Consultative referendums and democracy - assessing the short-term effects on political support of a referendum on a municipal merger

ABSTRACT Local consultative referendums are a widely used democratic innovation. Nevertheless, there is still limited knowledge about the local effects following a referendum, especially in terms of how the public reactions varies given if a citizen has been on the winning or on the losing side of the referendum. The purpose with this study is therefore to add to this line of research by assessing how a local referendum about a municipal merger affects external political efficacy and political trust within a local community. By analysing two cross-sectional datasets collected in a bilingual Finnish municipality before a merger referendum, in 2018 (N = 6,686), and after the referendum, in 2020 (N = 3,133), as a pseudo-experiment of effects of a municipal merger, we show that the aggregated levels of external political efficacy and political trust have increased and vary based on being on the winning or losing side of the referendum.


Introduction
Consultative referendums provide the citizens with a possibility of having a direct say in the decision-making processes for a specific policy issue, hence increasing the extent of political legitimacy for the subsequent political decision (Accetti and Oskian 2020;Qvortrup 2017). Nevertheless, the use of referendums has also been shown to contribute to polarising effects across the population, creating tensions within political communities (Van der Eijk and Rose 2021). This polarisation is derived from the notion that referendums unintentionally create winners and losers, and for the long-term sustainability of a political community it is subsequently considered crucial how the losers in a referendum react and deal with the defeat (Anderson et al. 2005;Blais and Gelineau 2007;Brummel 2020). Moreover, a reoccurring argument in favour of referendums is that the losers would easier accept political decisions if they have had a voice in the process (Hedrick and Gherghina 2020). Hence, referendums might contribute to improving the quality of democracy while simultaneously fostering disunity.
The effects on the population are subsequently expected to differ after a local referendum, based on whether the citizen was on the winning or losing side of the referendum issue. Winners have been shown to express higher levels of political support following a referendum while the effects among the losers are not as straightforward (see Esaiasson et al. 2019;Marien and Kern 2018). Whether a citizen is a winner or a loser is therefore based on the issue opinion. Moreover, whether the citizen was in favour or against the referendum as an element of the decision-making process is also expected to alter the effects on political support (Marien and Werner 2019). How a citizen relates to the organising of a referendum is therefore here used to reflect a process opinion. We thus argue that citizens' political support varies based on their issue-and process opinion, both prior to-, and following a local referendum. As the moderating effects of process opinion is an understudied element of the local referendum literature, our study aims at making an important contribution to that line of research (but see Esaiasson et al. 2019). Our research question is therefore whether these assumptions hold. Our broad research aim is subsequently to explore the relationship between political support, issue-and process opinion prior to and following a local level consultative referendum.
In order to answer our research question, the subsequent paper will test the effects of issue-and process opinions on political support by taking both a prospective and retrospective approach. By using two cross-sectional survey datasets collected in a bilingual municipal setting in Western Finland 2018 and 2020, prior to-, and following a referendum about a proposed municipal merger, it was possible to explore the relationship between issue-and process preferences and political support and compare the effects prior to, and after a consultative referendum. After this brief introduction, we continue with accounting for the relationship between referendums and political support and describe how the effects are expected to differ based on issue-and process-opinions. Thereafter we present the general research design and data together with the main findings. In the final part, our findings are discussed in relation to earlier findings.

Local referendums and political support
Direct democratic processes, such as referendums, are expected to enhance the levels of political support by increasing the levels of the perceived fairness of the decision-making processes within the political community (Esaiasson et al. 2019;Marien and Kern 2018, 5;Van der Eijk and Rose 2021, 105).
Following this assumption, studies have repeatedly found referendum experiences to contribute to creating higher levels of political support among the participants (Bauer and Fatke 2014;Bowler and Donovan 2002). Still, as the use of referendums challenges the foundations of a representative political system (Folkestad et al. 2021) it might also contribute to lower levels of political support, as it could be taken as evidence that the current political system does not work. As Dyck (2009) argued, 'direct democratic institutions put citizens in an adversarial relationship with their government'. The use of referendums might therefore also contribute to decreasing the authority and fostering distrust in political authorities (Citrin 1996;Dyck 2009;Voigt and Blume 2015).
Consultative local referendums tend to have a direct impact on the everyday lives of the citizens living in the local community. Moreover, by taking part in a democratic exercise at the local level, citizens get to experience direct democracy in working while influencing political outcomes (Marien and Kern 2018, 12). These local referendums are therefore usually perceived as important events (Hedrick and Gherghina 2020). There is, however, no guarantee that the advice given by the local population in the referendum will be followed at the political level if it is consultative (Folkestad et al. 2021;Jäske 2017). As the congruence between political decision-making and majority preferences is considered as a key requirement of democratic governance (Leemann and Wasserfallen 2016), these events can have important consequences for the political system. The workings of the local political community thus give the local population evidence of what to expect from a representative form of democracy, and during instances when the local politicians decide to organise a referendum but then still overrule the majority will, it should subsequently negatively affect various types of political support. Likewise, when the local politicians follow the outcome of a referendum, it should subsequently have a positive effect on the evaluation of democratic governance (Marien and Kern 2018). As such, local referendums give the population a glimpse into the actual relation between public opinion and political decision-making.
When measuring political support, scholars are interested in finding out whether a citizen is positively or negatively oriented towards a political object (Easton 1975, 436) and political efficacy and political trust are two types of attitudes that are considered as essential components of political support (Norris 2011). Political efficacy refers to citizens' perception of whether they understand and can influence the political processes (Boulianne 2019) and is here understood as the 'feeling that individual action does have, or can have, an impact upon the political process' (Campbell, Gurin, and Miller 1954, 187). Because of the complexity of the concept, it has come to be divided into an internal and external component (Bowler and Donovan 2002;Niemi, Craig, and Mattei 1991), with the internal component reflecting the extent of political self-confidence and political understanding and the external component reflecting assessments regarding the responsiveness of the political system (Niemi, Craig, and Mattei 1991, 1407-1408. This study will focus only on the external component of political efficacy, i.e., external efficacy. Whereas external efficacy refers to an evaluation of citizens' possibilities to influence politics, political trust refers to an assessment of whether the political representatives and institutions can be trusted to act in the best interest of the population (Boulianne 2019, 9). Consequently, political trust reflects an evaluation of whether a citizen judges the politicians and/or the political institutions to be trustworthy and is thus more relational in character than political efficacy (Iyengar 1980, 255). Moreover, a citizen might express high levels of trust in the judicial system while simultaneously expressing low levels of trust in the elected Members of Parliament (MPs). It is therefore crucial to differentiate between the objects of trust (Bauer and Fatke 2014;Easton 1965;Norris 2011). Furthermore, as trust in various objects has been repeatedly shown to be internally related (Levi and Stoker 2000), longer periods of well-performing local level politicians could be expected to contribute to fostering higher levels of political trust on a higher institutional level. Consequently, the experience of a local referendum might affect more system important political attitudes at higher system levels.

Political support, issue opinion and process opinion
There are two important elements related to a referendum to consider, with the effects on political support expected to be affected by an individual's position on these two elements: issue-and process opinion. In this context, issue opinion reflects how a citizen relates to the political question put to a referendum whereas process opinion reflects how a citizen relates to the use of a referendum as a political decision-making process. As previously mentioned, it is inevitable that a referendum creates winners and losers. Thus, following a referendum, citizens that were in favour of the eventual outcome should express higher levels of political support in comparison with citizens on the losing side (Esaiasson et al. 2019;Marien and Kern 2018; Van der Eijk and Rose 2021). Consequently, there is an array of studies focusing on the effects of the winner-loser gap following referendums, showing that the winners typically start to express higher levels of political support while the losers usually retain their pre-referendum levels of political support (Anderson and Tverdova 2001;Esaiasson 2011;Marien and Kern 2018). As the winners, furthermore, constitute a majority within the specific political community, it is therefore not surprising when the aggregate level effects often have been shown to be positive in terms of changes in political support following a referendum (Marien and Kern 2018). As both the external type of political efficacy and political trust are used to measure citizen's perceptions of the political system, these items have been shown to be strongly correlated (Anderson 2010;Esaiasson, A-k, and Turper 2015;Niemi, Craig, and Mattei 1991). Therefore, it is likely that the effects following a referendum should be similar and we expect it to matter for both types of political support whether the citizen was on the winning or losing side of the referendum issue.
The other important element to consider concerns process opinion, implying that it matters for political support whether the citizen was in favour or against the referendum as a part of a decision-making process. When the public approves of the referendum, it might be possible to create 'better losers' following the outcome. According to democratic theory, it is expected that when the citizens have been given the opportunity to influence a specific political decision, they should subsequently become more likely to accept an outcome that they themselves did not initially support (Anderson et al. 2005;Dahl 1989). Moreover, according to Esaiasson et al. (2019, 310), agreement on the decision-making procedures 'is a civilized way to solve collective decisionmaking problems', but the public is not necessarily a consultative part when deciding about the organisation of a referendum. Hence, when there is debate about the legitimacy of a decision-making process, the legitimacy of the final decision will subsequently also inevitably become questioned (Miller 1974). We therefore expect opinions about the referendum process to affect political support following a referendum. Following this overview, we present three initial hypotheses: H1a: The aggregate levels of political support increase following a referendum.
H1b: Following a referendum, the levels of political support increase more among citizens on the winning side in comparison with the citizens on the losing side of the referendum.
HIc: Following a referendum, the levels of political support increase more among citizens being in favour of the referendum in comparison with citizens being against the referendum.
It is also possible for a citizen to have been on the winning side while still being dissatisfied that the referendum was conducted in the first place, making the eventual positive causal effects on political support somewhat more uncertain. Still, it is more likely for a citizen to have been in favour of the referendum while still being disappointed with the eventual outcome, as the public has repeatedly been shown to be supportive of referendums as part of a political decision-making process (Bowler, Donovan, and Karp 2007;Font, Wojcieszak, and Navarro 2015;Schuck and de Vreese 2015). Hence, there are four different scenarios related to a referendum to consider with varying effects on political support, notwithstanding that two of them (issue-related) are here solely related to a post-referendum scenario. The expected effects from a referendum of being a winner/in favour or a loser/against are thus obvious. However, while we in line with Esaiasson et al. (2019) broadly expect winners being against the referendum to show moderate gains in political support we on the other hand expect losers still being in favour of the referendum to show moderate losses in political support. These four outcome scenarios are summarised in a fourfold below in Figure 1.
Furthermore, we intend to assess how issue-and process opinion and political support relate to each other and how this relation has changed after a referendum in comparison with the relation prior to the referendum. There is an array of studies focusing on what happens to political support among the losers following a referendum, but less so on the relationship between process opinion and political support following a referendum (but see Esaiasson et al. 2019;Marien and Werner 2019;Nadeau, Belanger, and Atikcan 2021). Leading up to a referendum, it is expected that the relation between issue-and process opinion should be strong as citizens are expected to be more positive towards a democratic process if they expect to gain from it. Hence, the process might just be perceived as a necessary means to an end. Therefore, even as support for referendums is generally high, citizens that expect to win should be even more supportive of it before the outcome based on instrumental considerations (Werner 2020) and become more supportive of referendums in general after a positive outcome (Brummel 2020). Even if, as Tyler (2006) argued, evaluations of decision-making processes and the eventual political outcomes are related, outcome satisfaction is expected to be the more important determinant for political support following a referendum (Esaiasson et al. 2019;Marien and Kern 2018). Consequently, even if both types of evaluations should matter, we expect issue preference to be the more important individual-level determinant of political support of these two. As previously mentioned, referendums create both winners and losers within a political community. However, leading up to a referendum, when the public tensions are expected to be higher, the relation between issueand process preferences and political support should be stronger than following a referendum when the referendum issue is settled. Hence, following a referendum, and when the referendum issue is finally politically settled, the relation between issue-and process opinion and political support should be weaker. We thus formulate our two final hypotheses accordingly:

Process
H2a: Citizens' issue opinion is more important than process opinion for predicting political support both prior to and following a referendum.
H2b: The effects of citizens' issue opinion and process opinion on political support are stronger prior to than following a referendum.
We now continue with presenting the local context and the research design guiding the remaining part of our paper.

Research design, context, methods and data
Proposed municipal mergers constitute typical contested issues (Setälä et al. 2020) and since municipalities create local identities merger discussions tend to evoke strong feelings (Zimmerbauer and Paasi 2013). Moreover, mergers are very complex issues for citizens to form an opinion on, as there is a multitude of elements to consider (Strandberg and Lindell 2020, 297;Lapointe, Saarimaa, and Tukiainen 2018, 514-516). Since 1991, it has been possible to organise referendums at the municipal level in Finland, and these can be either citizen-or local government-initiated (Local Government Act 656/1990, Section 24). Nevertheless, the final decision to conduct a local referendum remains solely at the local government level, where a majority of the municipal council needs to be in favour of the referendum. Moreover, when organising the referendum, the voters need to be given the option on the ballot of not supporting either of the alternatives, and the outcome of the referendum can only be advisory by law (Jäske 2017, 53). Of the 63 municipal level referendums during 1991-2021, 57 have been about a proposed municipal merger (Karv and Backström 2021), showing the importance placed by the elected local politicians in Finland of having public approval before making decisions about a municipal merger.
The two cross-sectional survey datasets for this study were gathered prior to and following a municipal level referendum on a proposed municipal merger in the municipality of Korsholm, Finland. Korsholm is a predominantly rural municipality located in the bilingual part of Western Finland and has around 19,500 citizens. The other partner of the merger, the city of Vaasa, is the 14 th largest city in Finland, with a population close to 70,000. Besides the traditional urban-rural factor, another important aspect that differentiates these two entities concerns the different language compositions. Korsholm is predominately Swedish speaking, as around 68% has Swedish as their first language. In Vaasa on the other hand, around 68% has Finnish as their first language. Hence, besides being a contested issue based on the urban-rural element there were apparent identity-based elements, derived from the language-aspect, included in the process of forming an opinion on the proposed merger that contributed to a polarising public debate (Setälä et al. 2020;Strandberg and Lindell 2020).
The referendum was held on the 18 th of March 2019 and the outcome was that 61.2% voted against the merger, 36.8% voted in favour of the merger while 2.0% chose neither, which is a mandatory option to include in local level referendums in Finland. The referendum attracted a significant turnout, as 11,329 out of 14,833 eligible voters participated in the referendum (turnout 76.4%). Earlier studies have shown that referendums that are considered as important attract a higher level of turnout (Garry, Marsh, and Sinnott 2005), and this constitutes evidence for the local importance placed on the merger issue. Following the results of the referendum, the municipal council consequently voted against the merger at the following municipality council meeting on 2 April 2019. Still, the decision to reject a merger was only reached after a vote in the municipal council, where 19 out of 43 elected representatives still chose to vote in favour of a merger. During the deciding vote in the municipal assembly, two of the pro-merger politicians chose to vote against the merger because of the referendum result, while another pro-merger politician chose to vote blank. This effectively altered the outcome through shifting the majority against a merger (Karv and Backström 2021); clearly showing the municipal residents that the majority will matters for political outcomes. Hence, this case is arguably interesting from a perspective of referendum effects as the referendum clearly affected the political outcome of the issue.
To test our hypotheses, we used public opinion data from two crosssectional surveys. The pre-referendum survey was conducted in collaboration with the municipality of Korsholm in March 2018, approximately 12 months prior to the referendum. The purpose of this survey was to give the local politicians information about what the population wanted included in the merger-agreement. The sampling frame for the survey was citizens aged 16 or older living in the municipality, in total 15,411 citizens. In total, 6,686 survey responses were received, giving us a response rate of 43.4%. The second survey was conducted in November 2020, approximately 18 months after the referendum. This survey had a strictly academic purpose as the merger issue at this point had already been settled in the municipal council. The sampling frame was here similar to the first survey, and in total 3,133 survey responses were submitted, yielding a response-rate of 21.5%. Young people, men, Finnish speakers and people living in the municipal centre were underrepresented among the respondents in both datasets (see Table A5). In order to account for this in our analyses, we apply a combined weight for gender, age, mother tongue and location in both datasets. This research design is thus based on a pre-and post-referendum test of the effects of issue opinion (merger opinion) and process opinion (referendum opinion) as determinants of political support.
It should be clearly noted that our possibilities to make inferences about the potential relationship between a referendum and political support based on our dataset is limited, given when the survey data was collected and the lack of suitable control groups (Gliner, Morgan, and Leech 2009;Leatherdale 2019). Firstly, since we use two cross-sectional datasets instead of panel data, our design is to be seen as a pseudo experimental design at best. Moreover, as the follow up survey data was collected 18 months following the referendum and around 30 months after the first survey it only allows for a nonexperimental evaluation about potential referendum effects. The main rationale for still conducting a follow-up survey 18 months after the referendum was that we expected that the local population might had become polarised to such an extent that time would not have healed all the wounds, as this was clearly a contested issue (Setälä et al. 2020). Hence, this research design still provides us with a possibility to contribute with valuable insights into potential polarising effects of a referendum based on a real-life event (see also Leatherdale 2019, 26). In the following section, the main variables are presented.

Variables
Our dependent variable is political support and is operationalised by using standard survey items measuring either external efficacy or political trust (see Appendix for survey items). To account for the referendum effects on external efficacy, we use two survey items: People can exert influence through voting (System influence) and Politicians do not care about the opinions of ordinary citizens (System responsiveness). Both items have been operationalised so that a higher value reflects the more efficacious answer in both instances on a four-point scale between 0-1 and these have been regularly included in similar studies for measuring external efficacy (Balch 1974;Niemi, Craig, and Mattei 1991). In order to account for the effects on political trust, we use two standard survey items measuring trust in politicians and trust in political parties, with responses operationalised on an eleven-point scale between 0-1, with higher values indicating higher trust. Using these four survey items included in two different datasets prior to and following a referendum within a specific local context enable us to account for the effects on four different types of political support following a referendum.
Our main explanatory variables concern individual opinions about the proposed merger and the referendum. Opinions about the merger are thus used to divide the population into perceived winners or losers (issue opinion). Still, to measure opinions about the merger we are forced to use survey items that differ between the two surveys. In the 2018 survey, the following survey item was used: What is your opinion about a potential merger between Korsholm and Vasa? Indicate your opinion on a scale between 0 and 10, where 0 means that you are entirely against a merger and 10 means that you are entirely in favour of the merger. Survey responses between 0 and 4 were here operationalised as being against a merger while responses between 6 and 10 were operationalised as being in favour of a merger. In the 2020 survey, the following survey item was used: How did you vote in the referendum about a merger with Vasa? Response options were in favour of a merger, against a merger, did not vote, don't want to say, and did not have the right to vote. In order to use these survey items for comparative purposes, both have been operationalised as binary variables, differentiating between citizens being/voting in favour of a merger and citizens being/voting against the merger, with a higher value indicating being against the merger. Through this procedure, we can divide our sample based on issue opinions.
Referendum opinion (process opinion) is measured through evaluations about the referendum as a part of the decision-making process, e.g., procedural satisfaction (see also Marien and Kern 2018). In order to measure referendum opinions, there is a similar situation as with the measurements of merger opinions, as different survey items have been used for comparison. In 2018, the following survey item was used: Do you think that a consultative referendum should be conducted about a municipal merger with Vasa? Response options were yes, no and don't know. However, in 2020 the following survey item was used: Do you think it was the right choice by the municipal council in Korsholm to organise a public referendum about a municipal merger with Vasa? Response options were still yes, no and don't know. Therefore, one of the survey items reflects an expectation while the other reflects an evaluation of a process. Notwithstanding this issue, both survey items reflect process preferences, and we see no reason why they are not compatible for measuring and comparing opinions towards the referendum as a part of the decision-making process. It should be noted that this is far from an optimal survey item to measure perceptions of procedural satisfaction, and survey items questioning whether the respondents find the referendum as a fair or appropriate decision-making procedure would have been more suitable (see here Esaiasson et al. 2019;Werner and Marien 2020). However, as the survey items reflect an evaluation of the decision-making process they are considered as suitable enough for our research purpose. Furthermore, in order to control for other individual level characteristics that have been shown to be associated with political support (Anderson 2010;Bowler and Donovan 2002;Schoon et al. 2010), we include covariates for age, gender, education and mother tongue (Swedish/Finnish) in our statistical analyses. Studies have also shown that language minorities are less politically active than the majority population (Hero and Tolbert 2004; but see Sandovici and Listhaug 2010) and that membership in a majority group is connected to higher levels of political support (Anderson and Tverdova 2001). Subsequently, political support has also been shown to be affected by context, and when members of a minority group within a heterogeneous population context does not feel that their interests are being sufficiently represented the levels of political support are likely to decline (Wolak 2018). We therefore expect members of the majority group in this local setting (Swedish-speakers) to express higher levels of political support. This especially following the referendum, as the members of the minority group would likely have become part of the majority group if the referendum outcome had been different.

Findings
We begin by checking for the relationship between merger-and referendum opinions in our two datasets and note that there appears to be a strong relationship between being against a merger and in favour of a referendum (see Table A6). We thereafter continue with comparing the levels of external efficacy and political trust in our datasets. Here we note that on aggregate, both indicators of external efficacy and political support have increased (see Figure 2). Independent-samples t-test were here conducted in order to test the significance of the differences in mean values between the two datasets. For System influence there was a non-significant difference, whereas the difference was significant for System responsiveness, Trust in politicians and Trust in political parties. Our findings further show that citizens who were against the merger were overwhelmingly more in favour of a referendum. However, in 2020, citizens in favour of a merger had become somewhat more positive towards the referendum, even though they lost. In terms of external efficacy, our findings show that the levels of efficacy declined among citizens being in favour of a merger, while the levels of efficacy increased substantially among citizens being against the merger. Moreover, the levels of political trust increased in each category. A separate regression model (not reported in detail here) also tested this through a three-way interaction term issue-opinion*processopinion*post-referendum, which was a significant predictor for all dependent variables. These initial findings partly confirm our hypotheses H1a, H1b and H1c.
We continue exploring the bivariate correlations between merger-and referendum opinion and we find them to be strongly correlated, although somewhat weaker in 2020 than in 2018 (see Table A7). Thus, to validate our hypotheses, we apply three multivariate linear regression models (OLS): two models including merger-and referendum opinion as covariates, together with sociodemographic control variables, for the 2018 (Model 1) and 2020 (Model 2) datasets respectively. For the final model (Model 3), we have pooled the two datasets. By using a dummy variable differing between pre-and postreferendum respondents and two interaction variables -one between postreferendum and merger opinion and one between post-referendum and referendum opinion -it is possible to assess the aggregated postreferendum effects on political support derived from merger-and referendum opinion. Furthermore, significant interactions have been post-hoc tested for marginal effects (see Figure A1 and A2) and are discussed in relation to the associated hypotheses.
Starting with the results for the effects on perceived system influence (see Table 1), our findings show that being in favour of the merger and the referendum both predicted higher levels of perceived system influence, even after accounting for sociodemographic predictors in 2018 and in the pooled analysis. However, neither merger opinion nor referendum opinion were significant predictors in 2020. This supports H2b. However, there is no difference in the beta-coefficients for issue-and process opinions prior to the referendum and, in the pooled analysis, the effect of process opinion is stronger than issue opinion. Both findings are contrary to H2a. It should also be noted that the direction of the causal effect of merger opinion on perceived system influence varies between the two datasets, as being against the merger in 2018 predicted lower levels of system influence but is insignificant in 2020. The dummy for post-referendum is significant in the pooled analysis, which supports H1a. The interaction-term for postreferendum*merger opinion is significant and the visualisation of it (see Figure A1) provides clear support for H1b whereas H1c is not supported (significance of marginal effects tested with Anova test of mean differences with post-hoc Bonferroni adjusted pairwise comparisons). Continuing with perceived system responsiveness (see Table 2), merger opinion turned out to be significant in both 2018 and 2020, showing that being against a merger predicts lower levels of perceived system responsiveness both prior to and following a referendum. Still, the effect was considerably weaker in 2020, which supports H2b. In terms of process-opinion, it had no significant effect in 2020 after being highly significant in 2018. Thus, the findings in Table 2 do not entirely support H2a. The post-referendum dummy is significant in the pooled model, which confirms H1a. Both interaction terms are significant and, when looking at marginal effects, they are in the directions predicted in H1b and H1c (see Figure A1 and A2).
Turning now to the other form of political support, political trust, and trust in politicians (see Table 3). Our findings show that being against the merger predicts lower levels of trust in politicians in both datasets, albeit with a weaker effect in 2020 than in 2018, while being in favour of the referendum was a significant predictor of lower trust in politicians solely in 2018. These findings are in line with H2a and H2b. The post-referendum dummy has a strong significant effect in the pooled analysis and thus confirms H1a. The interaction terms are both significant and the marginal effects are in the directions expected by H1b and H1c (see Figure A1 and A2 for interaction plots).
Finally looking at trust in political parties (see Table 4), merger opinion was a significant predictor in both datasets (stronger in 2018 than 2020), showing that being against the merger predicts lower levels of trust in political parties. Being in favour of the referendum was a significant predictor for lower levels of trust in political parties in 2018, but not in 2020. These findings support H2a and H2b. The post-referendum dummy is again a strong significant factor and confirms H1a. The interaction effects are also both significant and the exploration of marginal effects (see Figure A1 and A2) confirm H1b and H1c. Table 2. Individual-level factors as predictors of system responsiveness. In summary, the analyses depicted in Tables 1 to 4 provide strong support for H1, i.e., the level of political support increased after the referendum (tested with post-referendum dummy). Likewise, H1b is supported in all tables and thus shows that being on the winning side of the merger issue leads to an increase in political support. H1c is supported in all but Table 1, which rather strongly indicates that being in favour of the referendum is associated with an increase in political support after the referendum has been held. H2a does not receive conclusive support in the analyses, and it is uncertain whether issue-or process opinion is the main driver of political support. Often, they are strongly correlated -i.e., citizens are favouring the referendum if they believe it will provide the result they want on the issue -which is likely an explanation for this. Finally, H2b is supported in all analyses.

Discussion
From a democratic theory perspective, the losers in a referendum are expected to remain supportive of a political system when conditions about perceived procedural fairness are being met (Dahl 1989). Simply put: if the fight was fair, it should be easier to accept defeat. The empirical evidence, however, does not always seem to support this assumption (Esaiasson et al. 2019;Marien and Kern 2018). Even though the losers in the referendum studied here were able to have a say in the process, the polarising process could be argued to some extent inhibit positive effects in terms of political support (see here also Marien and Kern 2018). Still, given the relatively low Table 4. Individual-level factors as predictors of trust in political parties. effect sizes in all regression models a lot of the variance remains unexplained. However, interpretation of effect sizes is context dependent, and it is far from unusual for similar types of studies to report low levels of explained variance (Ferguson 2009;Rosnow and Rosenthal 2003).
Referendums are considered to function as a vital check on political power (Qvortrup 2017, 142), and prior to the referendum, being against the merger and in favour of the referendum both predicted lower levels of external efficacy and political trust. Hence, it is quite clear that the citizens that were against the merger expected to win in a referendum and hence supported it for instrumental reasons (Werner 2020) and were consequently also pleased with the referendum, following the eventual positive political outcome on their behalf. They did simultaneously express lower levels of political support in both datasets, and it was therefore not surprising that they did not trust the politicians to act accordingly to what they believed to be the majority will. Still, when institutions do follow the result of a referendum, as was the case in our local setting, the policies should be more aligned with the public preferences and the levels of trust increased accordingly.
Among citizens being against a merger and in favour of a referendum, the positive effects on political support were substantial. Studies have shown that participating in democratic innovations might foster higher levels of political trust (Bowler and Donovan 2002;Hero and Tolbert 2003), and our results partly adds support to this. Nevertheless, which side of the issue the citizens have been on determines the effects, which is in line with the findings presented by Marien and Kern (2018) and Esaiasson et al. (2019). There is thus a clear winner-loser effect at play here, with the winners becoming more positive towards the political system while the losers remained stable (see also Blais and Gelineau 2007). Additional analyses further confirmed that the effects of merger opinion and referendum opinion on local political trust and evaluations of the responsiveness of the local political system were similar to our findings towards the more general types of political support (see Tables  A8-A10). A core aspect of democracy is the alignment between voter preferences and government output (Leemann and Wasserfallen 2016), and it is worth speculating what the effects would have been given a situation where the municipal assembly would have gone against the majority of the public and voted in favour of a merger. It is likely that the negative effects on political support would have been substantial, as most of the public would then not consider the outcome as legitimate. This especially among the part of the public who adheres to a more populist, over an elitist (Schumpeter 1942), view of democracy.

Limitations
There are some substantial limitations with our datasets that have been previously mentioned but that needs to be addressed again. First, the lack of panel data and the fact that the second survey was collected 18 months following the referendum and around 30 months after the first survey, makes it challenging, to connect the post-referendum levels of political support directly to the outcome of the referendum process (Gliner, Morgan, and Leech 2009;Leatherdale 2019). Hence, the positive post-referendum values in terms of political support could also be partly explained by other events, such as a general trend towards higher political support amid the Covid-19 pandemic. This is also something that has been noted in many parts of Europe during the initial phase of the lockdowns (Bol et al. 2021;Schraff 2020). The long time-gap between the referendum and the post-referendum survey also creates a situation where it is uncertain to what extent the responses are derived from the referendum outcome and the merger process, and not evaluations of what has happened in the municipality since the referendum.
Hence, we are hesitant to draw any larger general conclusions about the potential effects derived from a referendum process on political support based on this survey data. Nevertheless, as our findings do clearly show that there is a winner-loser gap in terms of the effects in our 2020 dataset, we would still argue that our study contributes to that line of research. Nevertheless, if possible, future studies would benefit from panel data and from the pre-and post-referendum measurements to be very close to the referendum in order to raise the internal validity. Furthermore, a comparison between the respondents in the pre-and post-referendum datasets (see Appendix Table A5) indicates there being a risk of selfselection bias. This possibility was partly counteracted by using raked weighting in our analyses. However, a risk of there being systematic differences between respondents and non-respondents in our datasets remains. Future research should consider more analyses of non-respondents through follow-up surveys and the utilisation of nonresponse weights when having access to panel data. A further limitation of our analyses is that we focused on effects for all respondents as a whole and disregarded analyses of subsamples. Future studies should thus focus on potential effects that reside at sub-group level to further our understanding of referendum effects on political support.
Furthermore, we acknowledge that political trust and external efficacy can be used to predict merger opinion and referendum opinion, e.g., there is also a reversed causation between the main variables tested in our regressions. Thus, given the heavily contested issue this was to be expected as the start of the merger discussions, with the real possibility of a merger materialising, most certainly had a negative effect on both political trust and external efficacy among those opposing a merger. The start of the merger discussions at the political level was already a step too far for many opposing the merger, and a causal relation could therefore be expected. However, if the survey data had been collected before the start of the merger discussions at the political level instead, we are far from certain that there would have been a similar type of causal relation. Unfortunately, given the lack of adequate data we are not able to explore that possibility further.

Conclusions
Given the importance of political efficacy and political trust for the functioning of democracy, we would expect the elected representatives to be careful when using democratic innovations as a part of the decision-making processes. As a referendum creates winners and losers, it should be considered extra important that there are high levels of diffuse support (Easton 1965) among the citizens on the losing side to prevent dissatisfaction from turning into to more serious expressions of disappointment (Van der Eijk and Rose 2021, 105). Hence, if the goal is political stability in the local community, the best way to secure it over time is perhaps not to transfer the most important policy decisions straight to the public jury. As representative political systems could be considered fragile, it is therefore far from optimal to increase the level of uncertainty regarding where the political power finally lies. Consequently, by channelling decision-making powers directly to the citizens policy differences within the population becomes more apparent, thus contributing to polarisation. Nevertheless, as the effects on political support derived from issue preferences are clearly weaker following the referendum, another possible interpretation is that political support becomes less conditional after a referendum. Hence, as the importance of the referendum issue declines so does its relationship with political support. As our study has only focused on the local level and short-term effects, more research is solely needed to further assess the long-term effects on attitudes of importance at system level following a referendum experience. Don't want to say (x) Did not have the right to vote (x) 11. Do you think it was the right choice by the municipal council in Korsholm to organise a public referendum about a municipal merger with Vaasa?
Yes (1) No (0) Don't know (x)      Table A9. Individual-level factors as predictors of trust in municipal council.  Table A10. Individual-level factors as predictors of local system influence.