Abstract
This dissertation proposes a new approach to model-based meta-analysis that synthesizes Structural Equation Modeling (SEM) study results. Typically, model-based meta-analysis requires a pooled correlation/covariance matrix from observed correlation/covariance matrices. However, many SEM studies do not provide correlation/covariance matrices and thus are frequently excluded in model-based meta-analyses. To address this gap, the current study proposes using an implied covariance matrix derived from the estimated SEM parameters. The performance of this approach was evaluated using a Monte Carlo simulation, in which a number of factors were manipulated. Those include (1) two levels of correlation between factors, (2) two levels of factor loading, (3) four levels of between-studies variance, and (4) three levels of number of studies. Outcomes were the relative bias values of two parameters (i.e., factor loading and factor correlation) and their associated standard errors. The results indicate that the model-based meta-analysis using an implied covariance matrix performed reasonably well. Specifically, the estimated factor loading and factor correlation were more accurate, with the smallest standard errors, as compared to estimates based on a pooled correlation matrix. This dissertation demonstrates the validity of this alternative for model-based meta-analyses that use SEM studies in practice.