How the Replication Crisis Rewired the Grant Cycle

For most of the twentieth century, a grant application lived or died on novelty. The implicit contract between investigator and funder was straightforward: produce a surprising result, publish it, move on to the next surprising result. Reproducibility was assumed, not verified, and certainly not funded.

That assumption collapsed publicly and loudly. A series of large-scale replication efforts, most visibly the Reproducibility Project in psychology and parallel efforts in cancer biology, returned sobering numbers. Depending on how you defined successful replication, somewhere between one-third and two-thirds of published findings failed to hold under independent testing. The variance in those estimates is itself informative: researchers could not even agree on what replication meant, which told you something about how little the field had systematized the concept.

Funding agencies were not passive observers. The National Institutes of Health, the largest biomedical research funder in the United States, began revising application requirements starting around 2014 and 2015. The changes were structural rather than cosmetic. Investigators were now required to address rigor and reproducibility directly in the application narrative, to pre-specify statistical approaches, to report the sex of biological subjects as a variable rather than an afterthought, and to demonstrate that key reagents, particularly cell lines and antibodies, had been authenticated. None of this guaranteed better science, but it made the scaffolding visible to reviewers in a way it had not been before.

The shift in reviewer culture followed, unevenly. Study sections began penalizing applications that relied on a single pilot experiment as proof-of-concept, particularly when that pilot was underpowered. The old practice of running twelve subjects per arm to hit a p-value below 0.05 became harder to defend in review, not because the statistics had changed, but because reviewers were more likely to ask what the power calculation assumed and whether the effect size was biologically plausible or optimized for significance.

Private funders moved differently. Philanthropic science funders with faster grant cycles began explicitly bankrolling replication work as a category, not as a consolation prize for failed original research but as a first-class scientific activity. This was a genuine cultural shift. Journals had historically resisted publishing replications, and investigators had little career incentive to run them. Dedicated replication funding created a small but real market for the work.

The problems that remain are structural and largely unsolved. Most funding cycles still reward the lab that discovers over the lab that confirms. Career review committees at universities still count first-author publications in high-impact journals, and a clean replication in a mid-tier journal does not substitute for a Nature paper in a tenure file. Funding agencies changed the application form; they did not change what happens to a scientist who spends five years replicating rather than generating.

There is also a selection pressure problem that does not show up in any grant announcement. Labs with strong preliminary data and clean effect sizes are better positioned under the new rigor requirements than labs working on genuinely uncertain questions in hard-to-model systems. Reproducibility requirements, applied without context, can favor the tractable over the important.

The replication crisis did something real to funding policy. It moved reproducibility from the methods section into the scoring rubric. Whether that is enough to change what science gets done, as opposed to how applications describe the science, is a question the next decade of replication data will have to answer.