The Political Scar of Epidemics NBER WORKING PAPER SERIES THE POLITICAL SCAR OF EPIDEMICS Cevat Giray Aksoy Barry Eichengreen Orkun Saka Working Paper 27401 http://www.nber.org/papers/w27401 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 June 2020 We thank Nicolás Ajzenman, Chris Anderson (discussant), Belinda Archibong, Sascha Becker, Damien Bol, Ralph De Haas, Anna Getmansky (discussant), Luigi Guiso, Beata Javorcik, André Sapir, Konstantin Sonin, Dan Treisman, and webinar participants at the Bank of Finland, Comparative Economics Webinar series, EBRD, LSE and University of Sussex for helpful comments. We are also grateful to Kimiya Akhyani for providing very useful research assistance. Views presented are those of the authors and not necessarily those of the EBRD. All interpretations, errors, and omissions are our own. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research. NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications. © 2020 by Cevat Giray Aksoy, Barry Eichengreen, and Orkun Saka. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source. The Political Scar of Epidemics Cevat Giray Aksoy, Barry Eichengreen, and Orkun Saka NBER Working Paper No. 27401 June 2020, Revised in October 2020 JEL No. I1,N0,Z28 ABSTRACT What political legacy can we expect from the Coronavirus pandemic? Drawing evidence from past epidemics, we find that epidemic exposure in an individual’s “impressionable years” (ages 18 to 25) has a persistent negative effect on confidence in political institutions and leaders, but not in other institutions or individuals. We find similar negative effects on confidence in public health systems however, suggesting that the loss of confidence in political institutions and leaders is associated with healthcare-related policies. In line with this argument, our results are mostly driven by individuals who experienced epidemics under weak governments with less capacity to act against the epidemic, disappointing their citizens. We provide evidence of this mechanism by showing that weak governments took longer to introduce policy interventions in response to the COVID-19 outbreak. These results imply that the Coronavirus may leave behind a long-lasting political scar on the current young generation (“Generation Z”). Cevat Giray Aksoy European Bank for Reconstruction and Development Broadgate, 1 Exchange Square London EC2A 2JN United Kingdom aksoyc@ebrd.com Barry Eichengreen Department of Economics University of California, Berkeley 549 Evans Hall 3880 Berkeley, CA 94720-3880 and NBER eichengr@econ.Berkeley.edu Orkun Saka University of Sussex Business School Falmer, Brighton BN1 9RH United Kingdom and LSE o.saka@sussex.ac.uk 2 “Let me be blunt, too many countries are headed in the wrong direction. The virus remains public enemy number one, but the actions of many governments and people do not reflect this. The only aim of the virus is to find people to infect. Mixed messages from leaders are undermining the most critical ingredient of any response: trust.” (Tedros Adha nom Ghebreyesus, Director-General's opening remarks a t the media briefing on COVID-19 - 13 July 2020, World Health Orga nization) 1. Introduction Epidemics are stress tests for governments. Public officials and institutions face the challenge of assembling information and mounting effective interventions against a rapidly spreading and potentially fatal disease. They must communicate that information, describe their policies and convince the public of their trustworthiness. Fukuyama (2020) argues that the keys to success in dealing with COVID -19 are “whether citizens trust their leaders, and whether th ose leaders preside over a competent and effective state.” By way of example, Rothstein (2020) ascribes the greater success at containing the COVID-19 in Nordic countries compared to Italy to greater trust in government. Trust in government is not a given, however. Specifically, there is reason to ask how epidemic exposure itself will affect such trust. On the one hand, there is the “rally ‘round the flag hypothesis.” Trust in and support for political institutions and leaders tend to rise in the wake of actual and potential disasters (Mueller 1970, Baum 2002).2 On the other hand, trust in government may decline because public institutions and those charged with their operation fail to prevent or contain the pandemic. And in both cases the persistence of the effect is unclear. In this paper, we provide evidence on the effects of epidemics on trust in government.3 We use data on trust and confidence in governments, elections, and 2 For exa mple, Chanley (2002) shows that in the da ys a fter the 11 September 2001 a tta cks, public trust in the U.S. government rose to levels not seen since the 1960s. 3 There is limited evidence on other politica l impacts of epidemics a nd containment efforts. Ca mpante et a l. (2020) find that heightened concern about Ebola led to lower vote r turnout in the 3 national leaders from the 2006-2018 Gallup World Polls (GWP) fielded in nearly 140 countries annually.4 These are three related aspects of political trust. Questions about confidence in government elicit opinions about the political institutions and officials comprising government broadly defined. Questions about national leaders (leaders at the time of the poll, which is not necessarily the same as the time of epidemic exposure) elicit opinions about the head of state or government, namely the individual with the most influence over and most clearly associated with the actions taken by government. Questions concerning confidence in elections elicit views of the integrity and efficacy of the process by which those leaders are chosen. In practice, we obtain very similar results for all three dependent variables. We also use the average and the first principal component of these variables as a way of identifying their common element and again obtain very similar results. We link individual responses to the incidence of epidemics since 1970 as tabulated in the EM-DAT International Disasters Database. Building on work suggesting that attitudes and behavior are durably molded in what psychologists refer to as the “impressionable” late-adolescent and early-adult years (e.g. Krosnick and Alwin 1989, Giuliano and Spilimbergo 2014), we show that exposure to epidemics at this specific stage in the life course durably shapes confidence in government, elections and national leaders. United Sta tes but no evidence of a n anti-incumbent effect. Amat et a l. (2020) show that following the COVID-19 outbreak in Spa in, citizens expressed a stronger preference for technocratic governance and strong lea dership. Bol et a l. (2020) surveyed citizens of 15 European countries and found tha t the imposition of lockdown wa s a ssociated with a 2 percent increase in trust in government. Another body of research examines the impact of trust in government on epidemics a nd containment efforts. Marlow et a l. (2007) show that trust in government is a predictor of flu va ccine a cceptance by mothers in the United States. Using survey evidence from Liberia during the Ebola epidemic, Bla ir et a l. (2017) report that respondents who expressed low trust in government were less likely to ta ke precautions in their homes or a bide by government-mandated social dista ncing. 4 We group the terms confidence, trust, a nd approval under the general heading of trust. Confidence is the belief tha t certain future outcomes will obta in. Trust is vesting confidence in specific institutions or individuals for delivering those outcomes. App roval is a function of trust and other fa ctors, such as, in the present context, success in containing epidemics. Checkland, Marshall, and Ha rrison (2004) a nd Smith (2005), a lso working in a public health context, a rgue that confidence is something that is entrusted in systems (what we refer to here a s institutions), whereas trust is vested in individua ls (in the present context, lea ders). A further discussion of the rela tionship between trust a nd confidence is Ada ms (2005). 4 The effects are substantial: an individual with the highest exposure to an epidemic (relative to zero exposure) is 5.1 percentage points less likely to have confidence in the national government; 7.2 percentage points less likely to have confidence in the honesty of elections;5 and 6.2 percentage points less likely to approve of the performance of the national leader.6 These effects represent the average treatment values for the remainder of life. They decay only gradually and persist for two decades. These adverse effects are unique to political institutions and cannot be detected in the same individuals’ confidence in military, banks or media. Nor is the loss of political trust paralled by the loss of in-group or out-group trust in the same society. There is no evidence of a generalized decline in trust, in other words; our findings pertain specifically to trust in political institutions and leaders. Throughout, we control for other potentially confounding shocks that were experienced by individuals at the time of the epidemic. These include economic shocks (the growth and stability of the economy, inflation, GDP per capita and so on) and social and political shocks (internal conflict, external conflict, corruption scandals, democratic accountability, revolutions, assasinations, purges, riots, anti- government demonstrations and so on). We further incorporate fixed effects (country, year, age, cohort and country by year). We use the approach of Oster (2019) to establish that our results are unlikely to be driven by omitted variables.7 The effects we identify are specific to communicable diseases, such as viruses, that spread contagiously and where a timely and effective policy response is needed for containment. For non-communicable diseases, we do not see the same impact of impressionable-year outbreaks on subsequent views of the trustworthiness of 5 Rea ders may recall some discussion of how confidence in the presidential primary election in Wisconsin in 2020 might be a ffected by it occurring in the midst of COVID-19. Among the mechanisms highlighted in this deba te is the possibility tha t ma il-in ba lloting a nd other complications will slow the vote count a nd “invite a distrust of the election process” (Ad Hoc Committee for 2020 Election Fa irness a nd Legitimacy 2020). 6 The respective a verages of these three variables in our sample are 51 percent, 50 percent, a nd 50 percent. 7 The estimates suggest that for our results to be spuriously generated, the degree of selection on unobservables relative to observables needed to be 12 to 25 times (depending on the outcome) a s important for the outcomes as the included control variables. This is unlikely, since we control directly for va rious determinants of past a nd current political trust. 5 governments and leaders. This suggests that our finding of significant and persistent impacts reflects the success or failure of governmental authorities and agencies in putting in place timely and effective measures against contagion. We document that individuals exposed to epidemics in their impressionable years are less likely to have confidence in the public health system and the safety and efficacy of vaccination. The former is indicative of trust in the overall health policies of the government, while the latter reflects attitudes toward pharmaceutical interventions. These findings again suggest that the perceived adequacy of health- related government interventions during epidemics is important for trust in government generally. The magnitude and persistence of the effect depends on the strength of the government at the time of the epidemic. When individuals experience epidemics under weak governments, the negative impact on trust is larger and more persistent. This is consistent with the idea that such governments are less capable of effectively responding to epidemics, hence leading to a long-term fall in political trust. We substantiate this conjecture by considering this same conditioning factor, government strength, in the context of COVID-19. We show that government strength is associated with statistically significant improvements in policy response time. Finally, we show that our results are driven by the reaction to epidemic exposure in democracies. In democracies, residents sharply and persistently revise downward their political trust in the event of impressionable -year epidemic exposure. The same is not true, however, in autocracies. Evidently, citizens expect democratic governments to be responsive to their health concerns, and where the public -sector response is not sufficient to head off the epidemic they revise their views in unfavorable ways.8 In autocracies, in contrast, there may not exist a comparable 8 Consistent with this, Economist (2020) discusses tha t democracies typically respond more effectively to epidemics; our results suggest that when they disappoint this expectation, they are more severely punished. Below we a ddress a nd dismiss the a lternative interpretation that respondents in a utocracies a re more relucta nt to volunteer a la ck of trust or confidence in government. 6 expectation of responsiveness and hence little impact on political trust. In addition, democratic regimes may find consistent messaging more difficult. Because such regimes are open, they may allow for a cacophony of conflicting official views (Associated Press, 2020). This may result in a larger impact on trust when things go wrong. Our data cover some 750,000 respondents in 142 countries, which speaks to the generality of the findings. Our treatment variable, exposure to epidemics, is more plausibly exogenous than the man-made shocks employed in previous literature. Note that it is commonplace in the law to regard epidemics and pandemics as “Acts of God” and to invoke escape clauses in contracts. The number of people affected by a virus in different countries may still depend on country characteristic s. But there is also a random component in natural infection and mortality rates across different epidemics, which changes from virus to virus and thus brings randomness to our setting. Ebola was more deadly but less contagious, for example, than COVID-19. To be sure, trust in government may affect the severity of an epidemic (as we note in our opening paragraph). But as we show below, our findings are robust to using as the key explanatory variable a zero/one indicator for the occurance of an epidemic rather than its intensity. Section 2 reviews kindred literatures. Sections 3 through 5 describe our data, empirical strategy, and model. Section 6 presents the baseline results, while Section 7 reports a battery of robustness checks. Sections 8 and 9 then offer evidence on mechanisms and political behavior, respectively, after which Section 10 concludes. 2. Literature Our analysis connects up to several literatures. First, there is work in economics on the determinants and correlates of trust.9 Contributions here (e.g. Greif 1989, 9 In a ddition, there is work in politica l science and psychology. Levi and Stoker (2000) survey work in politica l science on how trust is conceptualized. They a rgue that trust is both rela tional and conditional. By rela tional, they mean that it involves a n individual making herself vulnerable to a nother individual, group, or institution (such a s government) that has the ca pacity to do her harm or to betray her. By conditional, they mean that trust is pla ced in specific individuals a nd institutions 7 Alesina and La Ferrara 2000, Nunn and Wantchekon, 2011) tend to focus on trust in other individuals (in-group and out-group trust) rather than trust in political institutions and leaders. Exceptions are Becker et al. (2016), Algan et al. (2017), and Dustmann et al. (2017).10 Becker et al. (2016) show that the historical presence of high-quality institutions (in regions previously governed by the Habsburg Empire) is associated with greater trust in government agencies today. 11 Algan et al. (2017) study the implications of the Great Recession for general trust and political attitudes (as well as for voting for anti-establishment parties), using regional data for Europe. They show that crisis-driven economic insecurity tends to be associated with lack of political trust. Dustmann et al. (2017) use data from the European Social Survey to identify economic and social characteristics associated with lack of trust in national parliaments and the European Parliament. They find that positive economic outcomes are important for trust in national parliaments, but that voters look to other competences when evaluating the trustworthiness of the European Parliament. Another literature analyzes how past experience shapes attitudes and behaviors. Malmandier and Nagel (2011) show that stock market returns experienced by an individual affect his or her subsequent financial risk-taking. Krosnick and Alwin over specific domains. Citizens may entrust their lives to their government during wa rtime or in a public hea lth emergency, for exa mple, but not otherwise. Work in psychology proceeds a long simila r lines. Thus, Ma yer et a l. (1995) also distinguish three dimensions of trustwo rthiness, which they denote a bility, benevolence, a nd integrity. By a bility, they mean the perceived technical competence of the trustee in a particular domain of interest. Perceptions of ability, therefore, consist, a s they put it, “of a subjective evaluation of the various skills a nd capabilities that may be needed for the trustee to a ctually a ccomplish what it is being trusted to do.” Benevolence derives from the extent to which the trustor believes the trustee is prepared to expend effort to protect the trustor. Integrity refers to the perception that the trustee follows a set of internalized values a cceptable to the trustor. All three a spects may be relevant to the problem at hand. 10 Other recent papers also analyze approval of leaders and governments, but they consider different independent variables. Margalit (2011) shows that job losses from import competition depressed the vote share of the incumbent president in 2004 a nd 2008 in the United States. Aksoy et al. (2018) show tha t tra de shocks affect political a pproval of governments a nd lea ders, Guriev et al. (2019) show that an increase in broadband mobile internet access reduces government approval, a nd Guriev a nd Treisman (2019) find that approval of lea ders is higher in non -democracies when media and internet a re restricted covertly, but approval ra tings fall when citizens observe censorship. 11 Specifica lly, they consider trust in the courts a nd police (one of which we a lso consider below). 8 (1989) and Osborne et al. (2011) show that partisanship and party affiliation are affected by past experience and, once formed, remain stable for long periods. Third, there is the literature, already noted, on the importance of the “impressionable years” in durably shaping attitudes and values. A seminal study pointing to the importance of this stage of the lifecycle is the repeated survey of women who attended Bennington College between 1935 and 1939 (Newcomb 1943, Newcomb, Koenig, Flacks and Warwick 1967), among whom beliefs and values formed then remained stable for long periods. An early statement of the resulting hypothesis is Dawson and Prewitt (1969); Krosnick and Alwin (1989), among others, then pinpoint the impressionable years as running from ages 18 to 25.12 When rationalizing the importance of the impressionable years, some scholars draw on Mannheim’s concept of the “fresh encounter,” suggesting that views are durably formed when late adolescents and early adults first encounter new ideas or events. Others invoke Erikson (1968) to suggest that individuals at this age are open to new influences because they are at the stage of life when they are forming their sense of self and identity. Still others suggest that attitudes are pliable at this stage of the lifecycle because views have not yet been hardened by confirmatory information (Converse, 1976). Spear (2000) links the literature on the impressionable years to work in neurology describing neurochemical and anatomical differences between the adolescent and adult brain, suggesting that these neurochemical and anatomical changes are associated with durable attitude formation. Niemi and Sobieszek (1977, p.221 et seq) suggest that only in the late adolescent years have young people 12 Some contributions to this litera ture suggest tha t a ttitudes toward, including trust in, other individuals a re instilled by parents at a very early age (see e.g. Erikson 1950; 1968), but that attitudes towa rd institutions, such a s the political institutions we a nalyze here, a re instilled by one’s peers, typica lly a t the juncture where adolescent lea ves the parental household. As we expla in below, we a lso checked whether epidemic exposure at a younger age had a significant effect on attitudes toward politica l institutions (in general it did not). This is not to deny that the family is a lso an important source of political idea s (the litera ture on political socialization surveyed by Niemi a nd Sobieszek, 1977, suggests that it in fa ct is), but to cla im that extra-familia l experience is a lso important. 9 developed “the cognitive capacity to deal with political ideas” and that the same can be said to some extent of individuals in their university years (p.222). In terms of applications, Giuliano and Spilimbergo (2014) establish that experiencing a recession between the ages of 18 and 25 has a significant impact on political preferences and beliefs about the economy. Using survey data from Chile, Etchegaray et al. (2018) show that individuals in their impressionable years in periods of political repression have a greater tendency to withhold their opinions, compared to those who grew up in less repressive times. Farzanegan and Gholipour (2019) find that Iranians experiencing the Iran-Iraq War in their impressionable years are more likely to prioritize a strong defense. Akbulut-Yuksel, Okoye, and Yuksel (2018) show that Germans in their impressionable years during the Nazi expulsion of Jews are less interested in politics later in adulthood, compared to the less exposed. Finally, we should mention two recent papers. Aasve et al. (2020), who use the approach of Algan et al. (2017) to study the impact of the 1918 -19 Spanish flu pandemic on social trust. Analyzing the General Social Survey for the United States, they find that individuals whose families emigrated to the Unite d States from a country with many Spanish flu victims display less trust in other people.13 Fetzer et al. (2020) use an experimental research design to establish that individuals’ 13 The nega tive impact on trust resembles our findings, a lthough their focus is trust in other people a s opposed to trust in politica l institutions a nd leaders. Aa sve et a l. (2020) have only one epidemic occurring a t one point in time and an unusually small sample (36 observations at country-year level). Historica l data on excess mortality are less a ccurate than modern data and the fact that the 1918-19 Spa nish Flu coincides with the end of the World Wa r I complicates the ca usal inference. Furthermore, a s discussed below, we fa il to detect a ny corresponding drop in trust towards out- group or in-group individuals in our setting. 10 beliefs about pandemic risk factors are associated with Covid-19 are causally related to their economic anxieties. 3. Data Our principal data sources are 2006-2018 Gallup World Polls (GWP) and the EM- DAT International Disasters Database. GWP are nationally representative surveys fielded each year starting in 2006 in about 150 countries, with responses from approximately 1,000 individuals in each country. Our full sample (depending on outcome variable) includes around 750,000 respondents in 142 countries.14 The outcome variables come from questions asked of all Gallup respondents about their confidence in the national government, their confidence in the honesty of elections, and their evaluation of the job performance of the incumbent leader:15 (i) “In (this country), do you have confidence in each of the following, or not: … How about the national government?” (ii) “In (this country), do you have confidence in each of the following, or not: … How about the honesty of elections?” (iii) “Do you approve or disapprove of the job performance of the leadership of this country?” 16 A visual summary of these variables is in Appendix Figure B.1-B.3. GWP provides information on respondents’ age, gender, educational attainment, marital status, religion, urban/rural residence, labor market status, and income. Controlling for employment status and income allows us to measure the impact of past epidemics on confidence in political institutions and leaders free of any direct effect on material well-being. 14 We drop observations for Na gorno-Karabakh, Northern Cyprus, Somalila nd, a nd Puerto Rico, as they a re not international recognised independent states. 15 We do not observe the respondent’s, lea der’s or government’s position on the left or right of the politica l spectrum. The politica l coloration of the government or lea der could in principle be incorporated into our setting. 16 These questions a re pa rt of the Ga llup “na tional institutions index.” If a respondent a sks for cla rifica tion or interpretation of the question, Ga llup surveyors are trained to answer “However you interpret the question,” or “It is wha tever the question means to you.” If a respondent a sks whether there is a more neutral response option than “yes” or “no,” surveyors a re trained to a sk whether “there is one tha t you lea n more towards.” 11 We also examine responses to three additional GWP questions: whether respondents have confidence in the military; confidence in financial institutions or banks; and confidence in media freedom. This helps to determine whether what we are capturing is the impact of epidemic exposure on trust and confidence in political institutions and political leaders specifically, as distinct from any impact on trust in society, its institutions, and its leaders generally. Data on the worldwide epidemic occurrence and its effects are drawn from the EM- DAT International Disasters Database from 1970 to the present. 17 These data are compiled from UN agencies, non-governmental organizations, insurance companies, research institutes, press agencies, and other sources. The database includes epidemics (viral, bacterial, parasitic, fungal, and prion) meeting one or more of the following criteria: • 10 or more deaths; • 100 or more individuals affected; • Declaration of a state of emergency; • Calls for international assistance. Our dataset includes 47 different types of epidemics and pandemics since 1970. This includes large outbreaks of Cholera, Ebola, and H1N1 and also more limited epidemics. Averaged across available years, H1N1, Ebola, Dysentery, Measles, Meningitis, Cholera, Yellow Fever, Diarrhoeal Syndromes, Marburg Virus, and Pneumonia were the top 10 diseases causing epidemic mortality worldwide. Many of these epidemics and pandemics affected multiple countries. 18 137 countries experienced at least one epidemic since 1970. This includes 51 countries in Africa, 40 in Asia, 22 in the Americas, 19 in Europe, and 5 in Oceania. 17 EM-DAT wa s esta blished in 1973 a s a non-profit within the School of Public Hea lth of the Ca tholic University of Louvain; it subsequently became a collaborating center of the World Health Orga niza tion. It a lso ga thers historical information on epidemics tha t took place before it was founded; however, those data a re patchy a nd bia sed towards well-recorded epidemics. Hence we only focus on epidemic cases that EM-DAT “live” collected a fter it wa s founded in ea rly 1970s. 18 Note tha t the EM-DAT International Disa sters Da tabase does not include da ta on non- communicable diseases. We employ separate data on non-communicable diseases below. 12 The most epidemic-prone countries in the dataset are Niger (25), Nigeria (25), Congo (22), Cameroon (21), Mozambique (20), Sudan (20), Uganda (20) and India (19). Advanced countries in our sample all experienced 5 or fewer epidemics.19 Each epidemic is tagged with the country where it took place. When an epidemic affects several countries, the database contains separate entries for each country. EM-DAT provides information on the start and end date of the epidemic, the number of deaths and the number of individuals affected, where the number of individuals affected is how many require assistance with basic survival needs such as food, water, shelter, sanitation, and immediate medical treatment during the period of emergency. Figure 1 provides a visual summary. We aggregate all epidemic-related information in this database at the county -year level and merge it with Gallup World Polls. In robustness checks, we employ a panel dataset on diseases from Institute for Health Metrics and Evaluation (IHME) and a dataset on recent epidemics from Ma et al. (2020). To explore underlying mechanisms, we use data from the Wellcome Global Monitor, Google Trends, the European Center for Disease Prevention Control, the Johns Hopkins Coronavirus Resource Center, and the Oxford COVID- 19 Government Response Tracker.20 Table 1 shows descriptive statistics for the outcome variables, country characteristics, and individual characteristics. Averaging across all country -years, nearly 50 percent of respondents say they have confidence in elections, have confidence in the national government, or approve of the performance of the leader.21 19 We do not provide the full country-year-epidemic list due to spa ce constraints. Interested rea ders can find the full list of epidemic cases used in our pa per online: https://www.dropbox.com/sh/cwe5n1ie8f4zmbl/AAAD9JdVnjXvQciAYX9ID7WOa?dl=0 20 See Appendix A for a dditional details on these data sources a nd our construction of variables. 21 There of course is very considerable heterogeneity within a nd a cross countries. For comparison, 72 percent respondents had confidence in the military, while only 60 and 54 percent had confidence https://www.dropbox.com/sh/cwe5n1ie8f4zmbl/AAAD9JdVnjXvQciAYX9ID7WOa?dl=0 13 4. Empirical Model To assess the effect of past epidemic exposure on confidence in government, elections and political leaders, we estimate the following specification: Yi, c, t, a, b = β1Exposure to epidemic (18-25)icb + β2Xi + β3Number of people affectedct-1 + β4Cc + β5Tt + β6Aa + β7Bb + β8Cc*Age + εict (1) where Yictab is a dummy variable for whether or not respondent i of age a and birthyear b in country c at time t approves or has confidence in an aspect of their country’s political institutions or leadership. Responses to all three questions are coded as dummy variables, with one representing a positive answer and zero otherwise. We estimate linear probability models for ease of interpretation. To measure the Exposure to epidemic (18-25), we calculate for each respondent the number of persons affected by an epidemic as a share of the population, averaged over the 8 years when the respondent was aged 18 to 25, consistent with the “impressionable years” hypothesis.22 Number of people affected controls for whether or not the individual is also ex posed to an epidemic contemporaneously. This is also calculated as the number of individuals affected by an epidemic as a share of the population in the country of residence in the year immediately prior to the interview.23 in ba nks a nd fina ncia l institutions a nd in the media , respectively. We use responses to these questions in pla cebo tests discussed below. 22 The effect of a n epidemic on younger cohorts may a lso depend on the nature of the virus (i.e., how letha l it is to the young). Unfortunately, EM -DAT does not contain information on the ages of the a ffected or of those who died. In addition, our treatment variable cannot differentiate between individuals who are themselves infected a nd individuals who may react to the infection of others. Thus, we ca n only ca lculate the average treatment effect a cross all types of epidemics operating through a combination of these channels. 23 This va ria ble is la gged to ensure that the independent variable is rea lized before the dependent va ria ble, since Gallup World Polls may interview individuals at any point in the year (not necessarily a t its end). 14 The vector of individual controls Xi includes indicator variables for urban residence and the presence of children in the household (any child under 15), and dummy variables for gender, marital status, employment status, religion, educational attainment, and within-country-year income deciles. We control for income before taxes in both log and log squared form.24 It is possible that prior epidemic exposure affects an individual’s responses partly by affecting his or her subsequent income. But we can rule out that prior exposure affects an individual’s responses solely by affecting his or her subsequent income by controlling for household income separately. A sense of the relative importance of this and other channels can be gained by comparing specifications with and without this income variable.25 We include fixed effects at the levels of country (Cc), year (Tt), and age (Aa). The country dummies control for time-invariant variation in the outcome variable caused by factors that vary cross-nationally. Year dummies capture the impact of global shocks that affect all countries simultaneously. Age dummies control for the variation in the outcome variable caused by factors that are heterogeneous across (but homogenous within) age groups. We also include country-specific age trends (Cc*Age) and cohort fixed-effects (Bb). A fully saturated specification includes also country-year fixed effects, which account for possible omitted country features that may change with time (such as GDP per capita, population, political regime, etc.). 26 We cluster standard errors by country and use sample weights provided by Gallup to make the data representative 24 These individual respondent controls a re important, since epidemics ma y ha ve a n effect depending on gender (Archibong a nd Anna n, 2017) a nd a va riety of other socioeconomic cha racteristics. Note tha t the income measure includes a ll wa ges a nd salaries, rem ittances from fa mily members living elsewhere, a nd a ll other income sources. Ga llup converts local income to International Dolla rs using the World Ba nk’s individual consumption PPP conversion factor. This ma kes income estimates comparable a cross countries. 25 In la ter pa rts of our a nalysis, we a lso control for the past GDP growth during a n individual’s impressionable years, which should in principle ta ke into a ccount any epidemic -induced change in income at the country and cohort level. We thank an anonymous referee for this suggestion. 26 This forces us to drop contemporaneous epidemic exposure, because it is perfectly correlated with the country-year effects. 15 at the country level. Finally, we limit our sample to individuals born in the same country in which they were interviewed by Gallup. 27 5. Threats to Identification One can imagine several potential threats to identification. First, estimates could be driven by factors that are specific to each cohort, since our treatment categorizes individuals in each country by year of birth. Some cohorts could have cohort- specific attitudes toward political institutions and leaders or be more or less trusting than others in general. Individuals born in the late 1940s and early 1950s, for example, may vest less trust in political institutions and leaders because they experienced the widespread protests against political repression in the late 1960s, their impressionable years. We therefore include dummies for year of birth so as to compare the individuals only within the same birth cohort. 28 Second, independent of the cohort effects, individuals may exhibit differential behavior across the life cycle. They may become more (or less) trusting as they age, for example. Political views and ideologies may change from more liberal when young to more conservative when older (Niemi and Sobieszek 1977). Age-specific factors may also matter if different generations were exposed to epidemics with different probabilities; given advances in science and improvements in national healthcare systems, one might anticipate that epidemics are less likely to be experienced by younger generations. We therefore include a full set of age-group dummies, which eliminates any influence on our outcome variables of purely age- related and generational effects. 27 We ca nnot guarantee that these individuals spent all of their impressionable years in their country of birth, but a ny measurement error a rising from this concern only stacks the cards a gainst us by lowering the precision of our estima tes. Furthermore, to th e extent that la rge epidemics push individuals to migra te to other countries not a ffected by the sa me epidemic, we may have a survivorship bias in our sa mple that leads us to underestimate the true effect of a past epidemic experience. 28 Including these dummies bia ses our estima tes downward if epidemics a re correla ted across countries a nd a ffect them simultaneously. In this ca se, a ny common effect of a n epidemic on a specific cohort will be subsumed by these cohort-specific dummies, a nd our trea tment will pick up the va riation in pa st epidemics only when they were staggered a cross countries. 16 Generational trends in political attitudes could be heterogeneous across countries. Some national cultures may be more flexible and open to change in individual values and beliefs, leading to larger differences across generations. We therefore include country-specific linear age trends. Third, any relevant omitted variable that varies across countries and years can bias estimates even when conventional country and year fixed effects are included separately. This issue arises when we observe individuals’ attitudes toward national political institutions and leaders. Because the identity of those leaders and the structure of those institutions may change over time, it can be difficult to separate these shifts in identity and structure from the treatment (i.e., the epidemic). For instance, even when approval of a leader declines following an epidemic, we may not capture this effect if the epidemic simultaneously triggers a change in the identity of the leader, bringing in someone for whom approval levels are higher. We address this by including dummies for each county-year pair. This eliminates all heterogeneity in our outcome variables tracable to country-specific time-varying factors, such as changes in the government or leader. Thus, the treatment only compares individuals within the same country and survey year, ensuring that these individuals face the same political institutions and leaders. This strategy also mitigates concerns that the results are driven by other structural differences between countries that are repeatedly exposed to epidemics and those that are not. Fourth, in any study of the impact of past experience on current outcomes, the underlying assumption is that the effect is durable and persistent. This is the essence of the “impressionable years” hypothesis. To the extent that this is not the case because the effect has a relatively short half -life, our empirical strategy will be biased towards failing to reject the null hypothesis of no effect. We explore this by tracing the impact of past epidemic exposure across different age groups and show that the effect persists at least for two decades while decaying gradually as individuals age. Hence, the full-sample estimates represent the average treatment effect across the whole life cycle after the impressionable years. 17 Fifth, although we fully saturate our specifications with fixed effects, there could still be other past exposures correlated with epidemics. To address this concern, we control for various past economic, political and social factors in the country in question in the individual’s impressionable years. Including these controls for other past conditions has no impact on the stability of our coefficients of interest. In addition, we use the methodology developed by Oster (2019). The results suggest that our findings are unlikely to be driven by unobserved variation. 6. Results Tables 2-4 report estimates of Equation (1). The dependent variables are a dummy indicating that the respondent has confidence in the national government (Table 2), a dummy indicating that the respondent approves of the performance of the leadership of his or her country (Table 3), and a dummy indicating that the respondent has confidence in the honesty of elections (Table 4). In all three tables, Column 1 reports estimates with country, year, and age group fixed effects. Column 2 adds the logarithm of individual income and its square, demographic characteristics, within country-year income decile fixed effects, and labor market controls. Column 3 adds country-specific age trends, while column 4 adds cohort fixed effects. Column 5 fully saturates the specification with country*year fixed- effects, non-parametrically controlling for all potentially omitted variables that can vary across countries and years. Column 1 of Table 2 shows a negative and statistically significant relationship between exposure to an epidemic in the individual’s impressionable years and current confidence in the national government. In contrast, the measure of contemporaneous epidemics is positive but statistically imprecise. Columns 2 to 4 show that the estimated effects change little as controls are added and that country- specific age-trends seem to be necessary for precisely identifying the effect of past epidemics in our setting. 18 Column 5 restricts all variation to within country -year observations and reports conservative estimates that are smaller in magnitude but still significant at 1 percent level.29 In our preferred model (Column 4), an individual with the highest exposure (0.032, that is, the number of people affected by an epidemic as a share of the population in individual’s impressionable years) relative to individuals with no exposure has on average 5.1 percentage points (-1.592*0.032) less confidence in the national government after his or her impressionable years.30 Given that the mean level of this outcome variable is 50 percent, the effect is sizable. Tables 3 and 4 report results for approval of the performance of the leader and confidence in the honesty of elections. The results on impressionable-year epidemic exposure have the same sign, statistical significance, and magnitude (a 6.2 percentage point decrease in approval of the political leader and a 7.2 percentage point decrease in the honesty of elections, where the mean outcome level is 50 percent). How persistent are the effects? We investigate persistence by estimating our baseline specification on the subsample of individuals closest to their impressionable years (that is, ages 26 to 35) and then repeatedly rolling the age window forward in a series of separate estimations. This permits us to observe how the coefficients change as we increase the distance between the age range in which impressionable individuals had exposure to epidemics and the age at which they are surveyed. If the effects are 29 It ma kes sense that the point estimates shrink when we only compare individuals within the same country and point in time. It is likely tha t both treatment and control groups in this setting must have experienced the same epidemics but only in different parts of their life cycle (impressionable vs non- impressionable yea rs). Hence, to the extent tha t epidemics ca rry nega tive effects for other experience windows, we a re only estimating the differential impact on individuals who were in their impressionable years during these epidemics, thus reducing the size of our point estimates. 30 Beca use epidemics a re rare events a nd our main independent varia ble of interest, Exposure to epidemic (18-25), is skewed to the right, it ma y not be a ppropriate to use its sta ndard deviation or mea n for understanding the effect size. 19 persistent, then the estimated coefficient should not change substantially as distance increases between the time of exposure and time of observation. Figure 2, based on Column 4 of Tables 2-4, shows the effect of epidemic exposure on the outcome variables. The effects on the base subsample (i.e., 26 -35) are more than three times larger than the point estimates for the full sample, confirming that the age groups closest to the experience window (i.e., 18-25) are disproportionately affected (compared to other age groups).31 For this base sample, the median time distance between the past experience window (median age: 21.5 years) and the subsample (median age: 30.5 years) is 9 years, hence documenting the effect of past epidemics in the medium term. When the model is re-estimated on successively older subsamples, the magnitude of the impact remains stable for the first six estimations following the base sample before decaying gradually. It nearly vanishes when estimated on the subsample of individuals aged 36 to 45, when the median distance between the experience window and the subsample is 19 years. On this basis, we conclude that epidemic experience during the impressionable years has persistent effects on political trust that can remain evident for two decades of adult life.32 Role of country characteristics We consider the baseline specification (Column 4 of Table 2) for various country subsamples. Each cell of Table 5 reports a separate regression. Each column shows the coefficient estimates for our main variable of interest: average epidemic exposure during the impressionable years. We report the baseline estimates for our main outcome variables in the top row. 31 We exa mine this specific point further below, where we compare impressionable year epidemic exposure with exposure when individuals a re younger a nd older than 18-25 (see Appendix Figure B.4). 32 We formally test the decay in the effect of epidemic exposure on political trust by interacting our ma in treatment variable with respondents’ a ge. Appendix Table B.1 confirms the earlier figures a nd shows that the negative effect of impressionable-period epidemic exposure is mitiga ted in later a ges. 20 The negative impact of epidemic exposure on confidence in the government and its leader is larger in low-income countries, although the difference across groups is not always statistically significant. This pattern is in line with evidence from Gómez et al. (2020), who find that people in the low-income countries see their governments more untrustworthy and unreliable in the context of public reactions to the COVID-19 pandemic. The negative impact of an epidemic also tends to be larger in countries with democratic political systems; the difference in coefficients for democracies and non-democracies is consistently significant at standard confidence levels. 33 An interpretation is that respondents expect democratically-elected governments to be responsive to their needs and are especially disappointed when such governments do not respond in ways that prevent or contain an epidemic. In contrast, the effect of prior epidemic exposure is insignificantly diff erent from zero in non- democracies, where there may be no similar presumption of responsiveness. In addition, democratic regimes may have more difficulty with consistent messaging. Because such regimes are open, they may allow for a cacophony of conflicting official views, resulting in a larger impact on confidence and trust. Either way, our results are driven by respondents in democratic regimes. 34 These results go some way toward addressing the issue of external validity in the context of COVID-19. The effects we report here are not limited to low-income countries, autocratic governments, or fragile democracies – the kind of regimes that are popularly associated with prominent epidemics such as Ebola. This suggests 33 We cla ssify political regimes ba sed on the most recent Polity5 dataset. Countries with Polity scores 5 a nd above are cla ssified as democracies. 34 This finding could a lso be expla ined by preference falsification, a phenomenon in which individuals’ responses to public surveys might be a ffected by social desira bility or implicit a uthoritaria n pressures (Kura n, 1987). Such bia ses could na turally a rise m ore often in non- democratic countries where survey participants feel the urge to hide their true beliefs, reducing the heterogeneity a cross respondents within the sa me country a nd time point. In a n unreported robustness check, we dropped ten per cent of the highest-ranking observations (in terms of approval of the leader) at the country-year level in our sample assuming that preference falsification -if exists- would be preva lent especially on these observations. We obta in simila r results implying that preference fa lsification by itself is unlikely to expla in the difference between democracies and a utocracies. 21 that our results may also have broader applicability to global pandemics such as COVID. 7. Robustness In this section we report further analyses establishing the robustness of our findings. Are the results driven by other past experience? The literature suggests that economic conditions (Hetherington and Rudolph, 2008), social conflict (De Juan and Pierskalla, 2016), and corruption (Anderson and Tverdova, 2003) also affect political trust. Appendix Tables B.2 and B.3 therefore consider whether our results are driven by other omitted economic , social and political exposures that individuals may have experienced in their impressionable years. In Appendix Table B.2 we include measures from the ICRG data set, which captures 12 aspects of national economic and political conditions.35 In particular, we include the following 12 indices to account for past economic, political, and social conditions: government strength, socio economic conditions, investment profile, internal conflict, external conflict, corruption, military presence in politics, 35 These a re (1) government strength - a n a ssessment both of the government’s a bility to carry out its decla red programs and its a bility to stay in of fice; (2) socioeconomic conditions - a n assessment of the socioeconomic pressures in a society that could constrain government action or fuel social dissa tisfaction; (3) investment profile - a n a ssessment of factors a ffecting risks to investment not ca ptured by other politica l, economic a nd fina ncial risk components; (4) interna l conflict - an a ssessment of political violence in the country a nd its a ctual or potential impact on governance; (5) external conflict - a n a ssessment of the risk to the incumbent government from foreign a ction, including both non-violent external pressure a nd violent external pressure; (6) corruption - an a ssessment of corruption in the politica l system; (7) milita ry in politics – a n a ssessment of the milita ry’s involvement in politics, even a t a peripheral level; (8) religious tensions – a n assessment of whether a single religious group seeks to repla ce civil la w by religious la w a nd to exclude other religions from the political a nd/or social process; (9) la w a nd order – a n a ssessment of the strength a nd impartiality of the legal system and popular observance of the la w; (10) ethnic tensions - an a ssessment of the degree of tension within a country attributable to ra cial, na tional, or linguistic divisions; (11) democratic accountability - a measure of how responsive government is to the people; a nd (12) bureaucracy quality – a n assessment of whether bureaucracy has the strength a nd expertise to govern without drastic changes in policy or interruptions in government services. 22 religious tensions, law and order, ethnic tensions, democratic accountability and bureaucracy quality. In Appendix Table B.3, we control for GDP growth, GDP per capita, inflation rate, political regime (Polity2 scores), assassinations, general strikes, terrorism/guerrilla warfare, purges, riots, revolutions, and anti-government demonstrations during the individual’s impressionable years. For all non-economic variables (excluding Polity2), we use the CNTS dataset in order to capture as many aspects of political conflict as possible. In both tables, we calculate the average values for each one of these dimensions during the impressionable ye ars of each individual. Including these past experiences as controls makes for smaller samples, since ICRG and CNTS cover only some of the countries and years in our main sample. None of these additional controls has much impact on the coefficients for past epidemics. Both the point estimates and statistical significance remain stable.36 Note that we cannot directly control for pre-epidemic levels of social and political trust due to lack of data availability.37 However, we do control for various factors that can explain both social and economic trust, therefore it is unlikely that our results can be explained by omitted variables bias or re verse causality. Nevertheless, we follow the method proposed by Oster (2019) to shed light on the importance of unobservables in Appendix Table B.8, where Panel A is based on the models with past exposure controls as in Table B.2 and Panel B is based on the models with past exposure controls as in Table B.3. 36 In a ddition Appendix Tables B.4 and B.5 show that we get simila r results if we were to control for the pre-existing values in the past (i.e., a ges 10-17) instead of impressionable years (i.e., ages 18-25) in order to make sure that the past controls themselves a re not influenced by the epidemic in the sa me experience window. Furthermore, our results remain qualitatively unchanged in Appendix Tables B.6 and B.7 a fter controlling for both impressionable-year experiences a nd country*year fixed effects a t the same time (à la Model 5 in Tables 2-3-4). 37 By interpola ting the corresponding va lues a cross a ll historica l wa ves of the World Va lues Surveys, we ha ve created a country panel da taset on various social a nd political trust va riables for the purpose of using them to control for pre-epidemic levels of trust in a country. However, due to poor country-year coverage in the old editions of the WVS, the size of our main Gallup sample falls by 95 percent to a bout 35,000 respondents. We, therefore, do not report the results a s we lack sta tistical power due to very sample size in these a nalyses. 23 We first reprint the baseline estimates for our main outcomes in the top row for comparison purposes. The second row of each panel then presents the estimation bounds where we define Rmax upper bound as 1.3 times the R-squared in specifications that control for observables following Oster (2019). The bottom row presents Oster’s delta, which indicates the degree of selection on unobservables relative to observables that would be needed to fully explain our results by omitted variable bias. The results in Appendix Table B.8 show very limited movement in the coefficients. The high delta values (between 12 and 24 depending on the outcome) are reassuring: given the wide range of controls we include in our models, it seems implausible that unobserved factors are 12 to 24 times more important than the observables included in our preferred specification. 38 Are the results unique to political institutions and leaders? It is important to establish that the relationship between epidemic exposure and subsequent views of political institutions and leaders is not simply part of a broader reassessment of social institutions and social trust (both in-group and out-group). If exposure to past epidemics worsens attitudes toward all national institutions and reduces social trust generally, it would be misleading to interpret the findings in Tables 2-4 as the effect of the epidemic exposure specifically on trust in political institutions and leaders narrowly defined. We therefore estimate similar models for outcomes related to views of other institutions. In Appendix Table B.9, outcome variables equal one if the individual has confidence in the military (column 1), in banks and financial institutions (column 2), and in media freedom (column 3); has relatives or friends to count on – a proxy for in-group trust (column 4); and has helped a stranger in the past month – a proxy for out-group trust (column 5). The first three variables represent the confidence in non-political insitutions in the same country, while the last two 38 The rule of thumb to be able to a rgue that unobservables cannot fully explain the treatment effect is for Oster’s delta to be over the value of one. 24 capture the potential change in individuals’ trust towards their in-group or out- group peers.39 There are no meaningful relationships between past epidemic exposure and any of these variables, consistent with our hypothesis that loss of trust by individuals with epidemic experience is specific to political institutions and leaders, and not a reflection of the general loss of trust in society and its institutions.40 Are the results driven by non-comparable samples? Not all Gallup respondents answered all three questions. Thus, the results could conceivably be biased by heterogenous, non-comparable samples across the three response variables. We therefore also consider only individuals who answered all three questions. We construct a new variable (“political trust”) that measures the average response of an individual across the three outcomes. We also construct a dependent variable that is the first principal component of these three variables. The results, reported in Appendix Tables B.10-B.11, confirm that our findings are robust across overlapping samples and alternative measures of political trust. 41 Are the results unique to impressionable years? One could argue that our treatment effect can be influenced by the potential differential response in individuals who may have experienced the same epidemics 39 As Ga llup does not ha ve direct questions on generalized (social) trust, we refer to these two va ria bles a s the closest proxies to measure the in-group a nd out-group trust. Alternatively, using a mea sure of individual donations or the civic enga gement index in Ga llup generates very similar results. 40 We understand that one could be concerned with media freedom in countries with low political trust a nd its potentially negative relationship with individuals’ confidence in media. However the media is not a political institution strictly defined, even though it ca n be influenced by politics. We ha ve no priors a bout how individuals might change their opinions a bout the media in the midst of a hea lth crisis. One could easily argue that individuals’ confidence in media may rise instead of falling if it functions well a s a tra nsmitter of life-saving information during the epidemic. Our results show tha t there is not much change in the long-term confidence in media, consistent with this - a priori - a mbiguous direction of the rela tionship. 41 In Appendix Table B.12, we a lso compare our 3 main outcome variables a s well a s 4 placebo outcomes (except the one on confidence in media which has a very small coverage in Ga llup) over the exa ct same group of individuals who have responded to a ll 7 questions. Aga in, we find that the loss of politica l trust a fter past epidemic exposure is unmatched by any of the alternative outcomes. 25 not during their impressionable-years but in other close experience windows before or after. Since these individuals will be categorised as counterfactuals in our setting, their potential differential response may drive our estimates upwards or downwards. In order to check this possibility, we re-estimate our specification with a focus on these alternative windows. Appendix Figure B.4 shows the effect of exposure in successive eight-year age windows (analogous to the eight-year window of ages 18 to 25).42 The analysis again considers our two composite dependent variables: the average of the three outcome variables and the first principal component of the responses. In both cases, the negative effect is only evident when epidemic exposure occurs in the individual’s impressionable years.43 This alleviates the concern in our setting that a counterfactual individual who experiences the same epidemic a little earlier or later than the impressionable age window may produce a differential response compared to an individual who has not experienced any epidemics at any of these windows.44 Are the results robust to alternative data for epidemics? We also analyze the recent large-scale epidemics reported in Ma et al. (2020), which constructs a country panel dataset starting in the early 2000s. This list of countries affected by post-2000 epidemics includes, at some point, almost all the countries in the world. For instance, H1N1 in 2009 alone infected more than 200 countries. Several aspects of this dataset make it less than ideal for our purposes. One is its 42 We repea t the a nalysis only for the first four windows a fter birth to make sure we ha ve age -wise comparable samples a cross separate estimations. It is important to keep in mind that as we check the la ter experience windows, respondents’ age at the time of the survey has to restricted to those older tha n the corresponding experience window. 43 We a ga in find the sa me for the three individual response variables. Results are availa ble upon request. Additionally, we checked the alternative experience windows rolling them by one year from 10-17 a ges to 18-25. We find tha t the effects increase in older-a ge windows a nd rea ch their ma ximum during a ges 16-23 before declining. 44 This interpretation is especially valid for the base-sample estimates (i.e., a ges 26-35) in Figure 2. In this subsample, only possible past experience windows a re from a ges 2 to 25 and hence, given the la ck of response in ea rlier a ge windows, it ca n be a rgued tha t our trea tment captures the hypothetical difference between an individual who experienced epidemics in their impressionable yea rs a nd another who never experienced a ny epidemics a t a ll. 26 short time span, which allows us to consider only individuals young enough to be in their impressionable years between 2000 and 2018.45 Another is that the dataset does not contain country-specific intensity measures and thus only can be used in dichotomous form. As will be clear later, epidemic intensity matters, in that only large epidemics in EMDAT dataset have a significant impact on political trust. At the same time, this list of recent epidemics buttresses our assumption of the exogeneity of our treatment variable, since the occurrence/start of an epidemic (as opposed to its intensity) is likely to be uncorrelated with country or cohort characteristics.46 In Appendix Table B.13, where we utilize this dataset, exposure to an epidemic (18-25) takes a value of 1 if the respondent experienced SARS, H1N1, MERS, Ebola, or Zika in his or her impressionable years. The results for confidence in elections and approval of the leader (as well as average and principal component proxies for political trust) are robust to the use of these alternative data. In line with our earlier results, the adverse impact of past epidemics is only evident in democratic countries. These results thus provide further evidence that the causal direction of the relationship runs from past epidemic experience to political trust later in life. Do countries with and without a pandemic display similar pre-trends? As mentioned earlier, Ma et al. (2020) provide a comprehensive dataset of pandemic events in this century. By creating an event-study setting around the dates on which a pandemic was declared by the WHO for a specific country, we can investigate whether countries experiencing pandemics exhibit the same pre -trends as other countries. We can also analyse how quickly the overall level of political trust changes after a pandemic. 45 This a lso means that we must drop all observations in Ga llup before 2008 -9 to ensure that the first impressionable-years cycle (2000-2007) is ca lculated before we apply this variable onto individuals. 46 As we show below, there is no evidence of a differential pre-trend in politica l trust between countries that were recently hit by an epidemic a nd those that were not. 27 To do this, we estimate the following model: Yi, c, t, a, b = β1LaggedPandemicict + β2Xi + β3Cc + β4Tt + β5Aa + β6Bb + β7Cc*Age + εict (2) LaggedPandemic is a dummy taking on a value of 1 if the WHO announced a pandemic for the country c in the year immediately preceding survey year t and 0 otherwise. This variable is lagged by one year to ensure that all respondents in the country experienced the pandemic (since Gallup surveys could be undertaken at any point of a year).47 Appendix Table B.14 shows that political trust starts declining immediately. In Figure B.5, we re-estimate the model changing the timing of the variable of interest. This helps to visualise the short-term response and also to check if the countries that were struck by a pandemic and those that were not shared similar trends in terms of their political trust levels before the pandemic hit the former.48 Countries with and without a pandemic share a common trend in the pre -event window; the divergence starts only after the pandemic hits. This supports the exogeneity assumption we made in a previous section in which we employed the occurence (rather than intensity) of recent epidemics as a shock to individuals’ impressionable years. Whereas there is no pre-trend prior to an epidemic infecting a country for the first time, the approval of the leader declines by more than 6 percentage points two years after. This aggregagre effect is large. It is comparable to the lifetime effect that we found for impressiomable-year exposures. 47 Here we do not include the pa st epidemic exposure variable a s we would like to ca pture the response of the whole population, ra ther only th ose for whom we ca n calculate the past experience window. 48 We conservatively restrict the event window a round the pandemic to plus/minus 2 years. This is beca use different pandemic events in Ma et al. (2020) may hit the same country in a matter of couple of yea rs, which complicates the identification in la rger event windows. 28 Is the response specific to communicable diseases? Poor public-policy responses to communicable diseases may have a powerful negative effect on trust in political institutions because those diseases can spread contagiously, making that policy response especially urgent. In contrast, non- communicable diseases may develop over longer periods and be driven by individual decisions and characteristics, such as lifestyles and demographics, instead of or in addition to government policy. Hence non-communicable diseases may not have equally powerful long-term effects on trust in political institutions. If they do, such effects should be smaller. Since the EM-DAT International Disasters Database does not include data on non- communicable diseases, we use data from IHME for the period 1990 to 2016.49 The communicable and non-communicable disease measures are population-adjusted and expressed in terms of Disability Adjusted Life Years Lost (DALYs). 50 As explained by Roser and Ritchie (2020), DALYs are a standardized metric allowing for direct comparison and summing of the burden of different diseases. We present results in Appendix Table B.15 for all countries in Column 1, for democratic countries in Column 2, and for non -democratic countries in Column 3. The top panel shows results for the outcome variable “ confidence in the national government,” the middle panel for “approval of the leader,” and the bottom panel for the “confidence in honesty of elections.” Each column in each panel is a separate regression in which we simultaneously include both types of past exposure (exposure to communicable and non-communicable diseases, respectively). There is a significant negative impact, as before, on confidence in the government and in elections of past exposure to communicable diseases. In contrast, we find no 49 Simila r to the previous exercise, this dataset is more limited than the EMDAT data that spans a much longer time period from the 1970s. 50 Communicable disea ses include dia rrhea, lower respira tory disease, other common infectious disea ses, malaria & neglected tropical disea ses, HIV/AIDS, and tuberculosis. Non -communicable disea ses include cardiovascular diseases, cancers, respiratory disease, diabetes, blood and endocrine disea ses, mental a nd substance use disorders, liver disea ses, digestive disea ses, musculoskeletal disorders, a nd neurological disorders. 29 statistically significant association between trust in these political institutions and exposure to non-communicable diseases during the impressionable years. The results thus confirm that the association we document is unique to communicable diseases. It remains the case, as before, that the full sample results are driven by respondents in democratic countries. Are large epidemics different? The effects we identify are larger for more severe epidemics. In Appendix Table B.16, we re-estimate our baseline model where, instead of the continuous variable reported in the top row, use indicators for the top 0.5 percent of exposures to epidemics, the top 1 percent, the top 2 percent, and the top 5 percent , each in a separate estimation. An epidemic exposure in the top 0.5, 1, or 2 percent of exposures causes a significant fall in an individual’s confidence in elections, the national government, and its leader.51 Moreover, the magnitude of the effect linearly increases with more intense experiences, which leads us to undertake the next analysis. Are the results driven by the intensive or extensive margin? In Appendix Table B.18, we distinguish the intensive and extensive margins of the treatment. For the extensive margin, we mean whether the effect is due to any level of epidemic exposure. To capture this, we construct a binary variable based on whether the number of persons affected by epidemics during the individual’s impressionable years is positive or zero. For the intens ive margin, we limit the sample to individuals with positive epidemic exposure in their impressionable 51 Rea ders may wonder how many democracies are included among the top 2 per cent of most severe epidemics. It turns out that there a re more democracies than autocracies in this limited sa mple. Democratic ca ses include Ja pan (1978), Botswana (1988), Ba ngla desh (1991), Peru (1991), Mozambique (1992), Pa raguay (2006) a nd Haiti (2010). In Appendix Table B.17, we estimate an intera cted model a nd find that the loss of political trust is la rger in those experience windows during which the epidemic-stricken country wa s relatively more democratic. 30 years. Approximately 55 percent of respondents in our surveys have no exposure to epidemics when impressionable and hence are dropped. Table B.18 shows that the treatment works via the intensive margin. It is not simply being exposed to an epidemic that generates the effect; rather, conditional on being exposed, the severity of the epidemic drives the results. When individuals with no epidemic exposure are excluded from the sample, the estimated effects of past exposure are, if anything, larger than in the full sample. Falsification We undertake two falsification exercises. Appendix Table B.19 focuses on the GWP subsample of individuals aged 30 or above who migrated to the country of interview in the previous 5 years. These individuals did not spend their impressionable years in the country of the interview. For falsification purposes, we assume that they did so (as opposed to spending those years in their country of origin). Second, Appendix Table B.20 assigns all individuals in the full sample to a random country for the calculation of their experience during impressionable years while keeping all else the same as in Tables 2-3-4. In both cases, we find no effect of these “made-up” and “randomly-assigned” treatments on political trust. Multiple hypothesis testing We also conducted multiple hypothesis testing by employing a randomization inference technique recently suggested by Young (2019). This helps to establish the robustness of our results both for individual treatment coefficients in separate estimations and also for the null that our treatment does not have any effect across any of the outcome variables (i.e., treatment is irrelevant), taking into account the multiplicity of the hypothesis testing procedure. The method builds on repeatedly randomizing the treatment variable in each estimation and comparing the pool of randomized estimates to the estimates derived via the true treatment variable. The 31 results presented in Appendix Table B.21 show that our findings remain robust both for the individual coefficients and the joint tests of treatment significance. Excluding potential “bad controls” One might worry that some of the individual characteristics (such as household income) are themselves affected by epidemic related economic shocks. We checked for potential “bad controls” (Angrist and Pischke, 2008) by excluding these individual characterisitics. Doing so does not substantively change the point estimates for our variables of interest (see Appendix Table B.22).52 Robustness to Alternative Treatment Definitions One might be concerned that large population may increase the intensity of the epidemic as well as the intensity of the epidemic affecting the population counts (through both mortality and immigration). We, therefore, checked the robustness of our results using population unadjusted treatment variable: the number of individuals affected by an epidemic averaged over the 8 years when the individual was aged 18 to 25. The results presented in Appendix Table B.23 show that our results are robust to this alternative definition. Ruling Out Influential Observations We rule out the importance of influential observations by plotting the coefficients of our preferred specifications as one year is omitted at a time. Appendix Figure B.6 shows that our coefficient estimates are quite stable even as a specific survey year is eliminated from our main sample in each iteration. We repeat a similar analysis with Appendix Figure B.7 in which we drop one random country at a time in each estimation for 15 consecutive trials (for illustration purposes) and again find that our estimates are not driven by a ny single country.53 52 We therefore keep these controls in our ba seline specification to a void omitted variable bia s. 53 Results a re similar for dropping any country within our sample and available upon request. We ha ve also undertaken a dfbeta a nalysis (unreported here) on all three main outcome variables and confirmed that the highest absolute dfbeta value among a ll observations in our sa mple is 0.04 and 32 8. Evidence on Mechanisms Weak, unstable governments with limited legislative strength, limited unity, and limited popular support presumably are less able to mount effective responses to epidemics. If they are prone to disappointing their constituents, we would expect the effects we identify to be strongest when the government in office at the time of exposure is weak and unstable, other things equal.54 To explore this hypothesis, we use ICRG data on government strength. These data are widely used in economics (see, for example, Knack and Keefer, 1997; Chong and Gradstein, 2007; Asiedu and Lien, 2011), political science and sociology (see, for example, Evans and Rauch, 1999; Grundler and Potrafke, 2019; Souva et al., 2008). They measure, for the period since 1984, the unity of the government, its legislative strength, and its popular support.55 We expect weak governments to perform poorly in epidemics, and conjecture that individuals will downgrade their confidence in government and trust in its leaders more severely as a result. We first calculate the average score for government strength in the individual’s impressionable years. We then construct an indicator that takes the value of 1 for this past experience if the observation is in the bottom half/tercile/quartile of impressionable-year government strength index scores across all respondents.56 We include this measure of impressionable-year government strength by itself in thus much smaller tha n the sta ndard threshold of 1.00 further a lleviating the concerns about influential outliers. 54 There is va st litera ture in political science on how fragmented and weak governments (such as multiparty coalitions) are “plagued” by a gency pro blems that may distort the policymaking process (Ma rtin a nd Va nberg, 2005). An economic example of this phenomenon has been shown on coa lition governments leading to excessive public spending due to reduced electoral a ccountability on the pa rt of the government parties (Vela sco, 2000; Bawn a nd Rosenbluth, 2006). Mia n et al. (2014) illustra te that governments become more polarized a nd weaker in the a ftermath of financial crises, which is likely to produce a deadlock in the parliament and decrease the chances of major fina ncial reform. 55 Wherea s in the ICRG da taset this index is la belled government sta bility, we refer to it as government strength, since we think this is a better name for what is essentially the implementation ca pacity of the incumbent government. 56 It is crucia l to include this variable categorically ra ther than in a continuous form to make sure tha t it is unlikely to respond to changes in the pandemic experience. 33 addition to interacting it with impressionable-year epidemic exposure to distinguish epidemic-specific and general effects. This leads to the following specification: Yi, c, t, a, b = β10Exposure to epidemicicb x Government strength icb + β9Government strength icb + β0 + β1Xict + β2Exposure to epidemicicb + β3Number of people affectedct-1 + β4Cc + β5Tt + β6Aa + β7Bb + β8Cc*Age + εict (3) The effect of exposure to an epidemic in Table 6 is more than twice as large if the epidemic is experienced under a weak government. The point estimates on the weak government dummy are small in magnitude and mostly statistically insignificant. This suggests that we are identifying not a “weak government effect” per se but rather the interaction with the effect of epidemic exposure in the presence of a weak government. Figures 3-5 show further evidence of the importance of government strength at the time of the epidemic. We again restrict the observations to the 26-35 age range and re-estimate the Equation (3) when rolling the age window forward. In each figure, the top panel shows the estimates for the total effect on individuals experiencing epidemics under weak governments, while the bottom panel shows the corresponding estimates for individuals experiencing epidemics under strong governments. For all outcomes, the negative impact on trust is larger and more persistent for respondents who experienced epidemics under weak governments. Again, this is consistent with the notion that these individuals became and remained more disenchanted with their country’s political institutions and leaders, insofar as those 34 institutions and leaders failed to adequately respond to the country -wide public- health emergency.57 Health policy at the time of the epidemic Governments’ pharmaceutical interventions, in particular their vaccination policies, have played an important role in the prevention of contagious disease. 58 Using data from GWP and the Wellcome Global Monitor, we therefore analyze whether attitudes regarding the health system and vaccination are affected by exposure to an epidemic. In the top panel of Table 7, the outcome is a dummy variable indicating that the respondent has confidence in the national healthcare system (via GWP).59 In the second panel, it is a dummy indicating that the respondent agrees or strongly agrees that “vaccines are effective.” In the third panel, it is a dummy indicating the respondent agrees or strongly agrees that “vaccines are safe.” In the fourth panel, it is a dummy indicating the respondents’ “children received a vaccine” that was supposed to prevent them from getting childhood diseases such as polio, measles, or mumps. In the final panel, it is whether the respondent agrees or strongly agrees that “vaccines are important for children to have.” The specification is otherwise as in Column 4 of Table 2. The results show that here too opinions are affected negatively by impressionable- year epidemic exposure. These results suggest that the same experience causing individuals to lose confidence in society’s capacity specifically to deliver adequate health outcomes also causes them to lose confidence in the political system and its 57 An a dditional implication of Tables 3-6 is tha t even individuals experiencing epidemics under strong governments display less politica l trust in the a ftermath. This finding is consistent with a model of lea rning where citizens may ex-ante over-trust their government (independent of whether it is wea k or strong) a nd where epidemics serve a s stress tests tha t ca n reveal new (nega tive) information a bout the government, thus correcting the initia l optimism. That our findings are genera lly stronger for less-educated people (see Ta ble 5) supports such a n interpretation. 58 The U.S. Centers for Disea se Control a nd Prevention lists va ccination as one of the “Ten Great Public Hea lth Achievements in the 20th Century” because of its impact on morbidity a nd mortality (Ba rra za et a l., 2018). 59 The exa ct wording of the question is a s follows: “In this country, do you have confidence in each of the following, or not? How a bout healthcare or medical systems?” 35 leaders more generally. In line with previous findings, Table 7 then shows that the negative impact of epidemic exposure is larger in countries with democratic political systems.60 Again consistent with earlier findings, Table 8 shows that individuals exposed to an epidemic in their impressionable years have more negative perceptions of health- related government policies if the epidemic was experienced under a weak government. Note that the sample is smaller since we use the Wellcome Global Monitor (2018) and ICRG covers only part of the countries and years in in our main sample. Despite much smaller sample size, 12 of the 15 interactions here are significant at the 95% confidence level. Evidence from COVID-19 Given the absence of internationally comparable data on policy interventions in response to past epidemics, we examine the association of government strength with policy interventions in the context of COVID-19. To do so, we investigate the relationship between government strength, measured as before, and the number of days between the date of first confirmed case and the date of the first COVID-19 policy (i.e. non-pharmaceutical intervention: school closure, workplace closure, public event cancellation, public transport closure, or restrictions on within-country movement) on a large sample of countries. We also provide case studies detailing the link between government strength and policy interventions for France, South Korea and the United Kingdom in Appendix C. Our sample consists of 78 countries that adopted non-pharmaceutical interventions between January 1, 2020 and March 31, 2020. We estimate OLS models, controlling for average Google search volume one week before the policy intervention to account for the possibility that public attention to COVID -19 60 These results a re in line with Legido-Quigley et a l. (2020), who argue that the integration of specific services like va ccination into the health system a s a whole a mplifies the capacity to a bsorb a nd adapt to health crises. 36 accelerates the non-pharmaceutical response. We also control for (log) cumulative own country cases one week before the policy, (log) cumulative own country deaths one week before the policy, (log) GDP per capita, (log) urbanization rate, (log) total population, (log) share of the population age 65 and above, Polity2 score, and a dummy variable indicating whether a country experienced an epidemic since 2000. Table 9 reports the results for the full sample in Column 1, for countries with above-median Polity2 scores in Column 2, and for countries with below-median Polity2 scores in Column 3.61 Although we make no causal claims, we find that government strength is associated with a statistically significant improvement in policy response time: a one standard deviation (0.765) increase in government strength reduces policy response time by three days. 62 This is a hint of why exposure to epidemic leads to major negative revisions of confidence in governments and trust in political leaders when governments are weak. According to Column 2, a one standard deviation (0.765) increase in government strength reduces the policy response time by four days in more democratic countries (those with above-median Polity2 scores). In contrast, there is little evidence that government strength reduces the policy response time in countries with below- median Polity2 scores. It is sometimes suggested that more democratic countries, where it is necessary to build a political and social coalition in support of restrictive policies, found it more difficult to respond quickly to the outbreak of COVID -19, compared to less democratic countries where “pseudo-democratic” leaders can move unilaterally to limit traditional political and civil rights and short-circuit democratic processes.63 Evidently, government weakness is mostly a problem in democratic societies, since this is there where it translates into a greater delay and less timely intervention. 61 We ca nnot split the sa mple into democracies vs. non -democracies because we ha ve only 10 countries in the non-democracy sample. This is why we instead split the sample b y below and above the median polity score. 62 Three da ys ca n make a substantial difference in the context of COVID-19, given the infection’s high ra te of reproduction when no non-pharmaceutical intervention is put in pla ce. 63 See for exa mple the discussion in Dia mond (2020). 37 9. Evidence on Political Behavior Even if epidemic exposure in one’s impressionable years affects self-reported trust in government, elections, and political leadership, it is not obvious that it also alters actual behavior. For example, one might expect that less confidence in elections leads individuals to vote less and take more political action through non -electoral means, (by participating taking place in demonstrations, participating in boycotts and signing petitions, for example).64 GWP lacks information data on such behavior. We therefore turn to the World Values Survey (WVS) and the European Social Survey (ESS). We use all available waves of the WVS covering the period 1981 -2014, as administered in more than 80 countries, where we focus on the democracies. We also consider annual waves of the ESS for the period 2002-2018 in over 30 countries. The WVS and ESS give us as many as 103,000 and 171,000 responses, respectively, depending on the question. We estimate our baseline model (Column 4 of Table 2) on several outcome variables related to individuals’ political behaviour Some of the results, in Appendix Table B.24, are consistent with the preceding conjecture.65 ESS respondents with epidemic exposure in their impressionable years are significantly less likely to have voted in recent national elections. Both WVS and ESS respondents are significantly more likely to have attended or taken part in lawful/peaceful public demonstrations. WWS respondents are significantly more likely to have joined boycotts and signed a petition. These are the type the 64 Ea rly evidence in the context of the recent COVID-19 crisis suggests that the young generation in US is more likely to sympathise with the George Floyd protests a nd more critical of the way US government is ha ndling the health crisis (Pew Research Center, 2020). 65 Note that we are not describing the self -reported behavior of the same individuals who, we showed a bove, self-reported less confidence and trust in elections, the national government, a nd the national lea der (where one might worry, there could be selective misreporting to minimize cognitive dissona nce). Rather, we a re analyzing completely different data sets where respondents are asked a bout a ctual political behavior a nd a ctions. This fa ct makes these additional findings especially striking. 38 responses one would expect from individuals rendered less confident in elections and other conventional governmental institutions.66 10. Conclusion In this paper we have shown that experiencing an epidemic can negatively affect an individual’s confidence in political institutions and trust in political leaders, with negative implications for this collective capacity. This negative effect is statistically significant, large and persistent. Its largest and most enduring impact is on the attitudes of individuals who are in their impressionable late-adolescent and early- adult years when an epidemic breaks out. It is limited to infectious or communicable diseases, where a government's success or failure in responding is especially important. It is the largest in settings where there already exist doubts about the strength and effectiveness of government. We also find that epidemic exposure in one’s impressionable years matters mainly for residents of democratic countries. Residents in democracies sharply revise downward their confidence and trust in political institutions and leaders following significant exposure, whereas the same is not true in autocracies. It may be that citizens expect democratic governments to be responsive to their concerns and that where the public-sector response is not adequate, they revise their attitudes unfavorably. In autocracies, there may not exist a comparable expectation of responsiveness. In addition, democratic regimes may find consistent messaging more difficult. Because such regimes are open, they may allow for a cacophony of conflicting official views, resulting in a larger impact on conf idence and trust. 66 Other results a re insignificant. There is no difference in the likelihood of never voting in national elections a mong WVS respondents a s a function of impressionable year epidemic exposure. Nor is there a ny difference among WWS respondents in the likelihood of having joined unofficial strikes or occupying buildings or fa ctories. Our a nalysis of these variables is necessarily ba sed on smaller sa mples, which may account for the contrast. However, the majority of the results where we have la rger sa mples a re consistent with the idea tha t not just self -reported trust but a ctual political beha vior are a ffected by epidemic exposure in the expected manner. 39 The implications are disturbing. Imagine that more trust in government is important for effective containment, but that failure of containment harms trust in government.67 One can envisage a scenario where low levels of trust allow an epidemic to spread, and where the spread of the epidemic reduces trust in government still further, hindering the ability of the authorities to contain future epidemics and address other social problems. As Schmitt (2020) puts it, “lack of trust in government can be a circular, self -reinforcing phenomenon: Poor performance leads to deeper distrust, in turn leaving government in the hands of those with the least respect for it.” 67 A releva nt study by Ajzenman et al. (2020) examines how political lea der’s words and actions a ffect people’s behaviour in the context of COVID-19 pa ndemic. The a uthors show tha t after Bra zil’s president publicly a nd emphatically dismissed the risks a ssociated with the COVID-19 virus a nd advises against isola tion, social distancing by residents in pro-government localities fall rela tive to pla ces in which pro-government sentiment is wea ker. 40 References Aasve, A., G. Alfani, F. Gaondolfi and M. Le Moglie (2020), “Epidemics and Trust: The Case of the Spanish Flu,” unpublished manuscript, Bocconi University. Ad Hoc Committee for 2020 Election Fairness and Legitimacy (2020), “Urgent recommendations in law, media, politics, and tech to advance the legitimacy of, and the public’s confidence in, the November 2020 U.S. elections,” Peltason Center for the Study of Democracy (April). Adams, B. (2005), “Trust versus confidence,” Defence Research and Development Canada Report CR-2005-203. Ajzenman, N., T. Cavalcanti and D. Da Mata (2020). “More than words: Leaders’ speech and risky behavior during a pandemic,” unpublished manuscript. Akbulut-Yuksel, M., D. Okoye and M. Yuksel (2018), “Social changes in impressionable years and the formation of political attitudes,” unpublished manuscript, Dalhousie University (February). Aksoy, C. G., S. Guriev and D.S. Treisman, D. S. (2018). “Globalization, government popularity, and the great skill divide,” NBER Working Paper no. 25062. Alesina, A. and E. La Ferrara (2000), “The determinants of trust,” NBER Working Paper no.621. Algan, Y., S. Guriev, E. Papaioannou and E. Passari (2017). “The European trust crisis and the rise of populism,” Brookings Papers on Economic Activity, 2017(2): 309-400. Amat, F., A. Arenas, A. Falcó-Gimeno, and J. Muñoz (2020). “Pandemics meet democracy. Experimental evidence from the COVID-19 crisis in Spain,” unpublished manuscript. Angrist, J. D., and J.S. Pischke. (2008). Mostly harmless econometrics: An empiricist's companion. Princeton: Princeton University Press. Anderson, C. J., and Y.V. Tverdova (2003), “Corruption, political allegiances, and attitudes toward government in contemporary democracies,” American Journal of Political Science 47(1): 91-109. Archibong, B. and F. Annan (2017). “Disease and gender gaps in human capital investment: Evidence from Niger's 1986 meningitis epidemic.” American Economic Review 107(5): 530-35. 41 Asiedu, E., and D. Lien (2011), “Democracy, foreign direct investment and natural resources” Journal of International Economics 84(1): 99-111. Associated Press (2020), “Trump, US officials send mixed messages on COVID - 19 risk to Americans” (25 February), https://www.fox2detroit.com/news/trump-us- officials-send-mixed-messages-on-covid-19-risk-to-americans Baum, M. (2002), “The constituent foundations of the rally -round-the-flag phenomenon,” International Studies Quarterly 46(2): 263-298. Bawn, K. and F. Rosenbluth (2006), “Short versus long coalitions: electoral accountability and the size of the public sector,” American Journal of Political Science 50(2): 251-265. Becker, S. O., K. Boeckh., C. Hainz and L. Woessmann, (2016), “The empire is dead, long live the empire! Long‐run persistence of trust and corruption in the bureaucracy,” Economic Journal 126(590): 40-74. Blair, R., B. Morse & L. Tsai (2017), “Public health and public trust: Survey evidence from the Ebola virus disease epidemic in Liberia,” Social Science Medicine 172: 89-97. Campante, F. R., E. Depetris-Chauvin R. Durante (2020), “The Virus of fear: The political impact of Ebola in the US,” NBER Working Paper no.26897. Bol, D., M. Giani, A. Blais, A. and P.J. Loewen, “The effect of COVID‐19 lockdowns on political support: Some good news for democracy? ” European Journal of Political Research. Forthcoming. Chanley, V. (2002), “Trust in government in the aftermath of 9/11: Determinants and consequences,” Political Psychology 23(3): 469-483. Checkland, K., M. Marshall and S. Harrison (2004), “Re-Thinking Accountability: Trust versus Confidence in Medical Practice,” Quality and Safety in Health Care 13: 130-135. Chong, A., and M. Gradstein (2007). “Inequality and institutions,” Review of Economics and Statistics 89(3): 454-465. Converse, P.E. (1976), The dynamics of party support: Cohort-analyzing party identification, Beverly Hills, CA: Sage. Dawson, R. and K. Prewitt (1969), Political socialization, Boston: Little, Brown & Co. https://www.fox2detroit.com/news/trump-us-officials-send-mixed-messages-on-covid-19-risk-to-americans https://www.fox2detroit.com/news/trump-us-officials-send-mixed-messages-on-covid-19-risk-to-americans 42 De Juan, A., and J.H. Pierskalla (2016), “Civil war violence and political trust: Microlevel evidence from Nepal,” Conflict Management and Peace Science 33(1): 67-88. Diamond, L. (2020), “America’s COVID-19 disaster is a setback for democracy,” The Atlantic (16 April), https://www.theatlantic.com/ideas/archive/2020/04/americas-covid-19-disaster- setback-democracy/610102/ Dustmann, C., B. Eichengreen, S. Otten, A. Sapir, G. Tabellini and G. Zoega (2017), Europe’s trust deficit: causes and remedies, London: CEPR Press. Economist (2020), “Out in the open: Covid-19 and democracy,” Economist (6 June): 77. Erikson, E. (1950), Childhood and Society, New York: Norton. Erikson, E. (1968), Identity, Youth and Crisis, New York: Norton. Evans, P., and J. Rauch (1999), “Bureaucracy and growth: A cross-national analysis of the effects of Weberian state structures on economic growth,” American Sociological Review: 748-765. Etchegaray, N., A. Scherman and S. Valenzuela (2018), “Testing the hypothesis of ‘impressionable years’ with willingness to self-sensor in Chile,” International Journal of Public Opinion Research 31: 331-348. Farzanegan, M. and H. Gholipour (2019), “Growing up in the Iran-Iraq war and preferences for strong defense,” MAGKS Joint Discussion Paper Series in Economics No.07-219, Philipps-University Marburg. Fetzer, T., Hensel, L., Hermle, J., & Roth, C. (2020). Coronavirus perceptions and economic anxiety. Review of Economics and Statistics, 1-36. Fukuyama, F. (2020), “The thing that determines a country’s resistance to the coronavirus,” The Atlantic (30 March), https://www.theatlantic.com/ideas/archive/2020/03/thing-determines-how-well- countries-respond-coronavirus/609025/ Giuliano, P., & A. Spilimbergo, (2013). “Growing up in a recession,” Review of Economic Studies 81(2), 787-817. Greif, A. (1989), “Reputation and coalitions in medieval trade: Evidence on the Maghribi traders,” Journal of Economic History 49(4): 857-882. https://www.theatlantic.com/ideas/archive/2020/04/americas-covid-19-disaster-setback-democracy/610102/ https://www.theatlantic.com/ideas/archive/2020/04/americas-covid-19-disaster-setback-democracy/610102/ https://www.theatlantic.com/ideas/archive/2020/03/thing-determines-how-well-countries-respond-coronavirus/609025/ https://www.theatlantic.com/ideas/archive/2020/03/thing-determines-how-well-countries-respond-coronavirus/609025/ 43 Gómez M, A. Ivchenko., E. Reutskaja and P. Soto-Mota (2020). “Behaviours, perceptions and mental wellbeing in high-income and low/middle-income countries at the beginning of COVID-19 pandemic,” unpublished manuscript. Guriev, S., N. Melnikov and E. Zhuravskaya (2019). “3G internet and confidence in government,” unpublished manuscript. Guriev, S. and D. Treisman (2019), “The popularity of authoritarian leaders: An empirical investigation,” unpublished manuscript. Grundler, K., and N. Potrafke (2019), “Corruption and economic growth: New Empirical Evidence,” European Journal of Political Economy 60, article 101810. Hale, T., A. Petherick, T. Phillips and S. Webster (2020). “Variation in government responses to COVID-19,” Blavatnik School of Government Working Paper no. 31. Hetherington, M. J., and T.J. Rudolph (2008), “Priming, performance, and the dynamics of political trust,” The Journal of Politics 70(2), pp.498-512. Knack, S. and P. Keefer (1997). “Does social capital have an economic payoff? A cross-country investigation,” Quarterly Journal of Economics 112(4): 1251-1288. Krosnick, J. and D. Alwin (1989), “Aging and susceptibility to attitude change,” Journal of Personality and Social Psychology 57: 416-425. Kuran, T. (1987), “Preference falsification, policy continuity and collective conservatism,” Economic Journal 97(387): 642-665. Legido-Quigley, H., Asgari, N., Teo, Y.Y., Leung, G.M., Oshitani, H., Fukuda, K., Cook, A.R., Hsu, L.Y., Shibuya, K. and Heymann, D., (2020), “Are high- performing health systems resilient against the COVID-19 epidemic?,” The Lancet, 395(10227): 848-850. Levi, M. and L. Stoker (2000), “Political trust and trustworthiness,” Annual Review of Political Science 3” 475-507. Ma, C., J.H. Rogers and S. Zhou (2020). “Global economic and financial effects of 21st century pandemics and epidemics,” unpublished manuscript. Malmandier, U. & S. Nagel (2011), “Depression babies: Do macroeconomic experiences affect risk taking?” Quarterly Journal of Economics 126(1): 373-416. Mannheim, K. (1928), “The problem of generations,” Essays on the Sociology of Knowledge, London, Routledge (translation 1952). 44 Margalit, Y. (2011), “Costly jobs: Trade-related layoffs, government compensation, and voting in US elections,” American Political Science Review, 105(1): 166-188. Marlow, L., J. Waller & J. Wardle (2007), “Trust and experience as predictors of HPV vaccine acceptance,” Human Vaccines and Immunotherapeutics 3(5): 171- 175. Martin, L. W. and G. Vanberg (2005), “Coalition policymaking and legislative review,” American Political Science Review: 93-106. Mayer, R., J. Davis & F. Schoorman (1995), “An integrative model of organizational trust,” Academy of Management Review 20(3): 709-734. Mian, A., A. Sufi and F. Trebbi (2014). Resolving debt overhang: Political constraints in the aftermath of financial crises. American Economic Journal: Macroeconomics 6(2): 1-28. Mueller, J. (1970), “Presidential popularity from Truman to Johnson,” American Political Science Review 64: 18-34. Murray, C. J., R.M. Barber, K.J. Foreman, A.A. Ozgoren, F. Abd-Allah, S.F. Abera and N.M. Abu-Rmeileh, (2015). “Global, regional, and national disability-adjusted life years (DALYs) for 306 diseases and injuries and healthy life expectancy (HALE) for 188 countries, 1990–2013: quantifying the epidemiological transition,” The Lancet, 386(10009): 2145-2191. Newcomb, T. (1943), Personality and social change: Attitude formation in a student community, New York: Dryden. Newcomb, T., K. Koenig, R. Flacks and D. Warwick (1967), Persistence and change: Bennington College and its students after 25 years, New York: Wiley. Niemi, R. and B. Sobieszek (1977), “Political Socialization,” Annual Review of Sociology 3: 209-233. Nunn, N., and L.Wantchekon (2011), “The slave trade and the origins of mistrust in Africa,” American Economic Review 101(7): 3221-52. Osborne, D., D. Sears and N. Valentino (2011), “The end of the solidly democratic south: The impressionable-years hypothesis,” Political Psychology 32: 81-107. Oster, E. (2019). Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics, 37(2), 187-204. 45 Pew Research Center (2020), “Younger adults differ from older ones in perceptions of news about COVID-19, George Floyd protests,” https://www.pewresearch.org/fact-tank/2020/07/09/younger-adults-differ-from- older-ones-in-perceptions-of-news-about-covid-19-george-floyd-protests/ (20 July). Roser, M. and H. Ritchie (2020), “Burden of disease”. Retrieved from: https://ourworldindata.org/burden-of-disease Rothstein, B. (2020), “Trust is the key to fighting the pandemic,” Scientific American (24 March), https://blogs.scientificamerican.com/observations/trust-is- the-key-to-fighting-the-pandemic/. Schmitt, M. (2020), “In the wake of its COVID-19 failure, how do we restore trust in government?” New America Weekly (23 April). Smith, C. (2005), “Understanding trust and confidence: Two paradigms and their significance for health and social care,” Journal of Applied Philosophy 22: 299- 316. Souva, M., D.L. Smith and S. Rowan (2008), “Promoting trade: The importance of market protecting institutions,” Journal of Politics 70(2): 383-392. Spear, L. (2000), “Neurobehavioral changes in adolescence,” Current Directions in Psychological Science 9(4): 111-114. Velasco, A. (2000), “Debts and deficits with fragmented fiscal policymaking,” Journal of Public Economics 76(1): 105-125. Young, A. (2019), “Channeling fisher: Randomization tests and the statistical insignificance of seemingly significant experimental re sults,” The Quarterly Journal of Economics 134(2): 557-598. https://www.pewresearch.org/fact-tank/2020/07/09/younger-adults-differ-from-older-ones-in-perceptions-of-news-about-covid-19-george-floyd-protests/ https://www.pewresearch.org/fact-tank/2020/07/09/younger-adults-differ-from-older-ones-in-perceptions-of-news-about-covid-19-george-floyd-protests/ https://ourworldindata.org/burden-of-disease https://blogs.scientificamerican.com/observations/trust-is-the-key-to-fighting-the-pandemic/ https://blogs.scientificamerican.com/observations/trust-is-the-key-to-fighting-the-pandemic/ 46 Figure 1: Average Number of People (per million) Affected by Epidemics, 1970-2017 Notes: This figure shows the number of people a ffected by epidemics (per million), a veraged a cross a ll a vailable years. Source: EM-DAT International Disa ster Da tabase, 1970-2017, UN Popula tion Da tabase, 1970-2017, a nd a uthors’ ca lculations. 47 Figure 2: Effects of Epidemics in Impressionable Years over Subsamples with Rolling Age-Windows Note: This figure shows the persistency of the effects on three main outcome variables by restricting the observations to the respondents who a re in the 26-35 a ge range a t the time of the survey (Base sa mple) and then repeatedly rolling this age window forward by one year for ea ch separate estimation. The specification is Column 4 of Table 2 and only the estimated coefficient on Exposure to epidemic (18-25) is plotted. Confidence intervals a re a t 95% significa nce level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 48 Figure 3: Effects of Epidemics on Confidence in Government over Subsamples with Rolling Age-windows (separately under weak and strong governments) Note: This figure shows the persistency of the effects on three main outcome varia bles by restricting the observations to the respondents who a re in the 26 -35 age ra nge a t the time of the survey (Base sa mple) a nd then repeatedly rolling this a ge window forwa rd by one yea r for ea ch separate estima tion. The specification is Equation 3/Table 6. The lower panel only plots the coefficient on Exposure to epidemic (18-25) whereas the upper panel plots the sum of the coefficients on Exposure to epidemic (18-25) a nd its intera ction with bottom qua rtile government strength dummy. Confidence intervals a re a t 95% significa nce level. Source: Ga llup World Polls, 2006-2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017. 49 Figure 4: Effects of Epidemics on Approval of the Leader Over Subsamples with Rolling Age-Windows (separately under weak and strong governments) Note: This figure shows the persistency of the effects on three main outcome varia bles by restricting the observations to the respondents who a re in the 26-35 age ra nge a t the time of the survey (Base sa mple) a nd then repeatedly rolling this a ge window forwa rd by one yea r for ea ch separate estima tion. The specification is Equation 3/Table 6. The lower panel only plots the coefficient on Exposure to epidemic (18-25) whereas the upper panel plots the sum of the coefficients on Exposure to epidemic (18-25) a nd its intera ction with bottom qua rtile government strength dummy. Confidence intervals a re a t 95% significa nce level. Source: Ga llup World Polls, 2006-2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017. 50 Figure 5: Effects of Epidemics on Confidence in Elections over Subsamples with Rolling Age-Windows (separately under weak and strong governments) Note: This figure shows the persistency of the effects on three main outcome varia bles by restricting the observations to the respondents who a re in the 26 -35 age ra nge a t the time of the survey (Base sa mple) a nd then repeatedly rolling this a ge window forwa rd by one yea r for ea ch separate estima tion. The specification is Equation 3/Table 6. The lower panel only plots the coefficient on Exposure to epidemic (18-25) whereas the upper panel plots the sum of the coefficients on Exposure to epidemic (18-25) a nd its intera ction with bottom qua rtile government strength dummy. Confidence intervals a re a t 95% significa nce level. Source: Ga llup World Polls, 2006-2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017. 51 Table 1: Sample Characteristics (1) Va ria bles Mea n (Standard deviation) Main dependent variables Confidence in national government 0.50 (0.50) – N: 760099 Confidence in honestly of elections 0.51(0.49) – N: 736679 Approval of the lea der 0.51 (0.49) – N: 719742 Ha ve confidence in the health system 0.62 (0.49) – N: 98283 Placebo outcomes Ha ve confidence in the milita ry 0.72 (0.45) – N: 730156 Ha ve confidence in the banks 0.59 (0.49) – N: 809972 Ha ve confidence in the media 0.54 (0.50) – N: 190167 Individual-level characteristics Age 41.58 (10.41) Ma le 0.47 (0.49) Tertia ry education 0.18 (0.38) Secondary education 0.50 (0.50) Ma rried 0.63 (0.48) Urba n 0.40 (0.49) Christia n 0.57 (0.49) Muslim 0.20 (0.40) Country-level characteristics Exposure to epidemic 0.002 (0.0015) Government strength 7.33 (1.26) Notes: Mea ns (sta ndard devia tions). This ta ble provides individual a nd a ggrega te level va ria bles a veraged a cross the 13 years (2006-2018) used in the a nalysis. The sa mple sizes for some variables a re different either due to missing data or because they were not asked in every yea r. 52 Table 2: The Impact of Exposure to Epidemic (18-25) on Confidence in National Government (1) (2) (3) (4) (5) Outcome ➔ Ha ve confidence in na tional government Ha ve confidence in na tional government Ha ve confidence in na tional government Ha ve confidence in na tional government Ha ve confidence in na tional government Exposure to epidemic (18-25) -1.073* -0.924 -1.614*** -1.592*** -0.508** (0.594) (0.576) (0.265) (0.262) (0.219) The number of people affected t-1 0.548 0.739 0.733 0.740 -- (3.478) (3.484) (3.457) (3.452) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income No Yes Yes Yes Yes Demogra phic characteristics No Yes Yes Yes Yes Income decile fixed effects No Yes Yes Yes Yes La bor market controls No Yes Yes Yes Yes Country*Age trends No No Yes Yes Yes Cohort fixed effects No No No Yes Yes Country*Year fixed effects No No No No Yes Observa tions 760099 760099 760099 760099 760099 R2 0.138 0.144 0.145 0.145 0.182 Mea n of outcome 0.50 0.50 0.50 0.50 0.50 Notes: * significa nt at 10%; ** significant at 5%; *** significant a t 1%. Outcome is a dummy variable indicating that the respondent has confidence in “national government”. Exposure to epidemic (18-25) defined as the a verage per ca pita number of people a ffected by an epidemic when the respondent wa s in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate assistance during a period of emergency (tha t is, requiring ba sic survival needs such as food, wa ter, shelter, sa nitation, a nd immediate medical a ssistance). Demographic characteristics include: a male dummy, a dummy for ea ch age group, dummy variables for marital sta tus (single, married), educational a ttainment (tertiary education, secondary education), religion dummies (Christia n, Muslim, and other religions), employment status (full-time employed, part-time employed, unemployed), a dummy variable for living in a n urban area a nd presence of children in the household (a ny child under 15). Income decile fixed-effects a re constructed by grouping individuals into deciles ba sed on their income rela tive to other individuals within the same country a nd year. Individual income includes a ll wa ges a nd salarie s in the household, remittances from fa mily members living elsewhere, and all other sources before taxes. Gallup converts local income to International Dolla rs using the World Bank’s individual consumption PPP conversion factor, which makes it comparable a cross a ll countries. Results use the Ga llup sa mpling weights a nd robust standard errors are clustered a t the country level. Source: Ga llup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 53 Table 3: The Impact of Exposure to Epidemic (18-25) on Approval of the Leader (1) (2) (3) (4) (5) Outcome ➔ Approval of the lea der Approval of the lea der Approval of the lea der Approval of the lea der Approval of the lea der Exposure to epidemic (18-25) -1.521*** -1.501*** -1.916*** -1.957*** -0.583*** (0.380) (0.369) (0.326) (0.330) (0.118) The number of people affected t-1 0.201 0.184 0.141 0.120 -- (2.696) (2.735) (2.710) (2.712) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income No Yes Yes Yes Yes Demogra phic characteristics No Yes Yes Yes Yes Income decile fixed effects No Yes Yes Yes Yes La bor market controls No Yes Yes Yes Yes Country*Age trends No No Yes Yes Yes Cohort fixed effects No No No Yes Yes Country*Year fixed effects No No No No Yes Observa tions 719742 719742 719742 719742 719742 R2 0.127 0.132 0.133 0.133 0.182 Mea n of outcome 0.51 0.51 0.51 0.51 0.51 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Outcome is a dummy variable indicating that the res pondent approves “the job performance of the lea der”. Exposure to epidemic (18-25) defined as the a verage per ca pita number of people a ffected by a n epidemic when the respondent wa s in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate assistance during a period of emergency (that is, requiring ba sic survival needs such a s food, wa ter, shelter, sa nitation, a nd immediate medical a ssistance). Demographic characteristics include: a male dummy, a dummy for ea ch a ge group, dummy variables for marital sta tus (single, ma rried), educational a ttainment (tertiary education, seconda ry education), religion dummies (Christia n, Muslim, and other religions), employment status (full-time employed, part-time employed, unemployed), a dummy variable for living in a n urban area a nd presence of children in the household (any child under 15). Income d ecile fixed-effects are constructed by grouping individuals into deciles based on their income relative to other individuals within the same country and year. Individual income includes a ll wa ges and sa laries in t he household, remittances from family members living elsewhere, and all other sources before taxes. Gallup converts local income to International Dolla rs using the World Bank’s individual consumption PPP conversion factor, which makes it comparable across a ll countries. Results use the Gallup sampling weights a nd robust standard errors are clustered at the country level. Source: Ga llup World Polls, 2006 -2018 and EM-DAT International Disa ster Database, 1970-2017. 54 Table 4: The Impact of Exposure to Epidemic (18-25) on Confidence in Elections (1) (2) (3) (4) (5) Outcome ➔ Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Exposure to Epidemic (18-25) -1.643** -1.481* -2.226*** -2.258*** -1.181*** (0.794) (0.811) (0.341) (0.339) (0.273) The number of people affected t-1 -3.734* -3.582 -3.645* -3.625* -- (2.203) (2.187) (2.195) (2.182) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income No Yes Yes Yes Yes Demogra phic characteristics No Yes Yes Yes Yes Income decile fixed effects No Yes Yes Yes Yes La bor market controls No Yes Yes Yes Yes Country*Age trends No No Yes Yes Yes Cohort fixed effects No No No Yes Yes Country*Year fixed effects No No No No Yes Observa tions 736679 736679 736679 736679 736679 R2 0.137 0.144 0.146 0.146 0.178 Mea n of outcome 0.51 0.51 0.51 0.51 0.51 Notes: * significa nt at 10%; ** significant at 5%; *** significant at 1%. Outcome is a dummy variable indicating that the res pondent has confidence in “honesty of elections”. Exposure to epidemic (18-25) defined a s the a verage per ca pita number of people a ffected by an epidemic when the respondent wa s in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate assistance during a period of emergency (that is, requiring ba sic survival needs such as food, wa ter, shelter, sa nitation, a nd immediate medical a ssistance). Demographic characteristics include: a male dummy, a dummy for ea ch age group, dummy variables for marital sta tus (single, married), educational a ttainment (tertiary education, seconda ry education), religion dummies (Christia n, Muslim, and other religions), employment status (full-time employed, part-time employed, unemployed), a dummy variable for living in a n urban area a nd presence of children in the household (a ny child under 15). Income decile fixed-effects a re constructed by grouping individuals into deciles ba sed on their income rela tive to other individuals within the same country a nd year. Individual income includes a ll wa ges a nd salaries in t he household, remittances from fa mily members living elsewhere, and all other sources before taxes. Gallup converts local income to International Dolla rs using the World Ba nk’s individual consumption PPP conversion factor, which makes it comparable across a ll countries. Results use the Ga llup sa mpling weights a nd robust standard errors are clustered a t the country level. Source: Ga llup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 55 Table 5: Heterogeneity (1) (2) (3) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Full sa mple -1.592*** (0.262) -1.957*** (0.330) -2.258*** (0.339) Ma les -1.153** (0.470) -1.351** (0.528) -2.014*** (0.379) Fema les -2.042*** (0.416) A -2.516*** (0.545) A -2.551*** (0.413) Low-income countries -11.181 (7.577) -20.701* (11.546) -11.753*** (4.145) High-income countries -1.212*** (0.262) -1.503*** (0.260) A -1.773*** (0.343) A Less tha n degree level -1.657*** (0.285) -1.753*** (0.295) -2.249*** (0.330) Degree level education 0.658 (1.242) A -5.120*** (1.328) A -1.071 (0.816) A Rura l -1.518*** (0.268) -1.377*** (0.265) -1.967*** (0.357) Urba n -3.015*** (0.781)A -6.195*** (1.452) A -4.049*** (0.893) A Low-income HH -0.226 (0.341) -0.112 (0.339) -2.527*** (0.485) Middle-income HH -3.015*** (0.781) -3.140*** (1.008) -2.207** (0.869) High-income HH -0.854* (0.457) -3.572*** (0.455) -1.559*** (0.389) Democratic countries -1.884*** (0.249) -1.587*** (0.301) -2.514*** (0.287) Non-democratic countries 3.097 (2.497) A 2.061 (2.529) A 0.880 (3.480) A Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. See notes to Table 2. A indicates statistically significant difference in each pair of means at p<.05. Results use the Ga llup sampling weights and robust standard errors a re clustered at the country level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 56 Table 6: The Role of Government Strength (1) (2) (3) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Exposure to epidemic (18-25)*MedianGov.Strength -4.033*** -1.092 -2.987*** (0.876) (0.849) (0.618) Exposure to epidemic (18-25) -0.235 -3.018*** -1.901** (1.038) (1.044) (0.833) Media nGov.Strength 0.014* 0.015* -0.000 (0.008) (0.009) (0.007) Exposure to epidemic (18-25)*BottomTercileGov.Strength -3.919*** -2.230*** -4.863*** (0.719) (0.629) (0.559) Exposure to epidemic (18-25) -1.048 -2.514*** -1.183* (0.808) (0.693) (0.698) BottomTercileGov.Strength 0.013* 0.023*** 0.002 (0.008) (0.008) (0.007) Exposure to epidemic (18-25)*BottomQuartileGov.Strength -3.578*** -2.027*** -4.643*** (0.748) (0.542) (0.521) Exposure to epidemic (18-25) -1.289 -2.657*** -1.373* (0.889) (0.640) (0.800) BottomQuartileGov.Strength -0.000 0.010 -0.002 (0.008) (0.010) (0.008) Observa tions 422523 394323 412051 R2 0.136 0.115 0.136 Notes: * significa nt a t 10%; ** significant a t 5%; *** significa nt a t 1%. The specification is Equation 3. See Ta bles 2-3-4 for va riable definitions. Results reported in each column a nd panel come from separate models. Results use the Ga llup sampling weights a nd robust standard errors are clustered at the country level. Source: Ga llup World Polls, 2006-2018, EM-DAT International Disa ster Database, 1970-2017, a nd the International Country Risk Guide. 57 Table 7: Impact of Exposure (Ages 18-25) on Attitudes towards Healthcare (1) (2) (3) Full-sa mple Democratic countries Non-democratic counties Outcome ➔ Confidence in healthcare Confidence in healthcare Confidence in healthcare Exposure to epidemic (18-25) -6.760*** (1.270) -6.543*** (1.649) -5.964 (4.084) Observa tions R2 95732 0.092 72793 0.098 22939 0.172 Outcome ➔ Va ccines are effective Va ccines are effective Va ccines are effective Exposure to epidemic (18-25) -1.178** (0.564) -1.699*** (0.554) -0.596 (0.470) Observa tions R2 81930 0.092 52638 0.072 25258 0.139 Outcome ➔ Va ccines are sa fe Va ccines are sa fe Va ccines are sa fe Exposure to epidemic (18-25) -1.685 (1.039) -2.703*** (0.672) -0.618* (0.341) Observa tions R2 81847 0.142 52612 0.117 25195 0.202 Outcome ➔ Children received a va ccine Children received a va ccine Children received a va ccine Exposure to epidemic (18-25) -0.339 (0.847) -1.432*** (0.417) 0.941 (0.650) Observa tions R2 67125 0.049 42415 0.056 21477 0.038 Outcome ➔ Va ccines are important for children to have Va ccines are important for children to have Va ccines are important for children to have Exposure to epidemic (18-25) -0.525 (0.566) -1.037* (0.549) -0.009 (0.295) Observa tions R2 83666 0.091 53623 0.084 25928 0.110 Notes: * significa nt a t 10%; ** significant at 5%; *** significant at 1%. Outcome is a dummy variable indicating tha t: the respondent a grees or strongly a grees that “vaccines a re effective” in the top panel; the respondent agrees or strongly a grees that “vaccines are safe” in the second panel; the respondent reports that their “children received a va ccine” that was supposed to prevent them from getting childhood diseases such as (such a s polio, measles or mumps),” in the third panel; the respondent agrees o r strongly a grees that “vaccines a re important for children to ha ve” in the bottom panel. Exposure to epidemic (18 -25) defined as the average per capita number of people affected by a n epidemic when the respondent was in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate a ssistance during a period of emergency (that is, requiring ba sic survival needs such a s food, wa ter, shelter, sanitation, a nd immediate medical a ssistance). Each specification inclu des country-fixed effects, year-fixed effects, demographic (a ma le dummy, a dummy for ea ch a ge group, dummy va riables for educational a ttainment (tertiary education, secondary education), religion dummies (Christian, Muslim, a nd other religions), a nd la bor market (full-time employed, pa rt-time employed, unemployed) characteristics, within-country income-deciles, dummy varia bles for living in a n urban area a nd presence of children in the household (a ny child under 15). Results use the Ga llup sampling weights a nd robust standard errors a re clustered at the country level. Source: the Wellcome Global Monitor, 2018 and EM-DAT International Disa ster Database, 1970-2017, a nd Polity5. 58 Table 8: The Role of Government Strength and Attitudes toward Healthcare and Vaccination (1) (2) (3) (4) (5) Outcome ➔ Ha ve confidence in hea lthcare Va ccines are effective Va ccines are sa fe Children received a va ccine Va ccines are important for children to ha ve Exposure to epidemic (18-25)*MedianGov.Strength -16.783 0.862 -3.554** -3.253*** -3.084*** (29.181) (0.981) (1.772) (0.610) (0.777) Exposure to epidemic (18-25) 1.071 -3.112*** -2.033 0.855 0.806 (35.099) (0.824) (1.843) (0.810) (0.777) Media nGov.Strength 0.023** -0.013* -0.011 -0.005 -0.003 (0.011) (0.007) (0.009) (0.004) (0.005) Exposure to epidemic (18-25)*BottomTerc.Gov.Strength -19.117 -1.815** -5.386*** -1.526*** -2.337*** (27.583) (0.762) (1.585) (0.405) (0.797) Exposure to epidemic (18-25) -3.716 -0.921 -1.090 -0.577 0.056 (26.485) (1.046) (1.510) (0.586) (0.585) BottomTercileGov.Strength 0.001 -0.004 -0.006 -0.005 -0.007 (0.009) (0.007) (0.008) (0.005) (0.006) Exposure to epidemic (18-25)*BottomQuar.Gov.Strength -49.140** -2.142*** -5.987*** -1.926*** -2.058** (23.329) (0.723) (2.099) (0.529) (1.024) Exposure to epidemic (18-25) 8.549 -1.057 -0.776 -0.350 -0.179 (20.633) (0.740) (1.722) (0.703) (0.776) BottomQuartileGov.Strength 0.004 0.005 -0.002 -0.001 -0.002 (0.010) (0.007) (0.008) (0.005) (0.005) Observa tions 49517 49799 49779 38702 50791 R2 0.110 0.078 0.133 0.048 0.091 Notes: * significa nt at 10%; ** significant a t 5%; *** significant a t 1%. The specification is Equation 3. See Table 7 for variable definitions. Results reported in each column and panel come from separate models. Results use the sampling weights and robust standard errors are clustered at the country level. Source: EM-DAT International Disa ster Da tabase, 1970- 2017, the Wellcome Global Monitor, 2018, and International Country Risk Guide. 59 Table 9: Government Strength and Policy Response Time to COVID-19 (1) (2) (3) Sa mple ➔ Full-sa mple Above Median Polity Score Below Media n Polity Score Government strength -3.611** (1.731) -5.357** A (2.560) -.0837 (2.077) [-2.764] [-4.231] [-0.062] Continent fixed effects Yes Yes Yes Country characteristics Yes Yes Yes Avera ge Google search volume one week before the policy Yes Yes Yes (log) cumulative own country cases one week before the policy Yes Yes Yes (log) cumulative own country deaths one week before the policy Yes Yes Yes Observa tions 78 39 39 Notes: * significa nt a t 10%; ** significant a t 5%; *** significant a t 1%. OLS regressions. Outcome variable is the number of days between the date of the first confirmed case and the date of the first COVID-19 policy (i.e. non-pharmaceutical intervention: school closure, workplace closure, public event ca ncella tion, public tra nsport closure, or restrictions on within -country movement) in the own country. Government strength is a n a ssessment of both the government’s ability to carry out its declared programs and its a bility to stay in office. It ra nges between 12 (maximum score) a nd 0 (minimum scor e) with higher scores indica ting better quality. Country characteristics include (log) GDP per ca pita, (log) urba nization ra te, (log) tota l popula tion, (log) share of population age 65 a nd a bove, Polity Score, a nd a dummy variable indicating whether a country experienced any epidemic since 2000. We a dd 1 to every country observation and then apply a logarithmic transformation. Bracket s report point estimates for one standard deviation (0.765) increase in government strength index. Robust standard errors are clustered at the country level. A indicates statistically significant differences between the pair estimates. The sa mple consists of 78 countries that ever-adopted non-pharmaceutical policy between 1/1/2020 and 31/03/2012. Source: EM-DAT, European Centre for Disea se Prevention Control, Google, Polity V, Oxford COVID-19 Government Response Tra cker, the International Country Risk Guide, World Ba nk. 60 Appendix A: Additional Data and Sources International Country Risk Guide Our data on institutional quality are from the International Country Risk Guide (ICRG). This measures 12 political and social attributes for approximately 140 countries from 1984 to the present. We focus on government strength, which is an assessment both of the government’s ability to carry out its declared programs and its ability to stay in office.68 Specifically, the index score is the sum of three subcomponents: (i) Government Unity; (ii) Legislative Strength; and (iii) Popular Support. In the original ICRG dataset, this measure is called as government stability. Throughout the paper, we refer to government stability as government strength as it captures the policy-making strength of the incumbent government. Scores for government strength range from a maximum of 12 and a minimum of 0. Wellcome Global Monitor The Wellcome Global Monitor (WGM) is a nationally representative survey fielded in some 160 countries in 2018. It is a global survey of how people think and feel about key health and science challenges, including attitudes towards vaccines and trust in doctors, nurses and scientists. WGM also provides information on respondents’ demographic and labor market characteristics. We use the Wellcome Global Monitor (WGM) to explore the mechanisms underlying our findings, and specifically whether these run through attitudes and feeling about the public health response to epidemics. Google Trends We use Google Trends data on searches to measure public attention paid to the COVID-19 pandemic. More specifically, we collected data on the volume of Google searches for “corona; korona; Wuhan virus; COVID; COVID -19,” translating these search terms into the official language of each country. We 68 Other institutional qua lity index measures cover democratic a ccountability, socioeconomic conditions, investment profile, internal conflict, external conflict, corruption, milita ry in politics, religious tensions, la w a nd order, ethnic tensions, and bureaucracy quality. 61 assemble these data on a daily basis at the country level for the period from January 1 through March 31, 2020. Observations are scaled from 0 (lowest attention) to 100 (highest attention). We exclude 21 countries where the internet is classified as “not free” according to Freedom House (2019). COVID-19 Related Cases and Deaths We obtain daily data on the coronavirus related cases and deaths by country from the European Center for Disease Prevention and Control (ECDC) and the Johns Hopkins Coronavirus Resource Center (JHCRC). There are minor reporting differences between the two sources. We use both datasets and create our measures of cases and deaths using the maximum value re ported in either dataset. Government Policy Responses We rely on the Oxford COVID-19 Government Response Tracker (OxCGRT) for information on public policy responses to the outbreak ( Hale et al., 2020). Specifically, we use the information on the following responses: (i) closing of schools and universities; (ii) workplace closures; (iii) public event cancellations; (iv) closing of public transport; (v) restrictions on internal movement. We again gather these data for the period between January 1, and March 31, 2020. Communicable and Non-communicable Diseases We distinguish communicable diseases (diarrhea, lower respiratory, other common infectious diseases, malaria and neglected tropical d iseases, HIV/AIDS, tuberculosis, other communicable diseases) from non -communicable diseases (cardiovascular diseases, cancers, respiratory disease, diabetes, blood and endocrine diseases, mental and substance use disorders, liver diseases, digestive diseases, musculoskeletal disorders, neurological disorders, other non- communicable diseases) using data from the Institute for Health Metrics and Evaluation. These data are at the country-level data and cover the period 1990- 2016. These measures are population -adjusted and expressed in Disability Adjusted Life Years Lost (DALYs), which is a standardized metric allowing for direct comparison and summing of burdens of different diseases (Roser and Ritchie, 62 2020). Conceptually, one DALY is the equivalent of one year in good health lost to premature mortality or disability (Murray et al. 2015). Country Characteristics Data on GDP per capita and urbanization rate come from the World Bank. We obtain the data on the total population and population by age from the Uni ted Nations. Data on political regime characteristics are from the Polity5 Series , with scores ranging from -10 to +10. We define 5 and above democracies. Political Behaviour We use the World Values Survey (WVS) and the European Social Survey (ESS) to measure political behavior. We use all available waves of the World Values Survey from 1981 to 2014. The dataset covers more than 80 countries and we use 6 variables to capture political behavior. In particular, questions aim to capture some forms of political action that people can take and asked as follows: please indicate whether you have done any of these things, whether you might do it or would never under any circumstances do it: (i) attending lawful/peaceful demonstrations; (ii) the respondent signing petition; (iii) joining in boycotts; (v) occupying buildings or factories; (vi) joining unofficial strikes. We code “have done” and “might do” as 1 and zero otherwise. We also use the question on whether the respondent vot ed in recent parliament elections. Additional data on political behavior come from the 2002 -2018 European Social Surveys. These surveys are fielded biannually in over 30 European countries. The key outcome variables we use come from questions asked to all ESS respondents: (i) during the last 12 months, have you taken part in a lawful public demonstration?; (ii) did you vote in the last national election? We code “yes” as 1 and zero otherwise. 63 The Cross-National Time-Series (CNTS) Data We use the following variables from CNTS data to control for individuals’ past domestic political experiences. The variable definitions are as follows: (i) Assassinations: any politically motivated murder or atte mpted murder of a high government official or politician; (ii) General Strikes: any strike of 1,000 or more industrial or service workers that involves more than one employer and that is aimed at national government policies or authority; (iii) Terrorism/Guerrilla Warfare: any armed activity, sabotage, or bombings carried on by independent bands of citizens or irregular forces and aimed at the overthrow of the present regime. A country is also considered to have terrorism/guerrilla war when sporadic bombing, sabotage, or terrorism occurs; (iv) Purges: any systematic elimination by jailing or execution of political opposition within the ranks of the regime or the opposition; (v) Riots: any violent demonstration or clash of more than 100 citizens involving the use of physical force; (vi) Revolutions: any illegal or forced change in the top government elite, any attempt at such a change, or any successful or unsuccessful armed rebellion whose aim is independence from the central government; (vii) Anti-government Demonstrations: any peaceful public gathering of at least 100 people for the primary purpose of displaying or voicing their opposition to government policies or authority, excluding demonstrations of a distinctly anti-foreign nature. 64 Appendix B: Additional Evidence and Analysis Appendix Figure B.1: Share of Respondents Who Have Confidence in Honesty of Elections Notes: This figure shows the share of respondents who have confidence in honesty of elections, a veraged a cross all a vaila ble years. Source: Ga llup World Polls, 2006 -2018. Appendix Figure B.2: Share of Respondents Who Have Confidence in National Government Notes: This figure shows the share of respondents who have confidence in na tional government, a veraged a cross all a vaila ble years. Source: Ga llup World Polls, 2006 -2018. 65 Appendix Figure B.3: Share of Respondents Who Approve the Performance of the Leader Notes: This figure shows the share of respondents who a pprove the performance of the lea der, a veraged a cross all a vaila ble years. Source: Ga llup World Polls, 2006 -2018. 66 Appendix Figure B.4: Effects of Epidemics in Alternative Treatment Years Panel A: Dependent variable is the average of all three outcome variables Panel B: Dependent variable is the 1 st principal component of responses Notes: This figure shows the trea tment effect for various a ge bands. Tha t is, we ca lcula te for each individual the number of people affected by an epidemic a s a share of the population, a veraged over the 8 years when the individual wa s 2-9 years old, 10-17 years old, 18-25 years old, a nd 26-33 years old. Each point estimate comes from four separate models. Specification is Column 5 of Table 2. Confidence intervals a re at 95% significance level. Results use the Gallup sampling weights a nd robust standard errors are clustered at the country level. Source: Ga llup World Polls, 2006-2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017. 67 Appendix Figure B.5: Short-term Effect of Epidemics on Political Trust Note: Epidemic year corresponds to the year in which World Health Organisation (WHO) declared one of the following pandemic/epidemic outbreaks for the country in which Gallup respondent resides: SARS, H1N1, MERS, Ebola, or Zika. Specification is the same as in Equation 2. Confidence intervals are at 90% significance level. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018 and Ma et al., 2020. 68 Appendix Figure B.6: Robustness to Dropping One Year at a Time Note: This figure shows the point estimates on Exposure to epidemic (18-25) variable on three main outcome variables while dropping one sample year at a time. The specificatio n is Column 4 of Tables 2, 3 and 4. Only the estimated coefficient on Exposure to epidemic (18-25) is plotted. Confidence intervals are at 95% significance level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disaster Database, 1970-2017. 69 Appendix Figure B.7: Robustness to Dropping One Country at a Time Note: This figure shows the point estimates on Exposure to epidemic (18-25) variable on three main outcome variables while randomly dropping one sample country at a time. The specification is Column 4 of Tables 2, 3 and 4. Only the estimated coefficient on Exposure to epidemic (18-25) is plotted. Confidence intervals are at 95% significance level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disaster Database, 1970-2017. 70 Appendix Table B.1: Persistency of the Effect (1) (2) (3) Outcome variable ➔ Ha ve confidence in na tional government Approval of the Leader Ha ve confidence in honesty of elections Exposure to Epidemic (18-25) -19.683*** (4.340) -18.251* (10.260) -17.498** (7.159) Exposure to Epidemic (18-25)*Age 0.464*** (0.104) 0.418* (0.250) 0.391** (0.167) The number of people affected t-1 0.649 (3.432) 0.039 (2.698) -3.693* (2.174) Country fixed effects Yes Yes Yes Yea r fixed effects Yes Yes Yes Age group fixed effects Yes Yes Yes Individual income Yes Yes Yes Demogra phic characteristics Yes Yes Yes Income decile fixed effects Yes Yes Yes La bor market controls Yes Yes Yes Country*Age trends Yes Yes Yes Cohort fixed effects Yes Yes Yes Observations R2 760099 0.145 719742 0.133 736679 0.146 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Outcome is a dummy variable indicating that the res pondent has confidence in “honesty of elections”. Exposure to epidemic (18-25) defined as the average per capita number of people affected by an epidemic when the respondent was in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate assistance during a period of emergency (that is, requiring basic survival needs such as food, water, shelter, sanitation, and immediate medical assistance). Demographic characteristics include: a male dummy, a dummy for each age group, dummy variables for marital status (single, married), educational attainment (tertiary education, secondary education), religion dummies (Christian, Muslim, and other religions), employment status (full-time employed, part-time employed, unemployed), a dummy variable for living in an urban area and presence of children in the household (any child under 15). Income decile fixed-effects are constructed by grouping individuals into deciles based on their income relative to other individuals within the same country and year. Individual income includes all wages and salaries in the household, remittances from family members living elsewhere, and all other sources before taxes. Gallup converts local income to International Dollars using the World Bank’s individual consumption PPP conversion factor, which makes it comparable across all countries. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disaster Database, 1970-2017. 71 Appendix Table B.2: Robustness to Controlling for Other Economic and Political Shocks (1) (2) (3) (4) (5) (6) Outcome ➔ Have confidence in national government Have confidence in national government Approval of the leader Approval of the leader Have confidence in honesty of elections Have confidence in honesty of elections Exposure to Epidemic (18-25) -3.589*** -3.417*** -3.926*** -3.944*** -4.373*** -4.219*** (0.585) (0.787) (0.487) (0.746) (0.636) (0.0849) The number of people affected t-1 0.847 0.876 0.872 0.698 -3.308* -3.354* (3.183) (3.019) (2.419) (2.218) (1.851) (1.701) Government strength (18-25) -- -0.001 -- -0.012* -- 0.006 (0.005) (0.007) (0.005) Socioeconomic conditions (18-25) -- -0.018*** -- -0.007 -- -0.018*** (0.006) (0.007) (0.006) Investment profile (18-25) -- 0.007 -- 0.010* -- 0.002 (0.006) (0.006) (0.006) Internal conflict (18-25) -- -0.007 -- -0.013** -- -0.002 (0.005) (0.006) (0.005) External conflict (18-25) -- 0.002 -- -0.001 -- 0.006 (0.005) (0.006) (0.004) Corruption (18-25) -- -0.009 -- -0.010 -- -0.005 (0.010) (0.010) (0.009) Military in politics (18-25) -- 0.021** -- 0.019* -- 0.010 (0.009) (0.011) (0.009) Religious tensions (18-25) -- -0.003 -- -0.005 -- -0.003 (0.011) (0.014) (0.010) Law and order (18-25) -- 0.030** -- 0.045** -- 0.041*** (0.015) (0.017) (0.014) Ethnic tensions (18-25) -- 0.011 -- 0.013 -- 0.005 (0.008) (0.010) (0.007) Democratic accountability (18-25) -- -0.005 -- -0.009 -- -0.016** (0.007) (0.010) (0.006) Bureaucracy quality (18-25) -- -0.017 -- -0.024 -- -0.022 (0.016) (0.021) (0.014) Observations R2 422523 0.136 422523 0.137 408564 0.139 408564 0.140 412051 0.137 412051 0.137 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018, EM-DAT International Disaster Database, 1984-2017, and ICRG 1984-2017. 72 Appendix Table B.3: Robustness to Controlling for Other Economic and Political Shocks (1) (2) (3) (4) (5) (6) Outcome ➔ Have confidence in national government Have confidence in national government Approval of the leader Approval of the leader Have confidence in honesty of elections Have confidence in honesty of elections Exposure to Epidemic (18-25) -1.879*** -1.743*** -2.274*** -2.204*** -2.519*** -2.185*** (0.502) (0.632) (0.515) (0.576) (0.348) (0.544) The number of people affected t-1 3.118** 3.077** 1.634 1.478 -1.900** -1.825** (1.374) (1.381) (1.540) (1.505) (0.800) (0.811) Assassinations (18-25) -- 0.006 -- 0.008* -- 0.002 (0.005) (0.004) (0.005) General Strikes (18-25) -- 0.010 -- 0.012 -- 0.005 (0.007) (0.009) (0.007) Terror./Guerrilla Warfare (18-25) -- -0.023* -- -0.015 -- -0.024** (0.012) (0.020) (0.011) Purges (18-25) -- 0.021 -- 0.035* -- 0.019 (0.015) (0.018) (0.015) Riots (18-25) -- -0.003 -- -0.000 -- -0.001 (0.004) (0.006) (0.003) Revolutions (18-25) -- 0.014 -- -0.006 -- 0.019* (0.013) (0.014) (0.011) Anti-gov. Demons. (18-25) -- -0.002 -- -0.001 -- -0.001 (0.002) (0.002) (0.002) GDP Growth (18-25) -- 0.001 -- 0.002 -- 0.001 (0.002) (0.002) (0.001) GDP Per Ca pita (18-25) -- -0.000 -- 0.000* -- -0.000 (0.000) (0.000) (0.000) Infla tion (18-25) -- 0.000 -- 0.000 -- 0.000 (0.000) (0.000) (0.000) Polity (18-25) -- -0.001 -- -0.001 -- 0.001 (0.002) (0.002) (0.002) Observations R2 429204 0.134 429204 0.134 398284 0.123 398284 0.123 415441 0.159 415441 0.159 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018, EM-DAT International Disaster Database, 1970-2017, and CNTS 1970-2017. 73 Appendix Table B.4: Robustness to Controlling for Other Economic and Political Shocks (Ages 10-17) (1) (2) (3) (4) (5) (6) Outcome ➔ Have confidence in national government Have confidence in national government Approval of the leader Approval of the leader Have confidence in honesty of elections Have confidence in honesty of elections Exposure to Epidemic (18-25) -3.478*** -2.205* -5.000*** -3.627*** -4.496*** -3.839*** (1.182) (1.153) (0.813) (1.040) (1.132) (1.002) The number of people affected t-1 0.795 1.060 0.426 0.315 -3.149* -3.017** (3.111) (2.672) (2.351) (1.957) (1.667) (1.258) Government strength (10-17) -- 0.002 -- -0.017** -- 0.010 (0.007) (0.008) -0.007 Socioeconomic conditions (10-17) -- -0.010 -- 0.006 -- -0.011 (0.009) (0.012) -0.008 Investment profile (10-17) -- -0.005 -- -0.002 -- -0.012 (0.009) (0.012) -0.008 Internal conflict (10-17) -- -0.003 -- -0.003 -- -0.011* (0.007) (0.007) -0.006 External conflict (10-17) -- -0.008 -- -0.019*** -- -0.002 (0.006) (0.007) -0.006 Corruption (10-17) -- -0.009 -- -0.015 -- -0.015 (0.015) (0.015) -0.015 Military in politics (10-17) -- 0.035* -- 0.034* -- 0.016 (0.014) (0.017) -0.012 Religious tensions (10-17) -- -0.036** -- -0.051** -- -0.034** (0.017) (0.020) -0.015 Law and order (10-17) -- 0.037** -- 0.059*** -- 0.049*** (0.019) (0.022) -0.016 Ethnic tensions (10-17) -- 0.015 -- 0.033** -- 0.012 (0.011) (0.016) -0.012 Democratic accountability (10-17) -- 0.001 -- -0.007 -- 0.004 (0.013) (0.016) -0.012 Bureaucracy quality (10-17) -- -0.036* -- -0.048** -- -0.03 (0.019) (0.024) -0.019 Observations R2 274953 0.135 274953 0.137 257901 0.113 257901 0.116 268600 0.135 268600 0.137 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018, EM-DAT International Disaster Database, 1984-2017, and ICRG 1984-2017. 74 Appendix Table B.5: Robustness to Controlling for Other Economic and Political Shocks (Ages 10-17) (1) (2) (3) (4) (5) (6) Outcome ➔ Have confidence in national government Have confidence in national government Approval of the leader Approval of the leader Have confidence in honesty of elections Have confidence in honesty of elections Exposure to Epidemic (18-25) -1.622*** -1.639*** -2.465*** -2.811*** -2.657*** -2.748*** (0.349) (0.537) (0.419) (0.596) (0.277) (0.430) The number of people affected t-1 3.236** 3.230*** 1.501 1.378 -2.348*** -2.277*** (1.254) (1.197) (1.279) (1.205) (0.647) (0.645) Assassinations (10-17) -- 0.006 -- 0.016 -- 0.012** (0.010) (0.013) (0.005) General Strikes (10-17) -- 0.028** -- 0.047*** -- 0.022** (0.013) (0.012) (0.010) Terror./Guerrilla Warfare (10-17) -- -0.042* -- -0.061** -- -0.004 (0.025) (0.027) (0.022) Purges (10-17) -- 0.012 -- 0.010 -- 0.02 (0.022) (0.021) (0.019) Riots (10-17) -- -0.001 -- -0.014 -- -0.005 (0.006) (0.008) (0.005) Revolutions (10-17) -- -0.054*** -- -0.039* -- -0.037** (0.019) (0.022) (0.015) Anti-gov. Demons. (10-17) -- -0.005 -- 0.003 -- 0.001 (0.007) (0.005) (0.005) GDP Growth (10-17) -- 0.003 -- 0.004 -- 0.004* (0.002) (0.003) (0.002) GDP Per Ca pita (10-17) -- -0.000 -- 0.000 -- -0.000 (0.000) (0.000) (0.000) Infla tion (10-17) -- 0.000 -- 0.000 -- 0.000 (0.000) (0.000) (0.000) Polity (10-17) -- -0.001 -- -0.004 -- -0.003 (0.002) (0.003) (0.002) Observations R2 315587 0.126 315587 0.127 293751 0.116 293751 0.117 306094 0.158 306094 0.159 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018, EM-DAT International Disaster Database, 1970-2017, and CNTS 1970-2017. 75 Appendix Table B.6: Robustness to Controlling for Other Economic and Political Shocks and Country*Year Fixed Effects (1) (2) (3) (4) (5) (6) Outcome ➔ Have confidence in national government Have confidence in national government Approval of the leader Approval of the leader Have confidence in honesty of elections Have confidence in honesty of elections Exposure to Epidemic (18-25) -0.613** -0.577** -0.502** -0.529** -1.269*** -1.293*** (0.253) (0.286) (0.197) (0.259) (0.191) (0.192) Government strength (18-25) -- 0.002 -- 0.006*** -- 0.002 (0.002) (0.002) (0.002) Socioeconomic conditions (18-25) -- -0.002 -- -0.001 -- -0.003 (0.002) (0.002) (0.002) Investment profile (18-25) -- 0.002 -- 0.002 -- 0.001 (0.002) (0.002) (0.002) Internal conflict (18-25) -- -0.002 -- -0.001 -- 0.003 (0.002) (0.002) (0.002) External conflict (18-25) -- 0.001 -- 0.002 -- 0.002 (0.002) (0.002) (0.002) Corruption (18-25) -- -0.005* -- -0.003 -- -0.003 (0.003) (0.003) (0.003) Military in politics (18-25) -- -0.002 -- -0.000 -- 0.002 (0.003) (0.003) (0.003) Religious tensions (18-25) -- 0.002 -- 0.007** -- -0.003 (0.003) (0.003) (0.004) Law and order (18-25) -- 0.003 -- -0.004 -- 0.006 (0.004) (0.004) (0.004) Ethnic tensions (18-25) -- 0.002 -- 0.000 -- -0.002 (0.003) (0.002) (0.003) Democratic accountability (18-25) -- -0.002 -- 0.001 -- -0.009*** (0.002) (0.003) (0.003) Bureaucracy quality (18-25) -- 0.009 -- 0.011* -- 0.009* (0.006) (0.006) (0.005) Observations R2 422523 0.174 422523 0.174 408564 0.166 408564 0.166 412051 0.170 412051 0.170 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 5 of Table 2 country*year fixed effects. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Sour ce: Gallup World Polls, 2006-2018, EM-DAT International Disaster Database, 1984-2017, and ICRG 1984-2017. 76 Appendix Table B.7: Robustness to Controlling for Other Economic and Political Shocks and Country*Year Fixed Effects (1) (2) (3) (4) (5) (6) Outcome ➔ Have confidence in national government Have confidence in national government Approval of the leader Approval of the leader Have confidence in honesty of elections Have confidence in honesty of elections Exposure to Epidemic (18-25) -0.630*** -0.607*** -0.765*** -0.623*** -1.346*** -1.198*** (0.184) (0.217) (0.158) (0.200) (0.159) (0.205) Assassinations (18-25) -- -0.001 -- 0.000 -- -0.004 (0.003) (0.002) (0.003) General Strikes (18-25) -- 0.002 -- -0.000 -- -0.003 (0.004) (0.005) (0.004) Terror./Guerrilla Warfare (18-25) -- -0.002 -- -0.006 -- -0.015*** (0.006) (0.004) (0.005) Purges (18-25) -- 0.025* -- 0.025 -- 0.007 (0.013) (0.018) (0.016) Riots (18-25) -- -0.003 -- 0.000 -- -0.001 (0.002) (0.002) (0.002) Revolutions (18-25) -- 0.016** -- 0.009 -- 0.021*** (0.007) (0.007) (0.007) Anti-gov. Demons. (18-25) -- 0.001 -- -0.001 -- 0.001 (0.001) (0.001) (0.001) GDP Growth (18-25) -- 0.000 -- 0.001** -- 0.000 (0.001) (0.001) (0.001) GDP Per Ca pita (18-25) -- -0.000 -- 0.000** -- 0.000 (0.000) (0.000) (0.000) Infla tion (18-25) -- 0.000 -- 0.000 -- 0.000 (0.000) (0.000) (0.000) Polity (18-25) -- -0.001 -- 0.000 -- 0.001 (0.001) (0.001) (0.001) Observations R2 429204 0.134 429204 0.170 398284 0.171 398284 0.171 415441 0.192 415441 0.192 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 5 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018, EM-DAT International Disaster Database, 1970-2017, and CNTS 1970-2017. 77 Appendix Table B.8: Robustness to Omitted Variables Bias (1) (2) (3) Outcome variable ➔ Ha ve confidence in na tional government Approval of the Leader Ha ve confidence in honesty of elections Panel A: Estimation model: Columns 2, 4 and 6 of Appendix Table B.2, which controls for various past economic and political shocks Exposure to Epidemic (18-25) -3.417*** (0.787) -3.944*** (0.746) -4.219*** (0.849) Bounds on the treatment effect (δ=1, Rmax=1.3*R) (-3.417, -3.844) (-3.944, -4.120) (-4.219, -4.635) Trea tment effect excludes 0 Yes Yes Yes Delta (Rmax=1.3*R) 11.60 24.24 19.02 Panel B: Estimation model: Columns 2, 4 and 6 of Appendix Table B.3, which controls for various past economic and political shocks Exposure to Epidemic (18-25) -1.743*** (0.632) -2.204*** (0.576) -2.185*** (0.544) Bounds on the treatment effect (δ=1, Rmax=1.3*R) (-1.743, -1.943) (-2.204, -2.317) (-2.185, -2.556) Trea tment effect excludes 0 Yes Yes Yes Delta (Rmax=1.3*R) 12.72 21.34 12.34 Notes: * significa nt a t 10%; ** significant a t 5%; *** significant a t 1%. Bounds on the Exposure to Epidemic (18 -25) effect are ca lculated using Stata code psa calc, which calculates estimates of treatment effects a nd relative degree of selection in linea r models a s proposed in Oster (2019). Delta, δ, ca lculates a n estimate of the proportional degree of selection given a maximum value of the R -squared. Rmax specifies the maximum R-squared which would result if a ll unobservables were included in the regressio n. We define Rmax upper bound as 1.3 times the R -squared from the main specification tha t controls for a ll observables. Oster’s delta indicates the degree of selection on unobservables relative to observables t hat would be needed to fully expla in our results by omitted variable bias. Results use the Gallup sampling weights and robust standard errors a re clustered a t the country level. Source: Ga llup World Polls, 2006-2018 and EM-DAT International Disa ster Database, 1970-2017. 78 Appendix Table B.9: Placebo Outcomes (1) (2) (3) (4) (5) Outcome ➔ Have confidence in the military Have confidence in banks Have confidence in media Have relatives or friends to count on Have helped to a stranger Exposure to epidemic (18-25) -0.542 0.147 -0.652 0.290 0.021 (0.442) (0.193) (0.610) (0.851) (0.281) The number of people affected t-1 2.210 0.118 -10.208** -1.134** -1.390 (3.284) (2.038) (4.817) (0.456) (1.796) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income Yes Yes Yes Yes Yes Demogra phic characteristics Yes Yes Yes Yes Yes Income decile fixed effects Yes Yes Yes Yes Yes La bor market controls Yes Yes Yes Yes Yes Country*Age trends Yes Yes Yes Yes Yes Cohort fixed effects Yes Yes Yes Yes Yes Observa tions 730156 809972 190167 902066 889981 R2 0.141 0.136 0.104 0.122 0.074 Notes: * significa nt at 10%; ** significant a t 5%; *** significant a t 1%. Outcome is a dummy varia ble indicating tha t the respondent ha s confidence in “milita ry”; “ba nks a nd fina ncial institutions”; “media freedom”. The specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Ga llup World Polls, 2006-2018 a nd EM-DAT International Disaster Da ta base, 1970-2017. 79 Appendix Table B.10: The Impact of Exposure to Epidemic (Ages 18-25) on the Average of All Three Outcome Variables (1) (2) (3) (4) (5) Outcome ➔ Avera ge of a ll three outcome variables Avera ge of a ll three outcome variables Avera ge of a ll three outcome variables Avera ge of a ll three outcome variables Avera ge of a ll three outcome variables Exposure to Epidemic (18-25) -1.365** -1.248** -1.855*** -1.867*** -0.705*** (0.565) (0.539) (0.264) (0.264) (0.155) The number of people affected t-1 -0.854 -0.779 -0.801 -0.803 -- (3.086) (3.065) (3.056) (3.051) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income No Yes Yes Yes Yes Demogra phic characteristics No Yes Yes Yes Yes Income decile fixed effects No Yes Yes Yes Yes La bor market controls No Yes Yes Yes Yes Country*Age trends No No Yes Yes Yes Cohort fixed effects No No No Yes Yes Country*Year fixed effects No No No No Yes Observa tions 636156 636156 636156 636156 636156 R2 0.169 0.178 0.180 0.180 0.230 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Outcome is a n average of all three main dependent variables: “honesty of elections”; “confidence in national government”; “approval of the leader”. Exposure to epidemic (18-25) defined a s the average per ca pita number of people a ffected by an epidemic when the respondent was in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate assistance during a period of emergency (that is, requiring basic survival needs such a s food, wa ter, shelter, sanitation, a nd immediate medical a ssistance). Demographic cha racteristics include: a male dummy, a dummy for ea ch age group, dummy variables for marital sta tus (single, ma rried), educational a ttainment (tertiary education, secondary education), religion dummies (Christia n, Muslim, a nd other religions), employment sta tus (full-time employed, pa rt-time employed, unemployed), a dummy variable for living in a n urban area and presence of children in the household (a ny child under 15). Inc ome decile fixed-effects are constructed by grouping individuals into deciles based on their income relative to other in dividuals within the same country and year. Individual income includes a ll wa ges a nd salaries in the household, remittances from family members living elsewhere, a nd a ll other sources before ta xes . Ga llup converts local income to International Dolla rs using the World Bank’s individual consumption PPP conversion factor, which makes it comparable a cross all countries. Results use the Ga llup sa mpling weights and robust standard errors are clustered a t the country level. Source: Gallup World Polls, 2006 -2018 and EM-DAT International Disaster Da ta base, 1970-2017. 80 Appendix Table B.11: The Impact of Exposure to Epidemic (Ages 18-25) on the 1st Principal Component of Responses (1) (2) (3) (4) (5) Outcome ➔ the 1st Principa l Component of Responses the 1st Principa l Component of Responses the 1st Principa l Component of Responses the 1st Principa l Component of Responses the 1st Principa l Component of Responses Exposure to Epidemic (18-25) -4.672** -4.269** -6.361*** -6.400*** -2.378*** (1.932) (1.841) (0.914) (0.913) (0.531) The number of people affected t-1 -2.619 -2.353 -2.424 -2.431 -- (10.804) (10.730) (10.694) (10.677) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income No Yes Yes Yes Yes Demogra phic characteristics No Yes Yes Yes Yes Income decile fixed effects No Yes Yes Yes Yes La bor market controls No Yes Yes Yes Yes Country*Age trends No No Yes Yes Yes Cohort fixed effects No No No Yes Yes Country*Year fixed effects No No No No Yes Observa tions 636156 636156 636156 636156 636156 R2 0.169 0.178 0.180 0.180 0.230 Notes: * significa nt at 10%; ** significant at 5%; *** significant at 1%. Outcome is the 1st Principa l Component of responses to the main dependent variables: “honesty of elections”; “confidence in na tional government”; “approval of the lea der”. Exposure to epidemic (18-25) defined a s the a verage per ca pita number of people affected by an epidemic when the respondent wa s in their impressionable years (18-25 years). The number of people affected refers to people requiring immediate a ssistance during a period of emergency (that is, requiring ba sic survival needs such a s food, wa ter, shelter, sa ni tation, a nd immediate medical a ssista nce). Demographic characteristics include: a male dummy, a dummy f or ea ch a ge group, dummy variables for marital status (single, married), educational a ttainment (tertiary education, secondary education), religion dummies (Christian, Muslim, a nd other religions), employment s tatus (full-time employed, part- time employed, unemployed), a dummy variable for living in a n urban a rea a nd presence of children in the household (a ny child under 15). Income decile fixed- effects a re constructed by grouping individuals into deciles based on their income rela tive to other individuals within the sa me country a nd year. Individual income includes all wa ges a nd sa laries in the household, remittances from family members living elsewhere, a nd a ll other sour ces before taxes. Ga llup converts loca l income to International Dolla rs using the World Bank’s individual consumption PPP conversion factor, which makes it comparable a cross a ll countries. Results use the Ga llup sampling weights a nd robust standard errors a re clustered at the country level. Source: Ga llup World P olls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 81 Appendix Table B.12: Robustness to Using Comparable Samples (i.e. sample of individuals who have responded to all 7 questions) (1) (2) (3) (4) (5) (6) (7) Outcome ➔ Have confidence in national government Approval of the Leader Have confidence in honesty of elections Have confidence in the military Have confidence in the banks Have relatives or friends to count on Have helped to a stranger The number of people affected (18-25) -0.570** -0.420*** -1.282*** -0.374 0.598** 0.454 -0.095 (0.242) (0.112) (0.224) (0.291) (0.249) (0.577) (0.239) Observa tions 558299 558299 558299 558299 558299 558299 558299 Country fixed effects Yes Yes Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Yes Yes Individual income Yes Yes Yes Yes Yes Yes Yes Demogra phic characteristics Yes Yes Yes Yes Yes Yes Yes Income decile fixed effects Yes Yes Yes Yes Yes Yes Yes La bor market controls Yes Yes Yes Yes Yes Yes Yes Country*Age trends Yes Yes Yes Yes Yes Yes Yes Cohort fixed effects Yes Yes Yes Yes Yes Yes Yes Country*Year fixed effects Yes Yes Yes Yes Yes Yes Yes Notes: * significa nt a t 10%; ** significant a t 5%; *** significant at 1%. Results use the Ga llup sampling weights a nd robust standard errors a re clustered at the country level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 82 Appendix Table B.13: Robustness to Alternative Epidemic Exposure Measure - Exposure to SARS, H1N1, MERS, Ebola, or Zika (1) (2) (3) (4) (5) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Coefficient on Exposure to Epidemic (18-25) (sta ndard error) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Avera ge of a ll three outcome variables the 1st Principa l Component of Responses Sa mple: Democratic countries -0.022 (0.020) -0.044*A (0.024) -0.041**A (0.017) -0.038** (0.019) -0.132**A (0.066) Observa tions R2 106530 0.137 102838 0.108 103551 0.135 94695 0.171 94695 0.171 Sa mple: Non-democratic countries 0.029 (0.021) 0.029* (0.016) 0.022 (0.022) 0.030* (0.016) 0.104* (0.056) Observa tions R2 47796 0.187 44273 0.183 45566 0.192 37849 0.254 37849 0.253 Notes: * significa nt a t 10%; ** significant at 5%; *** significant a t 1%. Exposure to epidemic (18 -25) takes a value of 1 if the respondent experienced SARS, H1N1, MERS, Ebola , or Zika when the respondent was in their impressionable years (18-25 years). Specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. A indica tes statistically significant difference in ea ch pair of means a t p<.05. Source: Ga llup World Polls, 2006-2018 a nd Ma et a l., 2020. 83 Appendix Table B.14: Contemporaneous Effects of Pandemic on Political Trust (1) (2) (3) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections La gged pandemic -0.028* -0.037** -0.015 (0.016) (0.018) (0.018) Country fixed effects Yes Yes Yes Yea r fixed effects Yes Yes Yes Age group fixed effects Yes Yes Yes Individual income Yes Yes Yes Demogra phic characteristics Yes Yes Yes Income decile fixed effects Yes Yes Yes La bor market controls Yes Yes Yes Country*Age trends Yes Yes Yes Cohort fixed effects Yes Yes Yes Country*Year fixed effects Yes Yes Yes Observa tions R2 987864 0.142 931469 0.131 950827 0.147 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Equation 2. Results use the Gallup sampling weights a nd robust standard errors a re clustered at the country level. Source: Gallup World Polls, 2006-2018 a nd Ma et a l., 2020. 84 Appendix Table B.15: Impact of Communicable and Non-Communicable Diseases on the Political Trust (1) (2) (3) Sa mple ➔ Full-sa mple Democratic countries Non-democratic counties Outcome ➔ Ha ve confidence in na tional government Ha ve confidence in na tional government Ha ve confidence in na tional government Exposure to communicable dis. (18-25) -0.368** (0.152) -0.426** (0.213) -0.054 (0.209) Exposure to non-communicable dis. (18-25) 0.175 (0.303) 0.132 (0.407) 0.037 (0.373) Observa tions R2 389882 0.157 267544 0.125 109651 0.182 Outcome ➔ Approval of the lea der Approval of the lea der Approval of the lea der Exposure to communicable dis. (18-25) -0.111 (0.179) -0.152 (0.263) -0.043 (0.252) Exposure to non-communicable dis. (18-25) 0.123 (0.336) 0.125 (0.545) 0.184 (0.369) Observa tions R2 370749 0.140 256154 0.099 100751 0.177 Outcome ➔ Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Exposure to communicable dis. (18-25) -0.515*** (0.176) -0.533** (0.243) -0.032 (0.207) Exposure to non-communicable dis. (18-25) 0.553* (0.305) 0.525 (0.379) 0.191 (0.373) Observa tions R2 377838 0.147 259328 0.130 106387 0.194 Notes: * significa nt at 10%; ** significant a t 5%; *** significant at 1%. Exposure to communicable diseases (18-25) ta kes a value of 1 if the respondent experienced communicable diseases (dia rrhea, lower respira tory, other common infectious diseases, malaria & neglected tropical diseases, HIV/AIDS, tuberculosis, other communicable diseases). Exposure to non-communicable diseases (18-25) ta kes a value of 1 if the respondent experienced non-communicable disea ses (ca rdiovascular diseases, ca ncers, respira tory disease, dia betes, blood a nd endocrine diseases, mental and substance use disorders, liver diseases, digestive diseases, musculoskeletal disorders, neurological disorders, other non- communicable diseases). Both measures a re population-adjusted a nd expressed in terms of Disability Adjusted Life Years Lost (DALYs), which is a sta ndardized metric a llowing for direct comparison a nd summing of burdens of different diseases. Conceptually, one DALY is the equivalent of one year in good health lost due to premature mortality or disa bility. Specification is Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. A indica tes statistically significant difference in each pair of means a t p<.05. Source: Gallup World Polls, 2006-2018 a nd Institute for Health Metrics a nd Evaluation, 1990 -2016 85 Appendix Table B.16: The Impact of Exposure to Epidemic (Ages 18-25) on Political Trust by Exposure Thresholds (1) (2) (3) Coefficient on Dummy Va riable (sta ndard error) Coefficient on Dummy Va riable (sta ndard error) Coefficient on Dummy Va riable (sta ndard error) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Ba seline - Exposure to Epidemic (18-25) -1.592*** (0.262) -1.957*** (0.330) -2.258*** (0.339) Top 0.5 per cent (exposure to epidemic, 18-25) -0.144*** (0.041) -0.131*** (0.038) -0.147*** (0.054) Top 1 per cent (exposure to epidemic, 18-25) -0.097** (0.038) -0.084** (0.040) -0.112*** (0.034) Top 2 per cent (exposure to epidemic, 18-25) -0.054** (0.024) -0.051** (0.023) -0.061*** (0.023) Top 5 per cent (exposure to epidemic, 18-25) 0.001 (0.016) -0.007 (0.021) -0.014 (0.014) Notes: * significa nt a t 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. Results reported in ea ch panel come from sepa rate models. Threshold dummies in ea ch row a re defined based on the continuous treatment variable (Exposure to Epidemic, 18-25). See notes to Ta ble 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Ga llup World Polls, 2006-2018 and EM-DAT International Disa ster Database, 1970-2017. 86 Table B.17: The Role of Democracy at the Time of the Epidemic (1) (2) (3) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Exposure to epidemic (18-25) * Democracy (18-25) -4.199** -3.624 -3.379** (1.685) (3.143) (1.592) Exposure to epidemic (18-25) -1.504*** -2.112*** -2.110*** (0.420) (0.419) (0.406) Democracy (18-25) 0.007 -0.003 0.015 (0.010) (0.011) (0.010) Observa tions 523072 489155 504686 R2 0.140 0.127 0.154 Notes: * significa nt a t 10%; ** significant a t 5%; *** significa nt a t 1%. The specification is Equation 3. See Ta bles 2 -3-4 for va riable definitions. Results reported in each column come from separate models. Results use the Gallup sa mpling weights a nd robust standard errors are clustered at the country level. Source: Ga llup World Polls, 2006-2018, EM-DAT International Disa ster Da tabase, 1970-2017, a nd the Polity5 dataset. 87 Appendix Table B.18: Impact of Exposure to Epidemics (Ages 18-25) on Political Trust – Intensive and Extensive Margins (1) (2) (3) (4) (5) (6) Intensive margin Intensive margin Intensive margin Extensive margin Extensive margin Extensive margin Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Exposure to Epidemic (18-25) -2.779*** -3.241*** -3.329*** -0.001 -0.009*** 0.001 (0.519) (0.735) (0.505) (0.003) (0.003) (0.003) The number of people affected t-1 -0.004 -0.450 -3.463 0.773 0.138 -3.574 (4.959) (4.043) (2.779) (3.457) (2.718) (2.182) Observa tions R2 351733 0.138 340226 0.119 342209 0.133 760099 0.145 719742 0.133 736679 0.146 Notes: * significa nt a t 10%; ** significant at 5%; *** significant a t 1%. For intensive margin, the sa mple is restricted to respondents with a ny epidemic experience in their impressionable years, a nd models a re re-estimated a s in Column 4 of Ta ble 2 . For extensive margin, Exposure to Epidemic (18-25) is re- defined a s a dummy taking the value of 1 when the continuous version is positive a nd zero otherwise ; a nd models a re re-estimated over the full sa mple as in Column 4 of Table 2. See notes to Table 2. Results use the Gallup sampling weights and robust standard errors are clustered at the country level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 88 Appendix Table B.19: Impact of “Made-up” Exposure on Immigrants’ Political Trust (1) (2) (3) (4) (5) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Avera ge of a ll three outcome variables the 1st Principa l Component of Responses Exposure to epidemic (18-25) -0.919 -5.915 -0.205 -1.475 -5.229 (2.100) (3.601) (2.639) (1.688) (5.994) The number of people affected t-1 -10.238 -13.867 -13.788 -6.929 -24.679 (15.302) (15.535) (16.258) (11.686) (41.658) Observa tions 4639 4306 4118 3611 3611 R2 0.229 0.229 0.282 0.322 0.321 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. Exposure to epidemic (18-25) defined as the average per ca pita number of people a ffected by an epidemic when the respondent wa s in their impressionable years (18 -25 years). The number of people affected refers to people requiring immediate a ssistance during a period of emergency (that is, requiring ba sic survival needs such a s food, wa ter, she lter, sa nitation, a nd immediate medical a ssista nce). Demographic characteristics include: a male dummy, a dummy for each a ge group, dummy variables for marital statu s (single, married), educational a ttainment (tertia ry education, secondary education), religion dummies (Christ ia n, Muslim, a nd other religions), employment status (full-time employed, pa rt-time employed, unemployed), a dummy variable for living in a n urban a rea a nd presence of children in the household (a ny child under 15). Income decile fixed-effects a re constructed by grouping individuals into deciles ba sed on their income rela tive to other individuals within the same country and year. Indiv idual income includes a ll wa ges and salaries in the household, remittances from family members living elsewhere, a nd all oth er sources before taxes. Gallup converts local income to International Dolla rs using the World Ba nk’s individual consumption PPP conversion factor, which makes it comparable a cross all countries. Results use the Ga llup s ampling weights and robust standard errors a re clustered at the country level. Source: Ga llup World Polls, 2006 -2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017. 89 Appendix Table B.20: Impact of “Randomly-Assigned” Exposure on Political Trust (1) (2) (3) (4) (5) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Avera ge of a ll three outcome variables the 1st Principa l Component of Responses Exposure to epidemic (18-25) 0.210 -0.250 -0.238 -0.040 -0.109 (0.390) (0.488) (0.439) (0.389) (1.348) The number of people affected t-1 0.734 0.320 -3.609* -0.625 -1.802 (3.450) (2.660) (2.157) (2.996) (10.483) Observa tions 668022 632661 647417 559274 559274 R2 0.146 0.133 0.145 0.180 0.180 Notes: * significant at 10%; ** significant at 5%; *** significant at 1%. Specification is Column 4 of Table 2. Exposure to epidemic (18-25) defined as the average per ca pita number of people a ffected by an epidemic when the respondent wa s in their impressionable years (18 -25 years). The number of people affected refers to people requiring immediate a ssistance during a period of emergency (that is, requiring ba sic survival needs such a s food, wa ter, she lter, sa nitation, a nd immediate medical a ssista nce). Demographic characteristics include: a male dummy, a dummy for each a ge group, dummy variables for marital statu s (single, married), educational a ttainment (tertia ry education, secondary education), religion dummies (Christ ia n, Muslim, a nd other religions), employment status (full-time employed, pa rt-time employed, unemployed), a dummy variable for living in a n urban a rea a nd presence of children in the household (a ny child under 15). Income decile fixed-effects a re constructed by grouping individuals into deciles ba sed on their income rela tive to other individuals within the same country and year. Indiv idual income includes a ll wa ges and salaries in the household, remittances from family members living elsewhere, a nd all oth er sources before taxes. Gallup converts local income to International Dolla rs using the World Ba nk’s individual consumption PPP conversion factor, which makes it comparable a cross all countries. Results use the Ga llup s ampling weights and robust standard errors a re clustered at the country level. Source: Ga llup World Polls, 2006 -2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017. 90 Appendix Table B.21: Multiple Hypothesis Testing (1) (2) (3) Outcome ➔ Ha ve confidence in na tional government Approval of the lea der Ha ve confidence in honesty of elections Exposure to epidemic (18-25) -1.592*** -1.957*** -2.258*** (0.262) (0.330) (0.339) The number of people affected t-1 0.740 0.120 -3.625* (3.452) (2.712) (2.182) Country fixed effects Yes Yes Yes Yea r fixed effects Yes Yes Yes Age group fixed effects Yes Yes Yes Individual income Yes Yes Yes Demogra phic characteristics Yes Yes Yes Income decile fixed effects Yes Yes Yes La bor market controls Yes Yes Yes Country*Age trends Yes Yes Yes Cohort fixed effects Yes Yes Yes Observa tions 760099 719742 736679 R2 0.145 0.133 0.146 Mea n of outcome 0.50 0.51 0.51 Ra ndomization-c p-values 0.020** 0.007*** 0.007*** Ra ndomization-t p-values 0.006*** 0.007*** 0.007*** Ra ndomization-c p-values (joint test of treatment significance) 0.008*** Ra ndomization-t p-values (joint test of treatment significance) N/A Ra ndomization-c p-values (Westfall-Young multiple testing of treatment significance) 0.013** Ra ndomization-t p-values (Westfall-Young multiple testing of treatment significance) 0.003*** Notes: * significa nt a t 10%; ** significant a t 5%; *** significa nt a t 1%. Randomization-t technique does not produce p-values for the joint test of trea tment significance. Results a re derived from 100 iterations. Specification is Column 4 of Ta bles 2 -3-4. Results use the Ga llup sa mpling weights a nd robust standard errors a re clustered at the country level. Source: Ga llup World Polls, 2006-2018 a nd EM-DAT International Disa ster Da tabase, 1970-2017 91 Appendix Table B.22: Robustness to Excluding Potentially Bad Controls (1) (2) (3) (4) Outcome ➔ Ha ve confidence in na tional government Ha ve confidence in na tional government Ha ve confidence in na tional government Ha ve confidence in na tional government Exposure to epidemic (18-25) -1.073* -1.733*** -1.728*** -0.506** (0.594) (0.262) (0.258) (0.223) The number of people affected t-1 0.548 0.576 0.581 -- (3.478) (3.453) (3.450) Observa tions 760099 760099 760099 760099 Outcome ➔ Approval of the Leader Approval of the Leader Approval of the Leader Approval of the Leader Exposure to epidemic (18-25) -1.521*** -1.933*** -1.991*** -0.580*** (0.380) (0.313) (0.316) (0.123) The number of people affected t-1 0.201 0.177 0.151 -- (2.696) (2.675) (2.679) Observa tions 719742 719742 719742 719742 Outcome ➔ Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Ha ve confidence in honesty of elections Exposure to epidemic (18-25) -1.643** -2.322*** -2.367*** -1.117*** (0.794) (0.362) (0.355) (0.255) The number of people affected t-1 -3.734* -3.775* -3.754* -- (2.203) (2.211) (2.198) Observa tions 736679 736679 736679 736679 Country fixed effects Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Country*Age trends No Yes Yes Yes Cohort fixed effects No No Yes Yes Country*Year fixed effects No No No Yes Notes: * significa nt at 10%; ** significant at 5%; *** significant at 1%. Results use the Ga llup sampling weights a nd robust standard errors a re clustered at the country level. Source: Ga llup World Polls, 2006 -2018 and EM-DAT International Disa ster Database, 1970-2017. 92 Appendix Table B.23: Robustness to Alternative Treatment (i.e., Population Unadjusted Number of Affected People) (1) (2) (3) (4) (5) Outcome ➔ Ha ve confidence in the government Approval of the Leader Ha ve confidence in honesty of elections Avera ge of a ll three outcome variables the 1st Principa l Component of Responses Exposure to epidemic (18-25) -0.081*** -0.100** -0.090*** -0.091*** -0.313*** (0.029) (0.043) (0.014) (0.030) (0.105) The number of people affected t-1 0.139** 0.223*** 0.035 0.136*** 0.479*** (0.060) (0.068) (0.039) (0.048) (0.170) Country fixed effects Yes Yes Yes Yes Yes Yea r fixed effects Yes Yes Yes Yes Yes Age group fixed effects Yes Yes Yes Yes Yes Individual income Yes Yes Yes Yes Yes Demogra phic characteristics Yes Yes Yes Yes Yes Income decile fixed effects Yes Yes Yes Yes Yes La bor market controls Yes Yes Yes Yes Yes Country*Age trends Yes Yes Yes Yes Yes Cohort fixed effects Yes Yes Yes Yes Yes Observa tions 770836 731758 746610 644795 644795 R2 0.149 0.135 0.146 0.184 0.184 Notes: * significa nt at 10%; ** significant at 5%; *** significant a t 1%. Results use the Gallup sampling weights a nd robust standard errors a re clustered at the country level. Source: Gallup World Polls, 2006-2018 and EM-DAT International Disa ster Da tabase, 1970-2017. 93 Appendix Table B.24: Evidence on Political Behaviour (1) (2) (3) (4) Outcome is ➔ WWS - Attending la wful/peaceful demonstrations WWS – Never voted in na tional elections ESS - Ta ken part in a la wful public demonstration ESS - Voted in recent na tional elections Exposure to epidemic (18-25) 16.412* (9.736) 5.488 (7.014) 53.041** (12.811) -134.497** (59.276) The number of people affected t-1 -14.926 (19.588) -0.005 (0.011) 10.109 (127.553) -270.948** (116.562) Observa tions 103681 32448 171889 128836 R2 0.127 0.101 0.051 0.110 Outcome is ➔ WWS - Signed a petition WWS - Joined in boycotts WWS – Occupied buildings or fa ctories WWS - Joined unofficial strikes Exposure to epidemic (18-25) 18.944** (7.811) 19.322** (9.176) -2.481 (5.330) -4.982 (8.972) The number of people affected t-1 -16.000 (25.386) -1.362 (18.196) -7.416 (13.027) 21.980 (15.969) Observa tions 103851 101088 39440 71851 R2 0.226 0.198 0.081 0.132 Notes: * significa nt a t 10%; ** significant a t 5%; *** significant a t 1%.. Exposure to epidemic (18-25) defined as the average per capita number of people a ffected by an epidemic when the respondent wa s in their impressionable years (18 -25 years). The number of people a ffected refers to people requiring immediate a ssistance during a period of emergency (that is, requiring ba sic survival needs such as food, wa ter, shelter, sa nitation, a nd immediate med ical a ssista nce). Demographic characteristics include: a male dummy, a dummy for ea ch a ge group, dummy variables for marital sta tu s (single, ma rried), educational a ttainment (tertiary education, secondary education), religion dummies (Christian, Muslim, a nd other religions), employment status (full-time employed, part-time employed, unemployed), a dummy varia ble for living in a n urban area and presence of children in the household (any child under 15). Income decile fixed-effects are constructed by grouping individuals into deciles based on their income rela tive to other individuals within the s ame country a nd year. Results use the sampling weights and robust standard errors a re clustered a t the country -wave level. Source: World Va lues Survey (WVS), 1981- 2014; European Social Survey (ESS), 2002-2018); and EM-DAT International Disa ster Da tabase, 1970-2017. 94 Appendix C: Case Studies on the Association of Government Strength with Policy Interventions in the Context of COVID-19 Appendix Figures C.1-C.3 show COVID-19 related developments in South Korea, France, and the United Kingdom. We choose these countries because they followed very different trajectories in terms of public attention, policy interventions, and the spread of the virus. South Korea, France, and the United Kingdom are broadly similar in terms of their GDP per capita, urbanization, and population age structure (median age in all three countries is roughly 41). But they differ in terms of government strength: the ICRG score is 8.25 for Sou th Korea, 7.5 for France, and 6 for the United Kingdom.69 The figures show the number of confirmed COVID-19 cases and deaths, public attention to COVID-19 as measured by Google Trends, and the date of the first non -pharmaceutical intervention (school closure, workplace closure, public event cancellation, public transport closure, or restrictions on within-country movement in the own country). We also report the number of days between the date of the first confirmed case and the date of the first COVID- 19 non-pharmaceutical intervention. In South Korea, public attention rose rapidly after the first domestic case. The government responded within 11 days of the first case with domestic interventions aimed at curbing the epidemic. In France and the UK, in contrast, public attention remained low for several weeks after the first reported case. In France, domestic restrictions were imposed only after 36 days, while the UK government waited 45 days before imposing the first restrictions. These slow reactions were associated with rapid growth in confirmed cases and deaths in both countries. Simple comparisons among countries are complicated by the existence of other influences, such as past exposure to epidemics.70 Still, these comparisons are suggestive of the idea that government strength is positively associated with the speed of response to the outbreak. 69 The rela tively low score for the UK may come a s a surprise to readers but it is worth noting that: (i) it registered a significa nt fall since the Brexit Referendum (8.46 wa s the 2015 score); (ii) ICRG’s government strength score include points for government unity, legisla tive strength a nd popular support. That the UK ha s had minority and coa lition governments may therefore a ccount for its ra nking. Recent anecdotal evidence a lso reflects the low government strength score of the UK. For exa mple, As the Economist wrote in June, 2020: “The painful conclusion is that Britain has the wrong sort of government for a pandemic—and, in Boris Johnson, the wrong sort of prime minister. Bea ting the coronavirus calls for a ttention to detail, consistency a nd implementation, but they a re not his forte.” See: https://www.economist.com/leaders/2020/06/18/britain-has-the-wrong-government-for-the-covid-crisis 70 Thus, it ha s been suggested that Asian countries responded quickly because of their past experience with Avian flu. 95 Appendix Figure C.1: COVID-19 Related Developments in South Korea ICRG Government Strength score: 8.25 Note: This figure shows daily measures of public a ttention to COVID-19 measured a s the share of Google searchers (left axis) a nd the number of COVID-19 cases and deaths (right a xis), a s well a s the dates of the first ca se, first death, and first policy in South Korea. Source: Google Trends (1/1/2020-31/3/2010), JHCRC (1/1/2020-31/3/2010), a nd ICRG (2018). 96 Appendix Figure C.2: COVID-19 Related Developments in France ICRG Government Strength score: 7.5 Note: This figure shows da ily measures of public a ttention to COVID-19 measured as the share of Google searchers (left a xis) and the number of COVID-19 cases and deaths (right a xis), a s well a s the da tes of the first case, first death, and first policy in Fra nce. Source: Google Trends (1/1/2020-31/3/2010), JHCRC (1/1/2020-31/3/2010), a nd ICRG (2018). 97 Appendix Figure C.3: COVID-19 Related Developments in the United Kingdom ICRG Government Strength score: 6 Note: This figure shows daily measures of public a ttention to COVID-19 measured as the share of Google searchers (left axis) a nd the number of COVID-19 cases a nd deaths (right a xis), a s well a s the da tes of the first case, first death, a nd first policy in the United Kingdom . Source: Google Trends (1/1/2020-31/3/2010), JHCRC (1/1/2020-31/3/2010), a nd ICRG (2018).