Optimal Design for Categorical Variables

Abstract

Optimal design research is important for behavioral research because it enables efficient estimation of parameters and maximum power with minimum sample size and costs of performing a study. In this paper, an elementary introduction to the concepts of optimal design research is given and illustrated by means of four different cases, namely, linear and logistic models, with and without dummy coding. Although different models have different optimal designs, a rule‐of‐thumb may be that if the independent variables are treated as being quantitative, efficiency will increase if the variance of the independent variable increases, and if the design is as orthogonal as possible. For qualitative independent variables, a design that has equal sample sizes for all categories is often optimal.