A Bayesian structural equation model in general pedigree data analysis

Abstract

Abstract Structural equation modeling (SEM) is a powerful, comprehensive, and flexible multivariate statistical method for modeling relationships between observed and latent variables. However, in genetic association analysis, frequentist approaches to fitting SEMs do not always lead to convergence and admissible solutions for complex models, categorical variables, complicated data structures such as pedigree data and small sample sizes. Accordingly, to conduct a SEM pedigree data analysis, Stan platform as a probabilistic programming language was applied in our study to propose a new version of the Bayesian approach that adopts Hamiltonian Monte Carlo (HMC) and data augmentation techniques. At first, a comprehensive simulation study was conducted to compare the precision of each parameter of the suggested method with that of the classic technique in terms of bias, alpha error rate, and coverage probability. After that, the method was applied to real data with a conceptual model including ordinal indicators in order to conduct genetic association analysis of two well‐known genetic markers with metabolic syndrome trait. The simulation findings revealed the proposed Bayesian method was a more efficient technique than MLE approach. Moreover, Bayesian approach yielded a better statistical performance in solving the problems than did classic approach.