2
Faculty Member, Department of Statistics, University of Isfahan
Abstract
Generally, inference in a generalized linear model is based on asymptotic approximations for the bias and covariance matrix of the parameter estimator. In experiments with small sample sizes, these approximations perform poorly because they yield biased estimators. The study of optimal designs in such experiments is of particular importance in the life sciences, and especially in pharmacy, in order to determine the dose level of a drug. In this regard, Talebi and Russell [11] studied optimal designs for small samples in the logistic model, considering the integral of the mean squared error of prediction. They limited their research to 2-point designs. In this paper, we generalize their designs to three points and also examine the role of bias in the introduced criterion.
