处理效应作为量化分析政策效应的工具，在经济学的各个方面有着广泛的应用。例如，在评价研发加计扣除政策对高技术企业创新的影响时，表明上看与其他实证类似，数据往往是一些企业享受加计扣除政策，而一些企业没有。但不同的是，一般是研发创新较强的企业更有可能去选择申请高技术企业认证，进而享受税收优惠，这种有选择的样本一般来说具有内生性，即内生选择问题。现有文献一般都需要加一些条件（如均值独立）或采用工具变量法。在较弱的、与实际更吻合的条件下给出处理效应的一致估计量并不多见，本项目尝试在非参数估计的框架下对该问题进行研究，得到稳健的估计量。更进一步，基于上述提出的估计量，本项目针对处理效应模型中常见的识别条件“非混淆假设”进行了检验，并提出了一个检验统计量。最后，本项目将这些方法应用于两项政策评价：一是，评价“加计扣除政策”对高技术企业创新的影响；二是，评价 “二孩政策”对学区房溢价的影响。

Treatment effect is widely used in microeconomics, especially in the program evaluation literature. For example, in the analysis of the impact of R&D expenditure deduction policy on high-tech industry，it seems that there is no difference from the other empirical studies, the data contains some firms enjoy the deduction policy while others are not. However, the difference is that firms that have high level R&D input are more likely to apply for high-tech enterprise certification, and thus enjoy tax benefits. That means, there exists self-selection in this case, if we ignore this effect, the commonly used estimation procedure will incur the inconsistency of the estimation. In the literature, to consistently estimate the treatment effect, strong restrictions are usually imposed. Among them, there are two commonly used approaches: one is ignobility of the treatment conditional on a set of covariates (conditional mean independence), the other is that instrumental variable is available. Thus one interesting research topic is to explore the estimator under the weaker restriction estimation, this research project intends to explore the robust estimator for the treatment effects. Furthermore, based on the above-mentioned estimator, we proposed a tests statistic for “unconfoundness” which is always used in the treatment effect model. Finally, we apply these methods to two policy evaluations: First, the effect of R&D expenditure deduction policy on high-tech industry; Second, the “two-child” policy on the premium of school district.