We propose a causal framework for decomposing a group disparity in an outcome in terms of an intermediate treatment variable. Our framework captures the contributions of group differences in baseline potential outcome, treatment prevalence, average treatment effect, and selection into treatment. This framework is counterfactually formulated and readily informs policy interventions. The decomposition component for differential selection into treatment is particularly novel, revealing a new mechanism for explaining and ameliorating disparities. This framework reformulates the classic Kitagawa-Blinder-Oaxaca decomposition in causal terms, supplements causal mediation analysis by explaining group disparities instead of group effects, and resolves conceptual difficulties of recent random equalization decompositions. We also provide a conditional decomposition that allows researchers to incorporate covariates in defining the estimands and corresponding interventions. We develop nonparametric estimators based on efficient influence functions of the decompositions. We show that, under mild conditions, these estimators are $\sqrt{n}$-consistent, asymptotically normal, semiparametrically efficient, and doubly robust. We apply our framework to study the causal role of education in intergenerational income persistence. We find that both differential prevalence of and differential selection into college graduation significantly contribute to the disparity in income attainment between income origin groups.

翻译：我们提出一个因果框架，用于通过中间治疗变量分解结果变量中的群组差异。该框架捕捉了基线潜在结果、治疗流行率、平均治疗效果和治疗选择中的群组差异贡献。该框架基于反事实公式化，能直接为政策干预提供信息。其中，治疗选择差异的分解成分尤为新颖，揭示了解释和改善差异的新机制。该框架将经典的Kitagawa-Blinder-Oaxaca分解用因果术语重新表述，通过解释群组差异而非群组效应补充了因果中介分析，并解决了近期随机均衡分解的概念性困难。我们还提供了条件分解，允许研究者纳入协变量定义估计量和相应干预措施。我们基于分解的有效影响函数开发了非参数估计量。研究表明，在温和条件下，这些估计量是$\sqrt{n}$-一致的、渐近正态的、半参数有效的且具有双重稳健性。我们将该框架应用于研究教育在代际收入持续性中的因果作用。研究发现，大学教育的流行率差异和选择差异均显著贡献于不同收入出身组别之间的收入差距。