20. Estimate lift when experiments are imperfect
Covers regression adjustment, matching, weighting, difference-in-differences, and sensitivity checks when randomized experiments are not available. You estimate incrementality while being clear about what the data can and cannot prove.