Estimation of heterogeneous treatment effects (HTE) is of prime importance in many disciplines, ranging from personalized medicine to economics among many others. Random forests have been shown to be a flexible and powerful approach to HTE estimation in both randomized trials and observational studies. In particular "causal forests", introduced by Athey, Tibshirani and Wager (2019), along with the R implementation in package grf were rapidly adopted. A related approach, called "model-based forests", that is geared towards randomized trials and simultaneously captures effects of both prognostic and predictive variables, was introduced by Seibold, Zeileis and Hothorn (2018) along with a modular implementation in the R package model4you. Here, we present a unifying view that goes beyond the theoretical motivations and investigates which computational elements make causal forests so successful and how these can be blended with the strengths of model-based forests. To do so, we show that both methods can be understood in terms of the same parameters and model assumptions for an additive model under L2 loss. This theoretical insight allows us to implement several flavors of "model-based causal forests" and dissect their different elements in silico. The original causal forests and model-based forests are compared with the new blended versions in a benchmark study exploring both randomized trials and observational settings. In the randomized setting, both approaches performed akin. If confounding was present in the data generating process, we found local centering of the treatment indicator with the corresponding propensities to be the main driver for good performance. Local centering of the outcome was less important, and might be replaced or enhanced by simultaneous split selection with respect to both prognostic and predictive effects.

Efficiently and flexibly estimating treatment effect heterogeneity is an important task in a wide To identify causal effects, the aforementioned approaches variety of settings ranging from medicine to marketing, operate under the exchangeability assumption, i.e., the assertion and there are a considerable number of that conditional on observed covariates, the treatment promising conditional average treatment effect assignment is as good as random. We propose a CATE estimators currently available. These, however, estimator, which using the framework of Tchetgen Tchetgen typically rely on the assumption that the measured et al. (2020), allows one to estimate causal effects in covariates are enough to justify conditional settings where conditional exchangeability fails, but one has exchangeability. We propose the P-learner, motivated measured a set of sufficient proxy variables. Our practical by the Rand DR-learner, a tailored twostage approach is motivated by the generic Neyman-orthogonal loss function for learning heterogeneous (Chernozhukov et al., 2018a) loss function from Nie & Wager treatment effects in settings where exchangeability (2021) and Kennedy (2020) that decouples nuisance given observed covariates is an implausible assumption, estimation and CATE estimation into two stages that can be and we wish to rely on proxy variables estimated (and tuned with cross-validation) by flexible lossminimizing for causal inference.