Optimal Design for Mixed Effects Models

• Activity Number: 507
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Linwei Hu*+ and John Stufken
• Companies: and University of Georgia
• Keywords: optimal design ; mixed model ; complete class ; information matrix ; linearization ; \Phi_p-optimality

Abstract

We consider optimal design problems for correlated data under mixed effects models. For independent data, complete class results have been given for many commonly used models. These results are very powerful because the search of optimal designs can now be restricted within a small complete class of designs, which greatly simplifies the problem. However, for correlated data, similar results haven't been established. In this paper, we will generalize the complete class results derived under independent data to correlated data under linear and nonlinear mixed effects models. We will also extend some other nice results to mixed effects models. Then we apply our method to the mixed effects biexponential regression model, to find its optimal designs. Finally, the robustness of design efficiency against miss-specifying the covariance structure of the random effects is investigated.