MethodsPublished 24 Jul 2026 · Reviewed 24 Jul 2026
Before calculating sample size, define the estimand
A sample-size formula cannot rescue an ambiguous population, outcome, contrast, time point, or handling of intercurrent events.
Prepared by: ProWrite Research Standards Desk
Review record: Internal source review completed; named independent expert review not yet published
- Write the population, treatment or exposure conditions, outcome, time point, and population-level contrast before choosing inputs.
- Distinguish a detectable effect from the smallest clinically meaningful effect.
- Record the source and uncertainty for every event rate, variance, correlation, loss-to-follow-up, and design-effect assumption.
- Plan sensitivity scenarios rather than presenting one number as certain.
Correction history
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
BiostatisticsPublished 24 Jul 2026 · Reviewed 24 Jul 2026
Five statistical claims a p-value cannot support
Statistical significance alone does not establish clinical importance, causality, absence of bias, model validity, or replicability.
Prepared by: ProWrite Research Standards Desk
Review record: Internal source review completed; named independent expert review not yet published
- A small p-value does not measure the size or importance of an effect.
- A non-significant result is not proof that two conditions are equivalent.
- Association does not become causal because confounders were entered into a model.
- Model output is not reliable until assumptions, missingness, multiplicity, and data quality are addressed.
Correction history
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
Research defensePublished 24 Jul 2026 · Reviewed 24 Jul 2026
What a resident should be able to teach back before defense
A defensible project requires the researcher to explain why the design and analysis answer the question—and where they do not.
Prepared by: ProWrite Research Standards Desk
Review record: Internal source review completed; named independent expert review not yet published
- State the research gap, primary objective, and primary outcome in plain language.
- Explain the sampling pathway and the most important selection-bias risk.
- Interpret effect estimates and confidence intervals before discussing p-values.
- Name limitations, their likely direction, and the claims that should therefore be avoided.
Correction history
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
PrivacyPublished 24 Jul 2026 · Reviewed 24 Jul 2026
De-identification is a risk review, not a single delete command
Removing names is necessary but may not be sufficient when dates, rare conditions, geography, free text, or linked variables can identify a person.
Prepared by: ProWrite Privacy and Security Contact
Review record: Internal source review completed; named independent expert review not yet published
- Start with authority, purpose, recipients, environment, and minimum necessary fields.
- Separate direct identifiers and linkage keys from the analytical working copy.
- Review combinations of indirect identifiers and small cells.
- Record residual risk, access rules, retention, and the person who approved release.
Correction history
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
Responsible AIPublished 24 Jul 2026 · Reviewed 24 Jul 2026
Responsible AI disclosure needs purpose, location, and verification
Naming a tool is not enough. Authors should document what it did, where its output appears, and how humans verified accuracy and originality.
Prepared by: ProWrite Research Standards Desk
Review record: Internal source review completed; named independent expert review not yet published
- Do not list an AI system as an author.
- Do not upload confidential manuscripts or identifiable data where confidentiality is not assured.
- Verify factual claims, calculations, citations, images, and wording against authoritative sources.
- Follow the target journal and institution even when their required disclosure is stricter.
Correction history
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
