Testing the Need for a Random Effects Modelin in a Two Compartment Model

The objective of this paper is to find a simple way to test whether random effects are needed in a nonlinear mixed effects model. We proposed a test statistic, approximately an F random variable, from the fixed parameter approach which compares the residual sum of squares from the full model and the reduced model. From the difference of exponentials model simulations, the empirical size of the test is slightly higher than the nominal level a. The test offers very good power for detection. The achieved power depends on the error variance, the population coefficient of variation (CV) of the random effects, and the number of random effects in the model. For a fixed error variance, power increases as the population CV increases and/or the number of random effects increases. From our sensitivity analysis the performance of these test statistics is similar when the modei has fwo rate constants that are almost equal, or when the model is close to a one compartment model