key: cord-312721-e6532xrl authors: Ruck, D. J.; Borycz, J.; Bentley, R. A. title: Cultural values predict national COVID-19 death rates date: 2020-07-17 journal: nan DOI: 10.1101/2020.07.17.20156091 sha: doc_id: 312721 cord_uid: e6532xrl National responses to a pandemic require populations to comply through personal behaviors that occur in a cultural context. Here we show that aggregated cultural values of nations, derived from World Values Survey data, have been at least as important as top-down government actions in predicting the impact of COVID-19. Whereas trust in institutions predicts lower COVID-19 deaths per capita, secular- rationalism and cosmopolitanism each predict more deaths. The effects of these cultural values register more strongly than government efficiency. This suggests that open democracies may face greater challenges in limiting a pandemic, and that all nations should consider their cultural values as actionable parameters in their future preparations. Combating the COVID-19 pandemic [1] in nations around the world has depended partly on efficient government response [2, 3, 4, 5, 6] , and partly on the behaviors of individuals [7, 8, 9, 10] . Since culture is the context for behavior [10, 11, 12] , effectiveness of government intervention on COVID-19 ought to reflect public trust in their institutions as well as other cultural values that vary substantially between countries [21, 14] . Here we estimate the observable effects of aggregated cultural attitudes in different countries on their COVID-19 fatality rates. Applying a two-stage factor analysis to World and European Values Survey data [19, 20] we previously derived cultural value factors-including secularrationality (RAT), cosmopolitanism (COS) and institutional trust (INST)-in over one hundred countries [14, 15] . We entered these cultural values into a matrix of country-scale covariates, X, among the 87 counties for which we also have data for government efficiency, from an established index [26, 24] , as well as logarithm of GDP per capita, population size, per cent urban population, and per cent of population aged 65 and over (see Materials and Methods). We first explore the variance structure in the covariate matrix X by principal component analysis, to see how the principal components correlate with per-capita COVID-19 deaths. We then use multivariate regression to explain how the individual covariates predict the residual variance in COVID deaths in the different countries,Ñ d : where the errors ✏ follow a negative binomial distribution and have a variance for a given mean, µ, of µ(1 + µ/r), where r is a dispersion parameter. As Figure 1 shows, the initial spread of COVID-19 was positively correlated with cosmopolitanism, COS (Adjusted r 2 = 0.229, p < 0.0001) and rational-secularism, RAT (Adj. r 2 = 0.217, p < 0.0001). Deaths correlate negatively with institutional confidence, INST (Adj. (Tables S1 and S2) . Per-capita COVID-19 deaths correlate positively with PC1 and negatively with PC2 ( Figure 3 ). This broadly suggests that socio-economic development has predicted more deaths while the effectiveness of institutions has reduced deaths. In between these major components lie the cultural factors of cosmopolitanism and secularrationalism (COS, RAT), which plot on a diagonal between PC1 and PC2 ( Figure 2 ). These two principal components explain 70% of the total variance, and are loaded significantly on each of the seven covariates. Notably, PC4 (7.8% of the variance) is loaded almost entirely on COS (Supplementary Materials). In the Supplement ( Figure S1 ) we show the effects on the PCA using additional, similar variables, such as Government Integrity, Judical Effectiveness, and Business Freedom (from the Heritage Foundation). These variables load primarily onto PC1, with Government Integrity having the biggest loading (Table S1 ). Figure 3 shows how COVID deaths correlate positively with PC 1 (Adj. r 2 = 0.390, p < 0.0001) and negatively PC2 (Adj. Having explored the principal components of variation, a multivariate regression helps disentangle the joint effects of cultural values, government efficiency and economic incentives. The effect of several covariates-COS, INST, GDP and government efficiency-increased between Day 10 and Day 50 (Table S4 ). The z-scores of these effects between 5 and 55 days after the outbreak ( Figure S2 ) reveal a change in covariate effects over the first two months of the outbreak (since deaths are a delayed effect, the timeline may be offset by about two weeks). Both cosmopolitan openness (COS) and and GDP increased in significance, while secular-rationality became less important over the first two months ( Figure S2 ). Institutional confidence (INST) remains near significance level (p = 0.05) for the entire two months, while government efficiency only became significant after the two months ( Figure S2 ). [2, 18] . The cultural value of secular-rationalism (RAT) predicted fewer deaths in the first 10 days of the pandemic (Table 1) , but had lost this effect after two weeks ( Figure S3 ). We could speculate that cultural faith in science may have enabled national governments to to take early preventative action [2] . In the first two months of the outbreak, higher GDP per capita predicted higher COVID-19 death rates (Table 1, Figure S2 ). Although we find evidence, using new data [2] , that higher GDP predicted a stronger and earlier government response to COVID-19 (Table S3) , the weakness of government efficiency in our main regression ( (Table 1) . Finally, a larger question is the resilience of cultures and democracies to unprecedented challenges and 'black swan' events [11] . While trust in institutions predicted fewer COVID-19 deaths and ought to facilitate government action, this value has been declining for decades in many countries [13] . Cultural values of cosmopolitanism and openness, which underlie the eco- Table 1 Cultural values, including secular-humanism (RAT), openness to minorities (COS) and trust in institutions (INST), were derived from multivariate statistics and the World and Euro-S13 pean Values surveys (WEVS) data from 109 nations [19, 20, 21, 13, 14, 15] . The WEVS data are derived from the same 64 questions in the five waves of these surveys at 5-year intervals since 1990, administered to 476,583 participants from 109 different nations. These data were compressed into multivariate factors in two steps. The first used Exploratory Factor Analysis (EFA) to identify nine cultural factors underlying the WEVS data. From the EFA step, we summarized the common variance in the WEVS data and thereby remove the portion of the total variance that is likely to be measurement error or other forms of statistical noise. The second step was to use the EFA factor loadings as weights for Principal Component Analysis (PCA) to derive principal components (PCs) that are orthogonal, which is advantageous for our subsequent regression models. We used the first three of these cultural components in in our analysis here: 'Trust in In- RAT , is correlated with secularism (r = 0.76), political engagement (r = 0.62), respect for individual rights (r = 0.59) and low prosociality (r = 0.45) [13] . This means that secularhumanist respondents to the WEVS are those who reported, for example, that religion is not important in their lives, that they are likely to attend protests or sign petitions, they only pay taxes when coerced and believe that homosexuality and divorce are justifiable [21, 13] . Openness to minorities, COS, is correlated with the exploratory cultural factors for 'openness to out-groups' (r = 0.78), 'openness to norm violators' (r = 0.78) and 'subjective wellbeing' (r = 0.43). High COS implies willingness to have neighbours that are immigrants, from another race, homosexual or from other stigmatized groups; as well as self-reporting a high level of happiness and satisfaction with life [13, 15] Figure S1 : Left: Scree plot for the PCA analysis discussed in the text. Right: Bubble grid showing the relative loadings on the PCA. S15 Figure S2 : Significance of covariate effects on number of cases as the COVID-19 outbreak progresses; where z value = effect size/standard error. Dotted line indicates the z score corresponding with a p value of 0.05 in a two sided test. For principal component analysis (PCA), we use the 'Factominer' and 'Factoextra' packages in R to compute the contributions (Table S1 ) and loadings (Table S2) Table S3 : Results of multiple regression log(Gov.resp d ) =~ X + ✏, where ✏ ⇠ N ormal(0, ) and covariate matrix X contains the covariates from the rows in this table. The vector~ are the effects of covaraites on government response, both 10 and 50 days after outbreak. In parentheses are heteroskedastic standard errors. Heteroskedastic adjusted significance: ⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05, † p < 0.10 Gov. resp., day 10 Gov. resp., day 50 Novel coronavirus (2019-nCoV) situation report -11 Variation in Government Responses to COVID-19, Version 6.0. Blavatnik School of Government Working Paper COVID-19: towards controlling of a pandemic A novel coronavirus emerging in China: Key questions for impact assessment The COVID-19 pandemic in the USA: what might we expect? Making decisions to mitigate COVID-19 with limited knowledge Controlling epidemic spread by social distancing: Do it well or not at all. it BMC Public Health 12 Modelling the influence of human behaviour on the spread of infectious diseases: a review Encouraging sanitation investment in the developing world: A cluster-randomized trial Social networks and health: New developments in diffusion, online and offline Cultural evolution, Covid-19, and preparing for what's next Cultural and institutional factors predicting the infection rate and mortality likelihood of the COVID-19 Pandemic Religious change preceded economic change in the 20th century Cultural prerequisites of socioeconomic development The cultural foundations of modern democracies Data availability Original cultural values data [14] generated for this research work have been archived within the Zenodo repository This research is supported by NSF grant 2028710. DJR was supported by a grant from the College of Arts and Sciences, University of Tennessee.