Volume 2,Issue 1
处理效应的异质性的识别
—— 基于因果机器学习的仿真和估计
既有因果机器学习的相关文献所提出的用于识别异质群体的平均处理效应（GATE）的方法未能考虑在协变量变化的情况下解释不同群体间处理效应的异质性。为了解决该问题，本文提出基于无偏机器学习（DML）的平衡组平均处理效应（BGATE）来衡量具有预先确定的协变量特定分布的组平均处理效应（GATE），通过计算两个BGATE 之间的差值来比较两个GATE 的值，从而更好地识别因果效应的异质性，最终将由协变量不同分布所导致的差异与由解释变量所导致的差异区分开来。该估计量在标准条件下具有N − 一致性和渐近正态的性质。通过对比DML、自动无偏机器学习（Auto-DML）和重新加权（Reweighting Approach）三种估计方法的仿真结果可知：如果已知DML 没有性能问题，如当倾向得分很极端时，建议采用DML 估计量，其模拟表现最好；如果已知DML 有性能问题，建议采用Auto-DML 估计量或重新加权方法。
[1]Chernozhukov V,Chetverikov D ,Demirer M, et al.Double/debiased machine learning for treatment and structural parameters[J].The Econometrics Journal,2018,21(01):C1-C68.
[2]钱浩祺,龚嫣然,吴力波.更精确的因果效应识别:基于机器学习的视角[J].计量经济学报,2021,1(04):867-891.
[3]Lechner M, Mareckova J.Comprehensive Causal Machine Learning, arXiv preprint arXiv:2405.2024:10198.
[4]Frölich M.Finite-Sample Properties of Propensity-Score Matching and Weighting Estimators[J].The Review of Economics and Statistics,2004,86(01):77-90.
[5]Chernozhukov V, Newey W, Quintas-Martınez V M. et al. Riesznet and forestriesz: Automatic debiased machine learning with neural nets and random forests[J]. In International Conference on Machine Learning,2022,:3901-3914. PMLR.
[6]郭峰,陶旭辉.机器学习与社会科学中的因果关系：一个文献综述[J].经济学(季刊),2023,23(01):1-17.
[7]Abrevaya J, Hsu Y C, Lieli R P.Estimating Conditional Average Treatment Effects[J].Journal of Business and Economic Statistics, 2015, 33(4):00-00.DOI:10.1080/07350015.2014.975555.
[8]Lechner M.Modified Causal Forests for Estimating Heterogeneous Causal Effects[J].Michael Lechner, 2018.DOI:10.13140/RG.2.2.13540.22405.
[9]Semenova V, Chernozhukov V.Debiased Machine Learning of Conditional Average Treatment Effects and Other Causal Functions[J].The Econometrics Journal,2021,24(02):264-289.
[10]Zimmert M, Lechner M.Nonparametric estimation of causal heterogeneity under high-dimensional confounding[J].Papers, 2019.DOI:10.48550/arXiv.1908.08779.
[11]Fan Q, Hsu Y C, Lieli R P, et al..Estimation of Conditional Average Treatment Effects With High-Dimensional Data[J].Journal of Business And Economic Statistics, 2022, 40(01):313-327.
[12]Farbmacher H, Huber M, Langen H,et al.Causal mediation analysis with double machine learning[J].The Econometrics Journal, 2022,25, 277-300.
[13]Rubin D B.Estimating causal effects of treatments in randomized and nonrandomized studies.[J]Journal of Educational Psychology, 1974,66, 688.
[14]Imbens W G,Wooldridge M J.Recent Developments in the Econometrics of Program Evaluation[J].Journal of Economic Literature,2009,47(01):5-86.
[15]Imbens W G.Nonparametric Estimation of Average Treatment Effects under Exogeneity: A Review[J].The Review of Economics and Statistics,2004,86(01):4-29.
[16]Belloni A, Chernozhukov V. Least squares after model selection in high-dimensional sparse models[J].Bernoulli,2013,19(02):521-547.
[17]Bach P, Schacht O, Chernozhukov V, et al.Hyperparameter tuning for causal inference with double machine learning: A simulation study[J]. In Causal learning and reasoning,2024:1065-1117. PMLR.