Further investigation on different indicators for LPA might be able to refine the classification of SNPs. We used −log10(p) as an index of association strength and used them as class indicator for LPA among other possible class indicators for the reasons described in the methods section. In a post hoc analyses, we fit the same LPA models to alternative class indicators: raw p-values, absolute values of beta coefficients, and absolute values of t-statistics. Beta coefficients were also standardized to make interpretation of resulting classes more straightforward, given that linear transformations of class indicators do not affect the estimation of classes (Magidson and Vermunt 2002). The results of 3-class LPA using beta, raw p-values and t statistics are provided in the supplementary material (Table S3). As expected, only LPA results using t statistics as indicators showed consistent patterns of classifications to the results from using −log10(p). Although −log10(p) may be more appropriate for LPA than alternatives considered, searching for class indicators that can better represent association strengths and distinguish between SNPs would be an important step to elaborate our initial results.
