Direct democracy and equality: context is the key

Abstract

Direct democratic instruments are increasingly applied in many European countries. They are subject to an ongoing public—and often highly controversial—debate. The question of how direct democracy relates to equality, i.e. if direct democratic instruments have the potential to foster equality or if they lead to more inequality, is crucial in this debate. Research has struggled to come up with a general answer to this question with scholars assuming equality as well as inequality-promoting effects of direct democratic outputs. In this paper, we try to shed new light on this controversial debate and argue that the impact of direct democracy on equality essentially depends on the social and political context in which the decisions take place. In order to examine the impact of context factors on direct democratic outputs, we first analyze the influence of country-specific variables such as the level of equality and the age of democracy—applying large-N multilevel logistic regressions. Yet, these statistical analyses offer no clear results: the effects of the country-specific variables are rather blurry and hard to make meaning of. The result suggests that other factors than the ones we included in the regressions are decisive for the effect of direct democratic instruments on equality. Therefore, in a second step, we demonstrate how country-and case-specific variables such as social and political characteristics impact the outcomes of direct democracy, using the same-sex marriage referendums in Ireland and Slovenia in 2015 as examples. We conclude that context factors are too complex to be grasped in a large-N, statistical analysis. This means that to understand how context influences the effects of direct democratic instruments on equality, for now, one has to go beyond the lens of statistical analysis and look at the respective cases in great depth and detail.

Introduction

Facing declining participation in elections and political parties as well as growing distrust in representative institutions, claims for enlarging direct democratic options are being made in democracies all over the globe. Likewise, the use of direct democratic instruments has been increasing during the last decades. Political parties from left to right, but also civic organizations advertise for letting the people decide on policies in direct votes. But little is known about the impacts of direct democratic instruments on equality. Theoretical considerations as well as empirical findings are mixed. While popular votes seem to foster equality in some cases (Feld et al. 2010; Frey and Goette 1998; Garry 2013), other cases hint towards the danger of making already influential citizens and groups even more influential—thus leading to increased inequality (Merkel 2011; Merkel and Ritzi 2017; Schäfer and Schoen 2013).

Researchers recently found that there have been more successful direct democratic votes in European democracies between 1990 and 2015 (national level) that aimed at fostering equality than at decreasing it. In detail, the equality-related record of direct democratic instruments seemed to depend on the country groups you look at: especially in Switzerland and Eastern Europe, direct democratic votes often lead to positive outputs when it comes to socioeconomic equality, while legal and political equality seem to be enhanced less often (Geißel et al. 2019a, 2019b).

But what are the relevant context factors for equality- or inequality-promoting outputs of direct democratic instruments? This article aims at identifying, whether there are general context variables that determine pro-equality outputs of direct democratic votes or whether we have to look at singular votes and their specific contexts.

The paper proceeds as follows: first, we define our use of the terms equality and direct democracy and explain why we expect a connection between both. Afterwards, we give an overview over the existing literature on the impact of direct democracy on equality and elaborate on current research gaps. Building on theoretical assumptions as well as empirical findings, we outline the variables that we suspect to influence whether direct democratic votes have pro-equality outputs. We proceed by presenting the dataset and our coding procedure and summarize descriptive findings. The dataset includes all popular votes in European democracies between 1990 and 2015 at the national level. At the heart of this paper lies, first, a multilevel regression analysis investigating the decisive context factors of direct democratic votes for having pro- or contra-equality outputs. Independent variables include the level of socioeconomic, legal and political equality, the age of democracy, and the direct democratic instrument employed.

Our multilevel regressions do not offer the general explanations we hoped for. They could not detect, why sometimes (and somewhere) direct democratic outputs were pro-equality and at other times (and places) they were contra-equality. On the contrary, we get confusing results that are hard to explain (see below). The missing of general factors explaining the equality-related outputs of direct democratic votes suggests turning to the concrete circumstances of singular cases. Therefore, in the second part of our analysis we illustrate the complexity of context factors based on two direct democratic votes. In order to be able to identify the decisive factors that made a difference, we look at votes on the same topic that took place in the same year, but in different countries and with different outputs. These are the votes on same-sex marriage in Ireland and Slovenia in 2015, resulting in a win for same-sex couples in the first case and in a loss in the second one. The literature points to decisive case- and country-specific factors. Whereas the history of the gay and lesbian rights movement, the campaigning and the Irish Citizen Assembly were important for the success of the minority in Ireland, the role of the Constitutional Court and cleavages between the rural and urban population (amongst other factors) resulted in their defeat in Slovenia. Although the topic and timing of the two referenda were similar, case- and country-specific factors seem to explain the difference in their outputs. Future research should analyze these and other cases more thoroughly to unveil the relevant mechanisms at work.

To conclude, our analyses offer no general explanations for why direct democratic votes end up sometimes in pro-equality outputs and sometimes in contra-equality ones. Whether a popular vote provides a pro- or contra-equality output seems to depend on the actual context of each vote. As a result, there is also no definite answer whether direct democracy is a curse or a blessing for issues of equality—depending on the context both scenarios are possible.

Theoretical considerations

Two important terms have to be conceptualized before presenting the state of the art and proceeding with our own analyses: equality and direct democracy. Furthermore, we have to outline why we expect direct democracy to have an impact on equality.

First, we understand equality as closing the gap between disadvantaged (A) and well-off groups (B) with regard to a certain benchmark (X) (see also Alexy 1986; Altwicker 2011; Westen 2016; Rawls 1971). In order to close this gap, groups might also be treated unequally—those who are worse off might get benefits at the expense of those who are better off (Altwicker 2011; Sartori 1992). Therefore, we consider direct democratic bills that aim at closing the gap and thereby making society more equal—either by proposing equal or unequal treatment of certain social groups—as pro-equality (cf. codebook in Appendix).

To investigate the effects of direct democracy on equality in detail, we include different equality dimensions in our analysis. While several studies focus on aspects of socioeconomic equality, others are more interested in the implications for minorities, entailing aspects of legal and political equality (see State of the Art below). We include the three dimensions of equality and define them in the following way:

• 1. Socioeconomic equality: equality regarding the socioeconomic status (aspects such as income, education, health, or property)

• 2. Legal equality: equality regarding the legal status of the inhabitants of a country

• 3. Political equality: equality regarding the scope of political influence (especially of minority groups).

Second, we define direct democracy as popular votes on specific issues. Thereby, we exclude elections or recalls of politicians. We differentiate between bottom-up votes (initiated by citizens), top-down votes (initiated by government or parliament) and mandatory ones (prescribed in the constitution).

In our study, we define successful direct democratic votes as votes that were adopted, i.e. won the majority of the votes and passed a possible quorum, i.e. foster equality directly. Of course, bills can have impacts on equality even when they fail at the ballot or are withdrawn by their initiators before coming to a vote. However, these indirect effects of direct democracy are hard to grasp in quantitative analyses and would require a different empirical approach, e.g. applying expert interviews. Therefore, we focus on the analysis of successful direct democratic outputs, i.e. adopted bills, and future research might examine indirect effects.

Third, there are different theoretical considerations as to why there are relations between direct democracy and equality. Two examples for potential relations might suffice here, more detailed discussions are provided below. Regarding socioeconomic equality, it might be assumed that in countries with high income equality, the number of ‘have-nots’ is rather small and the support for more welfare spending and income redistribution might be modest. But it could also be the opposite: citizens in such countries might favor even more welfare spending and income distribution, because they see many benefits of income equality. Also considering countries with high income-inequality, contrasting expectations are possible. Since the group of ‘have-nots’ is probably bigger, more welfare spending and income redistribution might find more support. But it is also possible that citizens in these countries endorse liberal instead of social concepts and disdain higher taxes. Another consideration stems from the assumption that disproportionately more well-off and middle-class people might cast a ballot compared to poor people. And the well-off have more resources to push their interests. This would result in direct democratic decisions favoring the better-off and undermining the interests of low SES groups (Merkel 2011; Merkel and Ritzi 2017). Also considering legal and political equality, a variety of considerations are discussed. Several arguments are put forward speculating that direct democratic votes would enhance or diminish legal and political equality.

To sum up, we expect direct democratic votes to have an impact on socioeconomic, legal, and political equality theoretically. The direction of this effect will probably depend on several context factors. We give a brief overview over existing research on respective topics and identify the research gaps that this paper addresses, before we elaborate on the factors we investigate in our analysis.

State of the art

A number of studies has investigated the impact of direct democratic options and their use on several aspects of equality during the last decades. Most of these studies focused either on socioeconomic measures or legal or political minority rights. As direct democracy has a long history in Switzerland and several US states, most studies focus on these countries. Comprehensive, cross-national analyses are almost missing. In addition, the impact of context factors is ignored in most studies, pointing at a large research gap.

Empirical findings on direct democracy’s impact on socioeconomic equality are still rare and rather mixed (Moser and Obinger 2007; Berry 2009; Feld and Kirchgässner 2000; Wagschal and Obinger 2000; Matsusaka 2004; Freitag and Vatter 2006). While Merkel (2011, 2015) bases his contra-equality assessment mostly on anecdotal evidence and assumptions, Berry (2014) indeed finds education spending to be lower and more unequal in US states with direct democratic options than in states without these. The theoretical expectation that resourceful groups use their influence to enforce their interests and prevent redistribution is also partly confirmed in case studies of German subnational direct democratic votes. These studies insinuate that the potentially higher turnout of the better-off can hinder pro-equality impacts of direct democratic instruments (Töller and Vollmer 2013; Schäfer and Schoen 2013). Feld et al. (2010) come to a more nuanced conclusion when comparing Swiss cantons with different direct democratic options: they find social welfare spending to be lower in cantons with more direct democratic options. However, contrary to the pessimistic conclusions about socioeconomic equality that might be drawn from this finding, the amount of social welfare spending had no influence on income equality. The level of income equality was not related to the extent of direct democratic options. Socioeconomic inequality was not affected by direct democracy. Finally, Blume et al. (2009) and Blume and Voigt (2012) detected in their cross-national analysis that the effect of direct democratic options on welfare spending depends on the applied direct democratic instruments: while countries with mandatory referenda spend less on welfare, expenditures are higher in countries allowing citizens’ initiatives at the national level.

Findings for direct democracy’s effect on legal and political equality are mixed as well. Several scholars studying the fate of minorities in the US states discovered that especially the rights of linguistic minorities and LGBTQ citizens can be at risk in direct democratic votes (Gamble 1997; Lewis 2013; Haider-Markel et al. 2007). In contrast, studies on Switzerland arrive at more optimistic conclusions. In Switzerland, the rights of certain minorities are often protected and enhanced by popular vote (Donovan and Bowler 1998; Hajnal et al. 2002; Frey and Goette 1998). The median voter theorem (Black 1948; Downs 1957)—presuming that direct democratic decisions are close to median voter attitudes—seems to provide a useful empirical answer. If the median voter is in favor of the minority concerned, this most likely results in minority-friendly outputs of direct democratic votes. In contrast, if s/he regards the minority as an out-group, its legal and political equality might be endangered by direct democracy (Matsusaka 2004; Töller and Vollmer 2013; Vatter and Danaci 2010). All in all, context factors like the kind of minority and their perception in society influence the popular vote considerably (Helbling and Kriesi 2004; Bollinger 2007; Vatter and Danaci 2010; Christmann and Danaci 2012).

Taking socioeconomic, legal, and political equality into account, Geissel, Krämling, and Paulus (2019a) reveal that between 1990 and 2015, bottom-up national votes in European democracies produced more contra- than pro-equality outputs, while it was the other way around for mandatory and top-down referenda. This contradicts the findings of Blume et al. (2009, 2012) only at first sight, as the negative record of bottom-up referenda exclusively stems from the legal and political dimensions of equality, whereas these referenda show a positive record for socioeconomic equality.

To get a first impression if context does play a role regarding the equality-output of direct democratic votes, a recent article evaluates national-level referenda in European democracies between 1990 and 2015 according to different European regions. Descriptive analyses reveal that socioeconomic equality is fostered more often in Eastern Europe and Switzerland, whereas political and legal equality are more often decreased in these countries. In the other European regions, pro- and contra-equality outputs are more balanced (Geißel et al. 2019b).

These findings point to the importance of context. But which factors are decisive? Research in this regard is scarce and patchy. For example, in their cross-national, comparative studies Blume et al. (2009) and Blume and Voigt (2012) took the age of democracy into account.¹ Other studies looked at indicators such as the level of equality, education, population size, and attitudes—e,g, the share of fundamentalists (Haider-Markel et al. 2007; Donovan and Bowler 1998).

Research on the effects of context variables on the equality-related influence of direct democratic votes is sparse. The literature concentrates on the following factors: (1) the level of socioeconomic, legal, and political equality in a country, and (2) how long the country has been governed democratically. In addition and for the sake of completeness, we discuss (3) how much importance citizens place on equality—although this variable cannot enter our regression analyses due to data problems. We outline the reasons for our selection below.

Besides these country-level factors, the direct democratic instrument employed can also be assumed to play a vital role in determining whether a vote has pro- or contra-equality outputs. Picking up on previous research, we investigate the effects of bottom-up, top-down, and mandatory referenda.

Why are these variables expected to play a crucial role in whether equality is fostered or decreased in direct democratic votes? First, the level of equality in a country is most likely a decisive context factor, as already mentioned above, and contradictory assumptions can be found. One could assume that the probability for pro-equality direct democratic outputs might be higher in countries with low levels of socioeconomic, legal and political equality. For example, socioeconomic inequality might increase grievances and thereby result in more pro-equality outputs. In line with this thought, high levels of equality might decrease the likelihood of pro-equality outputs, because citizens consider further equality-enhancing actions as unnecessary. Yet, it could also be the other way around: in countries with high levels of inequality, citizens might endorse liberal values and reject social ones. And high levels of legal and political equality might be necessary to initiate a successful direct democratic bill promoting equality. Furthermore, in countries with high inequality worse off groups might not possess the financial, legal and political resources to campaign in direct democratic votes. Looking at the voter turnout, in unequal societies the better-off might have considerably more means to push their interests. This could result in more contra-equality outputs (see for example Merkel and Ritzi 2017). As possible effects could run both ways, we refrain from formulating clear-cut hypotheses and instead perform exploratory analyses to figure out whether high levels of equality increase or decrease the likelihood of pro-equality outputs.

Second, the age of democracy is promising in several regards. It is an indicator for “the stability of political rights”, which most likely has an effect on how direct democracy and equality are related (Blume et al. 2009; Blume and Voigt 2012, p. 295). In systems, where rights have been instable and the rule of law is not implemented properly, citizens might consider popular votes on equality probably as less meaningful. However, it could also be the other way round and citizens living in countries with low stability of political rights are more inclined to vote for more equality, when they have the chance in a popular vote. The age of democracy might have also a socializing effect. When citizens are used to exercise power in long-term democracies, they might feel more efficacy to cast a pro-equality ballot in popular votes. But it might also be possible that citizens living in a young democracy are especially keen to take part in a pro-equality referendum. Again, we do not formulate a hypothesis on the direction of the effect and we apply an explorative approach.

Third, the last country-level variable is citizens’ attitudes towards equality. In line with the median voter theorem, we might expect that attitudes prevalent in society determine the results of referenda. The more importance citizens place on equality, the higher should be the probability that direct democratic votes have pro-equality outputs. In contrast to the variables mentioned above, the opposite effect is pretty unlikely: there are no reasons to expect direct democratic outputs leaning towards contra-equality in countries where citizens value equality to a high degree.

Fourth, we include the direct democratic instrument employed. Theoretical considerations as well as empirical findings on equality-promoting impacts of different instruments seem to be contradictory. Considering bottom-up instruments, on the one hand, we could expect them to be especially positive for legal and political equality, as they give marginalized groups the opportunity to put their interests on the agenda (Eder and Magin 2008). Similar arguments can be made regarding socioeconomic equality: bottom-up votes could be used by the poor to enhance their situation. On the other hand, bottom-up instruments could also be used by wealthy, dominating groups to suppress legal or political minorities and to increase their status and wealth, which is why bottom-up instruments might decrease equality (Gamble 1997; Vatter and Danaci 2010; Blume et al. 2009; Blume and Voigt 2012). Considering mandatory instruments, no assumptions are possible. Mandatory instruments often refer to changes in the constitution and thus more often to legal and political issues than to socioeconomic equality. But findings on equality related outputs are missing. Finally, we have no definite expectations regarding top-down referenda: governments might either use direct democratic instruments to improve their reputation by decreasing or increasing existing inequalities or by legitimizing the status quo via popular vote. Or governments might want to boost their popularity or answer to pressures from minority groups or the European Union by proposing equalizing measures. Our explorative analysis will show which direction of effect holds true.

In sum, we consider levels of socioeconomic, legal, and political equality, the age of democracy, attitudes towards equality, and the direct democratic instrument employed as crucial context factors. As a result of contradicting theoretical arguments as well as empirical findings, we refrain from formulating hypotheses regarding the direction of these variables’ effects on the equality-output of direct democratic votes.

Besides these rather general factors, which can be analyzed quantitatively, more specific circumstances of a popular vote might be decisive for a pro- or contra-equality output, for example, the role of the church or trade unions in the country, or specific events in the run-up to the referendum. These factors are difficult to consider in regression analyses. Therefore, we investigate them separately in a preliminary case study of two cases (see also Methodology and Data).

To sum up, we identify three research gaps: first, research on context variables influencing the effect of direct democratic votes on equality is scarce and patchy. Second, with few exceptions, existent studies focus on single countries, either doing case studies or comparing Swiss cantons or US states. Third, most of the literature only investigates direct democratic options and/or the numbers of referenda without looking at the concrete content of the bills. We address these research gaps by investigating the effects of context variables on the probability of successful direct democratic votes with pro-equality outputs in comparative perspective looking at the actual output. To account for votes in one country being more similar to each other than votes from different countries, we undertake multilevel logistic regression analyses. The methods employed for coding the outputs and for our analysis are described in the next section, which also entails descriptive findings.

Methodology and data

The database for our regression analyses are all successful direct democratic bills at the national level in European democracies between 1990 and 2015.² We coded their outputs according to the proposed measures aiming at increasing or decreasing socioeconomic, legal or political equality. Outputs are considered “pro-equality” if they aim at closing the gap between clearly disadvantaged and better-off groups by giving the greatest benefit to worse-off or diminishing benefits to better-off groups. In contrast, outputs are considered “contra-equality” if they aim at further increasing the gap by giving benefits to better-off groups or denying them to worse-off groups.

This coding procedure is straightforward. Considering socioeconomic equality, pro-equality outputs enhance the live of worse-off groups which are disadvantaged in terms of education, income, welfare, healthcare, housing or similar fields. Considering legal equality, pro-equality outputs give rights to groups which do not enjoy the same rights as other groups. Examples include long-term residents without citizenship or same-sex couples in countries where same-sex marriage or adoption are not possible. Pro-equality bills on the legal dimension enhance the rights of these groups. Regarding political equality, political minority groups’ interests and aims differ significantly from mainstream actors and they usually do not enjoy the influence and resources necessary to gain political power. Measures to improve political equality include for example a proportional instead of a majoritarian voting system or increasing public financial aid for parties (see Appendix, Codebook).

In total, 515 bills were voted upon between 1990 and 2015 in European democracies (considered free according to Freedom House 2020). 240 of those took place in Switzerland. One main factor for the varying frequency of direct democratic votes between the countries included is the availability of instruments (see Table 3 in the Appendix). Whereas for example Norway and the Netherlands allow direct democratic votes only on EU issues, Italy, Liechtenstein, and Switzerland provide far more options for such votes and as a result, popular votes occur quite regularly.

To arrive at clear and sound coding decisions, we have to exclude some direct democratic bills due to missing data, the complexity of the proposal, or the topic being direct democratic options themselves. After that, we are left with 379 direct democratic bills for our analysis, of which 131 were adopted, i.e. gained a majority at the ballot and passed a possible quorum. For the assessment whether a bill fosters or hinders equality, we use multiple data sources such as the bill proposal itself, NGO reports, newspaper articles, political science articles and legal texts. Our codebook with the questions guiding the coding can be found in the appendix. Of the 131 adopted bills, we coded the content of 46 bills as related to equality. 19 of these were contra-equality bills, whereas 27 were pro-equality ones. Table 1 displays the coding for the 46 adopted bills.

Table 1 Adopted pro- and contra-equality bills

More than half of the coded direct democratic bills stem from Switzerland. Yet, there are also a couple of votes from Italy, Liechtenstein, Ireland, Slovenia, Slovakia and San Marino, as Fig. 1 shows. This implies cross-national variation and thereby justifies comparative analyses.

Fig. 1

Number of coded direct democratic votes per country, 1990–2015

In addition, Fig. 2 shows a rather even distribution of adopted direct democratic bills related to equality per country. Of course, Switzerland again accounts for more than half of these, but the other equality-related outputs occur in numerous countries. This again suggests a comparative investigation.

Fig. 2

Number of adopted, equality-related direct democratic bills per country, 1990–2015

The descriptive findings mentioned in the State of the Art point to the importance of taking context into account when analyzing the impact of direct democracy on equality. Which of the factors mentioned above turn out to be decisive? To investigate this question thoroughly, we apply multilevel logistic regression analyses. This method is necessary because we analyze individual direct democratic votes nested in different countries, whereby the individual cases are group-dependent on the specific country contexts (Hox 2010; Steenbergen and Jones 2002; Gelman and Hill 2007). In addition, there might be a time factor at work here, because the likelihoods could differ depending on the year a vote takes place. Therefore, crossed-effects models are used, in which the intercepts are grouped by countries as well as by years. Our dependent variable is the probability that a successful direct democratic vote that relates to issues of (socioeconomic, legal, or political) equality has a pro-equality output. It is dichotomous and we therefore use logistic regressions (Long 1997). Taking all equality dimensions into account results in 46 cases, which enables multilevel logistic regressions with meaningful results. However, differentiating between the dimensions of equality results in models with only 22 cases related to socioeconomic equality, 8 cases related to political, and 16 cases related to legal equality. Models with such a low number of cases do not yield very reliable findings. Therefore, we focus on the models including outputs related to all dimensions of equality taken together. The Appendix includes results for single dimensions of equality. A Bayesian instead of a Maximum Likelihood framework is used to reduce biased results in models with small numbers of countries and years (Stegmueller 2013).

Turning to the independent variables, we start with the direct democratic instrument employed in a particular vote. Three dichotomous variables contain the information on whether a vote is bottom-up, top-down or mandatory. The data are taken from search engines such as www.sudd.ch and www.c2d.ch as well as from official sites.

To test the influence of the level of equality on the probability that a equality-related direct democratic vote has a pro-equality output, the country-level variables include measures of (1) socioeconomic, (2) political, and (3) legal equality. First, the level of socioeconomic equality is operationalized using the GINI score (World Bank 2018). Unfortunately, data are only available from the mid-2000s onwards for most countries. In order to be able to nevertheless include votes from the 1990s in our analysis, we therefore use the averages of the available scores between 1990 and 2015 for each country. GINI scores have been rather stable for most of the countries for at least the last 15 years of our analysis. This justifies the use of means scores, although of course we would have preferred to have scores for each year.

Second, the level of political equality is operationalized as the distribution of political power based on the “Political Equality”-index provided by the Varieties of Democracy project (Coppedge et al. 2018).³ In line with the GINI, we take averages of the means of both scores in our timeframe for each country. The scores for both indicators have been relatively stable for most of the countries over time.

Third, legal equality is measured using the “Equality before the law and individual liberty”-index from Varieties of Democracy (Coppedge et al. 2018).⁴ Again, the analyses include the average scores between 1990 and 2015 for each country. Similar to socioeconomic and political equality, legal equality has been quite stable for most of the countries included during our timeframe.

The variable ‘age of democracy’ contains the information on how long a country has been governed democratically, i.e. the number of years between the establishment of democracy and the respective direct democratic vote in a country.⁵

Unfortunately, due to missing data, we are not able to investigate the effect of citizens’ attitudes towards equality. Although data are available for some countries on the importance citizens place on this value in the Eurobarometer 82.3, it is missing for a large number of our cases (European Commission, Brussels 2018). Therefore, we have to postpone this part of the analysis to future research.

As noted above, we turn to a preliminary comparison of two cases in the second part of our investigation. By looking closer at the votes on same-sex marriage in Ireland and Slovenia in 2015, we hope to identify factors that are relevant for the opposite outcomes of the two votes but cannot be quantified and included in a regression analysis. The case selection of two votes on the same topic at the same time helps to keep some context factors constant and to concentrate on the special circumstances decisive for the differences in the outputs. These were the only two cases on the same issue at the same time in our dataset, therefore we opted to scrutinize them more closely. This first comparison will reveal whether case studies that are more detailed are a promising path for future research.

Results

Are there some general factors that determine if direct democratic votes produce pro-equality outputs? In order to answer this question, we present the results of Bayesian multilevel logistic regression models. The dependent variable is the probability that a pro-equality direct democratic bill succeeds at the ballot. Table 2 shows the results of six multilevel logistic regression models, taking all equality dimensions together. As we analyze a full sample of all direct democratic votes on the national level in European democracies 1990–2015, levels of significance are not reported.

Table 2 Bayesian multilevel logistic regressions of equality-related direct democratic outputs being pro-equality (all equality dimensions; mean coefficients with standard deviation in brackets)

First, it is important to mention the differing numbers of cases (votes as well as countries and years) across the models due to missing data on the stepwise included variables. While the aim should always be to hold the number of cases constant to be able to compare different models, this would have meant an even lower number of votes and countries already for the first models. This would have made their interpretation far more difficult. Therefore, we decided to compromise the comparability of the models for a better performance of the single ones.

Second, by concentrating on successful direct democratic votes that relate to one of the equality dimensions, the number of cases, countries and years is already very low in the first two models (46 votes in 12 countries and 20 years), and is further diminished when introducing more independent variables. Keeping this in mind, the results presented in Table 2 should be dealt with caution and taken as preliminary.

The results offer no clear picture and are not easy to interpret. The intra class correlations (ICC) in Model 1 point to low, but detectable variation between countries as well as between years—of the overall variation in successful bills being pro or contra equality, 11% stems from variation between countries and 5% stems from variation between years. This supports using crossed-effects models. As shown in the second and third row, compared to bottom-up votes, top-down and mandatory votes seem to foster pro-equality outputs.

The effects of the levels of equality are less straightforward: while higher levels of political and legal equality seem to increase the probability of pro-equality outputs (Models 4–6), higher levels of socioeconomic equality appear to decrease the probability of pro-equality outputs as soon as political and legal equality are controlled for (Models 4–6). This supports the assumption formulated above that socioeconomic grievances are necessary to motivate pro-equality outputs, but certain levels of legal and political equality are prerequisites for pro-equality bills to be initiated and adopted.

The age of democracy has a slightly positive effect—outputs are more likely to be pro-equality when democracy has been established for a longer time in a country. This supports the arguments formulated above that being used to exercise political power might increase the likelihood of pro-equality outputs. However, when age of democracy is controlled for, higher levels of political equality have a negative effect (Model 6). This suggests that political equality is higher in countries that have been a democracy for a longer time, and that this experience in democracy is the more equality-promoting factor.

Taken together, the analysis does not reveal unambiguous, central factors that determine the equality-output of direct democratic votes. Results for the single equality-dimensions are even more humbling: due to the low number of cases, some of the models cannot be estimated and those that can be estimated offer no clear results (see Appendix). Therefore, we do not discuss them in detail at this point.

To sum up, Bayesian multilevel logistic regressions do not offer us the enlightening insights about influential context factors for equality-fostering direct democratic votes that we hoped for. The causal mechanisms behind some of the calculated effects remain blurry.

Context factors: too complex to grasp

A quantitative statistical analysis does not seem to produce meaningful results with regard to the question of equality fostering—or hindering—outputs of direct democratic votes. Why? We assume that relevant context factors are plainly too case-specific and therefore nearly impossible to be grasped in a quantitative regression analysis. In fact, numerous context factors such as different actors, prevalent values, media coverage, integration of minority groups, and the institutional setting presumably play a role when it comes to direct democratic decision making, and all of these factors interact in very specific ways—depending on the case. In order to illustrate this, we will now briefly look at two cases of same-sex marriage referenda in Ireland and Slovenia, which both took place in 2015 but turned out differently due to different context factors and very case-specific dynamics. As mentioned above the following discussions of two examples cannot yet provide full-fledged case studies but give hints for future research. In-depth case studies would need more space and would require a second article.

The Irish referendum on same-sex marriage took place on 22 May 2015 and was approved by approx. 62% percent of the electorate, resulting in a legalization of same-sex marriage. As the policy involved a change in the constitution, a referendum was mandatory by law. For understanding the result, a large number of context factors have to be taken into account. As Elkink et al. (2016) point out, numerous factors played a crucial role for the result, among them vigorous and active campaigning, value shifts along the deep seated liberal conservative cleavage, generational effects, and the unique deliberative democracy experiment in the form of the Irish citizen assemblies. Other authors such as Murphy (2016), O’Sullivan et al. (2017), or Tobin (2016) point to other context factors. Among them are the historical development of the gay and lesbian rights movement in Ireland, the Irish Constitutional Convention preceding the vote (Murphy 2016), social structure and online campaigning in Ireland (O’Sullivan et al. 2017), or a lack of political determination to tackle the issue in the Irish Parliament (Tobin 2016). In addition, Ryan points to the decreasing influence of the Catholic Church in Ireland and the overall support for the referendum by the government (Fergus 2015).

While these are just some of the major potentially relevant context factors,⁶ the main point here is that all of these factors—under the given political and social circumstances at the time—played into the dynamics of the referendum so that in the end, the direct democratic vote produced a pro-equality output. Including all of these case-specific context factors in a meaningful quantitative large-n analysis is already an enormous challenge itself—and even if it could be done, the dynamics and interactions that these factors develop in a given political and social setting are simply too complex to be calculated in a statistical analysis.⁷

The same-sex marriage referendum in Slovenia took place on 20 December 2015 and was rejected with approx. 63% No-votes. Clearly, this referendum took place under entirely different social and political circumstances than the Irish counterpart. Again, there was a large number of context factors that played a role in shaping the referendum, its course, and its final output. Important context factors were, among other things, issue framing and campaigning (Krasovec 2015), the history of struggle of the Slovenian gay and lesbian movement (Kuhar and Ceplak 2016), the role of the Slovenian Constitutional Court, citizens attitudes (Kużelewska 2019) or cleavages between the rural and urban population (Krasovec 2016, p. 238). Again, these are just some of the main context factors, the list is not exhaustive. Rights of same-sex couples had already been put to a vote in 2012, when the conservative group “Civil Initiative for the Family and the Rights of the Children” successfully initiated a veto referendum on parliamentary legislation that granted same-sex partners more rights, e.g. regarding inheritance. Conservative opponents then launched another referendum after parliament adopted a law defining marriage as a “union of two” (Di Bari 2019). What we can see here, again, is that there are numerous, country- and case-specific context factors that play a role for the outcome of the referendum. Similar to the Irish example, these factors do not work isolated, but they produce interactions and certain dynamics within the given political and social frame. For example, in both cases the Catholic Church campaigned against extending the rights of same-sex couples, however with different results (Kużelewska 2019). In order to understand these dynamics, it seems more useful to draw on in depth case analysis, instead of statistical, large-N approaches. Our first comparison of the Slovenian and Irish votes can serve as a point of departure in this regard.

Conclusions

In this paper, we examine the context factors having an impact on direct democratic decisions with regard to equality related issues. In order to do this, we introduced three equality dimension, namely socioeconomic, legal, and political equality and conducted multilevel regression analyses including several context factors, namely levels of equality, age of democracy and different types of direct democratic instruments. Our regression analyses, however, do not reveal any clear results. The statistical results do not provide us with any meaningful insights on whether and how context factors influence the effects of direct democracy on equality. One problem was that the number of direct democratic votes related to equality is relatively low as is the number of countries these votes have taken place in. Although our study covers all popular votes, which took place in European democracies at the national level between 1990 and 2015, the effects we found should be taken as preliminary. Considering the direction of the effects and the causal mechanisms any results remain blurry and confusing.

We assume that this is due to the fact that context factors are plainly too case- and country-specific. In order to further elaborate on this argument, we looked at the two cases of same-sex marriage referenda that took place in Ireland and Slovenia in 2015. It became clear that in both cases there were case-and country-specific factors at work that had a massive impact on the output of the referenda. In Ireland, the historical development of the gay and lesbian rights movement, the Irish Constitutional Convention, and last but not least the Irish Citizen Assemblies were key factors for the pro-equality vote. It is rather obvious, that such specific context factors cannot be considered in a quantitative analysis such as a multilevel regression. In other words, the context of direct democratic votes is (for now) too complex to be grasped in a meaningful statistical, large-N analysis.

What does this mean for further research? We do not argue that, in general, large-N statistical analyses are meaningless when it comes to the question of direct democratic instruments and their impacts on politics and society. However, at least with regard to direct democracy and equality, the context is often so complex that it is necessary to look at single cases in great detail and depth. If we want to understand how direct democratic instruments impact equality, for now, we have to look at case- and country-specific context factors and the dynamics they develop in a given social and political setting. This will only be possible with in-depth case studies or case comparisons of a feasible number of cases. Our first, superficial comparison of the votes on same-sex marriage in Ireland and Slovenia can serve as a point of departure for future, more in-depth case studies. In the case of these two votes, an emphasis might be placed on interactions and dynamics between the Catholic Church and different societal actors and the differences in attitudes towards the LGBTQ + community in both countries. Case studies could also reveal the indirect effects of direct democratic bills that were withdrawn before a vote or were rejected at the ballot but might have been nevertheless influential. This would supplement our focus on successful, i.e. adopted, direct democratic bills. Another interesting comparison could focus on differences in the campaigns, in mobilization and ultimately in turnout between pro- and contra-outputs on similar topics. Finally, more research is needed on the effects of the different direct democratic instruments in different contexts, e.g. whether more or less equal countries prefer specific instruments.

Summing up, direct democratic instruments are neither a curse nor a blessing for equality. Whether direct democratic instruments have pro- or contra-equality effects is not inherent in the specific practice of direct democracy but depends on the context. More research is necessary to detect respective variables.

Appendix

See Tables 3, 4, 5, 6.

Table 3 Overview of available direct democratic instruments in European democracies

Table 4 Bayesian multilevel logistic regression of direct democratic outputs related to socioeconomic equality being pro-equality

Table 5 Bayesian multilevel logistic regression of direct democratic outputs related to political equality being pro-equality

Table 6 Bayesian multilevel logistic regression of direct democratic outputs related to legal equality being pro-equality

Codebook
Dimension	Guiding questions (Code pro-equality if any of these questions can be answered with “yes” except 1.8)
1. Socioeconomic equality	1.1) Does the bill propose measures that will increase income for low SES groups? (i.e., raising the minimum wage, give tax cuts to low income people, raising pensions etc.) 1.2) Does the bill propose measures that make (higher) education more affordable for low SES groups? 1.3) Does the bill propose measures that make healthcare more affordable for low SES groups? Does it lower patient contributions in the health care sector? 1.4) Does the bill propose measures that make housing more affordable for low SES groups? (i.e., raising housing subsidies, expand public housing, etc.) 1.5) Does the bill propose measures that expand social welfare programs? 1.6) Does the bill propose measures that abolish/lower other kinds of fees that are not proportionally rising with income? 1.7) Does the bill propose measures to invest in common goods mainly benefitting low SES groups? (e.g., public transportation) 1.8) Does the bill propose measures that increase the retirement age? (if yes code contra-equality)
2. Political equality	2.1) Does the bill propose measures that strengthen the political voice/powers of (political) minorities? 2.2) Does the bill propose measures that lead to a more proportional composition of parliament? (i.e., get rid of/weaken majoritarian voting procedures, get rid of certain % thresholds for parliamentary elections) 2.3) Does the bill propose measures that increase the media presence of (political) minorities? Does it propose measures against media monopolies of certain political actors?
3. Legal equality	3.1) Does the bill propose measures that give more legal rights to disadvantaged groups? (i.e., allowing same-sex marriage, allowing adoption for same-sex couples, allowing permanent residents without citizenship to vote in elections, etc.) 3.2) Does the bill propose measures that facilitate the way to citizenship? (i.e., for immigrants that are long term residents of the country, for children of immigrants that were born/raised in the country, etc.) 3.3) Does the bill propose measures that give more rights to immigrants/asylum seekers? Does it increase protection against deportation? 3.4) Does the bill propose measures that improve the legal status of foreign residents of a country? (i.e., allow them to buy property, allow them to work in certain professional fields, make them eligible to apply for social welfare programs/unemployment benefits, etc.)