What is the replication crisis, and what is being done about it?

The finding that a substantial proportion of published research findings cannot be reproduced when the studies are repeated — documented most prominently in psychology and medicine, and subsequently found across many fields.

What the evidence showed. Large coordinated projects attempting to replicate published studies found that a considerable share failed to reproduce the original result, and that effect sizes in successful replications were frequently much smaller than originally reported. Similar concerns arose independently in preclinical biomedical research, where industry attempts to build on published findings failed at high rates.

The causes, which are structural rather than a matter of individual dishonesty:

Publication bias. Journals preferentially publish positive, novel findings. Null results go unpublished — the file drawer problem — so the literature systematically overstates effects.

p-hacking. Analysing data many ways and reporting the analysis that reached significance. Frequently done without any intent to deceive, because the choices feel reasonable individually.

HARKing — hypothesising after the results are known, presenting an exploratory finding as a prediction that was tested.

Small samples, which produce imprecise estimates and — counterintuitively — mean that significant findings are more likely to be overestimates.

Incentives. Careers depend on publications in prestigious journals, which want striking results. Replication was until recently neither publishable nor rewarded.

What is being done:

Pre-registration, publishing hypotheses and analysis plans before collecting data, which removes the flexibility p-hacking depends on.

Registered reports, where journals accept a study on the basis of its method, before results exist — so publication cannot depend on the outcome. This is the most direct structural fix.

Data and code sharing, allowing reanalysis.

Larger samples and multi-site collaborations.

Reporting standards and journals accepting replications and null results.

Statistical reform, including attention to effect sizes and confidence intervals rather than significance alone.

Why it should increase rather than reduce confidence in science. The crisis was identified by researchers, using scientific methods, and has produced systematic reform. A field that could not detect its own errors would be in a worse position, not a better one.