The statistical tasks have distinct owners: data-adequacy assesses what data can support; analyze plans and executes approved analyses; interpret explains existing results; and sensitivity examines specific changes in assumptions.
Choose a starting point
| What you have | Appropriate task |
|---|---|
| Raw data only | Inspect structure, missingness and adequacy |
| A need for primary analysis | Prepare a reviewable plan before execution |
| Existing outputs | Interpret effects, intervals and clinical importance |
| Concern about an assumption | Define the uncertainty and plan sensitivity work |
Interpret existing results
Provide effect scales, units, reference groups, confidence intervals, populations and model details. Clinical-importance discussion also needs prespecified, sourced thresholds; record missing thresholds rather than inventing them.
Interpret these existing results without refitting models. For each outcome, explain effect size and interval uncertainty, separating statistical evidence from clinical importance. State where a clinical threshold has not been established.
Check direction, units and reference groups. A nonsignificant finding is not automatically evidence of no effect, and a P value alone does not replace interpretation of effect estimates and uncertainty.
If additional analysis is needed
Identify the concrete question, such as a missing-data assumption or analysis-population change. Specify alternative assumptions, required inputs, outputs and execution conditions in a plan, then work within the approved scope. Preserve scripts, logs, diagnostics and result versions.
Hand off to writing
Deliver an outcome-level interpretation table, threshold questions, calibrated wording and proposed additional analyses. results writes from verified numbers; discussion considers design and evidence; figures-tables checks consistency across displays and prose.
Interpretation example · Analysis example · Manuscript writing
These synthetic examples are teaching materials. Researchers and statistical professionals must review outputs; tool results do not certify study conclusions.