The gender gap in support for governments during the COVID crisis

ABSTRACT Using data covering most European Union Member States, we study how the support for national governments has unravelled in the first months of the COVID pandemic. Motivated by the growing evidence on the uneveneconomic impact of this crisis across genders, we study if suchunequal economic burden is related to differences in support for public authorities between men and women. While the support for national governments has overall faded in the period considered, the decline has been more pronounced for women and working women in particular. We find indication that the decline in support signals a shift in concerns among Europeans and women in particular, from the health emergency to the economic consequences of the pandemic. We impute up to a third of the widening gender gap in support for government to the shift of emphasis from the health to the economic dimension of the crisis.


Introduction
The rally around the flag triggered by the COVID-19 pandemic has now been analysed in several studies (Bol et al., 2021;Kritzinger et al., 2021;Schraff, 2021;Yam et al., 2020) and received attention at the beginning of the pandemic in the media too (The Economist, 2020). The theory argues that during national crises, people will tend to support their leaders, shrinking differences in terms of previous beliefs, ideological orientations, or partisanship (Dinesen & Jaeger, 2013;Hetherington & Nelson, 2003;Mueller, 1970;Newman & Forcehimes, 2010). The rise in support during crises predicated by the rally around the flag effect is of substantive importance in the context of the COVID-19 crisis. Low trust and support towards government can undermine compliance with measures directed at containing the spread of the virus (Bavel et al., 2020;Devine et al., 2021;Weinberg, 2022), undermining their effectiveness. Unfortunately, the first wave of the pandemic was followed by a second and a third and, at the time of writing, the so-called Delta variant is fast spreading across the globe. However, from a policy perspective, trust in and support for governments remain as crucial as during the start of the pandemic, as they are instrumental for the successful uptake of mass vaccination campaigns.
In this paper, we investigate how the support for 19 European governments has evolved in the six months that go from the peak/end of the first wave in Europe to the onset of the second. In a protracted crisis such as the COVID-19 pandemic, political economy arguments would suggest that socio-demographic groups who have been most affected should also be the ones who lower the most their support for government (Foster & Frieden, 2017). During this pandemic, empirical evidence has accumulated in the economic literature pointing to an uneven economic impact across genders (Adams-Prassl et al., 2020;Alon et al., 2020;Hupkau & Petrongolo, 2020;Sevilla & Smith, 2020), which has been echoed also by international organisations and agencies (European Institute for Gender Equality., 2021;ILO, 2021;IMF, 2021). We analyse whether this uneven economic impact between genders has also translated into a gender gap in support for the government. By focusing on gender, we add new evidence to a literature that has studied changes in support mainly based on labour market status or by education (Erkel & Meer, 2016;Foster & Frieden, 2017). To our knowledge, this is the first paper analysing differences in government support by gender, related to the unequal burden of the crisis and the containment measures.
This article connects to several strands in the literature. First, to the application of 'rally-around-the-flag' frameworks to the COVID-19 pandemic. In an early comparative analyses, Yam et al. (2020) found a strong relationship between the number of COVID-19 cases and executives' approval ratings on the same date. In the European context, Schraff (2021) argued that the sudden spike in political trust in the immediate aftermath of the pandemic was due to the scaring effect of the intensity of the pandemic rather than the reassurance coming from the effectiveness of governments' measures to control the pandemic. Such finding is in direct contrast with other cross-national evidence connecting the surge in support to the policy measures adopted by 15 European governments (Bol et al., 2021).
In the context of the rally-around the flag framework, heterogeneity in support is often explained in terms of political affiliation or ideology. As a second strand in the literature, we expand this focus on heterogeneity by looking at economic determinants of variation in support, and bring in political economy arguments to ground this expectation. In the context of the socalled euro-zone crisis, several studies analysed changes in support mainly based on labour market status (e.g., unemployed, highly skilled), or by education (low versus high educated) (Erkel & Meer, 2016;Foster & Frieden, 2017), under the assumption that these socio-demographic groups were among the worst affected by the crisis and thus more likely to withdraw their confidence in governments and institutions. Further, besides variation in sheer economic circumstances, how individuals rank economic concerns against other priorities is also likely to be a factor in determining overall levels of support in the context of crises. Oana et al. (2021) argue that, as a consequence of the pandemic, people may shift their political preference towards more safety and security, but this should be moderated by how people prioritise economic concerns to begin with. Therefore, the authors argue that people who prioritise economic concerns should have lower probabilities of preferring strong rather than mild lockdown measures, as the former would cause severe economic consequences.
A final theoretical consideration pertains to the purported duration of the rally around the flag, a question which has so far received relatively little attention. Altiparmakis et al. 2021 explicitly hypothesise that, during the pandemic, policy evaluations of government measures should strongly diverge based on trust and partisanship the more the crisis drags on. In more general terms, Kritzinger et al. (2021) suggest that we should think about rallies-around-the-flag as short outburst of support from a very heterogeneous coalitions of groups. The longer the crisis drags on, the higher the chances are that several groups will withdraw their support.
For our purpose, we leverage three European Parliament's special surveys on European citizens' attitudes and opinions during the COVID-19 pandemic, covering from April to October of 2020. While most previous studies focus on the immediate aftermath of the pandemic breakout, our data enable us to trail supportfor governments from the peak or end of the first wave (depending on the country), to the following period in which the pandemic subsided, and ending with the start of the second wave.
We start by showing that, overall, the support for most European governments has declined as the pandemic progressed, and this is true after we hold constant respondents' socio-demographics as well as epidemiological data such as the regional number of confirmed cases. We then zoom in on gender-based differences in support, noticing that women withdrew their support more profoundly than men, and especially women in the labour market. In parallel, by analysing how respondents placed themselves in a continuum where the health benefits of the restrictions are weighed against their economic costs, we show that, after the initial alarm caused by the pandemic, Europeans started to be concerned about the economic consequences of lockdowns and the pandemic. Again, we highlight how women had some of the sharpest reductions in their overall preferences for health versus economy. Finally, by including such health/economy preferences whenpredicting government support, we estimate that it accounts for about a third of the erosion insupport.

Materials and methods
Our empirical analysis leverages three data sources: i) three surveys commissioned by the European Parliament on attitudes during the pandemic; ii) publicly available epidemiological data maintained by the European Centre for Disease Prevention and Control (ECDC); iii) data on confirmed cases and policy measures collected by the Oxford COVID-19 Government Response Tracker (OxCGRT). While more details on these data sources are in the Appendix, we briefly introduce each of them below.

Survey data
We take our information on individual attitudes and personal characteristics from three European Parliament's Special Surveys on European citizens' attitudes and opinions during the COVID-19 pandemic conducted between April and October of 2020. Wave 1 was fielded in 21 EU countries between April and May 2020, whereas wave 2 and 3 were conducted in all 27 member states in, respectively, June and September/October. Each survey samples about one thousand individuals per country and the sample is limited to respondents aged between 16-54 or 64 depending on the country. The data has regional identifiers at different levels of geographical aggregation, which are also re-aggregated into macro-regionsa feature that we exploit to link these surveys to the regional number of cases collected by the ECDC.
This dataset provides both our two outcome variablessupport for the government, and the preferences over health/economic impactand the individual characteristics that we use to outline how our outcome variables have evolved differently for men and women and by labour market statuses within genders. In the dataset, the government support is recorded as an ordinal variable taking four values: totally support, tend to support, tend to oppose, and totally oppose. We recode this variable as a binary by grouping those who declare to totally or tending to support the government in one group and the other two categories in another.

Data on confirmed COVID-19 cases and policies
To account for the role that the intensity of the pandemic plays in influencing the variation in political support, we resort to two publicly available data sources: The Oxford COVID-19 Government Response Tracker (OxCGRT) (Hale et al., 2021), and the ECDC data on confirmed COVID-19 cases. From the former, we extract information on the Stringency Index, a composite indicator capturing the strictness of the policy measures enacted by governments in a variety of public policy domains, from transportation to education, to mobility. Further, we also include an indicator on school closure, to investigate whether home schooling may unevenly affect worklife balance across genders in the household. From the ECDC, we take the regional time series on the 14-day notification rate of newly reported COVID-19 cases per 100,000 populations by week, which is matched to the regions the respondents belong to.

Sample selection
We restrict our sample in several ways. First, we drop those individuals who declare a level of education inconsistent with their stated age. Second, we drop those who declare to be working and at the same time to be in fulltime education. Third, we drop all respondents from Lithuania, Estonia, Latvia, Cyprus, Malta and Luxembourg who were surveyed only twice. Fourth, we drop all respondents from Finland and Greece as the ECDC started collecting data for these two countries only after the first survey.
After applying these selection rules, our estimation sample consists of 57,471 observations collected in 19 EU Member States-Austria, Belgium, Bulgaria, Croatia, Czech Republic, Denmark, France, Germany, Hungary, Ireland, Italy, The Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain and Sweden-across three survey waves. The sample is fairly balanced across countries and waves with about 1,000 observations for each combination. Table A.1 reports our estimation sample by waves and country.

Empirical estimation
Our empirical analysis has two goals: i) describe how the support for European governments has evolved following the first months of the COVID-19 epidemic, and how it has evolved differently across genders; ii) understand if the economic uncertainty set off by the COVID crisis has influenced the forming of a gender gap in government support as the crisis unravelled.
In our empirical analysis, we proceed in two steps. First, we estimate the probability of supporting the national governments between survey waves and across genders by a series of linear probability models. Second, to study how priorities have shifted across survey waves, we use one question in the survey that asks respondents to express their position on a health/ economic continuum. Preferences over such political dimensions aim at capturing political preferences likely to be salient during a pandemic. Similar trade-offs have been analysed in political science literature on COVID19 (Oana et al., 2021) making our findings comparable to others.
The estimation of support for governments across genders. To characterise how the government support has evolved in general and by gender from waves 1 through 3 of the surveys, we estimate a linear probability model by OLS. The advantage of using a dummy variable instead of an ordinal variable as a dependant variable is that we can easily interpret the regression results as the change in the probability of support for a change in each independent variable. 1 Our baseline OLS model takes the form: where y it is an indicator variable taking value 1 if the respondent i in wave t expresses total support or a tendency towards support for the national government and 0 otherwise. s i is an indicator variable equal to 1 if the respondent is a woman and zero otherwise. w i is a categorical variable for each of the six possible work statuses: employed, self-employed, in education, inactive, unemployment and no answer that we include as five separate dummies in the regression. The vector x i includes age and education. γ c and τ t are country and survey wave fixed effects respectively. e ct captures either the severity of the epidemic in the two weeks preceding the fielding of the survey, or the severity of the stringency index, or the median school closure index in the month preceding the interview, depending on the specification, in country c. e it is the usual error term. How support evolves across survey waves is captured by interacting the gender dummy s, the work statuses dummies w and personal characteristics x with the wave indicators τ. Additionally, we allow for the interaction including gender to evolve differently in each survey wave depending on the work status of the individual by adding a triple interaction term and the related double interactions. Table A.3 reports the estimation results for our main parameters of interest for the LPM.
The estimation of the evolution of economic vs. health concerns. To estimate how the preferences over health versus the economy have evolved we resort to an ordered probit model where the ordinal dependant variable is a personal evaluation of whether the health benefits assured by the restriction measures in place at the time of the interview exceed their economic costs. Respondents can choose between six options where a '1' indicates the highest confidence that health benefits are at the moment trumping economic costs while a '6' indicates the opposite. Choices in between indicate intermediate positions between the two extremes.
We indicate the six ordered outcomes with k. In our model, the probability for the individual i to be in category k will depend on demographic characteristics x i and survey wave τ t . If we assume that the error term is normally distributed, we can estimate the probability of observing outcome k for individual i in the usual ordered probit fashion: where the parameter ψ k indicates the threshold for category k that are assumed to be strictly increasing (ψ k < ψ k+1 ∀k) and ψ 1 = −∞, ψ K+1 = ∞. In our specification, the vector x i contains our usual demographics (sex, age, occupation, and education) plus a country fixed effect. Our results of interests do not change if instead of a series of five dummies we include one categorical variable. Table A.4 reports the estimation results for our main parameters of interest for the ordered probit model.

Results
How the support for the government has changed between waves Figure 1 reports the share of respondents who declared to supporttotally or to some extentthe national government in each of the three surveys for each country in our sample (red lines and dots) and the pooled sample (pale blue lines and dots). Overall, the blue line makes it apparent that the proportionsof government support have decreased from April to October 2020, approximately going from 0.54 to 0.46. In the next section, we discuss if this erosion has been equally distributed across genders and socioeconomic strata.

Gender-based differences in support
In Figure 2 we show the evolution across our pool of 19 European countries of the probability of supporting the government for men and women from April (survey wave 1) to October (survey wave 3) 2020. These probabilities are estimated with the LPM described in Equation (1) where we control for demographic characteristics, an indicator capturing the intensity of the pandemic at the regional level in the two weeks preceding the interview, as well as indicators for the stringency of the policies to tackle the pandemic (see Table A.3). 2 Therefore, the results we show below take the severity of the pandemic into account.
The figure tells us that while support has declined for both genders, women have withdrawn their support more decidedly than men. We estimate that while the probability of supporting the government in April 2020 was practically identical for both men and women at around 55%, after six months of the pandemic this probability has dropped to less than 50% for men and less than 45% for women creating a 5 percentage point gender gap in support. Should economic considerations motivate the Figure 2. Probability of expressing support for government, by gender. Note: Predicted probabilities based on Equation (1). Standard errors are clustered at the regional level. The shaded areas represent 95% confidence intervals (Cameron et al. 2008).
sharper drop in support among women, we would expect women in the labour market to be pushing the drop for those women are those more likely to be affected by the restrictions to economic activities imposed throughout the continent.
In Figure 3 we report the gender gap in support and how it has evolved dividing the sample into five 3 work statuses: employed, self-employed, in education, inactive and unemployed. The markers represent the differences in the predicted probability of supporting the government between women and men estimated with Equation (1), while the bars represent the 95% confidence intervals. 4 As a preliminary remark, we need to point out that all work statuses show a decreasing trend in support for the national government for both men and women. In such context though, it is evident that the only two groups for whom the gender gap in support is statistically significant are the employed and the self-employed. Whereas support among the other three groups has evolved in parallel from Spring to Autumn 2020 for women and men, employed and self-employed women are around 5 percentage points less likely to support their national government than their men's counterparts. That gap did not exist in spring. This evidence conforms with the economic literature on the gendered impact of the COVID crisis. Over time, the pandemic put a comparatively larger burden on those women active in the labour market who were more likely to be employed in sectors heavily exposed to pandemic-containment measures.
A second distinctive feature of the pandemic has been its impact on the volume of home production. School closures 5 in particular have forced parents to re-organise the division of domestic work whose burden has often fallen on the women's shoulder. We expect women to be less supportive of measures increasing their share of home production and of governments responsible for that. We investigate whether this is the case in Figure 4 where we plot the difference in the probability of support between women and men in and out of the labour force and, among them, women with and without children. We observe no difference in the level of support between women and men out of the workforce, irrespective of whether they are parents or not (left-hand side plot). The blue markers, representing the gender gap within parents, is below the red line, but both confidence intervals of the predicted probabilities include 0. For those in the workforce (right-hand side plot), instead, the gender gap in support is null in wave 1, it grows in wave 2, and becomes significant by wave 3. While the gap tends to be larger for women with kids (blue line) in both Note: The red lines represent the difference in the probability of supporting the government between women and men without children. The blue lines represent the difference in the probability of supporting the government between women and men with at least one child. The vertical bars represent the 95% confidence interval (Cameron et al. 2008). All estimates come from our main regression in Equation (1)  the left-and right-hand side plots, in further analyses we fail to detect any statistically significant difference in government support between women with or without children, irrespective of whether they are in or out of the workforce. 6 In conclusion, in a context of a generalised drop in support for European governments women have withdrawn their support more sharply than men. This shift has been mostly driven by women employed in the labour market who have been at the receiving end of the most severe economic consequences of the COVID crisis. Instead, we do not find evidence of an effect of children on the widening gender gap in support over and above the effect of labour market status. In the next section, we show how this gender gap in government support has occurred in parallel with a more pronounced shift in economic concerns for women as compared to men.

Changing trade-offs
One of the most immediate and evident consequences of the public health containment measures put in place by almost all governments across the continent has been the partial halting of several economic activities. As a consequence, the containment measures have been perceived by many as imposing a trade-off between health benefits and economic performance (Oana et al., 2021). The surveys ask how respondents rank the health gains compared to the economic costs of the containment measures. We use this question to estimate how the relative rank has shifted as the crisis progressed.
We estimate how the probability has moved in time with the ordered probit model described in Equation (2) where the dependent variable is the personal position in the health-economy trade-off. In Figure 5 we report the marginal effect calculated at the mean values of all other covariates on the probability of declaring that the health benefits surpass the economic costs of the containment measures (red line) and vice versa (blue line) for women compared to men. 7 From the figure, we can see that while women had a 2 per cent higher chance to appreciate the health benefits of the containment measures over their economic costs in survey wave 1, they became more anxious about the possible economic damages by survey waves 2 and 3 when the difference compared to men disappeared.
The 2 per cent drop in support for the health benefit and the almost parallel uptick in the consideration of the economic damages might seem trivial, but they go a long way in accounting for the widening gender gap in support between waves 1 and 3. We assess how much this shift in emphasis from health concerns to economic costs can explain of the gender gap in support by estimating an extended version of our baseline model of Equation (1), where the ranking of health benefits vs economic costs is included directly on the right-hand side. 8 Those who are more concerned with the economic costs of the measures tend to support their national government less. Those who declare the highest level of concern for the economic costs of the measure are 32 percentage points less likely to support the government compared to the reference category.
For us though, the most revealing feature of Table A.5 is that by comparing the coefficient of the interaction term between the women dummy and survey dummies in the baseline model -which does not include the health/economic trade-offto the extended modelwhich does -, we can understand how much of the drop in support is captured by the introduction of this political preference in the model. Figure 6 displays the estimated coefficient for the two-way interactions in the baseline (in red) and extended (in light blue) models. Whereas in survey wave 2 the inclusion of the trade-off explains little of the (smaller) gap in support, by survey wave 3 the change of emphasis becomes more relevant. In our estimate, including information on the relative trade-off between health and the economy shrinks the gender gap in support in the third survey wave compared to the first one from around 4 percentage points to around 3 percentage points. In other words, the change in emphasis in the trade-off for women accounts for about a third of the widening gender gap in support recorded between survey waves 1 and 3.

Discussion
In this paper, we used data from three EP surveys focused on the opinions of European citizens related to the COVID-19 pandemic to study how the support for governments evolved during the first months of the pandemic. In particular, we were interested in understanding whether the harsh and uneven economic and financial consequences of the pandemic suffered by women correlate with their support for their national governments. The main expectation guiding the research was that, in the context of a protracted crisis such as COVID-19, support should dwindle for socio-demographic groups who suffered the most during the pandemic.
We show that the support for national governments has overall fallen during the pandemic, and that this drop is more pronounced for women in the labour market. Further, while women started in April as substantially more concerned about health rather than economy compared to men, by Autumn there was no difference between the two groups. When we include such preference over health versus economy in our analysis, it explains about a third of the gender gap government support.
Our data cover only the first six months of the pandemic, therefore this paper is necessarily silent on the reactions that the following waves of the COVID19 pandemic might have elicited in the European population. We hope that future research will monitor these shifts. Further, the pandemic  (2) in the baseline (red) and extended version (light blue). The bars represent the 95% confidence intervals (Cameron et al. 2008). Point estimates are reported next to the corresponding marker. Standard errors are clustered at the regional level. entailed rapid changes in both the epidemiological situation and the policy response to it, but we only have three survey waves to tackle this time dimension. The fact that the observational period is limited may have an impact on our estimates. For instance, while the gender gap shows a clear trend over time, its magnitude is small. However, and as explained in previous sections, many countries still had to enter a harsh second wave of the pandemic when the third survey was fielded (e.g., Germany). It is plausible that our estimates would have been larger were a fourth survey be fielded later during subsequent waves. The same reasoning applies to the gender gap in health v. economy preferences, or to the interactions of gender with occupation and family composition. Finally, while for presentational purposes in this paper we concentrated on the gender gap in government support pooled across European countries, this is just an average of the substantial variation occurring between states.
Another widespread limitation in times of coronavirus concerns measurements. The reader should be aware that some of the measurements that we use might be imperfect proxies of the actual policies and epidemiological situation. For example, our measurements of the policy response to the pandemic are countrywide. However, we know that often containment measures were adopted at the local level. We could not capture this sub-national variation in our analyses. More profoundly, companies executing surveys had to change ways of conducting them or mixing different methodologies to deal with unpredictable and rapidly changing circumstances.
Nonetheless, we believe that the evidence that we put forth suggests that governments should cushion the economic consequences of containment measures for specific exposed groups. In a context of extreme uncertainty and severe economic and financial hardship experienced by some segments of the European public, legitimacy for further pandemic-containment measures and especially for successful and protracted mass vaccination campaigns require a level of national cohesion that cannot be taken for granted.
In such context, we believe that gender gap in support is especially concerning as it exacerbates pre-existing inequalities. Indeed, our results indicate that the drop in support seems to be driven by women in the labour market, a group which has long been affected by structural disparities in several member states. Well-documented gender segregation in some of the occupations deemed essential during the pandemic, such as the education and care sectors, might have meant a hardening of the working conditions and environment, exerting additional pressures on an already strained work-life balance during the pandemic. While it is important to underline that this paper is not suited to draw any causal arguments in this regard, it suggests that the combination of all these factors may have led to steeper declines in government support for women compared to men. In policy terms, knowing that most of the decline in support among women occurred among those in the workforce means that two different channels of interventions could be triggered, namely a labour market-based one, and a mainstream one which is not labour market specific. When we include additional checks relative to non-labour market individual features, such as having dependent children or being single parents, our results do not change. This further strengthens the chances that workplace-based interventions targeting women and aiming at relieving some of the additional burdens experienced during the pandemic may bring back some of the confidence lost. To reiterate, to the extent that trust in and support for governments are deemed essential for effective policy responses during a pandemic, reducing the gender gap that we observed should be regarded as an integral part of such policies. Notes 1. For a textbook comparison of the Linear Probability Model (LPM) with nonlinear models such as logit or probit see Angrist and Pischke (2008). 2. Prompted by the suggestion of an anonymous reviewer, we explore the role that psychological stress might play in explaining the gender gap in support.
We include a score measuring psychological stress at the moment of the interview based on a series of questions inquiring about 'What feelings best describes [the respondents] current emotional status' during the interview in our baseline model. As we would expect, people declaring to be in distress (e.g. declaring feelings of 'fear', 'uncertainty', 'anger', 'frustration', or 'helplessness') tend to support the incumbent less and this difference is statistically significant. This inclusion does not affect the gender gap in support though. More details are in Table A.4. 3. For presentational purposes, we omit including the plot for the sixth value of the categorical variable, namely 'no answer'. 4. In Table A.3 we report all the coefficients of this regression. 5. We do not find any statistically significant effect for the categorical variable recording the school-closure policy in force nationally in the month preceding the interview. More details are in Table A.3. 6. We also explore whether single parenthood might affect the support for governments differently between men and women by estimating our baseline model and including a dummy for single parenthood and its interaction with a wave and a women indicator variable. We find no statistical difference in support between single fathers and single mothers, and between single parents and parents in a relationship. Results are available upon request. 7. The coefficients of the ordered probit regression are reported in Table A.4. The descriptive trend for this question is in Figure A.4. 8. We report the estimated coefficients in Table A.5.

Disclosure statement
The scientific output expressed does not imply a policy position of the European Commission. Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use that might be made of this publication. For information on the methodology and quality underlying the data used in this publication for which the source is neither Eurostat nor other Commission services, users should contact the referenced source. The designations employed and the presentation of material on the maps do not imply the expression of any opinion whatsoever on the part of the European Union concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries.