The role of social capital in COVID-19 deaths

For this study, we used data from open access databases and a survey. We collected COVID-19-related data from the website “COVID-19 Dashboard by the Centre for Systems Science and Engineering (CSSE) at Johns Hopkins University” [2]. This website has complied data from several important sources, such as the World Health Organization (WHO), European Centre for Disease Prevention and Control (ECDC), and WorldoMeters, which have documented COVID-19 case numbers, death numbers, recovered numbers, active case numbers, testing rate, case-fatality ratio and incidence rate from 188 countries by country and province/state. We identified 35,157,350 COVID-19 cases with 1,037,075 deaths at 11.00 AM on October 05,2020 from the CSSE database.

Social capital data were collected from a large-scale web-based and face-to-face survey of 100,956 respondents across all regions/provinces/states of 37 countries in 2017. The web-based approach was predominantly used in this survey to avoid interviewer bias caused by arbitrary factors. We derived questions in the questionnaire based on the sociology and psychology literature. Additionally, we contacted a survey company in each country to translate the questionnaire into the native language and conduct the survey, and we carefully assessed the surveys to ensure the accuracy of responses through translations and multiple checks by native survey conductors. We used the data of a previous year (2017) instead of during the COVID-19 pandemic to measure social capital in each considered country to gain an understanding of society and people’s behaviour in general conditions. Individuals’ responses change due to unexpected incidents and information. Therefore, we used social capital data collected during stable conditions and checked the influence on unexpected risk events, such as COVID-19 deaths. Data regarding population density, hospital bed numbers, and population aged 65 or older were retrieved from the World Development Indicators (WDIs) [36]. Data on country lockdowns were obtained from the website “National responses to the 2019-20 coronavirus pandemic” [37].

The most recent WDI country data was available for 2018. After merging our survey data with COVID-19 data and WDI country-level data, the study sample consisted of 765,875 deaths in 37 countries. Among these countries, 8 countries, China, India, the USA, Indonesia, Brazil, Russia, Mexico, and Japan, were separated into provinces/states because they had the highest populations in the sample. Therefore, the total number of observations was 294. The sample countries and cumulative COVID-19 deaths are described in Supplementary Table S1.

COVID-19 deaths were measured as the number of deaths per square kilometre (km²). The deaths per km² were calculated by dividing the number of deaths by the land area km² of 29 countries and the land area km² of the provinces of the 8 most populous countries in the sample.

Social capital was measured by four factors adopted from previous studies, namely, community attachment, social trust, family bond, and security, and were ordinal variables. The four variables were assessed using a single question based on a previous study conducted among the general public [38]. Community attachment is one of the main independent variables in this analysis. The respondents were asked to measure their attachment to the neighbourhood/community in which they currently reside on a 5-point Likert scale, where 1 = not attached at all, 2 = not really attached, 3 = neither attached nor detached, 4 = slightly attached, and 5 = very attached. The average of all items was created, and a higher score indicated that the respondents displayed slight attachment to the community. Social trust is another main variable, and the respondents were asked to measure the importance they attached to “being able to believe people/organizations”. Answers were rated on a 5-point Likert scale, where 1 = not at all important, 2 = not very important, 3 = neither, 4 = somewhat important, and 5 = very important. The average of all items was created, and the high score indicated that to the respondents, social trust was somewhat important. To measure family bond, another main variable, the respondents were asked to measure their feelings about family relationships in the following question: “Do you feel that the relationship with family is important?” A negative answer was rated as 0, and a positive answer was rated as 1. The average of all items was created, and the high score indicated that the respondents considered the family relationship important. For the final main variable, security, the respondents were asked to measure the safety of their neighbourhood according to the question “Do you think the people in your neighbourhood are safe?” Answers were rated on a 5-point Likert scale, where 1 = do not know, 2 = very dangerous, 3 = slightly dangerous, 4 = moderately safe, and 5 = very safe. The average of all items was created, and the high score indicated that respondents considered that the people in their neighbourhood were not safe. For further details, the survey questionnaire is provided in Supplementary Material 2.

Other explanatory variables are population density, population aged 65 or older, the interaction terms between population aged 65 or older and social capital-related factors, number of hospital beds, and country lockdown. Population density was measured by dividing the total population of each country by the land area km² of 29 countries and the land area km² of the provinces of the 8 most populous countries. The population aged 65 or older was measured by dividing the total population of each country by the land area km² of 29 countries and the total population of the provinces of the 8 most populous countries. The number of beds was measured per 1000 people. Country lockdown was a dummy variable measured by the number of days before a shutdown/lockdown decision was made or a stay-at-home order imposed after the first COVID-19 case was reported in China. If a country was locked down or a stay-at-home order was imposed after 50 days, it was called an early lockdown; if it was locked down after 100 days, it was called a late lockdown; and if a country was not locked down, it was referred to as having no lockdown. We used all the variables in log form except the number of beds and country lockdown to make the data conform more closely to the normal distribution and to improve the model fit.

Stata 16 software was used for all analyses. First, we conducted descriptive statistics to describe the centrality of the variables. Second, we conducted Pearson’s correlation analysis to assess the associations between the variables considered in this study. For the correlation analysis, we employed Cohen’s (1992) standard to determine whether the correlation coefficients were substantial, with r = .01, .03, and .05 representing small, medium, and large effect sizes, respectively [39]. Finally, we employed multiple linear regression to analyse the influence of various independent variables on COVID-19 deaths, including community attachment, social trust, family bond, and security (to measure social capital), population density, population aged 65 or older, community attachment with population aged 65 or older, social trust with population aged 65 or older, family bond with population aged 65 or older, security with population aged 65 or older, and number of beds and country lockdown. In the multiple regression analyses, COVID-19 deaths per km² were the dependent variable, and the main explanatory variable was social capital. We checked the influence of some independent variables on the dependent variable in this paper. We employed the multiple regression model, an econometric model for testing whether explanatory variables significantly influence the dependent variable, in our analysis. We regressed eight multiple linear regression models with four social capital-related variables, and the other explanatory variables were added one by one. Prior to using multiple regression, simple linear regressions were applied to investigate the correlation between COVID-19 deaths and the four variables used to measure social capital. We divided the sample into two groups: all samples and only the provinces/states of 8 countries. The goal was to examine whether the relationship between COVID-19 deaths and social capital-related variables varied by country. The correlation coefficient and p-value of the coefficient for the social capital-related variables were calculated in the analyses of both groups.

Table 1 summarizes the means, SDs, medians and ranges of COVID-19 deaths per km² and the regression covariates. For the 294 observations (29 countries and 264 provinces of 8 countries), the mean score for COVID-19 deaths per km² was 0.09 (SD = 0.54), meaning that the score of COVID-19 deaths per km² was relatively low. Regarding the social capital-related factors, the respondents indicated that they were slightly attached to their community (3.71 out of 5), with a median response of 3.64 (IQR 3.38–3.94), and the respondents felt that being able to trust their society was somewhat important (4.53 out of 5), with a median response of 4.57 (IQR 4.40–4.69). Moreover, the respondents reported that they felt that relationships with family were important (0.90 out of 1), with a median response of 0.92 (IQR 0.86–1), and they thought the people in their neighbourhood were slightly dangerous (2.95 out of 5), with a median response of 2.96 (IQR 2.77–3.21). The results for the mean and median of all four factors of social capital-related factors varied little. Furthermore, the mean score of population density was 540.47 (SD = 1793.92); the mean score of the population aged 65 or older per km² was 42.77 (SD = 146.77); and the mean score of the number of beds per 1000 people was 4.28 (SD = 4.21), which was relatively low compared with the mean score ranging from 0.53 to 13.05. However, the country lockdown policy showed that most of the countries decided to shut down/lockdown or impose a stay-at-home order 50 days after the report of the first COVID-19 case in China (2.04 out of 3).

The relationship between COVID-19 deaths per km² and social capital-related factors is illustrated in Fig. 1. Figure 1a and b demonstrate that COVID-19 deaths per km² were positively and significantly associated with community attachment for all samples (r = 0.27, p = 0.000) and for the provinces of 8 countries (r = 0.24, p = 0.000). Figure 1c and d also displays that the positive correlation between COVID-19 deaths per km² and social trust was significant for all countries (r = 0.29, p = 0.000) and for the provinces of 8 countries (r = 0.32, p = 0.000). In contrast, Fig. 1e and f shows a negative and significant correlation between COVID-19 deaths per km² and family bond for all countries (r = − 0.15, p = 0.014) and for the provinces of 8 countries (r = − 0.21, p = 0.001). Finally, Fig. 1g and h reveals a negative correlation between COVID-19 deaths per km² and security for all countries (r = − 0.07, p = 0.231) and for the provinces of 8 countries (r = − 0.06, p = 0.307).

The results of multiple linear regressions for predicting COVID-19 deaths per km² are shown in Table 2. According to our aim, we included social capital-related factors for all specifications (1–8) to check the robustness of the predictions of the social capital-related factors. In columns 1–7, we included the other explanatory variables one-by-one along with the social capital-related factors. In columns 3–6, we included social capital-related factors with the population aged 65 or older as interaction terms. In column 8, we included all explanatory variables simultaneously in the regression model.

Among the social capital-related factors, two factors, community attachment and social trust, were associated with more COVID-19 deaths per km². In contrast, two other factors, family bond and security, were associated with fewer COVID-19 deaths per km² in all specifications. Other explanatory variables, population density and population aged 65 or older per km², were associated with more COVID-19 deaths per km²; all interaction terms between population aged 65 or older and social capital-related factors were associated with more COVID-19 deaths per km². One additional bed per 1000 people was associated with fewer COVID-19 deaths per km². Early lockdown policies were associated with fewer COVID-19 deaths than late lockdown policy.

As robustness checks, although we included other explanatory variables with social capital-related factors in several multiple regressions, the sign of the influence of social capital-related factors on COVID-19 deaths was not changed. Furthermore, the robust standard error implies that there was no heteroscedasticity, and diagnostic tests confirmed normality and no multicollinearity in the regressions. In addition, we tested the correlation between COVID-19 deaths per km² and social capital-related factors by the 8 most populous countries separately. Most of the results were similar to those in Fig. 1. The results are presented in Supplementary Fig. S1.