The Robustness of Controlling the False Discovery Rate
Talkat Robust and High-Dimensional Statistics (Simons Institute workshop, 2018) · 2018-10-29..11-02·
summary
Shows that the Benjamini–Hochberg rule stays controlled under far weaker conditions than classical PRDS: any compliant procedure applied to null p‑values satisfying a mild independence/ordering property gives FDR ≤ Ck·q (explicit Ck that falls toward 1 as k grows), with constants provably optimal. Simulations on correlated Gaussian designs find BH rarely exceeds nominal q. Identifies the null p‑value distribution as the main driver of failures, analyzes noisy (differentially private) p‑value selection, and proposes a relaxed criterion called FDR consistency for dimension‑uniform control.
AI-written summary · about
chapters (11)
- 0:00Introduction to FDR and BH
- 1:13The BH procedure mechanism
- 3:21Theoretical limitations
- 5:24Bridging theory and practice
- 7:22Robustness through compliance
- 11:59Theoretical proofs and bounds
- 16:26Simulations and performance
- 18:26Relaxing dependency assumptions
- 25:08The concept of FDR consistency
- 29:54Conclusion and summary
- 30:58Q&A session
related talks
01h 07mStatistical Inference and Privacy, Part I
01h 03mStatistical Inference and Privacy, Part II
43mInformation-Theoretic Bounds and Phase Transitions in Community Detection and High-Dimensional Clust
01h 03mSimons Institute Live Stream
31mBias Meets Bayes: A Bayesian Perspective on Improving Model Fairness
42mDifferential Privacy
11mCausal Inference With Corrupted Data: Measurement Error, Missing Values, Discretization, And...
01h 10mThe Social Survey Statistician's Perspective