In the following, we will explore how different political decision procedures influence acceptance. Table 1 shows the summary statistics of the collected variables. At first, we will focus on the comparison between direct democracy and political parties, thereby pooling decisions made by SPD, CDU and the Parliament. The average reported acceptance for decisions resulting from direct democracy is slightly higher than decisions made by political parties (average acceptance of 3.5 vs. 3.37). Thus, we can find some evidence for direct democracy leading to greater acceptance at the aggregate level, albeit only weakly significant (p = 0.07 on a two-sided Mann–Whitney u-test). However, whether respondents perceive a decision as “acceptable” or not is influenced not only by the decision mode, but also by the respondents’ opinion on the topic and the decision. Two factors we are focusing on are the respondents’ agreement with the decision and the perceived importance of the topic.
Table 1 Summary statistics for the dataset
Figure 2 reports the average acceptance depending on agreement and importance levels. It reveals that average acceptance increases with average agreement for decisions arrived at by direct democracy (Spearman’s rank correlation ρ = 0.79 with p < 0.0001) as well as with the average agreement for decision made by political parties (ρ = 0.75 with p < 0.0001). Focusing on the importance of the topics the decision was about reveals the first differences between the two decision modes. Average acceptance is not significantly correlated with the importance of the topic in the case of direct democracy (ρ = − 0.01 and p = 0.887), but significantly negatively correlated in the case of political parties (ρ = − 0.13 and p = 0.014). This correlation mostly owes to the stark drop in acceptance for important and very important topics. Comparing the average acceptance levels of only very important topics reveals them to be 27 % higher for direct democracy—a significant difference between the outcomes of the two decisions modes (p = 0.0015 on a two-sided Mann–Whitney u-test).Footnote 4
To control better for these and other additional influences on acceptance we run a series of linear random-effects models presented in Table 2 in the next subsection. Observations are clustered by respondents over three different scenarios and are based on the smaller dataset wherein direct democracy and political parties are compared with regard to their acceptance levels. In the subsequent section, we include the decisions made by expert committees. The regression models in Tables 3 and 4 replicate our previous analyses with the full dataset.
Acceptance of outcomes from direct democracy versus political parties
The dependent variable in all models is the acceptability of the decision to the respondent. Personal agreement with the outcome of the collective choice and the importance of a topic are the most important control variables. We are primarily interested in the variation in acceptance conditional on decision modes and holding personal opinion on the issue constant. In this regard, Model 1 tests whether direct-democratic decisions are significantly more acceptable than decisions made by political parties (the reference group). The Direct Democracy variable is a dummy, which is set equal to 1 if the decision mode is direct democracy and 0 otherwise (i.e., if either SPD, CDU or Parliament was the decision mode). It thus captures the effect of decisions reached by direct democracy vis-à-vis decisions made by political parties. In line with our initial assumption, direct democratic decision procedures do not generate, per se, more acceptance than decisions made by political parties. This follows from the small and insignificant main effect for Direct Democracy. As one would expect, personal opinions on the issue measured by Agreement and Importance influence the acceptability of a collective choice. The more the respondents agree with the decision, the more acceptable it is, and the more important a decision is for them, the less they accept it if they disagree with the choice.
In a next step we analyze how the acceptance of a direct-democratic decision depends on the perceived importance of the issue. In Model 2, an interaction between the variables Importance and Direct Democracy is added. The interaction term is significantly positive, while at the same time the main effect of Direct Democracy turns significantly negative. Whether direct democracy or decisions made by political parties are more acceptable depends on the importance of the issue. For the lowest importance level, acceptance of a decision made by direct democracy is 5.3 percentage points less than for a decision made by a political party (p < 0.008). As importance increases, the acceptance score goes up by roughly 4 percentage points if the decision is made through a direct-democratic procedure instead of a party. Or, conversely, any form of party involvement in the decision-making process reduces the decision’s acceptability by 4 %, for an additional point on the importance scale. Thus, for very important topics, acceptance is 6.5 % greater for decisions made by direct democracy (p = 0.012).
Model 3 demonstrates that this effect is not driven by the subject-matter of the issue at hand. Three different decision scenarios were presented to all respondents: nuclear energy (Scenario 1), school graduation (Scenario 2), and religious education (Scenario 3). While the decision in Scenario 2 generates more acceptance overall than the other two decisions, this has no impact on the size and significance of the interaction effect. In addition, we include two additional control variables in order to check for the robustness of our findings. The Influence Vote term captures the extent of perceived political self-efficacy during the upcoming state-level election: voters who tend to think that the electorate can actually change politics and policies by means of voting for representatives are more likely to accept decisions in general. However, this perceived self-efficacy does not diminish the interaction between importance of the issue and decision mode. Even if voters tend to think that their voting for parties can make a difference, they are more likely to accept direct-democratic decisions if they are important to them.
The Vote Mass Party variable indicates the intention to vote for one of the two mass parties, SPD or CDU. One may argue that supporters of these mass parties may be more supportive of decisions that are made by precisely these parties and less skeptical than other voters even when it comes to important decisions made by these parties. This is clearly not the case; again, controlling for this variable does not alter the coefficient on the interaction effect. Our finding is not conditional on mass party preferences.
Models 1, 2, and 3 impose a linear functional form on the influence of importance; however, Fig. 2 suggests that this might not be true. In Models 4 and 5 we replicate our previous results without imposing a functional form on the importance variable. In Model 4 we include interactions between Direct Democracy and each level of Importance. The coefficients show the impact on acceptance compared to a decision made by a political party for a topic with the lowest level of importance. For that importance level, a decision generated by direct democracy results in reduction of acceptability by 0.27 points, translating into a 5.9 % lower acceptance rate (albeit only weakly significant, p = 0.07). In contrast to this, direct democracy leads to a 7 % greater acceptance rate for the highest level of importance (p = 0.044, determined by comparing the coefficients Direct Democracy = 0 × Importance = 5 and Direct Democracy = 1 × Importance = 5). Again, the model confirms that acceptance declines with greater importance, as demonstrated by the significantly negative coefficients for interactions with importance levels exceeding two. Model 5 confirms the results from Model 4 after adding controls for the scenario, perceived political self-efficacy and intention to vote for one of the two mass parties.
Taking decisions by expert committees into account
While Table 2 contrasts direct democracy with political parties, Tables 3 and 4 presents additional models that contrast direct democracy with representative democracy, that is, decisions made by expert committees are added to the group of representative decision procedures, so the Direct Democracy effect is tested against decisions made by political parties or expert committees. Model 6 demonstrates that direct democracy is significantly less acceptable for issues of low importance, but more acceptable than the reference group of parties and expert committees when important issues are at stake. In other words, this is not just a difference between direct democracies and parties, but more generally a difference between direct and representative democracy. In both decision-making arrangements, parties and expert committees, decisions are one step removed from the electorate, and citizens have less control over it. While for issues with very low importance this seems not to reduce the acceptability of the outcome, it does lower it for issues considered very important.
Table 2 Random effects regression with acceptance of a decision as the depended variable
Table 3 Random effects regression with acceptance of a decision as the depended variable
Table 4 Random effects regression with acceptance of a decision as the depended variable
Figure 2 visualizes these differences between direct democracy and the decision procedures based on intermediaries. While intermediaries perform better in terms of procedural acceptance for decisions of low importance to the respective voter (at importance level 1), direct democracy performs slightly better on average (level 4) and significantly better (level 5) when the issue at stake matters to the voter personally.
The remaining models provide additional checks for validity, omitted variable bias, and the functional form of the impact of importance. Model 7, for example, takes political parties out of the reference group and compares the different party configurations with the expert decision-making effect that is left in the baseline group. Separate effects are included for SPD, CDU and the majority of parties in the parliament. In Model 8 we include interaction terms with perceived issue importance for variables CDU, SPD, and Parliament, as well as controls for the scenarios, perceived political self-efficacy and intention to vote for one of the two mass parties.
Models 8 replicates Model 3 for the full dataset. It shows that the interaction effect between importance and direct democracy is not affected by the introduction of issue scenarios. As in Model 3, the positive effect of voters’ perceived self-efficacy does not change the result. Instead of Vote Mass Party, we introduce two separate control variables this time—Vote SPD and Vote CDU—as there are also separate model terms for SPD and CDU in the model specification. Neither of the control variables changes the main results presented above.
As an additional validity check, we exclude all observations for which the personal opinion of the respondent is strongly positive; that is, we exclude all observations in Model 9 where Agreement = 4 and run the analysis with the remaining observations. We would expect that those who strongly agree with the decision anyway should not have any reason to be dissatisfied with the procedure. Accordingly, the main effect should still hold for the remaining groups and not be driven by this potential artifact. And indeed, the exclusion of these observations does not alter the effect size or p value of the interaction term significantly. In other words: The observed effect results from those who disagree and are overruled and those who only “tend to” agree.
In Models 10 and 11 we again remove the functional form restriction on importance and include interactions between Direct Democracy and each level of Importance. In Model 10 we replicate Model 5 for the whole dataset; again we observe the same effects of direct democracy and importance on the acceptability of political decisions. The acceptance of decisions generally declines with increasing importance of the issue, but it does so at a considerably faster rate in systems with intermediary decision makers. As predicted, the latter seem to be more acceptable in cases where the issue is less important, while direct-democratic decisions attract significantly higher acceptance levels for important decisions. For issues of low importance, direct democracy leads to significantly lower acceptance than political representation (p = 0.042), while for important issues direct democracy leads to significantly higher level of acceptance (p = 0.016). Figure 3 visualizes the marginal effects of direct democracy for each level of importance as featured in Model 10 with all control variables included. In other words, it depicts how large the additional effect of Direct Democracy is in comparison to the other decision modes is for each level of importance. As the confidence intervals indicate, we do not see significant differences for moderate importance levels but we do see that the procedure does make for a significant difference between issues that are not considered important (1) and issues that are considered very important (5), on both ends of the scale. If an issue is considered very important, direct-democratic procedures lead to significantly higher acceptance rates.
Finally, Model 11 includes interactions between SPD, CDU, the majority of parties in Parliament and each importance level (not reported in the table). Note that the coefficients of the interactions between direct democracy and the importance levels are with respect to expert committees and the lowest importance level in this model. The effect demonstrated for the comparison between direct democracy and party decisions can be confirmed for the comparison of direct democracy and expert committees.