Social Inequality in Political Participation: The Dark Sides of Individualisation

West European Politics ISSN: 0140-2382 (Print) 1743-9655 (Online) Journal homepage: http://www.tandfonline.com/loi/fwep20 Social Inequality in Political Participation: The Dark Sides of Individualisation Klaus Armingeon & Lisa Schädel To cite this article: Klaus Armingeon & Lisa Schädel (2015) Social Inequality in Political Participation: The Dark Sides of Individualisation, West European Politics, 38:1, 1-27, DOI: 10.1080/01402382.2014.929341 To link to this article: https://doi.org/10.1080/01402382.2014.929341 Published online: 18 Jul 2014. Submit your article to this journal Article views: 2494 View related articles View Crossmark data Citing articles: 12 View citing articles Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalinformation?journalcode=fwep20 Download by: [37.44.193.101] Date: 21 November 2017, At: 12:07

West European Politics, 2015 Vol. 38, No. 1, 1 27, http://dx.doi.org/10.1080/01402382.2014.929341 Social Inequality in Political Participation: The Dark Sides of Individualisation KLAUS ARMINGEON and LISA SCHÄDEL Has the participatory gap between social groups widened over the past decades? And if so, how can it be explained? Based on a re-analysis of 94 electoral surveys in eight Western European countries between 1956 and 2009, this article shows that the difference in national election turnout between the half of the population with the lowest level of education and the half with the highest has increased. It shows that individualisation the decline of social integration and social control is a major cause of this trend. In their electoral choices, citizens with fewer resources in terms of education rely more heavily on cues and social control of the social groups to which they belong. Once the ties to these groups loosen, these cues and mobilising norms are no longer as strong as they once were, resulting in an increasing abstention of the lower classes on Election Day. In contrast, citizens with abundant resources rely much less on cues and social control, and the process of individualisation impacts on their participatory behaviour to a much lesser extent. The article demonstrates this effect based on a re-analysis of five cumulative waves of the European Social Survey. Social inequality in political participation is a mainstay of political sociology research. We know, for example, that higher income groups and the better educated participate more actively in democratic politics (Anderson and Beramendi 2008; Deth and Elff 2004; Gallego 2007, 2010; Verba et al. 1978, 1993, 1995), particularly in the USA (Alber and Kohler 2010). Inequality also tends to be less pronounced with regard to more conventional or institutionalised modes of political participation; that is, class differences are less palpable for activities such as voting than for less institutionalised modes of democratic engagement, such as participating in demonstrations. For a long time, electoral participation in Western European countries was only marginally affected by social position, as lower classes were successfully mobilised by their social groups and organisations, such as trade unions and political parties (Verba et al. 1978; Gray and Caul 2000). In fact, some authors argue that inequality Correspondence Address: klaus.armingeon@ipw.unibe.ch 2014 Taylor & Francis

2 K. Armingeon and L. Schädel in voting did not exist in Europe at least during the 1980s and 1990s (Teorell et al. 2007: 397; Topf 1995: 48). Recent studies, however, have revealed increasing inequality in various modes of political participation in Europe (Schäfer 2013a, 2013b; for Germany see Gattig 2006). Does this finding also apply to electoral participation across a larger set of countries and for a longer period of time? If so, how can we explain the widening gap in turnout between the upper and lower classes? These are the guiding questions of this article. In this article we focus on one crucial aspect of social position: educational attainment. We will show that there has been an increasing impact of educational attainment on turnout. We argue that this is an effect of individualisation (Beck 1986), which denotes processes of declining societal integration and social control. Social integration and social control are particularly important for members of lower social strata. Voting is a demanding activity, one which requires information, evaluation, and reasoned decision-making. Educational institutions strive to facilitate students to make well-informed, rational, and appropriate decisions in complex situations. Therefore, all other things being equal, the more highly educated a citizen is, the less difficulty he or she should encounter with regard to voting. Citizens with less education are thus disadvantaged in arriving at a voting decision. These decision problems can however be reduced if trustworthy elites and networks provide citizens with cues. Additional incentives for participation are offered if these networks sanction non-participation. Individualisation deprives citizens of such cues and sanctions, a process which affects citizens from the lower social strata to a greater extent than their better-off counterparts. This asymmetrical effect of individualisation explains why electoral participation is increasingly affected by educational attainment. The level of educational attainment has increased in the past decades, with a decreasing share of citizens with only basic education. Barro and Lee (2013: 187) report that in the 24 economically most advanced countries, 6 per cent of the population aged 15 and older had a tertiary education in 1950; 25 per cent had a secondary level of education, 60 per cent had reached the primary level, and 9 per cent had not completed any formal education. Sixty years later, we have quite a different picture: 32 per cent of the populations in the world s richest countries have completed tertiary education, 52 per cent secondary, 14 per cent primary, and only 2 per cent have not completed any form of formal education. The average duration of schooling has increased from 6 to 11 years in this period. The median citizen is clearly much better educated today than after World War II. Since education is correlated with the likelihood of electoral participation which we know from countless studies relying on the explanatory model of socio-economic status (see for example Verba et al. 1995) we would expect rising turnout. This has not however been the case. In the majority of democracies the general trend has been a decline in electoral participation. While the outflowing tide lowers all boats, some groups of citizens may be particularly affected by this general decline. We argue that these are the lower social strata which suffer from changing opportunities

Social Inequality in Political Participation 3 available for citizens when making an electoral decision and going to the voting booth. In the following section we outline our argument and then describe the growing impact of educational attainment on electoral participation. This description is based on 94 electoral surveys from eight Western European countries in the period between 1956 and 2009. Having demonstrated that participation has become increasingly unequal, we test our causal argument regarding the asymmetrical effect of individualisation on electoral participation in a cross-sectional analysis of the cumulative European Social Survey data, 2002 2010. The Argument Before delving into the discussion about how social position has increasingly impacted on electoral participation, it is very useful to first consider the normative relevance of such a discussion. From a minimalist or liberal view of democracy (Schumpeter 2010), social inequality in participation is not a major problem if citizens are not actively kept from participating. Furthermore, inequality of participation is seen as unproblematic as long as it does not impinge on the working of democracy, which, understood in a minimalist Schumpeterian perspective, is the allocation of power to decide in the course of a competitive process. In addition, liberal thinkers argue that the right to vote also includes the right not to vote: if, for example, members of lower classes should decide to abstain from voting, this is their free choice. The normative response posits that democracies are based on the ideals of political equality, appropriate representation of societal preferences, and responsiveness of government. From this perspective, equality of participation is a major criterion for the quality of a democracy. All social groups and all preferences should be represented in the election of legislators and executives. If the less advantaged do not vote, politicians have little incentive to represent their interests (see Mahler 2008). In addition, democracy is not only about the selection of governments, rather it is a process of deliberation among citizens. If certain groups of citizens are systematically underrepresented in these processes, the outcomes of deliberation will be biased (Schäfer 2010, 2013a, 2013b). In what follows we start from the normative position that equality of political participation is a key element for the quality of democracy. But how does social position affect political participation? Solt (2008, 2010) points to a mechanism behind the social inequality and participation relationship, drawing heavily on the famous analysis by E.E. Schattschneider. The main argument states that economic inequality powerfully depresses electoral participation of the less affluent, as politics is the realm of powerful and resource-rich citizens and the lower strata will find that their debates are of little interest to them (Solt 2008: 48; for a similar point see Shore 2014).

4 K. Armingeon and L. Schädel There is no need for a conspiracy to deprive other Americans of the meaningfulness of their votes. The process is automatic and largely unnoticed, a consequence of competition among wealthier citizens for political power: when more affluent people use their money to amplify their own positions in some debates, they drown out the voices of poorer citizens and so keep the issues they would raise from being discussed. (Solt 2010: 287; see also Gattig 2006) By implication, the greater the inequality, the more non-participation by members with fewer socio-economic resources there will be. This fits with the long-term observation that turnout has been declining while income inequality has increased (OECD 2011). On the other hand, given the expansion of secondary and tertiary education in affluent democracies over the past decades, the level of overall educational attainment has increased. This development should have given way to higher turnout rates. However, the average turnout has declined in mature democracies from about 85 per cent in 1960 and over 80 per cent in 1980 to about 75 per cent in 2011 (calculated from Armingeon et al. 2013). We argue that lower social strata indicated by the level of educational attainment depend more on cues and social control for political participation than higher strata. If people are not well integrated in socio-cultural groups, they have few chances to find shortcuts that help in making an electoral choice. In addition, if politically organised socio-economic groups no longer exist, we will also not find elite members offering such cues. The process of individualisation therefore deprives poorly educated citizens of cues that facilitate participation. In contrast, citizens with considerable educational attainment do not rely on such cues to the same extent. We expect that these processes of diminishing cues will have particularly dramatic consequences for those at the lower end of the social hierarchy. This argument is based on three established theories. The first theory predicts that people participate if they want to, if they can, and if they are asked to (Brady et al. 1995: 271). In this view, resources and mobilisation are key (Brady et al. 1995; Rosenstone and Hansen 1993; Verba et al. 1995). A major element of this explanation addresses social integration as a favourable condition for political participation. It rests on broad empirical evidence and a clear argument, which is usually phrased in social-psychological terms. Social integration describes citizens who live in families, who are members in interest organisations, or who feel a sense of belonging to social, cultural, or political milieus. Famous examples are the theories on voter alignments and societal cleavages (Lipset and Rokkan 1967). From this perspective, socio-cultural elites serve to mobilise members of their group to take political action and also provide shortcuts that reduce the costs of electoral decisions (Downs 1957). Leaders of trade unions, churches, and political parties appeal to their members during electoral campaigns. They may officially endorse candidates and parties

Social Inequality in Political Participation 5 or describe the dangers if a preferred party does not gain sufficient support. Likewise, respondents who live with their families have a greater likelihood of going to the voting booth compared to citizens who live alone (Hobbs et al. 2014; Powell 1986; Verba et al. 1995). Families reduce the complexity of making an electoral decision, as the decisions of other members can provide cues. Moreover, voting may be seen as a socially desirable behaviour within families. People who live alone, on the other hand, do not have the same discussion opportunities at the breakfast table on the morning of Election Day and no one will question their decision to stay in bed or head straight to the office instead of paying a visit to their polling station. There are thus at least two ways that this mobilisation occurs: through social control and by reducing the individual decision costs by making reliable suggestions. 1 This is where the theory of heuristics in political decision-making comes to the fore. This line of thought argues that in many, if not most decision situations, citizens rely on shortcuts ( cues ), since it is often difficult and time-consuming to obtain and analyse the information necessary to make a decision on one s own in a reasoned way. These heuristics can be very efficient and reliable. Imagine a blue-collar worker who has neither the time, nor the knowledge, nor the political interest to decide whether a particular social policy reform is in his or her best interest. Rather than taking part in discussions, reading the literature, and systematically evaluating the arguments, he or she can proceed as follows: Based on my experience, my interests as a worker are usually best served by trade unions and social democratic parties. Therefore and given that I trust these organisations and unless I have no major counter-arguments I can simply follow their guidelines in this decision. The explanatory power of such cue theories has been repeatedly shown in empirical analyses about attitudes toward the European Union (Hooghe and Marks 2005) or direct democratic choices in Switzerland (Kriesi 2005). Cues are particularly important when a reasoned decision is less feasible due to constraints of time, motivation, or lack of ability to obtain and evaluate information and make a decision in complex or ambiguous situations. Education aims to improve the students capacities in terms of problem-solving and rational decision-making. In addition, educational attainment is a major determinant or at least a correlate of social position. Therefore, citizens with low levels of education are frequently citizens in the lower social strata and also lack capabilities to make reasoned electoral decisions. They are therefore particularly dependent on the availability of cues. Finally, the third theory concerns the process of individualisation the decline of social integration and social control and the emergence of the critical citizens. Because their attachment to a close-knit social group has weakened, these critical citizens are less exposed to the forces of social control. Their political decisions are increasingly their own argument-based individual decisions, since acceptable cues from groups have become less available (Norris 1999). Not everyone who has severed or loosened her or his ties to social groups will however participate in politics. Some citizens will opt for non-participation, since they lack the motivation and the resources to make

6 K. Armingeon and L. Schädel their own reasoned decision when they are no longer strongly socially integrated (Gray and Caul 2000). Furthermore, they are also deprived of cues and controls from informal and formal networks. This is particularly a problem for citizens at the lower end of the social hierarchy who tend to have limited resources for obtaining and critically evaluating information on their own. In contrast, for more advantaged citizens, the reduced availability of cues and the lack of social pressure will have fewer negative consequences for political participation. These three arguments focus on the interaction of micro- and meso-variables. They do not take into account other time-specific and time-invariant yet countryspecific determinants of electoral participation (for an overview see Geys 2006). For example, past research has demonstrated that the level of voting depends on the size of population, the electoral laws and other political institutions of a nation, the closeness of an election (Powell 1980, 1986), the habit of voting, which reflects the political socialisation of generations at various points in time in the history of a democracy (Franklin 2004), the type of party composition (Anderson and Beramendi 2008), or even on the genetic composition of a national electorate (Fowler 2006; Fowler and Dawes 2008). In addition, overreporting of participation is greater in countries with high turnout (Karp and Brockington 2005). In the causal analysis in the fourth section we control for these country- and time-specific factors with time and country dummies and test for the robustness of findings by an alternative specification in a multi-level model. Educational Inequality in Voting Has the effect of educational attainment on voting increased over the past decades? We focus on educational attainment, something which is strongly correlated with social position. Therefore, turnout inequality among educational groups also implies turnout inequality between social groups. In addition, our causal argument states that citizens will withdraw from electoral participation if they have problems in terms of collecting and evaluating complex information, efficiently making decisions, and when social cues are no longer available. The extent to which citizens are able to make decisions in ambiguous situations is arguably closely linked to one s level of education. Hence educational attainment measures both the extent to which a citizen is dependent on cues and the social position of the citizen. As we do not yet have systematic empirical answers to this question regarding the long-term changes in the effects social position has on political participation, we re-analysed 94 national election studies in Western European nations. The first study we analyse dates back to 1956, with the most recent survey conducted in 2009. We only considered surveys that are more or less comparable with regard to the wording for the dependent variable (participation in the last national election) and with regard to ordinal-scaled information on educational achievement. These surveys are however not completely identical: They were

Social Inequality in Political Participation 7 designed by various institutions and were implemented using different sampling techniques and questionnaires. However, these are by far the best data we have for a longitudinal analysis covering more than five decades. Such election studies are available for eight countries: Denmark, Germany, Italy, the Netherlands, Norway, Sweden, Switzerland, and the UK. See the appendix for details on the surveys. 2 Inequality of voting is measured by simply comparing the mean turnout of citizens who have an education level above and below the median. In order to calculate the indicator we ordered all respondents by educational attainment and split the sample at the median. We then calculated the difference between the mean turnout in these two groups. This measure provides data that can be compared across different surveys with different operationalisations of educational attainment and across different countries with different educational systems. DT ¼ T u T 1 where T u and T l are the mean respective turnout rates of the upper and lower groups. In the following we explain how we calculated the figure on the basis of grouped data, as we have only a limited number of groups of educational levels. Imagine four groups, each with a share of 25 per cent of the total electorate: for example, 25 per cent has no education, 25 per cent only primary education, 25 per cent secondary, and 25 per cent tertiary education. The citizens without any formal education had a turnout rate of 60 per cent; those with primary education 70 per cent; with secondary education 85 per cent; and with tertiary education 95 per cent. We calculated the mean turnout for the two groups in the upper part of the distribution ((85 per cent + 95 per cent)/2 = 90 per cent) and subtracted the mean turnout of the groups in the two lower parts of the distribution ((60 per cent + 70 per cent)/2 = 65 per cent). In this case, the difference was 90 per cent 65 per cent = 25 per cent. However, in most cases the median citizen with regard to educational attainment was not exactly at the boundaries of two adjacent groups, and the educational groups were not of similar size. In this case we had to adjust by weighting 3 T l ¼ T i S i 50 þ T j 1 S i 50 for all cases where median in group i T u ¼ T j

8 K. Armingeon and L. Schädel And T l ¼ T i T u ¼ T j S j 50 þ T i 1 S j 50 for all cases where median in group j. T l = mean turnout of the lower educated group T u = mean turnout of the higher educated group M = Median of educational attainment T i = mean turnout of the group with the level of education i, with i below or equal M T j = mean turnout of the group with the level of education j, with j above or equal M S i = share of educational group i (as percentage of the electorate), with i below or equal M S j = share of educational group j (as percentage of the electorate), with j above or equal M Table 1 depicts the development of the difference between the turnout rates below and above the median citizen by educational attainment over the past 50 years. Two observations stand out: there are remarkable differences between countries with respect to the level of turnout inequality. In some countries inequality in participation is clearly on the rise (Germany, Norway, Sweden, and Italy). In Switzerland inequality dropped during the late 1970s and 1980s and afterwards increased to levels found in the 1970s. Finally, in the UK, Denmark, and the Netherlands there is no long-term increase of educational inequality in voting. This variation comes as no surprise, since this is a descriptive analysis of the extent of participatory inequality over time. One needs to correct for elections that mobilised large parts of the electorate and for those elections that were particularly important to certain social groups as well as for other idiosyncrasies of country and time periods. Aside from these differences between countries and time, the general result is an increase of educational inequality in voting. This is indicated by the mean differences calculated per decade. Can this general finding of increasing social inequality in electoral turnout be replicated if we consider the effect of social position measured by income category? While income inequality is an obvious indicator of social position, information on household income is difficult to measure in a reliable and valid way. Survey questions on income tend to be plagued by a particularly high rate of refusal and therefore the share of missing cases in statistical analyses is very high. Moreover, some doubts remain as to whether respondents know their net household income and if so whether they report these figures honestly. In addition, for longitudinal analyses we need high-quality representative surveys with a large number of respondents, such as the European Social Survey (ESS)

Social Inequality in Political Participation 9 TABLE 1 TURNOUT INEQUALITY IN EIGHT COUNTRIES, 1956 2009 (DIFFERENCE IN AVERAGE TURNOUT BETWEEN THE GROUPS WITH HIGH AND LOW EDUCATIONAL ATTAINMENT) Year Denmark Germany Italy Netherlands Norway Sweden Switzerland UK Mean per decade 1956 1.58 1.58 1961 1.65 0.84 1964 1.61 1.80 1965 0.01 2.02 1968 1.79 1969 0.13 0.98 1970 0.73 3.34 2.13 1971 0.81 2.83 15.32 1972 0.76 16.69 2.87 1973 3.06 2.15 1974 3.86 1975 15.51 1976 0.20 1.03 1977 1.93 2.34 2.85 1979 1.41 0.38 7.51 5.66 1980 1.69 2.00 1981 3.72 1.82 2.42 1982 4.65 3.96 1983 0.58 3.43 0.80 1984 1.31 1985 0.47 1.51 1986 1.95 1987 4.44 0.19 3.22 3.32 2.89 1988 4.92 0.76 1989 3.04 2.34 (Continued)

10 K. Armingeon and L. Schädel TABLE 1 (Continued) Year Denmark Germany Italy Netherlands Norway Sweden Switzerland UK Mean per decade 1990 5.80 2.70 4.06 1991 2.19 3.97 1992 0.47 2.21 1993 3.84 1994 2.78 6.38 2.33 5.32 2.32 1995 10.64 1996 0.57 1997 3.37 2.23 1998 1.22 2.89 5.02 4.55 1999 14.36 2001 0.47 4.39 6.55 4.94 2002 2.89 1.82 4.19 2003 2.45 10.31 2005 0.29 6.77 8.04 2006 8.09 2.00 2007 0.50 13.94 2008 1.67 2009 5.80 9.76 Sources: See Appendix Table A1. Turnout inequality: The difference of average turnout in two groups of the sample defined by their educational attainment. The samples were split at the median respondent by educational attainment. The indicator of turnout inequality is the difference between the average turnout in the group above and below the median. For weighing procedures see section 3.

Social Inequality in Political Participation 11 (http://www.europeansocialsurvey.org/) or the surveys of the International Social Survey Programme (ISSP) (http://www.issp.org/). Although we do not empirically base our argument on the effects of income on participation, such an analysis may serve as a useful secondary test in terms of whether we find the same types of statistical relationships as found for the effects of educational attainment on participation. The following analyses are based on the cumulative European Social Survey (2002 2010) and on the Political Action Study. The Political Action Study was conducted in the mid-1970s by a team led by Samuel Barnes and Max Kaase (available from http://www.icpsr.umich.edu/icpsrweb/icpsr/ studies/7777 and from http://info1.gesis.org) and it is one of the first high-quality comparative surveys of political behaviour that covers many countries. The ISSP surveys start only in the mid-1980s and are therefore not suited to compare developments over a long period of time. Arguably, the ESS surveys belong to the highest quality surveys and have particularly when cumulated for various waves a large number of respondents. A bivariate logistic regression of participation in the last election on the categories of net family income could be done for seven European countries, which were included in the Political Action Study (mid-1970s). In five cases (Austria, Germany, Finland, Great Britain, and the Netherlands) the coefficient had either an unexpected sign (i.e. lower income groups participated more than high income groups: Austria, Finland, Britain) and/or the coefficient was nonsignificant at the p < 0.05 level (Germany, Great Britain, Netherlands). Only in the case of Switzerland was the coefficient significant and in the theoretically expected direction. Replicating this analysis using the recent five cumulative ESS waves (2002 to 2010) for these countries yielded coefficients that were both significant and in the expected direction. Only in the case of Switzerland do we find that the likelihood of participating in elections increases significantly with income category and with the same magnitude as 30 years before, when the Political Action survey was fielded. A similar result was found by Gallego (2007: 14). Therefore, as in the analysis of effects of educational attainment, there is strong evidence that the political effect of income groups has generally increased over the past decades. The lower social strata tend to withdraw more from politics. Given the general development of decreasing overall turnout, democracies are increasingly representing the upper classes, silencing the poor and less educated. The Asymmetrical Effects of Individualisation on Electoral Participation In this section we test the hypothesis that the decline of social integration and social control in modern societies has deeper consequences for the electoral participation of lower as compared to upper social classes. The macro-data for voting support this explanation. In the long term, turnout has been declining and is accompanied by declining social integration

12 K. Armingeon and L. Schädel (Gray and Caul 2000). At the same time, social inequality in participation is on the rise and can most likely be explained by the socially asymmetrical effects of the decline in the importance of cues. These cues have become increasingly less prevalent due to reduced societal integration in networks as well as the declining social control exerted by social milieus and organisations. These reported correlations on the macro-level between social integration, social inequality, and turnout may however have other causes; therefore, we need direct individual-level evidence that citizens at the bottom of the social hierarchy are more dependent on social integration than their more advantaged counterparts. In order to show this effect, we specified an individual-level model estimated on the basis of the five cumulative waves of the European Social survey. 5 We selected all Western European countries for which surveys are available with the exception of Belgium and Luxembourg, which have systems of compulsory voting. 6 While a weight was introduced for design effects, no weight was applied for variance in the size of the sample or the frequency with which a country participated in the five waves of the ESS. The resulting data-set is a pooled set of individuals; international macro-variations are taken into account by controls for country-specific and period-specific characteristics (see below). The odds of having participated in the past national election was regressed on an indicator of social integration, an indicator of highest educational level, an interaction term of these two variables, and on a vector of control variables, as well as country and time dummies: PR (vote = 1) ¼ logit a þ b 1 education þ b 2 social integration þ b 3 education where social integration þ c i X n þ h i X o i¼1 i¼1 year of ESS wave þ control variables þ d i X m i¼1 countries vote: whether respondent participated in last national election, 1: yes; 0: no; education: highest level of education, ranging from 1 5 (rescaled to 0 1); social integration: a composite measure of social integration (rescaled to 0 1); control variables: age, gender, place of residence (1 big city to 5 rural), an indicator of political knowledge, level of turnout in respondent s country. The variable social integration is a composite indicator and is the sum of the following variables, each coded as 1 if yes and 0 if no. The respondent (1) lives with a partner; (2) is a member of a trade union; (3) attends church at least once a month;

Social Inequality in Political Participation 13 (4) feels close to a political party; (5) meets with friends, relatives, or work colleagues more once a month. These individual items indicate social integration. The reason for constructing a single indicator out of five items is that several individuals who might otherwise be very well socially integrated may have no chance to be integrated in one specific social network: e.g. employers cannot join trade unions, atheists will not attend church. The indicator has been rescaled from 0 to 1. The obvious problem is that while party identification is a particularly strong indicator of the feeling of belongingness to a politically organised socio-cultural group, we also know that party identification is an antecedent of electoral participation. Electoral participation may thus be endogenous to party identification. If a citizen has acquired the psychological membership in a political party and this is the basic meaning of party identification the likelihood of participation without any additional mobilising or facilitating effects of group membership (such as cues, social control, or reduced decision costs) will increase. We address this problem in a pragmatic way and estimated the model with both the indicator including party identification and with an indicator omitting party identification. In the latter case, party identification was introduced as a control variable, thereby introducing a very conservative test of our hypothesis. Age and gender are standard control variables in any model specification of political participation. Place of residence (re-scaled to 0 1) indicates aspects of social control. The indicator of political knowledge is a combined index of political knowledge and measures the extent to which the respondent follows politics on TV, the radio, and in newspapers. The variable was constructed by calculating the mean time spent watching, listening to, and reading about current events, ranging from 0 (no time spent) to 7 (more than 3 hours). Finally, the national level of turnout was integrated in the model, since higher levels of turnout tend to be characterised by less socially stratified participation (Gallego 2007; Schäfer and Streeck 2013; data on turnout are from Armingeon et al. 2013). Model 1 contains all control variables. In model 2 the variable for social integration was calculated with the omission of party identification, which was entered as an additional control variable. The area of residence could restrict the opportunity for non-integration or it may be correlated with social control. In small towns, for example, it may be difficult to avoid integration into some form of social network, and neighbours might notice if you do not show up at the voting booth. Therefore, in model 3 the area of residence was not included as an independent variable. Age of respondents and the measure of political knowledge time spent following the news on politics on TV, radio, and in newspapers are significantly correlated (about 0.30 in the full sample). This finding makes sense, as young people have on average less interest in politics. On the other hand, older people may spend more time watching TV and reading the newspaper. The knowledge variable may also capture some lifestyle

14 K. Armingeon and L. Schädel and life cycle effects and younger respondents may still follow news and current events but are likely to do so online. The analysis was therefore replicated in model 4 without the knowledge variable, yielding substantively the same results. Note that dummies for the year of the survey and country dummies, although included, are not shown in Table 2. The basic finding across all models provides support for our main argument: A lack of social integration affects less educated people s probability of political participation to a greater extent than people with higher levels of education. The effects are smaller when party identification is not considered to be an indicator of integration, but rather as an independent control variable. In this conservative test the coefficients remain highly significant and in the expected direction. Figure 1 visualises the marginal effects, with the upper and lower lines indicating the 95 per cent confidence interval. The graph was calculated on the basis of model 2 and demonstrates highly significant findings: The graph shows that the effect of education on turnout is considerably higher at low levels of social integration. The marginal effects were calculated by setting the independent variables at their means; the alternative calculation of average marginal effects leads to substantially the same conclusions. For an individual at the bottom of the educational hierarchy (0), a one unit increase in the indicator of social integration (which varies between 0 and 1) implies an increase in the probability to vote by more than 30 per cent if she/he is fully integrated in a socio-political group. By contrast, the probability to vote for an individual at the top of the hierarchy increases only by about 10 per cent if reaching the maximum values of social integration. The political participation of the less educated depends particularly on membership in social networks; and once these network affiliations decline, the lower classes tend to withdraw from politics. In fact, the decline of social networks is a characteristic of the process of societal modernisation, which includes the process of individualisation (Gray and Caul 2000). We tested for the robustness of these findings. First, we replaced the variable participation in national elections with the answer to the question How difficult or easy do you find it to make your mind up about political issues (ranging from 1 very difficult to 5 very easy ). If the argument regarding social control and cues is valid, we should be able to observe results similar to the voting analysis. As model 5 in Table 2 shows, this is the case. The marginal effects are displayed in Figure 2. 9 Our estimations in models 1 3 include country dummies, since our analyses include information from surveys covering only 15 countries. According to most methodological research, an N this small at the second level is much too small to estimate a robust multi-level (ML) model (Stegmueller 2013), although some authors use a similarly small number of countries in their analyses (Steenbergen and Jones 2002). Replicating model 1 as a ML model does not however alter the substantive results.

Social Inequality in Political Participation 15 TABLE 2 TURNOUT AND DIFFICULTY TO MAKE UP ONE S MIND ABOUT POLITICAL ISSUES AS A FUNCTION OF EDUCATION AND SOCIAL INTEGRATION (1) (2) (3) (4) (5) Voted last Voted last Voted last Voted last Difficult to make mind up; national election national election national election national election 1 difficult, 5 easy Education, 1.56 *** 1.41 *** 1.38 *** 1.45 *** 1.32 *** 5 levels (0.06) (0.06) (0.06) (0.06) (0.04) Index of social integration (including party identification) Interaction index of social integration* education (including party identification) Index of social integration (excluding party identification) Interaction index of social integration* education (excluding party identification) 3.37 *** (0.08) 0.93 *** (0.13) 1.81 *** (0.07) 1.86 *** (0.07) 1.85 *** (0.07) 0.36 *** (0.05) 0.59 *** (0.12) 0.61 *** (0.12) 0.64 *** (0.12) 0.39 *** (0.08) Party identification 1.16 *** (0.02) 1.16 *** (0.02) 1.18 *** (0.02) 0.54 *** (0.01) Age 0.03 *** (0.00) 0.03 *** (0.00) 0.03 *** (0.00) 0.03 *** (0.00) 0.00 ** (0.00) Gender (1 male, 0 female) 0.03(0.02) 0.02(0.02) 0.01(0.02) 0.01(0.02) 0.69 *** (0.01) Domicile (1 Big city 5 Rural) 0.06 *** (0.01) 0.08 *** (0.01) 0.08 *** (0.01) 0.04 *** (0.01) Indicator political knowledge 0.16 *** (0.01) 0.15 *** (0.01) 0.14 *** (0.01) 0.22 *** (0.01) Turnout last national election 0.01(0.00) 0.00(0.00) 0.00(0.00) 0.00(0.00) Constant 2.54 *** (0.40) 2.35 *** (0.40) 2.14 *** (0.39) 2.22 *** (0.39) Cut 1Constant 2.12 *** (0.13) Cut 2 Constant 0.09(0.13) Cut 3 Constant 1.46 *** (0.13) Cut 4 Constant 3.60 *** (0.13) Observations 124,035 124,035 124,244 124,554 109,833 Pseudo-R 2 0.15 0.16 0.16 0.16 0.07 Notes: Standard errors in parentheses. *p < 0.05; **p < 0.01; ***p < 0.001. Coefficients for country and year dummies included but not displayed. Database: Cumulative ESS waves 2002 2010 for 15 countries: Austria, Switzerland, Germany, Denmark, Spain, Finland, France, Great Britain, Greece, Ireland, Italy, Netherlands, Norway, Portugal, and Sweden.

16 K. Armingeon and L. Schädel FIGURE 1 MARGINAL EFFECT OF SOCIAL INTEGRATION ON VOTING AT VARIOUS LEVELS OF EDUCATIONAL ATTAINMENT (ALL COUNTRIES (MARGINAL EFFECTS AT THE MEANS)) FIGURE 2 MARGINAL EFFECT OF SOCIAL INTEGRATION ON EASE OF ABILITY TO MAKE UP ONE S MIND ABOUT POLITICAL ISSUES AT VARIOUS LEVELS OF EDUCATIONAL ATTAINMENT (ALL COUNTRIES, MARGINAL EFFECTS AT THE MEANS, DEPENDENT VARIABLES CONSIDERED TO BE INTERVAL SCALED) The results furthermore hold when logistic regressions were run countryby-country. There was always a similar significant marginal effect of social integration (index with party identification) on political participation, with the exception being the Italian case, where the effect of social integration on

Social Inequality in Political Participation 17 political participation at the highest level of education was no longer statistically different from 0 (p < 0.05). Finally, the indicator of social integration brings together very different aspects of belonging to social groups ranging from partnership, to socialising and party identification, to church attendance and union membership. Robustness tests were run with the single elements of the index of social integration (using otherwise model 2) and did not yield substantively different findings. Only in case of informal socialising were the marginal effects between the groups with the least and most education not statistically different from zero, although the relationship between the marginal effects of informal socialising on voting were substantively the same as in the other analyses. In general, and not surprisingly, the effects of the single components of the indicator of social integration were smaller than the effect of the combined index. The marginal plots of these analyses are included in the appendix. In the preceding analyses we showed that turnout inequality has increased on average in modern democracies. Second, we showed that social integration has a stronger positive effect on the likelihood to vote at a low level of education based on data from 2002 to 2010. But this does not imply that the inter-temporal and international variation of turnout inequality is caused by declining social integration. The available sets of surveys allowed only for the calculation of a coarse measure of turnout inequality. Changing social integration may cause some of this turnout inequality and its development. However, a considerable part of the variation could be caused by differences in the quality and comparability of these electoral surveys conducted between 1956 and 2009; by idiosyncratic elements, such as a particularly mobilising election at a certain point of time, or an electoral campaign which targeted particular social strata. Finally, some variation may also be caused by the changing overall level of education or the fact that those below the median educational attainment in the early 1960s may be hardly comparable to those below the median in 2009, in terms of both years and quality of schooling. All we can say is that, on average, turnout inequality between educational groups increased with considerable variation between countries. We also found that turnout inequality between income groups has increased. The first bit of evidence that the increasing turnout inequality is caused by decreasing social integration comes from descriptive figures on aspects of social integration for which we have information from aggregate and individual data (Table 3). These figures show that there is a general trend toward individualisation, but the magnitude of these changes differs between various dimensions and between countries. A sufficiently specified model that estimates the differences in turnout inequality between two consecutive elections with changes in various aspects of social integration and changes in relevant control variables would be too demanding for the available data in terms of data quality and number of cases. However, some bivariate correlations between turnout inequality between educational groups and indicators of social integration

18 K. Armingeon and L. Schädel TABLE 3 INDICATORS OF INDIVIDUALISATION, EIGHT COUNTRIES, 1960 2010 Denmark Germany Italy Netherlands Norway Sweden Switzerland UK Crude divorce rate 1960 1969 1.5 1.0 0.5 0.7 1.3 0.9 0.8 2000 2009 2.7 2.4 0.8 2.0 2.3 2.3 2.4 2.5 Difference 1.3 1.4 1.5 1.5 1.1 1.5 1.7 Crude marriage rate 1970 1979 6.3 6.3 6.8 7.7 6.7 5.0 6.1 7.8 2000 2009 6.7 4.7 4.4 4.7 4.8 4.8 5.4 4.8 Difference 0.5 1.6 2.5 3.0 1.9 0.2 0.7 3.0 Respondent lives with partner 1960s/1970s 0.83 0.72 0.76 0.77 0.68 0.72 0.75 1990s 0.71 0.69 0.59 0.71 0.68 0.55 0.65 Difference 0.12 0.03 0.17 0.07 0.00 0.18 0.10 Respondent attends church at least once a month 1960s/1970s 0.11 0.35 0.40 0.19 0.14 0.26 0.26 1990s 0.21 0.26 0.15 0.08 0.14 0.23 Difference 0.15 0.14 0.04 0.06 0.12 0.03 Respondent feels close to a political party 1960s/1970s 0.58 0.57 0.35 0.35 0.63 0.58 0.80 1990s 0.54 0.45 0.20 0.33 0.47 0.43 0.61 Difference 0.04 0.12 0.14 0.03 0.16 0.16 0.19 Trade union density 1960 56.9 34.7 24.7 40.0 60.0 72.1 36.1 40.4 2010 68.5 18.7 35.5 19.3 54.8 68.9 17.2 27.1 Difference +11.6 16.0 +10.8 20.7 5.2 3.2 18.9 13.3 Definitions: Crude divorce rate: The ratio of the number of divorces during the year to the average population in that year. The value is expressed per 1,000 inhabitants. Crude marriage rate: The ratio of the number of marriages during the year to the average population in that year. The value is expressed per 1,000 inhabitants. Trade union density: Number of trade union members in employment as a proportion wage and salary earners in employment. Sources: Marriage and divorce: Eurostat (Demography National data, Mariage and divorce), available at http://epp.eurostat.ec.europa.eu (accessed 14 April 2014); respondent lives with a partner, church attendance, closeness to a political party. The European Voter Database. Continuity Files of National Election studies in Denmark, Germany, Great Britain, the Netherlands, Norway and Sweden. Switzerland: Swiss national election studies, cumulated file 1971 2007. Union density: Visser (2011). support the assumption that stronger social integration is accompanied by lower turnout inequality. Trade union density is arguably a reasonable indicator of social integration and for which quality annual data are available (see Visser 2011). The correlation between the extent of turnout inequality (Table 1) and the level of trade union density in this election year in the respective country is 0.25 and highly significant (p = 0.01, n = 94). Gray and Caul (2000) argue that the level of turnout is strongly correlated with social integration. Therefore the level of turnout (see Armingeon et al. 2013) is a reasonable instrument for

Social Inequality in Political Participation 19 social integration. One of the problems with this instrument variable is, however, that the higher the turnout, the lower the likelihood of turnout inequality, since at a turnout level of 100 per cent there cannot be any turnout inequality between social groups. The correlation between turnout inequality (Table 1) and level of turnout is 0.63 (p = 0.00, n = 94). Taking the effect of very high turnout on turnout inequality into account, we considered only cases with lower turnout. We continue to find a strong correlation, although the actual level of turnout is far from from 100 per cent. For example, for all the cases with turnout < 85 per cent, the correlation is 0.59 (p = 0.00, n =59 cases). Looking only at the observations with turnout below 75 per cent, the correlation is 0.63 (p = 0.01, n = 18). Conclusion Socio-economic inequality in electoral participation has risen over the past decades in Western European countries. This is not only due to increased socioeconomic inequality, but also to the fact that the differences in societal position have a larger impact on political participation today as compared to the 1970s. This was the first finding of our descriptive analysis, wherein we re-analysed large representative electoral surveys in eight countries. The measure of social position was inequality in educational achievement; the measure of political participation was electoral participation. Our test was particularly conservative, since the spread in education between the upper and lower half of the sample has arguably become smaller over the past decades and therefore, all else being equal, we would expect a decrease in the social stratification of electoral participation rates. But how can we explain the fact that the participatory gap between those with the lowest and highest level of education has widened? The answer to this question can first be found within the established literature on social integration and participation. The stronger one s social integration and thus the stronger the social control, the greater the likelihood of political participation. Citizens who are socially integrated into groups and networks receive cues and experience social control. Cues reduce the costs of decision-making; social control increases the likelihood of negative sanctions if one fails to follow the group norm of participation. Citizens with abundant resources such as higher education and the related cognitive capabilities are not as relient on cues as citizens with fewer resources. Regarding political participation, the lower social strata in particular benefit from social integration. With the processes of modernisation we have also witnessed a decrease in social integration and social control. Socio-cultural milieus have eroded; psychological (in addition to formal) membership in political parties, membership in interest organisations, and church attendance were receding. These are all signals that indvidualisation within societies has increased. For a poorly educated citizen, the argumentative identification of his or her electoral choice was probably as difficult in the 1970s or 1980s as it is today. But 30 or 40 years