Statistical Limits of Causal Inference
summary
Covers three concrete vignettes: limits for personalized causal effects (when tau is smooth but mu0/mu1 are not), rates without any smoothness where one-step corrections / influence-function estimators give 1/n variance but square nuisance errors, and discrete-covariate ATE where polynomial-approximation / collision-based higher-order U-statistics beat plugins only by a log d factor. States a structure-agnostic minimax: first-order (double-robust) estimator is optimal unless extra smoothness lets U-statistics reach minimax n^{-8s/(4s+d)} rates; cites Jin & Syrgkanis for ATE lower bounds.
AI-written summary · about
chapters (10)
- 0:00Causal inference motivation
- 1:04Functional estimation basics
- 9:33Causal inference basics
- 25:31One-step corrections
- 35:48Double ML and nuisance
- 40:21Higher-order u-statistics
- 49:55Conditional treatment effect
- 56:41Error bounds and limits
- 1:04:59Discrete covariates models
- 1:10:17Q&A
related talks
33mSimple Yet Efficient Estimators For Network Causal Inference...
31mLearning And Testing Causal Models: A Property Testing Viewpoint
57mA New Perspective on High-Dimensional Causal Inference
43mRobustness in High-Dimensional Inference Tasks
45mThe Statistical Fairness-Accuracy Frontier
41mTransfer learning via local convergence rates of the nonparametric least squares estimator
28mData Driven Convergence in Statistical Estimators
01h 01mThe Mathematics of Causal Inference, with Reflections on Machine Learning and the Logic of Science