IsoLATEing: Identifying Counterfactual-Specific Treatment Effects with Cross-Stratum Comparisons
IsoLATEing: Identifying Counterfactual-Specific Treatment Effects with Cross-Stratum Comparisons
复制标题
IsoLATEing：通过跨层比较识别反事实的特定治疗效果
DOI：
--
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
Peter Hull
中科院分区：
文献类型：
--
作者：
Peter Hull
工具变量（IV）的因果效应估计可能难以解释时，反事实的治疗混合多种选择。我探索识别多个反事实特定的局部平均治疗效果，从一个单一的准实验，使用分层控制的仪器的相互作用。我得出的一般形式，这样的IV被估量，并建立识别下的意思是独立的编译器处理效果的分层。在较弱的条件独立性假设下，识别实现了一种新的非参数加权方法。我使用这个框架来估计GED认证的回报，样本中包括那些否则将获得传统高中文凭的人以及那些否则将辍学的人。理论结果也可能提供一种策略，以调整内源性损耗随机试验，我说明了这一点，通过重新分析的俄勒冈州健康保险实验。
Instrumental variables (IV) estimates of causal effects can be difficult to interpret when the counterfactual to treatment mixes multiple alternatives. I explore identification of multiple counterfactual-specific local average treatment effects from a single quasi-experiment using interactions of an instrument with stratifying controls. I derive the general form of such IV estimands and establish identification under mean-independence of complier treatment effects with respect to the stratification. Under weaker conditional independence assumptions, identification is achieved with a novel non-parametric weighting approach. I use this framework to estimate the returns to GED certification in a sample that includes individuals who would otherwise obtain a traditional high school diploma as well as those who would otherwise drop out. The theoretical results may also offer a strategy to adjust for endogenous attrition in randomized trials; I illustrate this through a re-analysis of the Oregon Health Insurance Experiment.