Designing to detect heteroscedasticity in a regression model

We consider the problem of designing experiments to detect the presence of a specified heteroscedastity in Gaussian regression models. We study the relationship of the D-s- and KL-criteria with the noncentrality parameter of the asymptotic chi-squared distribution of a likelihood-based test, for local alternatives. We found that, when the heteroscedastity depends on one parameter, the two criteria coincide asymptotically and that the D-1-criterion is proportional to the noncentrality parameter. Differently, when it depends on several parameters, the KL-optimum design converges to the design that maximizes the noncentrality parameter. Our theoretical findings are confirmed through a simulation study.