In simulation studies64, most tests behave similarly in many situations. However, the results may depend on assumptions used in simulated data. The relative power to detect association depends on factors such as the number and proportion of causal variants, their population frequency, and their effect sizes, as well as directionality of effects, the number of genes contributing to the trait, and fraction of causal genetic variation located in the exome. Statistical tests were developed with various combinations of these factors in mind and therefore are likely to be sensitive to different disease architectures. For example, the simulation framework used in development of the WSS test assumes effect size proportional to 1/x(1 − x) (where x is the population frequency of the causal allele), while Sequence Kernel Association Test (SKAT)66 simulation framework uses effect size proportional to −log(x), and the VT test simulations uses a demographic history model with a range of possible values of strength of selection leading to different relationships between effect size and x. These simulations were designed to demonstrate the strengths of each methods under different effect-size
