| name | lemma-inference |
| description | Quantify whether a difference or association is distinguishable from sampling noise using effect estimates, uncertainty intervals, tests, or power analysis. |
Statistical inference
Deliver
Report the estimand, effect estimate, uncertainty interval, population and
denominator, method and assumptions, practical interpretation, and any power
or multiplicity limitation.
Check
- Define the estimand, comparison, smallest meaningful effect, and sampling or
assignment process.
- Preserve pairing, clustering, repeated measures, and temporal dependence.
- Inspect sample sizes, missingness, imbalance, outliers, and distribution
shape.
- Use robust, clustered, paired, permutation, or bootstrap uncertainty when
the design requires it.
- Separate planned from exploratory tests and account for multiplicity.
- Stress-test the weakest assumption.
Do not equate significance with importance or causality, treat dependent
observations as independent, or report p-values without effect sizes.
For method selection and power detail, read
references/deep-guide.md.
