The document discusses a tree-based approach for addressing self-selection in causal research involving high-dimensional data, aimed at improving impact studies such as randomized experiments and quasi-experiments. It outlines the challenges of traditional propensity score methods in big data and proposes a new method that leverages classification and regression trees to identify confounders and analyze treatment effects more effectively. The authors present applications of this method in various domains, demonstrating its advantages over conventional techniques in detecting heterogeneous treatment effects and handling unbalanced covariates.