Addressing Endogeneity in Meta-Analysis: Instrumental Variable Based Meta-Analytic Structural Equation Modeling

In management research, meta-analysis is often used to aggregate findings from observational studies that lack random assignment to predictors (e.g., surveys), which may pose challenges in making accurate inferences due to the correlational nature of effect sizes. To improve inferential accuracy, we show how instrumental variable (IV) methods can be integrated into meta-analysis to help researchers obtain unbiased estimates. Our IV-based meta-analytic structural equation modeling (IV-MASEM) method relies on the fact that IVs can be incorporated into SEM, and meta-analytic effect sizes from correlational research can be used for MASEM. Conveniently, IV-MASEM does not require that each primary study measures all relevant variables, and it can address typical types of endogeneity, such as omitted variable bias. We clarify how the principles of IV-SEM can be applied to MASEM and then conduct three simulations to study the validity of IV-MASEM versus Univariate Meta-Analyses (UMA) and MASEMs that exclude IVs when the instruments were appropriate, inappropriate, and missing from a subset of primary studies. We also offer an illustrative study to demonstrate how to apply IV-MASEM to address endogeneity concerns in meta-analysis, which includes a new R function to test the qualifying conditions for IVs. We conclude with limitations and future directions for IV-MASEM.