Citizens’ response to a non-responsive government: the case of the Swiss Initiative on Mass Immigration

In times of contested globalization, democratic governments have increasing difficulties to reconcile international obligations with domestic political demands. Unresponsiveness to domestic constituents due to international constraints may threaten to undermine democratic legitimacy. We assess how citizens react to non-responsive governments in the case of a high-stake direct-democratic vote in Switzerland. The 2014 referendum on restricting immigration from the European Union failed in its implementation because of the EU's refusal to negotiate the free movement rights of its citizens. How did Swiss citizens adapt their policy preferences to this implementation failure? Drawing on original survey data, we show that citizens overwhelmingly did not adapt their policy preferences; rather, they rationalized the implementation failure in an effort to protect their ideological and partisan orientations. The results suggests that governments face major challenges to convey constrained policy choices to their citizens.


Introduction
Democratic governments appear to have growing difficulties to respond to domestic demands from their citizens due to constraints stemming from international interdependence and political internationalization (Mair 2013(Mair , 2014. So far, we have a limited understanding of how citizens respond to such constraints and governments' non-responsiveness resulting from them. Previous research has predominantly focused on how globalization constraints affect voting behavior and political support (Hellwig 2008;Kosmidis 2018;Linde and Peters 2020;Walter et al. 2018). In this article, we go beyond these studies and ask how citizens form their policy preferences when they learn about external constraints. Do people rationally adjust their policy demands in light of new information or do they engage in motivated reasoning to uphold their preferences?
To shed light on this question, we study a 2014 popular vote in Switzerland that amended the country's constitution with a new article requiring to curb immigration (henceforth: Initiative against Mass Immigration or IMI). A majority of Swiss voters approved this policy demand despite its violation of the free movement agreement between Switzerland and the European Union that constitutes a necessary condition for Switzerland's preferential access to the European single market. The implementation of this constitutional amendment asks nothing less than squaring the circle between the norm of adhering to international agreements and employing a popular policy demand that necessitates its violation. The Swiss government responded to this decision dilemma by muddlingthrough, i.e., keeping the free movement of people in place and addressing the popular demand by implementing domestic measures to strengthen employment instead (Armingeon and Lutz 2020). How did Swiss citizens react to this decision of sticking to international agreements over responsiveness to the people? How do they deal with information about external policy constraints regarding their policy preferences? Do they reconsider their own position in light of this information or stick to their original position by reinterpreting the circumstances so that it can be integrated into their existing political belief system?
The Swiss anti-immigration vote is a case of a high-stakes decision that mobilized a large number of voters and involved potentially major economic consequences for citizens, in the context of clear cues by political elites, an abundance of uncontroversial information in the public domain, a relatively high level of policy-specific knowledge, and citizens' direct participation through the means of a popular vote (Milic 2015;Sciarini et al. 2015). As almost a fifth of Swiss citizens are cross-pressured on the policy issue, i.e., they would like to limit immigration but also continue economic openness through the Bilateral Treaties, they are potentially in need of further information and arguments for their decision-making (Lauener et al. 2022). In addition, the vote took place in a country with one of the highest GDP per capita, a stable political system highly trusted by citizens, an excellent system of public education and professional mass media, with a citizenry experienced in taking frequently far-reaching political decisions by referendum for more than 150 years. Finally, recent research shows that the institution of direct democracy reduces the gap in satisfaction with democracy between winners and losers of a vote: even if a voter gets the short end of the stick, in direct democracies these citizens are more likely to be satisfied with democracy and hence can be more relaxed about the pros and cons of a ballot proposal (Leemann and Stadelmann-Steffen 2022). Under these conditions, we would expect citizens to be most likely to form their opinions based on argument-based instrumental rationality-"trying to process information as dispassionately as possible" (Flynn et al. 2017, p. 132) and consequently adapt their policy preferences when learning about external constraints. This in contrast to opinion formation based on voters' partisan and value orientation and motivated-reasoning when interpreting information about external constraints.
To shed light on how citizens responded to the non-implementation of their policy demand, this article analyses Swiss citizens' preferences three years after the approval of the constitutional amendment on mass immigration-that is, after it became clear that many casted their vote based on assumptions that did not materialize. We investigate which citizens continued to support the initiative and how we can explain citizens' voting preferences after the implementation failure. The in-depth analysis of original survey data reveals that public preferences did hardly shift and many citizens continue to support the failed policy. We show that even under favorable institutional and political conditions, cross-pressured citizens do not change their opinion in response to changed facts such as a policy failure. The beliefs of a substantial share of citizens are best explained by the prominent role of partisan cues and values. We find evidence that citizens who were initially in favor of the initiative re-interpreted the policy failure to render it consistent with their partisan and value orientation. This means that individuals protect their prior beliefs and use directionally motivated reasoning even in this least likely case for motivated reasoning. These findings are hard to reconcile with optimistic assumptions about citizens 'educated by initiative' (Smith and Tolbert 2004).
The article is organized in five sections: we start with a description of our case, the Swiss constitutional amendment on mass immigration, and its implementation. We then turn to a discussion of citizens' preference formation and derive theoretical expectations for our case. We then detail the research design and our main data base: a 2017 survey that asked respondents whether they would still vote for the initiative after the implementation failure. Our survey allows us to identify the support coalition of the failed policy and the determinants of citizens' voting preferences. Are they consistent with a rational and informed policy choice or rather explained by partisan heuristics and values? How is the implementation failure interpreted by voters? Our findings are presented in three steps. First, we analyze the policy preferences of Swiss citizens after the implementation failure and show that a significant number of citizens would still vote in favor of the initiative. Second, we test the explanatory factors behind citizens' voting preferences with a multivariate analysis. Finally, we assess whether and how Swiss citizens rationalized the implementation failure of the popular vote before drawing conclusions and discussing implications on how citizens form policy preferences in the context of external constraints.

The case: voting against European immigration
Our case is a popular vote on a constitutional amendment on immigration rules that took place in February 2014 in Switzerland and has been narrowly approved by voters with 50.3% of them voting 'yes'. European integration is one of the most salient political conflicts in Switzerland and as in other popular votes on the issue, the turnout (57%) has been significantly higher than on average (usually in the range between 40 and 50%). While all eight previous national referenda on the continuation of the Bilateral Treaties, including those on the extension of free movement of people to new EU Member States, were positive, this time a narrow majority implicitly rejected what Swiss politicians call the "Bilateral path" (Wasserfallen 2021). The ballot proposal was initiated by the right-wing populist Swiss people's party and sought to restrict immigration to the country by introducing an annual admission quota and a preference for Swiss citizens on the labor market. These new constitutional rules are in conflict with several stipulations of the Bilateral Treaty between Switzerland and the EU, establishing the free movement of people between the country and the European Union, thereby outlawing the discrimination of EU citizens on the Swiss labor market. Ending free movement would hence require to terminate or re-negotiate this treaty with the EU. To make things more complicated, this treaty involves a so-called 'guillotine clause' according to which the termination of the agreement would imply the self-executing termination of six major additional treaties between the EU and Switzerland. These treaties provide Switzerland a preferential access to the European single market and, arguably, are of great importance to Switzerland's economic growth and competitiveness. This conflict has been interpreted differently in domestic politics: while the opponents of the constitutional amendment have warned of an end to the Bilateral Treaties and made this risk a central element of their campaign (Armingeon and Lutz 2020), the supporters did not consider the interdependence among the treaties to be a major problem. Rather, they expressed the clear expectation that the EU would be willing to re-negotiate the Bilateral Treaties so that it would conform to the new constitutional rule. Many voters shared this expectation that the policy demand of immigration restrictions does not face relevant external constraints: 36% of all voters and 56% of 'pro'-voters thought that the initiative would not endanger the Bilateral Treaties (12% of 'pro-voters' have been undecided). Although in case of an actual incompatibility of immigration control and stable relations with the EU, four out of five 'pro'-voters preferred the former over the latter, the share of 'pro'-voters who started right from the beginning from the assumption of incompatibility was only 33% (Sciarini et al. 2014, p. 40). 1 The share of 56% of 'pro'-voters who assumed a compatibility of the Bilateral Treaties and immigration restrictions did not dramatically decrease afterward (Sciarini et al. 2015). The approval of the constitutional amendment has thus not been a deliberate rejection of the Bilateral Treaties. A sizeable and therefore crucial number of voters assumed that Switzerland's international obligations are no constraint to the implementation of immigration restrictions as demanded by the ballot proposal. However, this expectation is only plausible if one departs from the assumption that the EU would re-negotiate the free movement principle. In contrast to these beliefs among Swiss voters, the European Commission and the member states consistently communicated before the popular vote that they would not enter any such negotiations. This position was upheld thereafter.
The majority of Swiss politicians and the Swiss electorate preferred the Bilateral Treaties to the implementation of the new Constitution article. Consequently, the Swiss parliament opted for a policy measure that has been called an 'implementation light' and that seeks to improve the effectiveness of the public employment agencies without restricting European free movement. This measure's aim is to hire more domestic workers rather than recruit them abroad, addressing a key concern that motivated the vote on immigration restrictions. While being loosely related to the constitutional article, it did not implement its exact wording and objective of restricting immigration. In 2018, a clear majority of 56% of citizens expressed dissatisfaction with the popular vote's implementation and only 27% of citizens said that they were satisfied. 2 Nevertheless, the factual non-implementation and the unease about it did not cause significant organized opposition among the Swiss electorate and another attempt of ending free movement with the European Union clearly failed in a popular vote in September 2020.
In conclusion, the constitutional reform failed to be implemented because the assumption that the European Commission would pander to the popular vote and enter re-negotiations on the free movement of people did not come to fruition. There is little reason to believe that this assumption hold by many Swiss voters results from being misinformed or a lack of information on EU's intransigence. The major newspapers reported about the EU's lacking will to negotiate and leading Swiss politicians acknowledged publicly that negotiation attempts have failed (see Wasserfallen 2021). However, despite the implementation failure and the unambiguous information that there will be no re-negotiations with the EU, a high percentage of Swiss citizens continues to support the initiative despite its failed implementation according to our survey data. How can we explain this phenomenon?

Explaining popular support for a failed policy
To answer our research question of why citizens continue to support a policy whose implementation failed due to external constraints, we are inspired by the recent literature on preference formation, which provides important arguments that help us to identify explanatory factors and formulate our hypotheses. This literature identifies three modes of preference formation: the accuracy mode, the cold cognition mode, and the hot cognition mode (Colombo 2016(Colombo , 2018Flynn et al. 2017;Kriesi 2020;Lodge and Taber 2013). In the accuracy mode, citizens evaluate policy options by considering data and arguments when forming their opinion on the matter. This is an argument-driven, rational, or 'systematic' procedure ('slow thinking; ' Kahneman 2011). The important point is that citizens weigh arguments and information to arrive at a logically consistent solution. This systematic mode of preference formation is what classic democratic theorythe folk-theory of democracy (Achen and Bartels 2016)-would suggest. It also constitutes an underlying assumption in theories on deliberative democracy (cf. Colombo 2016) and public choice theories on the effects of direct democracy (Benz and Stutzer 2004). Evidence of its validity comes from analyses showing that most Swiss citizens vote in line with their preferences and are considerably knowledgeable (Milic 2012;Milic et al. 2014, Chap. 6;Milic 2015). Finally, citizens arguably learn to cope with the challenges posed by direct democracy. Over time, they are ever more able to follow a process of preference formation driven by arguments and deliberation (Smith and Tolbert 2004). In this vein, authors have argued that vote choice inconsistencies, such as those observed in the British Brexit referendum, would probably have been less likely in contexts with participants experienced in direct democracy, such as Switzerland (Grynberg et al. 2020). The way EU citizens formed their preferences with respect to the Brexit negotiations between the UK and the EU was also marked by considerable instrumental rationalism (Walter 2019). A first implication of this rationalist view on voting is that support for the IMI should represent citizens' rational preference. In our case, this means that voters give priority to immigration restrictions over the continuation of the Bilateral Treaties. Such a pattern of policy preferences may reflect citizens values of closure, partisan orientation toward the radical-right, but also their material interest as losers of globalization. In any case, a continuous support of the failed policy is for these voters a rational choice and consistent with their political preferences.

Hypothesis 1 Preferring immigration restrictions over EU market access increases continuous IMI support.
To what extent citizens are able to take such consistent vote choices should depend on how well they are informed about the ballot proposal. Behavioral scholars have known that the general population's levels of political information and political knowledge are very low since the first survey-based analyses of voting behavior (Berelson et al. 1954, pp. 307-310). This finding has been largely confirmed thereafter (Delli Carpini and Keeter 1996; Achen and Bartels 2016). While many studies show that Swiss citizens have considerable political knowledge and-in particular-are able to logically explain their vote choice (Colombo 2016), the extent and depth of their political sophistication remain unclear. Applied to our case of the IMI, citizens' voting preferences after a failed policy implementation, we would expect that whether citizens continue to support the policy should depend on their policy-specific knowledge. This should apply primarily to the cross-pressured voters. The better informed they are, the more likely that they recognize the implementation failure and adapt their policy preference accordingly. If systematic processing of information explains voters IMI preference, then we should see that continuous support for the IMI stems from voters who are informed about the policy and show consistent preferences.
Hypothesis 2 Knowledge about the IMI reduces citizens' continuous support for that policy.
There are two main explanations for why citizens' preference formation may not follow a systematic processing, but rather shaped by their pre-existing political orientations. First, citizens may follow the position of their preferred political party. Political parties provide cues to their voters that can use them as decision heuristic (Colombo 2016, Colombo 2018De Angelis et al. 2020;Kleider and Stoeckel 2019;Kriesi 2005;Morisi et al. 2021;Pannico 2017;Pannico 2020;Steenbergen et al. 2007; see also Boudreau and MacKenzie 2014;Lupia and Matsusaka 2004). This can be a case of 'cold cognition' where citizens use a heuristic to take decisions that they believe to be in their interest. But it can also be 'hot cognition' where citizens follow their partisan orientation in order to protect their political identity (Leeper and Slothuus 2014). These considerations lead to the following hypothesis: Hypothesis 3 Citizens that sympathise with IMI-supporting parties are more likely to express continuous support for that policy.
Second, citizens may follow their deep-seated values to take policy decisions in a process of directed motivated reasoning: respondents systematically bias "judgements in favor of automatically activated, affectively congruent beliefs and feelings" (Lodge and Taber 2013, p. 24) thereby protecting "their pre-established opinions and attitudes against the effect of new information" (Kriesi 2020, p. 2; see also Flynn et al. 2017;Herrmann 2017). Humans tend to protect their basic or prior beliefs, attitudes and choices. This should also apply when citizens learn new information about a policy proposal. Citizens with strong values of national sovereignty and closure to immigration should be motivated to re-interpret the implementation failure of the IMI that allows them to deny the external constraints and continue their support for the policy proposal. These citizens are motivated to defend their sovereignist beliefs that the constitutional amendment to restrict immigration can be implemented. In this perspective, continuous IMI support should be explained by whether the implementation failure conforms with or contradicts her basic convictions about Switzerland's status in the world. Such a value-based motivated reasoning should allow citizens to hold inconsistent preferences (support for initiative and support for Bilateral Treaties) by denying their inconsistency and rationalizing the implementation failure. Such a reaction denotes cognitive strategies consisting of biasing of decisions in favor of congruent beliefs and feelings (Lodge and Taber 2013, p. 24). This leads us to the following fourth hypothesis: Hypothesis 4 Citizens with closure values are more likely to express continuous support for the IMI.

Data and method
To test our hypotheses, we fielded an original survey based on a population-representative sample in Switzerland. Our questionnaire has been included in the 2017 MOSAiCH survey, a bi-annual cross-sectional and representative population survey of social attitudes that is administered by the Swiss National Data Archive in Lausanne (FORS). The first part of the survey consists of a total of 1066 face-toface interviews. We drop all respondents who were not Swiss citizens-having no voting rights-obtaining a sample of 877 respondents. Although the survey is representative, we assume that supporters of the constitutional amendment are underrepresented, because surveys by other polling institutions included slightly more individuals who voted "yes" than the around 40% in our sample. 3 The face-to-face survey was followed up by a written questionnaire that includes further question items relevant for our analysis and reduces the sample due to non-response to 684 respondents. Therefore, we apply post-stratification weights to all analyses to reduce the sampling error and potential bias from non-response. For a detailed list of all variables and their question wording, see Table 1 in "Appendix" in supplementary materials. In addition, we use data from the 2015 wave of the MOSAiCH survey (2 years earlier) that includes some of the same question items (trade-off preference, party cues, values). The analysis of this repeated cross-sectional survey data allows for a comparison of preferences before and after the implementation failure.
Our main outcome of IMI support is measured by citizens' vote intention of whether or not they would still vote in favor of the IMI after the implementation failure. 4 There are two reasons why we consider these statements about vote intention in our survey as sincere and valid. First, signatures for triggering a referendum on the acceptance of 'implementation light' have been collected (but did not reach the necessary threshold of 50,000 supporters). Therefore, a vote on this process was indeed a realistic scenario. Second, Swiss citizens are used to be called to the ballot box several times a year so that it is not very likely that respondents made just cheap talk, since they were experienced to express themselves in the Swiss direct democracy.
Our four independent variables measure respondents' policy-preference, policy knowledge, party cues and value orientation as explanatory factors for their intended vote choice and allows us to assess their explanatory power for continuous IMI support. We use the trade-off preference of citizens on whether they prefer the Bilateral Treaties with the EU over restrictions on immigration (or vice versa) as an indicator of informed decision-making. An IMI supporter preferring restriction for immigration over the continuation of the Bilateral Treaties is considered making a consistent decision, while an IMI supporter with the preference for the Bilateral Treaties over immigration restrictions is considered inconsistent since the latter preference cannot be reconciled with support for the IMI. The variable has four values (strongly prefer IMI, rather prefer IMI, rather prefer Bilateral Treaties, strongly prefer Bilateral Treaties). We expect that preferring immigration restrictions over the Bilateral Treaties should be an important predictor for continuous IMI support assuming informed decision-making.
As a second variable, we measure policy knowledge based on whether or not respondents correctly recall the content of the IMI. Respondents are provided with four responses (quota rule and limitation of immigrants; restrictive asylum policy; protection of wages; domestic employment measure) out of which they are asked to select the correct one (the first response). We create a dummy variable on whether a citizen has chosen the correct answer or not. As this variable is a very specific knowledge question, we use alternatively a slightly broader knowledge-index adding two knowledge questions related to European migration policy (which country was most restrictive on refugees in the 2015 refugee crisis and which institution defines the political goals of the European Union, see Table 1 in "Appendix" in supplementary materials for the detailed wording). Aggregating the three knowledge questions provides us with an index from 0 (no correct responses) to 3 (all responses correct).
The third independent variable are partisan cues of whether or not citizens were faced with a cue in support of the IMI (dummy variable). As a supportive cue we consider an attachment to the Swiss People's Party (initiator of the IMI) or to any one of four minor right-wing parties (Lega, SD, EDU and MCG) who express support for immigration restrictions. The remaining major parties as well as the Swiss government shifted their implementation preference to the continuation of free movement once all re-negotiation attempts with the EU have failed (Armingeon and Lutz 2020). Alternatively, we use a more fine-grained cue variable taking into account the strength of party attachment on a 0 to 4 scale.
The fourth independent variable captures the value orientation of citizens in terms of closure and openness of the nation state. To measure this latent concept, we aggregate four questions on citizens' vision for the country in terms of openness and closure (defending traditions, privileges for native citizens, prevention of foreign influences, preserving national independence). These are well established survey question designed to identify citizens with a strong national orientation and conservative values on sovereignty. The resulting index of closure values spans from 0 to 24 and approximates a normal distribution. Those in favor of closure are arguably predisposed to answer questions about IMI support in the affirmative since the mere title of the initiative immediately aligns it with the basic value of an independent and sovereign Switzerland. Our survey includes furthermore a series of question on how citizens interpret the implementation failure (degree of actual implementation, power balance between Switzerland and the EU, room for negotiations on the free movement of persons). These questions allow us to assess to what extent citizens engage in a motivated rationalization to protect their pre-existing beliefs. All independent variables are z-standardized for the analysis.
Our analysis is split in three parts. In a first descriptive part, we compare the supporters and opponents of the IMI based on their policy preferences, knowledge, party cues and value orientation. In the second part we estimate a multivariate regression model to identify the relative explanatory power of these determinants for citizens' intended vote choice. We follow the practice of using logistic regressions to model dichotomous outcome variable and calculate the marginal effects. In the third part, we assess whether the implementation failure has led to more consistent voting preferences or citizens rather rationalize the implementation failure in order to stick to their initial vote choice.

Descriptive analysis of supporters and opponents of the failed policy
In a first step, we analyze to what extent supporters and opponents of the IMI differ in their policy preferences, policy knowledge, party cues and value orientation-our explanatory variables. Figure 1 shows that there are pronounced differences between the supporters and the opponents of the IMI. Among those that oppose the IMI, more than 80% prefer the Bilateral Treaties over immigration restrictions. More surprisingly, still more than a third of IMI-supporters of the MEI do prefer the Bilateral Treaties over the core policy demand of the IMI. A majority of citizens that continue to support the IMI do not face a dilemma from the violation of the Bilateral Treaties as they consider them of a lesser priority. A significant share of them does, however, reveal conflicting preferences that they would vote for the IMI although they would give priority to the Bilateral Treaties. In contrast, a smaller difference is found for citizens' policy knowledge. The majority of both supporters and opponents correctly recall the content of the IMI vote, with a slightly higher level of knowledge among opponents than among supporters. Furthermore, we see that closure values are a dominant characteristic of IMI-supporters with over 80% expressing a leaning in that direction. Among IMI-opponents, this is only the case for a minority of 40%. Finally, party cues are received only by a small minority of voters. Around 5% oppose the IMI despite a supportive party cue, whereas more than a third of IMIsupporters are exposed to a party cue in line with this vote intention. This descriptive overview confirms our theoretical expectations, but also shows significant variation between the different determinants.

Testing the determinants of citizens' voting preferences
To identify the relative explanatory power of the different determinants, we estimate a logistic regression model with IMI vote intention as the dependent variable and four independent variables of trade-off preference, policy knowledge, party cues and value orientation that we identified as potential determinants of citizens' continuous support for the IMI (for the complete model output, see Table A-2 in the supplementary materials). To allow for a meaningful interpretation of the estimates, we report the average marginal effects (see Fig. 2). The results are in line with our expectations and the descriptive results shown above. We find strong and significant effect of citizens' value orientation and party cues. The more citizens have values of national closure and the stronger their attachment to political parties in support of the IMI, the more likely they are to support the IMI despite its implementation failure. The negative coefficients of citizens' trade-off preference and policy knowledge are also in line with our theoretical expectations. Knowing the content of the IMI does however have only a marginal and not significant effect on the probability of continuous IMI support. Preferring the Bilateral Treaties over immigration restriction predicts Fig. 2 Average marginal effects on IMI-support Note: plot displays the average marginal effects with 95% confidence intervals for the independent variables on IMI support. Estimates based on a logistic regression model with standardized independent variables (for detailed model output see Model 1 in Table A-2 in the supplementary materials) a significantly lower probability of continuous IMI support. These results remain substantially unaltered if we control for socio-demographic variables, such as age (in years), sex (male/female), and education (number of years of education after primary school). Among those variables only education exerts a significant (negative) effect. Overall, the multivariate analysis corroborates the pattern of the descriptive results and suggests that voters' response to the policy failure cannot be sufficiently explained by rational preference formation (driven by knowledge and preference consistency) as a matter of informed and consistent decision-making but, rather, by their values and partisan orientations.
To further assess these results, we conduct a series of robustness checks. First, we estimate our models using alternative operationalizations of the main independent variables. The model estimates remain largely unaltered when using a broader index to measure policy knowledge. The same is the case for using the more finegrained measure of party cues. Second, we analyze the interdependencies between the independent variables by including interaction terms into our models. Knowledge may not necessarily lead unconditionally to lower IMI support, but primarily among cross-pressured citizens as knowledge should prevent them from supporting IMI in inconsistency with their preference on the Bilateral Treaties. Citizens with accurate information should be more likely to hold consistent preferences. The interaction model confirms this expectation, knowledge has a significant negative effect on IMI support but only among those that prefer the Bilateral Treaties over immigration restrictions. This suggests that knowledge indeed leads to more preferenceconsistent vote choices. Nevertheless, the knowledge effect remains small compared to the main effects of trade-off preference, party cues and value orientation. We may further expect that knowledge moderates the effect of partisan cues and values. This is however not the case, suggesting independent effects.

Assessing the rationalization of implementation failure
In the third empirical part, we further assess how Swiss citizens responded to the IMI implementation failure. More specifically, we seek to answer whether citizens have adapted their preferences in response to the implementation failure or rather engaged in motivated reasoning to protect their pre-existing beliefs. If the latter is driving the continuous support for the IMI, we would expect that IMI-supporters rationalize the popular vote's failed implementation in a way that makes it compatible with their worldview and past vote choice. The information that the vote for immigration restrictions cannot be implemented threatens their existing beliefs and therefore creates psychological discomfort from the resulting cognitive dissonance. We can think of two rationalization strategies that can be used to avoid cognitive dissonance. First, citizens who supported the IMI could convince themselves that the initiative has been implemented and that they never wanted free movement of people to end but, rather, the employment service to improve. This resembles Aesop's fable of 'The Fox and the Grapes' where the hungry fox cannot reach the grapes and, instead of admitting her failure, she adopts the belief that the grapes are sour and, therefore, undesirable. In a similar vein, IMI supporters could adopt the belief that their vote choice has been implemented and that ending free movement is not desirable. Second, citizens could rationalize the failure as resulting from the Swiss government's unwillingness to implement the initiative and adopt the belief that its implementation would have been feasible if it had not been sabotaged by those in power. In this perspective, Switzerland could have forced the European Union into negotiations if it had taken a tougher stand.
We analyze the degree to which the IMI supporters share these two rationalizations of the initiative's failed implementation. More than a third of them believe that the popular vote has been implemented. However, only 3.4% hold the opinion that the popular vote was implemented in its entirety and the remaining 30% think that the vote has been partially implemented. Hence, almost no IMI-supporters believe that the implementation has delivered what they voted for and only a minority believes that the popular vote has been implemented at least partially. Furthermore, only 2.3% of IMI-supporters think that the popular vote's core demand was to improve the employment service. There is no evidence that supporters re-interpreted the popular vote's original demand in an attempt to rationalize the implementation failure. We find more consistent evidence for the second rationalization strategy: Four out of five supporters believe that Switzerland could have gotten a better deal if the government had negotiated more forcefully.
This pattern is confirmed by two additional questions focusing on Switzerland's negotiation powers where only around 20% of all supporters agreed with the statement that Switzerland lacked the negotiation power to demand a reform of the agreement on free movement. That this rationalization is ideologically driven is further corroborated by the fact that this rationalization is significantly higher among IMI supporters with the strongest ideological preference for national closure. We conclude that citizens who continuously support the IMI do so not because they have adapted their understanding of what the vote is supposed to achieve but because they rationalize the implementation failure as the result of an unresponsive government.
One might argue, however, that the need to rationalize the implementation failure should primarily apply to citizens that are strictly cross-pressured, i.e., IMIsupporters that nevertheless give priority to the Bilateral Treaties. 5 These voters should experience the strongest cognitive dissonance as their preferences appear irreconcilable. In contrast, those supporters who would give priority to immigration restrictions if they had to choose seem to be perfectly consistent and should thus engage less in the rationalization of the implementation failure. For this reason, we calculate the same figures presented above separately for these two subgroups of citizens: IMI-supporters that give priority to the IMI and those that give priority to the Bilateral Treaties. The results are presented in Fig. 3. The cross-pressured supporters are by around 20 percentage points more likely to consider the IMI to have been implemented (fully or rather), and they are more likely to hold the view that the employment measures were the initial objective of the initiative (4.2% vs. 2.3%). Regarding the second rationalization strategy, we see the opposite pattern: cross-pressured voters are less likely to believe that more would have been possible, that the EU is more dependent on Switzerland and that the Swiss government could have enforced negotiations. The overall pattern is confirmed that IMI supporters downplay external constraints and hold the view that there is no strict trade-off between the IMI and the Bilateral Treaties. While the rationalization of the implementation failure by re-interpreting the content of the popular vote is chosen more often by cross-pressured supporters, the rationalization of believing that it was the Swiss government that didn't want to implement the popular demand is stronger among those IMI-supporters with consistent closure preferences (prefer immigration restrictions over market access).
If these results show that Swiss citizens who are unimpressed with the empirical developments after the popular vote continuously support the IMI and have rationalized its implementation failure, we should observe opinion stability over time despite the highly salient turn of events. We assess this implication by comparing survey data from before and after it became clear that the implementation of the popular vote had failed. Voters were asked the same set of questions in 2015 and in 2017. This allows us to compare the support coalitions in these 2 years as well as the determinants of vote choice between immigration restrictions and the Bilateral Treaties (see Fig. 4). Around a third of IMI supporters preferred the Bilateral Treaties to the immigration restrictions. This share has Fig. 3 Rationalization strategies Note: the plots represent the share of IMI-supporters that agree (in blue) or disagree (in red) with a statement of rationalizing the implementation failure. The plots separate cross-pressured supporters (prefer Bilateral Treaties) and non-cross-pressured supporters (prefer IMI). (Color figure online)

Fig. 4 IMI-supporters by trade-off preferences
Note: blue stands for preferring immigration restrictions and red stands for preferring the Bilateral Treaties. Based on variable "Trade-off preference" (see Table A1  slightly increased over time (from 33.9 to 35.7%) and demonstrates that those supporters did not resolve their fundamental choice dilemma.
Finally, we compare the determinants of the vote choice in the two survey waves to identify stability and change over time. For that purpose, we estimate logistic regression models with the binary vote-choice as the dependent variable and three independent variables proxying for the different modes of preference formation. Since we lack the knowledge variable in the data from 2015, we include only respondents' trade-off preference between the Bilateral Treaties and the immigration restrictions (initiative) as a proxy for the accuracy mode as we would expect citizens' vote choice to be consistent if they base their decision-making on the systematic processing of arguments: If voters prefer the Bilateral Treaties over immigration restrictions, then they should not vote in favor of the IMI given that the two policies have been shown to be incompatible. As in the previous model, partisan cues are operationalized with sympathy with right-wing parties in support of the IMI, and values-with the preference for closure over openness. Additionally, we also control for age, sex and education. The comparison of effect sizes over time allows us to determine which explanatory factor has become more influential in the aftermath of the implementation failure. The results are shown in Fig. 5 (see Table A-3 in in the supplementary materials for the complete model output). The effects are as expected with a closure preference and radical-right sympathy increasing the odds of IMI-support and preferring the Bilateral Treaties over immigration restrictions reducing them. 6 The overall pattern of the three factors' explanatory power is largely similar across the two survey waves. While the effect of partisan cues has slightly increased over time, the effect of citizens' value orientation has slightly decreased. The effect of the trade-off preference-i.e., rational consistency-has not increased as one would expect in the case of rational belief-adaptation, but even shrunk in size. These results corroborate the previous findings that the popular vote's failed implementation did not change citizens' vote intention that would render their Fig. 5 Vote preference determinants before and after the implementation failure Note: Coefficient plot based on a logistic regression (see Table A-3 in the supplementary materials for the full model output), model coefficients shown with 95% confidence intervals choice more consistent with their policy preferences but, rather, made them rely on their ideological and partisan pre-dispositions to rationalize the implementation failure and stick to their initial vote choice.

Conclusions
Democratic policy-making in times of contested globalization presents often complex evaluation and decision tasks involving trade-offs. Governments need to balance political demands from their domestic constituents with external constraints, particularly their international obligations. In this article, we study the consequences of constrained policy-making for the demands and preferences of citizens. More specifically: Swiss voters approved a ballot proposal initiated by a right-wing populist party that demanded the re-negotiation of free movement of persons with the European Union in order to restrict immigration to the country. As the EU refused to enter such negotiations, the popular demand failed to be implemented. How did the Swiss citizens respond to this experience of an implementation failure due to external constraints? Why are many voters prepared to support a political project in a popular vote despite that their initial expectations did not materialize? We provide empirical evidence for the argument that a large share of citizens relies in their preference formation on deep-seated values and partisan attachment. Many citizens continue to support the failed policy proposal as it is in line with their values of national closure and the cues they receive from their preferred political parties. The dramatic implementation failure did thus not alter the views of Swiss citizens, but rather lead to motivated reasoning in the form of rationalizing the implementation failure that allows citizens to uphold their views. We conclude that only a small percentage of respondents appear to be consistent with the idea that they arrived at their conclusions in a process representative of the ideal type of the classic concepts of democracy. This is, many Swiss voters continued their support for a failed policy following their ideological and partisan bias instead of a rational updating of their policy preferences. This applies primarily to the crucial group of cross-pressured voters who supported the IMI but also preferred the Bilateral Treaties to immigration control.
Our analysis tackles a pressing question of democratic governance in the context of international interdependence and provides insights from an ideal-typical case where popular demands and external constraints collide. The study is however limited in the sense that we do not present empirical evidence that identifies the specific motivations behind citizens' preference formation. Nevertheless, we show a consistent pattern of persistent support for a failed ballot proposal that is in line with citizens' deep-seated values and accompanied by beliefs that help them to rationalize the implementation failure and to avoid cognitive dissonance.
Our findings bear implications on the prospects and limits of (foreign) policy referenda and more broadly on democratic legitimacy in times of contested globalization. The debate on the 'Initiative against Mass Immigration' was held in a nation that has a very long and a successful experience with direct democracy and voting on high-stakes policy issues. Switzerland is well-known for citizens' relative knowledge about the ballot proposals. In this particular case of immigration restrictions, information has been available in abundant amounts as it was not only easy to understand what the ballot vote was about but was also accompanied by an intense public discourse by politicians and the mass media. The political parties offered clear stances. The project was of utmost importance for citizens' economic interests. Hence, this was a least-likely case for a process of preference formation in which automatically activated and affectively congruent beliefs would dominate the decision process. However, even in this most likely case for a process of an accurate preference formation, we find strong evidence for the dominating influence of values and partisan orientation. While we offer further support to the argument about the outstanding role of motivated reasoning in Western democracies, our findings are of particular relevance to the political debate on the functioning of direct democracy and the role of referenda voting in domestic politics. The Swiss immigration vote demonstrates that direct democracy involving high stake issues of foreign policy and the external constraints of international cooperation are likely to disappoint the expectations of its domestic supporters and therefore unlikely to result in conflict resolution (cf. Hobolt et al. 2022). Supporters of direct-democratic decision-making have further to be aware of the limits of rational belief adaptation even under conditions presumably favoring deliberate political debates guided by norms of accuracy.
Funding Open access funding provided by University of Zurich.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.