The paper by Lee, Lessler, and Stuart explores the use of machine learning techniques, particularly classification and regression trees (CART), for estimating propensity scores in observational studies. They conducted simulations to evaluate the performance of various CART-based propensity score models, including logistic regression, pruned CART, bagged CART, random forests, and boosted CART, under different scenarios of non-linear and non-additive relationships between covariates and the exposure. The study found that while all methods generally performed well under conditions of non-linearity or non-additivity alone, ensemble methods (especially boosted CART) showed superior performance under conditions of both moderate non-additivity and moderate non-linearity. These methods provided better bias reduction and more consistent 95% confidence interval coverage compared to logistic regression. The results suggest that ensemble methods may be particularly useful for propensity score weighting, offering advantages over traditional logistic regression in terms of covariate balance and effect estimation.The paper by Lee, Lessler, and Stuart explores the use of machine learning techniques, particularly classification and regression trees (CART), for estimating propensity scores in observational studies. They conducted simulations to evaluate the performance of various CART-based propensity score models, including logistic regression, pruned CART, bagged CART, random forests, and boosted CART, under different scenarios of non-linear and non-additive relationships between covariates and the exposure. The study found that while all methods generally performed well under conditions of non-linearity or non-additivity alone, ensemble methods (especially boosted CART) showed superior performance under conditions of both moderate non-additivity and moderate non-linearity. These methods provided better bias reduction and more consistent 95% confidence interval coverage compared to logistic regression. The results suggest that ensemble methods may be particularly useful for propensity score weighting, offering advantages over traditional logistic regression in terms of covariate balance and effect estimation.