Machine learning for causal inference: on the use of cross-fit estimators
This addresses the challenge of reliable causal effect estimation in epidemiology when using machine learning, though it is incremental as it builds on existing cross-fit methods.
The study tackled the problem of using machine learning for causal inference, which can complicate statistical inference, by evaluating cross-fit estimators through simulations. The results showed that doubly-robust cross-fit estimators with machine learning substantially outperformed other methods in bias, variance, and confidence interval coverage when parametric models were misspecified.
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.