Meta-analysis of non-statistically significant unreported effects. (December 2019)

• Record Type: Journal Article
• Authors: Albajes-Eizagirre, Anton; Solanes, Aleix; Radua, Joaquim
• Is Part Of: Statistical methods in medical research. Volume 28:Number 12(2019)
• Page Start: 3741
• Page End: 3754
• Publication Date: 2019-12
• Subjects: Meta-analysis -- non-statistically significant unreported effects -- interval censoring -- multiple imputation; Medicine -- Research -- Statistical methods -- Periodicals; Research -- Periodicals; Review Literature -- Periodicals; Statistics -- methods -- Periodicals; Médecine -- Recherche -- Méthodes statistiques -- Périodiques; 610.727
• DOI: 10.1177/0962280218811349
• Languages: English
• ISSNs: 0962-2802

Abstract

Published studies in Medicine (and virtually any other discipline) sometimes report that a difference or correlation did not reach statistical significance but do not report its effect size or any statistic from which the latter may be derived. Unfortunately, meta-analysts should not exclude these studies because their exclusion would bias the meta-analytic outcome, but also they cannot be included as null effect sizes because this strategy is also associated to bias. To overcome this problem, we have developed MetaNSUE, a novel method based on multiple imputations of the censored information. We also provide an R package and an easy-to-use Graphical User Interface for non-R meta-analysts.