Introduction

The 2020 coronavirus disease 2019 (COVID‐19) pandemic poses an unprecedented challenge to established democracies. Governments face an apparent trade‐off between a large number of deaths and an economic shutdown. In response, most countries have enforced strict social confinement measures unthinkable outside war times, and usually referred to as ‘lockdowns’. Although there seems to be a consensus that such policies are the most effective means to reduce the incidence of severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2), which causes COVID‐19, they also strongly reduce civil liberties, erode social capital and bring about economic insecurity (Fetzer et al. Reference Fetzer, Hensel, Hermle and Roth2020). Hence, whether and how lockdowns affect political attitudes is far from being trivial.

As emphasised by Stephens (Reference Stephens2020) in the Financial Times, we observe, amid the COVID‐19 pandemic, ‘the return of government to centre stage [that] marks the close of an era in which power and responsibility migrated from states to markets’. In this new era, a key question emerges: when confronted with grave threats such as those caused by a major health crisis, do citizens trust the democratic system to respond? To answer this question, we study the short‐term effects of lockdowns on political attitudes in the context of Western European democracies. In particular, we investigate public support for decision makers, institutions and regimes. Such support, specifically directed towards political actors and more diffuse towards the political system, is key for the good functioning of democracy (Easton Reference Easton1975; Claassen Reference Claassen2020).

Major international crises can act as critical junctures or reinforce the status quo depending on how well‐established institutions perform, and how citizen perceive this performance. In this paper, we aim to speak to the literature on policy feedback that explores the loop from policies to mass attitudes, and to policies again (for a review, see Campbell Reference Campbell2012), by focusing on the effect of the main policy response to the COVID‐19 pandemic, that is, lockdowns. This literature has shown that the policies adopted in response to a major crisis can affect political support. First, they can rally citizens around the Head of the State when these citizens perceive that the policies indeed help solve the issue or at least bring some relief (e.g., Bechtel & Hainmueller Reference Bechtel and Hainmueller2011; Healy & Malhotra Reference Healy and Malhotra2009). This mechanism of retrospective performance evaluation leads to accountable governments. Second, they can alter support for institutions by shifting citizens' views about these institutions, which can in turn lead to regime change (e.g., Aidt & Leon Reference Aidt and Leon2016; Ruiz‐Rufino & Alonso Reference Ruiz‐Rufino and Alonso2017).

Unsurprisingly, there is so far no consensus in the literature about the political consequences of the COVID‐19 crisis. In a recent working paper, Leininger and Shaub (Reference Leininger and Shaub2020) exploit variations in the number of confirmed cases at the county level in Bavaria and show that the health crisis has benefited the dominant party. In a series of survey experiments and analysis of social media data in Canada, Merkley et al. (Reference Merkley, Bridgeman, Loewen, Owen, Ruths and Zhilin2020) find that the pandemic has been associated with greater partisan consensus and support for the government (see also Harell Reference Harell2020). These studies suggest that the COVID‐19 crisis has been reinforcing the democratic status quo. Yet, in a working paper that focuses on Spain, Amat et al. (Reference Amat, Arenas, Faló‐Gimeno and Muñoz2020) combine several pieces of experimental evidence to show that it has induced a strong national bias coupled with higher demands for techno‐authoritarian decision making. These results suggest that the crisis may mark the beginning of a paradigmatic shift in terms of democratic attitudes. In another working paper focusing on the United States, Gadarian et al. (Reference Gadarian, Goodman and Pepinsky2020) analyse responses to a unique survey experiment and find that support for strict confinement measures is strongly rooted in partisanship. Similarly, Grossman et al. (Reference Grossman, Kim, Rexer and Thirumurthy2020) analyse social media data and show that the partisan affiliation of the Governor who announced local lockdowns is also key to explaining public compliance to these lockdowns. We seek to contribute to this emerging literature by focusing on how lockdowns have affected political support across Western European countries in the days that followed their enforcement.

Using data from a unique online survey fortuitously fielded throughout the months of March and April 2020 on representative samples of the population of 15 Western European countries, a period in which seven countries started a national lockdown, we compare the political attitudes of respondents who filled in the questionnaire right before the enforcement date and those who did right after.Footnote ¹ We find that lockdowns have increased support for the status quo decision makers, institutions and regimes. The vote intentions for the party of the Prime Minister/President have gone up by about 4 per cent, and trust in government and satisfaction with democracy by about 3 per cent on the response's scale.Footnote ² However, consistently with the study of Gadarian et al. (Reference Gadarian, Goodman and Pepinsky2020), we find that lockdowns have had no effect on self‐reported ideology. Furthermore, there is little evidence of a rally‐around‐the‐flag effect (Hetherington & Nelson Reference Hetherington and Nelson2003; Mueller Reference Mueller1970) that would be entirely driven by the pandemic itself and its incidence. Overall, our study thus confirms the retrospective evaluation of performance mechanisms: it seems that citizens have understood that lockdowns were necessary and rewarded those responsible for them.

Data and treatment variable

Our main dataset comes from a web‐based survey that took about 15 minutes to complete.Footnote ³ The questionnaire included questions about political attitudes and preferences for electoral outcomes. It was fielded between March 2 and April 3 in 15 Western European countries (i.e., 15 members of the European Union prior the 2004 enlargement except Luxembourg, but with the addition of Norway), seven of which started a lockdown during the temporal interval of the survey. The dataset can thus be seen as an interrupted time series. To identify the confinement measures, as well as their enforcement dates, we rely on the data gathered by Hale et al. (Reference Hale, Webster, Petherick, Phillips and Kira2020).Footnote ⁴ In this dataset, lockdowns are defined as nationwide and strictly enforced social confinement, that is, citizens are forced (not simply encouraged) to stay home unless they have a valid reason to travel. According to this definition, a lockdown was enforced in Austria (March 16), Denmark (March 18), France (March 17), Italy (March 13), the Netherlands (March 24), Spain (March 15) and the United Kingdom (March 23). In the main analysis, we only include these seven countries. To reduce noise in the estimates, we do not include countries in which the containment policy was decided at the subnational level like Germany. Yet, in a robustness test, we include the other countries, and find that the results are essentially similar (see Appendix A.2 in the Supporting Information).

From March 2 onwards, we contracted the survey firm DyNata to contact a sample of 1,000 adults representative of the population of each country in terms of age, gender and education attainment. In such a survey, not all respondents fill in the questionnaire on the same day. The reason has to do with the way survey firms operate. They usually have a large database with the email addresses of potential respondents together with some sociodemographic information about them. On Day 1, they send an email with a link to the survey to a first set of potential respondents and make sure that their sociodemographic distributions match the one in the latest census data of the country. Of course, not everybody takes the survey on the same day, and some decide not to take it at all. Then, the survey firm sends another email to a second set of potential respondents a few days later. This operation is repeated until the sample reaches the targeted number of respondents. In our survey, it took until April 3 to achieve this goal.Footnote ⁵

A consequence of the survey firm's sampling strategy is that we have in our sample respondents who were invited to complete the survey before the lockdown enforcement date, and others who were invited to complete it afterwards. As a consequence of this, we have respondents who completed the survey before the lockdown, and other respondents who completed it after. This is our treatment variable. Note that we do not assume that respondents who filled in the questionnaire before the enforcement date did not expect a lockdown. The virus had already reached Europe at the time, and the policy was already common in other affected countries like China. Yet, the key difference between the ‘treated’ and ‘non‐treated’ respondents is that only the latter witnessed the government adopting strict confinement measures. Furthermore, Appendix A.2 in the Supporting Information shows that, in our sample, those who responded before and after the enforcement date are very similar on all sociodemographic variables that we can access (age, gender, education level, rurality and immigration background). In a series of supplementary tests, we also show that they are similar on unrelated political attitudes like ideology and political interest. Hence, although the lockdown was expected, the probability of taking the survey before or after it seems to be ‘as good as random’. In the rest of the paper, we thus talk about the effect of lockdowns on political support, treating this as an effective causal identification.

Outcome variables

We focus on three key outcome variables that pertain to different aspects of political support. In line with Easton's (Reference Easton1975) classic typology, we include indicators of both specific (i.e., related to an actor or a set of actors) and diffuse support (i.e., related to an institution or a regime). First, we estimate the effect of lockdowns on the probability to report, in the survey, a vote intention for the party of the current Prime Minister (or President in the case of France). We see this variable as an indicator of specific support. It is also in line with studies of the rally effect following a major event like a terrorist attack (e.g., Hetherington & Nelson Reference Hetherington and Nelson2003).Footnote ⁶ The results are similar, although slightly larger. Second, we estimate the effect of lockdowns on variables capturing more diffuse political support. In particular, we use the reported levels of trust in government (on a scale from 0 to 4), and satisfaction with democracy (on a scale from 1 to 11) in the survey. Note that the variable trust in government can be seen as either specific or diffuse support, depending on whether the respondent understood the term ‘government’ as referring to the current composition of the government or the institution.

We also include two extra outcome variables in the survey to run some supplementary tests: left–right ideology (1–11) and political interest (0–3). A priori, these political attitudes should not be affected by lockdowns, since they result from a long‐term socialisation process and are quite stable over time (Evans et al. Reference Evans, Heath and Lalljee1996; Prior Reference Prior2019). The question wording of the five outcome variables and descriptive statistics can be found in Appendix A.1 in the Supporting Information. We recode all of them to vary between 0 and 1 to facilitate the interpretation of the results.

Results

We use OLS regressions and four specifications to estimate the effects of lockdowns on political attitudes. They are all based on the following equation:

(1)$y_{i,c,t}=\alpha +\beta \times {\text{Lockdown}}_{i,c,t}+\gamma {\text{Deaths}}_{c,t}+\psi T_{i,c}+\varphi z_{i,c,t}+{\theta }_c+{\mu }_{i,c,t}.$

Our outcome variable is $y_{i,c,t}$. It refers to respondent refers to individual $i$ in country $c$ at time $t$ (=day of filling the survey). $\alpha$ is the intercept. Lockdown is a dummy variable taking the value of 1 if the individual was surveyed after the enforcement date in country $c$ and 0 if it was before it. $\beta$ is the main coefficient of interest, that is, the lockdown effect. We introduce country fixed effects ${\theta }_c$ and as an idiosyncratic error term ${\mu }_{i,c,t}$. We cluster robust errors at the country‐day level to account for unobserved heterogeneity over time and space. This is the Basic specification.

We further estimate an Extended specification, in which we account for the progressing incidence of COVID‐19 spread by controlling for the number of deaths attributed to the virus as reported in official statistics in country $c$ at time $t$ ($\gamma {\text{Deaths}}_{c,t}$). The number of confirmed cases has a correlation $\rho =0.95$ with the number of deaths. We thus exclude it to avoid multicollinearity. In addition, we introduce a time trend $T_{i,c}$, consisting in the number of days since the start of the survey in March 2, to disentangle the time‐varying dynamic of political support during the pandemic from the lockdown itself.

In a third Full specification, we introduce $z_{i,c,t},$ a vector of control variables, all measured in the survey. It includes gender (0‐male, 1‐female), age (15–99), rurality (0‐urban, 1‐rural), immigration background (0‐not born in the country, 1‐born in the country), education level (0‐no university degree, 1‐university degree) and electoral participation (0‐not voted at last election, 1‐voted at last election). The descriptive statistics of all these control variables can be found in Appendix A.1 in the Supporting Information.

Fourth, we also estimate a Balanced specification. This specification is similar to the Full one, except that we first use entropy balancing (Hainmueller Reference Hainmueller2012) to reduce any imbalance between respondents interviewed before or after the interview. Although imbalance is relatively small (see Appendix A.1 in the Supporting Information), we can still correct for it to increase internal validity. Concretely entropy balancing weights the distribution of covariates among non‐treated respondents to make it mimic the first and second moment of the equivalent distribution among treated respondents.

Figure 1 presents the regression results of the four specifications.Footnote ⁷ First, it shows that lockdowns have increased satisfaction with democracy by 2.8–3.2 per cent and trust in government by 2.4–3.2 per cent on the full response's scale (from the minimum to the maximum). Second, it shows that they have also increased vote intentions for the party of the Prime Minister/President between 4.1 per cent and 4.3 per cent. This effect reaches 6 per cent points when we consider all governing parties (see Appendix A.2 in the Supporting Information). These short‐term effects are most of the time statistically significant at least at a level of $p<0.05$. Appendix A.2 show the results for the two extra outcomes. It points towards a non‐ideological lockdown effect: the policy has affected neither left–right ideology, nor political interest. The coefficients of interest are extremely close to 0, and never statistically significant.

Figure 1. Effect of lockdown on political support. (A): Satisfaction with democracy; (B) Trust in Government; (C) Support for PM/ President party. Squares are OLS coefficient estimates. Horizontal bars are 95 per cent confidence intervals.

We conduct a series of supplementary tests to assess the robustness of the results. They are all described in Appendix A.2 in the Supporting Information. First, we reproduce the analysis in adding the remaining countries covered by the original survey. Second, we remove, one‐by‐one, each of the seven countries of the main analysis to make sure that the results are not solely driven by one of them. In both instances, the results are very similar. Third, we reproduce the analysis in changing the treatment variable by other policy responses of smaller scale to COVID‐19 pandemic, that is, school closure and workplace closure. We find a null effect.Footnote ⁸ This test shows that among all policies enforced in March 2020, only the most emblematic one, the lockdown, has had an effect on political support. Fourth, we conduct a permutation test to ensure that our results are not due to chance. For this test, we reproduce the analysis in randomly assigning a lockdown enforcement date to one of the dates of the survey. Out of 1,000 iterations, we observe that the placebo lockdown effect is almost always null.

Finally, we discuss the mechanism that we believe is likely to explain the positive lockdown effect on political support. First, we consider the possibility of a rally‐around‐the‐flag effect, according to which it is the health crisis itself, and its incidence, that has increased political support. Such an effect is common after a major crisis like a terrorist attack (Hetherington & Nelson Reference Hetherington and Nelson2003; Mueller Reference Mueller1970, for some counter evidence, see Balcells & Torrats‐Espinosa Reference Balcells and Torrats‐Espinosa2018). Yet, if there was such an effect in data, we should observe positive coefficients associated with the time trend variable and/or the one associated with the number of deaths. Figure 1 reveals that there is a small positive coefficient for time (not statistically significant), and a negative one for the number of deaths (not statistically significant). There is thus little of the existence of a rally‐around‐the‐flag effect in the temporal interval of the survey. Second, we consider the possibility that the increase in political support is due to an increased airtime of the Prime Minister/President. Indeed, the introduction of lockdowns usually went along with televised statements followed by unusually large shares from the public. In Appendix A.2 in the Supporting Information, we report an extra test, in which we change the independent variable to capture whether the respondents were surveyed before or after the announcement of the lockdown by the Prime Minister/President in the media. We find a positive effect on political support, but not a statistically significant one. This suggests that the expectation of the policy is not enough to spur policy support. All in all, we believe that the most likely mechanism is retrospective performance evaluation, according to which citizens have had to understand that strict social confinement measures were necessary and have increased their support for those responsible for this policy.

Discussion

In this research note, we analyse the responses to a survey conducted in several West European countries in March 2020, the month in which most governments made the difficult choice to enforce a strict containment to limit the spread of the COVID‐19 virus. Although this policy has important negative consequences for the population, both human and economic, we find that respondents who took the survey right after the enforcement date show higher levels of both specific and diffuse political support than those who took it right before. First, lockdowns have increased vote intentions for the party of the Prime Minister/President by about 4 per cent points. Second, they increased trust in government and satisfaction with democracy by about 3 per cent on the response's scale. These effects are particularly interesting when considering that these two groups of respondents are very similar in terms of sociodemographics and unrelated political attitudes like ideology and political interest. We also find little evidence that this effect is simply due to a rally‐around‐the‐flag effect brought about by the health crisis itself and its incidence.

Our findings bring some good news. It seems that citizens have understood that strict social containment was necessary, and have rewarded governments that decide to enforce it, at least in the short term. Furthermore, our findings suggest that it has had a positive spillover effect on support for democracy and its institutions. Perhaps, this is due to the realisation that governments were ready to make hard decisions which have prioritised the health of vulnerable individuals over economic interests. Whether this allows sustained action against COVID‐19 virus remains to be seen. Yet, it seems that this pandemic has maybe offered the opportunity to reconcile part of the population with its political leaders and democratic institutions.

Acknowledgements

This study used a dataset co‐funded by a group of scholars from King's College London, Université de Montréal, University of Toronto, Université Libre de Bruxelles, Karlstads University, University of Wien, Sciences Po Paris, University of California, Riverside, University College Dublin, University of Helsinki and Pompeu Fabra University. Damien Bol would like to thank Célyne Van Corven for accepting to fully take care of their baby for two extra days, so that he could work on this paper. Also, he would like to thank Alice for being kind enough to her mother during that time. Damien Bol and Marco Giani contributed equally to this work.