Dashboard: Social and political correlates of Covid-19

major_missing_because_of_controls <- 
  filter(df, !complete.cases(select(df, all_of(controls)))) %>% 
  filter(pop_tot>1 & !is.na(country) & country != "NA")

dropped <- major_missing_because_of_controls$country %>% 
  as.character %>% 
  unique 

major_missing_because_of_controls_pc <- 
  filter(df, !complete.cases(select(df, all_of(controls_deaths_capita)))) %>%
  filter(pop_tot>1 & !is.na(country) & country != "NA")

dropped_pc <- major_missing_because_of_controls_pc$country %>% 
  as.character %>% 
  unique 

##  Get measures and labels


families <- c("state_cap_vars", "pol_account_vars", "social_vars",
              "econ_vars",  "phys_vars", "epi_vars", "health_sys_vars")

for(v in families){  
  assign(v, filter(measures, family == v)$vars %>% as.character) 
  assign(paste0(v, "_labels"), filter(measures, family == v)$labels %>% as.character)
 }

Existing social science research has implications for the kinds of features of societies that might help explain variation in a society’s ability to respond to adverse shocks. In some accounts, for example, more democratic societies will be more responsive; in other accounts, state capacity or social cohesion are more important. We gather together major measures suggested by these literatures and report on the evolving covariance between these measures and the cross national distribution of reported Covid-19 burdens.

As of the most recent data date, 22 November, 2021:

• A small set of simple controls—total population (logged), healthcare quality index (GHSI), healthcare spending/capita, health data quality, and NA—explain 64% of the cross national variation in deaths (logged). We used a Lasso procedure to chose the controls that perform best at predicting Covid-19 mortality (see paper for more details).

• Given these controls:

  • many political and economic measures are (currently) not related with Covid-19 burdens in the directions suggested by prominent theories. For instance, wealthier, more democratic, and more prepared states currently suffer the greatest burdens.
  • A number of features highlighted in public discussions—such as the role of populist governments and women leaders—also do not find support.
  • Other accounts do find support however, including a set of measures of state capacity and measures of social trust and trust in government.

• These patterns may change as the pandemic evolves.

• See our paper discussing the literature and logics behind these analyses.

• See this paper for even more discussion as well as supplementary analysis on policy outcome measures.

• All data preparation code and sources available here. And merged complete data here.

Notes / warnings:

• The covariances we report should not be taken as estimates of causal effects but rather as summaries of covariation between measures.
• Cross national covariation can in part reflect differences in reporting standards and not just underlying variation in the disease.
• Currently outcome data is highly skewed. Moreover the scale of the outcome is changing as the disease spreads which in itself can produce larger coefficient magnitudes over time.

Overview

Deaths/million (logged)

Predicted and observed mortality

---

Each point in the figure shows the actual and predicted (log) per capita mortality for a given country. To predict mortality, we use four covariates selected by a Lasso procedure that maximize predictive accuracy as well an additional control for the quality of health data reporting. The controls are share 65+, respiratory disease prevalence, healthcare quality index (GHSI), healthcare spending/capita, and health data quality.