QPRC 2016

(More) Design of Experiments for Generalized Linear Models with Random Blocks

Edgar Hassler, Arizona State University
Douglas C. Montgomery, Arizona State University
Rachel T. Silvestrini, Rochester Institute of Technology

Abstract

Recent work in designing experiments for generalized linear models in the presence of random blocks raises important questions about two simplifications often used in quasi-likelihood approaches.  First, marginal approximations that neglect the effect of dispersion on the mean are employed in cases where closed forms of certain moments are unavailable (e.g. binomial responses).   Second, information matrix approximations often ignore information with respect to the randomeffect dispersion parameter itself.  The effect of the first simplification is demonstrated in several models.  Then, quasi-likelihood information matrices are constructed that take into account the information for the random effect dispersion parameter. The effect of the second simplification is then assessed over several models.