Picture yourself holding a study that found nothing. The randomisation was clean, the sample size adequate, the controls textbook. You ran the numbers three times. Zero. You now have roughly the same chance of publishing it in a top-tier journal as you have of getting a refund on a train that was merely late: the system is not designed to accommodate your particular kind of disappointment.
That outcome is not random. It is the product of internal governance decisions made by editors and review panels long before any specific manuscript arrived on their desks. The literature you see is not the literature that was produced. It is the literature that survived a filter, and the filter has bylaws.
The standing rules that shape the kill decision
Most journals operate with a two-stage structure. A handling editor, often a senior academic on the editorial board, makes a desk-rejection call within days of submission. If the paper clears that gate, it goes to two or three external reviewers. What most readers don't know is that the criteria the handling editor applies at that first gate are written down, debated, and voted on by the editorial board, sometimes annually, sometimes when a new editor-in-chief takes over. They function like standing policy.
Those criteria almost universally include some version of "novelty" and "impact." Both terms do real damage to null results. A study that confirms a drug doesn't work, or that a widely-cited correlation evaporates when you control for a third variable, is rarely described as novel by the people writing the checklist. It confirms and consolidates, where the policy demands that it advance. The policy doesn't say "reject null results." It says "prioritize novelty." The effect is the same, and that equivalence is not accidental.
Consider a worked example. Two researchers, call them Okonkwo and Lindqvist, run near-identical randomised trials testing whether a popular mindfulness app reduces cortisol levels in shift workers. Okonkwo's trial, conducted with 200 participants over twelve weeks, finds a statistically significant reduction of 11%. Lindqvist's trial, run the following year with 340 participants and tighter blinding protocols, finds an effect size indistinguishable from zero. Okonkwo publishes in six months. Lindqvist gets desk-rejected at three journals, survives one round of review at a fourth, and eventually deposits the paper in a preprint server where it accumulates 40 downloads. The governance detail that made the difference: the journal that rejected Lindqvist fastest had revised its editorial charter two years earlier to require that submissions "demonstrate a clear advance on the existing literature." Lindqvist's null result advanced the existing literature considerably. The charter didn't see it that way.
The reviewer incentive problem the board set in motion
External reviewers are typically unpaid, anonymous, and working under no formal time constraint beyond a soft deadline. Editorial boards set the culture of review by choosing who sits on their reviewer pools, how those reviewers are recruited, and whether reviewer behaviour is ever audited.
Most boards don't audit.
Boards that never track reviewer recommendations against eventual citation impact are boards that never learn whether their reviewers are systematically conservative about null results. The result is a known asymmetry: reviewers who recommend rejection of a null result face essentially no professional consequence. A reviewer who recommends acceptance of a null result that later turns out to be underpowered faces reputational exposure. The incentive structure is written nowhere but enforced everywhere, rewarding caution in one direction only. It is a thumb on the scale so consistent it has stopped feeling like a thumb.
Here is a question worth sitting with: if you have ever attended an editorial board meeting and watched the novelty criterion sail through a vote without a single dissenting voice, what exactly did you think you were ratifying?
What makes this a governance problem, distinct from a simple matter of human nature, is that boards have the tools to correct it and mostly choose not to deploy them. Registered reports, where journals commit to publish a study based on its methodology before results are known, have been adopted by several hundred journals. The effect on null-result publication rates at those journals is measurable and significant, with some estimates putting the share of null or mixed results in registered-report outputs above 60%, compared with roughly 5 to 10% in the standard literature at comparable journals. The mechanism works. Adoption remains slow because it requires a board to vote to constrain its own future editorial discretion, which is a vote most boards find uncomfortable. Institutional bodies seldom enjoy amputating their own authority, even a small piece of it.
The suppression of null results is sometimes framed as a bias problem, as if it were a collection of individual prejudices that better-intentioned people would overcome. It isn't, quite. It is a structural output of written and unwritten governance rules that boards enact, renew, and could revise. The literature you trust was shaped by meetings you were never in the room for, and the minutes were not circulated.