Semiparametric Efficient Test for Distributional Treatment Effects Invisible to Means
arXiv·low signal
Develops a semiparametrically efficient test for detecting treatment effects that are invisible to mean comparisons — changes in distributional tails, modes, or dispersion. A treatment may preserve average outcomes while changing rare-event risk. Relevant for A/B testing practitioners and causal ML researchers who need to detect effects beyond simple average treatment effects.