Meta-analysis of dependent effects using robust variance estimation
Rommel Bunuan · Texas ScholarWorks (Texas Digital Library) · 2017
Robust variance estimation (RVE) was recently introduced as a way to handle within-study dependence in effect sizes in a multivariate meta-analysis. Unlike with other meta-regression and meta-analytic synthesis procedures for handling effect size dependence, it is claimed that the RVE method does not require accurate knowledge of the within-study covariance to obtain unbiased and efficient estimates of effect sizes and meta-regression model parameters. In addition, the RVE method can also handle different types of effect size dependency arising from multiple-outcome (MO), multiple-treatment (MT), and multiple-treatment and multiple-outcome (MT-MO) design studies, among others. RVE is a relatively new technique and has been used only for synthesis of effect sizes from either MO or MT design studies. In this study, this estimation method was applied to synthesize effect sizes from MT-MO design studies under 2-outcomes and 3-outcomes scenarios, and its performance was assessed in terms of relative parameter bias (RPB), relative standard error bias (RSEB), root mean squared error (RMSE), and 95% confidence interval coverage rates (CR). Results showed that RVE provides unbiased and efficient estimates of treatment effects even when the within-study covariances between effect sizes were misspecified. Implications of the simulation results for applied researchers and recommendations for future methodological research are discussed.