Investigation of Conditions Required to Use Meta-Analysis Handling a Missing Variable in Structural Equation Modeling

• In Event: 53.067 - Statistical Research in the Contexts of Parametric Assumption Violations, Simulation, Structural Equation Modeling, and Ordinal or Nominal Data Analyses

Abstract

Structural equation modeling (SEM) is a widely used statistical technique in social science. However, missing elements in a covariance matrix to be used for the analysis often create major problems for model specification and estimation of parameters. This simulation study was conducted to investigate the conditions where meta-analytic method is effective for handling the missing elements in correlation matrix for SEM. Specifically, I investigated the performance of the method to maintain the precision of parameter estimation, model fit, Type I error rate as functions of (a) the number of primary studies (k) used in meta-analysis to estimate the missing elements, (b) the average sample size per primary study (n), and (c) the locations of missing variables.

Authors

• Jing Lv, East China Normal University

• Yukiko Maeda, Purdue University