Exact Inference in Meta-Analysis via Exact Confidence Distributions
*Lu Tian, Stanford University
Keywords: meta analysis; fixed effect model; random effect model; exact confidence interval; confidence distribution
The performance of the standard inference procedure in meta-analysis is sensitive to the parametric assumption, as well as the number of studies. When the number of studies involved is not big, the widely used DerSimonian-Laird interval for the average treatment effect may not be able to cover the true parameter at the desired nominal level. In this talk, we will construct a new class of exact confidence distributions, which can be used to yeild the confidence interval for the parameter of interest. This new meta-analysis technique can be implemented easily with study-level summary statistics as the standard method. However, in contrast to existing methods depending on asymptotical approximations, the validity of our proposal does not require the number of studies involved to be large. The proposed procedure has robust empirical performance based on our extensive numerical study. We will then use the procedure to analyze some real data examples.