Comparison between Fully Bayesian Hierarchical Meta-analysis and Classical Meta-analysis: A Monte Carlo Study Based on Correlation Coefficient.

Take the Pearson's correlation coefficient as an example, the difference between fully Bayesian hierarchical metaanalysis and classical meta-analysis was compared. Through the experimental design of following four factors 5 (research quantum) *3(effect size) *4(variance of heterogeneity) *2(two types of meta-analysis), simulation comparison was conducted. The results show that: when these two types of meta-analysis are analyzed with fixed effect, there was almost no difference between the point estimation and confidence interval; while when they are analyzed with random effect, regardless of the prior distribution of random effects, uniform or conjugate distribution, with the increase in research quantum, results of them tend to be consistent; the credible interval of fully Bayesian hierarchical meta-analysis is slightly wider than that of classical meta-analysis. Therefore, if the true effect is determined to be fixed effect, both two types of meta-analysis can be used. But when random effect is adopted, if the research quantum is less than 20, the fully Bayesian hierarchical meta-analysis should be the choice, and the selection of prior distribution should be determined according to practical situations; while if the research quantum is more than 20, both two types of meta-analysis can be used. [ABSTRACT FROM AUTHOR]