# Machine learning for causal inference: on the use of cross-fit estimators

arXiv: https://arxiv.org/abs/2004.10337 · HTML: https://mlsanity.com/paper/2004.10337/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 21 Apr 2020.

## Built on by (most cited first, 11 of 11)

- [Efficient adjustment for complex covariates: Gaining efficiency with DOPE](https://mlsanity.com/paper/2402.12980/) (2024)
- [Meta-Learners for Estimation of Causal Effects: Finite Sample Cross-Fit Performance](https://mlsanity.com/paper/2201.12692/) (2022)
- [Graph Machine Learning based Doubly Robust Estimator for Network Causal Effects](https://mlsanity.com/paper/2403.11332/) (2024)
- [Application of targeted maximum likelihood estimation in public health and epidemiological studies: a systematic review](https://mlsanity.com/paper/2303.07329/) (2023)
- [Performance of Cross-Validated Targeted Maximum Likelihood Estimation](https://mlsanity.com/paper/2409.11265/) (2024)
- [Generalized coarsened confounding for causal effects: a large-sample framework](https://mlsanity.com/paper/2501.03129/) (2025)
- [Three Routes to One Answer: Reconciling AIPW, TMLE, and Double Machine Learning for Applied Researchers](https://mlsanity.com/paper/2609.26142/) (2026)
- [Causal Bias Quantification for Continuous Treatments](https://mlsanity.com/paper/2106.09762/) (2021)
- [Policy learning for many outcomes of interest: Combining optimal policy trees with multi-objective Bayesian optimisation](https://mlsanity.com/paper/2212.06312/) (2022)
- [When Prediction Error Is Not Enough: Evaluating Nuisance-Function Prediction for Causal Estimation](https://mlsanity.com/paper/2609.00071/) (2026)
- [When Sample-Splitting Choices Matter in TMLE and AIPW: Cross-Fitting, Targeting, and Repetition](https://mlsanity.com/paper/2610.16192/) (2026)

## Abstract (arXiv)

Modern causal inference methods allow machine learning to be used to weaken parametric modeling assumptions. However, the use of machine learning may result in complications for inference. Doubly-robust cross-fit estimators have been proposed to yield better statistical properties. We conducted a simulation study to assess the performance of several different estimators for the average causal effect (ACE). The data generating mechanisms for the simulated treatment and outcome included log-transforms, polynomial terms, and discontinuities. We compared singly-robust estimators (g-computation, inverse probability weighting) and doubly-robust estimators (augmented inverse probability weighting, targeted maximum likelihood estimation). Nuisance functions were estimated with parametric models and ensemble machine learning, separately. We further assessed doubly-robust cross-fit estimators. With correctly specified parametric models, all of the estimators were unbiased and confidence intervals achieved nominal coverage. When used with machine learning, the doubly-robust cross-fit estimators substantially outperformed all of the other estimators in terms of bias, variance, and confidence interval coverage. Due to the difficulty of properly specifying parametric models in high dimensional data, doubly-robust estimators with ensemble learning and cross-fitting may be the preferred approach for estimation of the ACE in most epidemiologic studies. However, these approaches may require larger sample sizes to avoid finite-sample issues.