Related Experiment Videos
Meta-analytic methods of pooling correlation matrices for structural equation modeling under different patterns of
Carolyn F Furlow1, S Natasha Beretvas
1Department of Educational Psychology, University of Texas at Austin, Austin, TX, USA. cfurlow@gsu.edu
Psychological Methods
|July 7, 2005
Summary
This study compared three correlation synthesis methods for meta-analytic structural equation modeling (SEM) with missing data. Weighted-covariance GLS (W-COV GLS) and univariate z showed similar performance, with W-COV GLS offering slightly better parameter estimation.
Area of Science:
- Psychometrics
- Statistical Modeling
- Meta-Analysis
Background:
- Missing data is a common challenge in meta-analytic structural equation modeling (SEM).
- Accurate synthesis of correlation matrices is crucial for reliable SEM results.
- Different methods exist for handling missing data, each with potential biases.
Purpose of the Study:
- To compare the performance of three correlation synthesis methods under various missingness conditions in meta-analytic SEM.
- To evaluate the impact of different missing data mechanisms (e.g., missing at random, missing not at random) on parameter and fit index estimation.
- To assess the effectiveness of listwise versus pairwise deletion in conjunction with these synthesis methods.
Main Methods:
- Monte Carlo simulation techniques were employed to generate data.
- Three correlation synthesis methods were compared: weighted-covariance generalized least squares (W-COV GLS), univariate weighting with untransformed correlations (univariate r), and univariate weighting with Fisher's z-transformed correlations (univariate z).
- These methods were evaluated under conditions of listwise and pairwise deletion, and with varying degrees and mechanisms of missingness.
Main Results:
- W-COV GLS and univariate z demonstrated comparable performance, with W-COV GLS showing marginal advantages in parameter estimation and model rejection rates.
- Missing not at random data significantly biased correlation and SEM parameter estimates, leading to increased incorrect model rejections.
- Pairwise deletion inflated standard errors across all synthesis methods and increased incorrect rejection rates for univariate weighting procedures.
Conclusions:
- W-COV GLS is a robust method for synthesizing correlations in meta-analytic SEM, performing comparably or slightly better than univariate z.
- Missing not at random data poses a substantial threat to the validity of meta-analytic SEM findings.
- Researchers should exercise caution when using pairwise deletion due to its potential to inflate errors and rejection rates.