Site Tools


workshop:index

Workshop on Robust SEM for Non-normal and Missing data using WebSEM

by Zhiyong Zhang and Ke-Hai Yuan

To be presented at the 121st Annual Convention of the American Psychological Association at Honolulu, HI, August 4, 2013.

We welcome feedback on the workshop. Thanks.

Description

Being capable of modeling latent variables and measurement errors simultaneously, structural equation modeling (SEM) has become one of the most popular statistical methods in social and behavioral research, where missing data are common, especially when data are collected longitudinally. Many procedures have been developed for modeling missing data in SEM, most of which are normal-distribution-based maximum likelihood (NML). However, NML estimates (NMLE) can be very inefficient or even biased when data have heavy tails or are contaminated.

This workshop aims to introduce a newly developed two-stage robust method and related software for SEM analysis to handle practical data, possibly with both missingness and contamination (Yuan & Zhang, 2011, 2012). Technical details of the two-stage robust method will be first reviewed. Then, using real data from National Longitudinal Survey of Youth 1997 cohort, we will demonstrate how to use the free online software WebSEM to practically conduct robust SEM analysis

Topics

  • Demonstration of differences between NML and robust methods
  • Two-stage and direct robust methods
  • Structural equation model diagnostics of outlying observations
  • Bias and efficiency of parameter estimates
  • Performance of test statistics
  • Missing data and influence of auxiliary variables

Use of WebSEM

The online software WebSEM is available at https://websem.psychstat.org. By clicking the link, you will be directed to log in. If you have not registered as a user, please do so following the on-screen instruction.

Example data sets

workshop/index.txt · Last modified: 2014/07/30 11:20 (external edit)