191 0.354 0.292 –0.108 –0.127 –0.095 0.267 0.534** 0.438*
(0.177) (0.323) (0.302) (0.18) (0.291) (0.291) (0.191) (0.251) (0.236)
66 years or more –0.464** 0.592 1.047*** –0.093 0.311 0.335 0.113 0.413 0.385
(0.19) (0.421) (0.388) (0.189) (0.308) (0.307) (0.202) (0.283) (0.275)
Education 0.1*** 0.119* 0.016 0.103*** –0.053 –0.075 0.067* 0.108** 0.074
(0.033) (0.065) (0.056) (0.031) (0.057) (0.053) (0.038) (0.054) (0.048)
Household goods index 0.036 0.149*** 0.103** 0.046* 0.074 0.055 0.055** 0.105*** 0.085**
(0.027) (0.052) (0.048) (0.026) (0.046) (0.046) (0.027) (0.04) (0.036)
Evangelical –0.229 –0.426 –0.383 0.08 1.075*** 1.048*** 0.257 0.918*** 0.855***
(0.212) (0.397) (0.374) (0.224) (0.36) (0.36) (0.239) (0.325) (0.314)
Catholic 0.003 0.262 0.321 0.139 1.232*** 1.233*** 0.199 0.945*** 0.915***
(0.202) (0.396) (0.375) (0.202) (0.343) (0.343) (0.212) (0.302) (0.294)
Other religion 0.142 0.359 0.259 0.194 0.921*** 0.88*** 0.113 1.016*** 1.005***
(0.161) (0.291) (0.272) (0.159) (0.259) (0.258) (0.154) (0.216) (0.211)
Don’t care about gov. type / Dk –0.097 0.72*** 0.93*** –0.117 0.516*** 0.561*** –0.301*** 0.526*** 0.691***
(0.1) (0.2) (0.182) (0.102) (0.19) (0.183) (0.11) (0.197) (0.161)
Authoritarian gov. can be
preferable
0.19 1.842*** 1.797*** 0.327** 1.765*** 1.721*** –0.182 2.268*** 2.357***
(0.125) (0.215) (0.2) (0.135) (0.19) (0.186) (0.132) (0.192) (0.178)
Interest in politics 0.603*** 0.708*** 0.332***
(0.065) (0.074) (0.061)
Inverse Mills ratio 1.997*** 0.567 1.067
(0.452) (0.389) (0.679)
Sigma 2.563 2.311 2.172 2.166 2.009 1.927
Rho 0.779 0.261 0.531
N obs / N censored 964/271 890/251 957/182
***p< 0.01, **p< 0.05, *p< 0.1.
Table B. Heckman selection model for left–right ideological scale in 1998–2001.
1998 2000 2001
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Intercept –0.595** 1.248** 2.299*** –0.265 2.054*** 2.663*** –0.062 2.551*** 2.935***
(0.262) (0.519) (0.38) (0.314) (0.528) (0.428) (0.296) (0.493) (0.454)
Male 0.128 0.075 0.009 0.234** –0.145 –0.263* 0.051 –0.286* –0.338**
(0.091) (0.156) (0.141) (0.105) (0.155) (0.142) (0.099) (0.162) (0.159)
26–35 years 0.357*** 0.726*** 0.449** 0.198 0.128 0.083 –0.096 –0.476* –0.464*
(0.125) (0.229) (0.201) (0.164) (0.224) (0.217) (0.149) (0.251) (0.248)
36–45 years 0.355*** 0.606** 0.302 0.048 0.353 0.355 0.234 0.094 0.014
(0.137) (0.248) (0.217) (0.16) (0.225) (0.218) (0.154) (0.25) (0.245)
46–55 years 0.155 0.73*** 0.591** –0.019 0.054 0.092 0.115 –0.094 –0.14
(0.15) (0.275) (0.254) (0.17) (0.247) (0.241) (0.156) (0.254) (0.25)
(continued)
Bargsted and Somma 117
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

estimate tends to be larger than the coefficient of the out-
come equation of the Heckman model. In some cases
(e.g. in 1998 and 2005) the differences are quite large.
Lastly, a second piece of information justifying the
use of a selection model, instead of simply using OLS, can
be found in the parameter estimate of the inverse mills
ratio. As shown in the tables, the estimate of the inverse
mills ratio is significant at a 0.05 or lower in 8 out of the
11 surveys, and is significant at a 0.10 level in 9 out of
the 11 surveys.
3. Predictive capacity of the left–right scale
In the article we employ the left–right self-identification
scale as our measure of political preferences, and not the
more commonly employed vote recall or vote intention
question. In the article, we provide a theoretical justifica-
tion for this decision, as well as some bibliographical refer-
ences of previous work that employed the same measure.
Nonetheless, in this current appendix we offer some further
empirical results that show a very strong association
between respondents’ positions on the left–right scale and
their declared electoral preferences. Through this analysis
we seek to remove any possible doubt about the validity
of our chosen dependent variable.
In addition to the left–right self-location, the Latino-
barometer survey asked respondents since the year 2001
to state for which party they would vote if there were gen-
eral elections next Sunday. We decided not to use this vari-
able not only because it covers a narrower time period, but
also because it suffers from a very large proportion of non-
response. Indeed, if we add the proportion of responses
‘Don’t know’, ‘Does not vote’ and ‘No response’, this
comes to a total of 51 percent for the period covered
between the years 2002 and 2009 (but excluding 2003).
Despite this problem it is still worth exploring whether peo-
ple’s responses to the left–right scale and vote intention are
associated or not with the segment of the population that
declares both a position on the left–right scale and an elec-
toral preference. Table E provides the results of binary logit
models predicting vote intention for one of the two main
political coalitions (Alianza or Concertacio
´n).
7
Table F
gives more disaggregated estimates predicting vote prefer-
ence for either of the Alianza parties (UDI or RN), the
Christian Democrats (which are part of the Concertacio
´n,
but stand at the political centre), the Concertacio
´ncentre-
Table B. (continued)
1998 2000 2001
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
56–65 years 0.409** 0.764** 0.4 0.074 0.238 0.265 0.4** 0.174 0.058
(0.184) (0.316) (0.275) (0.19) (0.274) (0.267) (0.199) (0.31) (0.302)
66 years or more 0.471** 1.165*** 0.831*** –0.173 0.404 0.542** –0.017 0.026 0.023
(0.196) (0.332) (0.292) (0.185) (0.287) (0.275) (0.19) (0.326) (0.324)
Education 0.045 0.233*** 0.182*** 0.026 0.122** 0.102** 0.028 0.141** 0.113*
(0.039) (0.067) (0.06) (0.035) (0.051) (0.048) (0.039) (0.063) (0.062)
Household goods index 0.037 0.058 0.043 0.043 0.148*** 0.115** –0.043 0.089 0.089
(0.026) (0.048) (0.044) (0.034) (0.051) (0.048) (0.034) (0.058) (0.058)
Evangelical –0.101 –0.22 –0.072 0.353 0.358 0.224 –0.127 0.01 0.054
(0.2) (0.349) (0.32) (0.279) (0.368) (0.353) (0.238) (0.386) (0.382)
Catholic 0.008 0.737** 0.979*** 0.04 0.572* 0.584* 0.173 0.619* 0.553*
(0.175) (0.322) (0.295) (0.21) (0.324) (0.316) (0.22) (0.335) (0.33)
Other religion 0.192 0.839*** 0.826*** 0.117 0.553** 0.508** –0.081 0.857*** 0.893***
(0.134) (0.225) (0.206) (0.171) (0.252) (0.245) (0.155) (0.239) (0.235)
Don’t care about gov. type /
Dk
–0.413*** 0.296 0.819*** –0.378*** 0.908*** 1.152*** –0.178 0.972*** 1.111***
(0.1) (0.229) (0.168) (0.111) (0.218) (0.182) (0.11) (0.203) (0.192)
Authoritarian gov. can be
preferable
–0.017 2.185*** 2.175*** 0.159 1.882*** 1.853*** 0.2 2.272*** 2.263***
(0.136) (0.211) (0.19) (0.148) (0.187) (0.181) (0.128) (0.195) (0.193)
Interest in politics 0.44*** 0.459*** 0.641***
(0.059) (0.064) (0.07)
Inverse Mills ratio 2.056*** 1.285** 0.903**
(0.556) (0.599) (0.452)
Sigma 2.417 2.123 2.288 2.201 2.323 2.288
Rho 0.85 0.561 0.389
N obs / N censored 931/243 1006/170 875/202
***p< 0.01, **p< 0.05, *p< 0.1.
118 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

left parties (PS, PPD or PRSD), or for the far left coalition
Juntos Podemos (dominated by the Communist party). We
estimate coefficients separately for each year for which we
have data available, as well as a pooled model that includes
all the observations together.
8
We first consider the binary logit model predicting
coalition vote. Several important results emerge here.
First, for each year the left–right scale has a highly signif-
icant and substantially large coefficient. According to the
pooled model, a unit change on the left–right scale is asso-
ciated with a 171 percent increase in the odds of the Con-
certacio´n being favoured. Similarly, the McFadden
Pseudo R square and the percentage of correct predictions
from the model indicate an overall strong fit. Considering
the pooled model again, using the left–right scale alone
we can correctly predict 86 percent of cases. This is a
43 percent increase in the predictive capacity of the model
with the left–right scale compared to a null model that
only correctly predicts 60 percent of cases.
9
Second, there
is a remarkable level of stability in the size of the coeffi-
cients across each wave. While somewhat smaller in the
2002 survey, the left–right coefficient still involves a very
strong effect over electoral preferences. Indeed, for this
year a unit change on the left–right scale is associated with
a 97 percent increase in the odds of a Concertacio
´nparty
being favoured. Therefore, the individual’s position on the
left–right self-identificationscaleisnotonlyhighlypre-
dictive of their vote intention, but this statistical associa-
tion between variables tends to be relatively constant
across time.
Table C. Heckman selection model for left–right ideological scale in 2003–2005.
2003 2004 2005
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Intercept –1.806*** 1.945*** 2.718*** –1.453*** 1.289** 2.947*** –0.609** 2.385*** 3.401***
(0.251) (0.566) (0.461) (0.257) (0.588) (0.424) (0.261) (0.523) (0.431)
Male –0.002 –0.384** –0.428*** 0.35*** –0.028 –0.318** 0.126 –0.104 –0.214
(0.086) (0.167) (0.165) (0.087) (0.17) (0.147) (0.088) (0.152) (0.142)
26–35 years 0.125 0.43 0.406 0.17 0.288 0.257 0.278* 0.074 –0.112
(0.134) (0.27) (0.269) (0.135) (0.253) (0.238) (0.142) (0.25) (0.232)
36–45 years 0.084 0.403 0.371 0.109 0.545** 0.459* 0.211 –0.12 –0.272
(0.136) (0.274) (0.273) (0.136) (0.251) (0.235) (0.141) (0.252) (0.237)
46–55 years 0.367** 0.587** 0.477* 0.366** 0.641** 0.414* 0.171 –0.292 –0.412*
(0.147) (0.284) (0.279) (0.148) (0.269) (0.247) (0.149) (0.262) (0.247)
56–65 years 0.177 0.371 0.325 0.315** 0.488* 0.303 0.258 0.02 –0.12
(0.155) (0.315) (0.314) (0.158) (0.29) (0.269) (0.158) (0.28) (0.264)
66 years or more 0.463*** 0.782** 0.626* 0.311* 0.762** 0.692** 0.195 0.253 0.137
(0.165) (0.33) (0.322) (0.166) (0.321) (0.303) (0.168) (0.304) (0.289)
Education 0.138*** 0.186*** 0.125* 0.173*** 0.267*** 0.108* 0.078** 0.108* 0.027
(0.033) (0.07) (0.064) (0.035) (0.072) (0.059) (0.037) (0.065) (0.058)
Household goods index 0.03 0.067 0.06 –0.031 0.036 0.045 –0.037 0.062 0.067
(0.029) (0.061) (0.061) (0.031) (0.059) (0.055) (0.03) (0.051) (0.049)
Evangelical 0.131 0.736* 0.674* 0 0.883** 0.88** 0.075 0.269 0.266
(0.198) (0.396) (0.394) (0.21) (0.383) (0.359) (0.209) (0.352) (0.333)
Catholic 0.123 0.56 0.545 –0.007 0.738** 0.867*** 0.012 0.621** 0.642**
(0.164) (0.346) (0.346) (0.164) (0.323) (0.306) (0.157) (0.283) (0.27)
Other religion 0.245* 1.026*** 0.961*** 0.131 0.638*** 0.571*** 0.097 1.03*** 0.984***
(0.129) (0.259) (0.257) (0.128) (0.224) (0.208) (0.127) (0.213) (0.201)
Don’t care about gov. type / Dk –0.067 1.036*** 1.182*** –0.068 0.449** 0.757*** –0.337*** 0.703*** 1.083***
(0.094) (0.206) (0.197) (0.099) (0.208) (0.187) (0.098) (0.208) (0.176)
Authoritarian gov. can be
preferable
0.249* 2.041*** 2.003*** 0.218* 2.096*** 2.026*** 0.188 1.857*** 1.83***
(0.128) (0.225) (0.224) (0.132) (0.226) (0.209) (0.152) (0.241) (0.228)
Interest in politics 0.717*** 0.558*** 0.582***
(0.064) (0.056) (0.061)
Inverse Mills ratio 0.834** 1.799*** 1.62***
(0.348) (0.394) (0.41)
Sigma 2.38 2.347 2.344 2.091 2.298 2.106
Rho 0.35 0.767 0.705
N obs / N censored 828/369 845/349 911/287
***p< 0.01, **p< 0.05, *p< 0.1.
Bargsted and Somma 119
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

We also estimated for each wave the association
between the left–right scale and a more disaggregated ver-
sion of individuals’ vote preferences. Despite the different
measurement strategies, the results (shown in Table F) indi-
cate very similar results to those just discussed. The coeffi-
cients of the left–right scale are highly significant and large
for each year, and their magnitude remains relatively stable
across time, though once again with the partial exception of
the 2002 survey. The results from this table also show a
high degree of consistency between the magnitude of the
coefficients and the ideological location of each set of par-
ties. Indeed, the coefficients for the left–right scale predict-
ing votes for the centre-left parties (PPD/PS/PRSD) are
more negative than those associated with votes for the
centrist Christian Democrats; the coefficients associated
with the Juntos Podemos coalition are, in turn, even more
negative than the coefficients of the centre-left parties. In
other words, a unit increase in the left–right scale (with
higher values indicating a more right-wing position)
implies a larger reduction in the probability of voting for
the Juntos Podemos coalition than for the centre-left par-
ties, and an even larger reduction than voting for the Chris-
tian Democrats. These varying magnitudes perfectly reflect
the ideological alignment of the Chilean political parties.
Lastly, the fit statistics of the models in Table F are
lower than the ones observed in the simpler models predict-
ing vote preference for either of the main political coalitions.
Model assessment analysis indicated that the reduction of fit
is related to a certain inability of the model to clearly differ-
entiate respondents that vote for the Christian Democrats and
those who prefer the centre-left parties. This, of course,
should not be surprising given that both sets of parties are
Table D. Heckman selection model for left–right ideological scale in 2007–2009.
2007 2009
Selection
Eq.
Outcome
Eq. OLS
Selection
Eq.
Outcome
Eq. OLS
Intercept –1.184*** 1.388** 3.074*** –1.196*** 3.023*** 3.562***
(0.231) (0.681) (0.428) (0.256) (0.528) (0.435)
Male 0.17** 0.165 –0.001 0.127 –0.143 –0.202
(0.083) (0.166) (0.15) (0.086) (0.153) (0.148)
26–35 years 0.342** 0.169 –0.073 0.381*** –0.08 –0.175
(0.138) (0.278) (0.253) (0.137) (0.265) (0.259)
36–45 years 0.205 0.115 –0.026 0.436*** –0.054 –0.168
(0.13) (0.265) (0.247) (0.134) (0.261) (0.252)
46–55 years 0.308** 0.154 –0.114 0.453*** 0.096 –0.028
(0.145) (0.289) (0.264) (0.148) (0.283) (0.273)
56–65 years 0.26* 0.283 0.104 0.522*** 0.737** 0.574**
(0.151) (0.304) (0.282) (0.168) (0.306) (0.291)
66 years or more 0.198 –0.009 –0.129 0.278* 0.286 0.205
(0.165) (0.339) (0.317) (0.163) (0.311) (0.305)
Education 0.067** 0.161** 0.076 0.047 0.014 –0.002
(0.032) (0.066) (0.058) (0.033) (0.056) (0.054)
Household goods index 0.031 0.114* 0.075 0.025 0.145*** 0.128**
(0.029) (0.059) (0.054) (0.028) (0.052) (0.051)
Evangelical 0.372* 0.364 0.062 –0.272 –0.464 –0.347
(0.213) (0.426) (0.392) (0.208) (0.419) (0.409)
Catholic 0.299** 0.688** 0.461* 0.395** 0.523* 0.458
(0.143) (0.301) (0.276) (0.164) (0.283) (0.279)
Other religion 0.36*** 0.866*** 0.637*** 0.107 0.221 0.222
(0.116) (0.249) (0.223) (0.12) (0.214) (0.213)
Don’t care about gov. type / Dk –0.218** 0.349* 0.637*** –0.322*** 0.317 0.487***
(0.091) (0.201) (0.175) (0.091) (0.199) (0.175)
Authoritarian gov. can be preferable –0.039 1.513*** 1.55*** –0.043 1.133*** 1.155***
(0.109) (0.203) (0.189) (0.14) (0.238) (0.235)
Interest in politics 0.439*** 0.591***
(0.052) (0.057)
Inverse Mills ratio 1.616*** 0.68*
(0.469) (0.372)
Sigma 2.342 2.110 2.13 2.101
Rho 0.69 0.319
N obs / N censored 828/369 845/349
***p< 0.01, **p< 0.05, *p< 0.1.
120 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

members of the same coalition and share similar ideological
positions on many issues. Despite these difficulties, the left–
right scale still obtains a good fit. The McFadden Pseudo R
2
is never smaller tan 0.20, and the percentage of correctly pre-
dicted cases is always above 60. The pooled model, with 64
percent of correctly predicted cases, implies an 85 percent
increase in predictive accuracy compared to a null model
correctly predicting only 34 percent of cases.
10
4. Validating the household goods index
In our article we employ the household index as a socio-
economic measure of the survey respondent. It is calculated
as an additive index that counts the number of goods each
survey respondent reports possessing in his/her household.
It includes the following goods: television, refrigerator,
computer, washing machine, landline phone, car and also
hot water. This index has a Cronbach alpha of 0.72.
To validate this measure we compared it with a
household income variable using three different surveys
from the Centro de Estudios Pu
´blicos applied in 1996,
2001 and 2008 (using CEP surveys 32, 41 and 58). The
household income variable is an ordinal scale with 14
income ranges which vary across surveys.
Using these data it is possible to recreate the exact same
household goods index measure. As shown in Table G,
according to the CEP data both variables – household income
and the household goods index – are highly correlated,
with Pearson correlations ranging between 0.65 and 0.7.
Table E. Binary Logit for Coalition Vote Intention.
2002 2004 2005 2006 2007 2008 2009 All years
Intercept 4.358 7.455 6.86 6.319 5.668 6.063 6.485 5.881
(10.366) (10.918) (11.596) (10.805) (10.251) (9.419) (10.135) (–28.319)
Left–right scale –0.677 –1.231 –1.179 –1.058 –0.939 –1.172 –1.134 –0.999
(9.862) (10.162) (10.638) (9.959) (9.939) (9.33) (9.658) (26.631)
Log likelihood –175.866 –141.524 –198.338 –165.22 –149.815 –173.705 –152.292 –1191.495
Pseudo R
2
0.323 0.512 0.451 0.411 0.394 0.386 0.424 0.404
Correct predictions 0.843 0.881 0.877 0.883 0.823 0.807 0.86 0.859
N cases 395 455 567 446 373 410 401 3047
Notes: 1) Numbers in parentheses are tstatistics; 2) Model predicts vote for Concertacio
´n parties.
Table F. Multinomial logit for vote intention.
2002 2004 2005 2006 2007 2008 2009 All Years
DC
Intercept 2.608 5.64 4.402 4.663 3.786 3.338 5.229 3.975
(5.803) (8.008) (6.882) (7.67) (6.405) (5.176) (8.129) (18.389)
Left–right scale –0.49 –1.034 –0.943 –0.887 –0.719 –0.801 –0.997 –0.788
(6.876) (8.339) (7.803) (8.07) (7.259) (6.516) (8.477) (20.427)
PPD/PS/PRSD
Intercept 4.673 8.038 7.112 6.585 6.452 6.172 6.415 6.205
(9.889) (10.837) (11.494) (10.481) (10.094) (9.445) (9.508) (27.589)
Left–right scale –0.877 –1.468 –1.297 –1.234 –1.239 –1.312 –1.293 –1.185
(10.258) (10.675) (10.979) (10.382) (10.314) (9.924) (9.936) (27.705)
Juntos Podemos
intercept 3.575 7.595 6.284 5.802 6.053 5.377 6.442 5.526
(6.015) (8.81) (8.602) (7.929) (7.434) (6.91) (8.4) (20.392)
Left–right scale –1.215 –2.198 –1.889 –1.69 –1.923 –1.689 –1.95 –1.712
(7.822) (9.542) (10.182) (9.462) (8.219) (8.672) (9.727) (24.117)
Log likelihood –393.573 –379.903 –465.838 –430.127 –332.648 –374.7 –386.623 –2831.913
Pseudo R
2
0.223 0.323 0.290 0.247 0.284 0.261 0.269 0.262
Correct predictions 0.602 0.666 0.715 0.627 0.639 0.664 0.601 0.637
N cases 415 473 590 469 388 432 424 3191
Notes: 1) Numbers in parentheses are tstatistics; 2) Coalition for the Change parties as reference category.
Table G. OLS for household goods index.
1996 2001 2008
Intercept 1.646*** 1.429*** 1.132***
(0.066) (0.097) (0.123)
Household income 0.459*** 0.425*** 0.421***
(0.013) (0.013) (0.015)
N cases 1368 1280 1091
R 0.70 0.66 0.65
R
2
0.49 0.44 0.42
Bargsted and Somma 121
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

Acknowledgement
Previous versions of this article benefited from the comments of
participants at the 2012 conference of the World Association for
Public Opinion Research (Bogota´), the 2013 Conference of the
American Sociologial Association (New York), the colloquiums
at the Instituto de Sociologı´a (Pontificia Universidad Cato´lica
de Chile), as well as those made by Kyle Dodson, two anonymous
reviewers, and the editors of Party Politics.
Funding
We appreciate the support of a CONICYT grant from the Chilean
Ministry of Education (CONICYT/FONDAP/15130009) and the
Pastoral Direction of the Pontificia Universidad Catœlica de Chile
(grant 189/DPCC2011).
Notes
1. In section 3 of the online supplement we provide empirical
results that show a very strong association between respon-
dents’ positions on the left–right scale and their declared elec-
toral preferences.
2. The goods included in the index are television, refrigerator,
computer, washing machine, landline phone, car and, addi-
tionally, hot water. This index has a Cronbach alpha of 0.72.
3. Adding interest in politics into the outcome equation does not
produce any meaningful change in the estimates we present
below.
4. To be exact, we have data for all the years in the 1995–2009
period except 1999, 2002, 2006 and 2008.
5. Given that in our specification the cleavage variables appear
in both the selection and outcome equations, the marginal
effect of each covariate does not correspond to its coefficient
in the outcome equations (as in linear regression). Instead,
each independent variable has a direct effect captured through
its respective parameter in the outcome equation, and an indi-
rect effect captured through its estimated parameter in the
selection equation. Formally, the marginal for each indepen-
dent variable kcorresponds to:
@E½yijZ
i>0
@xik
¼kkð"Þi
where kand kare the parameters from the outcome and
selection equations, respectively, is the correlation between
the errors of the outcome and selection equations, and
i¼2
iþwiiwhere iis the inverse Mills ratio and wi
is the linear predictor of the probit selection equation. See
Green (2003: 783) for the full derivation. The marginal
effects shown in Figures 2 to 4 are calculated for an ‘average’
survey respondent who is male, between 36 and 45 years of
age, Catholic, chooses the ‘democracy is always preferable’
option and has a sample average value on education, the
household goods index and interest in politics. The confi-
dence intervals of each figure are normal-theory intervals
estimated using 1,000 samples from non-parametric boot-
strapping simulation.
6. There is one survey in 1995 that includes a 10-point left–right
self-identification scale, but its question wording is com-
pletely different from the surveys of 2004 and after.
7. Recall that both political coalitions date back to the first pres-
idential election held in 1989. Up to this day they have
remained stable and their party membership has remained
almost untouched. While the main parties included in each
coalition have not changed at all, some small parties have
dropped over the years. Work on congress roll-call votes
shows that members of the two main coalitions tend to vote
as ideological blocs (Aleman and Saiegh, 2007).
8. This analysis cannot be done for the years 2000, 2001 and
2003 given that the codes for the political parties are not
available from the Latinobarometer website. We do not cal-
culate the respective model for the surveys before the year
2000 given that the vote intention question was worded dif-
ferently, and therefore is not comparable to the question used
later.
9. In Table E we consider a prediction correct if the predicted
probability of voting for a Concertacio´n party is higher than
0.5 and the respondent declared that she would vote for the
Concertacio´n. Also, we consider a prediction correct if the
predicted probability of voting for the Concertacio´ n is lower
than 0.5 and the respondent declared that she would vote for a
Coalition for a Change party.
10. In Table F we consider a prediction correct if the party men-
tioned by the respondent is also the party with the highest pre-
dicted probability of being voted for by the respondent.
References
Achen CH (1986) The Statistical Analysis of Quasi-Experiments.
Berkeley, CA: University of California Press.
Agu
¨ero F, Tironi E, Valenzuela E and Sunkel G (1998) Votantes,
partidos e informacio´n polı´tica: La fra´gil intermediacio´n polı´-
tica en el Chile post-autoritario. Revista de Ciencia Polı
´
tica
19: 159–193.
Alema´n E (2008) Policy positions in the Chilean senate: An anal-
ysis of coauthorship and roll call data. Brazilian Political Sci-
ence Review 2: 74–92.
Alema´n E (2009) Institutions, political conflict and the cohesion
of policy networks in the Chilean congress, 1961–2006. Jour-
nal of Latin American Studies 41: 467–491.
Alema´n E and Saiegh SM (2007) Legislative preferences, politi-
cal parties, and coalition unity in Chile. Comparative Politics
39: 253–272.
Alvarez RM and Katz G (2009) Structural cleavages, electoral com-
petition andpartisan divide: A Bayesian multinomial probitanal-
ysis of Chile’s 2005 election. Electoral Studies 28: 177–189.
Alvarez RM and Nagler J (2000) A new approach for modeling
strategic voting in multiparty elections. British Journal of
Political Science 30: 57–75.
Barone C, Lucchini M and Sarti S (2007) Class and political pre-
ferences in Europe: A multilevel analysis of trends over time.
European Sociological Review 23: 373–392.
122 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

Blofield M and Haas L (2005) Defining a democracy: Reforming
the laws on women’s rights in Chile, 1990–2002. Latin Amer-
ican Politics and Society 47: 35–68.
Bonilla CA, Carlin RE, Love GJ and Me´ndez ES (2011) Social or
political cleavages? A spatial analysis of the party system in
post-authoritarian Chile. Public Choice 146: 9–21.
Carmines EG and Stimson JA (1989) Issue Evolution: Race and
the Transformation of American Politics. Princeton, NJ: Prin-
ceton University Press.
Chhibber P and Torcal M (1997) Elite Strategy, Social Cleavages,
and Party Systems in a New Democracy Spain. Comparative
Political Studies 30(1): 27–54.
Cox GW (1997) Making Votes Count: Strategic Coordination in
the World’s Electoral Systems. New York: Cambridge Univer-
sity Press.
Dalton RJ (2002) Citizen Politics: Public Opinion and Political
Parties in Advanced Industrial Democracies. Chatham, NJ:
Chatham House.
Dix RH (1989) Cleavage structures and party systems in Latin
America. Comparative Politics 22: 23–37.
Downs A (1957) An Economic Theory of Democracy. Addison-
Wesley Publishing Company.
Enyedi Z (2005) The role of agency in cleavage formation. Eur-
opean Journal of Political Research 44(5): 697–720.
Evans G (1999) The End of Class Politics? Class Voting in Com-
parative Context. Oxford: Oxford University Press.
Evans G and Tilley J (2012) How parties shape class politics:
Explaining the decline of the class basis of party support. British
Journal of Political Science 42: 137–161.
Evans J (2004) Voters & Voting: An introduction. London; Thou-
sand Oaks, CA: Sage.
Fontaine A (1995) Significado del eje derecha-izquierda. Revista
de Estudios Pu
´blicos 58: 79–135.
Garreto´n MA (2003) Incomplete Democracy: Political Democra-
tization in Chile and Latin America. Chapel Hill, NC: UNC
Press.
Haas L and Blofield MH (2005) Defining a democracy: Reform-
ing the laws on women’s rights in Chile, 1990–2002. Latin
American Politics & Society 47: 35–68.
Harbers I, de Vries CE and Steenbergen MR (2012) Attitude
Variability Among Latin American Publics: How Party Sys-
tem Structuration Affects Left/Right Ideology. Comparative
Political Studies 46: 947–967.
Heckman JJ (1979) Sample selection bias as a specification error.
Econometrica 47: 153–161.
Hetherington MJ (2001) Resurgent mass partisanship: The role of elite
polarization. American Political Science Review 95: 619–631.
Hunneus C and Maldonado L (2003) Demo´ cratas y nosta´ lgicos
del antiguo re´gimen. Los apoyos a la democracia en Chile.
Revista Espan˜ola de Investigaciones Sociolo
´gicas 120: 9–49.
Inglehart R (1990) Culture Shift in Advanced Industrial Society.
Princeton, NJ: Princeton University Press.
Inglehart R (1977) The Silent Revolution. Princeton: Princeton
University Press.
Kalyvas SN (1998) From pulpit to party: Party formation and the
Christian Democratic phenomenon. Comparative Politics 30:
293–312.
Kitschelt H, Hawkins KA, Luna JP, Rosas G and Zechmeister EJ
(2010) Latin American Party Systems. Cambridge: Cambridge
University Press.
Knutsen O (2007) The decline of social class? In: Dalton R and
Klingemann HD (eds) The Oxford Handbook of Political
Behavior, pp. 457–480. Oxford University Press.
Kriesi H (1998) The transformation of cleavage politics: The 1997
Stein Rokkan lecture. European Journal of Political Research
33: 165–185.
Lazarsfeld P, Berelson B and Gaudet H (1948) The People’s
Choice: How The Voter Makes Up His Mind in a Presidential
Campaign. New York: Columbia University Press.
Lipset SM (1963) Political Man: The Social Bases of Politics.
Garden City, NY: Doubleday - Anchor Books.
Lipset SM and Rokkan S (1967) Cleavage structures, party
systems, and voter alignments: An introduction. In: Lipset
SMandRokkanS(eds)Party Systems and Voter Align-
ments: Cross-National Perspectives,pp.1–64.NewYork:
Free Press.
Luna JP (2008) Partidos polı´ticos y sociedad en Chile. Trayectoria
histo´rica y mutaciones recientes. In: Fontaine A, Larroulet C,
Navarrete J and Walker I (eds) Reforma de los Partidos Polı
´
-
ticos en Chile, pp. 75–124. Santiago: PNUD, CEP y Cieplan.
Mainwaring S and Scully T (1995). Building Democratic Institu-
tions: Party Systems in Latin America. Stanford, CA: Stanford
University Press.
Mair P (2007) Left–right orientations? In: Dalton R and Klinge-
mann HD (eds) The Oxford Handbook of Political Behavior,
pp. 206–222. Oxford University Press.
Manza J and Brooks C (1999) Social Cleavages and Political
Change: Voter Alignments and US Party Coalitions. New
York: Oxford University Press.
Manza J, Brooks C and Sauder M (2005) Money, Participation,
and Votes: Social Cleavages and Electoral Politics. In: Janoski
T, et al. (eds), Handbook of Political Sociology, New York:
Cambridge University Press, pp. 201–226.
Manza J and Wright N (2003) Religion and political behaviour.
In: Dillon M (eds) Handbook of the Sociology of Religion,
pp. 297–314. New York: Cambridge University Press.
Manza J, Hout M and Brooks C (1995) Class voting in capitalist
democracies since World War II: Dealigment, realigment, or
trendless fluctuation? Annual Review of Sociology 21: 137–
162.
Mohseni P and Wilcox C (2007) Religion and political parties. In:
Haynes J (eds) Handbook of Religion & Politics, pp. 211–230.
London: Routledge.
Moreno AM (1999) Political Cleavages: Issues, Parties, and the
Consolidation of Democracy. Westview Press Inc.
Morales M (2010) Disolucio
´n de la identificacio
´n partidaria en
Chile. Documento de Trabajo Encuesta Nacional UDP, Uni-
versidad Diego Portales.
Bargsted and Somma 123
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from

Navia P (2009) The Chilean Left: Socialist and neoliberal. In:
Burdick J, Oxhorn P and Roberts KM (eds) Beyond Neoliber-
alism in Latin America? Societies and Politics at the Cross-
roads, pp. 17–41. Basingstoke: Palgrave Macmillan.
Nieuwbeerta P (1995) The Democratic Class Struggle in Twenty
Countries, 1956–1990. Amsterdam: Thesis Publishers.
Norris P and Inglehart R (2004) Sacred and Secular: Religion and
Politics Worldwide. New York: Cambridge University Press.
Ortega Frei E (2003) Los partidos polU
´ticos chilenos: Cambio y
estabilidad en el comportamiento electoral 1990–2000. Revista
de Ciencia Polı
´
tica 23(2): 109–147.
Posner DN (2004) The political salience ofcultural difference: Why
Chewas and Tumbukas are allies in Zambia and adversaries in
Malawi. American Political Science Review 98: 529–545.
Przeworski A and Sprague JD (1986) Paper Stones: A History of
Electoral Socialism. Chicago: University of Chicago Press.
Raymond C and Barros Feltch BM (2012) Parties, cleavages and
issue evolution: The case of the religious–secular cleavage in
Chile. Party Politics. Epub ahead of print 19 March 2012.
DOI: 10.1177/1354068811436056.
Roberts KM (2011) Chile: the Left after neoliberalism. In Steven
Levitsky and KM Roberts (eds), The Resurgence of the Latin
American Left, John Hopkins: Baltimore, pp. 325–347.
Scully TR (1992) Rethinking the Center: Party Politics in Nine-
teenth and Twentieth-Century Chile. Stanford, CA: Stanford
University Press.
Siavelis PM (2002) Coalitions, voters and party system transfor-
mation in post-authoritarian Chile. Government and Opposi-
tion 37: 76–105.
Tironi E and Agu
¨ero F (1999) Will the new political landscape in
Chile survive? Revista de Estudios Pu
´blicos 74: 1–17.
Tironi E, Agu
¨ero F and Valenzuela E (2001) Clivajes polı´ticos en
Chile: perfil sociolo´gico de los electores de Lagos y Lavı´n.
Revista Perspectivas 5: 73–87.
Torcal M and Mainwaring S (2003) The political recrafting of
social bases of party competition: Chile, 1973–95. British
Journal of Political Science 33: 55–84.
Valenzuela JS (1995) Orı´genes y transformaciones del sistema de
partidos en Chile. Estudios Pu
´blicos 58: 5–79.
Valenzuela JS and Scully TR (1997) Electoral choices and the
party system in Chile: Continuities and changes at the recovery
of democracy. Comparative Politics 29: 511–527.
Valenzuela JS, Scully TR and Somma NM. (2007) The enduring
presence of religion in Chilean ideological positioning and
voter options. Comparative Politics 40: 1–20.
Author biographies
Matı
´as A Bargsted is an Assistant Professor of sociology at the
Pontificia Universidad Cato´lica de Chile. He earned a Ph.D. in
Political Science from the University of Michigan, Ann Arbor.
His research focuses on public opinion, comparative politics, reli-
gious behavior, and quantitative methods. His work has appeared
in the American Journal of Political Science and edited volumes.
Nicola
´s M Somma is an Assistant Professor of sociology at the
Pontificia Universidad Cato´ lica de Chile. He earned a Ph.D. in
Sociology from the University of Notre Dame. His research focuses
on social movements, political sociology, and comparative-historical
sociology in Latin America.His work has appeared in Comparative
Politics, Sociological Perspectives, and The Sociological Quarterly,
and is forthcoming in Acta Sociologica and the Journal of Historical
Sociology.
124 Party Politics 22(1)
at PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE on March 30, 2016ppq.sagepub.comDownloaded from