ESMARConf2023: {PublicationBias, phacking, and multibiasmeta} tutorial

Presenter: Mika Braginsky
Authors: Braginsky, Mika; Mathur, Maya
Session: Tutorials 9
Title: Tools for accounting for within-study and across-study biases in meta-analysis
Abstract: Meta-analytic estimates can be systematically biased when the meta-analyzed studies are subject to within-study biases (e.g. confounding) and/or across-study biases (e.g. publication bias and phacking). Existing methods are limited in that they generally analyze only one bias at a time and can make unrealistic statistical assumptions. We present a novel toolkit for addressing these issues by conducting meta-analytic bias correction and sensitivity analysis. Our methods allow taking into account the separate and joint effects of both within-study and across-study biases. These tools allow the researcher to answer questions such as: “For a given severity of the bias(es) in question, how much would the estimate change?” and “How severe would the bias(es) have to be to attenuate the estimate to the null?”. The toolkit includes the R packages PublicationBias, phacking, and multibiasmeta, as well as accompanying Shiny apps (which can be accessed at metabias.io). The PublicationBias and phacking packages provide analyses for the corresponding across-study biases, while the multibiasmeta package provides analyses for the joint effects of publication bias and various within-study biases. We demonstrate how to conduct these analyses both in R code and in Shiny apps.
GitHub repository: https://github.com/mikabr/pubbias-app https://github.com/mikabr/multibiasmeta https://github.com/mikabr/phacking https://github.com/mayamathur/Publica...