DR-Learners for Generalizing CATEs from a Randomized Controlled Trial
Building on existing DR-Learners, we develop the gDR-Learner for generalizing RCT CATEs to a target population, using cross-fitted machine-learning nuisance estimation.
ProofPaper ↗
Key points
- Randomized clinical trials are essential for evaluating interventions, yet average treatment effects can obscure treatment effect heterogeneity, making conditional average treatment effects (CATEs) a natural target.
- We derive excess-risk bounds in which nuisance errors enter only as products, showing that the estimator achieves oracle-rate convergence despite slower individual nuisance estimation.
- We construct weighted conformal prediction intervals for individual treatment effects, accounting for covariate shift between the trial and target population.
- We illustrate the methods by reanalyzing the Coronary Artery Surgery Study, examining treatment effect heterogeneity by prior myocardial infarction and baseline ejection fraction.
Sources (1)
- [1]DR-Learners for Generalizing CATEs from a Randomized Controlled TrialarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 05:03 PM
Building on existing DR-Learners, we develop the gDR-Learner for generalizing RCT CATEs to a target population, using cross-fitted machine-learning nuisance estimation.
Randomized clinical trials are essential for evaluating interventions, yet average treatment effects can obscure treatment effect heterogeneity, making conditional average treatment effects (CATEs) a natural target.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 18, 2026Google Open-Sources Harness2, a Recursive Agent Harnessing Framework
- Oct 18, 2026Think Visually Across Space and Time: Grounded Reasoning for Video Segmentation
- Oct 17, 2026EdgeKernelBench: Unified LLM-Driven Kernel Generation for Edge Backends
- Oct 16, 2026DLite: History-free Speculative Decoding for Agentic AI
- Oct 16, 2026Tree-Aware Posterior Distillation for Long-Horizon Agentic Reinforcement Learning
- Oct 16, 2026Self-Evolving Coding Agents via Harness-Policy Optimization