Fig. 5From: How to perform prespecified subgroup analyses when using propensity score methods in the case of imbalanced subgroupsPower of the interaction test. Comparison of the power of the Gail and Simon quantitative interaction test by the number of patients (A) (Scenario 1) or the treatment effect (B and C) (Scenario 2). The number of patients is set to n=3000 in B and n=300 in C (Scenario 1). Robust estimate of variance was used for SMRW weighting when \(n=300\), rather than bootstrapping, due to the importance of convergence problems that made it impratical to compute itBack to article page