Each SNP genotype was converted in SOLAR to a covariate measure equal to 0, 1, or 2 copies of the minor allele (or, for missing genotypes, the weighted covariate based on imputation). These covariates were included in the variance-components mixed models for measured genotype analyses [26] versus null models that incorporated the random effect of kinship and fixed effects such as age, sex, their interaction and higher order terms. For the initial GWA screen, we tested each SNP covariate independently as a 1 degree of freedom likelihood ratio test. The p-value threshold for genome-wide significance (alpha = 0.05) was set at 1.01×10−7. The p-value threshold for genome-wide significance was computed for our family-based cohort that takes into account pedigree structure. The effective number of SNPs given linkage disequilibrium (LD) was calculated by the method of Moskvina and Schmidt [27] as implemented in SOLAR. LD was computed in SOLAR using all available information (all genotyped SNPs on all individuals). The average ratio of SNP effective number/actual number obtained from analysis of 1,989 non-overlapping bins of SNPs was used to calculate the
