WHO WANTS DIRECT DEMOCRACY?

Despite the broad interest in the effects of direct democratic institutions on political outcomes, there is surprisingly little literature on their origins. Most contributions to the literature on institutional origins emphasize the conflicting interests of ruling and nonruling political elites. For example, Smith and Fridkin (Reference Smith and Fridkin2008) argue that legislative competition, the organizational strength of parties, and third-party presence explain why state legislatures choose to devolve institutional power by delegating to citizens the power to amend state constitutions. In this argument, electoral competition and third-party strength are indicators of how easily members of the majority can be forced into supporting reform (see also Scarrow Reference Scarrow1997). In a similar vein, Chacón and Jensen (Reference Chacón and Jensen2020) argue that political elites benefiting from the overrepresentation of their districts oppose direct democratic institutions because such reforms would threaten their ability to further their political interests (for the Swiss case, see Kriesi and Wisler Reference Kriesi, Wisler, Guigni, McAdam and Tilly1999; Leemann Reference Leemann2023). Focusing on elite support for direct democratic institutions, Bowler, Donovan, and Karp (Reference Bowler, Donovan and Karp2002, 747) find that politicians from nonruling parties are more likely to support direct democracy, whereas politicians from the ruling party coalition “are significantly less supportive of expanding the scope of direct democracy in their nation” (see also Gherghina, Close, and Carman Reference Gherghina, Close and Carman2023).

In short, the existing literature on the institutional origins suggests that political groups dominant in the representative arena have little interest in adopting direct democracy, because such reforms reduce their level of political control. In contrast, nondominant groups support reforms that disperse political power. This literature, however, suffers from an important deficit. Little attention has been paid to the fact that direct democratic institutions are often introduced by popular vote (Kölz Reference Kölz2004; Smith and Fridkin Reference Smith and Fridkin2008). Whereas existing research can explain under what conditions legislatures opt to delegate to citizens the power of direct democracy, we do not know which citizens accept this invitation.

The rich and growing literature on popular attitudes toward direct forms of democracy offers important insights into this question. Initially, this literature emphasized that intrinsic motives drive democratic innovations. Intrinsic motives concern considerations of procedural justice or normative conceptions of democracy, which value political procedures as such and not because of the expected outcome effects (Landwehr and Harms Reference Landwehr and Harms2020, 878–9). At first, the debate centered around the question whether highly engaged and educated “critical citizens” or less interested “alienated citizens” are more likely to push for direct democracy (Coffé and Michels Reference Coffé and Michels2014; Dalton, Bürklin, and Drummond Reference Dalton, Bürklin and Drummond2001; Donovan and Karp Reference Donovan and Karp2006; Font, Wojcieszak, and Navarro Reference Font, Wojcieszak and Navarro2015; Norris Reference Norris1999; Webb Reference Webb2013). More recently, this literature has explored whether there is a relationship between populist attitudes and support for direct democracy (Bengtsson and Mattila Reference Bengtsson and Mattila2009; Bowler et al. Reference Bowler, Denemark, Donovan and McDonnell2017; Gherghina and Pilet Reference Gherghina and Pilet2021; Mohrenberg, Huber, and Freyburg Reference Mohrenberg, Huber and Freyburg2021; Werner and Jacobs Reference Werner and Jacobs2022).

In parallel, research has begun to examine instrumental motives for supporting direct forms of democracy. In case of instrumental motives, democratic innovations are interpreted as a “result of a cost-benefit analysis, with individuals preferring one procedure over another if it is likely to produce outcomes that further their material interests or substantial policy preferences” (Landwehr and Harms Reference Landwehr and Harms2020, 880). Similar to the literature on institutional origins discussed above, these contributions emphasize that citizens endorse democratic innovations because of perceived representation deficits and because of the belief that other procedures for collective decision making will bring about the desired outcomes. For example, using survey experiments, these contributions show that respondents are more likely to support direct forms of democracy if they believe to be part of the majority (Brummel Reference Brummel2020; Landwehr and Harms Reference Landwehr and Harms2020; Werner Reference Werner2020). Put differently, these respondents endorse direct democratic procedure because they expect to win in such votes. Other contributions highlight that long-term losers in electoral politics are more likely to favor direct democracy (Gabriel Reference Gabriel2013; Smith, Tolbert, and Keller Reference Smith, Tolbert and Keller2010). Hence, these respondents believe to be inadequately represented in parliament and seek alternative decision making procedures to align political outcomes with their substantial policy preferences.

There are thus interesting parallels between the recent literature on popular attitudes toward direct democracy and the literature on institutional origins, but two main questions remain. First, is there an objective basis for this perceived lack of representation? Or is this perception as well as support for direct forms of democracy the result of a third variable, for instance, an alienation from the institutions of representative democracy? Second, do people who are badly represented in parliament only favor direct democracy if they believe to be ideologically closer to the median voter? Put differently, in their support for direct democracy, do they also consider their own position relative to the median voter?

REPRESENTATION DEFICITS AND SUPPORT FOR DIRECT DEMOCRACY

In this section, we develop our argument to explain popular support for the adoption of direct democratic institutions. We argue that direct democratic institutions find the most support among voters that are inadequately represented in parliament and ideologically closer to the median voter than they are to the median MP.

Electoral systems can suffer from significant biases. Calvo (Reference Calvo2009, 257–8) differentiates between majoritarian and partisan biases of electoral systems. Majoritarian biases denote the additional seats parties obtain as a result of the winner-takes-all properties of majoritarian electoral systems. Partisan biases denote the seat benefits that a particular party obtains beyond those expected by any other party with an equivalent vote share. Such biases are common in majoritarian systems (Becher and Gonzalez Reference Becher and Gonzalez2019), but they can also be found in proportional representation (PR) systems (Walter and Emmenegger Reference Walter and Emmenegger2023). They can be the unintended consequence of electoral geography (Calvo and Rodden Reference Calvo and Rodden2015), but they can also be the deliberate result of malapportionment or gerrymandering (Grofman Reference Grofman2016). In short, the composition of parliament might not be representative of the voters. If the electoral systems’ biases are substantial, the discrepancy between the median voter and the median MP can be large. Moreover, in case of partisan biases, underrepresented groups are likely to struggle to overcome these biases.

The biased transmission from the median voter’s ideological position to the median MP’s ideological position cannot be attributed solely to mechanical distortions in the vote-to-seat translation. If minority groups’ preferred candidates have only a low chance of winning because they are systematically disadvantaged, some voters will desert these trailing candidates in favor of more promising but ideologically more distant alternatives. In other cases, minority candidates might not enter electoral contests with low chances of winning, leaving voters with less preferred candidates. In these ways, strategic voting and candidate entry exacerbate biases that advantage the ruling party (Cox Reference Cox1997). By implication, strategic voting makes the ruling party’s constituency ideologically more heterogeneous, especially as the electoral system’s biases increase. Consequently, there might be significant gaps between the ideological position of ruling party’s median MP and the ideological position of some of the party’s voters.

Direct democratic institutions offer underrepresented voters the opportunity to overcome these electoral biases. In direct democratic votes, the median voter prevails. If the composition of parliament is largely representative of the population, voters who feel badly represented have little reason to believe that the population would be more receptive of their political ideas. However, in case of electoral biases, underrepresented voters may be ideologically closer to the median voter than they are to the median MP. For such voters, direct democratic institutions offer the potential to break the political dominance of overrepresented groups by circumventing parliament and addressing voters directly. By having direct democratic institutions at their disposal, underrepresented voters gain the opportunity to have direct success against political authorities, force parliament into negotiation on policies, and mobilize new issues and political tendencies (Linder and Mueller Reference Linder and Mueller2021). In short, direct democracy has the potential to force political representatives to pay more attention to underrepresented voters (Leemann and Wasserfallen Reference Leemann and Wasserfallen2016).

Importantly, however, this logic applies only to voters who are closer to the median voter than they are to the median MP. These voters have a better chance to win a direct democratic vote (where the median voter prevails) than their representatives are to win a vote in parliament (where the median MP prevails). In contrast, this logic does not apply to voters who are ideologically far away from the median MP but even further away from median voter. Although these voters might also be badly represented, they are not better off under direct democracy, because—in terms of ideological distance—the median voter is even further away than the median MP. For such voters, there is little benefit in adopting direct democratic institutions.

We argue that the relative ideological distance to the median voter and to the median MP matters for popular support for the adoption of direct democratic institutions. In contrast, the division between supporters of opposition and ruling parties is of secondary importance. Certainly, in biased electoral systems, supporters of opposition parties are more likely to favor the adoption of direct democratic institutions. However, direct democracy can also be interesting for voters of ruling parties. Especially voters who support ruling parties for strategic reasons might experience significant gaps between their own ideological position and the ideological position of the median MP (Cox Reference Cox1997). For such voters, direct democracy can help align policy-making more with their own ideological position. Therefore, we expect that support for direct democratic institutions does not exclusively divide voters of opposition and ruling parties. Rather, our argument should also apply to supporters of the ruling party.

WERE VOTERS ABLE TO MAKE AN INFORMED DECISION?

Voters ideologically closer to the median voter than they are to the median MP should be more supportive of direct democratic procedures. But is it realistic to assume that voters at the time understood where they stood ideologically relative to the median voter and the median MP? We believe that it is.

Our empirical analysis examines the adoption of the popular initiative in Switzerland in 1891. All adult males were invited to participate in the popular vote. They had access to detailed information on political debates and discussions. Voting results were made publicly available at the level of municipalities. In fact, in our analysis, we rely on the same publicly available information (see Figures 7–9 in the Supplementary Material for examples). In addition, at the time of the vote, literacy levels were comparatively high in Switzerland, as nation-wide examination results demonstrate. All Swiss males had to attend compulsory initial military training where their reading and writing skills were tested. Among the 1879 cohort, only 1.6% of the servicemen were considered illiterate. By 1913, this share had declined to 0.04% of the cohort (Grunder Reference Grunder2015). Baten (Reference Baten2022, 6) shows that in comparative perspective, Switzerland had one of the highest levels of school enrollment and literacy in Europe in 1900. Newspaper coverage was also high. In 1896, Switzerland had 339 different newspapers, which corresponds to one newspaper per 8,600 inhabitants (Ritzmann-Blickenstorfer and Siegenthaler Reference Ritzmann-Blickenstorfer and Siegenthaler1996). For comparison, the corresponding number for Germany, the “birthplace” of newspapers, is one newspaper per 15,700 inhabitants (Wilke Reference Wilke and Wimmer1991, 76).

In the public debate preceding the vote, the representation deficit was highlighted as the primary motivation for demanding the popular initiative. As the proponents argued, there was an “alienation” between the government and its people, with the government no longer understanding what the people really want (Michel Reference Michel2021, 262). In contrast, the opponents bemoaned the weakening of parliament, questioned the population’s ability to formulate policy proposals, and warned that the popular initiative would be mostly used by peripheral groups to push for extreme policies (Kölz Reference Kölz2004, 640; Linder, Bolliger, and Rielle Reference Linder, Bollinger and Rielle2010, 70–2; Michel Reference Michel2021, 256–64). Consequently, organized groups, most notably political parties, tried to mobilize voters with reference to questions of representation. In this way, political parties were an important source of information about representation deficits and the relative positions of the median MP and the median voter.

Most importantly, however, Swiss voters had access to a simple heuristic that allowed them to compare their position to the median voter and the median MP. Since 1848, Swiss voters are regularly voting on constitutional revisions. Moreover, in 1874, Switzerland adopted the law referendum, which allows citizens to challenge laws previously passed by parliament. By 1891, Swiss voters had voted on several dozens of direct democratic proposals. Each referendum offers widely available and easy-to-understand information that allows voters to compare their own ideological position to the median voter and the median MP. The informational requirements for these comparisons are low. Any referendum that voters support indicates that they are ideologically distant from the median MP (who, by definition, must have supported the law/constitutional revision in parliament). Moreover, any referendum that voters support and win indicates that they are also part of the majority in the voting population. Put differently, they are ideologically closer to the median voter than they are to the median MP. In contrast, any referendum that voters oppose and lose indicates that they are ideologically closer to the median MP than they are to the median voter. In short, every referendum provides voters with easy-to-understand information on where they stand ideologically relative to the median voter and the median MP. Voters simply need to compare their own vote (yes/no) to the widely published referendum outcomes (yes/no).Footnote ⁴

In the period under investigation, referendums covered a wide range of topics, from questions about civil rights, public health, and domestic commerce to issues related to transportation, government operations, and macroeconomics (see Table A1 in the Supplementary Material for a list of referendums considered in the analysis). However, only about 7% of all laws passed by parliament are challenged by law referendums, although law referendums have an almost 50% chance of being successful (Linder and Mueller Reference Linder and Mueller2021, 131). This small share of challenged acts of parliament should not be surprising though. Law referendums are a costly political action. To launch a law referendum, political groups had to collect 30,000 signatures (today 50,000 signatures) within 100 days (Kölz Reference Kölz2004, 617). In the late nineteenth century, this number of signatures amounted to about 5% of the vote-eligible population. For this reason, law referendums only challenged the most salient and important acts of parliament. For the same reason, turnout in law referendums was on average higher than in elections (by 3.1 percentage points).Footnote ⁵ In short, referendums do not only provide clear signals on voters’ ideological positions relative to the median MP and the median voter, they also do so in case of the most salient political issues.

THE ADOPTION OF THE POPULAR INITIATIVE IN SWITZERLAND

In the period before the adoption of the popular initiative, Swiss politics was characterized by a conflict between radical–liberal groups (called the “left”) and conservative groups (called the “right”). These two camps structured political competition along a simple left-versus-right dimension and subsumed new issues into this primary dimension (Rovny and Whitefield Reference Rovny and Whitefield2019).Footnote ⁶

In the late nineteenth century, the Socialists were still a marginal group. In the 1890 election, socialist candidates ran in only 19% of all electoral districts and obtained only 3.4% of the vote (Gruner Reference Gruner1978, vol. 3, 387), and none of their candidates was elected to parliament.Footnote ⁷ The Socialists would become a relevant political force only in the twentieth century. With the Socialists’ emergence, the use of “left” for the radical–liberal coalition was discontinued. In the 1891 vote, the (still marginal) Socialists supported the introduction of direct democratic institutions because, like the Conservatives, they were suffering from partisan gerrymandering by the ruling radical–liberal coalition (Emmenegger and Walter Reference Emmenegger and Walter2021).

In these years, the radical–liberal coalition benefited from the electoral system’s biases and controlled a majority in parliament. Convinced to be politically underrepresented, the Conservatives regularly accused the ruling radical–liberal coalition of resorting to partisan gerrymandering to cement their political dominance (Emmenegger and Walter Reference Emmenegger and Walter2021; Gruner Reference Gruner1978; Natsch Reference Natsch1967). In the 1880s, the Conservatives recognized that direct democratic institutions might be a way to gain political influence.Footnote ⁸ In 1884, three prominent Conservatives submitted a parliamentary motion, demanding from the government to develop a proposal for the adoption of a popular initiative. Importantly, in this parliamentary motion, they explicitly lamented the biases resulting from the majoritarian electoral system and the, from their point of view, politically motivated design of electoral districts (Kölz Reference Kölz2004, 639–40). Put differently, the Conservatives believed to be ideologically closer to the median voter than they were to the median MP who was a representative of the radical–liberal coalition in this period.Footnote ⁹

The Federal Government and the radical–liberal majority in parliament opposed the proposal. However, in line with the literature on why members of the majority group in parliament opt to delegate to citizens the power of the initiative (Smith and Fridkin Reference Smith and Fridkin2008), not all members of the radical–liberal coalition were equally opposed. First, the radical–liberal coalition was heterogeneous. As the dominant political force of the time, it combined a variety of groups, representing, among others, the manufacturing industry, crafts, salaried employees, and farmers. Especially, farmers were dissatisfied with party leadership (Junker Reference Junker1968). Second, in some cantons, members of the coalition ran against each other in elections (Gruner Reference Gruner1978). Hence, there were considerable, region-specific tensions within the radical–liberal coalition. Finally, the coalition had reasons to believe that more direct democracy was popular among voters. Legislators facing strong electoral competition in their districts had incentives to support direct democratic institutions (Kölz Reference Kölz2004, 643).

Ultimately, thanks to some renegades from the radical–liberal coalition, the popular initiative found a majority in the lower chamber.Footnote ¹⁰ The popular initiative gave the population the possibility to propose and vote on constitutional articles if 50,000 male citizens of voting age sign the proposal (circa 8% of the voting population).Footnote ¹¹ However, given that the adoption of the popular initiative required changing the constitution, it was subject to a popular vote. On July 5, 1891, Swiss voting-age males were asked to cast their vote. The decision proved to be contentious. While 60.3% of the voters approved the adoption of the popular initiative, a sizable minority of 39.7% rejected the reform (Linder, Bollinger, and Rielle Reference Linder, Bollinger and Rielle2010, 70).

MEASUREMENT AND RESEARCH DESIGN

The data we use in this article stem from a large data collection effort of municipality-level support for direct democratic proposals. In addition, we rely on roll-call votes that have been previously collected by Bolliger and Zürcher (Reference Bolliger and Zürcher2004). In the next section, we first describe the customized item response theory (IRT) scaling model we use and how we measure ideological positions. We then describe in the subsequent section the other variables we use in the analysis (for replication data, see Leeman, Emmenegger, and Walter Reference Leemann, Patrick and Walter2025).

The IRT Model

Our main variables are the ideological positions of municipalities and members of parliament (MPs) on a liberal-conservative dimension. Using these variables, we measure how far away a municipality is from the center of gravity, that is, how much a municipality differs from the weighted median municipality, which we use to capture the median voter, and the median MP. To derive such measures, we need to scale municipalities and MPs in a common space and need votes to do so. We can rely on referendum votes for which we can observe vote shares for each municipality but also how the MPs voted on them in parliament. We use these referendum votes on a large variety of topics because they are also voted upon in parliament, thus providing bridging votes that allow us to map ideological positions of municipalities and MPs into the same space. Put differently, for each of these referendums, there is both a vote in parliament and a popular vote on the exact same question, which we can use identify the ideological positions of the median voter and the median MP. We include all votes for which we have data, which took place up until 1891 (see Table A1 in the Supplementary Material for a list of the 22 votes used in the analysis). The parliamentary votes always precede the popular votes.

Hence, in our analysis, we infer individual behavior from group-level data. However, we find it unlikely that our analysis suffers from an ecological fallacy. For starters, most municipalities were small and ideologically homogeneous.Footnote ¹² In addition, it is hard to imagine that voters who are well represented in federal politics but reside in badly represented municipalities would support the adoption of the popular initiative when the federal government itself was openly critical of the new direct democratic institution. Finally, as a robustness test, we have reestimated all models using only ideologically homogeneous municipalities. We dropped all municipalities from the sample in which the top two partisan returns were within 20 percentage points of each other. Table A10 in the Supplementary Material shows that our results are not affected.

We build on standard IRT models that are frequently used in political science (e.g., Clinton, Jackman, and Rivers Reference Clinton, Jackman and Rivers2004). These models are often applied to legislative behavior where we observe only a yes or no vote. When working with direct democratic ballot decisions, we of course observe the vote share per municipality. Therefore, we derive an alternative IRT estimator, which is based on outcome data either following a Bernoulli distribution or a binomial distribution. For the legislative votes, we can use the standard IRT model and define the probability of a yes vote on bill i as

$$ \begin{array}{rll}P({y}_{ij}=1)=\Phi ({\beta}_i\cdot {\theta}_j-{\alpha}_i),& & \end{array} $$

where $ {\theta}_j $ is the ideal point of legislator j, $ {\beta}_i $ is the discrimination parameter of bill i, and $ {\alpha}_i $ is the difficulty parameter of bill i. Where we depart from the standard model is when we derive the model for the municipalities. Here, we formulate it as a binomial model, which allows us to retain the same bill parameters. This enables us to locate municipalities in the same political space as the MPs.

To describe the distribution of municipality votes, we define k as the number of percentage points a ballot issue received, n is set to 100. We further define the probability of an individual vote by a citizen from a municipality to also be $ P({y}_{ij}=1)=\Phi ({\beta}_i\cdot {\theta}_j-{\alpha}_i) $ and this leads to the following model:

$$ \hskip-20em P(y)=\Big\{{\displaystyle \begin{array}{l}\Phi \left({\beta}_i\cdot {\theta}_j-{\alpha}_i\right),\hskip11em \mathrm{if}\hskip0.3em i\hskip0.3em \mathrm{is}\ \mathrm{a}\ \mathrm{MP},\\ {}\left(\genfrac{}{}{0.0pt}{}{n}{k}\right)\Phi {\left({\beta}_i\cdot {\theta}_j-{\alpha}_i\right)}^k{\left(1-\Phi \left({\beta}_i\cdot {\theta}_j-{\alpha}_i\right)\right)}^{n-k},\hskip0.4em \mathrm{if}\hskip0.3em i\hskip0.3em \mathrm{is}\ \mathrm{a}\ \mathrm{municipality}.\end{array}} $$

This model has the attractive feature that we estimate the same difficulty and discrimination parameters for a specific bill, regardless of whether it was voted upon in parliament or at the ballot box. We provide full details on identification and implementation in the Supplementary Material (see Section A1.2.1 of the Supplementary Material). The model assumes that there is one main dimension of political conflict. A factor analysis of all votes used in the analysis supports this assumption (see Figure 4 in the Supplementary Material). The scree plot of the eigenvalues shows that the different votes load strongly on the first factor, while no other factor achieves an eigenvalue of one or more. In addition, we inspect the discrimination parameters to ensure that there are no issue areas that do not load on this main dimension. We find among the five votes with the smallest absolute values for the discrimination parameter issues ranging from the state’s alcohol monopoly, the federal state’s right to print money, regulations for employing factory workers, or allowing the federal state to legislate on accident and health insurances. Some of these proposals featured a broad partisan consensus. Others already foreshadow the slowly emerging conflict between capital and labor (see Section A1.2.2 of the Supplementary Material). To show the robustness of the results, we later also present our estimation models based solely on data from municipalities where no socialist candidate ran for office (see Table A9 in the Supplementary Material).

In addition, we carry out two further robustness checks. First, we dichotomize the municipality data and run a standard binary IRT model (see Section A1.2.4 of the Supplementary Material). Second, we split the samples and estimate separate models for MPs and municipalities to see whether it is sensible to project both actors in the same political space (see Section A1.2.5 of the Supplementary Material). Both robustness checks are in line with our measurement strategy.

The converged model provides us with an ideological measure of municipalities and national MPs in the same political space. Figure 1 shows the distribution of ideal points of both groups. Whereas municipalities seem to follow a rather unimodal distribution with a large conservative tail, the MPs show a different picture. MPs fall into two modes, which correspond to the two main political groups at the time: a large progressive group to the left of most municipalities and a sizable conservative group. Figure 1 is in line with descriptions of political factions in parliament at the time. Numerically, the progressive Radicals dominate, whereas the Conservatives are the main opposition group. Between the Radicals and the Conservatives, there is a small group of Liberals, at the time sometimes referred to as the “Center” due to their centrist position between the Radicals and the Conservatives (Gruner Reference Gruner1978). Although there were relevant political differences between the Radicals and the Liberals, they were united in their opposition to the Conservatives (Altermatt Reference Altermatt2020, 81). Together, the Radicals and the Liberals formed a center-left majority coalition in parliament, while the Conservatives were in a permanent minority position. There were no Socialists in parliament yet.

Figure 1. Ideological Position of MPs and Municipalities in Common Space

Based on these ideological positions we can derive our ideological measures:

• • Absolute ideological distance between the municipality and the median MP ( $ \mathrm{abs}({\boldsymbol{\theta}}_{\mathrm{mun}}-{\boldsymbol{\theta}}_{\mathrm{median}\ \mathrm{MP}}) $ ): To account for how far away a municipality is from parliament, we derive the absolute distance to the median MP. This variable captures how well (low values) or how bad (high values) a municipality fares under a purely representative system.

• • Absolute ideological distance between the municipality and the median voter ( $ \mathrm{abs}({\boldsymbol{\theta}}_{\mathrm{mun}}-{\boldsymbol{\theta}}_{\mathrm{median}\ \mathrm{voter}}) $ ): We derive the position of the median voter by locating the median municipality. To do this, we weight each municipality’s ideal point by the size of the voting-eligible population.Footnote ¹³ In this way, the weighted median municipality captures the ideological position of the median voter. Subsequently, we measure the distance of each municipality to this weighted median.

• • Relative ideological distance ( $ \mathrm{abs}({\boldsymbol{\theta}}_{\mathrm{mun}}-{\boldsymbol{\theta}}_{\mathrm{median}\ \mathrm{MP}}) $ − $ \mathrm{abs}({\boldsymbol{\theta}}_{\mathrm{mun}}-{\boldsymbol{\theta}}_{\mathrm{median}\ \mathrm{voter}}) $ ): Our main independent variable puts the two absolute ideological distance measures into relation. This variable captures whether a municipality is ideologically closer to the median voter (positive values) or ideologically closer to the median MP (negative values).

EMPIRICAL RESULTS

All regression models are hierarchical linear models where the municipality-level vote share for the introduction of the popular initiative is the outcome variable. We recognize that municipalities are nested in districts and districts are nested in cantons. To account for this data structure, we include random effects for electoral districts and cantons. Table 1 presents all model specifications (the complete regression table is provided in Table A3 in the Supplementary Material). We also present various robustness tests in the Supplementary Material where we add interactions, additional control variables, or use fixed effects for electoral districts to show that the results are nonsensitive to model specification.

Table 1. Strategic Vote for the Adoption of the Popular Initiative

Note: Labor market structure: Local employment shares in first and second sectors. Party vote shares for Radicals, Conservatives, Democrats, and Socialists, and others (ref: Liberals). Constant included but not shown. $ {}^{***}p<0.01 $ ; $ {}^{**}p<0.05 $ ; $ {}^{*}p<0.1 $ .

Adopting the popular initiative empowers the median vote relative to the median MP. Therefore, we expect support for the popular initiative to increase when voters are ideologically closer to the median voter than they are to the median MP. Model 1 in Table 1 presents the purely ideological specification with our relative ideological distance measure (positive scores indicate that voters are ideologically closer to the median voter than they are to the median MP) plus the two control variables for the local labor market structure. Model 2 adds the dummy variable for ideologically extreme municipalities. In Model 3, we add information on the current extent of subnational direct democracy. Finally, in Model 4, we add vote shares of the different political parties based on the elections in 1890.

We find that support for the adoption of the popular initiative is higher among voters who are ideologically closer to the median voter than they are to the median MP. Models 2 to 4 show that these results hold regardless of the precise model specification, although—unsurprisingly—the effect becomes somewhat weaker when we control for party vote shares (which are important sources of information for voters about their possible underrepresentation). These findings suggest that voters understand the strategic implications of the adoption of the popular initiative in terms of furthering their political goals. The popular initiative is primarily useful for voters to gain political influence if their ideological distance to the median MP is higher than their ideological distance to the median voter. Only in this case can voters realistically hope for a more positive outcome in a direct democratic vote (made by voters) compared to a parliamentary decision (made by MPs). Our results suggest that voters are aware of these strategic considerations. To illustrate the effect size, one can look at what happens when the difference in distances increases by one standard deviation and compare that to a one standard deviation change in the vote share of Conservatives. The model-predicted increase in the yes-share for ideology is half the size of the modeled increase for one standard deviation increase of the conservative vote share.

Figure 2 explores whether our argument about differences in ideological distances also applies within groups. Table 1 has already shown that the effect of our relative ideological distance measure holds when controlling for electoral support at the municipality level. Figure 2 now zooms in on supporters of the Radicals. The Radicals were the politically most powerful group of the time and arguably the most advantaged by the contemporary electoral system’s biases. Supporters of the Radicals are thus particularly important in the context of this vote on the adoption of popular initiative. As argued above, we expect supporters of the Radicals, like all voters, to be more supportive of the popular initiative if their own ideological position is closer to the median voter than they are to the median MP.

Figure 2. Support for the Initiative and Radical Vote Share

Note: We show average support for the popular initiative in municipalities closer to the median MP (red) and closer to the median voter (blue). We added a jitter factor to the x-values to make the figure easier to read.

Figure 2 shows that support for the adoption of the popular initiative (y-axis) decreases with the vote share of the Radicals (x-axis). More importantly, however, Figure 2 shows that across the full range of vote shares for the Radicals (from 0% to 100%), municipalities ideologically closer to the median voter than they are to the median MP (blue dots and LOESS curve) show stronger support for the popular initiative than municipalities that are closer to the median MP than they are to the median voter (red dots and LOESS curve). Even at very high levels of electoral support for the Radicals, we can still observe this effect of relative ideological difference to the median voter and the median MP.Footnote ¹⁵