Preregistration as Cheap Talk: The Logic and Limits of Self-Enforcing Plans in Scientific Research

Preregistration—announcing a research plan before collecting data—is widely promoted as a remedy for p-hacking: running many analyses and reporting only the significant ones. But plans are not enforced: if researchers face no penalty for ignoring them, why should announcing one change anything? We show that announcing a plan can still change behavior: when a reviewer knows which tests were planned, reporting unannounced tests reveals that the planned tests failed, so reviewers discount it. Yet the researcher designs her own plan, and broad plans preserve flexibility for selective reporting while still conferring credibility. Preregistration can therefore increase the rate of false published results, though not always. Making preregistration reliably beneficial requires that reporting all preregistered tests become the norm.

2026