File Drawer Effect

The file drawer effect happens when studies with no clear results are hidden away and not shared. This makes the set of public studies look more positive than all studies actually done.

The file drawer effect refers to the systematic suppression of null or negative results from the public literature, producing biased sample of reported findings. This selective non-publication distorts meta-analytic inference and inflates apparent effect sizes across a research domain.

A company tests a new painkiller in twelve small trials. Eight show the drug barely works, but four show strong results. They publish only the four positive trials. Doctors reading the literature think the drug is highly effective, when in reality most tests found it unremarkable.

A meta-analyst pooling randomized controlled trials on a cognitive-behavioral intervention for anxiety finds a pooled Cohen's d of 0.65 across 24 published studies. Applying a funnel plot asymmetry test (Egger regression intercept p = 0.003) and a trim-and-fill procedure reveals an estimated 11 missing studies in the lower-left quadrant, reducing the adjusted effect size to d = 0.38—well below the clinically meaningful threshold of 0.50. Comparison with a prospective trial registry shows 19 registered studies that never produced a publication, confirming that selective non-publication inflated the apparent effect by roughly 70% and that meta-analytic inference based solely on the visible corpus is materially misleading.

Researchers prefer sharing studies that look successful, so null results stay unseen. Journals and reviewers also favor striking findings, making publication uneven.

Selective reporting arises from asymmetric incentive structures and editorial thresholds that weight novelty and statistical significance more heavily; these constraints bias submission and acceptance flows. Structural elements like journal scope and peer review criteria create propagation asymmetry that suppresses null-result visibility.

Share all study results openly, even if they seem boring. Use places that accept null results so they appear in the public record.

Implement registered reports and mandatory data repositories to ensure all outcomes enter the literature and reduce publication bias. Reform editorial incentives to reward methodological rigor over novelty.

Overestimated effect sizes; Misleading meta-analyses; Wasted replication efforts

A bad actor funding applied research (e.g., pharmaceutical, tobacco, or chemical industries) can deliberately run many small trials, selectively publish only those producing favorable outcomes, and bury null or adverse findings—engineering a literature that systematically overstates efficacy or safety. Regulatory bodies and systematic reviewers relying on published corpora will then draw biased conclusions without ever knowing the suppressed data exists. This mechanism can be further weaponized by controlling who holds the raw data, making independent auditing structurally impossible.

Pre-registration of study hypotheses, methods, and analysis plans before data collection creates an auditable record that exposes non-publication; registries such as ClinicalTrials.gov operationalize this at scale. Registered reports, in which journals commit to publication contingent on methodological quality rather than outcomes, directly sever the incentive link between result valence and visibility. Funnel plot asymmetry tests, trim-and-fill methods, and the Egger test applied during meta-analysis can statistically estimate and partially correct for publication bias when full study registers are unavailable.