Metodi non parametrici per la verifica di ipotesi in indagini multicentriche

Abstract

This paper deals with an application of the nonparametric combination methodology of dependent permutation tests (NPC) (Pesarin, 1990, 2001) to an epidemiological multivariate testing problem. The data analysed come from an observational multicentric study on adverse events related to two different categories of anti-depressive pharmacological treatments. The NPC methodology allows the resolution of complex multidimensional and multistrata testing problems also when the number of variables is greater than the number of observations, i.e. in cases where there are no degrees of freedom (see also Crowder & Hand, 1990; Diggle et al., 1996), and in presence of missing data. Under rather mild conditions the NPC methodology allows to obtain an exact, unbiased and consistent solution and asymptotically equivalent to the optimal parametric solution when the latter exists (Pesarin, 2001). An interesting feature of NPC is that it does not require to model the dependence structure underlying variables. A brief description of the NPC with extension to the case of both repeated measurements and stratified samples and the results of the testing problem are presented.