Title: Two modeling strategies for two-part latent variable model
Authors: Yemao Xia - Nanjing Forestry University (China) [presenting]
Abstract: Semi-continuous data often occur in the survey of economics and social sciences. In analyzing such data, a primary interest is to assess the the effects of observed covariates on the variability of responses. We extend the two-part regression model to the case where the unobserved heterogeneities are explained by the latent variable model. The information on latent factors is specified via latent variable model. We develop two estimation procedures for analyzing such data: one is based on robust moment estimation equation and the other is within the Bayesian framework. For the former, we establish two-step estimation procedure for the unknown parameters involved and investigate the asymptotic properties such as consistency and asymptotic normality; while for the latter, we design a Polya-Gamma Gibbs sampler in the Bayesian posterior sampling. We also assess model fits via constructing various related hypothesis testing procedures. Simulation studies were carried out to assess the performance of the two approaches, especially the robust behavior of estimates and tests when the underlying distribution assumptions are violated. A real data set is analyzed to illustrate the practical values of the proposed methodology.