Parametric versus nonparametric tolerance regions in detection problems (Q2463661)

The authors deal with the problem of choosing between parametric and nonparametric approaches which often occurs in statistical practice. A statistical quality control problem of detecting a change in distribution of independent sequentially observed random vectors is considered. The authors are interested in checking and comparing the detection performance provided by tolerance regions when the original density \(f\) belongs to a rather parametric family and \(f_n\) is either an estimator from that family or a nonparametric one. In the case of normal mixture models via a simulation study they conclude that the nonparametric methodology offers a simple, model-free and competitive choice for this type of problems. A result concerning the convergence rates of the error probabilities under a general model is also obtained.