AGENT SKILL

preregistration

Use when preregistering a study design (hypothesis, methods, analysis plan) before data collection on OSF Registries or AsPredicted, applying for ClinicalTrials.gov registration for interventional trials, or submitting a Registered Report.

License 许可
MIT
最近更新
2026-10-30
Stars
★ 8
Forks
2

01 前言 Front Matter

name
preregistration
license
MIT
description
Use when preregistering a study design (hypothesis, methods, analysis plan) before data collection on OSF Registries or AsPredicted, applying for ClinicalTrials.gov registration for interventional trials, or submitting a Registered Report.

03 SKILL.md 正文

<!-- metadata: category: open-science oversight: medium version: 1.0.0 author: Pradyumna Jayaram tags: [open-science, research] difficulty: intermediate -->

Preregistration

Hard rules

  1. No fabricated citations. Every cited work must resolve to a verifiable
  2. No claim without provenance. Every quantitative or factual claim
  3. No silent failure. Every script invocation, API call, or tool use must declare its exit status and what to do on non-zero. A skill that silently swallows errors is a violation.

When to use

  • Confirming a study design before data collection starts — to commit to analysis.
  • Responding to reviewers or funder concerns about p-hacking / analytic flexibility.
  • Required ClinicalTrials.gov registration for any interventional trial with a US site, NIH-funded, or FDA-regulated.
  • Submitting a Registered Report (peer review of the protocol before results).
  • Applying for a grant that asks for a preregistration plan.
  • Planning a replication study where the design must match the original.

When NOT to use

  • For preprints (posting a completed manuscript) — see ors-open-science-preprints.
  • For data deposits without a study plan — see ors-open-science-fair-data.
  • For code release (no hypothesis) — see ors-open-science-code-release.
  • For exploratory / pilot / feasibility work that is intended to generate hypotheses. Preregistration is for confirmatory research.
  • After data are already collected — preregistration is a pre-data commitment. You cannot preregister after the fact.
  • For purely observational / database / secondary-data analyses — these often use analytic pre-registration (specifying the analysis plan, not the data collection) on OSF.

Prerequisites

  • A complete study protocol (hypothesis, primary outcome, sample size, analysis plan).
  • ORCID iD for all authors (required for OSF and AsPredicted).
  • IRB/IACUC approval number (or proof of exemption) — most registries ask for it.
  • For ClinicalTrials.gov: a Unique Trial Identifier (NCT number, if an amendment) or a sponsor-designated protocol.

Core workflow

  1. Write the study protocol and analysis plan. Use a template (see "Document patterns" below). Be specific: sample size N, primary outcome variable, primary analysis model, alpha, power, correction for multiple comparisons.
  2. Choose the platform. Use the decision tree:
    • Basic: AsPredicted — simplest (one page, 10 questions).
    • Standard: OSF Registries — most comprehensive, searchable, indexed in Crossref.
    • Regulated: ClinicalTrials.gov — required by law for US interventional trials.
    • Publishing: Registered Report format via a participating journal.
  3. Upload to the platform. Fill in the structured fields; attach the protocol as a PDF.
  4. Timestamp. The platform mints a DOI and date stamps the record. This is your proof of priority.
  5. Collect the data. Do not look at the primary outcome until data collection is complete. No peeking (or disclose that you peeked).
  6. Run the analysis as specified. If the analysis must change, disclose the change and explain why. Post-hoc changes are allowed but must be disclosed in the paper.
  7. Link the preregistration to the paper. Include the DOI in the final manuscript; cite the preregistration in the methods. Check the box in the journal submission system.
  8. Update the record for any changes (protocol amendments, sample-size re-estimation, analytic modifications).

Document patterns

Pattern 1: OSF vs AsPredicted vs ClinicalTrials.gov decision tree

Is this a US-regulated interventional clinical trial?
├── YES → ClinicalTrials.gov (required by law)
│         • Interventional (drug, device, gene therapy, etc.)
│         • US site OR NIH-funded OR FDA-regulated
│         • Must register BEFORE first patient enrolled
│         • Required fields: NCT number, recruitment status,
│           primary/secondary outcomes, analysis plan
└── NO  → Is this for a Registered Report (journal publication)?
         ├── YES → Use the Registered Report format from the target journal
         │        (e.g., Cortex, Open Science Framework, BMJ Open Science)
         │        • Journal sends protocol out for peer review
         │        • In-principle acceptance conditional on methods
         │        • Results paper gets fast-track if methods are sound
         └── NO  → Does the field expect a one-page confirmation?
                  ├── YES → AsPredicted (simplest, 10 questions)
                  │       • Common in psychology, social sciences, behavioral bio
                  │       • Lightweight, fast (~15 min to complete)
                  │       • No DOI (but has a unique URL and timestamp)
                  └── NO  → OSF Registries (full-featured)
                        • DOI ( searchable in Crossref)
                        • 50+ fields, all searchable
                        • Templates for different designs (RCT, survey, etc.)
                        • The default for biology, neuroscience, most fields

Pattern 2: The minimum preregistration fields (any platform)

Every preregistration must include:

  1. Hypothesis (what you are testing)
  2. Primary outcome (exactly what you will measure)
  3. Sample size (N, or stopping rule)
  4. Inclusion/exclusion criteria (who is in the study)
  5. Analysis plan (the statistical model)

Optional but recommended:

  • Secondary outcomes (exploratory outcomes)
  • Power calculation (alpha, power, effect-size assumption)
  • Correction for multiple comparisons (correction method: Bonferroni, FDR, etc.)
  • Exclusion criteria (how you handle missing data, outliers)
  • Data exclusions (any QC criteria decided before seeing the data)
  • Analysis software (R, Python, JASP, etc.)

Pattern 3: OSF Registries preregistration template (abbreviated)

Most OSF fields map to this skeleton:

# Title
[Hypothesis in one sentence.]

# Hypothesis
[State the directional hypothesis: "We predict that X will increase Y..."]

# Primary outcome
[Name the variable; how it will be measured; units.]

# Secondary outcomes
[Optional: list up to 3 secondary exploratory outcomes.]

# Inclusion criteria
- Age range: [...]
- Population: [...]
- Other: [...]

# Exclusion criteria
- [...]
# Sample size
- Target N: [...]
- Power calculation: 80% power to detect d = [...] at α = 0.05

# Analysis plan
- Primary test: [e.g., linear mixed model with fixed effect of GROUP, random effect of SUBJECT]
- Covariates: [age, sex, baseline score]
- Alpha: 0.05 (two-tailed)

# Exclusion rules
- Any samples with < 80% missing data will be excluded.
- Outliers identified as > 3 SD from the mean will be flagged
  and sensitivity analyses run with and without them.

# Known deviations from the analysis plan
[None to date / Will be added if the protocol changes]

# IRB approval
- Protocol #[...] obtained from [Institutional Review Board name]
- Approval date: [...]

Pattern 4: AsPredicted (10 questions)

AsPredicted simplifies to 10 questions:

  1. What is the hypothesis? (1 sentence)
  2. What are the variables?
    • Independent variable(s):
    • Dependent variable(s):
  3. How many participants will you collect data from? (N, or description of stopping rule)
  4. When are you going to collect the data? (dates or "continuous")
  5. Is this a between- or within-subjects design?
  6. What analysis will you run to test the hypothesis? (specify the test, e.g., "t-test", "mixed-effects model")
  7. What does that analysis tell you?
  8. Any other analysis? (list any secondary/ exploratory)
  9. How can we contact you? (email, ORCID)
  10. Anything else?

Pattern 5: ClinicalTrials.gov required fields

For US interventional trials, you must register:

Field Description
Unique Trial Identifier NCT number (assigned on first save)
Brief Title ≤ 120 characters
Official Title Full title
Study Type Interventional
Primary Outcome Variable, measurement, time point
Secondary Outcomes Up to 10
Arms / Interventions Drug, device, procedure
Eligibility Inclusion/exclusion criteria
Contact / Sponsor Name, phone, email
Recruitment Status Not yet recruiting / Recruiting / Completed / etc.
Study Start Date Expected first enrollment
Primary Completion Date Expected last enrollment
Completion Date Expected final data collection
Summary / Description ≤ 5000 characters
Study Design Allocation, masking, intervention model
Masking Who is blinded
Intervention Model Single group, parallel, crossover
Number of Arms 1, 2, 3+
Primary Purpose Treatment, prevention, diagnostic, etc.

Register before first patient enrolled. Updates required within 21 days of a change (recruitment status, primary outcome, sample size). Results must be posted within 12 months of the primary completion date.

Pattern 6: Registered Report formats

Registered Reports come in three "formats" (not versions, but stages):

Format What is reviewed What is guaranteed
Stage 1 (Protocol) The study design, hypothesis, methods, analysis In-principle acceptance — if you run the study as specified, the results paper will be published (subject to minor review).
Stage 2 (Results) The results, with the protocol in the supplement Peer review of the full paper.
Stage 3 (Registered Report) Protocol + Results in one submission Two-stage review — protocol reviewed first, then results after data are collected.

Most journals offer Stage 1 as the default. The workflow is:

1. Submit protocol (preregistration + analysis plan)
2. Stage 1 peer review → in-principle acceptance (IPA) or revise and resubmit
3. Collect data (do not look at results yet)
4. Submit results paper with protocol in supplement
5. Stage 2 peer review → publication OR major concerns (rare if IPA was given)

Journals that offer Registered Reports include:

  • Cortex (and other Center for Open Science partner journals)
  • eLife (exceptions apply)
  • BMJ Open Science
  • F1000Research
  • Royal Society Open Science
  • Open Science Framework journals

Common pitfalls

Pitfall Why it fails Fix
Preregistering after data collection Not allowed; it's post-hoc, not pre-registration Preregister BEFORE any data are touched.
Changing the analysis without disclosure Reviewers consider this undisclosed p-hacking Any post-hoc changes go in the paper's limitations section with explanation.
AsPredicted for a clinical trial Not legally sufficient for FDA/NIH registration Use ClinicalTrials.gov for US interventional trials.
No sample-size justification Underpowered studies are the norm; reviewers ask Include a power calculation (G*Power, R, or simulation).
Vague primary outcome "We will measure gene expression" is not specific "mRNA levels of gene X in peripheral blood monocytes, measured by qRT-PCR, normalized to GAPDH."
Not disclosing peeked data Reviewers assume you looked; they will ask in review Write: "One interim analysis was conducted at N = X; the study continued as planned because..."
No IRB number Most journals require it; will be asked in review Get IRB approval before preregistering.
Confusing preregistration with clinical trial registration Different requirements, different platforms Clinical trials go to ClinicalTrials.gov.
Registered Report rejected at Stage 1 The design or analysis was not sound Revise per reviewer feedback; don't skip to a Stage 2 submission elsewhere without disclosing.
"We did not preregister" in a preregistered study The journal's check box will catch this Either preregister or say "preregistered, DOI: ..." in the methods.
Including exploratory outcomes as primary This is p-hacking; reviewers see through it Distinguish primary (confirmatory) from secondary (exploratory). The primary outcome gets the formal alpha; the rest are treated as hypothesis-generating.
Using an unregistered change as a limitation The fix is to disclose, not hide Add a section "Deviations from the preregistered protocol" in the paper.

Validation

A preregistration is complete when:

  • [ ] The hypothesis is stated as a directional prediction (not a question).
  • [ ] The primary outcome is defined with exact variable name, measurement method, and time point.
  • [ ] The sample size is justified with a power calculation (or an explicit Bayesian stopping rule).
  • [ ] The analysis model is named (e.g., "linear mixed model", "two-sample t-test", "Cox proportional hazards").
  • [ ] The platform has minted a timestamp and (for OSF) a DOI.
  • [ ] All co-authors are listed and have approved the submission.
  • [ ] The IRB/IACUC number is included.
  • [ ] The preregistration is linked in the final manuscript (DOI in the methods).
  • [ ] Any changes from the plan are disclosed in the final paper.

Open alternatives

Commercial / restricted service Open alternative Trade-off
ClinicalTrials.gov (required, free) — Required for US trials; no alternative.
AsPredicted (free) OSF Registries (also free) OSF is more comprehensive; AsPredicted is simpler.
PROSPERO (systematic reviews) OSF Preregistration for systematic review protocols PROSPERO is domain-specific for systematic reviews; OSF is general.
REDCap (clinical data capture) OpenClinica, Castor EDC REDCap is free for academics; OpenClinica and Castor are also free for academics.
SurveyMonkey (surveys) Qualtrics (institutional license), Limesurvey (open-source) Limesurvey is self-hosted, free; SurveyMonkey has a free tier.
Dropbox Paper (collaborative writing) HackMD / HedgeDoc, Quarto, Manubot Manubot is git-native, versioned.

References

Related skills

  • ors-open-science-preprints — to post the results after data collection.
  • ors-open-science-fair-data — to deposit the data after the paper is published.
  • ors-open-science-code-release — to release the code used in the analysis.
  • ors-omics-statistics- (multiple skills) — for statistical analysis methods.
  • ors-bioinformatics-clinical- (multiple) — for clinical data handling.

Changelog

  • 1.0.0 (2026-06-10): Initial adaptation by Pradyumna Jayaram. Synthesised OSF Registries help; AsPredicted FAQ; ClinicalTrials.gov fields; SPIRIT 2013; Registered Reports info (cRedi); NIH analytic guide; US OSTP memo. OSF vs AsPredicted vs ClinicalTrials decision tree; OSF template skeleton; AsPredicted 10 questions; Registered Report format comparison are original compositions.

Cross-references

Other skills in this category:

  • code-release
  • fair-data
  • licensing
  • preprints

04 同仓库更多 Skills