Conditional Average Treatment Effect Estimation

Conditional average treatment effect (CATE) estimation asks how a treatment changes outcomes for people or groups with different observed characteristics, supporting decisions about whom to treat and where resources are most effective. Common approaches include S-, T-, X- and R-learners, causal forests and trees, and doubly robust or orthogonal methods that combine outcome and treatment-assignment models; Bayesian and neural methods extend these tools to uncertainty estimation and complex representations. Researchers apply them to randomized trials and observational data, including marketing, health, policy, and survival settings, and evaluate both effect accuracy and the value of resulting treatment policies. A central challenge is making estimates and uncertainty reliable under confounding, weak treatment overlap, distribution shift, and noisy outcomes, while choosing benchmarks and ranking metrics that match the decision being made.