Keep methods reproducible and claims traceable

A methods section should let a qualified reader understand the population, materials, variables, procedures, exclusions, estimand, and analysis. Name choices that affect interpretation. Passive voice can be useful when the procedure matters more than the actor, but it should not hide who made a judgment or how a classification was assigned.

Link every reported estimate to the corresponding method. If a model changes across analyses, state the adjustment set and purpose. If missing data, multiplicity, measurement error, or sensitivity analysis affects the result, report it where readers can connect the issue to the estimate.

Separate observation from interpretation

Results sections should report direction, magnitude, uncertainty, and the comparison being made. Statistical significance alone is not an interpretation of practical or scientific importance. Give the estimate and interval where appropriate, preserve units, and avoid translating a non-significant result into proof of no effect.

SectionPrimary responsibilityCommon risk
MethodsDescribe design and analytic decisionsOmitting choices that shape the estimand
ResultsReport estimates and uncertaintyReplacing magnitude with significance labels
DiscussionInterpret within design limitsPresenting an association as a cause
ConclusionState the bounded contributionGeneralising beyond population or setting

Treat causal language as a design claim

Words such as caused, led to, reduced, improved, and prevented make claims about counterfactual change. They require more than a strong association. A credible causal statement depends on design, identification assumptions, temporal ordering, confounding control, measurement, and analysis. When those conditions are not met, use descriptive or associational language and state the remaining uncertainty.

Search terms such as cautious interpretation or observational association surface patterns whose evidence requirements can be checked before use.