Fig. 4From: How to perform prespecified subgroup analyses when using propensity score methods in the case of imbalanced subgroupsSimulations with an unknown confounder. Comparison of the “across subsets” or “within subsets” strategy in terms of bias (A), variance (B) and coverage (C) in the estimation of the treatment’s effect according to the presence of an unknown confounder (Scenario 1)Back to article page