Maximin Model Robust Discrete Choice Experiments

Discrete choice experiments have their genesis in conjoint analysis, which was based on first-ordermodels. For this and other reasons, discrete choice experiments in practice are usually designedfor estimation of main effects. In this paper, we explore the construction of maximin model robustdesigns when the experimenter is concerned about the presence of interactions. We consider threeclasses of models—main effects models, main effects models plus first-order interactions, and secondordermodels, and we construct designs that maximize the minimum efficiency of the design for thethree competing models. We do so first with standard linear models and then extend our analysisto nonlinear models and, in particular, to models for discrete choice experiments. We compare ourresults with existing approaches in the literature, such as Li et al., 2013.