Jump to text
On this page
Copy and source
Published source confirmed at last check

Source changed 2026-11-16 21:18:08 UTC · snapshot created 2026-11-16 21:19:40 UTC · last check 2026-11-17 19:15:10 UTC

RMP-CORRECTION — Repair an analysis and return its consequence without discarding a suitable method

  • Situation: A completed analysis contains a calculation defect, and a later change in the observations may also challenge the method’s fit.
  • Question: Is the defect in the computation or the method? Under what conditions would a comparison of methods answer a further question?
  • First useful result or blocker: A verified correction and any resulting change in the analysis or claim its users rely on. Under different conditions, the researcher may be able to select a method or explain which question about its suitability remains open.
  • Start with: RMP.9 — Decide Whether and How to Improve Research Methods and Practice, using the actual difficulty and receiving conditions.
  • Stop or return: Keep the suitable method after repairing the computation. If conditions have changed, report a proposed comparison as a plan until results are available.
  1. Locate the calculation error. In the calculation application, the method for the independent observations 2, 4, 6 and 8 specifies sample variance with denominator n−1. The mean is 5 and the sum of squared deviations is 20. Dividing by n gives the reported 5. The specified denominator gives 20/3. With RMP.4 the researcher can trace the computation and find the analyses that used it.
  2. Correct and verify the affected result. Use RMP.9 to repair the denominator and retain the method. With RMP.5, recalculate the affected analysis. With RMP.8, return any changed uncertainty or conclusion to those who relied on it. The known-answer check settles this implementation defect. Earlier analyses that did not use the faulty calculation retain their support.
  3. Reassess the method for a different target. Now the programme wants an equally weighted mean across independent sites. The numbers of dependent observations per site are unequal. Pooling all observations gives the sites different weights. Use RMP.9 to compare methods for the site-level quantity. A qualified statistician, existing simulation code and two investigator-days make it feasible to compare a t interval from independent site means with an interval obtained by resampling whole sites. The team has selected this comparison. Which procedure is suitable remains open until the comparison or other qualified evidence answers it. Restrict the unsupported uncertainty claim and reconsider the comparison if its cost, value, access or staffing changes.