Extending Egger's Regression: Detecting Outcome Reporting Bias in Meta-Analysis of Effects on Multiple, Dependent Outcomes

• In Event: AERA Poster Session 4
  In Poster Session: Division D Section 2 Poster Session 1

Abstract

Egger's regression model was recently modified (Pustejovsky & Rodgers, 2018) to preserve nominal Type I error rates although its power to identify publication bias was low. Outcome reporting bias (ORB) is a form of publication bias resulting from primary study authors’ reporting results for more (statistically) significant outcomes. ORB can occur when primary studies report effect sizes for multiple distinct but related outcomes. We extended the modified Egger’s regression test to test ORB while handling the dependence from multiple effect sizes per study. The extension enhances power by allowing simultaneous ORB tests using the full meta-analytic dataset rather than analyzing each outcome’s effect sizes separately. We compared methods for handling within-study dependence and the extension versus separate per-outcome Egger’s tests.

Authors

• Sunyoung Park, California Lutheran University

• Melissa Angelina Rodgers, American Institutes for Research

• Tasha Beretvas, The University of Texas at Austin