Socioeconomic Correlates of High Turnout Elections Dr Jill Sheppard School of Politics and International Relations The Australian National University Acton ACT 2601 Australia jill.sheppard@anu.edu.au This paper examines whether political equality (measured as comparable socioeconomic characteristics among voters and non-voters) can be achieved without compelling voter participation, and if so, what contextual factors can produce the highest rates of political equality in voluntary voting systems. This paper analyses data from the Comparative Study of Electoral Systems Modules 1 to 4, using mixed multi-level regression models to compare the role of socioeconomic characteristics in explaining voting behaviour at the individual level across compulsory and voluntary voting systems. The analysis focuses specifically on compulsory voting systems, and those voluntary systems with regularly high rates of voter turnout at national elections. Compulsory voting is regularly lauded for its levelling effect on political equality: by mandating voting, systems can ensure that traditionally underrepresented ethnic and racial, age, gender, and socioeconomic groups have equal voice in the electoral process. Additionally, the imposition of fines and similar punishments for non-voting imposes costs on non-voters. Such punishments are disproportionately costly to low socioeconomic groups. Compulsory voting also increases the likelihood that low-information and low-interest voters will cast incorrect ballots, reducing the quality of the final electoral outcome (in as much as the outcome reflects or diverges from voter preferences). Costs notwithstanding, Arend Lijphart, Barack Obama, and other notable actors have proposed compulsory voting as a solution to political inequality. However, high rates of voter turnout, and associated increases in political equality, can be achieved without compulsion: Malta, Denmark, Sweden, and Iceland all report high rates of voluntary voting at national elections. Paper prepared for presentation at the International Political Science Association World Congress, Brisbane, Australia, 21-25 July 2018
Introduction The body of research examining the effects of compulsory voting on political behaviour invariably finds that enforceable laws mandating than eligible citizens vote increases the rate of voter turnout among those eligible voters. That is, if a state can compel individuals to vote, more individuals than otherwise would vote (Birch 2009; Franklin 1999). Accordingly, compulsory voting laws constitute a silver bullet solution to declines in voter turnout over recent decades (Lijphart 1997). Voters in compulsory voting systems are more representative of general populations than in voluntary systems where voters tend to have significantly higher socioeconomic status (e.g. higher household incomes, rates of employment, educational qualifications, and occupational prestige) than non-voters (Quintelier, Hooghe, and Marien 2011; Leighley and Nagler 2014). Further, compulsory voting produces more stable party systems and more evenly distributed political knowledge than evident in voluntary systems (Sheppard 2015; Mackerras and McAllister 1999). On the other hand, there exists compelling evidence to suggest that compulsory voting can reduce the aggregate quality of votes cast in an election (as low-information and low-engagement voters are mobilised to vote), and affect electoral and policy outcomes in unanticipated ways (Dassonneville, Hooghe, and Miller 2017; Fowler 2013; Singh 2014). Indeed, Brennan (in Brennan and Hill 2014, 93), argues that voters in the lowest quartile of political knowledge are not just ignorant. They know less than nothing They make systematic mistakes on these basic questions [of political knowledge]. A random-answer generator would do better than the bottom quartile of voters. Monkeys pressing buttons would be better. Additionally and relatedly compulsion undermines democratic legitimacy by artificially inflating voter participation rates, while also undermining individuals rights not to vote (see for example Brennan, Chapter 2 in Brennan and Hill 2014). This study examines the prospect of achieving the positive political outcomes attributed to compulsory voting laws without incurring negative outcomes or violating individuals right to abstain from voting. To do so, I compare the socioeconomic characteristics of voters in compulsory voting systems with those of voters in voluntary systems. I narrow the focus to specifically those voluntary systems with similarly high rates of voter turnout as in compulsory systems. In other words, do voters in systems that for a range of reasons that are discussed in the paper facilitate high turnout resemble voters who are compelled to vote elsewhere? Can we replicate compulsory voting without forcing people to vote? I proceed by comparing aggregate voter turnout and the composition of that turnout between voluntary and compulsory systems. I then identify and discuss high-turnout voluntary systems, based 2
on data from the Comparative Study of Electoral Systems (CSES) Module 1, 2, 3, and 4 datasets. Analysing data from compulsory and high-turnout voluntary systems, I compare the socioeconomic correlates of individual-level and district-level voter turnout at elections between 1996 and 2016. Finally, I use multi-level regression analyses to compare the effects of socioeconomic factors educational qualification, household income, and employment status on voter turnout in compulsory and high-turnout voluntary systems. Who votes in compulsory and voluntary voting systems? Research into the effects of compulsory voting laws on voter turnout has largely accepted the proposition that higher aggregate turnout will invariably lead to lower stratification between voters and non-voters (Lijphart 1997). Certainly, given the strong and well-established predictive effects of socioeconomic factors such as education and household income on voter turnout, the assumption is well founded. However, conditions in which to conduct empirical tests remain elusive, short of introducing compulsory voting laws simultaneously across a range of democracies and nondemocracies. Contradictory evidence from recent studies suggests that establishing any true effects of existing laws on socioeconomic stratification between voters and abstainers is difficult. Hooghe and Pelleriaux (1998) argue that, among Belgian citizens, education levels positively predict political interest, which in turn predicts a willingness to vote under voluntary conditions. Their (tentative) conclusion follows that, in the event that compulsory voting laws were abolished in Belgium, those with little education and with lower professional status indicate that they would no longer vote [so] the electoral weight of those with higher status would increase substantially (Hooghe and Pelleriaux 1998, 424). De Winter and Ackaert s (1998) critique of Hooghe and Pelleriaux also finds that compulsory voting lowers turnout stratification in Belgium. Cross-nationally, however, Quintelier et al. (2011, 410) find that a system of compulsory voting is not effective in reducing inequality, since it simply raises the turnout level for all groups within society, without leveling any differences there might be between groups. While compulsory voting does not unequivocally reduce socioeconomic stratification, it does appear to mitigate differences in political engagement between voters and non-voters. Just as Sheppard (2015) argues that compulsory voting makes the distribution of political knowledge more equitable within a voting population, a wealth of studies find significant differences in political knowledge between voters (with high levels of knowledge) and non-voters (with low levels) in voluntary systems (Selb and Lachat 2009; Dassonneville, Hooghe, and Miller 2017). Similarly, although voters and nonvoters regularly report significant differences in rates of partisan identification (with voters more 3
closely attached to parties), compulsory voting laws mitigate that difference by increasing both aggregate rates and mean strength of party identifiers in a jurisdiction (Singh and Thornton 2013). Several studies find that the positive effects of compulsory voting laws on measures of political engagement are concentrated among the least educated voters (Carreras 2016; Sheppard 2015). In sum, where voting is voluntary, voters report higher educational qualifications, higher household incomes, greater occupational prestige, political knowledge and interest, and partisan identification than non-voters. Where voting is compulsory, political knowledge, political engagement, and party identification are spread through the act of compulsion, increasing rates and strength of each among individuals whom we would expect would not vote voluntarily. As the systematic underrepresentation of low socioeconomic status individuals among voters represents a persistent challenge to democracy, it behoves us as researchers to examine means of redressing that underrepresentation. Compulsory voting laws constitute a possible solution, but a heavy-handed one. The following section describes the data and methods I employ in asking whether voluntary systems with high rates of voter turnout have succeeded in reducing disparities between voters and non-voters without legally compelling individuals to vote. Data and methods This study uses data from CSES Modules 1 to 4, merged as a single dataset. The data were collected following 133 elections in 47 countries between 1996 and 2013, and the dataset comprises 258,491 individual observations. The (unmerged) are available from the CSES website. 1 The dataset includes a range of individual- and country-level measures. The CSES includes an ordinal measure of compulsory voting, reflecting the degree of enforcement in any given system: strong, moderate, or weak. 2 Jurisdictions (either national or sub-national) with compulsory voting laws and strict regimes of enforcement and/or sanctions are categorised by a strong dummy (see Table 1). Jurisdictions with less rigorous enforcement regimes, but with some evidence of enforcement, are categorised by a moderate dummy variable. Jurisdictions with compulsory voting laws but no history or regime of enforcement are categorised by the weak dummy variable. Countries with no laws compelling voting are coded as voluntary. 1 http://cses.org/datacenter/download.htm. R code to merge the datasets, recode, and clean missing data are available on request to the author. 2 A small number of elections (Belgium-Wallonia 1999, Estonia 2011, Netherlands 1998 and Thailand 2001) were incorrectly coded in the CSES dataset, and have been recoded manually to reflect actual voting laws. 4
The key dependent variable voter turnout is measured in two different ways. First, I model the effects of socioeconomic resources and compulsory voting on turnout at the individual level. That is, a binary measure of whether the survey respondent (claims to have) voted at the previous general legislative election, or not. While this the most straightforward measure available in the data, best capturing the behaviour I seek to study, it also has two fundamental problems. One, given the parameters of the study (jurisdictions with either high turnout with or without compulsory voting laws), the overwhelming majority of survey respondents I am interested in will have voted in the previous general election. Accordingly, I expect little variance on the key dependent variable. Two, there are well-documented errors inherent to self-reported measures of voter turnout, wherein survey respondents systematically over-report their propensity to vote (Selb and Munzert 2013). Moreover, over-reporting bias is stronger in systems with high aggregate turnout (Karp and Brockington 2005). Second, I model the predictive effects of socioeconomic resources and compulsory voting on aggregate turnout at the smallest available aggregated level in the dataset, namely district-level turnout. This measure likewise incurs two problems. One, it has similarly low variance to the distribution of voters cf. non-voters at the individual level (although this is mitigated somewhat by being a continuous variable, compared to the binary did/did not vote measure). Two, this modelling strategy involves using individual-level predictors to predict district-level outcomes, which is suboptimal. However, the combined modelling strategies (individual-level and district-level outcomes) should minimise the biases inherent to each outcome measure. Figure 1 displays the distribution of elections by district-level voter turnout (as a proportion of the voting age population) within the CSES Modules 1 to 4 dataset (hereafter the dataset ). Table 1: categorisation of elections by compulsory or voluntary voting laws and enforcement Compulsory: strong enforcement Australia 1996, 2004, 2007, 2013 Uruguay 2007, 2009 Belgium (Flanders and Wallonia) 1999, Belgium 2003 Switzerland (Schaffhausen only) 1990, 2003, 2007, 2011 Peru 2000, 2001, 2006 Compulsory: moderate enforcement Brazil 2006, 2010 Chile 1999, 2005 Italy 2006 Thailand 2001, 2007, 2011 Compulsory: weak/no enforcement Greece 2009, 2012 Mexico 1997, 2000, 2003, 2006, 2009, 2012 Albania 2005 Bulgaria 2001 Belarus 2001, 2008 Canada 1997, 2004 Croatia 2007 Czech Republic 1996, 2002, 2006 Denmark 1998, 2001, 2007 Spain 1996, 2000, 2004, 2008 Estonia 2011 Voluntary Great Britain 1997, 2005 Hungary 1998, 2002 Ireland 2002, 2007, 2011 Israel 1996, 2003, 2006 Japan 1996, 2004, 2007, 2013 Kyrgyzstan 2005 Lithuania 1997 Montenegro 2012 Netherlands 1998, 2002, 2006 Norway 1997, 2001, 2005 Portugal 2002, 2005, 2009 Romania 1996, 2004 Russia 1999, 2000, 2004 Serbia 2012 Slovakia 2010 Slovenia 1996, 2004, 2008 South Korea 2000, 2004, 2008 Sweden 1998, 2002, 2006 Switzerland (ex Schaffhausen) 1999, 2003, 2007, 2011 5
Finland 2003, 2007 France 2002, 2007, 2012 Germany 1998,2002, 2005, 2009, 2013 New Zealand 1996, 2002, 2008, 2011 Philippines 2004 Poland 1997, 2001, 2005, 2007, 2011 Taiwan 1996, 2001, 2004, 2008, 2012 Ukraine 1998 United States 1996, 2004, 2012 Figure 1: density of voter turnout rates (as a percentage of the voting age population) in the CSES Modules 1 to 4 dataset by compulsory voting system Where compulsory voting is most strongly enforced, voter turnout ranges from between 29.5 and 99.9 per cent of the voting age. Here, I treat the district with 29.5 per cent turnout (an electoral division at the 2007 Australian general election) as a coding error. No districts report between 29.5 and 61.9 per cent turnout in the dataset, and as Figure 1 shows, the overwhelming majority of districts with compulsory voting report turnout of 80 per cent and higher. Electoral districts with moderately enforced compulsory voting laws report a similar distribution of turnout, ranging from 74.2 to 95.1 per cent. Turnout under weakly enforced compulsory voting ranges from 23.4 to 82.2 per cent, while voluntary systems report turnout between 0.7 (in various districts at the 2009 South African parliamentary election) and 94.6 per cent. For the purpose of this study, I am only interested in those elections at which the distribution of turnout rates is comparable between compulsory and voluntary systems. Accordingly, I need to specify a minimum cut-off in turnout rates, so that only high turnout elections are included in the final analysis. An appropriate cut-off could be identified by quantile or by categorisation by country 6
or jurisdiction as consistently high turnout. However, as the distribution of district-level turnout is distinctly skewed among compulsory voting systems, I propose a cut-off of 80 per cent (see Figure 2). The subsequent dataset consists of 55,574 observations from 50 different elections. The countries included in the final dataset include Australia, Austria, Belgium (Flanders and Wallonia regions), Belarus, Brazil, Canada, Denmark, Germany, Great Britain, Iceland, Italy, Netherlands, Norway, New Zealand, Peru, Philippines, Romania, Sweden, Turkey, Taiwan, Ukraine, and Uruguay. So few weakly-enforced compulsory voting elections remain in the dataset that they are removed; they are too small to comprise a reasonable subsample for analysis. The distributions of individual-level turnout (with little variance) and district-level turnout (with greater variance) in the high-turnout subsample (hereafter the high-turnout dataset ) are shown in Figure 3. Note that the frequency of voters is similar between all three remaining forms of voting (strongly-enforced, moderatelyenforced, and voluntary), but the distribution of district-level turnout varies between the three categories. The relationship between socioeconomic resources and voter turnout is modelled as a linear mixed effects equation, 3 to account for a lack of independence between respondents within each election study. Elections are modelled as random effects. Fixed effects include controls at both the country and individual levels. To control for the expected effects of electoral systems (particularly of Figure 2: histograms of district-level turnout in elections conducted under strongly-enforced and voluntary voting in the CSES dataset 3 The LME4 package in R (Bates et al. 2014) was used for the analysis. 7
Figure 3: distributions of individual-level voters and district-level voter turnout in the highturnout dataset 8
proportional vis a vis plurality systems), district magnitude is included in the model. The model also controls for the number of days between the election and the administration of the survey, to account for any temporal effects of the election event. Regime type is measured by Polity IV scores from the year of the election, as included in the CSES Modules 3 and 4. I have coded Polity IV scores manually for Modules 1 and 2. 4 At the individual level, control measures are selected for model parsimony: gender, age, household income and educational attainment (standardised within each module to account for coding variations between modules) are included in the model. Household income and educational attainment are of particular interest given the research question. I model the differential effects of education and income by level of compulsory voting (including voluntary voting) as interaction effects. Analysis To test whether the levelling effects (i.e. minimising socioeconomic disparities between voters and non-voters) of compulsory voting can be replicated within voluntary voting systems, I model the effects of household income and educational attainment conditional on the strength of compulsory voting enforcement (alongside those control measures discussed above) on both measures of voter turnout. The full regression output is available in the Appendix to this paper. Here, I only present the predicted probabilities of voter turnout (at the individual and district levels) by education and income conditional on compulsory voting. Among high-turnout elections, education has the weakest effect on individual-level voter turnout where voting is both compulsory and strongly enforced (Figure 4). However, educational attainment retains a positive effect on individual-level turnout among all categories of compulsory and voluntary voting, meaning that voters are disproportionately well-educated even in compulsory voting systems. Perhaps surprisingly, the predictive effect of education is strongest in moderately-enforced compulsory voting systems. Given the literature on determinants of turnout under voluntary voting systems, we might expect education to have the strongest conditional effect in voluntary systems. However, these results suggest otherwise. Modelling the same factors to predict district-level turnout both reveals important differences in the marginal distribution of that outcome measure and suggests that individual-level education has no real effect on district-level turnout (Figure 5). There is no significant difference in rates of district-level turnout among the least and 4 http://www.systemicpeace.org/polity/polity4.htm 9
Figure 4: predicted probabilities of individual-level turnout by level of education conditional on compulsory voting Figure 5: predicted probabilities of district-level turnout by level of education conditional on compulsory voting 10
most educated respondents. Figure 5 does, however, reinforce the effect of strongly enforced compulsion in increasing aggregate turnout. Like education, household income shows a (weak) positive effect on individual-level turnout across all forms of compulsory and voluntary voting systems (Figure 6). The differences in the marginal effects of system are not significant, but to the extent that they are evident they reflect the effects of education: the effect of household income is weakest in strongly compelled systems, but weakest among moderately enforced compulsory systems. High-turnout voluntary systems in which we might expect the marginal effect of household income to be strongest, as the costs of voting are concentrated among those with the fewest resources report differences in voter turnout of only four percentage points between individuals in the lowest and highest household income quintiles. The two forms of compulsory systems report differences of three percentage points in turnout between individuals in those quintiles. Finally, Figure 7 shows that, as with education, individual-level household income has no effect on district-level voter turnout in any of the three systems, while reinforcing the effects of strongly enforced compulsion on aggregate turnout rates. Discussion and conclusion Given widespread public and elite opposition to compulsory voting (particularly where voting is currently voluntary), there is a normative imperative to examine whether the positive democratic outcomes associated with compulsion can be achieved without infringing citizens right to choose whether or not to turn out on election day. To that extent, this paper has examined CSES data from elections across a range of democratic, electoral, and legislative systems countries between 1996 and 2013. The quantitative analysis has compared compulsory systems with high turnout voluntary systems, finding that while individual-level educational attainment and household income have (weak) positive effects on individual-level turnout in each of voluntary, moderately-enforced compulsory and strongly-enforced compulsory systems, the marginal effects of those two socioeconomic resources on turnout is strongest in voluntary systems. However, compulsion does not mitigate the socioeconomic bias among voters (cf. non-voters) entirely, and this evidence has suggested that voluntary systems can closely simulate the voter populations of compulsory systems, at least in terms of socioeconomic resources. 11
Figure 6: predicted probabilities of individual-level turnout by level of household income conditional on compulsory voting Figure 7: predicted probabilities of district-level turnout by level of household income conditional on compulsory voting Clearly, such an analysis precludes other factors that characterise and maintain high-turnout voluntary systems. Among such factors include the threshold for voter registration, electoral competition and system, political and social culture, economic development, party system, and legislative settings (Jackman 1987; Blais and Dobrzynska 1998). The CSES dataset allows for consideration of some of these variables, but not all; future research should incorporate at least a comprehensive qualitative discussion of the role of institutional contexts, and at best both discussion and inclusion of such 12
contexts as variables in quantitative analyses. This study has controlled for the effects of district magnitude and democratic status of the country (at one year prior to the election), but accurate measurement of the full range of possible contextual factors remains elusive. Qualitative measurement can redress this problem to a degree, but not fully. Accordingly, interpretation of these and similar results should be mindful of its constraints. The mobilising role of political parties cf. electoral management bodies (particularly when armed with the capacity to sanction non-voters, as in Australia) is also central to high rates of voluntary turnout. Even in voluntary systems such as the United States, immense voter mobilisation efforts have done little to negate the decline in aggregate voter turnout there. Countries with high voluntary turnout, such as New Zealand, tend to have optimal combinations of institutional settings. For instance, New Zealand has a mixed-member proportional electoral system, allowing for both comparative party stability, high proportionality of electoral votes to legislative seats, and a low percentage of wasted votes. It has two strong major parties, with voter bases forged in a period of lower electoral proportionality, and comparatively high rates of partisan identifiers. Elections are held on Saturdays and the NZ Electoral Commission seeks to maximise participation in elections by lowering the costs of voting (for instance with expansive access to pre-election day polling booths). This paper has provided preliminary evidence that the positive effects of compulsory voting on the socioeconomic composition of the electorate can be replicated under voluntary voting laws. However, there is no silver bullet to achieving these outcomes. Where they have been achieved, institutional and political contexts have combined to create Goldilocks style environments: engaging enough that citizens will turn out to vote, but not so momentous that party stability is threatened or citizens start to lose faith in the political system. As always, the benefits of voting and participating in the electoral process more broadly should outweigh the costs. Compulsion balances this calculus by increasing the costs of abstention, without increasing the positive benefits of participation. The next stage of this research will more closely examine the combinations of institutional settings that characterise systems with consistently high voluntary turnout. 13
References Bates, Douglas, Martin Maechler, Ben Bolker, Steven Walker, Rune Haubo Bojesen Christensen, Henrik Singmann, and Bin Dai. 2014. Lme4: Linear Mixed-Effects Models Using Eigen and S4 (version 1.1-7). http://cran.r-project.org/web/packages/lme4/index.html. Birch, Sarah. 2009. Full Participation: A Comparative Study of Compulsory Voting. Tokyo, Japan; New York: United Nations University Press. Blais, André, and Agnieszka Dobrzynska. 1998. Turnout in Electoral Democracies. European Journal of Political Research 33 (2): 239 62. https://doi.org/10.1023/a:1006802916256. Brennan, Jason, and Lisa Hill. 2014. Compulsory Voting by Jason Brennan. Cambridge Core. June 2014. https://doi.org/10.1017/cbo9781139649742. Carreras, Miguel. 2016. Compulsory Voting and Political Engagement (beyond the Ballot Box): A Multilevel Analysis. Electoral Studies 43 (September): 158 68. https://doi.org/10.1016/j.electstud.2016.04.005. Dassonneville, Ruth, Marc Hooghe, and Peter Miller. 2017. The Impact of Compulsory Voting on Inequality and the Quality of the Vote. West European Politics 40 (3): 621 44. https://doi.org/10.1080/01402382.2016.1266187. Fowler, Anthony. 2013. Electoral and Policy Consequences of Voter Turnout: Evidence from Compulsory Voting in Australia. Quarterly Journal of Political Science 8 (2): 159 82. https://doi.org/10.1561/100.00012055. Franklin, Mark N. 1999. Electoral Engineering and Cross-National Turnout Differences: What Role for Compulsory Voting? British Journal of Political Science 29 (1): 205 16. Hooghe, M., and K. Pelleriaux. 1998. Compulsory Voting in Belgium: An Application of the Lijphart Thesis. Electoral Studies 17 (4): 419 24. https://doi.org/10.1016/s0261-3794(98)00021-3. Jackman, Robert W. 1987. Political Institutions and Voter Turnout in the Industrial Democracies. The American Political Science Review 81 (2): 405 23. https://doi.org/10.2307/1961959. Karp, Jeffrey A., and David Brockington. 2005. Social Desirability and Response Validity: A Comparative Analysis of Overreporting Voter Turnout in Five Countries. Journal of Politics 67 (3): 825 840. https://doi.org/10.1111/j.1468-2508.2005.00341.x. Leighley, Jan E., and Jonathan Nagler. 2014. Who Votes Now?: Demographics, Issues, Inequality, and Turnout in the United States. Princeton University Press. http://www.jstor.org/stable/j.ctt4cgcqb. Lijphart, Arend. 1997. Unequal Participation: Democracy s Unresolved Dilemma. The American Political Science Review 91 (1): 1 14. https://doi.org/10.2307/2952255. Mackerras, M, and I McAllister. 1999. Compulsory Voting, Party Stability and Electoral Advantage in Australia. Electoral Studies 18 (2): 217 33. https://doi.org/10.1016/s0261-3794(98)00047-x. Quintelier, Ellen, Marc Hooghe, and Sofie Marien. 2011. The Effect of Compulsory Voting on Turnout Stratification Patterns: A Cross-National Analysis. International Political Science Review 32 (4): 396 416. https://doi.org/10.1177/0192512110382016. Selb, Peter, and Romain Lachat. 2009. The More, the Better? Counterfactual Evidence on the Effect of Compulsory Voting on the Consistency of Party Choice. European Journal of Political Research 48 (5): 573 97. https://doi.org/10.1111/j.1475-6765.2009.01834.x. Selb, Peter, and Simon Munzert. 2013. Voter Overrepresentation, Vote Misreporting, and Turnout Bias in Postelection Surveys. Electoral Studies 32 (1): 186 96. https://doi.org/10.1016/j.electstud.2012.11.004. Sheppard, Jill. 2015. Compulsory Voting and Political Knowledge: Testing a Compelled Engagement Hypothesis. Electoral Studies 40 (December): 300 307. https://doi.org/10.1016/j.electstud.2015.10.005. Singh, Shane. 2014. Beyond Turnout: The Consequences of Compulsory Voting. Political Insight 5 (2): 22 25. https://doi.org/10.1111/2041-9066.12058. 14
Singh, Shane, and Judd Thornton. 2013. Compulsory Voting and the Dynamics of Partisan Identification. European Journal of Political Research 52 (2): 188 211. https://doi.org/10.1111/j.1475-6765.2012.02071.x. Winter, L. de, and J. Ackaert. 1998. Compulsory Voting in Belgium: A Reply to Hooghe and Pelleriaux. Electoral Studies 17 (4): 425 28. https://doi.org/10.1016/s0261-3794(98)00022-5. 15
Appendix: regression model output Base model Education model Household income model Fixed effects Strongly enforced CV -0.03 (0.05) (baseline) (baseline) Moderately enforced CV 0.05 (0.06) 0.02 (0.04) 0.07 (0.04) Voluntary voting (baseline) -0.01 (0.04) 0.04 (0.03) Education (standardised) 0.02 (0.01) 0.01 (0.01) 0.03 (0.01) Household income 0.01 (0.00) 0.01 (0.00) 0.01 (0.00) Education x moderate CV - 0.03 (0.01) - Education x voluntary voting - 0.04 (0.01) - Income x moderate CV - - 0.00 (0.00) Income x voluntary voting - - 0.00 (0.00) Party identification (strength) 0.01 (0.00) 0.02 (0.00) 0.02 (0.00) Age (years) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Female 0.01 (0.01) 0.01 (0.00) 0.00 (0.00) Polity IV (T-1) 0.02 (0.02) 0.01 (0.00) 0.01 (0.00) District magnitude 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Days post-election 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Intercept 0.49 (0.45) 0.61 (0.09) 0.58 (0.09) Random effects - - Election 3.78 (1.95) 0.00 (0.07) 0.00 (0.07) Residual 2.58 (1.61) 0.07 (0.26) 0.07 (0.26) AIC 2296.9 3174.4 3181.1 BIC 2390.9 3292.7 3299.4 Log likelihood -1135.5-1572.2-1575.6 N (respondents) 10208 19671 19671 N (elections) 19 31 31 Multi-level mixed regression analyses predicting individual turnout Cells show coefficient estimates, with standard errors/deviations in parentheses Missing data excluded listwise 16
Fixed effects Base model Education model Household income model Strongly enforced CV 9.83 (1.11) (baseline) (baseline) Moderately enforced CV 2.16 (1.31) -10.71 (1.01) -8.93 (1.25) Voluntary voting (baseline) -8.67 (1.25) -10.70 (1.01) Education (standardised) -0.14 (0.05) -0.33 (0.06) -0.07 (0.04) Household income 0.03 (0.01) 0.05 (0.01) -0.06 (0.01) Education x moderate CV - 0.70 (0.08) - Education x voluntary voting - 0.28 (0.09) - Income x moderate CV - - 0.29 (0.02) Income x voluntary voting - - 0.10 (0.02) Party identification (strength) -0.02 (0.02) -0.04 (0.01) -0.03 (0.01) Age (years) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Female -0.04 (0.03) 0.00 0.02) 0.00 (0.02) Polity IV (T-1) 0.54 (0.37) -0.19 (0.12) -0.19 (0.12) District magnitude -0.01 (0.00) -0.01 (0.00) -0.02 (0.00) Days post-election 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Intercept 60.70 (7.42) 97.39 (2.64) 97.41 (2.64) Random effects Election 3.78 (1.95) 4.45 (2.11) 4.74 (2.11) Residual 2.58 (1.61) 2.72 (1.65) 2.71 (1.65) AIC 38779.8 75746.1 75620.8 BIC 38873.8 75864.4 75739.1 Log likelihood -19376.9-37858.0-37795.4 N (respondents) 10208 19671 19671 N (elections) 19 31 31 Multi-level mixed regression analyses predicting district-level turnout Cells show coefficient estimates, with standard errors/deviations in parentheses Missing data excluded listwise 17