Optimal Design Robust to a Misspecified Model

Abstract

Usually, in the Theory of Optimal Experimental Design the model is assumed to be known at the design stage. In practice, however, more competing models may be plausible for the same data. There is a possibility to find an optimum design which takes into account both model discrimination (for a subsequent application of a hypothesis test) and parameter estimation. In order to avoid the problem of hypothesis testing, a different approach is proposed: to determine an optimum design which is useful for estimation purposes and is robust to a misspecified model. In other words, the optimum design is “good” for estimating the unknown parameters whether or not the assumed model is correct.