Fig. 3. Associations between weather variables, non-pharmaceutical interventions and the effective reproduction number.
Non-linear associations between (a) mean temperature (°C), (b) relative humidity (%), (c) absolute humidity (g/m3) and (d) OxCGRT Government Response Index and predicted Re difference. Curves and their 95% confidence intervals show the predicted difference in Re with respect to a reference value set to the value at the trough of the curve for meteorological variables (a–c), or for the OxCGRT Government Response Index = 50 (d). Two-sided Wald test p values and adjusted curves with 95% confidence intervals were obtained from multivariable meta-regression multilevel models adjusted by population (log scale), population density (log scale), GDP (log scale), % population >65 years of age, PM2.5 (μg/m3, log scale) and OxCGRT Government Response Index, with cities nested within countries. The marginal distribution along the x-axis represents the observed data for that covariate.