Causal Bandits Podcast Do Heterogeneous Treatment Effects Exist? | Stephen Senn X Richard Hahn S2E9 | CausalBanditsPodcast
15 snips
Jan 30, 2026 Stephen Senn, medical statistician focused on drug development and trials, and Richard Hahn, ASU statistics professor working on causal inference and regression trees, debate whether heterogeneous treatment effects are real and detectable. They discuss ethics of averaging, richer covariate measurement for discovery, machine learning on RCT data, trial design tradeoffs, and when subgroup findings become actionable.
AI Snips
Chapters
Books
Transcript
Episode notes
Heterogeneity Exists But Beware Oversearch
- Heterogeneous treatment effects almost certainly exist and matter in real contexts like medicine and design.
- Statisticians must balance enthusiasm for finding them with the risk of chasing spurious patterns.
Tamoxifen Search Sparked Personal Concern
- Richard Hahn recounts searching for subgroup tamoxifen results for his wife and finding missing subgroup analyses.
- The data existed but wasn't reported, motivating measurement of richer covariates in trials.
Doctor Variation Can Outweigh Patient Heterogeneity
- Searching for ever-smaller reference classes often increases variance and worsens estimates.
- Physician practice variation can dominate patient heterogeneity and deserves attention.
