Between 80 and 95% of published papers report positive results. The studies that found nothing sit in a file drawer — and the next lab pays to rediscover them.
Publication bias is the tendency to publish positive results and shelve null ones — an estimated 80–95% of published papers report a positive finding. The file drawer problem is its consequence: the studies that found nothing sit unpublished, so the literature shows what worked and hides what didn't.
Publication bias is the systematic tendency for studies with positive, statistically significant results to make it into journals, while studies that found no effect quietly don't. The numbers are stark: across fields, an estimated 80–95% of published papers report positive results. The world doesn't run that cleanly — most well-designed experiments produce a mix of confirmations and null findings. The gap between what is run and what is published is the bias.
The effect is cumulative. Each missing null result is a small distortion, but stacked across a literature they bend the consensus toward effects that are weaker — or absent — than the published record implies.
The file drawer problem is the mechanism behind the bias, named by the psychologist Robert Rosenthal in 1979. His point was uncomfortable: for every published study showing an effect, there may be many unpublished ones — sitting in researchers' file drawers — that found nothing. We can see the published tip; we can't see how big the submerged body of null results actually is.
The published literature is a survivorship sample. It records the experiments that worked and forgets the ones that didn't — which is exactly the information the next experimenter needs.
Incentives, all the way down. Journals compete for novel, positive, citable findings, so a clean "no effect" is a hard sell. Reviewers scrutinise null results more harshly. And authors, anticipating all of that, simply don't write up the negative study — self-censorship that never even reaches an editor's desk.
Underneath sits the same structural gap we keep hitting: there is no slot, no citation, and no career credit for "here's what didn't work." So the knowledge never enters the shared record. It lives in a notebook, a Slack thread, or a postdoc's memory, and it leaves when they do.
Publication bias is not just a publishing nuisance — it directly feeds the reproducibility problem. Meta-analyses built only on published, positive studies overestimate how strong an effect really is. Replication attempts burn months re-testing dead ends that other labs already ruled out privately. The negative results that would have saved that effort are the exact thing the literature discards.
At the bench, the leak is concrete. The reason a step only works below 8 °C, the antibody lot that gave nonspecific bands, the buffer that suppressed the signal — these are negative results too, and they almost never make it into a methods section. The next lab inherits the polished path that worked and none of the wrong turns. We make the case for fixing that in why negative results belong in your methods.
The field's answers operate upstream of publication. Preregistration and registered reports — where a journal commits to publishing a study based on its design, before anyone knows how it turned out — break the link between a positive result and acceptance. Dedicated venues for null findings help too. These are real, and they matter.
But there's a layer publishing reform doesn't reach: the method itself. On jnar, a negative result is a first-class object you attach to a protocol — the conditions, the observation, and your ORCID — searchable and versioned alongside the method. When a failure changes how the protocol is done, it shows up in the fix history instead of vanishing into a drawer. That doesn't abolish publication bias — but it keeps the methods-level negative results, the ones a replicator needs most, from being lost in the first place.
Publication bias is the tendency for studies with positive, statistically significant results to be published more often than studies that find no effect. Because journals, reviewers, and careers reward confirmatory findings, an estimated 80–95% of published papers report positive results — so the literature systematically over-represents what worked and under-represents what didn't.
The file drawer problem, named by psychologist Robert Rosenthal in 1979, is the flip side of publication bias: studies with null or negative results never get written up or accepted, so they end up unpublished in researchers' file drawers. The published record is the visible tip; an unknown body of null findings stays hidden.
Incentives. Journals favour novel, positive findings; reviewers are harder on null results; and authors anticipate rejection so they don't submit negative studies in the first place (self-censorship). There is no citation, slot, or career credit for 'this didn't work', so that knowledge never enters the shared record.
Partly because journals rarely accept them, and partly because researchers expect that and don't bother writing them up. The result is the same: a negative result that would save another lab months of wasted effort stays in a notebook or a file drawer instead of being shared.
It distorts the evidence base. Meta-analyses that only see published, positive studies overestimate effects; replication attempts waste effort re-testing dead ends that other labs already ruled out privately. The missing negative results are exactly the information a replicator needs most — and they're the part the literature throws away.
At the field level: preregistration and registered reports (where a journal accepts a study based on its design, before the results are known) and dedicated venues for null findings. At the methods level: treat negative results as first-class, attributed, versioned data attached to the protocol itself — so what didn't work travels with the method instead of disappearing.
