The paper proposes an estimator to make inference of heterogeneous treatment effects sorted by impact groups (GATES) for non-randomised experiments. Observational studies are standard in policy evaluation from labour markets, educational surveys and other empirical studies. To control for a potential selection-bias we implement a doubly-robust estimator in the first stage. Keeping the flexibility, we can use any machine learning method to learn the conditional mean functions as well as the propensity score. We also use machine learning methods to learn a function for the conditional average treatment effect. The group average treatment effect, is then estimated via a parametric linear model to provide p-values and confidence intervals. To control for confounding in the linear model we use Neyman-orthogonal moments to partial out the effect that covariates have on both, the treatment assignment and the outcome. The result is a best linear predictor for effect heterogeneity based on impact groups. We introduce inclusion-probability weighting as a form of cross-splitting and averaging for each observation to avoid biases through sample splitting. The advantage of the proposed method is a robust linear estimation of heterogeneous group treatment effects in observational studies.
View on arXiv