Synthetic control (SC) methods are commonly used to estimate the treatment effect on a single treated unit in panel data settings. An SC is a weighted average of control units built to match the treated unit, with weights typically estimated by regressing (summaries of) pre-treatment outcomes and measured covariates of the treated unit to those of the control units. However, it has been established that in the absence of a good fit, such regression estimator will generally perform poorly. In this paper, we introduce a proximal causal inference framework to formalize identification and inference for both the SC and ultimately the treatment effect on the treated, based on the observation that control units not contributing to the construction of an SC can be repurposed as proxies of latent confounders. We view the difference in the post-treatment outcomes between the treated unit and the SC as a time series, which opens the door to various time series methods for treatment effect estimation. The proposed framework can accommodate nonlinear models, which allows for binary and count outcomes that are understudied in the SC literature. We illustrate with simulation studies and an application to evaluation of the 1990 German Reunification.
翻译:合成控制(SC)方法通常用来估计在小组数据设置中对单一处理单位的处理效果。SC是为与被处理单位相对应而建造的控制单位的加权平均值,其重量一般通过递减(摘要)预处理结果和经测量的被处理单位与控制单位的共差来估计。然而,已经确定,在缺乏适当性的情况下,这种回归估计器一般效果不佳。在本文件中,我们引入了一种初步的因果关系框架,正式确定在册种姓的识别和推断,并最终确定对被处理单位的处理效果,其依据的观察是,未对建造SC作出贡献的控制单位可被重新用作潜在凝结物的替代物。我们认为,处理后处理结果与SC之间的差别是一个时间序列,它打开了各种治疗效果估计时间序列方法的大门。拟议的框架可以容纳非线性模型,允许在SC文献中未得到充分研究的二进和计结果。我们用模拟研究和对1990年德国再工业化的评估的应用作了说明。