Meta-Regression Methods for Publication Selection Bias: Simulations and Heckman Regression

Publication selection (PS) exists when editors, referees, or researchers have a preference for statistically significant results, successful clinical trials or for confirmation of their favored theory. Medical researchers have long been concerned about the potentially pernicious effects of PS (e.g., deaths associated with taking Paxil and Vioxx). Register or perish! As a consequence, the best medical journals now have explicit PS policies (Krakovsky, 2004). In economics, the biases produced by publication selection often exceed the magnitude of the underlying empirical phenomenon being estimated. In a forthcoming paper in the Oxford Bulletin of Economics and Statistics, I offer meta-regression methods that address the three fundamental issues of publication selection: • How can publication selection be detected? [FAT—Funnel Asymmetry Test] • How can an empirical effect be identified, regardless of publication selection? [PET—Precision Effect Test] • How can the magnitude of the effect be estimated in a manner that is robust to publication selection? [PEESE—Precision Effect Estimate with Standard Error