The Interplay of Policy, Institutions, and Culture in the Time of Covid-19

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


Policy Research Working Paper 9470
Leveraging data in 147 countries from the Johns Hopkins University, the World Bank, and other sources, we investigate how the COVID-19 pandemic spread and how the mortality rates are associated with preexisting vulnerabilities, the government's mobility restriction policy, institutions (democracy), and culture (i.e., trust and individualistic culture).While delay in domestic mobility restrictions is not significantly associated with pandemic outcomes, increasing delay in restricting international mobility is associated with higher pandemic mortality rate.Some pre-existing vulnerabilities are positively associated with the spread of the pandemic but not the mortality rate.However, when vulnerabilities are accompanied with delay in domestic mobility restrictions, the pandemic has higher mortality.Democracy is associated with lower policy delay in restricting mobility, lower pandemic mortality rate, and features better protection of the vulnerable population from pandemic harms.However, delay in domestic mobility restrictions have more adverse effects in democratic countries.Trust is not associated with worse pandemic outcomes, but its combinations with policy delay and obesity are.More individualistic countries do not feature different pandemic outcomes.However, in such countries, government delay in domestic mobility restrictions is associated with significantly higher pandemic mortality rate, and the marginal effects of the share of elderly on pandemic spread is greater.This paper is a product of the Development Research Group, Development Economics.It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world.Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp.The authors may be contacted at lxu1@worldbank.org.

I. Introduction
Is the spread of COVID-19 and the mortality it triggered related to different countries' government mobility restriction policies, democratic institutions, and cultural norms?We examine this important, timely, but underexplored question.Still ongoing, the COVID-19 pandemic has already ranked as one of the most adverse economic and human shocks in a century (Barro and Ursua 2020;Fukuyama 2020).Its consequences appear in every aspect of our lives.In the United States, for instance, aggregate consumer spending dropped precipitously by 31% in the first few months since its onset (Coibio et al. 2020).The number of active business owners in the U.S. plummeted by 22 percent from February to April 2020 (Fairlie 2020).Among all infectious disease outbreaks, this pandemic left the largest impact on the stock markets both in the U.S. and around the world (Baker et al. 2020;Ding et al. 2020a).In addition, the pandemic has had serious effects on income inequality, impacting especially the young and low-income earners (Adams-Prassl et al. 2020), Blacks and the Hispanics in the U.S. (Fairlie 2020;Benitez et al. 2020), employees in firms with high female employment shares and therefore women (Alon et al. 2020), and those employed in low-wage services (Bartik et al. 2020a, Chetty 2020).This pandemic could also have long-term adverse effects by affecting quality of schooling (via distance learning), beliefs about disastrous economic shocks in the future (Kozlowski et al. 2020), or confidence in political institutions and leaders, confidence in public health systems (Aksoy et al. 2020), or individuals' social trust (Aassve et al. 2020).
Yet countries differ greatly in terms of the pandemic's spread and mortality.In terms of the number of accumulated confirmed COVID-19 cases as measured in the John Hopkins University (JHU) COVID-19 data on November 19, 2020, several countries, such as Brazil, India, and the United States have had more than 5 million of cases.In contrast, 11 countries or regions have had fewer than 100 confirmed cases.For the 187 sample countries or regions in the JHU data, COVID-19 cases per million residents range from about 3 to 78,634, with the highest observed in Andorra, Bahrain, Belgium, Montenegro, and Qatar, all above 47,500.Meanwhile, Cambodia, Lao PDR, Tanzania, Qatar, and Vietnam have the lowest cases, all fewer than 19 per million residents.Countries around the world have also reported very different COVID-19 mortality rates (i.e., the number of COVID-19 deaths divided by the number of cumulative cases), ranging from 0 to 29 percent.What explains the vast differences in the pandemic outcomes?
The unprecedented challenges of the pandemic compel every country, no matter the economic and political systems, to fight it using all the means under its control.The direct means for these challenges is through government mobility restriction policies, which could deter the spread of the pandemic.A country can also try to contain the pandemic by drawing on its formal institutions such as voluntary compliance with social distancing based on the legitimacy of the government under democracy, or direct control under authoritarian systems to address these challenges.A country can further draw on its cultural norms, such as collective cultural traits and general trust among the citizens.
This ongoing pandemic allows us to examine the roles of the government, democratic institutions, and cultural norms in containing the pandemic.If we classify the tasks confronting a society into normal activities-such as growing an economy and provision of basic public goodsand handling unusual crises, dealing with the pandemic is clearly the latter.In dealing with these distinct tasks, the roles of the government, institutions and culture differ greatly.For one thing, in normal times the government could use its formidable bureaucratic machinery to manage many routine tasks.But in crisis, the bureaucratic machinery matters less because of the work stoppage, and the role of leadership would matter much more (Fukuyama 2020), as would institutions and culture.Indeed, recent research suggests that a large portion of social distancing was explained by private individual choices rather than government lockdown policies (for example, Goolsbee and Syverson, 2020).It is thus intriguing how the roles of the government, democratic institutions and culture matter for the pandemic outcomes.
In this paper, we combine the COVID-19 data by the Center for Systems Science and Engineering (CSSE) at JHU, downloaded on November 19, 2020, with cross-country data from the World Bank and other sources to examine the key factors that are correlated with the two indicators of pandemic severity: the pandemic spread (i.e., the number of confirmed COVID-19 cases per 1,000 residents), and the pandemic mortality rate (among the confirmed cases).The explanatory factors are multidimensional, including preexisting demographical and environmental vulnerabilities (i.e., the share of elderly population, obesity prevalence, urbanization, and air pollution), government policy responses (i.e., the number of days that the government initiated some form of closedown, both domestically and internationally), democracy, and culture (i.e., individualistic culture, and general trust).We investigate not only their separate associations with the pandemic outcomes but also their interactions.Allowing for their interactions is important.The drastically different pandemic outcomes are often not due to a specific factor; rather, they are likely due to a combination of several factors at the time of crisis, such as preexisting vulnerabilities combined with weak government responses, or with democratic institutions and cultural norms.
Otherwise, we should not observe such big differences in the pandemic outcomes, for instance, within the high-income-and-democratic countries sample, such as those between the United States, Germany, and South Korea.
Based on cross-country regressions and when appropriate, the spatial autoregressive models, we find that countries with demographic and environmental vulnerabilities tend to have a greater pandemic spread (as in the case of higher urbanization and air pollution), while the share of elderly is associated with lower pandemic spread.Government delay in restricting international mobility is associated with higher pandemic mortality rate.Increasing the number of days to impose international-travel closedown by one standard deviation (SD) is associated with an increase in pandemic mortality rate by 0.4 percentage points, or 17% of the mean mortality rate.
Furthermore, government delay in restricting domestic mobility is associated with higher pandemic mortality when the country has higher obesity prevalence and air pollution (i.e., SO2).
Increasing obesity by one standard deviation (SD) would imply a change in the marginal effects of DelayD on mortality (i.e., ) from -0.004 to 0.003.Similarly, increasing SO2 by one SD from the mean would imply a change in the marginal effect of DelayD on mortality from -0.008 to 0.003.
Contrary to the notion that authoritarian countries may have an advantage in combating the pandemic, 2 we find that democratic countries have done better in handling the pandemic.Relative to non-democratic countries, democratic countries have responded faster by having significantly lower delay in domestic closedown by about a month.They also have significantly lower pandemic mortality rate, by 0.9 percentage points, or 43 percent of the mean.They seem to be able to reduce the harm to the vulnerable population.In particular, the marginal effect of the share of elderly on pandemic spread is lower in democratic countries by about 100 percent; that of the share of obese on pandemic mortality rate is roughly zero in democratic countries, in contrast to 0.07 in nondemocratic countries.However, government policy delays appear to be more harmful in democracies.The marginal effect of the delay in domestic closedown on pandemic spread in democratic countries is 2.6 times higher than in non-democratic countries.
Contrary to what is portrayed in the media, we do not find that countries with stronger individualistic culture, on average, suffer worse pandemic outcomes.In countries of relatively high individualism, government delay in domestic mobility restrictions is associated with higher pandemic mortality rate, likely reflecting the negative spillover effects from individualists at the time of pandemic on the society; the marginal effects on pandemic spread of the share of elderly is many times greater, but when the harm is internalized such as in the case of SO2, the marginal adverse effect is much more muted.
Similarly, the level of trust in a society by itself is not associated with worse pandemic outcomes.However, when the government signals a lack of seriousness of the pandemic by policy delay, such delay has more pronounced positive association with pandemic spread in more trusting countries.Increasing general trust by one SD from its mean is associated with higher marginal effect of the delay in domestic mobility-restriction policy on pandemic spread, and the increase amounts to 19% of the marginal effect at the mean trust level.In high-trust (i.e., one SD above the mean trust) countries, the link between obesity and pandemic spread is more pronounced: at the mean level of trust, the marginal "effect" of obesity on pandemic spread is -2.99, and at the high level of trust, it is 0.03.
We add to the literature on the correlates of pandemic severity and containment.The existing studies have found that the pandemic spread and mortality rates are related to factors such as the share of elderly, the dependence on commuting, weather, population density, globalization, political ideology, and connectedness to the source country.It also finds other suspects not mattering such as the pollution level, obesity rates, ICU beds per capita, and poverty rates (Knittel and Ozaltun 2020;Desmet and Wacziarg 2020;Zimmermann et al. 2020).Relatedly, some have found that compliance with social distancing, a key determinant of the pandemic outcomes, is related to democracy, willingness to sacrifice for social objectives, collectivist cultural traits, community civic capital, trust, ideology and beliefs, gender, and government signals. 3We offer evidence that democratic countries on average handle better in the speed to act, in the reduction in pandemic mortality, and in protecting the vulnerable population.We add to the literature that it is often the interaction of multiple different aspects among the correlates that determine the pandemic outcomes, such as coupling of government policy delay with pandemic vulnerability, or coupling institutions, culture with pandemic vulnerabilities and with pandemic policy delay.For instance, pandemic outcomes are worse when coupling individualism with policy delay and with the share elderly, and coupling democracy with policy delay.

II.
Relating pandemic outcomes to vulnerabilities, democracy, and culture We now consider how the pandemic outcomes are related to country characteristics, preexisting vulnerabilities, democracy, and cultural norms.How would the income level be related to pandemic outcomes?Richer countries, trading more, have more cross-country and within-country human interactions.This could result in a greater COVID-19 spread.However, the effect on pandemic mortality should be ambiguous for our sample period.Though without any effective treatment for COVID-19-and vaccines are just beginning to be implemented around the worldhigher income countries have better coping mechanisms: they have, for instance, better access to personal protective equipment, more ventilators in emergency, better access to helpful supplements, better dwelling conditions, all of which would could reduce the mortality.We thus expect a positive correlation between the income level and the COVID-19 spread, but an ambiguous correlation with the pandemic mortality rate.
Countries differ greatly in their state capacity to implement and to enforce policies to contain the pandemic, capacities such as conducting quick testing, contact tracing, among others.
For the same de jure policy, some countries are better at enforcing it (Hallward-Driemeier and Pritchett 2015).With difference in state capacity, and in residents' readiness to comply with the government's containment policy, it is unclear whether the pandemic containment policy (i.e., mobility restrictions) would hinder the spread and reduce the mortality rate of COVID-19.It should be noted that the government has different capacity to enforce restrictions when such restrictions involve domestic or international mobility.Enforcing restrictions on international mobility, relative to those on domestic mobility, is much easier: just by limiting international flight and by locking the border, much are achieved; by contrast, monitoring domestic mobility of all residents is virtually impossible.We thus expect that international mobility restrictions are more effective and have stronger pandemic-deterring effects than domestic mobility restrictions.

Preexisting vulnerabilities
We expect a tendency to have greater COVID-19 damages in countries with greater preexisting vulnerabilities that are identified in the literature.First, once being infected of COVID-19, the elderly is found to be more vulnerable in terms of mortality rate (Liu et al. 2020; Knittle and Ozaltun 2020).In the meantime, the elderly tends to be less physically active and socialize less, which might contribute to less pandemic spread.Second, obesity is associated with greater vulnerability to COVID-19 in terms of mortality (Yang et al. 2020).Similar to the elderly, obese people are less active, and might have a tendency to reduce pandemic spread.Third, in more urbanized countries, a larger share of people lives in cities. Dense urban living conditions dramatically increase the chance of infections from daily encounters with other city dwellers, facilitating the spread and potentially increasing mortality from COVID-19.Finally, since air pollution harms people's immune systems and especially lung capacity, COVID-19 is likely to be more deadly where air pollution is worse (Fattorini and Regoli 2020).
In countries with stronger preexisting vulnerabilities, the pandemic outcomes should worsen with a delay of the government to restrict mobility--here we mean domestic mobility since international mobility restrictions affect much less people.With the government's delay to restrict mobility, social interactions are only limited by voluntary reduction in mobility, and COVID-19 could multiply faster and gain momentum to spread, especially in countries with higher vulnerabilities-just like wildfire in places with more dry woods.When hospitals are overwhelmed with capacity constraints facing rising number of COVID-19 cases, mortality rate would inevitably rise.

Democracy
Accountability institutions such as democracy, by themselves, should not have strong effects on pandemic spread.On the one hand, countries with such institutions are often thought to have more lengthy process and cumbersome procedures for decisive actions, which likely would hinder pandemic containment.Furthermore, such countries allow citizens to have more freedom and are less coercive, which again may increase the spread of the pandemic.On the other hand, legitimacy of the government in such countries likely implies better voluntary compliance with the social distancing recommendations.We thus expect an ambiguous relationship between the pandemic spread and democracy.Since democratic countries are on average more accountable to the voters and tend to deliver what voters want (Becker 1958, Stigler 1972, Wittman 1989, Acemoglu, Reed and Robinson 2014), they value citizen lives more highly, and would do better in trying to save lives from the pandemic, thus reducing the pandemic mortality rate.We thus expect democracy to be associated with lower pandemic mortality rate.
However, democratic countries are not destined to do well in any circumstances.Due to greater trust in government in democratic countries, the signals such as decisive actions or inactions that the government sends at the time of the pandemic may have stronger impacts on citizen behaviors than in non-democratic countries (Glaeser et al. 2020).Indeed, it is shown that public figures' views and communications significantly influence the general public's behavior toward the pandemic (Bursztyn et al. 2020).In democratic countries, a longer policy delay on pandemic containment would be interpreted as non-severity of the pandemic.Furthermore, a significant share of citizens with pre-disposed ideology have lower perception of pandemic risks, as in the case of Trump voters and Fox News viewers in the United States (Alcott et al., 2020;Barrios and Hochberg 2020;Desmet and Wacziarg 2020;Simonov et al. 2020).Some citizens, due to large losses associated with strict mobility restriction policies, are biased against believing the seriousness of the pandemic-like confirmation bias, they started with the belief/hope that lockdown would not be necessarily, and they then find government inaction as confirmation of their views.Government inaction and the share of biased citizens thus combine to fan the nonchalance toward the pandemic, worsening the pandemic outcomes.Furthermore, in countries with worse preexisting vulnerabilities to COVID-19, the non-compliance due to government inaction would have more adverse effects, worsening the adverse signaling effects on the pandemic outcomes.Thus, we expect that the negative effects of government policy delay on the outcomes would be especially severe in democratic countries, and in countries with worse preexisting COVID-19 vulnerability.

Culture
A country can also deal with the pandemic through its cultural norms.We first consider a key cultural norm: the individualistic culture (Hofstede, 1997).Countries with a strong individualistic tendency (relative to a collective tendency) are often viewed as likely faring worse in pandemic outcomes.Indeed, in individualistic-culture societies, it is more difficult to successfully organize collective actions (Gorodnichenko and Roland, 2015), which include a coordinated response to a pandemic.Furthermore, the anti-government-intervention attitudes in such countries are prevalent (Pitlik and Rode, 2017).At the time of the pandemic, individualistic citizens fully enjoy the benefits of socializing, but would not sufficiently internalize its costs of spreading the virus.Yet the cost of socializing is now especially high: the social cost of a COVID-19 infection such as the costs associated with potentially infecting others is shown to be likely more than three times higher than the private cost of own risk of infection (Bethune and Korinek 2020).Consistent with the importance of individualistic tendency during the pandemic time, residents in U.S. counties that demonstrate more willingness to sacrifice for social objectives are more likely to comply with social distancing measures (Ding et al. 2020).Similarly, with the same policy restrictiveness, countries with more collectivist cultural traits experienced larger declines in geographic mobility relative to their more individualistic counterparts (Frey et al. 2020).Citizens are found to voluntarily comply more with social distancing if individuals are more patient and have altruistic preference traits (Alfaro et al. 2020).Countries with individualistic culture thus could face heightened pandemic risks and worse pandemic outcomes.
Yet, individualistic culture also typically emphasizes personal responsibility and initiative.
This emphasis is often associated with higher economic growth and more innovations (Gorodnichenko andRoland, 2011, 2017), factors that may also push for ingenious ways to deal with the pandemic, such as by inventing faster testing.Less likely to produce "yes men," individualistic societies encourage the flourishing of diverse opinions, which may ultimately facilitate the emergence of the best solutions to the pandemic (Prendergast, 1993).These considerations may reverse the adverse effects of the individualistic culture on pandemic results.
We thus expect that the individualistic culture has an ambiguous relationship with the pandemic outcomes.
We expect, however, that when facing a longer delay in pandemic-containing policy, the more individualistic the countries are, the worse they fare in pandemic outcomes.A longer delay in restricting mobility would allow individualists, not internalizing the costs of the virus to others, more time to socialize and potentially spread the virus, and thus eventually cause more damage.
The effects of individualistic culture on pandemic outcomes, furthermore, likely depend on the demographic structure and other vulnerability factors.In general, countries with a higher share of elderly are likely to have lower pandemic spread, simply because the elderly do not group-socialize that much as the young.However, in countries with strong individualism orientations, the elderly retains more the socializing tendency of the general society, and would, relative to their young, reduce social interactions less than in collectivist societies.As a result, the negative impact of the share of the elderly on pandemic spread would be less pronounced in individualistic societies.In contrast, for vulnerabilities with harm that can be easily internalized, such as air pollution, individualistic people would be proactively stay away from the harm.As a result, air pollution's effect on social interactions and therefore pandemic spread would be lower in individualistic countries.
Another cultural norm that has implications for the pandemic outcomes is a society's level of general trust (Fukuyama, 1995).There are indications that in determining compliance with social distancing in this pandemic, individual choices and self-regulating behaviors play a crucial role.
In empirically explaining the compliance with social distancing in the U.S. counties, for instance, the reduction in mobility was mostly explained by individual choice rather than legal shutdown The literature on social trust indicates that the level of trust in a community is consequential.
Otherwise similar communities experience different levels of economic development, crime rates and health depending on the social bonds between its members (Borgonovi 2020, Fukuyama 1995, Knack 1997).Trust has two components: the social interaction component (i.e., trusting communities have more social organizations and social interactions) (Kawachi, Subramanian, and Kim 2008), and the norms-values component (i.e., trust entailing attitudes that facilitate collaboration).These two components have different implications for the pandemic.On the one hand, building and maintaining trust requires frequent social interactions, often via numerous civic organizations.While helpful to the society in normal times, frequent social interactions help spread the virus in the time of the pandemic.Compliance with social distancing, for instance, is found to be less for countries that traditionally socialize more (Ding et al. 2020b).The social interaction aspect of general trust thus likely leads to a positive relationship between trust and the pandemic spread.On the other hand, the norm component of trust could help deter the pandemic.Trust, part of civic capital, would facilitate a group to overcome the free-rider problem to accomplish socially valuable activities" (Guiso et al. 2011).In communities with higher trust, individuals value the community higher, and are found to be more likely to sacrifice themselves including complying with social distancing for the benefit of the community (Barrios et al. 2020;Borgonovi and Andrieu 2020).This community spirit would help contain the pandemic.Thus, we expect that the net association between general trust and pandemic severity is ambiguous.
A high-trust society tends to carry out government policies more effectively.A high density of civic organizations may help spread popular views and/or government policies and encourage conformity.In Sierra Leone, for instance, more social capital is associated with a greater respect of their chiefs (Acemoglu et al. 2014), indicating ease in implementing leader policies.Local social networks in German towns had aided the rise of the Nazi party (Satyanath et al. 2017).These papers suggest that a high level of trust may speed up pandemic spread under wrong signals from the government, such as when facing official signals undermining the severity of the pandemic.
When the government sends out signals that the pandemic risks are not that great-as in the case of a delay in restricting social mobility by the government-the strong trust in government may amplify this perception, and result in worse pandemic spread.We thus expect that government policy delay in restricting mobility results in greater damage on pandemic outcomes where there is more trust.
Relatedly, when a high trust and a high vulnerability are combined in a country, the greater social interactions in such societies facilitate the spread of the pandemic, resulting in more infections and/or higher mortality from the pandemic.We thus expect that high-trust countries experience higher pandemic damages when they also have a higher vulnerability.

III. Data and measurements
Our To capture preexisting vulnerabilities to the pandemic, we rely on several variables: the share of elderly in the population (Elderly) (from WDI),8 the share of urban residents in total population (Urban, from WDI), obesity prevalence (Obesity, from World Health Organization), and atmospheric SO2 concentration (SO2, from European Space Agency).Obesity captures the prevalence of obesity among adults (i.e., those with BMI greater than 30), measured in 2016 and from the World Health Organization. 9We use data on atmospheric sulfur dioxide (SO2) concentrations as a measure of air pollution.The SO2 data set is the Sentinel-5P near real-time high-resolution imagery, provided by the European Space Agency (ESA).10A complete list of variables and data sources is shown in Table 1.The summary statistics for the variables are in Table 2.

IV. Empirical specifications and results
To explain the pandemic outcomes across countries, we estimate the following regression at the country level: Here, i refers to a country.Y is the pandemic outcomes: the number of cumulative confirmed cases of COVID-19 per 1,000 residents, and the mortality rate from confirmed COVID-19 cases.X is basic country characteristics, including log GDP per capita, log total population, the tertiary dummies indicating the distance to China, and the average temperature.Delay includes both DelayI and DelayD, capturing both international and domestic mobility restrictions.Vulnerability includes the four preexisting factors that make the country more vulnerable to COVID-19: Elderly, Urban, Obesity, and SO2.Democracy is a dummy variable indicating the status of being a democratic country.Culture contains two variables: individualistic culture and out-of-group trust.
As discussed earlier, we use the imputation-augmented cultural variables to maintain sample size and its consistency across specifications.11In all specification, we also control for the continental dummies using the World Bank classification.12These continental dummies serve to hold constant factors related to geography, and cultural and institutional clusters.We include longitude and latitude in some specifications as the additional geography control variables.
An issue in modeling the pandemic spread is that there are strong spatial interactions.
Economists have increasingly recognized the importance of Tobler's law of geography-adjacent locations are more likely to share common characteristics than distant ones (Tobler 1970), which suggests that omitting spatial interactions could result in spurious correlation of location-specific variables (Kelly 2019).In the case of pandemic spread, spatial interactions and spillover could be especially relevant: countries near each other have more trade and social interactions, all facilitating pandemic spread.To accommodate general spatial interactions, we estimate the following spatial autoregressive model (SAC) to allow for cross-country interactions of both the dependent variable and the error term:13 Here Y is a vector of the outcome for all country.  is the ith row of W, which is the spatial weights matrix, for which we use the inverse-distance matrix-the weights are inversely related to the distances between the capital cities of countries.  is the error term, and the ith element of the vector .  is the error term in the   equation.We allow a country's pandemic outcome to be affected by those of other countries; we also allow for the error term to have spatial autocorrelation across countries as well.We report the estimates for  and , which indicates strength of spatial interactions.
Before proceeding, we need to underline the limitations of descriptive studies like this, as emphasized by Knittel and Ozaltun (2020).First, we have not seen the end of the pandemic tunnel, and we could only rely on data up to this point.Nevertheless, it is a good cutoff time to examine the COVID-19 behavior up to the end of our sample period (i.e., November 2020): right then the vaccines were becoming available, and COVID-19 behavior, from that point on, might behave quite differently from when vaccines were not available.Since then, country wealth and government capacity to buy vaccines and implement the vaccine program would figure more prominently, even dominate.Second, our paper is a descriptive paper that does not aim to address causal effects, and we can only present correlations of key factors with the pandemic outcomes.
We can only show how various key factors related to government pandemic policies, institutions and culture co-move with the pandemic outcomes.These correlations should offer insights about potential key factors that may affect the pandemic outcomes across countries.We can also use our intuition on bias direction to bound the effects of some variables.For instance, we believe that government delay variables are negatively correlated with the unobservable of the pandemic spread, which imply that the OLS estimates of the effects of delay variables on pandemic spread would be biased downward, and would form the lower bound of the causal effects.To the extent that the results make sense, and to the extent that events of the highest importance have the largest consequences and there is no time to wait for definitive causal study to guide our policies, descriptive studies like ours should strengthen some priors but discount other priors, and the reader is invited to offer their own interpretations.

Determinants of time to close down
Before analyzing the determinants of the pandemic outcomes, we first examine how DelayI and DelayD are related to other covariates for several reasons.First, it is useful to understand the circumstances associated with government policy delay in closing down.Second, in the pandemic outcome regressions, we recognize that the delay variables have the most serious endogeneity issue.
Other variables are significantly less so.Time-varying country characteristics such as GDP per capita and urbanization are measured for the previous year, not subject to reverse causality.Factors such as democracy, culture, and demographic vulnerability are pre-determined at the time of pandemic arrival.Understanding the nature of the delay variables thus would help us interpret the estimated association between government delays and pandemic outcomes.To this end, we relate the delay variables to other covariables in Table 3, reporting both the OLS and the SAC specifications.
Since both the SAC parameters,  and , are estimated to be statistically significant, there are significant spatial interactions in government pandemic responses.The positive  indicates that neighbors' pandemic-spreading factors tend to have similar effects on the home country; the negative  implies policy substitutions across countries-specifically, if neighbors are aggressively closing down, there is less need for the home country to do so.Given the significance of the SAC factors, our discussions focus on the SAC model here, but the qualitative results are very similar for the OLS and the SAC models.
There are indications that DelayI is related to the need for international travel and interactions.It rises with country population and urbanization, perhaps because both represent stronger need for international travel and interactions, and thus lead to stronger opposition to international close-down.Countries with more prevalent obesity are faster in closing down internationally.It is not significantly related to the income level, the distance to China, the average temperature; neither is it related to democracy, individualism, or trust.Figure 1 depicts the kernel density of DelayI by democracy dummy, and again, democracy is not visually related to DelayI.
The evidence seems to suggest that except for the need for international travel, DelayI is not strongly related to a country's institutional and cultural factors.
DelayD rises with country population and the share of elderly.It is also higher in more individualistic countries (not statistically significant with a t-statistics of 1.44).Interestingly, contrary to the notion that democracy deters decision speed, democratic countries have significantly lower DelayD by 0.7 standard deviation (SD) or a month-a large magnitude.Figure 2 depicts the kernel density of DelayD by democracy, and since it does not show clear differences in the kernel densities by democracy, the relatively prompt response of democratic countries must have emerged after conditioning on the covariates.
We conjecture that when estimating the impact of policy delay on pandemic outcomes, the selectivity of DelayD should be higher than DelayI because domestic closedown affects virtually everybody, while international closedown affects far fewer people.The estimates of the DelayI effect should thus be more informative about the true effects than DelayD.Indeed, reverse causality likely affects DelayD more than DelayI: when the pandemic is severe, DelayD is likely to be shorter, causing negative correlation of the unobservable with DelayD, making DelayD's estimate an under-estimation of the true effects.On the DelayI effect we have similar concerns, but to a much smaller extent.The estimates of the effects of DelayD on pandemic severity-and to a smaller extent those of DelayI-should form the lower bound of the true effects.

Basic country characteristics and time to close down
In Table 4 we report how the pandemic outcomes are related to country characteristics, government policy delays, and vulnerabilities.We proceed step by step.For the pandemic spread and the mortality, we first include only basic country characteristics and DelayI and DelayD in columns (1) and (1').We add vulnerabilities in columns ( 2) and (2').In columns ( 3) and (3'), we then add longitude and latitude to examine the robustness of the previous specification to further geographical controls.We allow the interaction of DelayD with vulnerabilities to see how delay in government policies interact with inherent vulnerabilities in columns ( 4) and (4').We do not interact DelayI with vulnerabilities because DelayI has less impact, and in exploratory runs, we do not find consistently interaction effects between DelayI and vulnerabilities.For columns ( 2), (2'), (4), and (4'), we have tried both the OLS and the SAC specification, and to save space, we report only the preferred specification-that is, the OLS when the two SAC parameters (, ) are both insignificant, and the SAC otherwise.
Most of the results in columns ( 1) and (1') that link country characteristics and policy delays to pandemic severity make sense.More populous countries tend to have higher pandemic mortality rate.Richer countries have higher pandemic spread, reflecting the role of higher density and more international trade; they also have lower mortality rate, possibly due to better healthcare facilities.Perhaps surprisingly but upon reflection not so, DelayD is associated with lower pandemic spread.As discussed earlier, reverse causality for DelayD is serious: worse pandemic spread would lead to faster government response and a shorter DelayD, resulting in spurious negative correlation between DelayD and pandemic spread.In contrast, DelayI is positively associated with pandemic mortality rate.Given the effects of DelayI is likely under-estimated, the causal effects of DelayI are likely positive on pandemic mortality.Increasing DelayI by one SD is associated with an increase in pandemic mortality rate by 0.004, or 17% of its mean, a pretty large effect.
Columns (2) and (2') add vulnerabilities, and columns (3) and (3') further add longitude and latitude as controls to examine the robustness.We obtain two main findings.First, the share of elderly and obesity prevalence are both associated with lower pandemic spread, perhaps because the elderly and the obese are less physically mobile and have fewer social interactions, reducing the risks of passing the virus.On average, neither the share of elderly nor obesity prevalence is significantly associated with higher pandemic mortality rate.Second, and in contrast, urbanization ratio and SO2 are both positively correlated with pandemic spread (and not with pandemic mortality rate).Increasing the urbanization ratio by one SD (0.23) is associated with increasing pandemic spread by 27 percent (using column (3)); doing the same for SO2 (1.01), 41 percent.
Columns ( 4) and (4') further add the interaction terms between DelayD and vulnerabilities.
A longer domestic mobility-restriction delay, when coupled with higher obesity or SO2 level, is associated with higher pandemic mortality rate.Increasing obesity by one standard deviation (SD) would imply a change in the marginal effects of DelayD on mortality (i.e., ) by from -0.004 to 0.003.Similarly, increasing SO2 by one SD from the mean would imply a change in the marginal effect of DelayD on mortality from -0.008 to 0.003.

Democracy, culture, and their interactions with government delay and vulnerabilities
We now examine how democracy and culture contribute to pandemic outcomes.In columns ( 1) and (1') of Table 5, we add these new variables directly.In columns ( 2) and (2'), we allow each of the three variables interact with DelayD and vulnerabilities. 14To simplify the presentation and to avoid multicollinearity, we do not include those interaction terms that are far from being statistically significant-we have explored and found that for the significant interaction terms, whether or not including those insignificant interaction terms does not affect the results on the significant interaction terms.As before, we only report the results for the preferred specification between the OLS and the SAC model.
From columns (1) and (1'), countries differing in individualism do not differ much in the pandemic outcomes.Nor do countries with greater level of general trust have systematically different pandemic spread or mortality.Democratic countries, relative to non-democratic countries, have no higher pandemic spread, but significantly lower pandemic mortality rate, by 0.9 percentage points, or 43 percent of the mean, a fairly large effect.
Figures 3 and 4 plot the kernel density of the pandemic outcomes by democracy, and the significant lower pandemic mortality rate of democratic countries is also evident.Examining the details of pandemic mortality, we find that the maximum value of pandemic mortality rate for democratic countries is 3.8%, which is below that of 14 non-democratic countries that have values ranging from 4.1% to 29.2%.Democracy thus appears to be more effective in containing the harm from the pandemic.
Columns (2) and (2') suggest fairly strong interactions between institutions, culture, and government delay.DelayD has more pronounced positive association with pandemic spread in democratic countries.The marginal effect of DelayD on pandemic spread in democratic countries is higher by 0.47, 2.6 times higher than that in non-democratic countries.The results are consistent with the notion that the government mobility-restriction policies have signaling effects on the pandemic severity, especially in democratic countries.Glaeser et al. (2020) show that a longer Delay signals to the public that the government does not consider the pandemic to be sufficiently serious and dangerous, worsening voluntary social distancing.Our results thus suggest that the signaling effect is stronger in democratic countries.
DelayD has more pronounced positive association with pandemic mortality rate in more individualistic countries.Increasing individualism by one SD from the mean is associated with an increase in the marginal effects of DelayD on pandemic mortality rate by 0.009 (i.e., 0.108*0.08),from a marginal effect around zero at the mean level of individualism (i.e., -0.0007).The results are consistent with the notion that individualist people, valuing the self and personal freedom more, do not internalize the cost of passing the virus to others, ultimately causing a spike in the pandemic mortality rate.
DelayD has more pronounced positive association with pandemic spread in more trusting countries.Increasing general trust by one SD is associated with an increase in the marginal effect of DelayD on pandemic spread by 22 percent (i.e., 2.569×0.085),or about 19% of the marginal effect at the mean trust level.This is consistent with the scenario that trusting societies feature more social interactions, and delay in mobility restrictions thus yield a greater harm.
Vulnerabilities have different association with pandemic outcomes depending on democracy.Demographic vulnerabilities-both the share of elderly and the share of the obeseare associated with less adverse pandemic outcomes in democratic countries than in nondemocratic countries.The negative marginal effect of the share of elderly on pandemic spread is more pronounced in democratic countries by roughly 100% (i.e., from -13.8 in non-democratic countries to -27.6 in democratic countries); the marginal "effect" of the share of the obese on pandemic mortality rate is roughly zero, in contrast to 0.07 in non-democratic countries.In contrast, SO2 shows the opposite tendency.The marginal "effect" of SO2 on pandemic spread in democratic countries is slightly higher at 2.89 (relative to 2.29 in non-democratic countries).This is an interesting yet also puzzling finding.
Vulnerabilities' association with pandemic outcomes also depend on a country's individualistic tendency.When a country is more individualistic, its share of elderly is more positively associated with pandemic spread: the marginal effect of elderly on pandemic spread is 0.54 at the mean individualistic culture score, and 3.43 when the score increases by one SD.In contrast, when a country is more individualistic, its SO2 level is less positively associated with pandemic spread: the marginal effect of SO2 on pandemic spread is 0.18 at the mean Individualistic Culture, and -0.25 when it increases by one SD.
In high-trust countries, the link between obesity and pandemic spread is relatively more positive.At the mean level of trust, the marginal "effect" of obesity is -2.99, and when trust increases by one SD, it is 0.033.High trust and its consequent social interactions thus offset the natural tendency for obesity to reduce social interactions.
Interestingly, and perhaps reflecting its robust role in spreading the pandemic and its role in reducing pandemic mortality, urbanization's associations with the pandemic outcomes do not hinge on democracy and culture.

V. Conclusions
The number of COVID-19 cases per million residents ranges from 3 to 78,634 across countries, and the ratio of 90 th to 10 th percentile in the COVID-19 mortality rate is more than 11.Our contributions center on our investigation of the relationship between such large differences across countries and preexisting pandemic vulnerabilities, the delay of government mobility restriction policy, democracy, and cultural norms.We allow these categories to interact with each other.We find that preexisting vulnerabilities such as urbanization and air pollution increase the pandemic spread.Contrary to what many think, on average, government delay in domestic mobility restriction, and social norms related to individualistic culture and general trust do not have significant associations with the pandemic outcomes.
Contrary to the notion that democracy slows down decisions, democratic countries responded faster by one month in imposing mobility restrictions.Democratic countries, relative to non-democratic countries, have significantly lower pandemic mortality rate, by 0.9 percentage points.Democratic countries also appear to be able to reduce the harm to the vulnerable population such as the elderly and the obese.However, government policy delays appear to be more harmful in democracies.Individualistic culture and general trust both amplify the adverse effects of government policy delay on pandemic outcomes, and sometimes amplify the positive link between pandemic outcomes and some vulnerabilities (such as a high share of elderly and of obesity prevalence).
Our research is subject to the caveat that all the conclusions are just correlations.This being said, our research offers suggestive evidence that decisive government policies are important where the country suffers from strong preexisting vulnerabilities and where the citizens respond to government actions more responsively such as in democracies.It is also suggestive that it is not individualistic culture, trust, and policy delay themselves per se that are decisive in explaining adverse pandemic outcomes; rather, it is likely their interactions with preexisting vulnerabilities.
At the time of the pandemic, some advantages during normal times, such as democracy, or a hightrust culture, when combined with preexisting vulnerabilities and/or policy delay, may have extra adverse effects.In conclusion, key institutional and cultural traits have quite different effects during normal times and crises, and in different contexts such as different levels of preexisting demographic vulnerabilities.

Individualistic culture
The Individual Empowerment index (version 2), obtained from WVS.It measures the extent to which the people in a society are mentally and habitually empowered to make their own choices and to pursue them in their actions.For each economy, use the latest wave of survey covering survey years of 1996-2014.
Higher value indicates higher level of individualistic culture in the country.We replace the missing value by the computed average in the cell of continent-income-group (each continent is divided to two income group based on GDP per capita).

Trust
Trust in unknowns and people of a different nationality and religion.Obtained from WVS.Higher value indicates higher level of out-group trust in the country.For each economy, use the latest wave of survey covering survey years of 1996-2014.Higher value indicates higher trust level of the country.We replace the missing value by the computed average in the cell of continent-income-group (each continent is divided to two income group based on GDP per capita).significance at the 10, 5, and 1 percent levels.Heteroskedasticity-corrected standard errors in columns.We have used the Delay I to interact with the four risks variables, but found less significant results, which indicate that the Delay I works on the spread and mortality more independently than the Delay D.
For models (2), ( 4), (2'), and (4'), we have examined whether  and  are significant, and we report the preferred specification between OLS or the spatial model.2), we have also tried using the subsample with the culture variables not missing, which has 58 observations.The results are quite similar for the culture and democracy variables, except that the coefficient of Individualistic Culture is positive and significant (with a coefficient of 6.21).Since doing this would imply the loss of more than half of the sample for which all other variables are available, we opt to use the sample that imputes missing cultural variables.In all the regressions, the missing indicators for the cultural variables are statistically insignificant.
For each regression, we have tried the OLS version and the spatial model, and we report only the preferred version: that is, the OLS if the two terms representing spatial interactions are both insignificant; and the spatial model if either one is statistically significant.

(
Cronin and Evans 2020;Goolsbee and Syverson 2020).Since individual choices are shaped by the level of trust toward the general public, we now consider how general trust affects the pandemic outcomes.
key data source is the COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at JHU, which contains the latest data including confirmed COVID-19 cases and deaths in 188 countries (or regions) from the beginning of COVID-19 outbreak.It is updated daily, and we use data up to November 19, 2020.We construct two key COVID-19 outcomes: the number of confirmed cases per 1,000 residents, to measure the pandemic spread; and the death rate among the COVID-19 infected, to measure the pandemic mortality rate.The pandemic mortality rate captures pandemic severity, and it sheds light on the amount of consumption a society is willing to give up to contain the pandemic(Hall, Jones and Klenow 2020).The dataset on government policy responses to the COVID-19 outbreak is the Oxford COVID-19 Government Response Tracker (OxCGRT), 4 which tracks government policy responses to COVID-19 across 147 countries and time.We use it to define international and domestic mobility restriction delay (denoted as DelayI and DelayD), respectively.DelayI is delay in controlling the international travel, equals to the number of days, divided by 100, from the day that the economy discloses the first COVID-19 case to the day that the government first announces the policy to restrict the international travel.DelayD is delay in implementing domestic closedown policy, equals to the number of days, divided by 100, from the day that the economy discloses the first COVID-19 case to the day that the government first announces their domestic closedown policy, which include school closing, workplace closing, public events cancelling, restrictions on gatherings, public transport closing, stay-at-home requirement and restrictions on internal movement.Smaller DelayI and DelayD indicate prompt government response.We rely on the World Development Indicator (WDI) dataset from the World Bank to obtain basic country-level characteristics including GDP per capita and the total population. 5For all timevarying variables, we choose the latest year that the observations are mostly available.In addition, we control basic geographic and weather conditions of a country.Since the first large outbreak happened in China, we control for the distance to China-with two dummy variables of the distance being in the middle and the top tercile of the distance distribution.6Since temperature plays a role in pandemic spread, we also control for the average temperature since the inception of the COVID-19 on a country until November 2020.In a sensitivity check, we also control for the longitude and latitude of capital of a country.We add data on democracy and trust from several sources.A dummy variable indicating the status of democracy is from Freedom House.Indicators of out-group trust and individualistic culture are obtained from the World Value Survey (WVS).Out-group trust ("Trust") measures the extent of trust in strangers and people of different nationalities and religions; a higher value implies stronger general trust.Individualistic culture (Individualistic Culture) is captured by the individual empowerment index (version 2) in WVS, which measures the extent to which the people in a society are mentally and habitually empowered to make their own choices and pursue them in their actions.Since countries with higher individual empowerment are usually viewed as being more individualistic, this index serves well as a proxy of individualistic culture.For countries having more than one wave of survey covering a period of 1996-2014, we use the data from the latest wave.Trust and Individualistic Culture have significant missing incidence, available in less than half of the number of countries for all other variables.To ensure a reasonable sample size, and to keep the samples with different set of explanatory variables similar, we opt to impute the missing values of Trust and Individualistic Culture as the mean of the continent-income-level cell of countries.7In addition, we add their missing indicators to allow for sample selection associated with missing incidences.We do not find the missing indicators to ever be statistically significant, indicating non-selectivity of their missing incidences.

Figure 1 .
Figure 1.Democracy and Delay in Restricting the International Travel

Figure 4 .
Figure 4. Democracy and Pandemic Mortality Rate

Table 1 : Variable definitions and Sources
Hale et al. (2020) cases per 1,000 residents.Calculated based on original coronavirus data from Johns Hopkins Coronavirus Resource Center.Updated to November 19, 2020.In regressions, we use Log (1+Cases per 1000 residents) as the dependent variable.Delay in controlling the international travel, equals to the number of days, divided by 100, from the day that the economy discloses the first COVID-19 case to the day that the government first announces the policy to restrict the international travel.Lower number indicates prompt government response.Calculated based on the dataset constructed byHale et al. (2020).Delay DDelay in implementing domestic close-down policy, equals to the number of days, divided by 100, from the day that the economy discloses the first COVID-19 case to the day that the government first announces their domestic closedown policy, which include school closing, workplace closing, public events cancelling, restrictions on gatherings, public transport closing, stay at home requirement and restrictions on internal movement.Lower number indicates prompt government response.Calculated based on the dataset constructed byHale et al. (2020).the country or special region to China is in the middle (top) tercile in the distance to China.

Table 3 :
The Determinants of Policy Delay

Table 5 . Effect of Democracy and Culture
Notes: *, **, *** represent statistical significance at the 10, 5, and 1 percent levels.Heteroskedasticity-corrected standard errors in columns.Country Char.means basic country characteristics including log(GDP PC), log(population), average temperature, and two dummies indicated the distance to China, and missing indicators for the cultural indicators.For column (1) and (