WHO DOES NOT PARTICIPATE IN ELECTIONS IN EUROPE AND WHY IS THIS?: A multilevel analysis of social mechanisms behind non-voting

Voter turnout seems to have been on the decline during the last few decades in most industrial democracies. Looking at the European elections, which are generally characterised by high rates of non-voting, the percentage of people who abstain from voting has gradually risen, from 38 percent in 1979 to 57 percent in 2009.¹ Regarding national elections, the same problem can be observed. In Switzerland the rate of non-voting regarding national parliamentary elections increased from over 30 percent in 1944 to about 55 percent in the late 1990s. In West Germany, non-voting – which used to range between 10 and 20 percent – rose to the level of 20 percent during the 1980s and remained on this level. The United Kingdom has suffered a major and sharp increase to 40 percent in the 1990s.² Although these rising non-voting rates must be interpreted regarding country-specific events, as well as characteristics of the electoral systems, and although there are also signs that this trend of turnout decline has been interrupted in places when particular campaigns attracted high attention (e.g., the US elections in 2008; the German elections in 2002), these examples symbolise a social problem.

A declining voter turnout is a social problem since a lack of voting may be understood as a lack of democratic representation. Low voter turnouts decrease the legitimacy of the elected government and therefore decrease the degree of acceptance of governmental decisions. Political extremism may be strengthened and the stability of society as well as the quality of democratic civic life (Sheerin 2008) diminished. Therefore, we will ask: who are the non-voters and what characterises them? And how can voter turnout be increased in Europe? To derive some useful solutions for this problem, we employ an empirical perspective looking at individual characteristics of non-voters as well as aspects of the electoral systems in Europe. The data base is the European Social Survey (ESS) which covers 24 European countries. Our theoretical framework consists of a theory triangulation combining assumptions from socioeconomic theories (education), postmaterialist theories (cohort) as well as socio-psychological theories (political interest, political efficacy, trust, satisfaction) and political theories (characteristics of electoral system).

Since issues both on the individual level and on the macro level of society in 24 countries will be analysed, a multilevel design (HLM; Bryk and Raudenbush 2002) is used. A main reason for employing multilevel methods of analysis is the problem of dependent individual observations. The individual voters analysed here are nested in country-specific electoral systems. Therefore, their attitudes and behaviours are not independent, but depend on the country characteristics.

A main problem when analysing non-voting by use of ESS data is that the dependent variable ‘non-voting’ is based on self-reported voting behaviour. Respondents may over-report their voting behaviour due to a tendency to give socially desirable answers. A comparison between the turnout rates in the data set and actual turnout rates indicates that there may have been some over reporting, which may vary across countries. On the other hand, the willingness to take part in a survey might be highly correlated with the probability of voting. However, according to Sigelman (1982), the self-statement can be used without causing major validity problems, since the variables that discriminate between voters and non-voters are empirically the same as the variables that discriminate between actual voters and misreporters. Since our analyses focus on an explanation of non-voting rather than a comparison of percentages of non-voting, the bias caused by such effects of social desirability is considered as reasonable.

First we will explore some theoretical determinants of voting behaviour to deduce hypotheses on non-voting, then we will categorise the political systems of the 24 European countries to give some background information. We will then present the results of multilevel analyses before concluding the paper with a summary and discussion section.

There is a huge debate on the factors influencing non-voting, ranging from motivational factors, across socioeconomic variables to family structure. We will focus on both motivational factors (e.g., political efficacy, political interest, trust, satisfaction) and sociodemographic factors, that is, factors that are not directly of a motivational nature, like education, cohort and gender (Oppenhuis 1995) on the micro level, as well as on institutional factors on the macro level. Although voting is often understood as a political action based on social class, and is therefore strongly related to social class (Heath et al. 1996; Müller 1999; Weakliem and Heath 1999), the concept of social class itself will not be discussed as a determinant of non-voting, since the concept of class voting is linked to the question of how one is about to vote more than to the question if one is about to vote.³

Since an A-P-C analysis – modelling age, period and cohort effects simultaneously (Hadjar and Schlapbach 2009a) – is not possible due to the limited number of ESS data waves, cohort effects may be interpreted both in terms of political socialisation experiences and in terms of effects of the position within the lifecycle (age effect). Cohort is defined as a generation that is characterised by a certain educational level and shared socialisation experiences and therefore has a common general world view (Mannheim 1928/1972). One argument for turnout decline is provided by postmaterialism theory (Inglehart 1977, 1990), where non-voting appears to be a result of value change. Postmaterialists are more distant from traditional political institutions (e.g., parties) and therefore cannot be mobilised to take part in elections. For postmaterialists, participation in elections and sticking to traditional partisan alignments (i.e., class or religion-specific voting behaviour) often are no longer adequate means to deal politically with concerns like the environment, human rights or social equality. Instead unconventional forms of political participation gain importance (Dalton 1984).

Since political orientations and behaviour develop and increase during the lifelong socialisation process, the finding of differences between cohorts may also be routed back to an age effect. Although there is a core phase of political socialisation during late adolescence, knowledge and reflection competences as well as political involvement, political attachments and interests increase over the life cycle (Oppenhuis 1995) – leading to the empirical finding that voter turnout rises with age (Topf 1995; Dalton 2005). According to lifecycle theory, political concerns mature with age – when people raise children, have debts, own property and have to deal with taxes, they are more likely to vote and deal with politics (Strate and Parrish 1989; Russell et al. 2002).

Education may be first of all understood in terms of cognitive abilities, although it is also an indicator for social position and status in the hierarchic stratification of society. Both cognitive characteristics and the higher socio-economic position (e.g., higher income) result in an improvement of participation and shape opportunities in several realms of the private and public life (Baumert 1991: 347) and in particular regarding genuine political competences (Dalton 1984, 1996; Inglehart 1990). More highly educated people have easier access to politics and political issues, as they are characterised by more advanced competences in recognising, understanding and reflecting on political issues (Almond and Verba 1963: 380f; Verba and Nie 1972). Additionally, education is a causal determinant of moral development and responsibility (Lind 2006). Due to this, more highly-educated people have a stronger sense of civic duty to participate in elections (Almond and Verba 1963: 380f). People with a higher level of education also may be socialised in a social environment that supports civic norms of political participation and democracy (Nie et al. 1996).⁴ Although voting seems to be influenced by socio-economic factors on the individual level, Franklin (2004) puts in doubt that there is a link between education and turnout on the societal level. Exploring the Swiss case as an example, he suggests that voting is a habit that is learned rather than subject to social position; the crucial point is not whether or not people are more-highly educated, but whether or not such a voting habit is socialised (Franklin 2004). However, from our perspective, it may be assumed that educated people are more likely than non-educated people to learn the voting habit.

Since the education effect may not be the same for all cohorts, respectively age groups, interaction effects of year of birth by educational level will be introduced into the analyses. The changing composition of the educational groups might close the gap between more highly educated people and less educated people in political attitudes and behaviour. The slightly higher influx of people of working-class origin into post-compulsory schooling may lead to a heterogenisation of the school student population at intermediate and upper secondary schools, while the low-level schools become more and more homogenised in their student composition, since only migrants and children of lower strata remain there (Solga 2002; Hadjar and Schlapbach 2009a). Whereas the intermediate and highest educational groups may become more distant from politics, the lowest educational group may keep its negative distinction. Therefore the voting behaviour of the highly educated may become more similar to that of the low educated.

Findings of previous studies mainly suggest that women are more likely to be non-voters than men (e.g., Dalton 1996). On the other hand, results from some societies (e.g., Denmark, The Netherlands) point to the opposite: in those countries men are less likely to vote (Oppenhuis 1995). These ambivalent findings suggest that there are other factors behind the gender-gap in voting behaviour, like education and social status or profession. When such factors are modelled simultaneously with gender, gender effects disappear (see Norris 1991, regarding the UK). Considering gender differences in motivational factors of voting – like the lower political efficacy and interest in politics of women – and facilitative factors like education (Campbell 2006), women are expected to be non-voters more often than men.

Psychological theories attempting to explain non-voting provide mechanisms that may be situated between socio-economic or temporal variables and voter turnout as the dependent variable. Such theories focus on individuals’ psychological dispositions, like political efficacy, political interest, apathy, alienation or partisan dealignment (Campbell et al. 1954; Sheerin 2008). We will focus on political efficacy, political interest and trust as independent variables. Partisan dealignment, respectively party identification (cf. Verba and Nie 1972), as a prominent concept to explain non-voting, will not be included into the analyses, since alignment and voting behaviour are highly connected. The strong dealignment–non-voting link covers the effect of other important variables and will therefore be excluded.⁵

This concept refers to the degree a person believes in his or her own ability to understand politics and is convinced that individual political action does have an influence on political processes. A lack of political efficacy appears empirically to be a major cause of low voter-turnout – according to classic or contemporary studies by Campbell et al. (1954), Abramson and Aldrich (1982) or Becker (2004). Internal efficacy refers to individual competences, skills and resources to deal with politics (i.e., higher individual control beliefs), external efficacy refers to the individual perception that political institutions ‘are responsive to one's attempt to exert political influence’ (Clarke and Acock 1989: 552; cf. Lane 1959; Becker 2004). Political efficacy is a component that is also theorised in enlightened rational choice approaches on voting behaviour (Becker 2004), since competences are linked to the subjective expected probability of successfully gaining a benefit from political behaviour (e.g., voting). According to rational choice theories on voting, individuals will vote if they believe that their vote is highly likely to make a difference. This belief depends on the structure of political institutions (Jackman 1987), as well as previous positive voting experiences (Clarke and Acock 1989).

Political interest may be defined as the degree ‘politics arouses citizens’ curiosity’ (van Deth 1990: 278). It refers to a conative – and therefore action- and influence-related – component of the attitudinal system, and is thus closely linked to political behaviour including political participation (Breckler 1984). A lack of political interest reduces the willingness for political participation of individuals. If individuals are interested in politics, they are often involved in political discussions with their family and friends, and follow political media, and they are more likely to vote (Sheerin 2008). Lack of political interest as a cause of low voter turnout has already been recognised in a classical election study by Lazarsfeld et al. (1948).

Trust in a country's political system is a part of the individual's evaluation of the political system (Bühlmann and Freitag 2006). Trust ‘reflects evaluations of whether or not political authorities and institutions are performing in accordance with the normative expectations held by the public’ (Miller and Listhaug 1990: 358). If a person does not trust the political system, then the likelihood he or she will participate in any political action (e.g., voting) will be lowered (e.g., Ragsdale and Rusk 1993; Pattie and Johnston 2001; Bélanger and Nadeu 2005; Grönlund and Setälä 2007). Putnam (2000) provides theoretical backing for this assumption, arguing that trust is the main basis of a democratic society: people will only vote if they are confident that the political system is responding in some way to their voting behaviour. Therefore, trust is also linked to the subjective expected probability of the outcome (e.g., political change) as an element of subjective expected utility theory of voting (Becker 2004).

A key factor of political mobilisation is satisfaction. Klages (1984) interprets dissatisfaction with the political system and governmental policies as the main determinant that supports the genesis of political interest and political participation. An increasing perception of problems leads to a decrease of ‘apolitical trust’ in the state – resulting in a higher political interest and participation level and therefore in lower rates of non-voting. In contrast, it could be assumed that people who are highly satisfied with the government and the political system are less likely to be non-voters, because they see voting as their ‘civic duty’ (Goodin and Roberts 1975). Based on the latter consideration and empirical findings showing that satisfaction with democracy decreases the likelihood of non-voting (Grönlund and Setälä 2007), it may be hypothesised that satisfaction with political institutions and politicians has a negative impact on the likelihood of non-voting.

Macro level determinants include four (institutional) issues of the electoral system that may affect political efficacy and voter turnout; existence of a compulsory voting law, disproportionality factor, years of democratic experience and opportunities for direct democracy (e.g., referendums).

Compulsory voting clearly increases voter turnout, in particular in countries where compulsory voting is enforced, which include Belgium and one canton of Switzerland (Schaffhausen). However, countries where voting is compulsory by law, but where this law is not enforced by sanctions (e.g., Luxembourg, Greece, Italy) may also have higher voter turnouts, since the duty to vote is highly internalised by the citizens.

Another macro level factor influencing turnout is electoral disproportionality. Although people in modern democracies are aware of the low impact of their single vote on the election outcome, the shape of the disproportionality of the electoral system determines the actual and perceived impact of the vote – in particular regarding smaller parties. The probability of non-voting should increase in countries that are characterised by a high disproportionality (Jackman 1987). Recent findings suggest that not only the likelihood of voting but also political efficacy is influenced by the disproportionality of a country's electoral system (Karp and Banducci 2008). This indicates that the effect of a country's disproportionality factor on individual non-voting is mediated through political efficacy.

The next macro level factor under consideration is the length of a country's democratic experience. As stated before, a crucial point of non-voting is political efficacy. In countries where the democratic system does not have a long tradition, there is a higher probability that people have a lower political efficacy. The rationale behind this assumption is that a political learning process has to be stimulated in order to raise political efficacy (Madsen 1978). In other words, democratic experience has to be accumulated to enforce democratic behaviour like voting. Non-voting should be more common in countries with a shorter democratic experience.

The voter turnout may also be affected by the extent of participation opportunities. Tolbert and Smith (2005) postulate an ‘educative effect’ of direct democracies: elements of direct democracy in Switzerland may go along with an enhanced civic engagement, more political trust and a mobilisation of parties and interest groups – eventually leading to an increase in overall political activity and therefore increasing voter turnout. According to the categorisation of Gross and Kaufmann (2002); Kaufmann and Waters 2004; cf. The Initiative & Referendum Institute Europe 2005), participation opportunities are highest in countries that are rated as ‘The Avantgarde’ and ‘The Democrats’ (including Switzerland, Italy, Slovenia, Ireland, Denmark, Slovakia and The Netherlands). In such countries, elements of direct democracy are implemented in the institution, frequently used and approved by the parliament. Other European countries are rated as ‘The Cautious’, ‘The Fearful’, ‘The Hopeless’ and ‘The Tail Enders’, because referendums are not common here or are often ignored by governments and parliaments. However, it remains unclear whether these opportunities actually strengthen political efficacy and turnout regarding parliamentary elections.^.6

From the preceding arguments and empirical findings, the following hypotheses are derived which will be now tested empirically. Hypotheses 1 to 7 are based on micro level assumptions, while hypotheses 8 to 11 are concerned with macro level influences on the rates of non-voting in the countries analysed:

The probability of non-voting:

To analyse non-voting behaviour in an international perspective, the European Social Survey (ESS) 2006 will be used. The ESS data is suitable for an international comparison because the measures do not vary between countries. Since the Czech Republic, Greece, Luxembourg and Italy did not take part in the 2006 survey, ESS data-sets of 2004 will be used to represent these countries. All sub-samples were reduced to control for socialisation experiences – to include people with a relatively fixed general educational level and to avoid selection effects regarding very old people. The sample consists of people who were born between 1919 and 1978; that is, aged between 28 and 87 years.

Since not all contributors to the ESS data-set had been able to provide all individuals the same chance of selection and therefore country-specific sample bias is to be noted, the data has to be weighted for the analysis employing a design weight. When the whole sample of all national subsamples is employed, a population size weight will be used to ensure ‘that each country is represented in proportion to its population size’ (Norwegian Social Science Data Service 2005: 1).

The dependent variable ‘non-voting’ was measured by the question: ‘Some people don't vote nowadays for one reason or another. Did you vote in the last national election?’ in regard to the last election of the primary legislative assembly. Possible response categories were ‘yes’, ‘no’ and ‘not eligible for vote’. The variable was dichotomised into the categories ‘non-voting’ (1) and ‘voting’ (0), thereby omitting all the people who had not been eligible to vote at the last election. This measurement is based on self-reporting by the respondents – a problem already discussed in the introductory section.⁷

Regarding the position of people in society and their abilities, educational level is included in the analyses. People were asked for the highest level of education they had achieved. The different educational systems and types of degree only allow a simplified educational variable to be included in the analyses, comprising three categories. ‘Low education’ includes people with no school certificate, those who have completed compulsory schooling or who hold an intermediate school certificate. ‘Intermediate education’ refers to people who hold a university entrance certificate, an upper school certificate or an advanced vocational certificate. ‘High education’ applies to people who hold a tertiary-level certificate, i.e., undergraduate and post-graduate degree (CASMIN level 3a, 3b, see Brauns and Steinemann 1999).

Cohort is included in the models to compare cohort-specific characteristics (e.g., socialisation). On the one hand, for descriptive analyses, cohorts will be bundled into six dummy variables to reduce complexity. On the other hand, to determine a trend regarding non-voting over all birth cohorts, the year of birth will be included. Since no Age-Period-Cohort analysis can be estimated due to the short time period of the survey, the cohort effect may be also strongly affected by age effects that result from the change of social positions within the lifecycle. Gender is also included in the models – category ‘1’ stands for female gender, ‘0’ for male.

Political efficacy was measured by two items regarding internal political efficacy (Clarke and Acock 1989; Becker 2004): ‘How often does politics seem so complicated that you can't really understand what is going on?’ and ‘How difficult or easy do you find it to make your mind up about political issues?’.⁸ The response had to be given according to a Likert scale ranging from 1 to 5. Political interest was measured in the classical way ‘by asking people directly how interested they are in politics’ (Gabriel and van Deth 1995: 395). People had to decide between four categories: ‘not at all interested’, ‘rather less interested’, ‘rather strongly interested’ and ‘very strongly interested’. Since the level of the scale is rather ordinal, for the analyses, this variable was dichotomised into two categories: ‘0’ stands for ‘rather less and not at all interested’ and ‘1’ for ‘rather strongly and very strongly interested’. Trust (mean score) consists of three evaluations regarding trust in the parliament, the legal system and politicians (‘Please tell me on a score of 0–10 how much you personally trust each of the institutions’) following the theory of political trust by Gamson (1968). The factor satisfaction with politics is a mean score that comprises of three items: ‘Now thinking about the government, how satisfied are you with the way it is doing its job?’, ‘On the whole, how satisfied are you with the present state of the economy’ and ‘And on the whole, how satisfied are you with the way democracy works?’. People had to respond according to a Likert scale ranging from 0 (extremely dissatisfied) to 10 (extremely satisfied).

The analysis of non-voting will include macro level factors. Such indicators have been gathered at other sources and matched with the ESS data, whereas (a) the existence of forms of direct participation, (b) compulsory voting, and (c) the question whether a country is an ‘old democracy’ (i.e., democratic system established before 1989) is covered by dummy variables, and (d) the disproportionality index is a metric one. This index, created and theorised by Gallagher (1991; Gallagher and Mitchell 2008), indicates the disproportionality between the distributions of votes and of seats. The disproportionality scores have been gathered regarding the last national election taking place before the ESS 2006.

In Table 1, a list of all countries included in the analysis is presented, which also indicates the country characteristics regarding the analysed macro level influences on non-voting as well as the aggregated non-voting level.

To explain non-voting by micro and macro level variables, HLM models will be estimated. Variables will be included stepwise, since this strategy is more adequate to analyse the genuine effects of the variables at different levels and to explore social mechanisms. Results of random intercept HLM models are shown in Table 2. Since the dependent variable ‘non-voting’ is binary (Bernoulli), odds ratios are calculated. The odds ratios can be interpreted as the impact of a variable on the likelihood of non-voting. An odds ratio above ‘1’ indicates a positive influence of an independent variable on the likelihood of abstention, an odds ratio below ‘1’ a negative influence.

In Model 1, only socioeconomic micro level influences are modelled. Regarding education, people of a high level of education turn out to have the lowest probability of being non-voters. People of an intermediate educational level are characterised by a higher probability of being non-voters, but they still differ significantly from the group of low-educated people who show the highest rate of non-voting. Non-voting increases over cohort succession: whereas the older birth cohorts (1929–1938 and 1939–1948) have a lower non-voting rate, the non-voting rate of the 1969–1978 cohort is double the non-voting rate of the reference cohort (1919–1928). There is no significant gender effect – the non-significant odds ratio indicates a slightly lower non-voting rate for women.

In Model 2, macro level effects are introduced in addition to the micro level influences explored so far. Whereas the existence of instruments of direct democracy as well as the disproportionality index does not seem to have any influence on non-voting, old democracies and countries with compulsory voting show a lower rate of non-voting. When we include the macro level effects, the effects of micro level variables stay nearly the same: the ‘low educated’ and the youngest cohorts turn out to be most likely to have abstained from voting at the last national elections, but with no gender differences.

In Model 3, the dummy cohort variables have been substituted by a (mean-centred) metric cohort variable (year of birth). This is necessary, since interaction effects of cohort (year of birth) and education are also included in the models to analyse the change of the education gap in non-voting regarding consecutive cohorts. Additionally, social psychological variables are included, namely ‘political efficacy’, ‘political interest’, ‘political trust’ and ‘political satisfaction’. Due to the introduction of these variables, some effects lose power. The effects of the newly introduced variables indicate that political efficacy, political interest, political trust and political satisfaction reduce the probability of non-voting significantly. The macro level variables ‘elements of direct democracy’, ‘old democracy’, ‘compulsory voting’ and ‘disproportionality index’ show no significant impact on non-voting. The micro level effects of education also decrease after introducing the social psychological variables, although they remain significant in character. The significant and positive cohort effect indicates that younger cohorts abstain from voting more often than older cohorts. The significant interaction effect ‘cohort · tertiary education’ may be interpreted as follows: among tertiary educated people, the increase in non-voting is stronger than among intermediate or low educated people over cohort succession. Therefore the education gap in non-voting – the higher-educated are less likely to be non-voters – decreases slightly regarding consecutive cohorts. The non-significant and negative gender effect on non-voting becomes significant in the complex Model 3, when social psychological variables are modelled simultaneously. This indicates a suppression effect that often occurs when antipodal indirect effects are modelled which otherwise suppress the direct effect (MacKinnon et al. 2000). Taking into account women's level of political interest or political efficacy, they have an even lower voter-turnout, that is, a higher non-voting level, than expected.

In answer to the question of who the non-voters are, education turns out to be a very important factor in the European countries that have been analysed in this study. ‘Low’ and ‘intermediate’ educated people are more likely to be non-voters. This finding is in line with hypothesis 2a. The theoretical exploration of the link between cognitive ability and voting behaviour (e.g., Almond and Verba 1963) seems to be empirically supported.

Another important predictor of non-voting seems to be cohort. In support of hypothesis 1, members of younger cohorts are more often non-voters than members of older cohorts. Behind these changes regarding the consecutive cohorts may be a value change, a shift from materialist and conventional participation values to postmaterialist and unconventional participation values (Inglehart 1977, 1990). Empirically, suboptimal preliminary analyses using values by Schwartz (1992) in the ESS suggest that people who prefer universalism/prosocial values more often abstain from voting than others, although the significant link between both variables is rather weak (r=0.068).⁹ However, in future analyses both arguments may be tested against each other by including postmaterialist values as well as conventional and unconventional political behaviour in an empirical model.

Since political orientations and behaviour develop and increase during the lifelong socialisation process, the finding of differences between cohorts may also be attributed to an age effect. Although there is a core phase of political socialisation during late adolescence, knowledge and reflection competences as well as political involvement, political attachments and interests increase over lifecycle (Oppenhuis 1995). According to lifecycle theory, political concerns mature with age – when people raise children, have debts, own property and have to deal with taxes, they are more likely to vote and deal with politics (Russell et al. 2002).

Another interesting finding is that the education gap in non-voting decreased over cohort succession, due to a strong increase in non-voting among tertiary-educated people. This may also be an expression of the cognitive mobilisation that most strongly affected people with a higher educational level. The more highly educated became more critical regarding society, the political system and government over cohort succession (Hadjar and Schlapbach 2009b), and maybe also more critical about the electoral system, leading to a higher rate of non-voting. Another possible ad-hoc interpretation may be that this finding is an expression of the changing composition of the student population, namely an increasing heterogeneity of the higher educational group, which now also comprises people from families with no academic background. The greater participation in tertiary-level education among those from the lower classes whose class habitus (Bourdieu 1984) is rather distant from politics, may decrease the overall rate of voting among those people classed as ‘high’ educated (cf. Hadjar and Schlapbach 2009a).

According to the complex analyses (Model 3), a gender effect can be noted: women are more likely to be non-voters if their internal political efficacy, political interest, political satisfaction and political trust are considered simultaneously. This gender difference, which becomes obvious only in the complex model, may be due to the fact that women's right to vote in most of the European countries was introduced later than men's right to vote. This argument follows the thesis of Madsen (1978) that democratic and voting tradition has an impact on voter turnout.

Non-voting is also positively influenced by a lack of motivational or psychological resources (e.g., Campbell et al. 1954): non-voters have a lower internal political efficacy (hypothesis 4), a lower interest in politics (hypothesis 5), a lower trust in parliament, the legal system and politicians (hypothesis 6), and a lower satisfaction with politics (hypothesis 7) than voters. Regarding odds ratio coefficients, political interest and political efficacy seem to be more important than the other social psychological variables.

Macro level characteristics, like elements of direct democracy (hypothesis 10) or the disproportionality index (hypothesis 11), did not turn out to be influential in complex analyses. Whereas in less complex analyses, ‘old democracies’ and countries with a compulsory voting law had a significantly lower non-voting level – in support of hypotheses 8 and 9 – these effects vanished when we simultaneously measured social psychological variables like political efficacy or political interest on the individual level. Maybe these variables function as mediator variables that mediate the influence of the macro level variables on non-voting, a conclusion that supports the findings of Karp and Banducci (2008). People who live in a country with a long democratic tradition in the sense of Madsen (1978) have a higher political efficacy as well as a higher political interest – leading to a lower probability of abstaining from voting. The results of additional analyses modelling country dummy variables on the macro level show that there are huge country differences in non-voting which are all significant. This indicates that the four analysed characteristics of the electoral systems are not crucial for non-voting rates, but other country-specific characteristics might be – which may also be period-specific. Period effects are expressions of societal events, political developments, value climate and societal conditions, both on the structural and ideological level that affect all birth cohorts in a specific country (Hadjar and Schlapbach 2009a). Causes of period effects may be very polarising election campaigns, the introduction of controversial laws, unemployment, or a general distance from politicians after political scandals.

There are some limitations of this analysis: (a) Comparability of educational degrees: since the educational systems of the different countries are all distinctive in terms of stratification and standardisation (Shavit and Müller 1998), the cognitive skills specific to level of education may differ between the countries analysed. ‘Low educated’ people in the Ukraine may not necessarily exhibit the same educational level, and hence cognitive skills, as ‘low educated’ people in Switzerland. (b) Effects of classification: the definitions of educational categories – the effects of classification of country-specific educational certificates – may bias the results. However, the main argument may not be affected, since the three-step educational classification used is rather rough and it is focussed on differences between educational groups. (c) Self-reported voting behaviour: again, it must be pointed out that self-reported rather than actual voting behaviour has been analysed, although this bias may be acceptable for the reasons discussed in the introduction and method sections. (d) As already noted, a separation of cohort and age effects is not possible due to the short time period that is covered by the ESS. Longitudinal data, in particular panel data, that cover long time periods and major parts of individual lifecycles, are needed to determine the actual impact of age (lifecycle), cohort (socialisation) and period (societal events, value climate) influences on non-voting.

Finally, what has to be done to bring people back to the polls? Continuing the educational expansion seems to be a meaningful measure, as well as raising interest in politics and increasing internal political efficacy. Whereas political interest may be increased by election campaigns and a broad media coverage, internal political efficacy may be fortified through education in general, as well as political education and political information. To increase feelings of trust, government politics need to be more transparent and responsive.

Andreas Hadjar, PhD/habil., has been a lecturer in the Sociology of Education Department, University of Bern (CH) since 2004. He will become a professor at the University of Luxembourg in September 2010. Main research interests: sociology of education, political sociology, social research methods and gender. He studied sociology and journalism at Leipzig University (DE) and Glasgow University (UK). From 2000 to 2004, he worked at the Sociology Department, Chemnitz University of Technology (DE). Recent international publications: Expected and Unexpected Consequences of the Educational Expansion in Europe and the US (Haupt, 2009; co-editor), ‘Educational expansion and interest in politics in temporal and cross-cultural perspective’, European Sociological Review 25, 2009 (co-author); ‘The socialisation of hierarchic self-interest?’, Young 16, 2008 (co-author).

Michael Beck has been a research assistant at the Institute of Educational Science, University of Berne since 2007; studied sociology, educational sciences and psychology at the University of Karlsruhe (TH) and University of Mannheim, 1998–2007; Diploma in social sciences at the University of Mannheim in 2007. Research interests: inequality of education, social stratification, political participation.