Question

Explain why stratification is used in holdout estimation. What purpose does it serve?

Answer 1

Please read on to answer the questions. If you have doubts please let me know.

Stratification ensures that each "strata" of data is representative of all strata of the data.

Generally this is done in a supervised way for segmentation and aims to ensure each strata or class is (approximately) equally represented. Hence, when we have holdouts , we ensure through startification that the 2 groups of "control" and "test ( otherwise known has holdout) " are essentially the same. Results of holdout estimation would be robust when the holdout and controls groups are essentially similar - stratitification does this for us.

If the "holdout" and "control" groups are not the same out estimates of holdouts may not be precise / accuate.