Publication bias

Publication bias is the systematic difference between published research and all research conducted, often arising when results influence whether or how studies are published.

File drawer problem: The tendency for studies with null or unfavorable results to remain unpublished. Unpublished null studies accumulate outside the visible literature, creating a classic form of publication bias.
Meta-analysis: A statistical synthesis that combines results from multiple studies addressing a shared question. It can produce misleading pooled estimates when the studies available for synthesis are selectively published.
Preregistration: The public recording of a study's planned methods and analyses before its results are known. It makes planned studies visible even if their eventual findings are unremarkable.
Confirmation bias: The tendency to seek, interpret, or remember evidence in ways that favor existing beliefs. It shapes how evidence is interpreted, whereas publication bias shapes which evidence becomes available.
Egger's regression test: A statistical test for funnel-plot asymmetry in a meta-analysis. It detects one possible signal of missing studies, but asymmetry is not specific to publication bias.
Outcome reporting bias: The selective omission or alteration of study outcomes after results are known. Bias can arise within published studies when unfavorable outcomes are left unreported.
Systematic review: A structured review that uses explicit methods to identify and assess research on a question. Its search cannot fully recover studies that were never made public.
Registered report: A publishing format in which a study is reviewed before data collection, with publication based on the question and methods. Acceptance before results are known reduces incentives to publish only striking findings.
Selection bias: Systematic distortion caused by how individuals, observations, or studies enter an analysis. Publication bias is a research-literature form of this broader selection problem.
Trim-and-fill method: A meta-analytic method that estimates missing studies and recalculates a pooled effect. Its correction depends on assumptions about why studies are missing.

Linked from 45 pages