Learning the Brexit Lesson? Shifting Support for Direct Democracy in Germany in the Aftermath of the Brexit Referendum

Abstract The June 2016 Brexit referendum sent international shock waves, possibly causing adjustments in public opinion not only in the UK, but also abroad. We suggest that these adjustments went beyond substantive attitudes on European integration and included procedural preferences towards direct democracy. Drawing on the insight that support for direct democracy can be instrumentally motivated, we argue that the outcome of the Brexit referendum led (politically informed) individuals to update their support for referendums based on their views towards European integration. Using panel data from Germany, we find that those in favour of European integration, especially those with high political involvement, turned more sceptical of the introduction of referendums in the aftermath of the Brexit referendum. Our study contributes to the understanding of preferences for direct democracy and documents a remarkable case of how – seemingly basic – procedural preferences can, in today's internationalized information environment, be shaped by high-profile events abroad.


Introduction
Calls for democratic innovations to improve opportunities for both citizen participation and effective problem-solving abound across the political spectrum. While the value shift since the 1970s (Inglehart 1990) has led to higher, and perhaps excessive, expectations for participation among predominantly progressive 'critical citizens' (Norris 1999), more recently, both right-wing populists and climate activists have been criticizing representative decision making as inadequate, and have called for more direct forms of citizen participation. Within political science, the literature on democratic innovations has developed a new strand that studies determinants of preferences for alternative forms of political decision making.
Among the democratic innovations that are being discussed in academia and beyond, referendums constitute a particularly interesting case. Referendums provide an option to take decisions with binding force quickly and efficiently in order to settle a controversial issue and remove it from the agenda. Moreover, the effect of replacing a legislative decision with a referendum is transparent if majority opinion on the matter is known. This relative transparency of outcome effects has enabled researchers to isolate instrumental motives for referendum support Werner 2020).
While referendums are used in many countries at national, sub-national and local levels of government, there is a recent case that stands out in the magnitude of its consequences: the Brexit referendum on UK membership of the European Union (EU). The referendum campaign, the vote on 23 June 2016 and the repercussions of the decision have received considerable public attention across the EU and beyond. How has learning about the UK case affected public opinion towards referendums outside the UK?
This research note presents novel evidence from a German panel survey (GLES 2018) which shows that learning from the Brexit referendum affected support for direct democracy even outside the UK and among those who were not subjected to the decision. We find that supporters of EU integration turned more critical of the introduction of referendums in Germany in the aftermath of the Brexit referendum. Adding to recent research on the instrumental nature of referendum support, our study is the first to uncover evidence of transnational learning: individuals adjust their referendum support not only in reaction to the outcome of referendums in which they participated (Brummel 2020), but also in reaction to referendum outcomes abroad. Through this mechanism, the Brexit referendum has contributed to disenchantment with direct democracy among highly educated left-liberals.
The following section briefly summarizes the state of the art, theorizes the effects of the Brexit referendum on referendum support in Germany and derives hypotheses. We then present our data and the empirical results. Finally, we draw conclusions and discuss implications.
Instrumental Preferences towards Direct Democracy: Theoretical Arguments, Existing Findings and the Case of the Brexit Referendum How do preferences over democratic decision-making procedures evolve, and what role do instrumental motives play in this evolution? There now exists a wealth of studies on process preferences and, more specifically, referendum support. We do not provide a full literature review here, but concentrate on how we contribute to research on instrumental motives for referendum support.
From the 1970s to the end of the last century, direct democracy had been favoured by welleducated, politically interested and left-leaning 'critical citizens', and promoted by progressive political parties, such as the German Green Party (Dalton, Bürklin and Drummond 2001;Norris 1999). However, Hibbing and Theiss-Morse (2002) pointed out that not only may 'critical citizens' support referendums, but also politically disinterested and right-leaning 'stealth democrats'. More recently, direct democracy has been championed by populists seeking to replace representative decision making with a procedure that seemingly brings the 'will of the people', or volonté générale, to bear on decisions without contortion (Werner and Jacobs 2021).
While several studies show that process preferences are to a considerable extent intrinsic and driven by normative conceptions of democracy (see, for example, Landwehr and Steiner 2017;Werner and Jacobs 2021), there is also mounting evidence for referendum support to be driven by instrumental motives Werner 2020). Motivation to support referendums may be said to be instrumental when it is contingent upon 'outcome favourability', that is, the belief that they will bring about decisions one favours. If we can pinpoint changes in outcome favourability beliefs resulting in changes in referendum support, we have evidence for support being instrumental.
However, process preferences may be considered instrumental on two levels; first, there is support for a referendum to be held on a specific issue (specific referendum support); and, secondly, there is support for referendums to be implemented as an alternative form of decision making (generalized referendum support). Specific instrumental referendum support is contingent on the belief that there is a majority of voters for the option one prefers Werner 2020). Generalized instrumental referendum support, by contrast, is contingent on the belief that the procedure at hand by and large tends to promote one's own desired outcomes (Harms and Landwehr 2020; Werner 2020).
But how do citizens form such beliefs about the outcome effects of referendums? One possibilitystudied by Brummel (2020) is that citizens learn from experiencing outcomes of referendums. Based on panel data collected around five referendums, Brummel finds that those whose preferred option lost in the referendum turn more critical of referendums in general, at least in the short term. 1 While the reverse holds among winners only in some of the cases studied, these findings indicate that citizens update their generalized referendum support in response to the outcome of referendums in which they participated.
We extend this line of research and ask whether citizens may also learn from the results of referendums in which they neither had a chance to participate nor were subjected to outcomes, that is, whether citizens update their procedural preferences with evidence from other countries. To explore this possibility, we will make use of panel data in order to pinpoint intra-individual changes in referendum support and their underlying mechanisms in an exemplary case: the Brexit referendum and the observation of its outcome in Germany.
The result of the Brexit referendum on 23 June 2016 was a shock not only to Remainers in the UK, but to pro-Europeans everywhere in the EU. Apparently, pro-Europeans had been significantly misjudging majority support for their preferences. It is plausible to expect an event with such salience and ramifications to affect belief and preference formation not only among British citizens, but also among European citizens outside the UK (see Delis, Matakos and Xefteris 2020).
Previous research has shown this regarding 'substantive' preferences: Walter (2021) has studied the impact of Brexit on attitudes towards European integration outside the UK, showing that voters update their policy preferences on disintegration, resulting in either contagion or deterrence, depending on pre-existing preferences (see also Hobolt et al. 2022). Effects of the Brexit experience on citizen attitudes outside the UK are, as Walter also finds, much stronger among better-informed voters: to update one's preferences, one has to take in the relevant information.
Did the Brexit referendum also affect process preferences, specifically generalized support for referendums? A study by Rojon and Rijken (2021) alludes to this possibility. Comparing crosssectional surveys from a few European countries in 2012 and 2017, Rojon and Rijken find that referendum support decreased among the 'winners of modernization', that is, among socio-economically advantaged and politically satisfied individuals. This finding is in keeping with circumstantial evidence: progressive political actors have seemingly come to favour deliberative modes of citizen participation over directly democratic ones, for example, the German Green Party dropped the call for nationwide referendums from its party manifesto in 2020 after forty years.
We argue that the Brexit referendum might have played an important role in this seemingly dwindling support for direct democracy among progressives. Specifically, we argue thatin line with the findings cited earlierpre-existing substantive preferences on European integration, in tandem with the signal sent by Brexit, significantly shifted referendum support even outside the UK. Assuming outcome favourability to drive referendum support and process preferences to be updated in light on new evidence, our general hypothesis is: Hypothesis 1 (H1): Change in support for the introduction of referendums in Germany after the Brexit referendum depends on an individual's attitude towards EU integration.
Given that Brexit must be seen as a signal for pro-Europeans that majority opinion is not aligned with their own preferencesand that losing may cause larger adjustments of preferences for direct democracy than winning (Brummel 2020)we further expect the adjustment to be strongest in this group: The results in Brummel (2020) are from panel surveys in which respondents were reinterviewed directly after the referendums took place, which raises the question of how long lasting these effects are.
Hypothesis 1a (H1a): After the Brexit referendum, individuals with favourable attitudes towards EU integration become less supportive of introducing referendums in Germany.
Finally, we consider the role of political information, as previous research has shown change in referendum support to be shaped by the information people receive during referendum campaigns (Schuck and de Vreese 2011). Similar to Walter (2021), we assume that in order to update (process) preferences, individuals must first take in the relevant information. Thus, we hypothesize a conditioning effect of individuals' levels of political information: Hypothesis 2 (H2): The adjustment of referendum support based on EU integration attitudes is stronger among politically informed individuals.
The German panel data allow us to study these hypotheses within a context where transnational learning in line with outcome favourability may have been facilitated. The German constitution provides for referendums at the federal level only in exceptional circumstances (reorganization of the federal structure), and the last national referendums date back to the Weimar Republic (1918-33) (Altman 2019). Having had few(er) own experiences with direct democracy, Germans plausibly held less crystalized attitudes towards it, and this plausibly left them especially susceptible to the signal sent by the Brexit referendum. Moreover, the question of whether to introduce citizen-initiated referendums at the national level has been politically debated during the last decade in Germany. We thus expect German voters to have thought about the issue, developed initial stances towards it and potentially adjusted these in light of the Brexit referendum.

Data and Measurement
We use a panel dataset from the German Longitudinal Election Study (GLES 2018), including 2,725 German eligible voters who were interviewed online in eighteen waves between June 2013 and March 2018 in the context of the federal election campaigns in 2013 and 2017, as well as in between in yearly waves. For our outcome variable, we make use of an item on referendum support that ran in seven of the waves. Respondents were invited to rate their agreement on whether 'citizens should be able to initiate binding referendums at the federal level' on a fivepoint scale. Most crucial for us are the waves in the October of 2015 (about eight months before the Brexit referendum) and 2016 (four months after the Brexit referendum). These allow us to observe shifting support for direct democracy in the direct aftermath of the Brexit referendum, which we expect to be driven by individuals' attitudes towards EU integration. We measure individuals' attitudes towards EU integration by their position on whether 'European unification [should] be pushed further to establish a common European government soon or [whether it is] already going too far', recorded on a seven-point scale. To ensure that this attitude is itself exogenous to the Brexit referendum, we took answers from September 2013 (Wave 6), the last wave before the Brexit referendum in which this item was included. 2 To measure political involvement, we draw on nine factual political knowledge questions, also from before the Brexit referendum. We distinguish two groups of roughly equal size: individuals who answered at least eight of the nine questions correctly ('high political knowledge') and those who did not ('low political knowledge'). In Online Appendices D and E, we show that results are similar when using an alternative cut-off for political knowledge and when using self-rated political interest instead. 2 In Online Appendix F, we show that results are similar when using a contemporaneous measure of EU support.

Results
We start with a descriptive look at the evolvement of referendum support. Figure 1 shows the mean of referendum support, measured on a scale from 0 ('disagree') to 4 ('agree'), for all available waves. The first panel includes all respondents; the other three panels distinguish respondents by whether they want more or less European integration, or neither. We used only respondents who provided an answer in all seven waves to rule out that changes reflect panel composition.
With the overall mean hovering around 2.8 to 3.0, support for the introduction of binding referendums at the federal level in Germany is high throughout the observed period. In the overall sample, there is a slight tendency for smaller levels of support after the Brexit referendum. Yet, a clear pattern emerges only once we separate respondents by their level of support for EU integration. Among those in favour of more EU integration, we see a pronounced dip in support from October 2015 to October 2016, in parallel with the June 2016 Brexit referendum (from 2.75 to 2.41). The visual pattern supports the interpretation that this is a (persistent) Brexit referendum effect: there is a crisp drop from October 2015 to October 2016, but there is no trend pre-dating the Brexit referendum, nor is there any other similarly large shift in between two consecutive waves. Among those preferring less EU integration, referendum support is mostly flat over time. 3 As a result of this differential development, what was a small gap between supporters Notes: Mean with 95 per cent confidence intervals. Respondents grouped by attitudes towards EU integration. 3 Yet, especially among the Eurosceptics, referendum support is higher in October 2015. Given the overlap between immigration sceptics and Eurosceptics in Germany (Mader et al. 2020), a plausible explanation for this bump lies in the populist mobilization in the context of the 'refugee crisis'. Accordingly, those critical of the refugee influx might have demanded more direct citizen input to remedy what they perceived as low responsiveness to majority opinion (see Landwehr and Harms 2020;Werner, Marien and Felicetti 2020). As change in referendum attitudes between October 2015 and October 2016 may thus also depend on citizens' attitudes towards immigration, we carried out robustness checks including immigration attitudes as an additional control. The results, reported in Online Appendix C, are similar to those reported in the following. and opponents of further EU integration in July 2013 (0.33) has turned into a large gap in September 2017 (0.93).
To test our hypotheses more thoroughly, we now turn to regression models. In the main text, we present results from an approach that focuses on change between the two waves of October 2015 and October 2016. This approach is in line with the visual patternwhich indicates a concentrated dip in referendum support among Europhiles between these waves and persistence thereafterand allows us to run simple ordinary least squares (OLS) regressions. Specifically, we run regressions of the following form: Thus, we regress the first difference in referendum support of an individual i on their attitude towards EU integration (which we include as a categorical variable), while controlling for the prior level of referendum support and a set of k control variables. In Online Appendix G, we present results from alternative fixed-effect panel data models that use data from all the available waves. This alternative specification leads to similar conclusions. It also shows that the bifurcation in referendum support does not pre-date the Brexit referendum and that the shifts reported in the following persist over time. Figure 2 shows predicted change in referendum support from three regression models: Model 1 only controls for the lagged level of referendum support; Model 2 adds sociodemographic controls (age, gender, Eastern Germany, education and incomefull details are provided in Online Appendix A); and Model 3 adds voting intention (as measured in the pre-Brexit referendum wave) to rule out that initial effects we see just reflect party cueing. From the party-cueing perspective, individuals' referendum support might shift in reaction to their preferred parties adjusting their positions on referendums, rather than individuals adjusting their referendum support directly in response to the outcome of the Brexit referendum and based on their own attitude towards EU integration.
The results in Figure 2 are virtually the same across these three specifications. In line with H1, we see instrumental preference updating: how individuals adjust their referendum attitudes depends on their attitudes towards European integration. In line with H1a, this adjustment is driven by individuals with pro-EU attitudes reducing their support for referendums. Among the strongest supporters of further European integration, the models predict a substantial decrease of about 0.7 scale points, everything else equal. However, we do not see a corresponding increase at the low end of support for EU integration. Among the strongest opponents of further European integration, the models predict essentially zero change in referendum attitudes. This points to an asymmetry, with 'losers' downgrading their support for direct democracy but winners not upgrading theirs (see Brummel 2020). In between, the effects move gradually in line with our argument: the more in favour of EU integration, the more negative the change in referendum support. Now, we move a step further and ask whether the adjustment is stronger among politically informed individuals, as expected by H2. To investigate this, we augment Model 3 from earlier by interacting the EU integration attitude with the dummy variable for low versus high political knowledge. In line with H2, the predicted probabilities in Figure 3 reveal that the pattern from earlier is mainly driven by politically informed individuals. Among the politically informed, change in referendum support decreases monotonically with one's level of support for further EU integration. Among those with lower degrees of political involvement, the pattern is much less consistent. Only low-information individuals with very high levels of support for further EU integration (= 6) adjust their support for the introduction of binding referendums in Germany markedly downwards. It seems that holding an intense attitude compensates for lower levels of political information.
Reading these results from a different angle, we see that the effect of political information levels is conditional on attitudes towards EU integration. It is not the case that higher levels of information always go along with a decrease in referendum supportwhich would be the case if all informed individuals received similar messages on the deficiencies of the Brexit referendum, such as the role of misinformation in the campaign or the problems related to unequal turnout, and reacted similarly to those. Rather, what we see is that higher political knowledge goes along with dwindling referendum support only for those who are (mildly) in favour of more EU integration. Thus, these results support our reasoning that how individuals react to the Brexit referendum is fundamentally shaped by the outcome of the referendum in conjunction with individuals' own views on European integrationand that this reaction is stronger for politically informed individuals with stronger exposure to the signal.

Conclusion
Our study shows that German citizens have 'learned the Brexit lesson' and adjusted their generalized referendum support in line with the evidence on outcome favourability that the Brexit case provided. The adjustment is contingent upon substantive preferences towards European integration and upon political information: support for referendums dwindles among pro-Europeans, and most consistently so among the well informed.
Two implications of our findings strike us as particularly worth noting. First, the fact that information about the Brexit referendum apparently updates procedural preferences in Germany shows us that in an internationalized information environment, we should no longer expect support for democratic decision-making procedures to be driven only by experiences with and evaluations of one's own political system. If an event like the Brexit vote has ramifications for procedural preferences far beyond the UK, we may increasingly expect other remarkable experiences with specific decision-making procedures and democratic innovationssuch as electoral reforms or deliberative mini-publicsto inform preferences over alternative procedures in other countries as well.
Secondly, by providing further evidence on how referendum support is contingent upon substantive preferences and perceived outcome favourability, our study goes some way in explaining the growing disenchantment with direct democracy among left-liberals. In the German case, the Brexit referendum contributed to a swift reshuffle of partisan support for direct democracy: from July 2013 to September 2017, the average difference between voters of the radical-right Alternative for Germany (AfD) and the left-liberal Greens has quadrupled from 0.4 to 1.6 scale points, for example. The shift is also reflected in a rising educational divide in support for direct democracy: in September 2017, mean referendum support was 0.6 scale points higher among the lower educated than the highly educateda difference three times as large as in July 2013 (see Online Appendix H).
An important limitation of our study in this regard is that we study a single country only. As argued earlier, repercussions of the Brexit referendum in Germany might have been amplified by the limited experience of German citizens with direct democracy. However, leveraging subnational variation in this experience across German states in a supplemental analysis, we find no evidence that the response was stronger in states with less direct democracy (see Online Appendix I). We thus anticipate that the Brexit referendum has led to similar shifts in other EU countries, at least in the majority of member states with low to moderate levels of experience with direct democracy (see Altman 2019). The Brexit referendum might thus have contributed to an emerging divide over direct democracy between political camps in not just Germany. Future research should continue to pay close attention to this divide, studying its consequences, as well as the instrumental and intrinsic motives that underlie it.