Preference threshold procedure for statistical selection : robustness against deviations from normality

In this paper we investigate the robustness of the preference threshold procedure introduced by Verheijen Coolen and Van der Laan with respect to deviations from the normality assumption. The preference threshold procedure is a combination of the standard indifference zone selection and the subset selection the two standard approaches for statistical selection In a number of simulation studies the preference threshold procedure is applied on populations coming from Normal Logistic Laplace Weibull Gamma and Lognormal distributions. We compare the results on basis of the observed number of correct selections.