In registered reports, researchers submit a design for review; the journal decides whether to publish before results are available. How and when does the registered report approach improve on pre-registration and standard post-implementation peer review? And how should researchers make use of this new way to publish research? We consider these questions in light of a Bayesian learning model that incorporates researcher incentives. By withholding the publication decision until the study is completed, the conventional review process gains access to more information (the finding, but also details of the design as implemented) and limits moral hazard risk. But it may also encourage distortion (e.g. p-hacking) by publication-seeking researchers, reducing what can be learned. Registered reports help address this, and may also improve the average design of implemented projects both by selecting promising designs for implementation and generating more feedback for authors at the design stage. We conclude with advice for researchers, editors, and reviewers about writing and assessing registered report submissions.
