Title: Reliability of meta-analysis studies
Authors: Stan Young - CGStat (United States) [presenting]
Abstract: Claims coming from scientific studies, observational and experimental, are reliably estimated to fail to replicate over 50\% of the time. All these studies contain statistical analysis in support of claims made. There is a need for an analysis strategy to evaluate claims made in studies to ascertain their reliability. A meta-analysis uses multiple studies to address a common question. The idea is to use these multiple studies to cross-check the claim rather than to attempt to evaluate an individual study. A combination of simple techniques is used: multiple testing and multiple modeling are examined in individual studies; heterogeneity of effects across studies is evaluated using graphical methods; publishing pressures and biases are discussed. Examples from environmental epidemiology are presented. It is found that many claims made in the literature lack statistical support. The benefit is that any claim made in a meta-analysis can be examined for reliability.