Comparison of Meta-Analytic Methods for Synthesizing Correlations for Meta-Analytic Structural Equation Modeling With Missing Data

• In Event: 67.040 - Missing Data Considerations, Solutions, and Issues

Abstract

While two-stage structural equation modeling (TSSEM) has been frequently used in meta-analytic structural equation modeling (MASEM), missing correlations oftentimes can be problematic. To address this issue, we used robust variance estimation (RVE) and a recent approach combining multilevel modeling with RVE (MLM-RVE). Currently, no study has compared these three approaches in synthesizing correlation matrices with missing data. This study demonstrates discrepancies in use of RVE, MLM-RVE, and TSSEM regarding parameter and standard error (SE) estimation and parameter coverage rates associated with the elements in the pooled correlation matrix. Findings indicated that both RVE and TSSEM provided unbiased parameter estimates for pooled correlations across studies. However, RVE provided unbiased SE estimates, whereas substantial bias was found in the use of TSSEM.

Authors

• Kejin Lee, University of Nebraska - Lincoln

• Tasha Beretvas, The University of Texas at Austin