The Brexit learning effect? Brexit negotiations and attitudes towards leaving the EU beyond the UK

ABSTRACT Britain’s decision to leave the EU did not go unnoticed by the remaining EU states. Previous studies have shown that the assessment of the Brexit decision shortly after the referendum took place influenced the voting behaviour of citizens in a hypothetical EU referendum held in their country. This research note goes one step further by examining whether citizens’ willingness to leave the EU changed during the following three years of prolonged Brexit negotiations. To this end, Eurobarometer and ESS data are used to descriptively trace public votes in hypothetical referendums on EU membership over time. In addition, a cross-sectional analysis of Eurobarometer data collected two years after the referendum finds that citizens’ assessment of the British Brexit experience is strongly associated with attitudes towards EU membership. If Brexit is seen as the right decision for Britain, withdrawal of one’s own country from the EU is seen as more attractive although reassuringly for the EU, most people in the EU do not hold this belief.


Introduction
On the 23rd of June 2016 the citizens of the United Kingdom (UK) voted to leave the European Union (EU). On 29th March 2017 the British government invoked article 50 of the Treaty on European Union (TEU), which initiated the British EU exit process. Almost three years later Britain's exit from the EU, commonly known as Brexit, has finally happened. The leaving process was prolonged by extensive negotiations between the British government and the European Union as well as on-going divisions within British politics. Brexit was postponed three times prior to its enactment on 31st January 2020. On the 12th of December 2019, however, the United Kingdom held a national election initiated by the incumbent Prime Minister Boris Johnson. In this election the governing Conservative Party gained a substantial majority of seats in the House of Commons and was thus able to finally push Brexit through parliament. As a result of Britain leaving the EU the community has lost one of its largest member states. Immediately after the long-awaited agreement on a Brexit deal on the 24th of December 2020, the EU Commission President Ursula von der Leyen stepped in front of the press and announced the end of the negotiations. However, instead of being enthusiastic about the outcome, the parties involved seemed weary and relieved about the ending of the protracted negotiations: At the end of successful negotiations I normally feel joy. But today I only feel quiet satisfaction and, frankly speaking, relief. I know this is a difficult day for some. And to our friends in the United Kingdom I want to say: parting is such sweet sorrow. (Von der Leyen 2020) The Brexit negotiations have clearly illustrated the interdependence of EU states and the realization that a departure from the alliance comes with costs for all parties involved (Gerhards et al. 2019, 238). From the beginning the EU adopted a hard line towards the UK in the negotiations (European Council 2020). The aim of this political response was twofold. On the one hand, the EU wanted to prevent other EU member states from also holding a public membership referendum and leaving the community. On the other hand, the EU wanted to reduce the potentially negative consequences of Brexit for remaining member states. As Donald Tusk stated on 29th April 2017, the EU tried to do 'everything we can to make the process of divorce the least painful for the EU' (European Council 2020). Likewise, the majority of the EU public believed that Britain's exit from the EU would harm the community (Stokes 2016, 2). In line with that, citizens with more positive feelings towards the EU and their country's EU membership were particularly likely to take a 'more uncompromising negotiation stance towards the UK' (Walter 2019).
The stakes for the EU thus appear to be high with respect to the possible impact of Brexit on the degree of commitment to continued membership in other states. 1 It is important therefore to examine whether EU citizens' beliefs about Brexit might also have affected their opinion on the desirability of their own countries following a similar trajectory to the UK. Previous studies have shown that holding European Community (EC) referendums in some countries resulting in popular backing for the community enhanced public support for the EU in the rest of the community (Eichenberg and Dalton 1993, 523). Therefore, we might expect that perceptions of the Brexit referendum and its consequences for the UK have played a role in citizens' support for EU membership in fellow states.
The evidence on this is limited. De Vries' (2017) study of Brexit's effect on EU support in other EU member states compared respondents' willingness to leave the EU in a pre-and post study of the 2016 referendum. Her findings indicated that the Brexit vote sent a negative signal to the remaining EU states and their citizens. As a result, levels of leave voting in a hypothetical referendum decreased. She suggested that the EU should develop this negative interpretation of the referendum's impact and avoid the risk of Eurosceptic actors framing Brexit as a success story (De Vries 2017, 40). However, as De Vries' second survey wave was collected immediately after the Brexit referendum, in August 2016, it cannot capture the impact of prolonged Brexit negotiations on public support for EU membership. This research note therefore goes a step further and contributes to an understanding of the longer-term impact of Brexit by examining citizens' willingness to leave the EU after years of prolonged negotiations. It builds on De Vries' study by complementing her work with evaluations of more recent data from across the EU. Our research question asks to what extent has the British decision to leave the EU influenced voting behaviour in hypothetical membership referendums in other EU states in the years following the Brexit referendum?
We examine this in two ways. First through (limited) descriptive time series provided by the Eurobarometer and the European Social Survey (ESS) capturing citizens' vote choice in a hypothetical membership referendum from 1996 to 2019. Second, given the lack of panel data, we run an exploratory cross-sectional regression analysis based on 2018 Eurobarometer data. The time series reveals a sustained decline in public support for leaving the EU from 2016 onwards, a development that could not be observed in De Vries' work due to her shorter investigation period. Hence, our work provides a longer-term take on the implications of De Vries's study. The cross-sectional models suggest that cost-benefit calculations as well as European identity variables predict public support for leaving the EU. However, once we include respondents' evaluation of Brexit in our analysis, it has by far the strongest effect: If the British exit decision is considered to be the right choice, the likelihood of voting in favour of leaving the community increases substantially.
Temporal variations in public support for leaving the EU To assess the development of public support for leaving the EU over time we use a hypothetical referendum question from the Eurobarometer surveys undertaken in all EU member states. 2 Additionally, we draw on 2016/2017 3 data from the European Social Survey 4 that also examined respondents' voting behaviour in a hypothetical EU membership referendum. 5 Figure 1 shows the weighted 6 distribution of leave voters over time (in per cent). We differentiate between the UK, longstanding EU member states and later joining accession countries. 7 The significances given for this descriptive figure are based on independent t-tests. Figure 1 shows that even in the 1990s a far higher proportion of UK citizens wanted to leave the EU compared to other member states. This was also the case in 2016/17. In contrast, there are only minor differences in the proportion of leave votes between more established member states and accession countries in 2016/2017. Two years later, citizens' willingness to leave the EU had significantly fallen within both the UK (−12.1 per cent) and other longstanding EU states (−3.5 per cent). For accession states we see no significant changes between 2016/2017 and 2018. In 2019 the proportion of leave voters dropped significantly in both longstanding member states (−0.6 per cent) as well as in accession countries (−2.1 per cent). For this last time point, there was no significant change visible in the UK.
Taken as a whole these changes indicate that following De Vries' (2017) findings of a downturn of public support for leaving the EU immediately after the 2016 Brexit referendum, leave voting continued to decrease in EU countries in the following years, particularly in longstanding member states. Whether this is a result of the Brexit process is difficult to evaluate with solely aggregate data. We can however examine this possibility with an individual level analysis. Then, at very least, if we find no link between evaluations of Brexit and leave voting we can discount that influence on public responses.

Examining individual variation in desire to leave the EU
Research on public support for the EU has a long tradition and explanations vary over time and regional contexts. EU support can, for instance, be traced back to levels of education and cosmopolitan communication (Inglehart 1970), values (Inglehart, Rabier, and Reif 1987), European identity (Baglioni, Biosca, and Montgomery 2019; Carl, Dennison, and Evans 2019), party attachment (Gabel 1998) or rational cost-benefit calculations (Eichenberg and Dalton 1993;Gabel 1998). What could have influenced the recent decline in public willingness for leaving the EU? Several explanations have been proposed to account for voting behaviour in EU membership referendums. Citizens are more likely to vote to leave the EU when a pro-remain government has been in office for a long period of time, when emotional words are used on the ballot paper, when turnout is high and, most importantly, if voters assume their personal and/or national situation might improve once the country exits the community (De Vries 2017, 2018Qvortrup 2016, 61). The last point refers to De Vries' (2018) benchmark theory. When applying the theoretical expectations of the benchmark theory to public support for leaving the EU after the Brexit referendum, citizens might want to exit the community when they expect their country's alternative to EU membership to be better. However since the likely situation of a country outside the EU is uncertain, voters count on so-called 'benchmarks' when it comes to assessing alternatives to EU membership. These benchmarks might for instance refer to the economic situation of a country outside the EU or to experiences made by former member states that have already left the community (De Vries 2017, 40-41).
Since public support for leaving the EU had dropped within two to three years following the Brexit referendum, some explanations for the downturn are more likely than others. In general, economic evaluations are known to be rather volatile, at least compared to identity factors such as party attachment or European identity which have shown to be relatively stable over time (Boomgaarden et al. 2011, 259;Gabel and Whitten 1997, 92;Lubbers and Scheepers 2010). Positive evaluations of the national economy can enhance support for the EC/EU (Anderson and Reichert 1995;Eichenberg and Dalton 1993, 527-528;Gabel and Palmer 1995;Gabel and Whitten 1997). A recent and comprehensive analysis concludes that the 'core preference for Europe is about membership in the EU and benefits from it, alongside judgments about the trustworthiness of key institutions and a general view of the EU's image' (Anderson and Hecht 2018, 627). Moreover, as noted in the introduction, declining support for leaving the EU might be due to the difficult political and economic situation within the UK since the referendum as well as the UK's long-lasting negotiations with EU fellow states and EU politicians in Brussels. Conversely, voters outside Britain are more likely to be willing to leave when they evaluate Britain's EU exit process positively. We can test this idea by examining individual-level evidence that enables us to compare the impact of this perception with other potential explanations of a preference for leaving the EU.

Data and operationalization
In the absence of panel data enabling the tracking of individual change in attitudes of EU citizens we next use available cross-sectional data to augment the (limited) descriptive time series shown above. For the individual-level analysis we use data from Eurobarometer wave 90.1 collected in September 2018 (European Commission and European Parliament 2018). This results in 26,474 survey respondents across the EU (without UK).
The dependent variable is the response to a question on how respondents would vote in a hypothetical EU membership referendum: 'If a referendum was held tomorrow regarding (OUR COUNTRY)'s membership of the EU, how would you vote?' 8 We chose this instrument for two reasons. First, it allows a direct comparison to De Vries's (2017) analysis. Second, it does not require assumptions about the links between expressions of attitudinal support for EU membership and the likelihood of voting to leave. Citizens might think that their membership in the EU is a bad thing but might still vote to remain in the EU for a variety of reasons that could be associated with an EU exit such as resulting job insecurity, restricted freedom of travel, the common currency etc.. As a result, attitudinal support for EU membership ('membership is a bad thing') and voting to leave the EU on average only correlate at 0.5 across the national populations (see also table A-2 in the web appendix).
Our main independent variable asks for respondents' opinions on whether the British referendum decision was the right choice for Britain. This aspect is measured with the following item: 'From today's perspective, would you say the British people made the right choice in voting to leave the EU?' 9 Consequently, we do not measure the effect of the protracted Brexit negotiations in a direct way, i.e. we do not use a variable that explicitly asks for this evaluation. Instead, we rely on respondents' assessment of the Brexit decision two years after the referendum date and assume the answer not only relates to the direct outcome of the referendum, but also to political developments that have followed.
In keeping with studies of public support for exiting the EU (for instance Clarke, Goodwin, and Whiteley 2017; De Vries 2017; Hobolt 2016), we control for a range of competing sources of leave voting preferences: citizens' evaluation of whether Brexit is good/ bad for the EU, attitudes towards European solidarity, if EU membership is a good or bad thing for the country in which a respondent lives, if that country benefits from EU membership, attitudes towards illegal migration, trust in political parties, evaluation of the single currency, left-right self-placement as well as socio-demographic variables: gender, age, education and employment status. A description of all variables included in the analysis is shown in the web appendix (table A-1). Across all countries 85 per cent of respondents do not choose to leave the EU; the majority of citizens also thinks that Brexit was not the right choice for the UK (64 per cent), and that Brexit is bad for the EU (56 per cent). Correlations between these responses are shown in table A-2. The highest correlations are between general evaluation of a country's EU membership and perceived benefits gained from EU membership (0.60) and the referendum question and Brexit evaluation (0.51). 10 Figure 2 shows the marginal effects produced by the logistic regression analyses with (country) clustered standard errors. 11 If the confidence intervals do not cross the vertical line, we find a significant effect of the independent on the dependent variable. Model 1 (M1) shows the marginal effects of the regression analysis without citizens' Brexit assessments for the UK. We see that if citizens think that Britain leaving the EU is bad for the community, they less likely give their vote to the leave option. In line with that, the more respondents are in favour of their country's EU membership and think their country has benefited from the EU, the lower the probability of voting to leave. The same applies to respondents' attitudes towards the single currency, the Euro, as well as support for solidarity as a value of priority within the EU. If respondents think that solidarity within the EU is important, they less likely vote to leave. Attitudes towards migration, political trust and all socio-demographic variables, except for being female that shows a negative impact, are not significant. Model 2 (M2) adds the question on whether Brexit was the right choice for Britain to Model 1 (M1). This variable displays by far the strongest effect: If someone thinks that the British made the right choice in voting to leave the EU, they are more likely to vote for the leave option in a hypothetical EU referendum in their own country. We also see that other variables lose some of their explanatory power once we control for Brexit evaluations for the UK. In this case, respondents' support and evaluation of their own country's EU membership as well as their assessment of whether Brexit is 'a disadvantage for the EU' have a much-reduced effect on the leave vote compared to model M1.

Results
In line with previous literature, cost-benefit calculations (country benefits from membership) as well as European identity variables (EU solidarity & evaluation Brexit for the EU) predict public support for leaving the EU. However, when we include respondents' belief that the British made the right choice in voting to leave the EU, it becomes by far the strongest predictor and weakens the effects of other considerations. More than two years after the Brexit vote, assessments of Brexit and its consequences for Britain appear to be far more powerfully associated with public opinion than any other factor. 12

Established versus accession countries
We further checked to see if these patterns apply similarly across different sectors of the EU. As already shown in Figure 1, we differentiate between longstanding EU states and newer accession countries. 13,14 Research indicates that citizens in original member states are more likely to rely on their political values for EU support whereas those in later accession countries are more sensitive to the timing of referendums and the political and economic situation at the time (Gabel 1998, 352), although the impact of duration of membership tends to decline over time (Anderson and Reichert 1995, 242). Eastern EU member states entering the community from 2004 onwards, displayed high levels of support for European integration in the 2009 EP elections (De Vries 2013), but this changed with the outbreak of the European Sovereign Debt Crisis. The introduction of bailout packages for southern EU fellow states facing financial and economic difficulties was associated with increasing public opposition towards EU integration in eastern EU states. By 2014, support for EU membership was highest in the EU's founding member states compared to countries in both the south and the east (Hobolt and De Vries 2016). Figure 3 presents the models separately for EU longstanding and accession states 15 (for the regression tables see table A-3 in the web appendix). Respondents' evaluations of whether Brexit was the right choice for Britain have a significantly stronger effect on voting to leave the EU in EU accession states compared with longer-standing EU members. Moreover, we see a significantly more pronounced impact of respondents' evaluation of their country's EU membership on their potential leave vote in accession countries. Our findings are thus consistent with previous research in finding that newer member states are more volatile with regard to their support for EU membership and are more sensitive to daily political and economic developments (Anderson and Reichert 1995;Gabel 1998;Serricchio, Tsakatika, and Quaglia 2013). In accession countries, instrumental rationales to leave or stay in the EU outweigh feelings of belonging. Hence, crises have a higher likelihood of impacting public support for EU membership. The Brexit effect is enhanced in the way that other effects are, but without indicating any radical difference in its relative impact compared to other instrumental factors. Accession countries are simply more responsive to current costs and benefits coming along with EU membership.
How do these patterns relate to the overtime trends shown in Figure 1? There we saw that the time trend for accession countries is shallower than that for established states. Moreover, in more established countries the change occurred earlier (from 2016/17-2018). This is the case even though instrumental reasons (including Britain's Brexit experience) are more closely linked to voting intentions for leaving the EU in countries with a shorter EU history. One possible explanation for this could be that the willingness to leave the EU decreased in these groups of countries for different reasons. It may be that the more established member states evaluated the Brexit experience directly in terms of its impact on the EU as a common actor, whereas economic considerations may have prevailed in accession countries of later EU enlargement rounds. At the beginning of the Brexit negotiations, the economic consequences for the UK were not (and are still not) so clear, but over the years a clearer idea has emerged as to what future trade relations might look like and the extent to which the Brexit decision will have consequences for Britain.

Discussion
We have investigated whether Brexit and the ongoing Brexit negotiations have affected public support for EU membership outside Britain. We predicted that citizens' Brexit evaluations would evolve over time as more information emerged on the political processes following the referendum. Trend data on public support for a country's EU membership indeed reveal a decline in hypothetical leave voting outside Britain since the Brexit referendum. In accession states, public willingness to leave the EU only declined between 2018 and 2019, whereas this trend was observed earlier in longstanding EU-member states. We then examined what might have induced these changes in EU support. For this we were only able to use cross-sectional data, so we cannot confidently make causal inferences, but the analysis indicates that citizens' evaluations of Britain's Brexit experience are strongly associated with their votes in hypothetical EU referendums. If Brexit is seen as the right decision for Britain (36 per cent of non-UK respondents in 2018), withdrawal of a citizen's own country from the EU becomes more attractive.
We can thus conclude that it is possible that the Brexit referendum was the starting point of an ongoing learning experience undermining people's desire to leave the EU. The drop in support for leaving the EU was not large, especially in accession states, but at very least our findings go against the idea that Brexit increased the desire to follow suit among EU citizens. That should be reassuring for the EU and the majority of its citizens.
Our study inevitably has shortcomings and we have openly acknowledged its exploratory nature. The individual-level analysis is only correlational and further work might usefully examine these issues using panel data or experimental designs in order to more firmly assess the causal relationship between the Brexit negotiations and hypothetical voting behaviour. We also have to assume that the main independent variable in our individual-level analysis captures respondents' evaluation of both the Brexit referendum and the subsequent exit negotiations. To the degree that they do not, we may be underestimating these effects.
At the same time, the Eurobarometer survey we use does not include all other potential influences on referendum votes. For example, European identity is not measured directly, nor are questions on party affiliation, voting behaviour, authority attitudes or the perceived economic situation of a country. The inclusion of such factors might in contrast reduce the net impact of the Brexit evaluation. Future studies might be able to examine these counter-veiling influences. It was also not possible to compare models before and after the Brexit referendum. However, our work builds on the paper by De Vries (2017), which does examine data from immediately before and after the referendum.
To conclude, so far Brexit and its at times torturous progress seems to have been associated with a moderate weakening of public desire to leave the EU elsewhere in the community. Given that citizens and political parties across the EU are waiting to see what further turmoil will occur (Ferrera 2017) we may need to revisit this assessment at future points. After all, Eurosceptic attitudes did not disappear just because of Brexitalthough Eurosceptic parties recently started to tone down their claims with respect to leaving the European Union (Taggart and Szczerbiak 2018;Van Kessel et al. 2020). Instead it seems likely that the EU is holding its breath until the consequences of Brexit completely unfold. Brexit can still turn into a 'success story' for Eurosceptic parties and citizens but so far that has not proven to be the case.

Notes
1. The prolonged Brexit negotiations and Britain's problems with leaving the EU did not go unnoticed by national political parties. Populist radical right parties as well as Eurosceptic parties more generally started to tone down their claims for leaving the EU, more than likely as a response to uncertainty concerning the success of Britain's exit process (Van Kessel et al. 2020). However, if the UK manages to survive exit unscathed, it is possible that other countries might follow its example (Taggart andSzczerbiak 2018, 1211 A comparison of the models is shown in the web appendix (figure R-1). Since the results differ only minimally, we kept the approach that retains more cases. 9. (1) Yes, definitely (2) Yes, probably (3) No, probably not (4) No, definitely not (5) Refusal (6) Don't Know. We recode responses 1 and 2-1 to indicate that Brexit was the right choice and categories 3 and 4-0 indicating that Brexit was not the right choice; 5 and 6 are treated as missing values. 10. Unfortunately, the database does not contain information on European identity and party attachment.
11. The regression tables are attached in table A-3 in the web appendix. 12. We also ran additional robustness tests. These included adding terms to the regression analysis in Model 2 to examine if respondents' opinion on whether 'British people made the right choice voting for Brexit' interacted with levels of political interest/ discussion of EU matters for predicting the vote in an EU referendum. Citizens who were more interested in politics and EU matters might have followed the news on the Brexit referendum and the following negotiations more closely. As a result, their willingness to leave the EU might have dropped more substantially than those of uninformed citizens. However, no significant effects were found. 13. As an additional robustness check, we also differentiated between European Sovereign Debt Crisis creditor and debtor states, but found no significant differences between them (see figure R-2 in the web appendix). The same applies to comparing states with strong Eurosceptic parties (more than 15 per cent support in the last national election) or with Eurosceptic parties in government (see R-3 and R-4 in the web appendix) to countries without or only weak Eurosceptic parties. 14. In addition to group comparisons, we also checked whether our analyses across all countries could be driven by individual outliers. Therefore, a jackknife analysis was conducted, which provides information on the variability of the regression coefficients in the individual states. This is presented in table R-1 in the web appendix indicating that even the most extreme cases do not unduly influence the results. As an illustration, we also present the results with the countries with the highest (Sweden) and lowest (Romania) coefficients for the main independent variable excluded from the analysis (see R-5 in the web appendix) as a further indicator of the robustness of the findings. 15. We are able to include more countries in this analysis than were shown in Figure 1, as we have data for all EU countries in 2018. See table A-1 in the web appendix for an overview of the countries included in each group.