Power, FDR and conservativeness of BB-SGoF method

Castro Conde, Irene; de Uña-Álvarez, Jacobo

Abstract:

Beta-binomial sequential goodness-of-fit (or BB-SGoF) method for multiple testing has been recently proposed as a suitable modification of the sequential goodness-of-fit (SGoF) multiple testingmethod when the tests are correlated in blocks. In this paper we investigate for the first time the power, the FDR and the conservativeness of BB-SGoF in an intensive Monte Carlo simulation study. Important features such as automatic selection of the number of existing blocks and preliminary testing for independence are explored. Our study reveals that (a) BB-SGoF method roughly maintains the properties of original SGoF in the dependent case, reporting a small value for the probability that the number of false positives exceeds the number of false negatives with p value below γ ; (b) BB-SGoF weakly controls for FDR even when the beta-binomial model is violated and the number of blocks k is unknown; and that (c) the loss of power of the automatic selector for the number of blocks relative to the benchmark method which uses the true k varies depending on the proportion and the type (strong, intermediate or weak) of the effects, being strongly influenced by the within-block correlation too.

Year:

2015

Type of Publication:

Article

Keywords:

False discovery rate; Monte Carlo simulations; Multiple comparison procedures; Sequential goodness-of-fit; Serial dependence

Journal:

Computational Statistics

Volume:

30

Pages:

1143-1161

Month:

December

DOI:

http://dx.doi.org/10.1007/s00180-015-0553-2