Public support for government responses against COVID-19: assessing levels and predictors in eight Western democracies during 2020

Abstract

In order to halt the spread of COVID-19 governments have engaged in policies that are both economically costly and involve infringements of individual rights. In democratic countries, these policy responses have elicited significant debate but little is known about the extent to which the responses are supported or opposed by the broader public. This article investigates how citizens across eight Western democracies evaluate the specific policies imposed by their governments to contain the COVID-19 pandemic. The study relies on large-scale, longitudinal surveys that are reflective of the national populations (total N = 124,062). On this basis, it is investigated how pandemic-specific and broader political attitudes correlate with support for government responses during a significant part of 2020, a period marked by pandemic restrictions in all the countries. Medium to high levels of support for the government’s responses are found in all eight countries. Beyond the regular voters of the government, support is driven by individuals high in interpersonal trust and self-assessed knowledge about COVID-19. This may suggest that halting the spread of COVID-19 is viewed as a collective action problem and mobilises support from those who know how to act and who trust others to act similarly.

Supplemental data for this article can be accessed online at: https://doi.org/10.1080/01402382.2021.1925821 .

KEYWORDS:

• COVID-19
• policy support
• voting
• interpersonal trust
• collective action

Acknowledgements

FJ, AB and MBP designed the study; FJ, AB, MFL and MBP collected the data; FJ analyzed the data; FJ, AB and MBP drafted and revised the paper. All data and required code is publicly available in a repository at the web page of the Open Science Framework: https://osf.io/vuaw5/. We are grateful for research assistance from Magnus Storm Rasmussen.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Utilizing a rolling panel sample design, the survey firm Epinion sample adult respondents using online panels. Panelists are compensated via lotteries for gift certificates. In our study, the median interview length, across all countries, was 9 minutes. Among the panelists invited to take our survey, the response rate (calculated as the fraction of complete responses over invited, eligible participants) was between 18% (Hungary) and 38% (Denmark). The survey was conducted in line with the national ethical guidelines for conducting survey-based research involving human subjects. Informed consent was obtained from each participant at the beginning of the survey.

2 We run a number of robustness checks, as described here and below. We first rerun the models described above while also allowing the slopes of the predictors to vary over time (i.e. by data wave). This allows us to investigate whether the strength of the associations between support and the psychological predictors changes in time. Except for societal worry, we find no consistent time trends in the strength of the associations (see Section B.1 in the OA for details). Therefore, we focus on the models that pool observations over time in the main text. We discuss the association between support and societal worry in detail below. Second, we re-estimate the fourth (attitudinal) model while including (a) left–right self-placement as a covariate and (b) the personality trait variables as covariates. Overall, the results replicate those presented in the main text (see Section B.2 in the OA). Third, we also replicate our results while employing post-stratification weights. Weights have been calculated by the data provider, and include data on party choice, region, education, age and gender interactions, house type and household size. These weights ensure that our samples are informative of the population in these respects. Importantly, the weighted results remain fundamentally unchanged as compared to the unweighted results presented in the main text (see Section B.3 in the OA for details). Fourth, and finally, we rerun the main analyses, while shifting our outcome from support for the government response to an alternative outcome that asks respondents about their confidence in the government (see section B.4 of the OA). All results replicate those presented in the main text.

3 In public opinion models, knowledge is commonly included as moderator of the effects of other variables (Zaller 1992). In Figure OA6, we therefore model potential interactions between knowledge and the remaining attitudinal correlates. Overall, these analyses show that the correlation between support and knowledge is strikingly homogeneous over the range of both interpersonal trust and societal worry, while the correlation varies somewhat across the range of personal worry. Specifically, we observe a statistically significant interaction between knowledge and personal worry of about 6.5 percentage points (p < 0.0001), indicating that the correlation between support and knowledge is about 6.5 percentage points stronger among individuals high on worry compared to individuals who are unworried. Specifically, among the unworried the marginal correlation between support and knowledge is 0.14, while it is 0.205 for the most worried (please see details in section B.5 of the OA). Knowing what to do generates stronger support for the government’s efforts among those who also fear COVID-19.

4 The estimator gives an unbiased estimate of the causal influence of the attitudinal variables on support on the assumption that the response support of individuals had followed parallel trends in the absence of changes in these variables. In other words: absent a change in attitudes, all individuals would have experienced similar developments in support. This assumption can be violated in two ways described above: reversed causality and omitted variable bias. With many waves of data, we are able to indirectly assess the plausibility of the parallel trends assumption. Section C.1 in the OA offers evidence for the plausibility of this assumption.

5 Note that the estimated effects correspond to moving from the minimum to maximum on each of the correlates. Evaluated against a 2 standard deviation increase, personal worry increase support by about 1.7 percentage points, knowledge increase support by about 3 percentage points, while interpersonal trust increase support by about 2.8 percentage points. While these effect sizes are small in an absolute sense, the estimated effects are nonetheless relatively large compared to an overall standard deviation of 32 percentage points on the support measure. This, for example, means that a two standard deviation increase in knowledge moves support by 0.1 of a standard deviation.

6 In addition, we repeated the models reported in 3 replacing daily COVID-19 infection counts with daily number of COVID-19-related deaths. This latter metric may be more comparable across time and countries. This replacement does not change any of the reported conclusions.

7 As with the individual-level fixed effects estimators, the causal interpretation of our estimates rely on the parallel trends assumption. In Figure OA10 in the OA, we test the plausibility of the parallel trends assumption by including leads on the effects. Crucially, the estimated coefficients on the leads are close to 0 and far from conventional levels of statistical significance. Moreover, the estimated impacts of the immediate effects remain fundamentally unchanged upon control for the leads. In other words, we do not observe an increase in support that precedes changes in the predictors. This corroborate the design-based identification strategy and, hence, the causal interpretation of the estimated effects.

8 As an additional robustness check, we also included a first-order autoregressive (AR1) structure into the model. This did not change the reported conclusions. Substantially, it should also be noted that the inclusion of a lagged dependent variable is commonly used to assess the potential long-term effects of the independent variables. Specifically, long-term effects are calculated by dividing the coefficient of each independent variable by 1 minus the coefficient of the lagged dependent variable. Interpersonal trust and knowledge gives long-term effects of about 4 and 7 percentage points, respectively, which are similar to the coefficients in the multivariate and bivariate models. This indicates that the predictors have no additional impact on support beyond their immediate effects. In Figure OA10, we included lags on the effects of each predictor. These lags similarly capture long-term impacts on support. The conclusions from these models are similar: there is very little evidence that the predictors impact on support beyond their immediate effects.

9 Note that including such unit-specific time trends is another common way of gauging the plausibility of the parallel trends assumption: if the estimated effects remain similar after inclusion of the unit-specific trends, it corroborates the assumption. The fourth column of Table 3 shows that it makes no difference whether we include the unit-specific trends or not and, hence, this provides further support for the robustness of our results.

Additional information

Funding

This study has been funded by grant CF20-0044 from the Carlsberg Foundation to Michael Bang Petersen.

Notes on contributors

Frederik Jørgensen

Frederik Jørgensen is a Postdoctoral Researcher at the Department of Political Science at Aarhus University. [[email protected]]

Alexander Bor

Alexander Bor is a Postdoctoral Researcher at the Department of Political Science at Aarhus University. [[email protected]]

Marie Fly Lindholt

Marie Fly Lindholt is a Scientific Assistant at the Department of Political Science at Aarhus University. [[email protected]]

Michael Bang Petersen

Michael Bang Petersen is a Professor at the Department of Political Science at Aarhus University. [[email protected]]