New parametric and nonparametric multiple test procedures for high-dimensional data
Modern techniques in biomedical research as microarrays or computer based imaging techniques often yield extremely high-dimensional data for a patient. We propose several procedures for separate tests with all variables controlling the experimentwise type I error in a parametric as well as in a nonparametric setup. These procedures utilise the idea that all variables should have a similar scale. Otherwise the procedures are less powerful but the type I error is still under strong control.
Various modifications of the basic procedures weaken the power-dependence on the assumption of equal variances. All procedures are very simple to implement. They are demonstrated here in a microarray data set, comparing their performance with standard techniques.