Political Regime and COVID 19 death rate: efficient, biasing or simply different autocracies ?
PPolitical regime and COVID 19 death rate: efficient, biasing or simply different autocracies?
Guilhem Cassan ∗ and Milan Van Steenvoort † January 26, 2021
Abstract
The difference in COVID 19 death rates across political regimes hascaught a lot of attention. The “ efficient autocracy ” view suggests that autoc-racies may be more efficient at putting in place policies that contain COVID19 spread. On the other hand, the “ biasing autocracy ” view underlines thatautocracies may be under reporting their COVID 19 data. We use fixedeffect panel regression methods to discriminate between the two sides ofthe debate. Our results show that a third view may in fact be prevailing:once pre-determined characteristics of countries are accounted for, COVID19 death rates equalize across political regimes. The difference in deathrate across political regime seems therefore to be primarily due to omittedvariable bias. ∗ University of Namur, CEPR, DEFIPP, CRED and CEPREMAP. Email: [email protected] † Maastricht University. Email: [email protected] are grateful to Jeremie Decalf, Romain Lutaud, Glenn Magerman, Marc Sangnier and Vin-cenzo Verardi for helpful discussions and suggestions. We thank seminar participants at UNa-mur. Guilhem Cassan thanks CEPREMAP and the FNRS for financial support. Research onthis project was financially supported by the Excellence of Science (EOS) Research project ofFNRS O020918F. All errors remain our own. a r X i v : . [ ec on . GN ] J a n Introduction
While democratic countries have previously been shown to overperfom comparedto autocracies with respect to health outcomes (Franco et al., 2004; Besley and Ku-damatsu, 2006; Kudamatsu, 2012; Bollyky et al., 2019; Pieters et al., 2016), datashows that, in the specific case of the COVID 19 pandemic, democratic countriesmay be fairing much worse (Sorci et al., 2020), as illustrated in Figure 1. Indeed,50 days after the beginning of the pandemic , democratic countries’ COVID 19death rate is on average larger than that of non democratic countries by approxi-mately 3.9 per 100,000. That is, 50 days after the beginning of the pandemic in ademocratic country, the fatality rate in a democracy is on average 7.3 times largerthan in an autocracy.A debate (Ang, 2020) has emerged trying to unpack the reasons behind suchwide differences across political regimes: a priori, all other things equal , the polit-ical regime should not be related to the spread of a disease. We distinguish threemain hypothesis to explain this difference.A first interpretation relates to the relative efficiency of social distancing mea-sures in democracies and autocracies. Some have argued that democracies may beless well equipped to implement and enforce social distancing policies (Cepaluniet al., 2020; Sorci et al., 2020), or that they may be implementing them with a suboptimal timing (Sebhatu et al., 2020). That is, in this view, autocracies are moreable to implement social distancing measures. We will refer to this interpretationas the efficient autocracy hypothesis.A second interpretation is that there may be voluntarily misreporting of COVID Defined as when the number of cases reaches 0.4 per 100,000 in a country.
29 data, in particular by non democratic countries. For example, Tuite et al.(2020a) report that Egypt may have underreported its number of cases, Tuiteet al. (2020b) report that Iran may also have underreported its number of cases,while Kavanagh (2020) discusses that China’s political regime may have hinderedits initial response to the pandemic. In this view, there are systematic differencesbetween the real and the reported death rate. These differences are voluntary andsystematically linked to the type of political regime. We will refer to this interpre-tation as the biasing autocracy hypothesis.A third interpretation has caught less attention (Ashraf, 2020): democraciesand autocracies tend to have systematically different characteristics apart fromtheir political regimes. These differences, once accounted for, may in fact be suf-ficient to explain the difference in both the real and reported death rate. Thiswould leave the contributions due to voluntary under-reporting or differences inpolicies to matter only marginally. An example of such differences would be thatautocracies tend to have much younger populations (and therefore, a much smaller real death rate, all other things equal) but also a lower ability to test (and there-fore, a much smaller reported death rate, all other things equal). We refer to thisinterpretation as the simply different autocracy hypothesis.The three aforementionned hypotheses are not mutually exclusive, and simplereduced form econometric approaches can help measuring how much each of themcontributes to explain the differences observed across political regimes. Take thecase where the econometrician only observes a reported death rate rather thanthe real death rate but can observe the variables determining COVID 19 real and reported death rate. Also assume that there are two such types of variables: fixed3haracteristics (say, the share of the population aged 65+ who would determinereal death rate or the number of hospital beds per capita who would determineboth real and reported death rates) and policy response. Under the efficient au-tocracy hypothesis, regressing the reported death rate on a measure of democracyand controlling for all fixed parameters would lead to a positive and significantcoefficient on democracy. However, further controlling for policy response in theregression should bring the coefficient on democracy close to zero and render itnon significant. That is, all the differences observed between democracies and au-tocracies in their reported death rate, once fixed characteristics are accounted for,would be due to the difference in policy response across these two types of regime.In this case, there may be a difference between the real and the reported deathrate, but this difference is not systematically linked to the political regime. In fact,these results would indicate that the policy response of autocracies is better thanthat of democracies, from the perspective of COVID 19 death rate.Under the biasing autocracy hypothesis, in a regression of reported death rateson a measure of democracy and all relevant controls (including policy response),the coefficient on democracy should be positive and significant. That is, despitecontrolling for all relevant characteristics and policy response, there still is a sys-tematic difference between democratic and non democratic countries which is notaccounted for. In that case, the only reason why a difference may remain would bedue to systematic underreporting of casualties by non democratic regimes. Thiswould be due to the fact that the difference between the real and the reported death rate is always larger for autocracies. While the real death rate would be We call fixed characteristics the variable that are pre determined and can not be changedin the time horizon of interest in the paper. In the long run, all characteristics determining thedeath rate such as, say, the GDP per capita, can of course be considered at least partly as anoutcome of the political regime (Acemoglu et al., 2019). Note that our methodology is neutral with respect to which political regime may be biasing reported death ratesremain different even when controlling for the characteristics influencing non vol-untary under reporting .We use daily level data of COVID 19 death rates of 137 countries for the first50 days of the epidemic and resort to simple reduced form econometric methodsusing the panel structure of the data. First we start by looking into the evolutionof daily total death rates across political regimes, using a regression with no con-trols except country fixed effects (Regression 1). We then include controls for fixedcharacteristics of countries that are likely to determine the real COVID 19 deathrate and allow them to matter differently across time (Regression 2). Finally, wealso include controls for the stringency of social distancing measures and allowthese to matter differently across time (Regression 3). Comparing Regression 2 toRegression 3 addresses the efficient autocracy hypothesis: any difference betweenthe coefficient on democratic regime between Regressions 2 and 3 would be due tothe differential in policy response across political regimes. An increase would in-dicate that autocracies implement more stringent social distancing measures thatare successful in decreasing the death rate. Comparing Regression 1 to Regression3 addresses the biasing autocracy and the simply different autocracy hypothesis:once all controls for both fixed characteristics and policy response are accountedfor, does the difference between autocratic and democratic regimes remains ( bias-ing autocracy hypothesis) or vanishes ( simply different autocracy hypothesis)?Our results indicate that the inclusion of controls for country characteristicsand policy response is in fact enough to remove almost all cross regime differencein COVID 19 mortality rates. In particular, the population susceptibility to die and allows for democratic regimes to be underreporting more than non democratic regimes.
In order to investigate our hypothesis, we assemble a dataset that comprises in-formation on daily cases and deaths in the first 50 days of the pandemic for 137countries. Our dependent variable, the daily country-level total number of re-ported cases and reported deaths due to the COVID-19 virus is from Dong et al.(2020) . Our main variable of interest, the classification of political regimes alongthe autocratic-democratic scale, comes from the Polity IV project (Center for Sys-temic Peace, 2015).Under the simply different autocracy hypothesis, accounting for the differencesin characteristics of countries would suffice to explain the difference in reportedmortality rates across political regimes. We therefore collected an extensive arrayof country level variables. To proxy for income and health infrastructure differ-ences, we gathered data on gross domestic product per capita in 2018 from theWorld economic outlook survey (IMF), and completed it with the World Factbook(CIA). Furthermore, information on the number of available hospital beds (perthousand inhabitants) is retrieved from the World Bank, to account for differences Last accessed: 16.10.2020
6n health infrastructure that may drive the mortality difference (actual and re-ported death rates).To capture differences in demographic characteristics which may explain thespeed of the spread of the disease, we use data on countries’ total population anddensity in 2019 from the World Bank, and data on countries’ urbanization ratein 2019 from the World in Data website. To control for the effect of geographicalcharacteristics, we collect data on the latitude and longitude of each country’scapital from the World Cities Database, and classify each country according to itsWorld Bank region. Finally, to control for population risk of mortality, we includethe share of population aged 65+ (from the World Bank) and, since air pollutionhas been shown to be associated with COVID 19 death rates (Zhu et al., 2020),we use summary exposure values to ambient ozone pollution and ambient particlematter pollution from the Global Burden Disease dataset (2017).To test the efficient autocracy hypothesis, we use information on countries’different COVID 19 containment policies from the “Variation in Government Re-sponses to COVID-19” dataset (Hale et al., 2020). This dataset includes a dailypolicy stringency index based on the aggregation of 17 policy indicators .Given that the data on our dependent variable is at the daily level, this allowsus to construct a panel dataset that comprises a total of 137 countries , classifiedas either democratic or non democratic, for which we have information on all the These are: East Asia and Pacific, Europe and Central Asia, Latin America and Caribbean,Middle East and North Africa, North America, South Asia and Sub-Saharan Africa. Policy indicators include policies with respect to closures or movement restrictions as wellas economic and health system policies. Last accessed: 16.10.2020. See Appendix A.2 for the list of the countries present in our dataset and their classificationas democratic or non democratic. Given the country-day panel structure of our data, we resort to fixed effect panelreduced form econometric methods to look into the differences in COVID 19 casu-alty rates across political regimes and time. This method allows us to remove theinfluence of all time invariant differences across countries by including countriesfixed effects. This further allows us to control for an extensive set of countries’pre-determined characteristics and for differences in containment policies acrosscountries.We specify the following regression equation, which we run using OrdinaryLeast Squares: Because our outcome of interest is death per capita, it makes sense to use cases per capitarather than the absolute number of cases to determine the beginning of the pandemic. InAppendix A.4, we show that results are robust to using alternatives thresholds. In Appendix A.3,we show that the timing of the beginning of the pandemic does not seem to differ significantlyacross political regimes. eathRate ct = (cid:80) Tt =1 β t ∗ democratic c ∗ time.f rom.start t + (cid:80) Tt =1 α t ∗ time.f rom.start t + (cid:80) Tt =1 δ t ∗ X c ∗ time.f rom.start t + (cid:80) Tt =1 γ t ∗ Y ct ∗ time.f rom.start t + δ c + ω ct + (cid:15) ct (1) DeathRate ct is the log of daily declared total deaths per 100,000 inhabitantsin country c plus one, t days after the beginning of the pandemic in country c. democratic c is a dummy indicating that the Polity IV score of country c is positive. time.f rom.start t is a set of fixed effect for each day since the beginning of the pan-demic. The interaction of democratic c with time.f rom.start t allows us to trackday by day the evolution of the difference in death rates across political regimes, astandard approach in economics (see Duflo (2001) or Cassan (2019) among others). X c is a large set of controls for countries’ pre determined characteristics: GDPper capita, number of hospital beds per 1000, population, density, urbanizationrate, share of population aged 65+, summary exposure value to particle matterspollution, summary exposure value to ambient ozone pollution, as well as for thesquare of these variables, World Bank regions fixed effects, latitude and longitude.We interact all these variables with the time.f rom.start t fixed effects to allowtheir effect to vary over time. Y ct is a measure of country policy response to the pandemic. It is a stringencyindex of governmental response (as measured at t-15 to allow for lags in its effect)We also include the square level of this variable to allow for non-linear effects.Furthermore, we interact these variables with the set of time.f rom.start t fixedeffects, to control for their time varying effect. δ c is a set of country fixed effects.9inally, ω ct is a set of day of the week fixed effect interacted with time.f rom.start t fixed effects, to control for variations in reporting across days of the week.We perform this regression iteratively. First, we do not implement any of the X c and Y ct controls (Regression 1). This allows us to see the evolution of thedifference in casualty rates across political regimes when no confounding factorsare accounted for. Then, we implement X c but not Y ct (Regression 2). This willallow us to see how much of the difference across political regimes survives oncethe different pre-determined characteristics of countries are accounted for. Finally,we add the Y ct policy response controls (Regression 3).This iterative procedure allows us to address the different sides of the debateon the role of political regime in fighting COVID 19. Comparing Regression 2 toRegression 3 addresses the efficient autocracy hypothesis: any difference betweenthe β t coefficients on democratic regime between Regressions 2 and 3 would be dueto the differential in policy response across political regimes. An increase would in-dicate that autocracies implement more stringent social distancing measures thatare successful in decreasing the COVID 19 death rate.Comparing Regression 1 to Regression 3 addresses the biasing autocracy and simply different autocracy hypothesis: once all controls for both fixed characteris-tics and policy response are accounted for, do the β t coefficients remain positive( biasing autocracy ) or do they equalize to zero ( simply different autocracy )? Notethat these hypotheses are not mutually exclusive: autocracies may well be effi-cient, biasing and different at the same time. Our methodology allows to capturethis possibility: if the β t coefficients decrease but remain large and significantwhen passing from Regression 1 to Regression 3 and change but remain large and10ignificant between Regression 2 and Regression 3, then this would support thesimultaneous presence of the three hypotheses. Figure 2 presents the β t coefficients from Equation 1 for all three versions of thespecification. The first panel presents the results of Regression 1, when no con-trols excepting country and day of the week fixed effects are included. The deathrates start diverging across political regimes roughly 10 days after the beginningof the pandemic. After 50 days, the β t coefficient reaches 0.5, which represents135% of the mean.The second panel includes controls for pre-determined characteristics inter-acted with day fixed effects. The β t coefficients become precisely estimated zeros.That is, once countries’ differences in characteristics are taken into account, thedifference in death rates across political regimes does not survive. Hence, we donot find support for the biasing autocracy hypothesis with our methodology.The third panel adds controls for countries’ policy response to the pandemic.As a result, our coefficients of interest β t remain virtually unaffected. That is, ourresults do not support the efficient autocracy hypothesis. Therefore, once sys-tematic differences across countries’ characteristics and policy responses are takeninto consideration, the differences in death rates apparent in Figure 1 and in thefirst panel of Figure 2 vanish. The reason why reported COVID 19 death rates In Appendix A.4, we show that results are robust to using alternatives definition of the startof the pandemic in a country. To test if these results are driven by an outlier country, we run Regression 1, 2 and 3 137times, removing one country at a time. We plot the 50 coefficients of interests of each of these411 regressions in Figure 6 of Appendix A.5. It can be seen that the results are robust to theomission of any single country. simply different autocracy hypothesis is prevailing. That is, our results do notsupport neither the fact that autocracies are more efficient at controlling the pan-demic nor that they are voluntarily under reporting casualty more often.
Having seen that the inclusion of controls is sufficient to remove the “politicalregime” effect on COVID 19 death rate, we now move to a related question: whichcharacteristics are contributing to closing the COVID 19 death rate gap betweenautocratic and democratic gap? In order to do so, we group our control variablesin five categories:- Geographical controls (latitude, longitude, World Bank region fixed effects)- Wealth controls (GDP per capita, hospital beds per capita). These will proxyfor the quality of the health system in the country, and will likely influence boththe real and the reported death rate.- Demographic controls (population, density, urbanization rate): these are likelyto influence the speed of the spread of the pandemic.- Population fragility controls (share of population aged 65+ and exposure to pol-lution): these are likely to influence the lethality of COVID 19 for a given spreadof the disease.- Policy response.We run Regression 1, removing each of these groups of controls one at a time.Figure 3 presents the β t coefficients for each of these regressions. It also includesthe coefficients of the original results of Regression 1 for comparison (we do not12eport the confidence intervals on these coefficients for readability).Two set of controls seem to matter most. The first set of controls are the con-trols related to population fragility (which contain notably the share of populationaged 65+). The second set of controls that seems to affect the coefficient is thegeographic controls set. That is, the main reason why mortality rates from COVIDseem to differ across political regimes may be that autocratic regime tend to havea population that is less susceptible to die from COVID 19 and to be located inregions in which COVID 19 appears to be less lethal. Interpreting the geographiccontrols is not straightforward, but it is reasonable to say that they seem to indi-cate that country characteristics correlated with geography and not captured byour extensive set of controls may play an important part in explaining COVID 19mortality. A few remarks are in order to help interpret our results. First, one should keepin mind that the variables that we consider pre-determined characteristics, suchas the GDP per capita are only pre-determined in the time horizon that we areconsidering. Over the long run, they are an outcome of the political regime. Seefor example Acemoglu et al. (2019), who show that democracy causes growth. Inthat sense, our results do not take into consideration the long term effect of polit-ical regimes on the variables that may determine COVID 19 death rates.For instance, a better health care system will lead to both a lower real deathrate (infected individuals are better treated) and a higher reported death rate (in-fected individuals’ death is better attributed to COVID 19). If, as has been argued13n the literature (Franco et al., 2004; Besley and Kudamatsu, 2006; Kudamatsu,2012; Bollyky et al., 2019; Pieters et al., 2016), democracies tend to have betterhealth care policies; in the long run, the health care system (which we consider aspre-determined) will be better in democracies because of the political regime, whichwill causally affect both real and reported death rates across political regimes.Second, our focus is only on COVID 19 death rates. Arguably, however, onemay have wanted to study death rates from all causes rather than just from COVID19. Even in times of pandemic, governments should aim at preserving the healthof their citizens from all sources of harm, not from one specific cause. Given theattention given to COVID 19 death rates, a pro-democracy argument would bethat while there does not seem to be differences across political regimes for COVID19 death rates, this may hide the fact that autocracies have focused on decreasingCOVID 19 death rate at the expense of death from other sources.One could develop this idea even further and argue that democracies havehigher COVID 19 mortality rates because they are better at preventing non COVID19 deaths, leading to a population which is on average older and therefore morelikely to die if infected by COVID 19. This question can unfortunately not betackled with the available data, and we leave it to future research (when mortalitydata from all causes will be available for a sufficient number of countries), but notethat the differences in countries’ population’s susceptibility to die from COVID 19upon contamination seem to be one of the main drivers of the difference in COVID19 mortality rates across political regimes, all other things equal.Third, our findings do not contradict previous studies on under reporting ofCOVID 19 data, in particular country specific studies. Indeed, because of the sta-14istical analysis used, our results do not imply that no single country underreportedor manipulated its COVID 19 mortality data. However, our results do address thewidespread idea that autocracies are systematically and willingly under reportingCOVID 19 casualties. What our results do indicate is that under reporting (byany political regime) is primarily due to the different characteristics of countriesthat are correlated with the political regime rather than a direct causal effect ofthe political regime.That is, a plausible interpretation is that autocratic governments may well un-der report data while not manipulating it. One could argue that even if autocraciesare under reporting COVID 19 death rates, this may be primarily driven by theiroverall incapacity to link death to its cause rather than to a direct attempt at datamanipulation. The low reported COVID 19 death rate in autocracies may in partbe due to the lower level of development of both the public health infrastructureand the statistical apparatus of autocracies. However, and this goes back to ourfirst point, over the long run, public health infrastructure and statistical apparatusmay well be determined by the political regime.
In this paper, we investigated the COVID 19 death rate gap between democraticand autocratic countries. We formulated three main hypotheses based on the pre-vious literature: the gap can be due to the fact that autocracies are more efficientat implementing restricting policy measures; that autocracies are underreportingtheir COVID 19 data and that autocracies simply have different characteristicsthat can explain the death rate gap. Our analysis, relying on simple econometric15ools, allows to make progress in the debate around the sources of the observeddifferences in COVID 19 death rates across political regimes.We show that once pre-determined characteristics and policy responses aretaken into account, COVID 19 death rates do not exhibit any difference acrosspolitical regimes: the coefficients on democracy become precisely estimated zeros.Our results therefore do not show support neither for the efficient autocracy norfor the biasing autocracy hypotheses, as we do not find evidence that autocraciesare neither systematically better at preventing COVID 19 death nor that they aremore often under reporting casualties.Our findings indicate that democracies and autocracies are simply different ,and that these differences are sufficient to account for all the observed differencesin COVID 19 death rates. Finally, we analyzed which controls matter most inexplaining the difference in COVID 19 death rates. We found that characteristicsrelated to the vulnerability of the population to the disease and geographical con-trols appeared to be of significant importance.
References
Acemoglu, D., Naidu, S., Restrepo, P., and Robinson, J. A. (2019). Democracydoes cause growth.
Journal of Political Economy , 127(1):47–100.Ang, Y. Y. (2020). When covid-19 meets centralized, personalized power.
NatureHuman Behavior , 4:445–447.Ashraf, B. N. (2020). Devastation caused by covid-19: Is democracy to blame?
Working Paper . 16esley, T. and Kudamatsu, M. (2006). Health and democracy.
American EconomicReview , 96(2):313–318.Bollyky, T. J., Templin, T., Cohen, M., Schoder, D., Dieleman, J. L., and Wigley,S. (2019). The relationships between democratic experience, adult health, andcause-specific mortality in 170 countries between 1980 and 2016: an observa-tional analysis.
The Lancet , 393(10181):1628 – 1640.Cassan, G. (2019). Affirmative action, education and gender: Evidence from India.
Journal of Development Economics , 136(C):51–70.Center for Systemic Peace (2015). Polity IV Project.Cepaluni, G., Dorsch, M. T., and Branyiczki, R. (2020). Political regimes anddeaths in the early stages of the covid-19 pandemic.
Working Paper .Dong, E., Du, H., and Gardner, L. (2020). An interactive web-based dashboardto track covid-19 in real time.
The Lancet Infectious Diseases , 20(5):533 – 534.Duflo, E. (2001). Schooling and labor market consequences of school constructionin indonesia: Evidence from an unusual policy experiment.
American EconomicReview , 4(91):795–813.Franco, Á., Álvarez-Dardet, C., and Ruiz, M. T. (2004). Effect of democracy onhealth: ecological study.
BMJ , 329(7480):1421–1423.Hale, T., Webster, S., Petherick, A., Phillips, T., and Kira, B. (2020). Oxfordcovid-19 government response tracker.Kavanagh, M. M. (2020). Authoritarianism, outbreaks, and information politics.
The Lancet Public Health , 5(3):e135 – e136.17udamatsu, M. (2012). Has Democratization Reduced Infant Mortality in Sub-Saharan Africa? Evidence from Micro Data.
Journal of the European EconomicAssociation , 10(6):1294–1317.Pieters, H., Curzi, D., Olper, A., and Swinnen, J. (2016). Effect of democraticreforms on child mortality: a synthetic control analysis.
The Lancet GlobalHealth , 4(9):e627 – e632.Sebhatu, A., Wennberg, K., Arora-Jonsson, S., and Lindberg, S. I. (2020). Ex-plaining the homogeneous diffusion of covid-19 nonpharmaceutical interventionsacross heterogeneous countries.
Proceedings of the National Academy of Sci-ences , 117(35):21201–21208.Sorci, G., Faivre, B., and Morand, S. (2020). Why does covid-19 case fatality ratevary among countries?
SSRN .Tuite, A. R., Ng, V., Rees, E., Fisman, D., Wilder-Smith, A., Khan, K., andBogoch, I. I. (2020a). Estimation of the covid-19 burden in egypt throughexported case detection.
The Lancet Infectious Diseases , 20(8):894.Tuite, A. R., Sherbo, R., Watts, A., Fisman, D., and Khan, K. (2020b). Estimationof coronavirus disease 2019 (covid-19) burden and potential for internationaldissemination of infection from iran.
Annals of Internal Medicine , 10(172):699–701.Zhu, Y., Xie, J., Huang, F., and Cao, L. (2020). Association between short-termexposure to air pollution and covid-19 infection: Evidence from china.
Scienceof The Total Environment , 727:138704.18
Appendix
A.1 Data sources and descriptive statistics
Table 1 presents the data sources used to compute the variables exploited in ouranalysis.Table 2 presents the descriptive statistics for our variable. Our final datasetcomprises a total of 137 countries for which we have all the variables and for whichwe observe 50 days of data since the beginning of the pandemic.
A.2 Polity score, by country
Table 3 presents the polity IV score of all countries that are included in our sample.Countries whose score is higher than 0 are classified as democratic.
A.3 Political regime and start of the pandemic
Our analysis focuses on the evolution of the death rates across time since thebeginning of the pandemic in each country. We verify whether the political regimedetermines when these first contaminations are reached, which may imply thatthe timing that we rely on is biased. In order to do so, we run the following OLSregressions: time . to . start c = α + β ∗ democratic c + X c + (cid:15) c (2) time . to . start c is the number of days between the start of the pandemic incountry c and the th of January. We specify three variations of Equation 2:if the country has declared 4 or 6 per 100,000 cases or has reached 100 cases.Table 4 presents the results. We find no statistically significant difference across19olitical regimes for the start of the pandemic. Note that the number of countriesreaching both 100 total cases and for which we observe 50 days since the first 100case is 135: two countries do not reach 50 days after the first 100 cases in our datacompared to our main sample. A.4 Alternative definitions of the start of the pandemic
Figures 4 and 5 reproduce Figure 2, using 0.6 cases per 100,000 and 100 casesreported cases as the definition of the start of the pandemic in a country, respec-tively. Results remain unaffected by this change in definition of the beginning ofthe pandemic. Note that the number of countries reaching both 100 total casesand for which we observe 50 days since the first 100 case is 135: two countriesdo not reach 50 days after the first 100 cases in our data compared to our mainsample.
A.5 Robustness check: removing one country at a time
In order to test if an outlier country is driving our findings, we run each regression133 times, removing one country at a time. We plot the 50 coefficients of interestsof each of these 399 regressions in Figure 6, which replicates Figure 2. It can beseen that the results are robust to the omission of any single country.20 a b l e : D a t a s o u r ce s . D a t a S o u r c e C O V I D D e a t h R a t e D o n g e t a l. ( ) C O V I D C a s e s D o n g e t a l. ( ) D e m o c r a t i c P o li t y I V p r o j ec t( C e n t e r f o r S y s t e m i c P e a ce , ) S tr i n g e n c y I nd e x o f P o li c y R e s p o n s e V a r i a t i o n i n G o v e r n m e n t R e s p o n s e s t o C O V I D - ( H a l ee t a l., ) . G r o ss D o m e s t i c P r o du c t p e r c a p i t a W o r l d E c o n o m i c O u t l oo k , I M F ( ) + W o r l d F a c t b oo k , C I A ( ) Sh a r e o f + W o r l d B a n k ( ) P o pu l a t i o n D e n s i t y W o r l d B a n k ( ) P o pu l a t i o n W o r l d B a n k ( ) U r b a n i z a t i o n R a t e W o r l d i n D a t a H o s p i t a l B e d s p e r W o r l d B a n k Su mm a r y E x p o s u r e V a l u e t o A i r P o ll u t i o n G l o b a l B u r d e n o f D i s e a s e ( ) Su mm a r y E x p o s u r e V a l u e t o A m b i a n t O z o n e P o ll u t i o n G l o b a l B u r d e n o f D i s e a s e ( ) L a t i t ud e a nd L o n g i t ud e W o r l d C i t i e s D a t a b a s e W o r l d R e g i o n s W o r l d B a n k a b l e : D e s c r i p t i v e s t a t i s t i c s M e a n s d M i n M a x D e m o c r a t i c . . . . L og t o t a l d e a t h s p e r , . . . . Sh a r e o f + . . . . G D PP e r C a p i t a23 , , , P o pu l a t i o n i n m illi o n . . . , . H o s p i t a l b e d s p e r . . . . P o pu l a t i o n D e n s i t y . . . , . U r b a n i z a t i o n r a t e Su mm a r y e x p o s u r e v a l u e t oa m b i e n t o z o n e p o ll u t i o n - A g e s t a nd a r d i ze d Su mm a r y e x p o s u r e v a l u e t oa m b i e n t p a rt i c u l a t e m a tt e r p o ll u t i o n - A g e s t a nd a r d i ze S tr i n g e n c y t - O b s e r v a t i o n s , able 3: Polity Score, by country Country Polity Score Country Polity ScoreAfghanistan -1 Kuwait -7Albania 9 Kyrgyz Republic 8Algeria 2 Latvia 8Argentina 9 Lebanon 6Australia 10 Liberia 7Austria 10 Libya -7Azerbaijan -7 Lithuania 10Bahrain -10 Luxembourg 10Bangladesh -6 Madagascar 6Belarus -7 Malawi 6Belgium 8 Malaysia 7Benin 7 Mali 5Bhutan 7 Mauritius 10Bolivia 7 Mexico 8Botswana 8 Moldova 9Brazil 8 Mongolia 10Bulgaria 9 Morocco -4Burkina Faso 6 Mozambique 5Burundi -1 Myanmar 8Cabo Verde 10 Nepal 7Cambodia -4 Netherlands 10Cameroon -4 New Zealand 10Canada 10 Nicaragua 6Central African Republic 6 Niger 5Chile 10 Nigeria 7China -7 Norway 10Colombia 7 Oman -8Costa Rica 10 Pakistan 7Croatia 9 Panama 9Cuba -5 Paraguay 9Cyprus 10 Peru 9Czech Republic 9 Philippines 8Denmark 10 Poland 10Djibouti 3 Portugal 10Dominican Republic 7 Qatar -10Ecuador 5 Romania 9Egypt, Arab Rep. -4 Russian Federation 423ountry Polity Score Country Polity ScoreEl Salvador 8 Saudi Arabia -10Estonia 9 Singapore -2Eswatini -9 Slovakia 10Ethiopia 1 Slovenia 10Fiji 2 Spain 10Finland 10 Sri Lanka 6France 9 Sudan -4Gabon 3 Suriname 5Gambia, The 4 Sweden 10Georgia 7 Switzerland 10Germany 10 Syrian Arab Republic -9Ghana 8 Tajikistan -3Greece 10 Tanzania 3Guatemala 8 Thailand -3Guinea 4 Timor-Leste 8Guyana 7 Togo -2Haiti 5 Trinidad and Tobago 10Honduras 7 Tunisia 7Hungary 10 Turkey -4India 9 Uganda -1Indonesia 9 Ukraine 4Iran -7 United Arab Emirates -8Iraq 6 United Kingdom 8Ireland 10 United States 8Israel 6 Uruguay 10Italy 10 Uzbekistan -9Jamaica 9 Venezuela -3Japan 10 Vietnam -7Jordan -3 Yemen, Rep. 3Kazakhstan -6 Zambia 6Kenya 9 Zimbabwe 4Korea, Rep. 8 24 able 4: Time to first cases and political regime
Heteroskedasticity-robust standard errors in parentheses * p < .10 ** p < .05 *** p < .01.Controls included are: GDP per capita, population, density, urbanization rate, share of 65and above, number of hospital beds per capita and the square of all preceding variables,latitude, longitude, World Bank region fixed effect. igure 1: Evolution of COVID 19 data reporting by political regime, time since first 0.4 casesper 100,000Total deaths per 100,000. Country level average per type of regime. Democratic countries aredefined as a country with a polity score > igure 2: Evolution of COVID 19 log death per 100,000 since 0.4 cases per 100,000. 95% CI.No Controls Controlling for pre-determined characteris-ticsControlling for pre-determined characteris-tics and policy responseStandard errors are two way clustered at the country and day level. igure 3: Coefficient on “Democracy” and introduction of controlsNo Population fragility controls No Geographic controlsNo Economic controls No Demographic controlsNo Policy Response controlsStandard errors are two way clustered at the country and day level. igure 4: Evolution of COVID 19 log death per 100,000 since 0.6 cases per 100,000. 95% CI.No Controls Controlling for pre-determined characteris-ticsControlling for pre-determined characteris-tics and policy responseStandard errors are two way clustered at the country and day level. igure 5: Evolution of COVID 19 log death per 100,000 since 100 cases. 95% CI.No Controls Controlling for pre-determined characteris-ticsControlling for pre-determined characteris-tics and policy responseStandard errors are two way clustered at the country and day level. igure 6: Removing countries one by one: Evolution of COVID 19 log death per 100,000 since0.4 cases per 100,000.No Controls Controlling for pre-determined characteris-ticsControlling for pre-determined characteris-tics and policy response CI not reported for readabilityigure 6: Removing countries one by one: Evolution of COVID 19 log death per 100,000 since0.4 cases per 100,000.No Controls Controlling for pre-determined characteris-ticsControlling for pre-determined characteris-tics and policy response CI not reported for readability