Two teams both report that an intervention increased student motivation, p = 0.03. The first team wrote down before collecting data which motivation scale they would use, which test they would run and whom they would exclude. The second tried four scales, two outlier rules and three models, then reported the combination that looked best. On paper the results are identical. As evidence, they are worlds apart.
Preregistration lets readers see which team you are without having to take your word for it.
What preregistration is, and is not
A preregistration is a time-stamped research plan, stored publicly or under embargo until a set date, created before you collect or look at the data. It is not:
- A promise never to explore. You can, as long as exploratory analyses are labelled as such.
- A vow never to change the plan. You can, as long as you report it transparently.
- Only for experiments. Secondary data analyses and qualitative studies have their own templates.
What to include
The popular templates on OSF and AsPredicted revolve around these questions:
| Section | Specific enough | Too vague |
|---|---|---|
| Hypothesis | Students in the paired-learning condition score higher on the final test than those studying alone | Paired learning affects outcomes |
| Dependent variable | Score on a 40-item test, range 0–40 | Learning outcomes |
| Sample size and stopping rule | 120 students; stop when reached or on 30 November; based on a power analysis for d = 0.5 | As many as possible |
| Exclusions | Exclude completions under 5 minutes or failing both attention checks | Exclude poor-quality data |
| Primary analysis | Linear regression controlling for baseline score, two-sided, α = 0.05 | Appropriate statistics |
The test of sufficient detail: someone else holding your preregistration and your data should arrive at exactly one result, without having to choose between options.
Where to register
- OSF Registries: many templates for different designs, with embargoes of up to four years.
- AsPredicted: a short eight-question form suited to small experiments.
- Clinical trial registries such as ClinicalTrials.gov or those in the WHO network: effectively mandatory for trials, and many medical journals will not consider an unregistered trial.
- PROSPERO: for systematic review protocols with a health-related outcome.
Registered Reports: peer review before results exist
Registered Reports go a step further. You submit the introduction, methods and analysis plan before collecting data. If Stage 1 review goes well, the journal grants in-principle acceptance: it commits to publish regardless of the results, provided you follow the plan. Stage 2 review checks only whether you did what you said and whether the interpretation is sound.
For doctoral researchers the attraction is obvious: a null result still becomes a paper. The trade-off is waiting for Stage 1 review before data collection, which has to be built into the thesis timeline.
When you have to deviate
Almost every study deviates somewhere: a school withdraws and the sample shrinks, a scale shows poor reliability, the data violate an assumption. The right response:
- Record what changed, when and why, ideally before running the affected analysis.
- Include a “Deviations from preregistration” table in the paper: original plan, what you actually did, the reason, and the likely impact on results.
- Where possible, report both the planned and the adjusted analysis.
Explained deviations do not weaken a paper. Silent ones, discovered when a reviewer compares the article with the registration, damage trust in everything else.
Common myths
- “Someone will steal my idea.” You can embargo the registration until publication, and the timestamp helps prove you had the idea first.
- “My study is exploratory, I have no hypotheses.” You can still register data collection, exclusion rules and research questions. Declaring a study exploratory is itself a form of transparency.
- “It takes too long.” Most of the content already exists in your proposal or ethics application. The real extra effort is thinking the analysis through, which you should do anyway.
- “You can't preregister secondary data.” You can, with a statement of what you have already seen in the dataset.
A three-session workflow
- Session one: write hypotheses and variables in measurable terms. Draft a hypothetical results sentence: “Group A scored x points higher than group B.”
- Session two: write your analysis code and run it on simulated data. This is the fastest way to expose vague spots in the plan.
- Session three: ask a colleague to read it and keep asking, “If the data look like this, what will you do?” Every time you answer “it depends”, you have found something to specify.
Attach the code that runs on simulated data to the registration if you can. Readers will not need to guess what you meant.
Your next step: open the AsPredicted form and spend thirty minutes answering its eight questions for your next study. Whatever you cannot answer is a gap in your design, whether or not you ever submit the registration.
Câu hỏi thường gặp
Is preregistration mandatory?
For clinical trials it effectively is. For other research most journals encourage rather than require it, though a growing number of journals and funders favour preregistered studies.
Can I preregister after collecting data?
You can, if you have not analysed the data and you state clearly when registration happened relative to data collection. The evidential value is lower than registering beforehand.
What is the difference between preregistration and Registered Reports?
Preregistration means you archive your own plan. In a Registered Report the journal peer-reviews the plan before data exist and commits to publish regardless of results, provided you follow it.
Is an OSF preregistration public immediately?
Not necessarily. OSF lets you embargo a registration for a period, after which it becomes public automatically.
Can qualitative research be preregistered?
Yes. Qualitative templates cover research questions, sampling, data collection and analysis approach, and the researcher's own position.
