Conditionally Dependent Dichotomous Latent Class Models with Covariates
時間
星期四 下午2:00~4:00
地點
數學系1樓21113演講室
摘要
Biomedical and social psychology researchers are increasingly using latent class models with covariates to analyze the relationship between multiple categorical outcomes and associated covariates. Latent class models assume that multiple variables within the same latent class are independent, known as the conditional independence assumption. However, this assumption may not always hold true in certain cases, necessitating an increase in the number of latent classes to satisfy the assumption of conditional independence. This will result in redundant parameters without meaningful interpretation. To relax this assumption, we incorporate high-order correlations among the measurement items within latent classes and transform marginal probabilities into a saturated log-linear model. We use an iterative proportional fitting algorithm to estimate the joint probabilities, and then use the EM algorithm and the Levenberg-Marquardt algorithm for parameter estimation. This approach is also applicable for handling mixtures of correlated binary models with covariates.
Keywords: latent class models, iterative proportional fitting, Levenberg-Marquardt algorithm.
彰化市進德路1號 
04-7232105 ext 3207 
04-7211192