Pre-registration
Declaring what you will measure and how before you look at the data, which is what separates a test of a hypothesis from a search for one.
When not to use it
- For genuinely exploratory work, which should be labelled exploratory rather than dressed as confirmatory.
- As a substitute for good design. Registration constrains reporting, not the quality of the question.
- As proof of independence. A registered study by an interested party is still a study by an interested party, and both facts matter.
Reach for something else instead
- Registered reports — peer review before data collection, with acceptance independent of the result.
- Held-out evaluation — release the test set only after the analysis plan is fixed, which enforces the same discipline structurally.
- -
Further reading
- Ricciardi et al. (2025), The COMPARE Study, Annals of Surgery — a PROSPERO-registered, PRISMA-following synthesis, which is what makes interested-party research assessable.
- Bouthillier et al. (2021), Accounting for Variance in Machine Learning Benchmarks — why analytic choices made after seeing results move conclusions.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Treating registration as a quality mark without comparing registered and reported outcomes.
- Writing a plan vague enough to permit any analysis, which technically registers and practically does not.
- Omitting what would count as a negative result, which is the commitment that does most of the work.
At a glance
Where this sits
A starting point. Nothing needs to come before it.
Computed from the prerequisite graph, not assigned. How this works