Factor analysis is commonly used in many elds to establish relationships among manifest variables in terms of few underlying latent factors. Besides the elucidation of these factors, of importance in many applications is their subsequent analysis, for instance, in regression settings given some covariates. Motivated by genetic association studies of autism spectrum disorders, this research revisits common dependence measures for multivariate ordinal data, and investigates robustness of factor analysis and factor scores regression under various distributional settings. We rst demonstrate numerically, a comparison of dependence measures for multivariate ordinal data and investigate the robustness of polychoric correlation estimation under settings with asymmetric dependence patterns and varying degrees of skewness in marginal distributions. To accommodate such general joint distributions, we make use of factor copula models in data generation. These models o er a very general and exible framework to study relationships among manifest and latent variables. Hence, we propose an alternative strategy to quantify the scores on the latent variables using factor copula scores and investigate the performance of the proposed approach in comparison to the traditional factor model both in factor scores estimation and their association testing for a given covariate. Our ndings suggest that the traditional factor analysis is considerably robust to violations of distributional assumptions. Factor copula analysis yields very similar results to those based on polychoric correlation, with a slight power gain in association testing.