Studies finding no effect are submitted less often and accepted less often than studies finding one.
Studies finding no effect are submitted less often and accepted less often than studies finding one. The result is that the published literature overstates effects systematically, before any individual author does anything wrong. It is a property of the system rather than of the papers in it.
The missing studies are, by definition, not there to count. Detecting the bias means inferring the shape of what is absent from the distribution of what is present, which is why the standard checks are indirect and imperfect.
Plotting effect size against precision should produce a symmetric scatter if nothing is missing. Asymmetry suggests small studies with null results are absent. The test is weak when there are few studies, and asymmetry has other causes, so it is a prompt to think rather than a verdict.
Requiring trials to be registered before they begin creates a list of studies that should exist. Comparing registrations with publications reveals what was never written up, and that comparison is now itself a research literature with uncomfortable findings.
A study can be published and still hide results, by reporting the outcomes that worked and omitting those that did not. This is harder to detect than non-publication and is the reason registered outcome lists matter as much as registered studies.
Funding source correlates with published outcome, but academic groups face their own pressure toward positive findings. Treating publication bias as a commercial problem alone misses most of it.
Any synthesis of published work inherits the bias of what was published. A meticulous review of a selected literature produces a precise summary of a distorted sample, which is more dangerous than an obviously sloppy one.
Check whether a review searched trial registries and grey literature rather than journals alone, and whether it reports a funnel plot or similar check. Reviews that did neither are reporting the published record rather than the evidence.
The claims on this page about the literature is a selected sample before anyone starts reviewing it can be checked against sources outside this site: