Calibrated Inference: Statistical Inference That Accounts For Both Sampling Uncertainty And...
summary
Proposes a distributional uncertainty model where data come from a two‑stage random reweighting that produces a single global delta scaling factor inflating all variances; proves asymptotic normality under that regime and shows delta is estimable from between‑estimator variation when several asymptotically linear, scientifically equivalent estimators exist. Builds calibrated CIs (so‑called Yujin intervals) and an R toolchain including a CALM R package for GLMs; simulations recover ~95% coverage while standard CIs under‑cover.
AI-written summary · about
chapters (17)
- 0:00Intro
- 2:00Nature
- 3:52Random Decision Making
- 7:25Distributional Uncertainty
- 8:38Linear Regression
- 11:40Asymptotic Regime
- 12:55Math Incoming
- 17:35When Does This Assumption Hold
- 18:10Symmetry Assumption
- 19:25Sampling Bias Random confounding
- 21:10Additional assumptions
- 23:35Inference
- 28:15Numerical Example
- 29:50HighLevel Remarks
- 31:50Getting Started
- 34:05Philosophical Comments
- 40:35Discussion