Simulation study power results.

Simulation parameters are described in Table 3. 1000 replicates were generated for each scenario. Assumption violating scenarios are described in the text. Simulations on violation of model assumptions use the same parameters as “base”. Where nonlinear transformations of fitness effects are used, the variance of transformed fitness effects is rescaled to be the same. Power is proportion of p less than .05. Min p is Bonferroni corrected minimum p-value; CMC is the method of [20]; CAST is the same method ignoring common variants; Weighted is the method of [21]. Optimal Mean uses our implementation of , but not variance components. See methods text “Simulation Studies” for how CAST, CMC, and Weighted were modified to be closer to the data generating mechanism. Columns with “1%” or “5%” involve dichotomizing variants at the specified frequency threshold. CMC, Weighted, and Optimal Mean treat variants above that threshold as free regression parameters. Our method (EMMPAT) likelihood ratio p value is estimated from 500 permutations.