Comparative Study on Software for Analysing Non-normal Data in Structural Equation Modeling

• 발행기관: 한국자료분석학회
• 저자명: Sang Oak Dong Changwan Kang Kyu Kon Kim
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.14 No.3, 1129~1138쪽, 전체 10쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2012.06.30

영문 초록

Structural equation modeling (SEM) techniques are considered today to be one of the major components of applied multivariate statistical analyses and are used by biologists, economists, educational researchers, marketing researchers, medical researchers, and a variety of other social and behavioral scientists. In addition, many softwares have been developed for SEM and, over time, have grown steadily. However, a great obstacle for its wider use has been its difficulty in handling non-normal variables within the framework of generalized linear models. In this study, we introduce the five softwares for SEM with non-normal variables and compare them through some simulations. In the comparison results, most of the softwares analyzed by recognizing binary data in the continuous data. However, some softwares were found to use a special covariance matrix to analyze data. For continuous data, we recommend Mplus’s WLS and R’s WLS, and for categorical data, we recommend LISREL’s WLS using the tetrachoric correlation and R’s WLS using the ploycor package.

목차

1. Introduction

2. Method for Non-normal Data

3. Simulation

4. Conclusion and Proposal

Reference

키워드

SEM ADF WLS Model-fit Index Standard Estimate Standard Error