Get Started
Home
Topics
Search
Library
Research questionHow should treatment-effect heterogeneity and sample size guide sample splitting for individual effects in causal forests?Sample splitting can reduce overfitting, but it also leaves less data to detect and estimate treatment-effect heterogeneity. The balance between these risks may change with the strength of heterogeneity and the available dataset size.
Economics
Machine Learning
Research Paper
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Honesty in Causal Forests: When It Helps and When It HurtsThe source examines honest estimation in causal forests using more than 7,000 benchmark datasets. It reports that honesty can reduce individual treatment-effect accuracy, particularly with substantial heterogeneity and sufficiently large datasets, and frames honesty as a form of regularization rather than a universally appropriate default.research paper · Sep 3, 2026
Related questions
How can we reliably select a heterogeneous treatment effect estimator without observing treatment effects?How can extreme quantile treatment effects be estimated under heavy tails while respecting location invariance?How can we recover a target site's full treated-outcome distribution from controls alone under cross-site heterogeneity?How should limited samples be allocated across classes to estimate class-weighted risk for rare events in dependent sequences?