Political Regime and Suspected COVID-19 Death Data Manipulation

The COVID-19 pandemic—the worst one since the Spanish flu—has dramatically changed the world, with a great number of people still suffering and dying from the disease. Some scholars argue that the pandemic has severely damaged democratic countries, mainly because these cannot intervene in their citizens’ lives, as opposed to their authoritarian counterparts. Another study challenges this view and suggests that authoritarian countries manipulate data on COVID-19-related deaths. This paper aims to determine which view is more persuasive using cross-national data. This article uses statistical evidence to reveal that authoritarian countries are likely to manipulate the data. The result implies that, with this successful manipulation, authoritarian states can strengthen citizens’ support for their governments through the COVID-19 pandemic. * I wish to thank Masataka Harada (Fukuoka University), Masaaki Higashijima (Tohoku University), Kentaro Hirose (Waseda University), and Masaru Kohno (Waseda University) for their helpful comments. 2 Introduction The number of COVID-19 deaths is reported to exceed one million all over the world. Some argue that people, especially in democratic countries, face a tradeoff between freedom and health (Harari 2020; Koyama 2020). Recently published papers also reveal that democratic countries suffer from COVID-19 deaths more than authoritarian states (Cepaluni et al. 2020; Cheibub et al. 2020). Graph 1 supports these arguments to some extent. It shows the number of COVID-19 deaths on the vertical axis as reported by Johns Hopkins University and the level of Polity2 on the horizontal axis from the Polity V Project. 1 The number of deaths is the total sum since the first confirmed cases were reported (as of July 31, 2020). The correlation coefficient between the two variables is 0.2680, and it is statistically significant at the 1% level. This moderate positive relationship suggests that the arguments should be correct. However, is this relationship truthful? This article attempts to answer this question. 1 The Center for Systems Science and Engineering (CSSE) at Johns Hopkins Un iversity (2020) COVID 19 Data Repository, https://github.com/CSSEGISandData/ COVID19, accessed on August 8, 2020, Marshall, M. G., Jaggers, K., and Gurr, T. R. (2020) Polity V Project, Political Regime Characteristics and Transitions, 1800-2018. Center for Systemic Peace, http://www.systemicpeace.org/inscrdata.ht ml, accessed on August 8, 2020.


Introduction
The number of COVID-19 deaths is reported to exceed one million all over the world.
Some argue that people, especially in democratic countries, face a tradeoff between freedom and health (Harari 2020;Koyama 2020). Recently published papers also reveal that democratic countries suffer from COVID-19 deaths more than authoritarian states (Cepaluni et al. 2020;Cheibub et al. 2020). Graph 1 supports these arguments to some extent. It shows the number of COVID-19 deaths on the vertical axis as reported by Johns Hopkins University and the level of Polity2 on the horizontal axis from the Polity V Project. １ The number of deaths is the total sum since the first confirmed cases were reported (as of July 31, 2020). The correlation coefficient between the two variables is 0.2680, and it is statistically significant at the 1% level. This moderate positive relationship suggests that the arguments should be correct. However, is this relationship truthful? This article attempts to answer this question. Why can these arguments be questioned? For example, it is reported that Alexander Lukashenko, the Belarusian President, underestimated the risk for  spreading across the country. The president did not take any appropriate measures to prevent the pandemic in the country. As this case implies, authoritarian governments do not necessarily take a decisive measure immediately. However, the country has one of the lowest death rates in Europe (Karáth 2020). This is incredible. If the COVID-19 confirmed cases in authoritarian countries are not reduced, the argument that the government's intervention can reduce COVID-19 deaths by reducing confirmed cases is not persuasive.
In another case, Cepaluni et al. (2020) show a statistical analysis in which political regime is positively correlated with COVID-19 deaths. Their analysis includes confirmed cases as a control variable, which is also positively correlated with COVID-19 deaths, as expected. As it is puzzling that the political regime variable is statistically significant even after controlling for the confirmed cases, this leads to the question above. It is often argued that, if authoritarian governments can reduce death cases by stringent intervention, the total confirmed cases must be reduced first; however, the statistical analysis shows a significant effect of political regime on deaths, even after controlling for the confirmed cases. This result implies that a factor other than the confirmed cases significantly affects COVID-19 deaths in authoritarian countries.
It is difficult to imagine that the medical system in authoritarian states can work better than that in democratic countries. Scholars have highlighted that people in democratic countries are likely to have better health than their authoritarian counterparts (Wang et al. 2020;Gerring et al. 2020). Another possibility is that authoritarian countries manipulate death data. Kapoor et al. (2020) analyze the data's moving average and reveal that the data are unnaturally produced. This paper aims to determine which view is more persuasive using cross-national data.

Determinants of Confirmed Cases and Non-Pharmaceutical Interventions
First, this study considers the determinants of confirmed cases and non-pharmaceutical interventions (NPIs)-such as lockdowns-by utilizing data from the Johns Hopkins University and the University of Oxford (Hale et al. 2020)   the same results as model 1. These results suggest that the political regime variables do not affect the COVID-19 confirmed cases.

Determinants of COVID-19 Deaths and Data Manipulation
If authoritarian governments do not make stricter interventions in their citizens' daily lives to combat COVID-19 compared to democratic governments, why do authoritarian ４ This analysis does not consider how swiftly the government responds, which may make a difference (Cepaluni et al. 2020;Cheibub et al. 2020).
(3) The dependent variable is the number of COVID-19 deaths. The control variables are almost the same as the previous analysis above, except for an additional variable-the number of physicians per 1000 population from WDI data. This analysis uses negative binomial regression. Models 5 and 6 in Table 3 analyze the relationship between the political regime and COVID-19 deaths. Model 5 shows that Polity2 is positively correlated with COVID-19 deaths, and the coefficient is statistically significant even after controlling for the number of confirmed cases. This result suggests that the more democratic a country is, the more deaths it suffers. Model 6 shows the result from Polyarchy and almost the same as Model 5. These results are consistent with those of Cepaluni et al. (2020).
If authoritarian countries cannot reduce COVID-19 deaths by reducing the number of confirmed cases, why are fewer deaths registered in those countries? Is the medical system in authoritarian states better than that in democratic countries? A study suggests that Belarus, one of the authoritarian countries, has a large hospital capacity, leading to a lower death rates (Karáth 2020). That may be possible. However, the results of this study show the robust significance of the political regime variables on average even after controlling for the number of physicians. Moreover, People in democratic countries tend to have better health than those in authoritarian counterparts (Wang et al. 2019;Gerring et al. 2020), and it is peculiar that COVID-19 is an exception. data. The project regards the missing values in the WDI data as the government's unwillingness to disclose its country's internal affairs. This index can be a proxy for data transparency.  Table 4 shows the results of the analyses, including the Transparency Index as an additional control. Models 7 and 8 are the same as models 5 and 6, except for the Transparency Index. These models indicate that the political regime variables are no longer statistically significant with additional control; however, the Transparency Index is also not statistically significant. Cepaluni et al. (2020) also include this variable to were independent from the Soviet Union are excluded because of the first condition as they tend to be corrupt, which may systematically bias the results. Cepaluni et al. (2020) do not seem to recognize this potential problem. Then, the next analysis drops the observations in which the data on the Transparency Index are missing from models 5 and 6 to consider this possibility. Table 5 shows the results. Models 9 and 10 are the same as models 5 and 6, except for the number of observations. These results indicate that the political regime variables are not statistically significant at the conventional standard, although the Transparency Index is not controlled for. These models strongly suggest that the missing countries on the Transparency Index affect the results. ５ The next section examines the characteristics of these countries.
５ It is without a rational reason to exclude the countries in which the Transparency Index data are missing from the observations. On the other hand, the countries are systematically missing to some extent, according to Hollyer et al. (2014). In that sense, it is not totally unreasonable.

Extraordinarily Low Case Fatality Rate in Authoritarian Countries
The case fatality rate (CFR) is the proportion of deaths from a particular disease compared to the total number of confirmed cases. This indicator is often used to assess the severity of the disease and is affected by the test numbers. In this sense, it means that if the rate is low, it is usually thought to be not due to the intervention by a government to reduce the confirmed cases but by other factors such as the level of medical infrastructure, human resources, nutrition condition, and so on after the infection. Table 6 shows the difference in CFR among the no-missing and missing countries on the Transparency Index divided by the level of Polity2. ６ and CFR is clearly positive in the data missing countries on the Transparency Index, and the correlation coefficient is 0.5793. The same patterns can be seen in graphs 4 and 5. These graphs clearly reveal that the lower the level of Polity2, the lower the CFR, especially in the data-missing countries. Additional regression results also support these patterns. Table 7 shows the relationship between political regimes and COVID-19 deaths and CFR in the data-missing countries. The dependent variables in models 11 through 14 are the number of deaths. The political regime variables are positively correlated with the number of deaths at a statistically significant level. The dependent variables in models 15 and 18 are the CFR. These models also confirm a similar relationship, except for model 18. These results are tough to interpret. However, the variable of physicians is not statistically significant in all these models, and the confirmed cases continue to be significant in all the models in which the variable is included.
These results suggest that a factor other than the government's interventions or medical systems possibly affects the COVID-19 deaths.
In conclusion, it is strongly suspected that some authoritarian countries, especially in the data missing countries on the Transparency Index, manipulate data on COVID-19-related deaths. Kapoor et al. (2020) are more persuasive than other studies such as Cepaluni et al. (2020), which argue that authoritarian countries have been able to manage the COVID-19 problem more effectively.

Changing Trends
The analyses above are based on the data available on July 31, 2020. The situation seems to change now in October. Table 8 shows the results of the relationship between the political regime and COVID-19 deaths and CFR with the latest available data (October 10, 2020). ７ Models 19 through 22 reveals that both Polity2 and Polyarchy are no longer statistically significant. These results mean that the (superficial) advantage of authoritarian countries is now diminished.
The reason for this is unknown. Authoritarian leaders may no longer attempt to manipulate the data. Anyway, the argument that authoritarian countries have succeeded ７ The descriptive statistics are in Appendix 2.
in reducing COVID-19 deaths much better than democratic countries is no longer persuasive now. The tradeoff hypothesis between freedom and health is dubious. which view is more persuasive using cross-national data and reveals statistical evidence that authoritarian countries are likely to manipulate the data. This result implies that the tradeoff between freedom and health is superficial and misleading. It is probable that authoritarian countries only overstate their performance; nevertheless, authoritarian states can strengthen citizens' support for their governments through the COVID-19 pandemic by successfully manipulating the data.
This study has some limitations. For example, Cepaluni et al. (2020) and Cheibub et al. (2020) utilize daily data, which can make a more nuanced analysis possible to capture daily fluctuations in the prevalence of COVID-19 as well as the government's interventions. However, almost all other variables included in the analysis are yearly data, and it is not easy to determine which are more appropriate for analyzing the phenomena.
Another caution is that this study does not consider the effects of the number of tests. It may be possible that authoritarian countries conduct much more tests than democratic states, and that may affect the results. But only the 87 countries' test data are available. ８ The number of observations is too small, much fewer than the ８ Our World in Data, https://ourworldindata.org/coronavirus-testing, accessed on October 18, 2020.
Transparency Index data. This study regards the inclusion of the data into the analyses as meaningless in that aspect.