Fig. 2. Heterogeneity of exposure effect estimated using out-of-bag prediction.

a Heterogeneity in the effect of COVID-19 exposure on substantial post-acute full-time sick leave by deciles of estimated conditional risk difference (CRD). Observations are ranked in ten separate folds, using a causal forest model with observations clustered by folds. Risk differences (RD) are estimated using augmented inverse propensity weighting (AIPW). Error bars show two-sided 95% confidence intervals. b Distribution of conditional risk differences from n = 88,818 individuals, estimated using out-of-bag prediction from the causal forest model. The dashed vertical line shows the average risk difference.