Speaker: Lifeng Lin, Assistant Professor, Florida State University
Title: Innovative statistical methods for assessing publication bias
Abstract:
Publication bias (PB) frequently appears in many research fields, and it seriously threatens the validity of evidence synthesis. This talk presents several innovative statistical methods for assessing PB. On the one hand, examining funnel plots’ asymmetry has been popular to investigate potentially missing studies and PB direction. Most funnel plots present treatment effects against their standard errors (SEs). The contours depicting studies’ significance levels can be added to distinguish PB from other confounders. However, some effect measures (e.g., odds ratios) are associated with their SEs even if no PB exists. As such, SE-based funnel plots likely lead to false-positive conclusions when such association is nonnegligible. We present a Bayesian approach to examining PB with controlled false-positive rates. We also introduce contours for sample-size-based funnel plots to supplement the appraisal of PB. On the other hand, although many statistical tests have been proposed to detect PB, they make dramatically different assumptions about the cause of PB. Therefore, they are powerful only in certain cases that support their particular assumptions, while their powers may be fairly low in many other cases. We propose a hybrid test for PB by incorporating various tests’ benefits to maintain relatively high powers across different mechanisms of PB.
