Why should a data-subset refutation use repeated random subsets rather than only the first rows?
StatisticsLevel: IntermediateCausal stress testing
Why should a data-subset refutation use repeated random subsets rather than only the first rows?
Hints
- Ordered rows can change the target population.
- Smaller samples are noisier.
Answer
The first rows may represent a different period or population. Random subsets probe sampling stability; their increased uncertainty must be included in the comparison.
Source and licence: DoWhy — data subset refutation · MIT · answer: Quantoros editorial