Create an outcome-to-output map
List each research objective beside the corresponding outcome, analysis output, table and manuscript paragraph. Mark an objective with no reported result, and a result with no explained purpose, for discussion. This map helps your team find omissions without quietly changing the planned study. Keep the original analysis files and note which version generated the numbers used in the manuscript.
Explain the analysis population
Track the numbers screened, included and analysed, with reasons for differences. For each percentage, identify the denominator and check whether missing observations changed it. In the invented example below, 30 responses among 80 people with available data is 37.5%; using all 100 enrolled people would answer a different question. The author must explain which population the reported percentage represents.
Make methods and assumptions traceable
For each analysis, check that the reported method, variables, units and group definitions match the output. Record questions about pairing, repeated observations, adjustment or assumptions for the analyst. Explain how missing data were handled. The SAMPL guidance is useful for checking whether the statistical description is sufficiently explicit; the correct method still depends on the design and data.
Report magnitude and uncertainty
A p-value alone does not describe the size or precision of a result. Identify the estimate and an appropriate uncertainty interval where relevant. Name the measure and its units, and distinguish adjusted from unadjusted results. If an interval is absent from the approved output, request it from the analyst rather than deriving an unsupported value from the prose.
Reconcile tables and narrative
Check the abstract, main results and tables side by side. Mark inconsistent sample sizes, reversed group labels, unexplained units and different rounding of the same result. Distinguish standard deviation from standard error in headings and notes. A correction should be traceable to the approved output. Do not change an inconvenient number merely to make two sections agree.
Return a query log for scientific approval
Finish with a list of questions, their locations, the supporting file and the person responsible for resolving them. Separate a typographical fix from a request for re-analysis. Confirm changes with the research team before closing the log. Use the reporting guideline for the study design in addition to this checklist, and check that the conclusion reflects the final approved results.
Worked examples
These invented teaching examples show an editing decision. They are not client records or research results.
- Before
- 30% responded; 100 participants were enrolled, but response data were available for 80.
- After or author query
- Author query: Was the result 30 of 80 participants with available data (37.5%)? Explain the 20 missing observations and confirm the analysis denominator.
All numbers are invented for this teaching example. The edit must not silently substitute a denominator.
- Before
- The groups were significantly different.
- After or author query
- Author query: Which outcome and comparison? Provide the estimate, units, uncertainty interval and test result from the approved analysis output.
The reporting query asks for evidence; it does not fabricate a significant result or an interval.