‘You Can ’ t Always Get What You Want ’ : The effects of winning and losing in a referendum on citizens ’ referendum support

In Western democracies, many citizens support the use of referendums. However, as referendums create satis- fying outcomes for citizens with majority views, they could generate ambivalent feelings among voters for the minority option. Little is known about the effects of winning or losing a referendum for citizens ’ referendum support. This article analyses multiple-wave survey data from five referendums in Bavaria (Germany), Finland and the Netherlands. The findings show that losers ’ referendum support decreases in nearly all cases, but there is very limited evidence for an increase of winners ’ referendum support. Nevertheless, the results clearly indicate a winner-loser gap, suggesting that referendums have relatively more positive effects for winners ’ than for losers ’ referendum support. As such, this article extends previous findings of the non-stability of referendum support. As the legitimacy of democratic institutions depends upon losers ’ consent, these findings have important implica- tions for the democratic potential of referendums.


Introduction
Over the last decades, the use of referendums has increased worldwide (Altman, 2017;Qvortrup, 2017). Various opinion surveys suggest that citizens in Western democracies are generally favourable towards referendums (e.g. Donovan and Karp, 2006). However, recent referendums in e.g. Colombia, Hungary, the United Kingdom and the Netherlands have reignited the debate about the desirability of this direct democratic instrument (Hendriks et al., 2017). For proponents, the referendum is a useful addition to representative democracy or even "the salvation of democracy", while critics fear that introducing referendums harms and undermines the system of representative democracy (Hendriks et al., 2017, pp. 14-17;Van der Meer, 2017, pp. 92-95).
Some are further concerned that referendums lead to a "tyranny of the majority", in which the 51 percent decides and the voice and rights of minorities become oppressed (Hendriks et al., 2017, p. 16). Referendums create winners and losersespecially in referendums on EU membership, secession or other crucial political decisions, the outcomes could provide important advantages for those who win, but the costs could be high for those who lose. It is however unclear what happens to referendum support after citizens are confronted with a victory or a defeat in a referendum: does winning a referendum strengthen citizens' support for referendums and does referendum support erode when it turns out that one can also lose referendums?
While little is known about whether and how winning and losing in a referendum affect citizens' attitudes towards referendums, the notion of a winner-loser gap in elections is well-established. After elections, those who vote for a winning candidate or party ('winners') are generally most satisfied with the way democracy works, while 'losers' are most likely to be dissatisfied with democracy and to favour changes of the existing electoral rules (see e.g.: Anderson et al., 2005;Anderson and LoTempio, 2002;Anderson and Tverdova, 2001;Blais and G� elineau, 2007;Esaiasson, 2011). Compared to the large bunch of articles and books about winning and losing in elections, the literature about winning and losing in referendums, to my knowledge, only consists out of two published articles (Marien and Kern, 2018;Nadeau et al., 2019). These studies have investigated the effects of winning or losing a referendum for political attitudes as political support and satisfaction with democracy. As these studies do not consider citizens' support for referendums as an outcome variable, it is unknown whether and to what extent a victory or a defeat in a referendum affects the support for the institution of a referendum.
As such, this article is the first to contribute to the rather unexplored line of research by studying the effects of losing and winning in referendums on citizens' referendum support. Its central research question reads as: To what extent does winning or losing a referendum affect citizens' referendum support?
To answer above question, the study combines insights from the literature about electoral winners and losers with empirical multiplewave public opinion data from three national referendums in the Netherlands, the EU membership referendum in Finland and one referendum on a federal state level in Bavaria (Germany). Besides studying five referendums in three countries, this article's methodological strength lies with its use of panel data, as it makes the statistical analyses very suitable for establishing causal inferences and for controlling for confounding factors (Daniller, 2016;Esaiasson, 2011;Singh et al., 2012).
By answering this question, this study provides further empirical ground for a relatively new and emerging stream of literature that focusses on the fluidity and dynamics of citizens' levels of referendum support. Whereas extant literature has paid considerable attention to the question who supports referendums, there is limited knowledge about the stability of and changes in citizens' views towards referendums. Predominantly, the literature has focused on which citizens are most supportive of referendums, resulting into competing explanations and lively debates (e.g. Bengtsson and Mattila, 2009;Dalton et al., 2001;Donovan and Karp, 2006;Hibbing and Theiss-Morse, 2002;Norris, 1999). Within this overwhelming body of literature, the stability of citizens' preferences for referendums did not receive much attention (cf. Werner, 2019).
However, various authors have stressed the existence of a non-stable component in citizens' referendum preferences. For example, Werner (2019) showed that referendum support is more fluid and context-dependent than generally assumed and that instrumental considerations shape citizens' preferences for decision-making via referendums: they are influenced by the issue of a referendum and by the individual's perceptions of having a majority opinion. Building upon these insights, this study further explores the stability of public opinion towards referendums by analysing how citizens' referendum support is (re)shaped after being confronted with favourable or unfavourable outcomes of this direct democratic form of decision-making.
Studying the effects of winning and losing in referendums on referendum support is also relevant for evaluating the functioning of referendums. In a democracy, 'you can't always get what you want'. Within democratic societies, it is therefore crucial how losers deal with their loss in decision-making. After all, it is of great importance for the legitimacy and stability of democratic institutions, as referendums, that they could rely on the support and consent of losers, even despite disappointment with its political outcomes (cf. Anderson et al., 2005, p 7;Blais and G� elineau, 2007). This article's outline will be as follows. The article first provides an overview of the scientific debate about referendum support, followed by an elaboration upon the literature about winning and losing in elections and referendums. In the remaining part, the research design will be discussed, while the article ends with the presentation of the results and a conclusion and discussion. The most important finding of this article is that losing a referendum has negative consequences for losers' support for referendums, while a victory in a referendum does not necessarily lead to more referendum approval among winners.

Public support for referendums: the scientific debate
Support for forms of direct democracy, such as referendums, is generally high among citizens in Western democracies. Both in Europe and the United States, national surveys report that solid majorities of citizens favour introducing national referendums (Schuck and De Vreese, 2015;Smith et al., 2010;Werner, 2019). However, some groups of citizens are more supportive of direct democracy than others. An important issue in the academic literature is how to explain differences in citizens' referendum support. In this debate, two main views exist: the cognitive mobilization (new politics) hypothesis and the political dissatisfaction hypothesis (see e.g.: Bengtsson and Mattila, 2009;; Coff� e and Michels, 2014; Donovan and Karp, 2006;Schuck and De Vreese, 2015).
The cognitive mobilization (new politics) hypothesis assumes that popular support for direct democracy has increased because of societal modernization: citizens have become higher educated and postmaterialist values have become priorities in Western societies over materialist values. Because of their post-materialist values, citizens are more sceptical about hierarchical authorities, such as parties and representative systems, and would favour a new participatory style of politics with an emphasis on direct action and participation. Because of their higher education, citizens will also feel more competent to directly take part in decision-making (Norris, 1999, see also: Bengtsson andMattila, 2009;Dalton et al., 2001). Therefore, the cognitive mobilization hypothesis claims that support for direct democracy and referendums will be higher for those with a higher education, with more cognitive skills, better access to political information and higher levels of political interest and efficacy.
On the contrary, the political dissatisfaction hypothesis supposes that referendum support is higher for citizens that are dissatisfied and frustrated by representative democracy (Bengtsson and Mattila, 2009). These citizens favour changes of the existing political system and the referendum provides them an alternative to support (Schuck and De Vreese, 2015). The political dissatisfaction hypothesis is in line with the larger referendum support among voters of protest parties (Dalton et al., 2001) and among those with populist attitudes (Jacobs et al., 2018a). Further, as the lower educated citizens are suggested to be most dissatisfied with politics, they would be more strongly in favour of referendums (Coff� e and Michels, 2014).
The empirical evidence for these hypotheses, however, is diffused. For the case of Germany, Dalton et al. (2001) found a strong link between political dissatisfaction and support for direct democracy. In contrast with the cognitive mobilization hypothesis, support for direct democracy decreases as education and political interest increase. More recent studies provide further evidence for the political dissatisfaction hypothesis (Bengtsson and Mattila, 2009;Coff� e and Michels, 2014). On the other hand, Bowler et al. (2007, p. 360) showed "that approval of direct democracy comes from the politically engaged, as much as, if not more than, it comes from people who are politically disaffected or angry about how democracy is currently working" in an analysis with cross-national data from sixteen nations. In a study of referendum support in six Western democracies, Donovan and Karp (2006) found mixed evidence for the cognitive mobilization hypothesis. In New Zealand, Canada and Switzerland, support was higher among those who were more interested in politics. On the contrary, the effects of political interest in Norway were reversed. Furthermore, no evidence was found that education has a positive impact on support for direct democracyin four of the six countries, the effects were negative. Donovan and Karp (2006, p. 685) therefore argue that "neither the cognitive mobilization theory nor the political disaffection theory explains much of the variance in levels of support".
Although the cognitive mobilization and political dissatisfaction hypotheses provide competing explanations for citizens' levels of referendum support, both theories hold a similar assumption that referendum support is relatively stable within individuals (Werner, 2019). As such, they both account for differences in referendum support between individuals, but are poorly suited to explain differences in referendum support within individuals.
Recent research however suggests that referendum support has a fluid, non-stable and context-dependent component, rather than being a fully stable and constant attitude over time (e.g. Smith et al., 2010;Werner, 2019). In a survey experiment among U.S. citizens,  show that the relatively high levels of referendum support are ethereal. As well as for other electoral reforms, they found that referendum support significantly dropped when the reform proposal was framed in terms of risk or (potential) loss by slightly rephrasing the question wording of the survey items . Furthermore, Smith et al. (2010) find that referendum support is also shaped by short-term electoral fortunes: being a winner or loser in the electoral arena is an important force that can either increase or decrease support for a national referendum. Based upon multiple-survey data, they show that U.S. citizens "supporting the losing presidential candidate (…) are considerably more likely to favour a national referendum after the election, but not before, as their loser status had not yet been conferred upon them" (Smith et al., 2010, p. 519).
In a recent contribution to this body of literature, Werner (2019) argues that individuals take instrumental considerations into account when expressing support for forms of political decision-making, such as referendums. According to Werner (2019, p. 14), referendum support could not fully be considered as a stable political attitude, but further include a non-stable and context-dependent component and "citizens seem to think more pragmatically about decision-making procedures than previously assumed". The results of the study show that citizens' support for the use of referendums varies across policy proposals and that support for a certain policy change and the belief to hold a majority opinion on a policy proposal increase citizens' support for the use of a referendum on these specific policy proposals. Most citizens are neither a full opponent or full supporter of referendums, but rather belong to the category of "switchers", being more supportive of referendums about certain topics than about other issues (Werner, 2019).
Next to switching from referendum opponent to referendum supporteror vice versabased upon the policy proposal at hand, citizens' preferences for referendums could further change due to their experiences with referendums and specifically with winning or losing a referendum. Whereas Werner (2019) focused upon how individual expectations of achieving a desired policy outcome affect referendum support, this study moves its attention to the effects of individual experiences with (un)desired policy outcomes in the aftermath of referendums. Although the worldwide use of referendums is clearly on the rise since the 1960s (Altman, 2017;Qvortrup, 2017), it is still not tested to what extent winning or losing a referendum (re)shapes citizens' preferences for this direct democratic way of decision-making.

Winners and losers in elections and referendums
While little attempt has been made to study the effects of citizens' status as a winner or loser in referendums, a large body of literature has examined differences between electoral winners and losers. Several scholars have investigated the effects of people's status as winner or loser in elections for attitudes towards government, democracy and political institutions. Their empirical findings are very consistent: differences between winners and losers have been established for a wide range of political attitudes among a wide variety of democracies. (see e.g.: Anderson et al., 2005;Anderson and LoTempio, 2002;Esaiasson, 2011).
Winners of an election, those who vote for a winning party or a winning (presidential) candidate, have the most positive attitudes and evaluations about the political system (Anderson et al., 2005). An electoral victory positively affects satisfaction with democracy (Anderson and Guillory, 1997) and further increases levels of political efficacy and political trust (Anderson and LoTempio, 2002). In a similar logic, losers are most likely to be dissatisfied with the status quo and to favour changes in the existing institutional arrangements (Anderson and Tverdova, 2001;Smith et al., 2010). Furthermore, losers have generally less faith in the fairness and efficacy of elections and show less support for democratic principles (Anderson et al., 2005, pp. 47-49).
From a policy perspective, the winner-loser gap is explained by the fact that winners are more likely to expect the implementation of their preferred policies and are, therefore, happier, more trustful and more satisfied than losers (Anderson et al., 2005, p. 3). However, as Singh (2014) and Singh et al. (2012) argue, the effects of winning or losing elections could further be largely psychological. Where winning evokes positive emotions, loss is accompanied by negative feelings, such as sadness, disappointment or even anger (Singh, 2014;Singh et al., 2012).
Compare it to a soccer match: fans of the winning team will enjoy the game more, while supporters on the losing side could become disappointed and more inclined to criticize the referee and his decisions and could even turn away from soccer. These emotions are not limited to soccer stadiums, but are also present at election day (cf. Singh, 2014).
Despite evidence for the winner-loser gap, a main question remains under which conditions losers are more likely to consent to the election outcome and voluntarily accept their defeat (cf. Esaiasson, 2011). Empirical data shows that the negative effects of losing are conditional: Anderson et al. (2005) found that losers are under some circumstances remarkably willing to accept an unfavourable electoral outcome and they consent to being governed by those on the winning side. According to Anderson et al. (2005, Chapter 10), these graceful attitudes of electoral losers are affected by both a country's political context (e.g. political and electoral institutions, length of democratic experience) and voters' individual characteristics and attitudes.
Studies about the effects of winning and losing in referendums are seldom. One of the few exceptions is the work of Marien and Kern (2018). By studying voters' opinions before and after a local referendum in a Belgian neighbourhood, they find that levels of political support among winners of a referendum increase after the vote and those of losers remain stable. Changes in political support were not driven by the perceived fairness or perceived influence of the referendum. Another study by Nadeau et al. (2019) focusses upon which losers are more willing to gracefully accept their defeat in a referendum and confers legitimacy to the winning option of the referendum for the case of the Brexit referendum in the United Kingdom. They find that so-labelled "graceful losers" were often "soft" Remain supporters who were more critical of the EU, less emotionally engaged in the debate, more optimistic about the consequences of a Leave victory, and torn between the two options until the end of the campaign. According to Nadeau et al. (2019), these graceful losers form 37 percent of the entire population of Remain voters during the Brexit referendum. These graceful losers reported levels of democratic satisfaction that were way higher than those of 'sore' losers and that were even slightly higher than those of the winners of the Brexit referendum.
These findings suggest that a similar winner-loser gap exists for referendums as for elections, but some nuances should be made. Marien and Kern (2018) find that losing the referendum did not have implications for losers' levels of support, while Nadeau et al. (2019) show that in the case of the Brexit referendum nearly two-thirds of the losers refused to endorse the result of the referendum and that losing a referendum has negative effects for their levels of satisfaction with democracy.
Nevertheless, both studies are solely based upon a case study of one single referendum. These studies have investigated the effects of winning and losing a referendum for various attitudes, as views about the legitimacy of the referendum, political support and satisfaction with democracy. However, neither one of the two has looked into the effects of winning or losing a referendum for citizens' views about the institution of a referendumthus far, it is an issue that remained undiscovered in the small bunch of academic literature about the effects of winning and losing in referendums.

Methods
This section discusses this article's research design, cases and data. Most research on the political attitudes among electoral winners and losers has drawn upon cross-sectional post-election surveys to aggregate differences in perceptions between winners and losers. Cross-sectional data is however poorly suited to explain changes in perceptions among winners and losers. To trace the dynamics of winners' and losers' attitudes during an electoral or referendum process, panel data is needed with a before-after design which has identical measures of system support or other attitudes in both waves of the survey (cf. Daniller, 2016;Esaiasson, 2011). Some studies in this field have used such a longitudinal survey design (e.g. G� elineau, 2007;Daniller, 2016;Singh et al., 2012). As such an approach is very suitable to make causal inferences, this study uses a before-after design with multiple-wave survey data as well.
Hypotheses are tested by using public opinion data collected around five referendums in three different countries: the Dutch referendum about the Intelligence and Security Services Act (2018), the Dutch referendum about the Ukraine-EU Association Agreement (2016), the Bavarian non-smokers protection referendum (2010), the Dutch European Constitution referendum (2005) and the Finnish EU membership referendum (1994). By drawing upon insights from multiple referendums, this study enables to test whether the findings hold beyond the context-specific elements of a single referendum. Its comparative framework allows to inference relationships that exist among multiple conditions in which referendums could take place. The selection of these five cases has two major advantages, as these cases (1) reflect referendums that are hold in national settings that are largely typical and representative for most European democracies in terms of organizing referendums and (2) also differ in multiple aspects to provide greater probability that the findings are not biased by the case-specific features of the referendums under study.
By selecting the Dutch and Finnish referendums, this study follows Jacobs's (2018, p. 490) call to referendum researchers to move beyond the "few fairly extreme 'usual suspects' countries" as Switzerland and the United States, that regularly organize referendums, and to focus on "the broader set of countries that occasionally hold referendums". The Netherlands offers such a case as the country is rather representative for most European democracies where referendums are not a frequent part of political decision-making, but occasionally occur (cf. Werner, 2019). These three referendums were the first national referendums in the country since the beginning of the nineteenth century (Hendriks et al., 2017). Also Finland belongs to this large set of countries with occasional experience with national referendums, although the use of direct democracy in Finland is -at a national levelmore sparse than in the Netherlands. The 1994 EU membership referendum was the first and only national referendum in Finland since 1931 (Suksi, 1996). Furthermore, as EU issues are a popular topic for referendums in various countries (see: Hendriks et al., 2017, Chapter 6), the three EU-related referendums in this study provide examples for these bunch of referendums.
In addition to the Dutch and Finnish referendums, the Bavarian nonsmokers protection referendum was selected as a deviant case that provides information about the winner-loser dynamics in a referendum in a federal state with relatively large experience with direct democracy. Although national referendums are not possible in Germany, local and regional referendums are very common and Bavaria is a well-known case of a Bundesland with large referendum experience (Glaab, 2013). As such, the selection of these five referendums largely reflects a wider population of referendums in a European context. Furthermore, the five referendums are different in other aspects that allows to test the same hypotheses under different conditions. The case selection includes two referendums that have been initiated by parliament, two that have been initiated by citizens to reverse a parliamentary decision and one that has been initiated by citizens to propose new legislation. As Table 1 shows, the margins of defeat are quite similar for four of the five referendums with the exception of the most recent Dutch referendum. The referendums do further largely differ in terms of turnout rates. Two of the referendums resulted in a majority for the "Yes" vote, while the other three had a "No" victory as its outcome. Crucially, for all referendums, panel data was available including the relevant information on the dependent variable in both a pre-and postreferendum wave.
The following datasets have been used for the purposes of this study. For the Dutch Intelligence and Security Services Act referendum, data was used from the National Referendum Survey that was fielded among the Dutch LISS panel (CentERdata, 2018a(CentERdata, , 2018b(CentERdata, , 2018cJacobs et al., 2018b). In the weeks before the referendum, two short surveys were distributed among panel members, while a longer survey was held shortly after the day of the referendum. For all of these three waves, 2.000 to 2.300 respondents completed the questionnaire. For the Dutch Ukraine referendum, data was obtained from the Election Survey Ukraine referendum, which was also conducted among members of the LISS Panel (CentERdata, 2016a(CentERdata, , 2016b(CentERdata, , 2016cJacobs et al., 2016). Similar as the 2018 referendum survey, the survey was fielded twice during the campaign period in the run up to the referendum, followed by a post-referendum questionnaire directly after the vote. In these waves, the N varies between 2.300 and 2.600 -a number of 1.856 respondents participated in at least one of the pre-wave surveys and the post-wave survey.
With regard to the Bavarian non-smokers protection referendum, public opinion data was collected by Infratest dimap at the behest of Otto-Friedrich-Universit€ at Bamberg (Hilmer and Hofrichter, 2010). A pre-referendum survey was conducted by 4.000 respondents and 2.003 of them participated in the post-referendum survey. Data from the Dutch Referendum Study 2005 (Aarts andVan der Kolk, 2005/2007) was used in case of the Dutch European Constitution referendum. Here, a sample of the electorate was surveyed by GfK Benelux and the survey consists out of a pre-referendum and a post-referendum measurement -1.224 respondents participated in both waves. For the Finnish EU membership referendum, data was used from the referendum study of the Finnish Social Science Data Archive (Pesonen and S€ ankiaho, 2002). In the pre-referendum wave of the survey, 1.559 respondents participated and 1.316 conducted the post-referendum survey. To deal with missing observations, a listwise deletion of respondents with missing data on the key variables of interest has been employed for all cases (see: Allison, 2001).
In this study, citizens' support for referendums serves as dependent variable. Table 2 shows that such attitudes have been measured differently across the various surveys. In the 2016 and 2018 Dutch referendum studies and the 2010 Bavarian referendum study, items are related to citizens' general approval of referendums. In the Finnish referendum survey, however, items were included that were more specifically formulated or related to the EU membership referendum. The 2005 Dutch referendum study includes survey items that measure general referendum approval and a specific attitude about the European Constitution referendum itself. There are further differences in terms of level of measurement: some items are measured on a dichotomous scale, while others on a five-point Likert scale. However, all items have in common that they were asked before and after the referendum among largely the same respondents. If necessary, survey items have been recoded in such a way that the highest values reflect the most positive views towards referendums and that the lowest values reflect the lowest Note: For the two most recent Dutch referendums, blank votes are officially valid and are considered as part of the referendum outcome. As an effect, combining the percentages of both "Yes" and "No" votes does not precisely equal 100 per cent.
levels of referendum support. 1 To distinguish winners and losers, this study uses the post-referendum measurement in which respondents indicate what they had voted in the referendum. Respondents who indicated to have voted for the majority option are classified as "winners" and those who indicated to have voted for the other option as "losers". Such an approach is however debated, as respondents' recall of their vote choice in a post-referendum survey is self-reported behaviour and does not reflect actual vote choice per se. Some, as Daniller (2016, p. 154), use pre-election preferences instead of post-election reported vote choice to differentiate between winners and losers to ensure that "respondents' choices were not potentially tainted by knowledge of the election outcome". Based upon pre-election and post-election Eurobarometer surveys, Anderson et al. (2005) however show that large majorities of respondents report the same voting choice before and after the election. Anderson et al. (2005, p. 35) therefore argue that "assuming that there is some degree of measurement error inherent in any measure of vote choice, there seems to be little difference between classifying winners and losers on the basis of vote recall (past vote) or vote intention (future vote)." In addition to winners and losers, referendums create a third group of citizens: the non-voters. Studies into the effects of winning and losing in elections have only paid limited attention to non-voters (Rich and Treece, 2018). The percentage of non-voters at referendums is often larger than that at elections (Leiniger, 2015). However, studying the opinions of non-voters provides several challenges. The non-response from non-voters is a common problem in survey research, very often resulting into their underrepresentation in surveys. The problem of underrepresentation of non-voters is even larger for multiple-wave survey data. As Jacobs et al. (2016, p. 69) explained for the case of the Dutch Ukraine referendum, respondents who participate in the pre-wave survey prior to the referendum were better informed about the referendum than most of their fellow citizens and, therefore, more likely to vote. As a result, the numbers of non-voters in these surveys are relatively small and it is often unclear to what extent the sample is representative of the total population of non-voters. For the sake of completeness, this study reports levels of referendum support among non-voters before and after referendums, but these results need to be interpreted with a certain degree of caution. Finally, the very small numbers of respondents who indicated that they voted blank or invalid were excluded from the statistical analyses.
The study includes control variables for demographic background. As respondents' scores of post-level referendum support can be compared with prior information about their support levels, it reduces to a large extent the need to control for background variables that may account for differences between respondents in referendum support. However, to ensure that post-referendum changes are explained by citizens' status as winner or loser and that these changes are not caused by differences in underlying demographic characteristics between these groups, there will be controlled for standard demographic variables as gender, age and level of education. Specifically, level of education is suggested to have a significant effect on citizens' support for referendums (Coff� e and Michels, 2014).
In addition, this study also controls for two political attitudes that are suggested to affect citizens' referendum support: political interest and political trust. Both attitudes are linked to one of the two main explanations for differences in referendum support between individuals. In line with the cognitive mobilization hypothesis, citizens with higher political interest will be more supportive of referendums than those with lower levels of interest Donovan and Karp, 2006;Norris, 1999). In contrast, the political dissatisfaction hypothesis suggest that referendum support mainly comes from citizens with the lowest levels of political trust Dalton et al., 2001). Although the focus of this study is on differences in referendum support within individuals and not between individuals, it is important to take differences in political interest/political trust between citizens into account. As levels of political trust and political interest are likely to 1 ¼ Government and Parliament should have decided 2 ¼ Referendum was the right thing In your opinion, should politicians abide by the outcome of the referendum or would you be willing to let Parliament decide against the outcome? 1 ¼ Abide by the outcome 2 ¼ Parliament can decide 1 As the 2018 and 2016 Dutch referendum surveys consisted out of two prereferendum waves, respondents' prior-level support for referendums is reflected by the mean score of these two observations. For respondents with only a score for one of the pre-wave measures, the particular score is used. As such, the sample size will be enhanced and the number of missing observations will be reduced.
be unequally distributed between winners and losers of some of the referendums, it is important to control for differences in political interest/political trust between winners and losers. Unfortunately, data about political interest was missing for respondents in the case of the two most recent Dutch referendums: these statistical analyses were therefore run without controlling for political interest. As this study uses multiple datasets, it is important to note that there are differences in measurement of these indicators across the different cases. An overview of measures for all control variables could be found in Appendix A1.

Results
For each of the five referendums, Tables 3a and 3b shows the levels of referendum support of winners, losers and non-voters shortly before and after the referendum. The results confirm to a large extent that losing a referendum decreases referendum support. An increase of winners' referendum support is only partially supported by the findings.
After the Finnish referendum on joining the EU, referendum support among winners significantly increases (an increase of 3.4 percentage points, McNemar test, p ¼ 0.035; and an increase of 15 percentage points, McNemar test, p < 0.001). For the Dutch European Constitution referendum, winners' general and specific approval of holding the referendum about the new Constitution also increases (F ¼ 15.837, p < 0.001, and an increase of 6.5 percentage points, McNemar test, p < 0.001). On the other hand, referendum support remains at the same level among winners of the Bavarian non-smokers protection referendum (F ¼ 0.726, p ¼ 0.395, F ¼ 0.215, p ¼ 0.643) and among winners of the Dutch Intelligence and Security Services Act referendum (F ¼ 0.204, p ¼ 0.652). For winners of the Dutch Ukraine referendum, there is even a significant decrease in their referendum approval (F ¼ 40.958, p < 0.001).
These findings show remarkable differences between the effects of a referendum victory for citizens' generic approval of referendums and their specific evaluation and support of the referendum at stake.
Whereas winning a referendum does not have a clear effect on overall referendum support, the results clearly indicate that citizens are more positive about holding the referendum about a specific policy proposal after they have won a direct vote about that particular issue. However as most of the surveys include only measures for general referendum support or only measures for specific referendum attitudes, an alternative explanation is that case-specific elements and context of a referendum cause these different effects of winning a referendum. As the 2005 Dutch referendumthe only case with information about both attitudes shows an increase in both winners' general referendum approval and specific approval for the referendum about the European Constitution, this alternative explanation is more plausible.
The five cases offer greater clarity about the negative effects of losing a referendum for referendum support. After four of the five referendums, support for referendums decreases significantly among losers after their defeat. There is only one exception to the rule: while losers of the Dutch European Constitution referendum became less supportive of the referendum about the Constitution after the results of the vote were known, their approval of referendums in general did not significantly drop after their defeat (F ¼ 2,742, p ¼ 0.098).
The effects of a referendum for non-voters' views are not uniform across the five cases. The findings of the Dutch Ukraine referendum and the Bavarian non-smokers protection referendum suggest that non-voters will be more negative about referendums after holding a referendum. After the Dutch Intelligence and Security Services Act referendum and the Finnish EU membership referendum, non-voters' referendum support decreased as well, but the effects were insignificant. On the contrary, the Dutch European Constitution referendum even shows a significant increase in non-voters' general approval of referendums.
Importantly, these findings further illustrate that most citizens are positive or fairly positive towards the institution of a referendum in the run-up to these referendums. In nearly all cases, citizens' referendum support remains relatively high after the vote. Although there is clear Scores reflect mean scores (with standard deviations). All indicators were measured at a five-point Likert scale with 1 ¼ lowest level of referendum support and 5 ¼ highest level of referendum support. *p < 0.05. **p < 0.01. ***p < 0.001 (Repeated Measures-ANOVA).

Table 3b
Specific support for referendums before and after referendums.
Netherlands (2005) Finland (1994) Finland (1994) Approval of referendum about the European Constitution Referendum was the right thing Politicians should abide by the outcome Based upon above findings, can we now thus say that losing a referendum (generally) has a negative effect on referendum support, while winning a referendum only occasionally seems to have a limited positive effect on referendum support? Such a position could be somewhat misleading because of the occurrence of ceiling effects and floor effects. These effects make it difficult to compare between shifts in the perceptions of winners and losers. For example, the chance of a significant increase in referendum support becomes smaller as respondents already indicate higher values of referendum support in the survey wave prior to the referendum (a ceiling effect). Conversely, a floor effect means that the chance of a decrease in referendum support becomes smaller as respondents already have little support for referendums. Given the high levels of referendum support, there is a high chance of a ceiling effect for winners.
To be able to control for the higher values of referendum support prior to the referendums and to neutralize ceiling effects and floor effects, multiple linear and logistic regression models are very useful. Here, post-level referendum support serves as a dependent variable, while referendum support before the referendum is included as an independent variable in the regression models. The models include dummy variables for losers and non-voters, with winners used as the reference category. This makes it possible to compare the effects of a referendum on referendum support for these groups. In addition, these models take into account the socio-demographic differences between winners, losers and non-voters in terms of gender, age and level of education, as well as their differences in political trust and political interest. As a result, the effects could be attributed to the experience of winning and losing in referendums and not to possible differences in the composition of the groups of winners, losers and non-voters. Table 4 presents the results of these regression analyses for all cases. First of all, it shows that the support of citizens prior to the referendum is an important and significant predictor of their referendum support after the vote. In addition, the results show a clear gap between winners and losers. For each of the five referendums, it appears that losers have a significantly greater chance to report a decrease in referendum support than winners. In relative terms, a victory in a referendum clearly has a less negative impact on support for referendums than a defeat in a referendum.
In addition, three of the five cases show a "winner-non-voter gap": after the two most recent Dutch referendums and after the Bavarian referendum, there was a significantly greater chance of a decrease in referendum support among non-voters than among winners. The Finnish referendum shows that non-voters were less likely than winners to agree with the statement that politicians should abide to the outcome of a referendum, but the effect did not occur for their views about whether organizing a referendum was a good thing to do. The Dutch European Constitution referendum shows no significant differences between winners and non-voters in their shifts in referendum support.
Next to winning and losing, also political trust has an effect on postreferendum changes in citizens' preferences for referendums. Except for the Finnish referendum, four of the five referendums show that Table 4 The impact of a referendum victory or defeat on post-referendum changes in levels of support. Netherlands (2018) Netherlands ( Netherlands (2005) Finland (1994) Finland (1994) Approval of referendum about the European Constitution Referendum was the right thing Politicians should abide by the outcome

Coeff. (SE) Coeff. (SE) Coeff. (SE)
Prior-level support 2.095*** (0.21) 2.542*** (0.21) 1.529*** (0.20) Loser À 1.494*** (0.24) À 0.460* (0. Note: the dependent variable (DV) is respondents' post-level referendum support. All variables have been recoded to range between 0 and 1. The first rows report unstandardized coefficients from linear regression models, the second rows report unstandardized coefficients from logistic regression models. *p < 0.05. **p < 0.01. ***p < 0.001. organizing referendums has more negative consequences for referendum support of highly political trustful citizens than for those with the lowest levels of political trust. These effects are regardless of people's status of a winner, loser or non-voter in a referendum. Political interest does not account for much of the post-referendum changes in referendum support: only in the case of the Bavarian referendum, politically interested citizens were more likely than the least interested ones to become more supportive for federal referendums after the vote. In some cases, some of the socio-demographic variables affect postreferendum changes in referendum support. For the two Dutch EUrelated referendums, the referendums have a significant negative impact on the preferences of higher educated citizens, compared to those with lower levels of education. However, education has no effect on changes in referendum support for the other three referendums. The cases of the Bavarian and Finnish referendums suggest that referendums have more positive consequences for the referendum support of younger people, but these effects were only found for one of the two dependent variables that have been studied for these cases.

Conclusion and discussion
Various studies have underlined that vast majorities of citizens in Western democracies approve the use of referendums. Some however suggest that referendum support is rather fluid than stable Smith et al., 2010). Negative or positive experiences with institutions could reshape people's perceptions and political attitudes. Although the number of referendums has increased over the last decades, still very little was known about the effects of referendums and -more specifically, the effects of winning and losing in referendumson citizens' support for referendums.
This article provides limited evidence that a referendum victory leads to an increase of referendum support among winners, but stronger evidence that a referendum defeat results into a decrease in losers' support for the referendum. Table 5 summarizes the main findings of this study for all of the five cases. A decrease in losers' referendum support was found for all cases, although the Dutch referendum in 2005 only reports a decrease in losers' levels of support for one of the two survey items. Empirical evidence for an increase in winners' referendum support is limited. Higher levels of winners' referendum support are only found for the Finnish referendum in 1994 and for the Dutch referendum in 2005. However, all five referendums show a winner-loser gap in post-referendum changes in referendum support, suggesting that being on the winning or losing side in a referendum matters for referendum support. In addition to these main findings, experiencing a referendum also has a more negative impact on referendum support of citizens with the highest levels of political trust than for those with less trust. The findings further indicate that referendum support is relatively high across citizens. Despite the possible "referendum hangover" of losers, most of them still stay positive or relatively positive towards referendums after their defeat.
It is remarkable that the evidence for an increase in winners' referendum support is less strong than for a decline in losers' referendum support. A possible explanation for the strong effects of losing and the minimal effects of winning is a psychological one: negative experiences and emotions have a larger influence on us than positive events (Baumeister et al., 2001). People have an aversion to losing and therefore estimate losses greater than profits -"losing twenty dollars hurts more than finding twenty dollars pleases" (prospect theory, see: Vis, 2011, p. 335). On the other hand, the minimal effects of a referendum victory could also be explained on a statistical basis by the existence of a ceiling effect. In a relative sense, however, there appears to be a winner-loser gap for all five referendums. While support for referendums among winners did not often increase in an absolute sense, all five cases turned out to be significantly less negative for winners' referendum support than for losers' referendum support.
Overall, these findings show that insights about electoral winners and losers could also be applied to referendums and that studying political attitudes of winners and losers of referendums provides an interesting line of research (see also Marien and Kern, 2018;Nadeau et al., 2019). This study largely confirms that support for an institution decreased as people are confronted with the negative consequences of the institution (cf. Smith et al., 2010). It shows that citizens' referendum support is not solely based upon normative principles of citizens about the procedures of political decision-making, but also partly driven by citizens' experiences with referendums in practice and specifically by the outcomes that referendums produce for them.
This study does however not reject theories that treat citizens' referendum views as stable attitudes, such as the cognitive mobilization or the political dissatisfaction hypotheses. The findings also show that citizens' referendum support has a stable part, as prior-levels of referendum support are still a very strong predictor for citizens' views towards referendums in the aftermath of the vote. As such, this study's findings illustrate Werner's (2019, p. 3) notion that it is plausible "that citizens have a baseline attitude towards referendums which is shaped by values and evaluation of politicians generally", but "that process preferences also have a non-stable, context-dependent component that has so far remained understudied".
Still, this article faces two limitations. Its first limitation is that it only looks at the short-term effects of winning and losing referendums on referendum support. This does not necessarily mean that winning or losing a referendum has long-term effects for citizens' referendum preferences (cf. Van der Meer and Steenvoorden, 2018). Further research with long-term data is needed to determine to what extent winning and losing in referendums have an effect on citizens' referendum support in the long run.
As a second limitation, it should be stressed that this article only investigates five cases, due to the limited availability of suitable data. The question remains to what extent the results for these five referendumseach held within a European consensus democracyare generalizable. Further research could provide an answer to this question by studying the effects of winning and losing in referendums in other contexts, such as for referendums outside of Western Europe or for referendums within majoritarian democracies (i.e. the United Kingdom or the United States).
Over and above that, it would be of particular interest to trace the dynamics of a winner-loser gap over multiple referendums in countries in which referendums are a more regular part of the political decisionmaking process, such as Switzerland, Ireland and some U.S. states. The effects of winning and losing referendums could be very different in these settings, because of the occurrence of repeated losing and due to possible learning experiences. Repeated losing in referendums might substantively harm the support for referendums. With regard to elections, Anderson et al. (2005, p. 64) found that "persistent losing leads to drops in positive evaluations of the political system". On the contrary, having more experience with referendums also increases the odds that citizens are accustomed with both winning and losing in referendums. As a result, citizens are more likely to have undergone a "learning process" in which they have learnt to lose gracefully (cf. Anderson et al., 2005, pp. 91-92). This leads to an alternative expectation that the effects of losing a referendum will be weaker in countries with more referendum experience. Next to country context, further research could consider several other context variables that result into diverging effects of winning and losing in referendums. For example, the effects of winning or losing a referendum for citizens' referendum support might depend upon whether the outcome of a referendum shifts away from that status quo, rather than it maintains the status quo (Werner, 2019) or whether the government has won or lost. As governments could deal in multiple ways with negative referendum outcomes (Schimmelfennig, 2019), its implementation of a result might modify winning and losing effects. Another factor to consider is the tone of referendum campaigns: as electoral campaigns affect people's views of the fairness of an election (Wolak, 2014), campaign experiences could matter for winning and losing effects in referendums as well. Especially dirty and hard campaigns could strengthen the effects of losing a referendum. With data from a larger set of referendums, further research could investigate patterns that explain when and why the effects of winning and losing in referendums for citizens' referendum support are most likely to occur. This type of research is further relevant, as it could help to identify characteristics of referendums that could minimize the negative consequences of losing a vote for referendum support.
Despite these limitations, this article delivers an important contribution to the academic debate about support for direct democracy and it confirms recent findings that citizens' referendum support has a nonstable and fluid component. To a considerable extent, citizens' support for referendums is (re)shaped by their experiences with winning or losing a referendum. These results also have implications for the normative debate about the potentials and pitfalls of referendums. As the legitimacy of democratic institutions heavily relies upon losers' willingness to approve the decision-making process after their defeat, it is of important concern that the negative effects of losing a referendum on referendum support are so strong. Decreases in losers' referendum support could potentially undermine the democratic legitimacy of referendums. Whereas high levels of referendum support in Western democracies are most often interpreted as a reason in favour of organizing referendums, this study shows that holding referendums is not a panacea. A key challenge is to retain losers' support for the institution of a referendum, even despite their frustration or dissatisfaction with its disappointing outcomes.

Declaration of competing interest
The author declares that he has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding information
The author's work on this article was supported financially by the Netherlands Organization for Scientific Research, as part of the NWOvidi project Calibrating Public Accountability (452-14-008).