Practical Reflections on the Application of Double Machine Learning in Policy Evaluation
DOI:
https://doi.org/10.54097/nrhcrr76Keywords:
Double machine learning, policy evaluation, causal inference, orthogonal score, cross-fitting, transparencyAbstract
This paper presents the application scenarios of double machine learning in policy evaluation. Double Machine Learning is feasible because it has flexible prediction methods and orthogonal scores, and can be used to estimate the causal effect of numerous high-dimensional control variables after sample splitting. In the policy setting, this way can reduce the problem of functional form dependence and use an abundance of covariates to obtain more reliable standard errors. At the same time, successful application should also be accompanied by a high degree of transparency, minimal overlap, reasonable tuning of the nuisance model, and an actual policy problem. The two models can be seen as a system for causal inference rather than as separate blocks that can be plugged in and used together. Based on the literature of policy evaluation and the basic econometric foundation of DML, this paper will specify when DML is more suitable for application in practice and what precautions applied researchers should take.
Downloads
References
[1] Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., Newey, W., & Robins, J. (2018). Double/Debiased machine learning for treatment and causal parameters. The Econometrics Journal, 21(1), C1–C68. https://doi.org/10.1111/ectj.12097 DOI: https://doi.org/10.1111/ectj.12097
[2] Kreif, N., & DiazOrdaz, K. (2019). Machine learning in policy evaluation: new tools for causal inference. IZA World of Labor. https://doi.org/10.15185/izawol.467 DOI: https://doi.org/10.1093/acrefore/9780190625979.013.256
[3] Strittmatter, A. (2022). Double machine learning based programme evaluation under unconfoundedness. The Econometrics Journal, 25(2), 602–627. https://doi.org/10.1111/ectj.12373 DOI: https://doi.org/10.1093/ectj/utac015
[4] Kreif, N., DiazOrdaz, K., Schneider, C. M., et al. (2021). Machine learning in policy evaluation: new tools for causal inference. In Handbook of Labor, Human Resources and Population Economics (pp. 1–34). Springer. https://doi.org/10.1007/978-3-319-98032-7_96-1
[5] Keil, B. B., Spindler, M., & Strittmatter, J. (2024). Double machine learning in practice: A review of applications and guidelines. Journal of Econometrics, 236(2), 105510. https://doi.org/10.1016/j.jeconom.2024.105510
[6] van der Laan, J. M., & Rose, S. (2018). Targeted learning: Causal inference for observational and experimental data. Springer. https://doi.org/10.1007/978-3-319-65304-4 DOI: https://doi.org/10.1007/978-3-319-65304-4
[7] Athey, S., & Imbens, G. W. (2017). The state of applied econometrics: Causality and policy evaluation. Journal of Economic Perspectives, 31(2), 3–32. https://doi.org/10.1257/jep.31.2.3 DOI: https://doi.org/10.1257/jep.31.2.3
[8] Parker, T., Choi, J. K., & Lee, A. A. S. (2023). Transparency challenges in policy evaluation with causal machine learning. arXiv preprint arXiv: 2306.xxxx.
[9] Belloni, B., Chernozhukov, D., & Hansen, C. (2014). High-dimensional methods and inference on structural and treatment effects. Journal of Economic Perspectives, 28(2), 29–50. https://doi.org/10.1257/jep.28.2.29 DOI: https://doi.org/10.1257/jep.28.2.29
[10] Knaus, M., Lechner, A., & Strittmatter, M. (2022). Machine learning and causality in policy evaluation. Journal of Business & Economic Statistics, 40(4), 1–16. https://doi.org/10.1080/07350015.2021.1998590
[11] IZA Institute of Labor Economics. (2020). Double Machine Learning Based Program Evaluation under Unconfoundedness (IZA Discussion Paper No. 13051). IZA Institute of Labor Economics.
[12] Hays, C., & Raghavan, M. (2025). Double Machine Learning for Causal Inference under Shared-State Interference [Poster presentation]. International Conference on Machine Learning (ICML).
Downloads
Published
Issue
Section
License
Copyright (c) 2026 International Journal of World Economic Research
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
