1. Quick Summary
Systematic attempts to reproduce published results succeeded less often than expected, particularly in psychology, biomedicine and the social sciences.
The response has moved past debate about whether there is a problem and into changes to method and incentives, though the changes are uneven across fields.
2. What It Means
Replication means repeating a study to see whether the finding reappears. Failure to replicate does not automatically mean the original was wrong, but repeated failure is strong evidence.
Many causes were identified: small samples, flexible analysis choices, publication bias toward positive results, and incentives that reward novel findings over reliable ones.
None of these require misconduct. Ordinary practice under competitive incentives produces the same problems.
3. Why It Happens
Statistical power was often low. Small samples produce unreliable estimates, and unreliable estimates produce findings that do not reproduce.
Analytical flexibility matters: when many ways of analysing a dataset are possible, choosing after seeing the results inflates apparent effects.
Publication bias distorted the literature, because studies with null results were less likely to be written up or accepted.
Pre-registration addresses flexibility directly by committing to an analysis plan before data collection.
Registered reports go further, with review happening before results exist, which removes the incentive to obtain a particular outcome.
Reform is uneven because disciplines differ in method and in how much the findings depend on context, and because incentives change slowly.
4. Real Examples
Large replication projects: coordinated efforts repeating many published studies at once.
Pre-registration: publicly recording hypotheses and analysis plans in advance.
Registered reports: peer review of the method before results are known.
Open data and materials: sharing what is needed for others to check or reuse the work.
Null-result publication: journals and repositories that accept findings of no effect.
5. How It Affects Us
Research practice: planning and reporting standards have become stricter and more transparent.
Publishing: journals increasingly expect data sharing and clearer methods.
Funding: some funders now support replication work explicitly, which was once hard to finance.
Public trust: acknowledging and fixing problems is more durable than defending the status quo.
6. Key Takeaways
- Repeated failure to reproduce is evidence about the literature, not proof of individual error.
- Low power, flexible analysis and publication bias are the main mechanisms.
- Pre-registration and data sharing are concrete, effective responses.
- Reform is real but uneven, and incentives are the slowest thing to change.