Cao et al.17 considered a joint test of mean and variance differences by using a full likelihood approach based on linear regression models. The likelihood ratio test (LRT) statistic follows an asymptotic chi-square distribution under the null hypothesis. The LRT approach can increase power but is more sensitive to model assumptions such as normality. Therefore, Cao et al.17 proposed a parametric bootstrap method to calculate “honest” p values at the cost of computational efficiency. For both the LRT and distribution methods, implementation difficulties arise for multivariate (e.g., gene-based, gene-set, and pathway) analyses (see Appendix A).