A Bayesian estimation approach to analyze non-Gaussian data-generating processes with latent classes

Several methods have recently been proposed to discover a complete causal structure, that is, all the causal directions, under the assumption that disturbance variables have non-Gaussian distributions. However, the estimation results can be biased if there are "latent classes." Latent classes are unobserved discrete variables that have more than one observed child variables. Data that has been generated from different processes are mixed in the presence of latent classes. Several methods have been proposed to estimate the causal structure in the presence of latent classes [12], but all of these are affected by local optima. Therefore, in this paper, we propose a new estimation approach that can solve this problem. The structure of this paper is as follows.