We fit LPAs with two to ten classes and examined classifications of SNPs across LPAs with different number of classes, based on most likely class memberships. The purposes of these analyses were to: (1) determine the optimal number of classes of SNPs and (2) examine whether classifications of SNPs follow the pattern predicted by twin studies. Assuming that SNPs corresponding to internalizing and externalizing genetic factors exist, we expected that SNPs in a class associated with both AD and ASP would be divided into individual classes of AD and ASP, respectively, as the number of classes increases. In contrast, we hypothesized that SNPs associated with MD would initially form their own class and would not overlap with SNPs in classes for AD and/or ASP by increasing the number of classes. To guide the selection of the optimal number of classes, we used both information criteria and the bootstrap likelihood ratio test (BLRT) (McLachlan and Peel 2004). The Akaike information criterion (AIC) (Akaike 1987) and Bayesian information criterion (BIC) (Schwarz 1978) penalize the complexity of models (i.e. the number of parameters
