Wednesday, December 30, 2024. 

Time: 16:10-17:00

venue: Mathematics Building Room 527

Speaker:  Dr. Lin, Zong-Lin (Institute of Statistics, National Yang Ming Chiao Tung University)

Abstract:

Latent class models (LCMs) assume conditional independence among item responses; however, this assumption is often violated in practice, leading to estimation bias and poor model fit. Although previous studies have relaxed the assumption of conditional dependence, they have not considered the inclusion of covariates. The objective of this paper is to obtain stable parameter estimates for LCMs with covariates, relaxing the assumption of conditional independence, using generalized estimating equations (GEE). We construct quasiscore functions to estimate model parameters and pairwise covariances between item responses, treating these covariances as nuisance parameters and estimating them using conditional odds ratios. To address convergence issues associated with the Fisher scoring algorithm in LCMs within the GEE framework, we incorporate the generalized method of moments to adjust the estimation process, providing a robust alternative for stable parameter estimation and reliable results.