Optimal and Efficient Designs for Generalized Linear Models

• Activity Number: 6
• Type: Invited
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): John Stufken*+
• Companies: The University of Georgia
• Keywords: Geometric approach ; Locally optimal designs ; Non-linear models ; Logistic model ; Probit model ; Loglinear model

Abstract

The literature on optimal design for linear models is very well developed. In spite of seminal advances over the past decades, tackling the optimal design problem for non-linear models has proven to be much more difficult. Many of the available results for non-linear models have been obtained by clever use of the geometric approach, an approach inspired by Elfving's seminal contribution for linear models. In recent years it has however been possible to obtain more general results for many models through use of an analytic method. The models covered by this new approach include generalized linear models for binary and count data, such as logistic models, probit models, and loglinear models. We discuss the basic idea underlying the analytic approach, present key results obtained by its use, discuss its limitations, and offer some open problems in this area.