报告题目：Nonlinear structural equation modelling using H-likelihood 

报告时间：2017年4月26日(星期三) 13:30-14:15 

报告地点：尚贤楼1楼108报告厅 

报告主持人: 曹春正 副教授 

欢迎广大师生踊跃参加！ 

报告内容简介： 

Nonlinear structural equation modelling (SEM) has becoming widely used in practice. Due to its nature of non-normally distributed indicators, various distribution analytic approaches have been proposed. In this study, an alternative approach, the h-likelihood approach, is proposed under the normality assumption. The h-likelihood approach is computationally more appealing than the Bayesian approach and the Gauss-Hermite quadrature-based methods. The normal assumption-based h-likelihood approach is further extended to fat-tailed latent distributions. We also propose a two-step variant to be more robust for a skewed latent distribution. A simulation study shows that the proposed approach produces accurate parameter estimates if the distribution is correctly specified and a lower bias when the distribution is incorrectly specified. 

数学与统计学院 

2017-04-25