RMP.9:5.2 - Select a worthwhile comparison without inventing its result

A programme now estimates an equally weighted mean across independent sites, with repeated observations within each site. Site sample sizes differ and observations within a site are dependent. The old pooled-observation analysis does not answer the intended site-level question with defensible uncertainty under the case premises.

The team first fixes the target. Two attainable candidates use the same mean of site means but construct uncertainty differently: a t interval from independent site means and an interval obtained by resampling whole sites. The names of these procedures do not settle their adequacy when there are few sites, skewed outcomes and plausible heterogeneity.

A qualified statistician, existing simulation code and a bounded two-day allocation make a discriminating comparison attainable. The programme will reuse the chosen analysis across several studies, and a wrong uncertainty claim would materially affect that work. Include design, code checks, computation, failed runs, Monte Carlo uncertainty, interpretation and transition in the burden; the two-day availability is a case premise, not a general cost estimate.

Select the simulation around relevant generating situations, the shared estimand and the performance required by the intended use. Preserve adverse and inconclusive results. Both candidates may prove inadequate for the proposed reach, leading to a narrower claim or stopped use.

The completed result here is a justified decision to obtain evidence. Method qualification and the choice of a winning procedure remain unresolved until actual results exist. If the access, capability or worth premises change, reconsider this acquisition decision.