【摘要】 借助观测数据评估经济政策的因果效应时，处理变量通常具有内生性，即处理变量与回归误差项具有相关性，从而导致回归系数不相合。本文提出基于可加逻辑回归的二元内生处理效应估计方法，具体而言，采用可加逻辑回归模型，有效刻画内生变量与工具变量的非线性关系，同时利用组Lasso方法选择重要工具变量。利用所得倾向得分近似二元内生处理变量的最优工具变量，可以有效降低处理效应估计量的方差。此外，为弥补单选法在模型选择上的不足，采用双选法来识别可能因单选法而被遗漏的重要控制变量，以减少遗漏变量偏误。本文证明所提出的二元内生处理效应估计量具有相合性和渐近正态性，蒙特卡洛模拟显示该方法有良好的有限样本表现。在实证分析方面，应用该方法研究身体健康是否会影响个人收入等问题。更多还原

【Abstract】 When estimating the causal effects of economic policies on the outcome variable using observational data, treatment variables often exhibit endogeneity, meaning that the treatment variable and the regression error term are correlated, resulting in the inconsistency of regression coefficients. This study proposes a method for estimating binary endogenous treatment effects based on an additive Logistic regression model. Specifically, this study employs the additive Logistic regression model, which effectively characterizes the nonlinear relationship between the binary endogenous variable and instrumental variables in the first stage. Meanwhile, this study utilizes the group Lasso method to select important instrumental variables. This study approximates the optimal instrumental variables for binary endogenous treatment variables using the estimated propensity score, which effectively reduces the variance of the treatment effect estimator. Furthermore, to address the issue of omitted variable bias arising from imperfect model selection under the single variable selection method, the proposed method employs a double-selection approach to identify important control variables. Theoretically, this study demonstrates that the proposed estimator for the binary endogenous treatment effect is consistent and asymptotically normal. Monte Carlo simulations illustrate the method’s excellent finite sample performance. In empirical analysis, this study applies the proposed method to examine whether physical health affects individual income.更多还原