70 – Generalized Linear and Nonlinear Mixed Models
Statistical Power of Approximate Tests in Two-Stage Nonlinear Mixed Models
Jeff Burton
Pennington Biomedical Research Center
Julia Volaufova
LSU Health Sciences Center
Robbie Beyl
Pennington Biomedical Research Center
William D. Johnson
Pennington Biomedical Research Center
Here, we investigate power of tests about fixed effects parameters in nonlinear mixed models. For obtaining estimates of unknown model parameters needed for construction of test statistics, a two-stage approach is implemented. The tests of interest that shall be compared are the large-sample Wald and likelihood ratio tests as well as several approximate F-tests. For the Wald and approximate F-tests, we modify the test statistics by developing an approximation to the large-sample covariance matrix of estimates of fixed population parameters based on the Fisher information matrix. In contrast to previous research in which null properties of tests were investigated (Burton & Volaufova, 2014), our focus here is on comparing power of tests using various configurations of fixed model parameters under the alternative hypothesis. A simulation study using a one-compartment pharmacokinetic model is used for illustration.