Quantile treatment effects revisited: Uncovering the distributional consequences of a welfare experiment

Heterogeneous effects of welfare reforms on earnings, transfers, and income have been established
theoretically and empirically. Evaluation studies often focus on quantile treatment effects (QTE), which
rely on the marginal distributions of potential treatment and control outcomes. Parameters that depend on
the joint distribution of potential outcomes, such as quantiles of the distribution of treatment effects
(QDTE), receive less attention. We propose a strategy to identify these parameters. We leverage the
property that, under random assignment, rank correlation coefficients between actual treatment and
predicted control state outcomes are identical, irrespective of whether predictions are based on treatment
or control units. To identify QDTE, we assume that all permutations of observation units satisfying this
property are equally likely. Rearranging quantiles yields a generalized version of quantile treatment effects
(GQTE). We employ a reweighting approach for identification under strong ignorability. We test the
predictor strength and demonstrate that highly predictive covariates yield unbiased, consistent, and
asymptotically normal estimators. Our analysis of Connecticut’s Jobs First program reveals initial income
increases for a larger fraction of participants than previously recognized. Long-term gains were at least
twice as large as those derived from conventional QTE and concentrated at the lower end of the
distribution.