Worksheets and research resources · Page 46
Search and filter all worksheets and research resourcesChoose One Eligible Outcome Result Without Cherry-Picking
Prespecify rules for instruments, time points, and analyses when a study reports several eligible results.
Compare Methods on the Same Simulated Datasets
Preserve replicate pairing when evaluating differences in performance and computational behavior.
Label Percentage Denominators in Tables and Charts
Make row, column, eligible-population, and multiple-response percentages distinguishable.
Check Exchangeability Before a Permutation Test
Match the allowed rearrangements to the design and null hypothesis rather than shuffling every row.
Report pilot estimates with useful uncertainty
Explain what a small pilot can estimate and why an imprecise effect should not become a definitive trial assumption.
Audit Treatment Overlap Before Trusting a Weighted Comparison
Identify where the observed data do not support the treatment contrasts required by the estimand.
Check Population Margins Before Poststratifying a Survey
Verify categories, reference populations and weighted totals before accepting calibrated weights.
Read Precision–Recall Results in Their Population Context
Connect positive predictive value and sensitivity to the event frequency and threshold being evaluated.
Flag Predictions Outside the Observed Predictor Support
Distinguish interpolation from unsupported combinations before presenting model-based predictions.
Plan What a Prediction Model Does When an Input Is Missing
Validate the actual missing-input workflow instead of assuming complete predictors at deployment.
Treat Recalibration as a Model Update That Needs Evaluation
Preserve the original validation result while testing a justified update to predicted probabilities.
Scale Predictors for Numerical Stability Without Losing Units
Record transformations and back-transform coefficients so numerical convenience does not change the reported effect.