Answers are grounded in this paper and cite their sources.
This problem shows how the key ignorability-of-treatment assumption used in estimating treatment effects can be violated when certain factors are included among the covariates. The case considered is when there are J + 1 treatment levels, treatment is randomized with respect to potential outcomes, but the distribution of included covariates differs across treatment levels.
No takes yet. Share an insight, caveat, or question.
Jeffrey M. Wooldridge (2005) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: