Publication Bias

Publication bias is the systematic tendency for studies with a positive, statistically significant, or novel result to get published, while studies that find no effect sit unpublished — psychologist Robert Rosenthal called this the "file drawer problem" in 1979, since a null result is exactly the kind of finding that ends up in a drawer rather than a journal.

The mechanism runs at every stage of the pipeline: journals prefer to publish exciting findings because that's what gets cited and read, researchers are less motivated to write up and submit a null result because it's a harder sell and career rewards favor novelty, and a null finding is, correctly, less newsworthy on its own — though the aggregate absence of null results is exactly what distorts the record. The effect compounds statistically: if twenty independent labs test a hypothesis that's actually false, ordinary chance means roughly one of them will still turn up a "significant" result at the conventional threshold, and that's the one that gets published, while the other nineteen null results disappear from view — so the literature can look like a clear positive finding when it's actually noise plus a filter.

The distortion is visible and quantifiable: meta-analyses use a funnel plot, which should show a symmetric spread of study sizes and effect sizes if no bias operates, and a lopsided funnel — missing small studies with null or negative results — is the classic fingerprint of a filtered literature. Preregistration, where a study's hypothesis and analysis plan are locked in before data collection, is the standard fix, because it makes it possible to notice how many registered studies never got a paper. Publication bias is a structural, non-fraudulent partner to P-Hacking: p-hacking distorts individual studies from the inside, while publication bias distorts which studies the world gets to see at all, and the two compound each other in exactly the same direction.