A final limitation is related to multiple testing. A central issue in the field of genetic association studies is a lack of consensus on how to deal with multiple, non-independent analyses. Several strategies have been considered, including the standard Bonferroni correction for the total number of SNPs being analyzed, modifications of the Bonferroni correction based on the estimated number of independent SNPs being analyzed (Gao et al., 2010; Nyholt, 2004), evaluation based on rank-ordering of p-values (Helgadottir et al., 2007), false discovery rates (Benjamini & Hochberg, 1995), and Bayesian methods (Wakefield, 2007). See Ziegler and colleagues (2008) for a recent review of these issues. In the present study, this issue is compounded by the use of multinomial logistic regression, where the prediction of differential probabilities of membership across three discrete trajectory classes yields test statistics for two non-redundant pairwise comparisons (DL vs. DM & DL vs. SH) per analysis. Moreover, in addition to the question of the number of independent analyses on which to base a correction, it is not even clear whether corrections should be made with respect to
