Hypothesis Testing Problems In Meta-analysis In Finite Sample Cases

Meta-analysis provides a quantitative method for combining results from separate independent studies about the same scientific problem with identical treatment.Current thesis aims to study the heterogeneity test for random effects models in meta-analysis.An equivalent form of the exact distribution of the restricted likelihood ratio test(RLRT)statistic is developed under the null based on the spectral representation.Parametric bootstrap is employed to simulate the exact distribution of RLRT and the testing procedure under finite sample setting is developed accordingly.Simulation studies show that the proposed method outperforms the Q-test and RLRT developed based on asymptotic argument in finite sample situations and has very comparable performances when sample size is large.For demonstration purpose,the proposed method is adopted to analyze the bacillus Calmette-Guerin(BCG)vaccine data.The method developed in current study provides a useful remedy for improving the performance of asymptoticbased tests and can be considered as an optional instrument for testing the potential heterogeneity in the context of meta-analysis.