标题： 逆条件概率加权在聚类数据中的因果推断 显示英文标题

标题： Inverse Conditional Probability Weighting with Clustered Data in Causal Inference

摘要： 估计聚类数据中的平均治疗因果效应通常涉及处理未测量的聚类特定混杂变量。这些变量可能与测量的单元协变量和结果相关联。当忽略这些关联时，因果效应的估计可能会有偏倚。通过利用充分统计量，我们提出了一种逆条件概率加权（ICPW）方法，该方法对（i）未测量的聚类特定混杂变量与协变量之间的相关性和（ii）未测量的聚类特定混杂变量与结果之间的相关性都具有鲁棒性。提出了ICPW方法的假设和条件。我们建立了所提出估计量的渐近性质。为了说明起见，进行了模拟研究和案例研究。 显示英文摘要

摘要： Estimating the average treatment causal effect in clustered data often involves dealing with unmeasured cluster-specific confounding variables. Such variables may be correlated with the measured unit covariates and outcome. When the correlations are ignored, the causal effect estimation can be biased. By utilizing sufficient statistics, we propose an inverse conditional probability weighting (ICPW) method, which is robust to both (i) the correlation between the unmeasured cluster-specific confounding variable and the covariates and (ii) the correlation between the unmeasured cluster-specific confounding variable and the outcome. Assumptions and conditions for the ICPW method are presented. We establish the asymptotic properties of the proposed estimators. Simulation studies and a case study are presented for illustration.