Principal stratification provides a robust causal inference framework for the adjustment of post-treatment variables when comparing the effects of a treatment in health and social sciences. In this paper, we introduce a novel Bayesian nonparametric model for principal stratification, leveraging the dependent Dirichlet process to flexibly model the distribution of potential outcomes. By incorporating confounders and potential outcomes for the post-treatment variable in the Bayesian mixture model for the final outcome, our approach improves the accuracy of missing data imputation and allows for the characterization of treatment effects across strata defined based on the values of the post-treatment variable. We assess the performance of our method through a Monte Carlo simulation study where we compare the proposed method with state-of-the-art Bayesian method in principal stratification. Finally, we leverage the proposed method to evaluate the principal causal effects of exposure to air pollution on social mobility in the US on strata defined by educational attainment.

翻译：主分层为健康与社会科学中比较处理效应时调整后处理变量提供了稳健的因果推断框架。本文提出一种新颖的主分层贝叶斯非参数模型，利用相依狄利克雷过程灵活建模潜在结果的分布。通过将后处理变量的混杂因素与潜在结果纳入最终结果的贝叶斯混合模型，我们的方法提升了缺失数据插补的准确性，并能够刻画基于后处理变量取值所定义各层级的处理效应特征。我们通过蒙特卡洛模拟研究评估本方法的性能，将所提方法与主分层领域最先进的贝叶斯方法进行比较。最后，我们运用所提方法评估美国空气污染暴露对社会流动性的主因果效应，该分析基于教育程度定义的分层进行。