FIGURE 1.

Results from simulations of a continuous outcome, comparing different methods in terms of the mean squared error for the patient‐specific treatment effect. Scenarios C1‐C28 are shown in pairs, differing only in the number of studies. Within each pair, the two scenarios explore the same type of outcome, have the same number of covariates and effect modifiers, equal heterogeneity and magnitude of effect modification. Scenarios C29‐C36 are shown in pairs, differing only in the assumed heterogeneity (τ). Scenarios are described in detail in Section 2 of the Appendix. A, GLMM‐oracle; B, GLMM‐full; C, STEP; D, LASSO; E, ridge; F, BAYES‐LASSO; G, SSVS