Abstract:
Conventional multivariate meta-analysis models assume that both between-study random effects and within-study random errors follow a normal distribution. However, this assumption makes the model susceptible to outliers and unusual observations. To enhance the model’s robustness and its ability to handle anomalies in both random effects and random errors, this paper introduces a multivariate random effects meta-analysis model based on the
t-distribution. In this model, both the random-effect term and the random error term are assumed to follow multivariate
t-distributions. The paper details the ECM algorithm along with three accelerated versions of this algorithm and builds the
t-
t model for multivariate meta-analysis using maximum likelihood estimation. Numerical simulations and case studies demonstrate that the PX-ECME algorithm has the highest effciency among the proposed algorithms. The experiments not only confirm the robustness of the
t-distribution modeling but also demonstrate that the
t-
t model can effectively identify outliers and distinguish their origins.