Dichotomous phenotypes were modeled as latent normally distributed variables with a threshold above which an individual was considered "affected". It is notable that when applied to dichotomous phenotypes variance components linkage analysis can yield inflated LOD scores if distributional assumptions are violated (Duggirala et al., 1997). To protect against this potential bias, allele-sharing probabilities for a simulated locus under the null hypothesis of no linkage were generated across 250,000 trials (Duggirala et al., 2001), an approach shown to yield appropriate Type I error rates (Jung et al., 2006). The distribution of simulated LOD scores was used to calculate point-wise estimates of significance according to the formula p = (r+1)/(n+1) where r = the number of simulated values greater than the observed value and n = the number of simulations (North et al., 2002). As a further protection relative- -pair analyses utilizing the Kong and Cox (1997) statistic as implemented in MERLIN (Abecasis et al., 2002) were conducted for each reported linkage peak to test whether evidence for linkage was robust across analytic methods. All p-values were interpreted using guidelines suggested by Lander and Kruglyak (1995).
