Participatory processes and their outcomes: comparing assembly and popular vote decisions

Abstract How do face-to-face, assembly processes, and non-face-to-face, popular vote processes impact the decisions made by citizens? Normative discussions of the comparative merits of these two broad types of participatory decision-making processes partly rely on empirical assumptions concerning this question. In this paper, we test the central assumption that assemblies lead to decisions that are more widely supported by participants than popular votes. We do so by analyzing 1,400 decisions made through these processes on the highly salient issue of municipal mergers in Swiss municipalities since 1999. We find that assembly decisions are consistently made by larger majorities than popular vote decisions and that this relationship is significantly mediated by turnout. This suggests that higher levels of agreement in assemblies mainly result from selection biases – with fewer dissenting citizens participating in assemblies than in popular votes – rather than from internal dynamics in assemblies.


Introduction
The limitations of conventional representative systems, structured around elections, to realize democracy have been increasingly acknowledged by political actors and researchers. Many contend that political practice would come closer to realizing democratic principles if participatory processesnamely processes empowering ordinary citizens to take part in decision-making on policy issueswere included in democratic systems alongside elections. Such processes can take two main forms. They can gather citizens in proximity to discuss, deliberate, or make decisions on issuesin which case they are 'face-to-face' (Mansbridge, 1983, xi) [or 'observable' (Sartori, 1987, p. 111), or 'restricted' (Saward, 1993, p. 18)] participatory processes, such as town hall meetings, popular assemblies, and mini-publics. These processes all: (i) require that citizens are simultaneously present in the same location (assembled); and (ii) entail discursive elements ahead of the decision-making momentwhich can rank from structured and facilitated deliberation to, at least, the oral provision of information on the issue to be decided on. Alternatively, participatory processes can enable ordinary citizens to participate in decision-making processes without entering an assembly, as members of the mass publicin which case they are 'non-face-to-face' (Mansbridge, 1983, xi), 'greater-than-observable' (Sartori, 1987, p. 111), or 'unrestricted' (Saward, 1993, p. 18) participatory processes such as referendums and initiatives. These processes do not require: (i) that citizens are gathered in a specific place, nor that; and (ii) discussions take place in the presence of all participants before the vote.
Selecting one form of participatory process or the other is widely expected to influence both decision-making processes and their outcomes. Existing studies have extensively focused on the effects of these two kinds of processes on citizens, for instance in terms of gaining knowledge, increasing participants' readiness to change opinion or to empathize with others, and developing civic skills (see e.g. Fishkin, 1995;Grönlund et al., 2010;Colombo, 2018;Muradova, 2021). But these designs are also expected to influence the substance of the collective decisions made by participants. In particular, face-to-face processes have long been assumed to support the formation of larger agreement and to be more conducive to widely supported decisions than non-face-to-face processes (see e.g. Mansbridge, 1983, p. 274, xi;Parkinson, 2001;Qvortrup, 2005).
In this article, we test whether this theoretical expectation can be observed in political practice by comparing the impact of two specific instances of face-to-face and non-face-to-face participatory processes on the level of agreement in real-world decisions made by citizens on local political boundaries in Switzerland. We add to the existing literature, which has relied on single-case studies or on experimental data (see e.g. Setälä et al., 2010;Suiter and Reidy, 2020), with unique behavioral data enabling us to compare 1,400 decisions made by citizens in face-to-face municipal assemblies and in non-face-to-face popular votes on the conflictual issue of municipal mergers since 1999. Our results provide empirical support to the theoretical expectation that decisions made through face-to-face processes are approved by larger majorities than decisions made through popular votes. Yet, we find that the link between participatory process and level of agreement in the decision is significantly mediated by turnout, which is substantially lower in assemblies than in popular votes. This suggests that higher levels of agreement in assembly decisions do not result from internal, common interest-generating dynamics of face-to-face processes. Rather, larger majorities appear to result from the relative absence of dissenting citizens from assembly settings: a portion of the dissenting citizens who take part in popular votes do not take part in assemblies.
This study serves to clarify how institutional design choices impact the substance of outcomes in participatory processes in ways that can inform debates on democratic innovations (Mutz, 2008;Saward, 2021). It complements existing research by uncovering a missing link between the input and the output of participatory processes: while most studies find that assemblies have a negative effect on turnout compared to popular votes processes (Schaub, 2012;Stadelmann-Steffen and Dermont, 2016), we demonstrate that this significantly impacts the level of agreement on the resulting decisions. Our study also highlights the need to consider political authorities' motivations for experimenting with or preserving specific participatory processes. Since results do not only matter in terms of which decision has been made, but also in terms of the relative majorities for the different options (elected representatives might have a larger leeway to implement a proposal with 65% support than with 51% support), they might have strategic incentives to favor processes that generally lead to higher levels of agreement. Finally, our results can inspire further research on voluntary municipal mergers in other countries such as Finland, Norway, or Japan and inform political debates about how territorial reform can gain the support of the population.

Participatory processes and level of agreement
In normative debates comparing the value of face-to-face and non-face-to-face participatory processes, which we will call 'assembly' and 'popular vote' processes for simplicity, it is often assumed that the former are more prone to sustain the development of well justified, fair, common interestoriented collective decisions than the latter (see e.g. Mansbridge, 1983, pp. xi, 273-5;Gutmann and Thompson, 2004, pp. 30-31;Fishkin, 2009, p. 78). As a result, it is generally expected that decisions made in face-to-face processes will be adopted by larger majorities than decisions made in popular votes, reflecting the widespread acceptance by participants of the policy decisions made. The expected difference in the outcomes is not a difference in kindparticipants voting 'yes' in one setting and 'no' in the otherbut in degreelarger majorities voting 'yes' or voting 'no' in one setting than in the other. In what follows, we introduce the institutional variations between assembly and popular vote processes that have been invoked to explain higher levels of agreement in face-to-face decision-making. Most explanations focus on dynamics within assemblies ('internal mechanisms'); one focuses on dynamics prior to assemblies ('external mechanism'). Each mechanism can be categorized as supporting a beneficial or a cautionary interpretation of larger majorities in assembly decisions. Indeed, the value of high levels of agreement on policy proposals depends on the extent to which it reflects the actual acceptance of policy proposals by decision-makers, rather than manipulation or the suppression of disagreement.
Internal mechanisms: proximity, information, decision-rule Proximity One reason to expect that assembly decisions will be adopted by larger majorities than popular vote decisions is that assembly participants are in physical proximity with one another.
The beneficial expectation is that the opportunity for human contact and for communication through oral and body language provided by the assembly would promote empathy, enhance participants' readiness to change their mind, and increase incentives to consider common instead of personal interest and to form collective judgments (Mansbridge, 1983, p. 272;Barber, 1984, p. 271;Fishkin and Laslett, 2003, p. 54;Muradova, 2021). This in turn would support the formation of wider agreement among participants. In contrast, citizens casting their ballot in isolated ways in popular votes, without interpersonal exchange that could 'create a mood of cooperation and of responsiveness to each other' (Schaub, 2012, p. 310, fn8), would lack such incentivesor even face opposite ones: for Mansbridge, popular vote processes often 'discourage the development of common interests by encouraging people to register their personal preferences privately, without having to participate in public debate ' (1983, xi). Lower levels of agreement in popular votes' results are thus to be expected.
The positive relationship between proximity and agreement level has also been explained in a more cautionary way. It highlights the darker side of proximity, which can produce empathy and foster the generation of common interest, but also enable group pressure and discrimination in ways that sustain dynamics of internal exclusion (Young, 2000;Wojciechowska, 2019). Impatience, annoyance, disapprobation, or even disgust expressed through body language can discourage certain participants to publicly express disagreement, especially those from marginalized groups with less practice in articulating their preferences rapidly and effectively. Thus, face-to-face decision-making processes might produce larger majorities by suppressing conflict, especially in contexts with large inequalities. In contrast, lower levels of agreement in popular vote decisions could be explained by the fact that 'referenda provide the distance between opponents that the average citizen requires in moments of conflict' to feel able to express dissenting views (Mansbridge, 1983, p. 275).

Information
A second reason to expect that assembly decisions will gather the support of larger majorities than popular vote decisions is related to the way each process structures information provision to citizens ahead of the vote. In assemblies, all participants are provided the same information through debate or discussion before being able to vote; in popular votes, even when voters receive the same information in a voter-guide, the debate happens in a mass campaign that can provide different information to different voters.
The beneficial explanation for how this would lead to higher levels of agreement in assemblies is that these processes would facilitate citizens' acquisition of information. Because all participants at least hear the same discussions before the vote without having to actively search for it (Schaub, 2012, p. 310, fn8;Stadelmann-Steffen and Dermont, 2016, p. 101), and thus have to reflect on the reasons provided by the speakers (Fishkin and Laslett, 2003, p. 12), they would be better able to identify common interests and vote in their support than when they need to rely on impersonal news media before casting their ballot as members of the mass public (Fishkin, 1991, p. 21;Parkinson, 2006, p. 164).
A more cautionary explanation suggests that higher levels of agreement could reflect the incompleteness of the information provided in assemblies, especially in terms of giving visibility to disagreement. Those daring to speak up in an assembly process and able to react immediately are generally highly skilled people whose views might be closer to those of authorities, thus making the availability of dissenting views and arguments scarce. For instance, in an analysis of the deliberative quality of statements made at the Glarus citizen assembly in Switzerland, Gerber and Mueller (2018) show that members of the political elite are much more likely to make statements with a high deliberative qualitywhich are presumably also more convincingthan ordinary citizens. In this scenario, assemblies would lead to wider agreement because of the lack of dissenting information, whereas the results of popular votes would display more disagreement because voters would have received their information in more controversial public campaigns taking place over longer periods of time, thus better reflecting the varieties of views present in the political community (see Young, 1990, p. 227).

Decision-rule
The outcomes of assembly processes should finally be more widely supported than those of popular vote processes because of the decision-rule: assembly votes are generally open, whereas popular votes are secret. 1 In the beneficial explanation, the large majorities brought about by open vote would result from the adoption of a more responsible behavior by assembly participants. While secret voting would free decision-makers from a sense of accountability, lifting possible hurdles to casting uninformed or prejudiced ballots, open voting would incentivize them to make decisions justified by reasons and favoring common interests Engelen and Nys, 2013;Vandamme, 2018). It could also enable voters to cast a ballot informed by others' decisionsfollowing, for instance, the voting behavior of trusted fellow-citizens in the open vote (Mueller et al., 2021) (an information not readily available to voters in popular votes). As these effects would apply to all present members, higher levels of agreement can be expected.
In the cautionary explanation, open voting would cause broader agreement by suppressing disagreement  through an element of group pressure already mentioned in our discussion of proximity. For some participants, especially more vulnerable ones in relatively small polities, standing out publicly as 'the dissenter' in an assembly composed of one's communitypossibly including one's employer, for instancecan badly effect their lives beyond the assembly (Bufacchi, 2001;Gerber and Mueller, 2018). In particular when they expect to be in a clear minority, participants have incentives to abstainnot raise their handor even to vote against their own preferences to follow the majority instead of standing their ground. Dissenting preferences are easier to express with secret ballots (Mansbridge, 1983), which would explain lower levels of agreement in popular vote decisions. 1 Another, decision-rule related explanation often used in the literature is the binary aspect of popular vote decisions (see Wagenaar, 2019). Parkinson for instance considers that the "yes/no" format of referendums or initiatives is conducive to more conflict and polarization: "the referendum is an inherently majoritarian device: it splits the range of opinion on an issue into two camps, favouring conflict over compromise" (2001, p. 136). However, assembly decisions can also be binaryand the ones under scrutiny in this paper are, as we explain in the next section. Therefore, we do not expand on this explanation here.

External mechanism: selection bias
Beyond the effects of face-to-face processes on the internal dynamics of decision-making processes, a cautionary explanation also insists that assembly and popular vote processes can differentially impact who decides to participate in these processes in the first place. To be sure, and unless assemblies are mini-publics for which participants are randomly selected, self-selection biases are present in both assemblies and popular votes, whereby some categories of citizens are more likely to participate than others (Sciarini et al., 2016;Stadelmann-Steffen and Dermont, 2016;Wojciechowska, 2019). Yet, it is expected that assemblies reinforce self-selection bias compared to popular votes in ways that can affect levels of agreement. This is because, compared to casting a secret ballot in a popular vote, participating in assemblies entails additional costs (see Schaub, 2012, p. 312;Stadelmann-Steffen and Dermont, 2016, pp. 99-100;Stadelmann-Steffen and Gerber, 2020, p. 417). Assemblies are more time-consuming than popular votes: citizens need to gather in a specific place for a sometimes unknown duration. Especially for citizens with non-standard working hours or with care obligations, this can also entail higher monetary costs, for example, to pay for elderly care. And, as highlighted in the previous sub-section, participating in assemblies requires more from citizens: all have to vote publicly, and some might have to actively debate certain issues.
While their specific nature depends on institutional design details, 2 these additional costs are commonly expected to reinforce self-selection biases that suppress diversity more likely in assemblies than in popular votes (Young, 2000;Grönlund et al., 2010;Wojciechowska, 2019). On the one hand, these costs might be considered too high for some categories of citizensespecially citizens with lower levels of resources or abilities and citizens who disagree with the expected majority and fear group pressure. For them, staying away from the process will likely be even more attractive when they believe that the result will not be close or that they are in the minority on an issueeither because their investment would not have a sufficient effect, in a rational choice logic (Stadelmann-Steffen and Gerber, 2020), or because they might want to avoid disagreeing publicly (Mansbridge, 1983). 3 On the other hand, other categories of citizens will likely be overrepresented. This can be the case for citizens who can easily afford the higher participation costs, and thus might already share similar issue positions by virtue of belonging to the same social group; or it can be the case for those who believe they will be part of the majority in a specific assembly. In addition, when turnout in assemblies is generally low and the issues put to the vote are known in advance, special interest groups can easily hijack them and 'gain a majority by simply mobilizing their members and thus [ : : : ] reach decisions serving their particular interests at the expense of the common good' (Schaub, 2012, p. 311). In all these likely scenarios, the higher self-selection bias induced by assemblies entails that the decisions made through these processes will display higher levels of agreement than the decisions made in popular votesnot because participants end up agreeing as a result of being assembled face-to-face, but because those who disagree were not present in the first place.

Hypothesis
All these possible explanatory mechanisms, summarized in Table 1, give reasons to expect that the level of agreement will be higher in assembly than in popular vote decisions. The hypothesis we test in this article thus reads as follows: H 1 : The level of agreement is higher in assembly decisions than in popular vote decisions 2 For instance, compensation schemes for citizens or pre-defined schedules might lower participation costs. 3 We thank the anonymous reviewers for bringing the rational choice explanation to our attention.
Determining which of the discussed explanations for this hypothesis is most likely at work matters to interpret the democratic value of higher or lower levels of agreement. If larger majorities are the result of the beneficial explanations, we might praise it as an indicator that participants could widely agree on an issue, setting their personal divergences apart to support the common interest. But if the cautionary explanations explain higher levels of agreement, they would be highly problematic from a democratic perspective, signaling that the interests of dissenting members of the political community were suppressed. Unfortunately, our data does not allow us to systematically test which of the internal explanatory mechanisms are at play. Yet, the case at handwhich we introduce in the next sectionallows us to gauge the relative importance of internal and external mechanisms in explaining higher levels of agreement in assemblies.

Case description
We test the hypothesis (H 1 ) by comparing the level of agreement in decisions made through two specific kinds of assembly and popular vote processes at the local level in Switzerland. The country is known for being one of the few political systems that includes popular vote processes at all levels of government since the 19 th Century. It also has a long tradition of open, face-to-face legislative assemblies at the cantonal (the intermediate government tier in the Swiss federal system) and especially at the municipal level (the lowest government tier in the Swiss federal system), which continues to this dayparticularly in small municipalities and in German-speaking cantons (Ladner and Haus, 2019). 4 What interests us here is that both kinds of processes have been used at the same level of government, namely the local level, to make decisions on the same issue, namely municipal mergers. The local political landscape in Switzerland has indeed been marked by the 'wave' of municipal mergers in developed countries (Baldersheim and Rose, 2010), reducing the number of local jurisdictions by a quarter since the end of the 1990s. These territorial reforms took place on a voluntary basisfollowing a bottom-up approach that has become more widespread since the new millennium (Strandberg and Lindell, 2020;Folkestad et al., 2021). In other words, municipalities themselves forged merger coalitions, with financial or administrative support provided by higher The average population size of municipalities with an assembly is 2,500 inhabitants, whereas the average size of a municipality with a parliament amounts to 10,000 inhabitants (Ladner, 2016). government tiers (regional or national) (Baldersheim and Rose, 2010); local executives developed merger projects, proposing to amalgamate two or more local jurisdictions into a new municipality or to incorporate one or more jurisdictions into an already existing one; and the citizens of each municipality could approve or reject these merger projects (Strebel, 2019)either in assemblies, or in popular votes (see Table 2 for a comparison of the two institutional designs).
The assembly designs we study are so-called municipal assemblies. In Swiss local government, the legislative body is either an elected parliament or a municipal assembly open to all citizens (never both). Both bodies are equally competent to scrutinize, debate, and approve the actions of local executives on issues such as local budgets, zoning, or local infrastructure projects. In his comprehensive study of municipal assemblies in one Swiss canton, Rochat (2019) reports that the average duration of a municipal assembly, which generally takes place two to three times a year, is between 75 and 90 minutes. 5 For each agenda item, which is announced to citizens in advance, the local executive elaborates on its proposal before inviting assembly participants to speak up and propose amendments. Yet, participants use this discursive option sparsely: in only 32% of assemblies, there are 11 or more utterances by participants (Rochat, 2019, p. 19). Unlike in face-to-face processes like mini-publics, whose 'package' includes random selection, facilitation, and information provision (Mutz, 2006, p. 61), these communicative exchanges are also neither structured nor facilitated. After this opportunity for discussion, the participants to the assembly openly vote on the proposal made by the executive, as well as possible amendmentsunless a secret vote is requested, which is very rare (2% of all assemblies) (Rochat, 2019, p. 22).
The popular vote processes used to make merger decisions in Switzerland are mandatory referendums. 6 These processes are required by law; the legislative body's decisions on certain policy areas or of a certain legal status (such as constitutional amendments) cannot be implemented without the support of a majority of voters in a popular vote. A few weeks before the popular vote, citizens receive their ballots with accompanying information on the issues at stake and a public campaign involving both Yes and No supporters unfolds in the weeks preceding the vote. Unlike cantonal-level assemblies, which take place outdoors, municipal assemblies take place indoors. 6 Note that the decisions of municipal assemblies can often also formally be subject to optional referendums. This process enables nonelected actors to demand that specific legislative decisions are put to a popular voteif they succeed in collecting enough signatures within a specified amount of time. However, this process is extremely rarely used in practice to contest municipal assembly decisions in general (see Rochat, 2019, pp. 33-34); and, based on the responses of all cantonal departments responsible for local affairs, no optional referendum was used to contest municipal assemblies' decisions on municipal mergers. The cantonal department of municipalities of Grisons mentioned one casethe merger of Ilanzwhere signature collection against the assembly decision was started but not enough signatures were collected and hence no popular vote was held. Therefore, we do not discuss this formal option further.
Citizens have a two to three weeks window before referendum day to mail in their ballot, or they can cast it in person on the day of the popular vote. The result of the vote is made public after all ballots have been cast.
In both processes, participants could decide between the same two options: adopt a merger project proposed by their local authorities to merge with one or more other municipalities; or reject the merger project and preserve the status quo. These options were pre-set; since merger projects are elaborated by representatives from several municipalities jointlyalike international treaties (see Wagenaar, 2019, p. 2)there is no possibility to propose amendments, even in assemblies.
Our data includes 1,400 decisions made by citizens through participatory processes on municipal mergers since 1999. It provides us with unique real-world behavioral data to study the impact of participatory decision-making processes on decision outcomes and an ideal setting to test the hypothesis formulated above for four reasons.
First, Swiss municipal merger processes are very standardized and similar across municipalities, cantons, and time. For one, the starting conditions for local governments to develop merger projects are very similar: all municipalities are given financial incentives by the cantons to merge [lump-sum payments in the event of a merger, usually dependent on the number of inhabitants and the fiscal strength of the involved municipalities (Kaiser, 2014)], and all municipalities could seek potential merger partners and forge merger coalitions themselves. Then, the citizens of all municipalities vote on the same issue, and with the same options: the adoption or rejection of a merger contract elaborated by executive authorities of merging municipalities, which specifies various details concerning the merger process and the post-merger municipality, such as the name and the flag of the post-merger municipality, the location of the local administration and public schools, or the local tax rate. Our focus on these most similar cases thus ensures that we measure the effects of institutional designs on outcomes and not differences emanating from the issues at stake (see Stadelmann-Steffen and Dermont, 2016, p. 112).
Second, Swiss merger decisions provide us with exogenous variation with respect to our independent variable, assembly vs. popular vote decisions. Local executives who propose the merger cannot choose the process through which the decision is made on a case-by-case basis: rather, the ratification process is pre-determined by cantonal or municipal regulations. Hence, the choice of decision procedures is exogenous to the proposal voted on in the case at hand, which is an important prerequisite to study the impact of decision procedures on decision outcomes.
A third reason why Swiss municipal merger decisions are a well-suited case for our study is that they are very widespread and provide us with a high number of vote decisions on the same issue, but through different decision-making processes. A third of the 1,400 votes on municipal mergers was held in municipal assemblies, the rest in popular votes.
Finally, focusing on municipal mergers enables us to study the impact of specific processes on the decisions made when citizens actually participate in these processes. Indeed, decisions on municipal mergers tend to mobilize citizens. Whether or not to merge with another municipalityand thus dissolve or enlarge existing political communitiestouches upon the fundamental question of political boundaries. Merger decisions can also particularly mobilize citizens in that they are final (see Chambers, 1998, pp. 159-160) in the sense that they are legally binding for the municipality and that, once a merger is accepted and implemented, it is very hard to reverse it and secede from the new municipality. In our data, we observe an average turnout of 57% in municipal merger decisions, which is very high for Swiss standards. Municipal mergers mobilize 63% of voters on average in local popular votes, which can be compared to the 46% average turnout in national popular votes (generally assumed to mobilize more because of higher stakes) in the same period. And about 40% of voters join municipal assemblies on mergers, which is four times higher than the 9% average turnout generally observed in assemblies (see Ladner, 2016, p. 35;Rochat, 2019).

Operationalization and estimation
Our dependent variable is the level of agreement in merger decisions; it corresponds to the size of the majority that made a merger decision reported in official documents and newspaper reports. To operationalize this, we calculate the distance from the decisive 50% threshold. The variable can thus take values from 0, which signifies maximum disagreement (i.e. a perfectly split electorate), and 50, which signifies maximum agreement (i.e. unanimity).
In addition to the independent variable presented in the previous sectionwhether a merger decision is made in a popular vote or in an assemblywe also include several additional factors in our analysis, which could influence the level of agreement in a merger vote. 7

Turnout
Higher turnout in any voting decision is often a sign of higher politicization and contestation of a decision and might hence translate into lower levels of agreement. Moreover, as we have discussed above, assemblies and popular votes might affect turnout in different ways. Unfortunately, turnout was only reported in official documents and newspapers for 869 of our 1,400 merger decisions. Yet, for 429 additional decisions, we could approximate turnout by dividing the number of valid votes in a merger decision (i.e. the combination of yes and no votes cast) by the number of eligible voters in national-level popular votes in the same year. 8 This approximation appears to be fairly reliable: for cases for which we have both the number of valid votes and the reported turnout (N = 476), Pearson's correlation coefficient between reported and approximated turnout amounts to .99 (see also Figure A.1 in the online appendix).

Size of a municipality
Scholars have argued that levels of agreement on all issues are higher in small communities than in large onessimply because larger size increases the chances of a diverse citizenry (Dahl and Tufte, 1974). This might be reflected in the levels of agreement in merger decisions.

Relative benefits within the merger project
Levels of agreement in merger decisions might also be influenced by a municipality's relative position in a merger coalition. There might be less controversy among citizens in municipalities whose situation would change little or be clearly enhanced than in municipalities in which citizens could lose from a mergerfor instance, in relatively small municipalities that make up a small fraction of a merger coalition (Jakobsen and Kjaer, 2016) or in municipalities with lower tax rates than their merger partners (Strebel, 2019). We thus include a municipality's relative size in a merger coalition, as well as the difference in a municipality's tax rate to the population-weighted coalition mean in our analysis. 7 Descriptive statistics for all variables used in the analysis can be found in Table A.1 (online appendix). 8 We adapted our measure of approximated turnout to account for the fact that Swiss citizens living abroad are able to vote at the cantonal and local level in some cantons, but not in others. We did not use the approximated turnout measure for cantons and municipalities where foreigners have the right to vote, because we only have the information on the number of eligible voters at the national level, where foreigners are not eligible to vote. Our approximation does not account for abstention (not raising a hand or blank ballots), which are not included in the numbers of valid votes; this entails that we might slightly underestimate turnout. However, for the 349 cases for which we have the information on abstention, its mean percentage amounts only to 1.3%. Hence, the underestimation of turnout through the approximated turnout measure is minimal (see also Figure A.2 in the online appendix).

Local symbols
We control for changes in symbols that are important for the local identity of citizensan additional potential source of controversy in merger votes. An important aspect of citizens' local identity is the name of their municipality, to which citizens are often very attached (Soguel and Silberstein, 2015). A merger project might be more controversial in municipalities whose name would disappear after a merger than in municipalities whose name (or part of it) would remain.

Previous merger experience
Previous experiences with municipal mergers might also impact the level of agreement in a merger decision. About 12% of the municipalities in our data were once part of a merger coalition that was not implemented due to a lack of popular support, and 5% of them have already merged before and are involved in a subsequent merger project. This 'merger legacy' might increase the level of agreement, either because citizens in a municipality that already merged might be less skeptical of a renewed mergerthey already know the drill and accept change more easilyor because local executives of municipalities that previously failed to merge might be more careful to craft more acceptable merger projects.

Legislative body
To capture potential longer lasting effects of different types of legislative institutions on the extent to which citizens are divided over local issues, we also include a control variable for whether a municipality's legislative body is an assembly or whether it is a parliament.

Time
Finally, to account for unobserved factors which vary across time, we include time-period dummy variables.
Our unit of analysis is a decision made by a municipality involved in a merger project. Since municipalities are nested in merger projects, and merger projects in cantons, we use three-level hierarchical linear regression models to test our hypothesis, with merger decisions as level−1, merger projects as level−2, and cantons as level−3. This allows us to account for these multiple levels of nestedness instead of treating merger decisions that belong to the same merger project and canton as independent observations. The significant intra-class correlation of the level of agreement shows that this modeling strategy is appropriate. The share of variation in level of agreement amounts to 5% at the cantonal level and to 32% at the merger project level (see Table A.2 in the online appendix).

Assembly decisions are more broadly supported
Is there support for the hypothesis that the level of agreement is significantly higher in face-to-face, assembly decisions than in non-face-to-face, popular vote decisions on Swiss municipal merger projects (H 1 )? While most of the decisions made in both assemblies and popular votes are in support of the merger projects (85%), there seems to be a difference in degree, whereby the size of the majorities supporting merger proposals is larger in assemblies than in popular votes. 9 Figure 1 shows the distribution of the level of agreement in merger decisions taken in assemblies and in popular votes. The x-axis depicts the dependent variable 'level of agreement', where 0 means a perfectly split electorate and 50 means a perfectly unanimous decision. The figure reveals a remarkable 9 The very high acceptance of merger projects could reflect particularities of the Swiss context, in which many consultations take place before an issue is put to a vote to citizens (Mendelsohn and Parkin, 2001, p. 9;Lijphart, 2012;Qvortrup, 2018, p. 190).
concentration of unanimous decisions in assemblies, but not in popular votes. This is a clear indication that larger majorities support decisions made in face-to-face processes than in non-face-to-face processes.
However, Table 3 makes clear that we should not jump to conclusions about the effect of these processes on the level of agreement. It shows the results of three multilevel regression models: a bivariate model including only the main independent variable (1), a multivariate model with all control variables except turnout (2) and a full model including turnout (3). Models 1 and 2 suggest that decisions made in assemblies lead to higher agreement than decisions made in popular votes, and thus corroborates our hypothesis. On average, merger decisions taken in a municipal assembly are 9 percentage points more distant from the 50% threshold than popular votes in model 1 and 5.7 percentage points in model 2. Yet, this changes when we include turnout in the specification (model 3). Here, assembly decisions lead to slightly lower agreement than popular vote decisions. Turnout itself has a statistically significant and substantive negative effect on the level of agreement: if turnout increases, the level of agreement decreases.
Before we turn to a more in-depth assessment of this result, we provide evidence for its robustness. To alleviate the concern that the differences between assemblies and popular votes may be due to some unobserved factors at the level of the municipality or the merger project, we can focus on 55 municipalities in which the exact same merger proposals were voted on in a sequence of an assembly and a popular vote. This enables us to compare the level of agreement across different procedures for the same constituency voting on the exact same proposal. 10 If we calculate the difference in level of agreement between the assembly and the subsequent popular vote, we find that the level of agreement was equal or higher in the assembly than in the popular vote in 53 out of 55 municipalities (see Figure A.3). This is further face validity for Note: 0 on the x-axis signifies a situation of maximum disagreement (50:50), whereas 50 signifies a situation of unanimity (100:0).

10
In all of these cases, municipal assemblies first decided on the merger and the popular vote was a mandatory, preregulated, step for its ratification. If the municipal assembly rejects the merger project, the merger process normally stops there (before the mandatory popular vote). assembly decisions producing higher levels of agreement than popular vote decisions. 11 Figure 2 12 presents the results of a multilevel regression analysis, where decisions represent the unit of analysis, which are nested in municipalities (level−2) and in merger projects (level−3) and which employs the same control variables as the models in the main analysis. 13 Note. * p< .05; ** p< .01; *** p< .001. Coefficients are obtained throughmixedcommand in Stata; standard errors in parentheses.

11
It is also noteworthy that the mean level of agreement in assemblies that made the final decision to merge and in assemblies that preceded a popular vote is the same (see Figure A.4) as well as the mean level of agreement in popular votes that were preceded by an assembly and those that were not ( Figure A.5). 12 For the remainder of this paper, we present the main results in figures. All tables can be found in the online appendix. 13 Given that the 55 municipalities only stem from 3 cantons, we could not include cantons as level-4 in the analysis and instead had to include them as dummy variables. Figure 2 clearly shows that the results are robust to this alternative specification. While we find a positive and significant effect of assembly decisions on the level of agreement in a bivariate and in a multivariate specification without turnout, this effect becomes insignificant when turnout is included. Turnout itself exerts a negative and statistically significant effect on the level of agreement. 14 Turnout mediates the effect of participatory processes How can we explain that the positive effect of assembly decisions on levels of agreement vanishes once turnout is included in the regression model? There are two possible explanations. The first one is spurious correlation. This would mean that the newly included variableturnoutimpacts both the independent and the dependent variabledecision procedure and level of agreementand hence the previously observed association between the two hinged on both being caused by the new variable. This explanation does not make sense here, because the turnout in a merger decision can logically not impact the decision procedure used to make that decision.
The second possible explanation is mediation. In our case, this would mean that the effect of the decision procedure on level of agreement runs through turnout: assembly votes mobilize fewer voters, and lower mobilization is in turn associated with higher levels of agreement. Following our theoretical expectations, this would entail that the internal dynamics of assemblies play a very minor role in explaining the variations in level of agreement between assembly and popular vote decisions. Rather, these variations would result from the external mechanismthe fact that different people self-select to participate in these two processes.
Our data appears to confirm this interpretation. On the one hand, the participatory process used significantly impacts turnout: a multilevel regression model with turnout as a dependent Note: Dots are regression coefficients; lines represent 95% confidence intervals. Full results in Table A.3 in the online appendix.
14 Given that some municipalities in our dataset were involved in a merger project multiple times, we also conducted a restricted analysis that only involves municipalities with exactly one merger project participation (see Table A.4). The results remain the same in this alternative specification. Moreover, we also check for multicollinearity issues. The variance inflation factors for all variables remain below 5 which suggests that there are no multicollinearity issues (see Table A.5).
variable shows that turnout is on average 27 percentage points lower in assemblies than in popular votes (see Table A.6 in the online appendix). This is in line with existing studies (Schaub, 2012;Stadelmann-Steffen and Dermont, 2016), but adds to them by showing that assemblies clearly mobilize fewer voters than popular votes even when average mobilization is exceptionally high.
On the other hand, the mediation model presented in Figure 3 shows that the direct effect from decision procedure to level of agreement is not statistically significant (see Imai et al., 2011). Instead, making decisions in assemblies negatively affects turnout, and turnout negatively affects the level of agreement in a merger decision. This indirect effect of assemblies on level of agreement amounts to 9.5 percentage points (−26.86*−.354) and is significantly different from 0 with 99.9% confidence. In other words, turnout fully mediates the effect of decision procedures on level of agreement.
We can interpret this result as clear evidence that the main reason why we observe higher levels of agreement in assembly decisions than in popular votes is the external mechanism of selfselection bias. This gives us two reasons to be cautious not to interpret higher levels of agreement as a sign that the policies adopted in assemblies are better justified or simply more acceptable than the decisions made in popular votes. First, the internal mechanisms of face-to-face processes presented in section 1.1 seem to play very little role in explaining higher levels of agreementeven when the issue put to the vote is one that tends to mobilize more citizens than is generally the case. 15 The beneficial (or cautionary) effects of being in proximity, of common information provision, and of open voting can thus not explain higher levels of agreement. Second, this result suggests that higher levels of agreement reflect the fact that participants in assemblies are less diverse than participants in popular votes.
Indeed, while the available data does not enable us to determine who stays away from assemblies and why, our results suggest that the high costs of participating in an assembly do not affect all citizens equally. If all citizens were equally disincentivized from attending the assemblies, we would see no stable difference in levels of agreement between assembly and popular vote decisions. The presence of this stable difference suggests that assemblies specifically discourage dissenting citizens to take part in participatory decision-making. Following our theoretical framework, this can be explained by the fact that the anticipation of being in the minority tips the balance in favor of staying away from the assemblyeither because the costs of attending are perceived too high when the result cannot be impacted by one's vote or because the prospect of having to express disagreement publicly is discouraging (see also Stadelmann-Steffen and Gerber, 2020).
As a result, the views of dissenting citizens are less likely to be appropriately represented in assemblies than in popular votes, an inequality of presence that should be of concern for Note: ***p<.001. Full results in Table A.7 in the online appendix.

15
The level of turnout might indeed also reflect higher levels of politicization and conflict. However, this should not impact our results since there is no reason to assume that the politicization of a merger proposal is systematically higher or lower in assemblies or popular votesparticularly since local executives have no possibility to choose the decision-making process used.
democrats (Schaub, 2012;Wojciechowska, 2019). To be sure, it might not be so problematic in the case of municipal mergers, because these encompassing territorial reforms do not systematically harm the interests of specific groups. The relocation of public infrastructureschools or administrative servicesor changes in local tax rates might impact certain groups more than others, but they will do so in different ways depending on the specific merger project and on the specific municipality in which these groups are located. In addition, in the cases included in our database, including these dissenting views would rarely have changed the outcome of the decision. This, however, might not be the case for other municipal assembly decisions, especially when turnout is much lower than for municipal merger decisions and when decisions are made on issues that create clearer winners and losers.

Within assembly effects?
The results so far suggest that the effect of participatory processes on levels of agreement mainly results from self-selection. However, this does not entail that the internal mechanisms introduced in section 2.1 have no effect at all on the level of agreement in the decisions made. Therefore, and while we are not able to test the separate effects of the three internal mechanisms highlighted in Section 1 with the available data, we wish to close this section by shedding tentative light on this question by coming back to Figure 1. Indeed, this figure showed that a much larger proportion of assembly decisions were taken unanimously, with 100% of the votes. In general, unanimous decisions are viewed with suspicion by democrats, because it is very unlikely that a group of people with different identities, income levels, professions, activities, and beliefs can unanimously agree on anything.
We can examine these unanimous decisions more closely by distinguishing them from nonunanimous decisions and by regressing this binary variable on the same indicators as in the previous models. Figure 4 shows the average marginal effects of deciding in an assembly on the probability of making a unanimous decision based on a multilevel logistic regression. We can see that, even when turnout is included in the analysis, the probability that the outcome of an assembly is a unanimous decision is 2 percentage points higher than in a popular vote. This is a very substantive effect, considering that the percentage of unanimous decisions among all merger decisions amounts to 5.3%. Assemblies thus lead to more unanimous decisions independently of turnout.
This suggests that the internal dynamics of assemblies increase the chances of unanimity. We cannot determine whether these dynamics are of the beneficial or of the cautionary kind. It is possible that physical proximity, unified information, and open votes enable the identification of common interest and hence unanimous agreement among participants. However, the design of municipal assemblies gives many reasons to warrant democrats' suspicion towards these unanimous decisions. Unanimity could reflect some genuine form of agreement in highly structured environments in which information is provided to small groups of participants by experts and stakeholders, discussions among participants are moderated by facilitators and focused on reason-giving, and options are available to amend the proposal decided on to accommodate participants' preferences and requests. But the kind of assemblies under scrutiny here do not include any of these features: they involve relatively large numbers of citizens; they gather them for a short period of time; they do not include structured deliberative processes among participants (no expert inputs, no facilitation, no small group discussions); and they offer no options to amend the merger project on the table ('take it or leave it'). Thus, in the absence of design elements that could help developing proposals to which all really agree and, most importantly, which could reduce the effect of the structural inequalities in these processes (Chambers, 2009, p. 339), group pressure dynamics cannot be discarded as an explanation for the much higher share of unanimous decisions observed in municipal assemblies than in popular votes.

Conclusion
With an analysis of 1,400 real-world decisions on Swiss municipal mergers made by citizens either at municipal assemblies or in popular votes, this article shows that decisions made in municipal assemblies exhibit higher levels of agreement than decisions made in popular votes. Our analysis suggests that, contrary to what has been sometimes assumed in democratic theory, the internal dynamics of face-to-face assemblies are not the main explanation for this difference. Rather, the effect of participatory processes on levels of agreement is significantly mediated by turnout: assemblies lead to much lower turnout than popular votes and lower turnout is associated with higher levels of agreement. Thus, even if the door to participation is not wide open in popular votes (Parkinson, 2001, p. 133), it appears to be even less open in municipal assemblies. While further research on municipal assemblies' internal mechanisms and their role in explaining the high number of unanimous decisions observed is needed, our findings are in line with existing research, which has shown that assemblies lead to significantly lower turnout than popular votes (Schaub, 2012;Stadelmann-Steffen and Dermont, 2016), and they go beyond these results in two ways.
First, our study suggests that the citizens who do not walk through the doors of assemblies are citizens who disagree with the majority. Future research shedding some light on whether they stay away from assemblies because they do not want to publicly voice their dissenting opinion or because they simply do not see the point in attending an assembly when their vote is very unlikely to make a difference would be highly useful in guiding efforts to reduce the participation hurdles that keep citizens away from assemblies. The unwillingness to pay the higher costs of attending an assembly might be overcome if municipalities offer some form of compensation or free time for one's participation [as a few municipalities do (Stadelmann-Steffen and Dermont, 2016, p. 100)], whereas measures to restructure these processes such as introducing facilitation or secret voting might be necessary if dissenters are primarily discouraged by having to disagree publicly.
Second, our findings demonstrate that differences in turnout in assemblies and popular votes have a stable impact on the decisions made. In the Swiss municipalities we study, this difference is one of degree in the size of majorities, and not in kind (yes/no). In other words, the absence of dissenting citizens from assemblies does not seem to affect the nature of the decisions made. However, the high level of support for the merger projects in participatory processes could reflect Note: Dots are regression coefficients,lines represent 95% confidence intervals. Full results in Table A.8 in the online appendix. specificities of Swiss local politics, with small and quite homogeneous communities and a tradition of consensus democracy where conflict is often dealt with in a very early phase of the decisionmaking process (Lijphart, 2012). In more conflictual, majoritarian settings, the kind of outcome could well depend on which participatory process was usedand who stayed away from the process. Studies focusing on this would be highly valuable to test the validity of our findings beyond the Swiss case. What this study reaffirms is that participatory processes, like any political process, are not outcome-neutral. Expanding knowledge about how they impact outcomes, how this impact affects political elites' decisions to introduce one or the other participatory process in specific contexts, and how these processes' possible biases could be compensated will help to inform projects to deepen democracy.