Stratified sampling
Strata may represent risk categories or class labels. Sampling design can improve representation of important groups; estimates may require appropriate weighting if sample fractions differ. In model evaluation, stratified folds can preserve class proportions, but this does not replace time-aware splitting when chronology matters.
Worked example
A dataset contains 900 routine and 100 high-risk cases. A 10% proportionate stratified sample selects 90 routine and 10 high-risk cases.
Common mistake
Assuming a disproportionate stratified sample has the same population proportions without weighting, or using class stratification to justify leakage across time.
Where it appears in your exams
| Qualification | Exam | Relationship | Period |
|---|---|---|---|
| ACCA | MA | Direct | September 2026–June 2027 |
| FIA | FMA | Direct | September 2026–June 2027 |