Figure 5.

Percentage of rejections of the likelihood ratio test (LRT) between a gamma frailty model and a proportional hazard model compared to the test for nonproportional hazards (ZPH), when the data are simulated with an unobserved common risk following a lognormal distribution with expectation 1 and variance 0.25 and an increasing Weibull baseline hazard with shape α = 0.8. The rows correspond to the total sample size (300,900,1500) and the columns to the three main simulation scenarios: clustered failures, clustered failures where the observed covariate only varies between clusters, and recurrent events. β ₁ indicates the strength of the time‐dependent covariate effect [Colour figure can be viewed at wileyonlinelibrary.com]