At odds? How European governments decided on public health restrictions during COVID-19.

OBJECTIVES: This study aimed to understand how politics, economics, and public health restrictions affected each other during the COVID-19 pandemic.
METHODS: We use seemingly unrelated regressions on a monthly data set of government approval ratings, the stringency index, the time-dependent reproduction number (R), and unemployment, allowing the residuals in each regression to be correlated with each other. We also conduct sensitivity tests using weekly data and the growth in polls.
RESULTS: The study covers 27 European countries from April 2020 to April 2021. A unit increase in the R and COVID-19 cases per million increases the stringency index by 23.742 and 4.207, respectively; a unit increase in stringency boosts the incumbent's popularity by 0.384; the poll positively affects the stringency index; stringency has negative effects on the R; and the poll and stringency index have opposite effects on unemployment.
CONCLUSION: Political and economic pressures did not hinder the government from introducing stronger measures.

Background

COVID-19 has had an unprecedented impact on public health, causing a mass outbreak accompanied by a considerable number of deaths. Where the number of patients with COVID-19 soared past the capacity of a nation’s healthcare system, patients with COVID-19 could not be treated with proper management, and the death rate increased steeply. Moreover, the medical repercussions of the pandemic reached beyond the patients who were directly affected by the disease: by limiting patient access to hospitals, the pandemic has led to delays in treatment and diagnoses for diseases other than COVID-19 as well.

Containing the spread of the virus-induced disease was, therefore, a central task for many governments across the world in 2020 and 2021. To contain the infectivity of COVID-19, national governments imposed a range of restrictions from the mandatory wearing of masks to more stringent restrictions such as lockdowns. Public health decisions, however, were also inevitably interlocked with economic as well as political considerations. On the one hand, politicians have been penalized in the polls for steep rises in infection. On the other hand, the very restrictions that were seen as effective for curbing the infection of COVID-19 were arguably afflicting the economy4, 5, 6 and politically agitating citizens. In March 2021, Prime Minister Modi announced a sudden national lockdown on India, for instance, severely damaging the economy. Citizens did not comply with lockdown restrictions, leading Modi to eventually ease the lockdown, and India’s COVID-19 situation spiraled out of control.

As vaccinations rolled out, some scholars had prematurely predicted that social distancing would be over by the fall of 2021. However, although vaccines may have reduced both the number of new infections and the severity of the illness, the rise of the much more infectious Omicron variant and the possibility that yet another highly infectious virus may arrive at our doorsteps in the coming years call for an enhanced readiness against global pandemics. We contend, therefore, that a thorough investigation of the determinants of COVID-19 health restrictions is still wanting. Indeed, echoing previous fears about the politicization of public health policies, some medical doctors have cautiously raised concerns that the field of medicine is becoming “co-opted into a political program of population control,” even urging public health scientists to “play politics.” If we are unable to resolve this state of disquietude, the next pandemic might just have to be fought with an army of disillusioned and politicized medical professionals.

An impending question, therefore, is whether political concerns had really hindered the timely implementation of necessary public health restrictions. In an attempt to unpack the possible standoff between politics and public health, we carefully analyze the relationship between unemployment figures, government approval ratings, infectivity status, and the stringency of non-pharmaceutical interventions to understand the real impact of strong restrictions such as lockdowns on political and economic considerations and vice versa. Acknowledging the mutually endogenous nature of variables, we compare fixed effects models with seemingly unrelated regression (SUR) models that evaluate the relationship among the empirical equations.

Methods

To estimate the determinants of our four dependent variables (government approval ratings, the stringency index, the time-dependent reproduction number, and unemployment), we first conduct four fixed effects models using the monthly averages of the variables. However, if fixed effects models could account for country-specific baseline variations, they do not address the possibility that the error terms of each regression may be correlated with one another. Therefore, we also use SUR models allowing the residuals in each regression to be correlated with each other. As sensitivity tests, we apply the same set of analyses using the weekly averages of the variables.

Data source and study population

Our study compares 27 European countries from April 2020 to April 2021. The full list of countries is available in the appendix (Appendix, Table A1). The time-dependent reproduction number, R, is defined as the number of secondary infections that arose from a typical primary case in a completely susceptible population and is used as a measure of infectivity. Our article uses Arroyo-Marioli’s (2021) real-time estimates of the effective reproduction number. In addition to the reproduction number, our models also include the number of new cases per million. By doing so, we could compare the government’s sensitivity to the reproduction number and raw number of new cases. For the population of interest, we also gathered poll data concerning the public’s approval of their government. These data were drawn from the Poll of Polls data set, which uses a statistical method called the Kalman filter to aggregate polls from different sources: a series of measurements taken over time are combined to accurately estimate the popularity of an incumbent at a given point in time. To estimate the public’s support of their government, we used the public’s intention to vote for the incumbent party in the case of majority parliamentary governments, the public’s intention to vote for the largest party in the case of coalition governments, and the public’s approval of the incumbent president in the case of presidential systems. To measure the restrictiveness of a government’s COVID-19 policies, we use the Stringency Index developed by Oxford University’s Blavatnik School of Government. Finally, we use the seasonally adjusted unemployment data from Eurostat to account for the economic considerations behind public health restrictions during the pandemic. Unlike other data, which are available on a daily basis, unemployment data were only available by month.

Our main models use monthly averages, and we drop the first 2 months that represents the start of the virus, so the data could have high variability. As a result, our data set includes around 340 observations of monthly unemployment, the public’s approval of its government, the stringency index, and the time-dependent reproduction number over a period of 12 months.

Main analysis

First of all, the relationship between the independent and dependent variables varies considerably depending on country-specific contexts. Depending on whether a polity is a multiparty system or a two-party system, for instance, the baseline level of support for the incumbent (or largest incumbent) party would vary considerably. To account for this variation, we first use a set of fixed effects models. However, the variables we seek to analyze are also causally intertwined. Unemployment can affect the approval ratings, the stringency index, and the reproduction number; approval ratings can affect the stringency index and the reproduction number; the stringency index could affect unemployment, approval ratings, and the reproduction number; the reproduction number could affect unemployment, approval ratings, and the stringency index. As a result, if we independently conduct regressions using one of these variables as the dependent variable and the others as the independent variables, the error terms could be biased by the correlation among the independent variables as well as by reverse causality. To account for the range of new COVID-19 cases number, we treated the number as log scale. To accommodate this causal complexity, we use SURs to allow the error terms of these regressions to be correlated with one another.

Sensitivity tests

As sensitivity tests, we run the same set of models using weekly averages instead of monthly (Table 2 ). These tests allow us to compare long-term vs short-term considerations. Because unemployment data were only available on a monthly basis, unemployment is excluded in weekly model. Another set of sensitivity tests run the same models using monthly averages of the growth rate in polls rather than the absolute values of polls (Appendix, Table A1). Because of country-specific political contexts, some countries have inherently higher or lower polls than others; by using the growth in polls, we can discard such country-specific differences.

Table 2

The results of Fixed Effect and SUR models with weekly dataset.

	Model 5 Poll	Model 6 Stringency	Model 7 R
	Coefficient	Coefficient	Coefficient
Fixed effect model
Pollt-1		−0.033 (−0.814, 0.748)	−0.012∗∗ (−0.025, −0.000)
Stringencyt-1	0.008 (−0.033, 0.049)		−0.014∗∗∗ (−0.017, −0.012)
Rt-1	−0.237 (−1.281, 0.808)	−7.600∗∗∗ (−10.688, −4.513)
Ln (New cases per Mil.)	−0.789∗∗∗ (−1.282, −0.297)	3.071∗∗∗ (1.631, 4.511)
Constant	33.805∗∗∗ (31.447, 36.163)	59.256∗∗∗ (29.678, 88.833)	2.372∗∗∗ (1.922, 2.822)
SUR
Pollt-1		2.386∗∗∗ (2.221, 2.551)	0.060∗∗∗ (0.055, 0.066)
Stringencyt-1	0.399∗∗∗ (0.372, 0.426)		−0.027∗∗∗ (−0.030, −0.025)
Rt-1	12.307∗∗∗ (11.224, 13.390)	−20.724∗∗∗ (−23.554, −17.894)
New cases per Mil.	−0.243∗∗ (−0.317, −0.168)	1.092∗∗∗ (0.864, 1.319)
Constant	−6.562∗∗∗ (−8.927, −4.196)	7.206∗∗∗ (2.227, 12.186)	0.915∗∗∗ (0.788, 1.042)
N	1593	1595	1622

Notes:N is the sample size. ∗P < 0.10, ∗∗P < 0.05, and ∗∗∗P < 0.01.95% of confidence intervals are in the parentheses. New cases per million is logged variable.

Results

The included nations and their baseline characteristics are described in the Appendix (Table A1). Table 1 reports the results of each regression derived with the fixed effects (top) and SUR (bottom) models. As expected, the error terms of dependent variables are significantly correlated (Appendix, Table A2), affecting both the size and significance of the variables. Consequently, we will be using the SUR models in our analysis. In Model 1, the SUR indicates that the stringency index, unemployment rate, and number of new cases of previous month affect the poll in the present month. A unit increase in stringency raises the poll by 0.384, a 1% increase in unemployment reduces the incumbent’s popularity by 3.938, whereas one-unit increase of COVID-19 case reduces the poll by 1.044. By contrast, the reproduction number has no significant effect on the polls. In Germany, for instance, the first wave of stringent restrictions substantially increased the public’s support of the incumbent government, whereas the reproduction number had no clear effect on the polls (Fig. 1 a). In Model 2, higher reproduction numbers, more new COVID-19 cases, greater unemployment, and better polls all result in more stringent public health restrictions. A 0.1 increase in the reproduction number raises stringency by 23.742, one-unit increase of COVID-19 case pulls the stringency up by 4.207, a 1% increase in unemployment increases stringency by 7.084, and a unit increase in polls creates a 1.805 hike in the stringency index. In Models 3 and 4, the reproduction number and unemployment are each understandably affected by the other variables. As expected, the reproduction number decreases by 0.016 as the stringency index of the previous month increases by one unit. As the case of Slovenia may illustrate, countries that were slow to implement stringent public health restrictions suffered sharp rises in infectivity (Fig. 1b).

Table 1

The results of fixed effect and SUR models with monthly data set.

	Model 1 Poll	Model 2 Stringency	Model 3 R	Model 4 Unemployment
	Coefficient	Coefficient	Coefficient	Coefficient
Fixed effect model
Pollt-1		−0.685∗ (−1.427, 0.056)	0.000 (−0.005, 0.006)	−0.054∗∗∗ (−0.083, −0.024)
Stringencyt-1	−0.005 (−0·045, 0.035)		−0.008∗∗∗, (−0.009, −0.007)	−0.010∗∗ (−0.018, −0.002)
Rt-1	0.254 (−0.968, 1.476)	5.432∗∗∗ (1.501, 9.363)		−0.557∗∗∗ (−0.769, −0·345)
Unemploymentt-1	−1.186∗∗∗ (−2.040, −0.332)	−9.297∗∗∗ (−11.747, −6.847)	0.108∗∗∗ (0.066, 0.151)
New cases per Mil.	−0.728∗∗∗ (−1.222, −0.234)	3.590∗∗∗ (2.139, 5.041)		−0.086∗∗∗ (−0.145, −0.028)
Constant	41.741∗∗∗ (36.275, 47.207)	125.965∗∗∗ (88.312, 163.618)	0.790∗∗∗ (0.418, 1.161)	10.106∗∗∗ (8.618, 11.593)
SUR
Pollt-1		1.805∗∗∗ (1.435, 2.176)	0.025∗∗ (0.002, 0.047)	−0.243∗∗∗ (−0.270, −0.217)
Stringencyt-1	0.384∗∗∗ (0.308, 0.461)		−0.016∗∗∗ (−0.025, −0.007)	0.099∗∗∗ (0.080, 0.118)
Rt-1	0.662 (−1.727, 3.050)	23.742∗∗∗ (17.904, 29.581)		−0.210 (−0.799, 0.380)
Unemploymentt-1	−3.938∗∗∗ (−4.373, −3.504)	7.084∗∗∗ (5.580, 8.588)	0.110∗∗ (0.020, 0.199)
New cases per Mil.	−1.044∗∗ (−2.051, −0.037)	4.207∗∗∗ (2.276, 6.138)		−0.338∗∗∗ (−0.588, −0.089)
Constant	37.765∗∗∗ (31.049, 44.480)	−85.914∗∗∗ (−110.671, −61.158)	0.505 (−0.280, 1.289)	10.010∗∗∗ (8.315, 11.705)
N	342	343	343	340

Notes: N is the sample size. ∗ indicates P < 0.10, ∗∗ indicates P < 0.05, and ∗∗∗ indicates P < 0.01.95% of confidence intervals are in the parentheses. New cases per million is logged variable.

Fig. 1

Time trends (COVID-19 and poll data) in Germany and Slovenia.

Sensitivity analyses

Table 2 summarizes the results of our alternative analysis using weekly averages. As with the main analysis, there are considerable correlations among the error terms, creating biased estimates for the fixed effects models (Appendix, Table A3). Accordingly, our analysis will again be based on the SUR models. The results closely follow those of the main analysis. As with the monthly analysis, more stringent public health restrictions increase the popularity of the government of 0.399 while suppressing the reproduction number of 0.027, and governments are punished for rising numbers of COVID-19 infections by 0.243. Also confirming the weekly results, higher polls and more COVID-19 cases increase the stringency index by 2.386 and 1.092, respectively. In contrast to the monthly analysis, however, the reproduction number of the previous week has negative effects on the stringency index of the present week. As the responses in Slovakia and Finland illustrate, governments hesitate to conduct strict health policy until there is a large number of new COVID-19 cases, although a rise in the reproduction number precedes a peak of new cases (Fig. 2 ). Finally, Appendix Table A4 presents the results from our second set of sensitivity tests, using the growth of polls in place of the absolute values. The results do not substantially differ from those of our main analysis.

Fig. 2

Governments hesitate to increase restrictions unless there is a clear surge in cases.

Discussion

Principal findings

A global pandemic of unprecedented scale, COVID-19 has brought into light the complex, intertwined nature of political behavior during a public health crisis. While a varying intensity of non-pharmaceutical interventions has been introduced at the advice of health experts, restrictions to freedom of movement also negatively affected the economy4, 5, 6 and agitated the citizens under confinement. The public’s dissatisfaction with public health restrictions was often so high that they culminated in rule-defying mass demonstrations19, 20, 21 across the globe. Against this background, political decision-makers were arguably under a constant dilemma of whether to prioritize politics or health. The significance of our research, therefore, lies in its attempt to unpack the relationship between political and economic considerations, health restrictions, and health outcomes.

Our main analysis finds that the political dynamics did not hinder stronger public health restrictions, given the same number of new COVID-19 infections. Because the error terms were highly correlated, we used SUR models to analyze the data. First of all, governments imposing more stringent policies were actually rewarded at the polls: a unit increase in the stringency index boosted the incumbent’s growth in popularity by 0.384, whereas one additional new case per million reduced the poll by 3.938. Contrary to what the mass demonstrations against COVID-19 restrictions may have led us to believe, the vast majority of the public seems to approve of stronger measures. Second, governments introduced stronger restrictive measures as necessary, even as they suffered from high levels of unemployment. A 0.1 rise in the reproduction number and one additional new case per million raised the stringency index by 23.742 and 4.207, respectively. A unit increase in the polls and a 1% rise in the unemployment rate each resulted in a 1.805 and 7.084 surge in the stringency index. Our empirical analysis indicates that both the public’s perception of the incumbent’s performance and the government’s self-evaluation of their own performance depends heavily on the number of new cases rather than the reproduction number.

Interestingly, the poll affects the reproduction number positively and the unemployment figure negatively. It is possible that this outcome was caused by an exogenous variable such as the government’s economic response to the pandemic: expansionary economic policies can increase the popularity of the government while also increasing the reproduction number and reducing unemployment. As our ’s focus is on physical public health restrictions rather than economic policies, however, fully explaining Models 3 and 4 is beyond the scope of this article. It also seems that public health restrictions become, in time, less and less popular. Fig. 1a illustrates, for instance, that Germany’s first lockdown was more popular in the polls than later restrictions of similar scale.

Furthermore, the results of the sensitivity analysis highlight a potential difference in medium-term and long-term considerations in pandemic decision-making. Yet such an interpretation deserves considerable caution. Because unemployment is not included in the weekly analysis, the apparent differences between weekly and monthly observations may be attributable to the omission of this key variable from the sensitivity tests.

Comparison with other studies

To the authors’ knowledge, this is the first study that attempts to unpack the complex interdependence of politics, economics, and public health decisions during the COVID-19 pandemic. Most existing studies analyzing the effectiveness of non-pharmaceutical interventions treat political and economic factors as control variables at best.22, 23, 24, 25 When non-health-related variables did enter the causal framework, scholars have focused on the impact political and economic considerations have on the effectiveness of public health restrictions and vice versa without acknowledging that the causal arrow could head in multiple directions.26, 27, 28, 29 The main innovation of our research is that we explicitly admit the possibility that the variables may be causally intertwined in a multitude of ways: the effect of health restrictions on the reproduction number is contingent on the effect of political and economic factors on the restrictions, which, in turn, is affected by the effect of the reproduction number and the health restrictions have on the political and economic variables.

Limitations of the study

Our study is limited in its scope. First of all, our statistical method is unable to offer detailed analysis about specific countries or periods. Moreover, because data were not readily available in other regions of the world, our study is restricted to comparing 27 European nations, whose long history of democratic political institutions and practices arguably renders them exceptional. As a result, the findings may also have limited external validity outside of this region. Among younger democracies, for instance, the public’s support of the incumbent government may have less to do with the efficacy of non-pharmaceutical interventions and more to do with clientelistic linkages. Therefore, to establish whether the relationships we identify could be observed more generally, future studies would need to extend the analysis beyond the European region.

Finally, our study is also limited in its ability to explain the poll’s effect on unemployment and the reproduction number. Because of our specific focus on physical—rather than economic—interventions during the pandemic, we have been unable to fully explore the underlying reasons behind the poll’s apparent effect on unemployment and infectivity.

Conclusion and policy implications

With new, more infectious variants on the rise; the world is yet to witness a complete end to the prolonged COVID-19 pandemic. Although our study is limited in its geographic scope, it illustrates a novel attempt to disentangle the complex relationship between political and economic considerations and health. As we expected, the popularity of incumbent politicians, unemployment, health restrictions, and the reproduction number are causally intertwined with each other. If health professionals had expressed concerns about the politics surrounding this pandemic, , , our study reveals that politicians and health professionals in Europe were not at odds with each other.

Our data illustrate that as a government grows more popular, it could implement more stringent health restrictions, curbing the spread of COVID-19. Moreover, when governments increased the stringency of their policies, they were not punished by the polls; to the contrary, our analysis reveals that politicians were rewarded for their implementation of strict rules. In other words, governments, with greater public support behind their backs, implemented stricter health restrictions that had greater impact on the reproduction number. Although further research must examine the external validity of our findings, these results indicate that the governments and citizens of Europe generally did not demand looser health restrictions for the sake of short-term economic gains.

Author statements

Ethical approval

None sought.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Competing interests

None declared.