key: cord-0464363-3ex5ihmh
authors: Mellacher, Patrick
title: The Impact of Corona Populism: Empirical Evidence from Austria and Theory
date: 2020-12-29
journal: nan
DOI: nan
sha: 6c12addb398e84d105a32117184e34d0ab291a6c
doc_id: 464363
cord_uid: 3ex5ihmh

I study the impact of opposition politics aimed at downplaying the threat of Covid-19. Exploiting a policy U-turn of a major Austrian right-wing party (FPOe), I first show that beliefs regarding the health risks of Covid-19 of FPOe voters vs. others diverged after the turn using a difference-in-differences approach. Using aggregate-level data, I study whether weekly Covid-19 deaths per capita are significantly positively correlated with support for the FPOe on the regional level. By linking aggregate- and individual-level data, I show that imputed regional beliefs about the economic and health impact of Covid-19 have a significant effect on cases and deaths per capita. Paradoxically, the FPOe vote share is significantly positively correlated with deaths per capita after the turn, but not with the reported number of infections. I hypothesize that this can be traced back to a self-selection bias in testing, which causes a correlation between the number of"corona skeptics"and the share of unreported cases after the turn. I find empirical support for this hypothesis in individual-level data from a Covid-19 prevalence study that involves information about participants' true vs. reported infection status. I finally study a simple heterogeneous mixing epidemiological model and show that a testing bias can indeed explain the apparent paradox of an increase in deaths without an increase in reported cases.

Following Acemoglu et al. (2013) and applying it to the case of the corona pandemic, populism can be defined as an anti-elitist view that receives significant support, but ultimately has adverse effects for the majority of the population. 1 Situations in which costs are mainly external and/or difficult to comprehend seem to be particularly susceptible to such populism. This is neither news for scholars who study views on (policies against) climate change, nor for epidemiologists who witness seemingly ever-growing doubt against vaccines, e.g. in the case of the measles.

The Covid-19 pandemic, however, has put a spotlight on these views as an imminent danger for society, as health care systems around the world had been brought to the brink of collapse.

In such a situation, governments must rely on compliance with containment efforts, as well as more or less on voluntary social distancing. Corona populism is, more succinctly, politics aimed at downplaying the threat of COVID-19.

If the level of support for such populist views is too high, a democracy has difficulties to implement policies that internalize these externalities effectively -witness the yellow vest protests against the carbon tax in France and, e.g., the protests of "Querdenken" against Covid-19 containment policies in Germany (Lange and Monscheuer 2021). Unfortunately, relying on individual responsibility to reduce the level of negative externalities seems to be particularly hopeless in such situations. As the dangers caused by corona populism grew apparent, it has received scholarly attention across scientific disciplines (Alashoor et al. 2020; Brubaker 2020; Eberl et al. 2021; Lasco 2020; Pevehouse 2020) .

We can hypothesize that a) supporters of political parties which adopted corona populism are more likely to underestimate the threat posed by COVID-19, as experimental evidence suggests that voters are more likely to adhere to the policy stance of their own party (Grewenig et al. 2020 ), a view that has been long supported by political scientists (e.g. Campbell et al. 1960; Kam 2005; Bechtel et al. 2015) 2 and b) that these beliefs translate into p. 3 / 81 behavioral differences between supporters of corona populist parties and the rest of the population, i.e. lower compliance with containment measures and less social distancing as shown by e.g. Allcott et al. (2020) for the American case. If this is true, the support for corona populist parties in a given community can help to predict the size of its COVID-19 outbreak.

In this paper, I study whether the policy stance of the Austrian right-wing populist freedom party (FPÖ) had an effect on the regional differentiation of the pandemic in Austria with regard to the FPÖ vote share. The FPÖ was the first political party to demand that the Austrian government should take drastic measures against COVID-19 (APA OTS 2020a). By the end of April, however, the FPÖ made a U-turn and demanded to "end the Corona madness" (APA OTS 2020b) by which they meant the containment measures taken by the government. In the end of November 2020, one representative of the party even went so far as to advise people not to participate in a mass testing program announced by the Austrian government to be held before Christmas because testing positive would mean that you would have to spend Christmas home alone (APA OTS 2020c). On January 31 st 2021, three MPs of the FPÖ participated at a banned demonstration against the lockdown (APA OTS 2021a). Later on, leading FPÖ politicians argued against the introduction of the "green pass" (APA OTS 2021b), as well as against vaccinating teenagers (APA OTS 2021c) and against compulsory vaccinations (APA OTS 2021d).

The case of the FPÖ is particularly interesting for two reasons: First, the party, its predecessor VdU and various splinter groups have won seats in every parliamentary election since 1949, when most former members of the Nazi party were allowed to vote again, and participated in five coalition governments. It has thus a longer and more stable tradition than other rightwing parties in Europe. At the same time, the party could never hope to achieve a majority in parliament on its own and it is thus not as established as the Republican party of the US.

Second, its clear policy stance subject to a U-turn at the end of the first wave of infections is a natural experiment that allows for a prime opportunity to identify the effects of partisan policy specifically compared to confounding factors that are merely correlated with partisan support.

Previous research on the effects of political polarization and populism on beliefs, behavior, and public health outcomes during the pandemic has mainly concentrated on the US. Allcott et al. (2020) show using mobile phone data on the county level that democratic counties exercise more social distancing (also confirmed by, e.g., Baradaran Motie and Biolsi 2021), but p. 4 / 81 also record more cases and deaths per capita than republican counties. Controlling for a large number of covariates, Gollwitzer et al. (2020) however find that Trump-leaning counties do not only exercise less social distancing, but that this is also linked to higher growth rates in the number of cases and fatalities. Allcott et al. (2020) also confirm that individual beliefs about the severity of Covid-19 are linked to self-reported social distancing using data from an online survey with US participants.

Further investigating what drives these differences, Fan et al. (2020) document that there are partisan differences in social distancing behavior and beliefs, which also depend on differences in news consumption using data from an online survey. Wu and Huber (2021) use a regression analysis of survey data to show that partisan differences in self-reported social distancing disappear once they control for beliefs and social norms. Bisbee and Lee (2020) show that Republican-leaning counties were more likely to practice social distancing when Trump emphasized the risks of Covid-19 on his Twitter profile. As seen in their analysis, however, Trump sent at best a mixed message about the severity of Covid-19, making causal analysis difficult.

Research on other countries than the US is sparser. Barbieri and Bonini (2021) show that a higher vote share for the Italian right-wing party Lega is associated with lower social distancing using regional mobility data. Like Trump's course, the Lega's policy was characterized by a zigzag course: first downplaying the pandemic, then agreeing to a lockdown, followed by a call for a fast re-opening. In February 2021, the party entered into a "national unity" coalition government. Eberl et al. (2021) show that "populist" attitudes -which they define as being anti-elitist, people-centered and having a "Manichean outlook" (following Hawkins and Rovira

Kaltwasser 2018) -are positively correlated with Covid-19 conspiracy theories in Austria using data from the Austrian Corona Panel Project (Kittel et al. 2020 (Kittel et al. , 2021 . Eberl et al. (2021) emphasize, however, that such views are to be found everywhere in the left-right spectrum and not tied to voters of the FPÖ specifically. Charron et al. (2022) show that excess mortality is higher in European regions where elite polarization is stronger in the dimension of European integration, which they argue proxies the strength of populism.

With this paper, I aim to contribute to our understanding of human behavior in the spread of infectious diseases. My research is mainly related to two strands of the scientific literature:

First, I contribute to the empirical research about the effects of politics on behavioral responses to the pandemic (e.g. Allcott et al. 2020; Fan et al. 2020; Gollwitzer et al. 2020; Milosh et al. 2021) , which is dealt with in a part of a broader body of literature on the causes and effects of behavioral differences in the pandemic (e.g. Bai and Brauer 2021; Barrios et al. 2020; Brzezinski et al. 2020; Bursztyn et al. 2020; Chernozhukov et al. 2020; Jung et al. 2020; Papageorge et al. 2020; Wright et al. 2020) .

I add to this literature by a) exploiting a clear policy U-turn of an opposition party that helps to identify the effects of partisan policy in contrast to other factors that are merely correlated with support for populist parties. My analysis suggests that the turn had an impact on the micro-level (i.e. beliefs of FPÖ voters vs. non-FPÖ voters), as well as on the meso-level (infections and deaths in districts with a high FPÖ vote share vs. low FPÖ vote share) with likely implications on the macro-level.

I further add to this literature by b) linking individual-level data on beliefs with district-level data on infections and deaths using demographic characteristics of the respondents and the districts. My analysis is complementary to individual-level evidence on the role of beliefs and self-reported social distancing (Allcott et al. 2020; Fan et al. 2020; Wu and Huber 2021) and

shows that imputed beliefs regarding the health and economic impact of Covid-19 have a significant impact on the distribution of cases and deaths per capita. More precisely, beliefs in a high health risk are associated with a lower number of cases, whereas beliefs in high risks to the economy are associated with a higher number of cases and deaths per capita. I argue that this relationship has been fostered by framing containment policy as a trade-off between economy and public health. Hence, people who are particularly concerned about the economy may be more skeptical about containment policies.

Finally, I add to the empirical literature by c) showing that "corona skepticism" is significantly correlated with the share of undetected cases, i.e. the dark figure, in Austria using an individual-level data source that includes information about policy views and true infection status determined by highly specific tests. This result suggests that estimates regarding the true number of infections (in contrast to the reported number of infections) must be corrected for political factors. This in particular has important implications on policy that uses a "traffic light" approach to regionally vary containment stringency based on data on reported infections as it has been in use, e.g., in Austria and Germany. p. 7 / 81 towards government policy, hence allowing me to connect beliefs with epidemiological characteristics on an individual level.

In this subsection, I use data from the Austrian Corona Panel Project (Kittel et al. 2020 (Kittel et al. , 2021 to explore how the policy switch affected individual beliefs of people who responded that they voted for the FPÖ (the treatment group) versus people who responded that they are enfranchised, but did not vote for the FPÖ (the control group). This data source includes information about partisan affiliation, demographic factors and a wide array of questions on beliefs and perceptions from, hence providing a comprehensive overview of the evolution of beliefs on the Covid-19 pandemic in Austria. While the panel is not balanced, i.e. not every respondent participated in every survey "wave", new respondents are filled in to match the demographic and political characteristics of those who have to be replaced. Every wave includes approximately 1500 survey respondents. In my analysis, I use data from waves 1-28, omitting wave 5, as it was conducted during the policy switch of the FPÖ on the 27 th of April 2020.

I focus on five survey questions that are initially coded as five-point Likert scales. These questions focus on the appropriateness of current government containment policies, as well as about the perceptions about the danger that Covid-19 poses to the private and public health and economic situation. The same questions were asked in every wave analyzed. I recoded the Likert scales to create 10 dummy variables which are 1 for initial values for 1,2 (such as low public health danger) or 4,5 (such as high danger to personal economic situation).

This recoding exercise helps to a) account for potential different behavior at the lower and upper end of the distribution, and b) facilitate the statistical analysis with the help of probit regressions. Figure 1 shows the evolution of beliefs of declared FPÖ voters (respondents who declared that they voted for the FPÖ in the last national elections) vs. non-FPÖ voters (all other respondents) according to this data, omitting one wave that was conducted during the FPÖ policy switch.

A quick graphical analysis suggests that health perceptions were closely aligned between the two groups before the switch, but diverged after the policy switch in the sense that FPÖ voters were more likely to believe that Covid-19 poses i) a low danger to public health and ii) own p. 8 / 81 personal health. Furthermore, FPÖ voters seemed to have been less likely to believe that it poses a high danger only after the policy switch.

On the other hand, there are pre-existing differences between the two groups with regard to their beliefs about the economic impact of the crisis and their policy views. FPÖ voters seemed to have been more likely to believe in a stronger impact on the personal economic situation (and less likely in a weaker impact) even before the policy switch and these differences persists after the switch. Curiously, FPÖ voters have been more likely to oppose government policy on Covid-19 in both directions, i.e. they were more likely to believe that government action against Covid-19 is exaggerated and that it was too lax than other survey respondents.

However, the policy switch seemed to have had a coordinating effect in this matter, as the support among FPÖ voters for the view that the measures are exaggerated increased drastically after the switch, while the support for the opposing view diminished to a point where FPÖ voters were less likely to hold that view than the rest of the population.

The Impact of Corona Populism p. 9 / 81 I then proceed to an econometric analysis based on a difference-in-differences approach using probit regressions and controlling for potentially confounding factors such as age and gender. More precisely, I estimate the following model:

Where , is the respective (binary) dependent variable, is the intercept, is a vector of (dummy) control variables, a dummy variable which is 0 for survey participants who declared to have voted for the FPÖ at the last national elections, a dummy variable for policy switch, which is 0 before the 27 th of April (i.e. waves 1-4) and 1 afterwards, is another vector of (dummy) control variables, which accounts for potential differential impact of sociodemographic variables before and after the policy switch (i.e. potential endogeneity, because the demographic characteristics are themselves correlated with support for the FPÖ), are time (=wave) fixed effects and , is an error term. I am particularly interested in the coefficient , as it aims to evaluate whether the policy switch had an effect on the beliefs of FPÖ voters vis-à-vis non-FPÖ voters in a particular dimension. Models (1) and (6) are specified in a very simple way by excluding any additional control variables. Models (2) and (7) include gender-, age-and education-specific dummy variables that show that: First, men were more likely to think that the containment strategy is exaggerated, but that the difference between men and women became smaller after the policy switch. This effect could be driven by gender differences in risk perceptions (Gustafsod 1998) and in the willingness to take risks (Charness and Gneezy 2012).

Second, young people below the age of 25, as well as (only after the switch) people above the age of 64 were less likely to think that government policy is exaggerated than the reference group (i.e. those between 25 and 64). Furthermore, people above 64 were less likely to think that the government response is too lax before the switch, but this age-specific effect reversed after the switch. This result may be driven by the different nature of the infection waves in Austria. As I will show in the next subsection, only very few people died during the first wave of infections (i.e. before the policy switch), which was rather well-contained whereas the second wave in autumn of 2020 was particularly deadly. Hence, the experience with a lesscontained spread of the virus may have caused men and people who are at risk due to their age to realize the necessity for (some) virus containment policies. The result that the youth has been less critical about government policy is at first glance counterintuitive, as they are at least risk from Covid-19. However, previous research from has shown that pro-social motives are more powerful in encouraging social distancing behavior than personal ones (Jordan et al. 2021) . Hence, young people may be more inclined to comply with social distancing, as they are convinced that they are staying at home for others.

Finally, education matters: University graduates were overall less critical towards government policy than the reference group (i.e. people who finished at maximum compulsory education), although the share of university graduates who believed that government response is too lax increased significantly after the policy switch relative to the reference group. Furthermore, p. 12 / 81

people who completed at most an apprenticeship were less likely to believe that government policy is exaggerated before the switch, but more likely to do so after the switch. Education is a proxy for two different factors: First, their type of job may influence their exposure to the virus and the crisis both with regard to health and the economy (i.e. more or less "rational" factors which may be counteractive). Second, a higher level of education was shown by Eberl et al. (2021) to decrease the level of science skepticism and the belief in conspiracy theories (i.e. "irrational" factors).

Finally, models (3), (4), (7) and (8) incorporate the beliefs about the danger that Covid-19 poses to personal and public health, as well as to the own economic situation and the economy as a whole. Expectedly, respondents who believe that Covid-19 poses a high risk to personal or public health are more likely to believe that government policy is too lax.

Conversely, beliefs in low health risks are associated with the belief that government response is exaggerated. The opposite relationship can be observed for respondents who believe that

Covid-19 poses a high danger to their personal economic situation or to the economy as a whole, as they were more likely to think that the government response was exaggerated and, in the case of the belief in a high danger to the economy, less likely to think that it was too lax.

The opposite is true for respondents who believe that Covid-19 poses a low danger to their personal economic situation. This result may be explained by the fact the public and scholarly discourse often portrays containment policy as a trade-off between economic and public health outcomes (e.g. Mendoza et al. 2020), even though this view has not been shared unanimously among the scientific community (e.g. Bethune and Korinek 2020). Hence, people who fear the economic fallout of the Covid-19 crisis in particular may be more inclined to be skeptical about containment policies. p. 14 / 81 In table 2, I show probit regressions with different dependent variables, namely beliefs connected to a low danger to health and economy. The simple model (1), as well as the model involving socio-demographic control variables (2) agree that FPÖ voters did not differ significantly from non-FPÖ voters with regard to the belief that Covid-19 poses a low danger to personal health before the policy switch. After the switch, however, FPÖ voters were more likely to adhere to this belief. Both the simple model (3) and the model (4) controlling for socio-demographic characteristics agree that FPÖ voters did not significantly differ from others in their belief that Covid-19 poses little danger to public health. Again, FPÖ voters are more likely to hold that belief after the switch in both models.

Models (5) and (6) show that FPÖ voters have been less likely to believe that Covid-19 poses little danger to their personal economic situation even before the switch and suggest that the policy switch did not have an impact in this regard. Finally, models (7) and (8) suggest that FPÖ voters did not differ with regard to the belief that Covid poses a low danger to the economy as a whole before or after the switch.

Men were more likely to believe that Covid poses a low danger in all dimensions, whereas young people were more likely to believe in a low personal danger. Curiously, respondents above the age of 64 were more likely to believe that Covid-19 poses little danger to them personally after the switch than the reference group (i.e. people aged 25-64). This counterintuitive effect may stem from three causes: 1.) this group is usually retired and hence arguable more able to practice social distancing, 2.) it had sooner access to vaccines 4 , 3.) potential psychological reasons such as denial. People above the retirement age were also more likely to believe that Covid hat little impact on their personal economic situation, most probably because they are almost all of them are retired and believed that the crisis would not have an impact on their pensions.

p. 16 / 81

Finally, university graduates were more likely to believe in a low danger to personal health than the reference group before the switch, but the differences disappeared afterwards. The initial effect may be driven by the differing teleworking capabilities. (3) and (4) suggest that this is the case for the high danger to public health. While FPÖ voters did not significantly differ with regard to their beliefs in a high danger to their personal economic situation (see models 5 and 6), they have been more likely to believe that Covid poses a high danger to the economy as a whole throughout the observation period (see models 7 and 8).

The results regarding gender, age and education mostly mirror the results from table 2. My main results are robust in each model. My analysis suggests that the policy switch:

1.) Increased the likelihood that FPÖ voters believed that government response against Covid-19 is exaggerated. This is not true for any other party.

Covid-19 is too lax. This is not true for any other party.

p. 20 / 81 3.) Increased the likelihood that FPÖ voters believed that Covid-19 poses a low danger to their personal health after the policy switch. This is also true for SPÖ (the social democratic party) voters, although it is important to note that SPÖ voters are the only political group that have been less likely to believe in a low danger to their personal health before the policy switch. Thus, the reduction for SPÖ voters represented a move to the mean belief, whereas the reduction for FPÖ voters represented a move away from the mean belief.

4.) Increased the likelihood that FPÖ voters believed that Covid-19 poses a low danger to public health after the policy switch. This is not true for any other party. 5.) Reduced the likelihood that FPÖ voters believed that Covid-19 poses a high danger to public health after the policy switch. This is also true for SPÖ voters -but again, SPÖ voters are the only political group that had been significantly more likely to hold this belief before the switch. Hence, this effect could also be driven by a regression to the mean belief.

My difference-in-differences analysis of the panel survey data from the Austrian Corona Panel Project (ACPP) (Kittel et al. 2020 (Kittel et al. , 2021 suggests that the FPÖ policy switch affected the beliefs of FPÖ voters vis-à-vis others with regard to i) health perceptions, and ii) policy views, but did not significantly affect their perceptions regarding the economic consequences of the crisis which also play a role in explaining policy views.

My main analysis relies on probit regressions that control for time fixed effects and various socio-demographic factors (age, gender, education) and allows the control variables to have a changing effect that over time suggests that the policy switch affected the beliefs and perceptions of FPÖ voters vs. non-FPÖ voters regarding health risks and government policy.

In my second study, I investigate, whether the policy stance of the FPÖ had an effect on the regional differentiation of the evolution of the pandemic in Austria. In order to do so, I draw on district-level data on the number of infections and deaths, which are available for a daily basis (BMSGPK 2022a). Studying county-level data is a standard approach followed by e.g. To get a first graphical intuition of the evolution of the pandemic in communities with a low or a high FPÖ vote share, I split the time series dataset into two groups, one for districts with a FPÖ vote share below or equal to and one above the median share of this party. Figure 1 shows the mean and standard deviation of the cumulated number of infections per 1,000 inhabitants and deaths per 100,000 inhabitants over time. This exercise suggests that districts, in which the FPÖ fared relatively well at the last national elections received slightly lower damage in the first wave of infections, reporting lower numbers of cases and deaths. In the second infection wave starting in autumn 2020, however, the cumulated death toll in these districts surpasses the total number of deaths in the other districts, indicating that the second wave hit districts with a high FPÖ vote share much harder. The differences between the two groups then seem to be stable until the autumn of 2021, when the "Delta" variant hit Austria.

However, we do not observe the same clear trend in the cumulative number of reported cases per capita, as districts with a low FPÖ vote share continued to have a higher number of cumulative cases until the beginning of 2021, when there was already a clear difference with regard to mortality (see fig. 1 ). Even afterwards, districts with a high FPÖ vote share only had a small and potentially not statistically significant edge in terms of infections. While the result regarding the number of deaths would indeed suggest that the FPÖ U-turn had an impact on the regional differentiation of the pandemic, the result regarding the number of cases is counterintuitive: after all, nobody can die from Covid-19 without contracting it. I develop an explanation for this phenomenon subsequently.

In order to confirm whether this graphical intuition is also statistically significant, especially when considering district-specific characteristics which may drive this pattern such as, e.g., the age structure of the population, I then turned to panel regression analysis.

In order to do so, I first create a balanced panel data set of weekly data on infections and deaths based on the daily data. The FPÖ campaign against the containment measures started on the 27 th of April 2020, i.e. calendar week 18. I thus created a dummy variable which is 1 beginning with the 18 th calendar week of 2020 and 0 before. 5 The data analyzed in this paper range until 31st of December 2021, which corresponds to the end of the infection wave surrounding the "Delta" variant.

I then merged the dataset on infections and deaths with data on the results of the last national elections (2019) on the district level as a proxy for the influence of the FPÖ as well as regional data about vaccinations (BMSGPK 2022b (BMSGPK , 2022c . Finally, my analysis in the preceding subsection, which utilizes data from the Austrian Corona Panel Project (ACPP) (Kittel et al. 2020 (Kittel et al. , 2021 shows that FPÖ voters a) have differed from non-FPÖ voters before and after the switch with regard to perceptions about the danger to the economy and personal economic situation, and b) that some beliefs regarding health and economy changed after the policy switch for some socio-demographic characteristics.

This result suggests that a two-way fixed effects estimator may be prone to an endogeneity problem, as some of these characteristics are correlated with the support for the FPÖ (such as, e.g., the level of education). In order to address this issue, I imputed district-specific beliefs by linking the individual-level survey data to the aggregate-level administrative data according to three observable characteristics: gender, age, education. It is important to note that this exercise will inevitably also account for some of the impact of corona populism and hence p. 23 / 81 tend to underestimate the effect of the FPÖ policy switch exactly due to the correlation of the FPÖ vote share with these characteristics. Nevertheless, this exercise is important to help to establish a "lower bound" of the impact of the FPÖ policy switch.

In order to impute the regional belief indices, I first split the survey respondents into eight different demographic groups and computed mean values for the perceptions in each of the 10 dimensions used in the analysis in the preceding subsection for each group and wave. These criteria were chosen based on a) their relevance as explanatory factors of beliefs and perceptions as shown in the previous subsection, and b) data quality, as i) district-specific Austrian data does not allow us to fully disentangle age, gender and highest education and ii) young people likely did not complete their education yet, which would distort the meaning of, e.g., the category "highest education: compulsory schooling". If the mean belief for group ℎ with regard to dimension in week is ,ℎ, and the share of group ℎ in district is ,ℎ , then the imputed belief with regard to dimension d , , is determined according to equation 2 6 : 6 Since the data on infections and deaths is weekly, but the surveys from the ACPP were not collected on a weekly basis, I linked the two datasets by assuming that the beliefs in a given week that lies between waves X and X+1 are equal to those of wave X, if the date of this week's Sunday is closer to the start date of wave X than to the start date of wave X+1, and equal to the beliefs of wave X+1 otherwise.

p. 24 / 81 , , = ∑ ,ℎ ,ℎ, ℎ (2)

Using the combined dataset, I then estimate two-way fixed effects models of the following type, again using the fixest package (Bergé 2018) for the programming language R (R Core Team 2020):

Where , is the dependent variable, either the number of deaths per 100k inhabitants or the number of reported cases per 1k inhabitants, , a vector of control variables, a dummy indicating the policy switch, i.e. 1 after the FPÖ policy switch and 0 before, the FPÖ vote share in the specific district. District fixed effects are denoted with and time fixed effects with . , , and are coefficients, and is a vector of coefficients. Finally, , is the error term.

In all models, I am interested in the coefficient , as it tells us whether the FPÖ vote share conditional on the corona populist turn of the FPÖ has an impact on the dependent variable,

i.e. cases or deaths. It is important to note that this model does not allow an interpretation about whether increased support for the FPÖ predicts an increase in cases or deaths in total, as the time-invariant effect of the FPÖ-vote share is dummied out. 7

This subsection presents the results of the models on the number of i) cases per 1k inhabitants, and ii) deaths per 100k inhabitants. The first model includes the lagged cumulative number of cases per 1k inhabitants, as well as the lagged number of deaths as control variables. The lagged number of deaths is significant and negative, which suggests that people react to an increased perceived danger posed by Covid-19 by changing their behavior, i.e. by practicing social distancing. Unexpectedly, we do not find a "herd immunity" effect from the (lagged) number of cumulated previous cases in this model.

Models (2) and (3) the imputed belief that Covid-19 poses a high danger to public health reduces the number of deaths, whereas the imputed belief that Covid-19 poses a high danger to the economy increases the number of fatalities. The first coefficient reflects the interplay of two counteractive tendencies: on the one hand side, people who are more exposed to the virus may rationally believe that they are more at risk, hence increasing the number of infections.

On the other hand, people who are more risk-aware may be more inclined to practice social distancing, wear masks, get vaccinated etc. It is obvious from the regression results that the second effect prevails with regard to the number of cases. The second finding, i.e. that the belief that Covid-19 poses a high danger to the economy increases the number of cases, is less obvious, but also in line with the expectations based on the results and reasoning introduced in the previous subsection. As pandemic policy is often portrayed as a trade-off between saving the economy and saving lives, the belief that Covid-19 poses exceptionally high economic costs is related to the belief that government policy puts too much emphasis on containment policy.

Models (4)-(7) then also account for vaccinations, which have played a role in Austria since the beginning of 2021. Model (4) includes the cumulative number of second dose vaccinations (which is only available on the state-level as a time series). As expected, the respective coefficient is significant and negative due to an increase in herd immunity. Models (5)- (7) differ from model (4) 2022 and assuming that this share stays constant, thus matching it with the state-level time series readily available. While this imputation does not come without costs, as the relative distribution of vaccines may have changed over time, this imputation increases the significance and absolute value of the respective coefficient, hence providing credibility to the mechanism. Interestingly, the expected herd immunity effect from the cumulative number of previous cases is significant once we account for vaccinations. Again, this effect is stronger for the imputed district-level vaccination timeline, lending credibility to the imputation procedure.

Models (6) and (7) then also control for the imputed belief indices. While there are no notable changes for model (7), a belief in high risks for the economy as a whole is not significant anymore in model (6), whereas a belief in high risks for the personal economic situation becomes significant. This result suggests that, as one might have expected, the impact of corona "skepticism" is partly transmitted through the channel of vaccination rates. Table 5 shows the results of the panel regressions predicting the number of deaths per 100,000 inhabitants. The coefficient of interest, i.e. the effect of the interaction term between the policy switch dummy and the FPÖ vote share, is positive and statistically significant at the 0.1% level for any model. This result suggests that the corona populist turn of the FPÖ did have an impact on the regional distribution of deaths in Austria in the sense that the number of Covid-19 deaths per capita after the policy switch are correlated with the district-level support for the FPÖ.

Every model also includes the lagged dependent variable (i.e. the number of deaths per 100k inhabitants in the previous week), which is significant and positive. While deaths do not necessarily produce new deaths (save for a Zombie apocalypse), the lagged dependent variable is a proxy for other important factors with regard to the evolution of the pandemic such as infections, ICU capacities etc.

Models (2) and (3) also feature the imputed belief indices as described above. Only one index is significant, namely the belief that Covid-19 poses a high danger to the personal economic situation. This belief can be seen as a proxy for a certain type of skepticism (as reasoned above), and accordingly increases the number of deaths.

The fact that the vote share of the FPÖ is strongly correlated with deaths after the policy switch, but not with cases per capita, seems at first glance to be paradoxical and to sow doubt on the hypothesis that the corona populist turn of the FPÖ contributed to the spread of the virus. However, Covid-19 tests have largely been conducted on individuals who self-reported their symptoms or who are named as being close contacts. Thus, they have in one way or another often been voluntary, in particular before the introduction of compulsory tests needed for certain activities in the beginning of February 2021, which means that there may have been a self-selection bias in testing. 8 We can hypothesize that people who underestimate the virus (the "corona skeptics") or even deny the existence of the virus (the "corona deniers") are less likely to report an infection and 8 From the beginning of February 2021 to the beginning of March 2022, either Covid-19 tests or a "green pass", i.e. proof of testing, vaccination or immunity due to a previous infection, were required for a varying number of activities. However, much of the testing regime relied on the use of self-tests that a) offer relatively low sensitivity and b) that can easily be manipulated. Furthermore, the largest relative increase in mortality with regard to the FPÖ vote share seems to have happened before February 2021, as can be seen in fig. 2. p. 29 / 81 to name contacts. In this case, the number of deaths per infection in such communities would be higher.

In order to conduct a first test on this hypothesis, models (4)-(6) also control for the number of cases per capita in the two preceding weeks. Holding cases constant, the mortality rate may diverge for two reasons: first, due to a higher true infection fatality rate, given by, e.g., different age structures of the population (or, more precisely, the infected share of the population), and second due to a higher perceived infection fatality rate given by a higher share of undetected cases (i.e. a higher dark figure). It is interesting to see that the coefficient of interest only changed marginally. Since the age structure of the population did not change during the policy switch, this result hints to the fact that the policy switch indeed increased the share of undetected cases, i.e. the dark figure. This result would also explain the non-result regarding an impact on the number of cases: While the policy switch did not have an impact on the number of reported cases, there is reason to believe that it did have an impact on the true number of cases.

I further explore the existence of such a testing bias empirically in subsection 2.3, and explore whether such a testing-bias is indeed able to explain the puzzling result that I do find a significant impact of the policy switch on the regional distribution of deaths, but not reported cases, in a stylized epidemiological model in section 3. 

Appendices B-E show the results of several robustness checks. The first robustness check concerns a potential endogeneity due to heterogeneous evolution of the pandemic prior to the policy switch. The overview of cases and deaths shown in figure 2 suggests that those 50% of the districts, in which the FPÖ achieved its strongest outcomes recorded less deaths during the first, but more deaths in particular during the second wave. Thus, one could hypothesize that the underlying mechanism driving the results is in fact not a political, but an epidemiological one: if more people became infected and/or died during the first wave, the subsequent waves could be milder either due to herd immunity or due to increased awareness of the danger of the virus, and the FPÖ vote share could merely correlate with this underlying mechanism. I explore this hypothesis in appendix B and show that my results are robust to including an interaction term of the intervention with either the number of deaths or the number of cases before the intervention as an additional control variable, even though such a counteractive effect indeed exists. In a related concern, the social share of the population affected by the coronavirus may differ between the first and the subsequent waves, e.g. the second wave may have hit more elders than the first wave, thus increasing the case fatality rate. I thus include interaction terms between various demographic groups and the policy switch to explicitly account for this potential demographic endogeneity instead of relying on the belief indices. My results are robust to such an inclusion, although the population above the age of 64 seems to have been hit harder in the infection waves following the first one (i.e. after the policy switch).

In the second type of robustness checks, presented in appendix C, I check whether the results change when I include interaction terms between the policy switch and the vote share of other parties. All results are robust to the inclusion of interaction terms with any other party. Only a few additional results are significant for the number of deaths per 100k inhabitants, namely the interaction term between the policy switch and the vote share of a) the social democratic party (SPÖ), where it is positive, b) the major governmental party (ÖVP), where it is negative, as well as c) the combined vote share of minor parties, where it is positive in some models. It is important to note, however, that these results are not robust to the robustness checks introduced in appendix B and thus are likely an artifact of epidemiological considerations.

p. 32 / 81

Alashoor et al. (2020) study social distancing behavior in the United States using a survey and show that, for their respondents, not partisanship, but their vote in the presidential elections 2016 mattered for compliance with social distancing measures. Respondents who voted for Trump in 2016 were less likely to follow social distancing rules, if their attitude towards social distancing was negative. The last Austrian presidential elections before the Covid-19 crisis were also held in 2016, where the candidate of the FPÖ, Norbert Hofer, lost in the run-off against the candidate of the Greens, Alexander van der Bellen. In the third robustness check, I hence test whether including an interaction term between the vote share for Norbert Hofer and the policy switch influenced the results. Interestingly, this exercise even increases the effect size for the interaction term of the FPÖ and the policy switch with regard to the number of deaths, whereas the coefficient for the new interaction term is not significant.

In the final robustness check, presented in appendix E, I use alternative model specifications in order to explore the robustness of my results in the face of the dynamic panel bias (Nickell 1981) , which comes into play for fixed effects models that include an autoregressive term where the underlying data has a small t and large n. Since t>n in my panel, the bias should not be of a great concern. Nevertheless, I estimate i) static panel models (tables E1 and E2), i.e. models without an autoregressive term, as well as ii) pooled models without any fixed effects, but including the autoregressive term. This procedure is suggested by the literature (see e.g.

Angrist and Pischke 2009) because it helps to establish boundaries of the true effect, since static panel models will overestimate the coefficient, whereas pooled models will underestimate it.

My results are found in all but two models that do not account for fixed effects or the number of cases, but do include the belief indices as derived from the Austrian Corona Panel Project. This is not a reason for great concern, as a) this estimator aims to provide a lower bound to the true level of the coefficient, b) introducing the beliefs will also underestimate the true effect size, as the FPÖ switch had affected beliefs in crucial dimensions and the demographic characteristics used to link the two data sources are correlated with the FPÖ vote share (as argued above), and c) this model specification is not very sophisticated as it omits not only time-invariant characteristics of the districts, such as the age structure, but also time-variant characteristics of the epidemic such as lockdowns, mutations etc. My results regarding an p. 33 / 81 increase in the dark figure (i.e. positive effect on the number of deaths when holding the number of cases in the previous two periods constant), however, even hold in those settings.

The aggregate-level evidence provides a puzzling result, as it suggests that the FPÖ policy switch has affected the regional distribution of deaths, but not of cases per capita. I hypothesized that this is due to a self-selection bias in testing that causes a correlation between the FPÖ vote share and the share of undetected cases, and found some evidence for this hypothesis in the aggregate-level data. This subsection now aims to establish a link between "Corona skepticism" and the true infection status, as well as the dark figure, on the individual level.

While there is no Austrian individual-level data source that combines partisan affiliation with infection status, the Covid-19 prevalence study conducted in November 2020 (Paškvan et al. 2021 ) provides information about the individuals' stance on Covid-19 containment policy, true infection status (determined with a PCR test), past infection status (determined with an antibody neutralization test) and reported infection status. This study was conducted by the Austrian statistical office (Statistik Austria) and its participants were randomly chosen to form a statistically representative sample.

These data thus allow to test the hypotheses that "corona skeptics", i.e. people who are opposed to containment policies, are more likely to contract Covid-19 and less likely to test themselves.

In order to conduct this analysis, I first transform the 5-point Likert scale answer on a question related to the individual's policy stance to the binary variable "corona skeptical", where a 1 covers the views that the policy measures against Covid-19 are "definitely exaggerated" or "rather exaggerated", whereas 0 implies that the individual thinks that the measures are "suitable", "rather insufficient" or "definitely insufficient". In order to test whether the graphical intuition holds, I estimate the following model using probit regressions:

Where is a binary variable describing whether a) a positive PCR/antibody test has been conducted or b) individuals were already known to be infected prior to the study, is a binary variable describing whether an individual is "corona skeptical" or not as discussed above, is the coefficient of interest and is an error term. Finally, the fourth model shows that corona skeptical individuals were more likely to have gone through a past infection. Again, all of these results support the self-selection bias hypothesis. Furthermore, the fact that there is no significant difference between regarding reported infections seems to mirror the aggregate-level results. symptoms of Covid-19. Accordingly, only people in the quarantine compartment may die. Holding constant the fraction of infected who will eventually die (i.e. the true infection fatality rate), the fraction of quarantined who die (i.e. the reported infection fatality rate) depends on the fraction of non-critical cases who opt to get tested voluntarily, i.e. on the fraction of critical cases in the quarantine compartment.

2.) I split the compartments governing the susceptible, the infected and the quarantined to incorporate two different groups: one group showing low compliance (the corona skeptics) and another showing high compliance (the majority). I consider differences in a) social distancing, and b) the propensity to get tested. I also consider the case of homophilic mixing, i.e. that individuals of a certain group are more likely to get into contact with members of their own group than members of the other group (which is why I need two different compartments for the infected). (2008), which in turn largely follows p. 37 / 81 Nold (1980). 9 In contrast to comparable models such as the homophilic mixing model proposed by Ellison (2020), the model used in my paper is able to replicate the standard homogenous mixing model as a special case if the behavior of the two types of agents (especially the basic reproduction numbers 01 and 02 respectively) is equal. The model is most closely related to the one proposed by Bai and Brauer (2021) , who also study a SEIR model populated by two types of agents who differ in their basic reproduction number and add a quarantined compartment. In contrast to their model, however, my model also considers differences in testing between the two groups (to account for the Austrian empirics), as well as the case of homophilic mixing. Their model, on the other hand, also features an exposed compartment in order to capture pre-symptomatic infections.

The laws of motion between the different compartments in my model are given as follows,

where denotes the susceptibles of group i, the infectious, the quarantined, the size of group i at period 0, the number of infectious contacts from a member of group i, ℎ the homophily of social contacts, the propensity to get tested, 0 the basic reproduction number of group i, 1 the duration the illness, and the fraction of detected cases who eventually die:

̇( ) = ( 1 ( ) + 2 ( )) + ((1 − 1 ) 1 ( ) + (1 − 2 ) 2 ( ))

̇( ) = ( 1 1 ( ) + 2 2 ( )) (9)

p. 38 / 81

In order to better disentangle the effects of behavioral differences of the two groups, I make the following practical assumption: Individuals of both groups are equally likely to die as a result of an infection with a probability of . We can thus set the probability that a quarantined person dies at = and set a lower boundary for , as I assumed previously that at least all critical cases are tested, i.e. ≥ .

Let us first consider the case of homogenous mixing, i.e. ℎ = 0 and 0 = 01 = 02 . In this case, we can immediately see that = = 1 + 2

. Thus, if we normalize the population to 1, i.e. 1 + 2 = 1, the dynamic governing the susceptibles collapses to the dynamic of the classical one-group SIR framework, i.e.:

In such a case, differentiating between two infectious compartments is unnecessary (see proposition 1).

Proposition 1: Suppose two groups who mix homogeneously and only differ with regard to their propensity to get tested. Such a difference can only affect both groups equally (relative to their share) in terms of deaths or the sum of infected and quarantined.

Proof: Differentiating eq. (5) with regard to I1 and I2 yields the same results, hence the relative share of as part of the total infected does not have an impact on the evolution of : Nevertheless, we could use a homogenous mixing framework to consider differences in the propensity to get tested (and subsequently get quarantined), i.e. a different evolution of 1 and 2 .

As is common in this stream of literature ( While result 1 shows that a self-selection bias in testing can indeed be sufficient to explain the aggregate-level phenomenon even in a simple homogenous mixing framework, this model has two apparent problems: First, proposition 1 shows that the total share of infected people cannot differ between the two groups within a given population (e.g. within a district), i.e. skeptics could not be affected in a different way than non-skeptics in a given district, which seems to be questionable. Second, this model assumes that there are no behavioral differences between the two groups except for the different testing behaviors. In order to address these issues, I extend my model to heterogeneous mixing in the following two subsections.

As soon as the two subpopulations engage in different activity patterns, i.e. 01 ≠ 02 , homogenous mixing is implausible. If, for instance, group 1 only has one infectious contact per day, whereas group 2 has five, members of group 2 cannot on average have 2.5 infectious contacts with members of group 1, if the two groups are equal-sized. The specification by Brauer (2008) , which provides the basis of my model, accounts for this fact. If activity patterns differ, but mixing is not homophilic, it is proportionate, i.e. members of a specific group meet members of another specific group according to their relative population shares and basic

The Impact of Corona Populism p. 41 / 81 reproduction numbers as specified above. As a result, outcomes for both groups cannot be used interchangeably anymore. Instead, we must trace 1 and 2 separately.

Result 2: Suppose two groups 1 and 2 who mix proportionately and only differ with regard to their basic reproduction number ( 02 > 01 > 0). Then an increase in a) the basic reproduction number of 2 02 or b) the share of group 2 2 increases the cumulative number of infected for both groups, but the increase (relative to their group size) is stronger for group 2. 

Finally, I consider the case of homophilic mixing (ℎ > 0), i.e., when "corona skeptics" are even more likely to meet each other than given by their relative basic reproduction numbers. Such p. 42 / 81 a homophily in mixing could be created, for instance, by the anti-lockdown protests, which were frequently held in Austria. These settings are of particular interest, as they surely also increase the number of social contacts of the group of skeptics. Figure Before the policy switch, declared FPÖ voters did not exhibit statistically significant differences with regard to their beliefs about personal or public health impact of the Covid-19 crisis. They were, however, significantly more likely to believe that it poses a high danger to the economy.

FPÖ voters were furthermore more likely opposed against government policy even before the policy switch, although their opposition was stronger in both dimensions, i.e. they were more likely to believe that government policy was exaggerated and more likely to believe that it was too lax. I then investigated whether the FPÖ policy switch could have had an effect on the regional distribution of infections and deaths in Austria based on the district-level support for the FPÖ at the last national elections using two-way fixed effects models with different specifications and a wide array of robustness checks.

My regression analysis suggests that the policy switch had a significant impact on the regional distribution of Covid-19 related deaths, but not of the regional distribution of (reported)

infections. This result is puzzling, as nobody can die from Covid-19 without contracting it first.

I hypothesized that the solution to this puzzle can be found in a self-selection bias inherent to the Austrian containment policies, in particular before the introduction of mandatory testing in February 2021: The policy stance of the FPÖ caused their voter base to take the virus less seriously, who then did not only practice less social distancing, but also reported their symptoms less often, which means that they were less likely tested. Hence, the true infection rate among FPÖ voters could be higher than those among others due to an increased share of unreported cases, i.e. dark figure. I found some evidence in favor of this hypothesis by an increase in the case fatality rate, and I further explored this hypothesis using a different data source, as well as a theoretical framework later on in the paper.

In a subset of models, I combined the aggregate-level dataset with the individual-level data from the Austrian Corona Panel Project to impute regional beliefs about the Covid-19

pandemic and to account for a potential endogeneity with regard to "skeptical" beliefs. I showed that both imputed health perceptions and beliefs about the economic fallout are significantly correlated with Covid-19 cases and deaths per capita. While the belief that Covid-19 poses a high risk to public health is associated with a decrease in the number of cases, but an increase in the case fatality rate (which suggests some degree of rationality in perceptions and behavior), beliefs in high economic risks are associated with an increase in the number of cases, deaths and in the case fatality rate, suggesting that this belief is indeed a proxy for at least some kind of "corona skepticism".

The Impact of Corona Populism p. 45 / 81

In order to further explore the self-selection testing bias hypothesis, I then turned to individual-level data from the Austrian Covid-19 prevalence study conducted in November 2020 (Paškvan et al. 2021 ) and found that "corona skeptics", i.e. people who think that the containment measures are exaggerated, are more likely to test positive for Covid-19, more likely to be an undetected prior and current case, but not significantly more likely to be an officially known case. All of these findings support the self-selection bias hypothesis.

Finally, I tested whether the proposed mechanism, i.e. the testing bias, can indeed explain the aggregate-level outcomes in a simple theoretical framework. In order to do so, I extended the classical SIRD to incorporate testing (with corresponding quarantine), heterogeneous behavior and heterogeneous mixing. This model is populated with two groups who may behave differently: the corona skeptics and the majority, where the former group has a lower propensity to get tested (which is empirically calibrated using the data from the Austrian Covid-19 prevalence study) and may have a higher basic reproduction number. I explored the properties of this model for three cases: a) homogenous mixing, b) proportionate mixing, and c) homophilic mixing.

My analysis showed that even a simple homogenous mixing setting allows for a situation in which an increase in the share of skeptics increases deaths, but has a non-linear inverted Ucurve shaped impact on reported infections (i.e. increases them for lower shares of skeptics, but decreases them for higher ones). I further showed that homophily in mixing reduces the degree of non-linearity of the impact of an increase in the number of "skeptics" on cases and deaths. Homophily in mixing seems to be plausible, as a) Austrian government policy had a special focus on reducing the transmission in situations where homophily can be expected to be low (e.g. compulsory mask-wearing in shops, public transport and schools etc.), whereas b) large-scale protests organized by "corona skeptics" may have significantly increased the number of social contacts in particular among "skeptics" (see Lange and Monscheuer 2021 for an analysis of the German case).

Perfect homophily in mixing hence enables an increase in deaths without a change in reported cases in a special case. However, the simulations also show that the parameter space that allows for a relationship that may be indistinguishable from the special case is much larger, in particular if the empirical distribution of cases is noisy. This result suggests that this mechanism is relevant beyond the special cases. Hence, a testing bias as it can be found in p. 46 / 81 individual-level data can indeed explain the apparent aggregate-level paradox that the corona populist turn of the FPÖ seemed to have influenced the regional distributions of deaths (as the FPÖ vote share is correlated with deaths per capita after the policy switch), but not reported cases (as there is no statistically significant relationship with the FPÖ vote share).

The research presented in this paper can be extended in numerous ways. In particular, it seems promising to study corona populism and skepticism in a more complex model. One way to go would be to increase the complexity within the SIR framework. Another option would be to turn to a new class of models. Agent-based models such as the COVID-Town model (Mellacher 2020) are capable of modeling the spread of the virus via social networks and explicitly modeled heterogeneous agents, who can follow sophisticated behavioral rules. This level of analysis can be expected to be highly useful to better understand the impact and evolution of corona skepticism. For instance, it may make a big difference whether a corona skeptic faces many customers or is introverted and unemployed. However, this method can also help to better understand the emergence and dynamics of corona skepticism, e.g. by modeling heterogeneous risk perceptions, willingness to take risks, or even opinion dynamics of corona skepticism or corona populism. I hope to be able to study some of these questions in the future.

Acknowledgements: A previous version of this paper was featured in the preprint series -22405.028 -22226.109 -18528.156 -19490.093 -18598.105 -18550.562 -17462.432 -16958.660 Std.Errors by: wave & respid low danger X X high danger X X * p < 0.05, ** p < 0.01, *** p < 0.001 low danger X X high danger X X * p < 0.05, ** p < 0.01, *** p < 0.001 X X X * p < 0.05, ** p < 0.01, *** p < 0.001 Table A7 : Opinion on government policy controlling for Greens

(2) (3) (4) (5) (6) low danger X X high danger X X * p < 0.05, ** p < 0.01, *** p < 0.001 low danger X X high danger X X * p < 0.05, ** p < 0.01, *** p < 0.001 -27726.314 -27480.863 -20757.320 -20500.110 -27423.567 -26943.694 -8132.444 -8013.248 Std. -22378.774 -22218.946 -18577.675 -19468.734 -18629.698 -18586.206 -17515.334 -17015.594 Std. -27723.223 -27461.962 -20646.523 -20393.426 -27422.251 -26943.231 -8133.122 -8016.988 Std. -20593.715 -20584.217 -20591.356 -20367.336 -20357.553 -20365.486 Std. I tested whether introducing an interaction term between the vote share of one of the other parties (the conservative ÖVP, the greens, the social democratic SPÖ, the liberal NEOS, as well as the combined vote share of minor parties such as the FPÖ splinter group BZÖ) and the policy switch changes the results in order to investigate whether the observed relationships could merely be correlated with opposition to the ÖVP-greens government, as one of the reviewers of a previous version of this paper suggested. My results are robust in all 65 models.

In order to keep this paper within a reasonable length, however, I only report the results tables where the interaction term of the intervention and the vote share of the other party is statistically significant (p<0.05). (2) (3) (4) (5) (6) * p < 0.05, ** p < 0.01, *** p < 0.001 

Freiheitliches Corona-Leugner-Trio Hafenecker

FPÖ -Kickl/Steiner: Opposition muss Epidemie-und Covid-19-Maßnahmengesetz gemeinsam im Bundesrat blockieren

Coronavirus: Gesundheitsminister Mückstein hält an Impfempfehlung für Kinder und Jugendliche fest

FPÖ -Kickl: Österreich ist mit heutigem Tag eine Diktatur!

V (2022a). modelsummary: Summary Tables and Plots for Statistical Models and Data: Beautiful, Customizable, and Publication-Ready

marginaleffects: Marginal Effects, Marginal Means, Predictions, and Contrasts

The Effect of Face Mask Use on COVID-19 Models

County-Level Determinants of Social Distancing (or Lack Thereof) during the COVID-19 Pandemic

Populism and Political (Mis-) Belief Effect on Individual Adherence to Lockdown during the COVID-19 Pandemic in Italy

COVID19: Erroneous Modelling and Its Policy Implications

Civic capital and social distancing during the Covid-19 pandemic

How to Design Virus Containment Policies? A Joint Analysis of Economic and Epidemic Dynamics Under the p

Heterogeneity in disease-transmission modeling

Socio-demographic factors associated with self-protecting behavior during the Covid-19 pandemic

The COVID-19 Pandemic, International Cooperation, and Populism

Optimal COVID-19 quarantine and testing policies

R: A language and environment for statistical computing. R Foundation for Statistical Computing

Dauersiedlungsraum Abgrenzung

Corona-Cluster nach Hochzeit mit 700 Gästen

Wirtschaftsdaten auf Bezirksebene

Poverty and economic dislocation reduce compliance with covid-19 shelter-in-place protocols

Partisan differences in social distancing may originate in norms and beliefs: Results from novel data

Paket Bevölkerungsstand -Politischer Bezirk

Preliminary estimation of the basic reproduction number of novel coronavirus (2019-nCoV) in China, from 2019 to 2020: A data-driven analysis in the early phase of the outbreak

 -22578.227 -22398.654 -18665.767 -19617.586 -18627.493 -18578.190 -17502.045 -17012.048 Std.