Some Attention for Attenuation Bias

on YouTube ↗31m

" summary

The talk covers attenuation bias (measurement error/regression dilution) that shrinks model coefficients toward zero when features are noisy, and explains orthogonal regression as a traditional correction while stressing that some knowledge of feature noise is required to correct it. It then shows how the presenter’s company handles attenuation bias in geo experiments for causal inference, including code for their proprietary adjustment, and relates the topic to imputation for missing variables and cautions about random regression imputation (per Andrew Gelman) due to attenuation effects.

AI-written summary · about

" chapters (6)

1. 0:00Attenuation bias intro
2. 3:36Attenuation bias in regression
3. 8:59Attenuation bias in ML
4. 13:14Solutions for attenuation bias
5. 20:45Noise in binary effects
6. 27:06Key takeaways

" related talks

• 27mUnmeasured Spatial ConfoundingGeorgia Papadogeorgou · Learning from Interventions 2022
• 43mRobustness in High-Dimensional Inference TasksJelena Bradic · Robust and High-Dimensional Statistics (Simons Institute workshop, 2018)
• 44mInvariance, Causality and Novel RobustnessPeter Bühlmann · Robust and High-Dimensional Statistics (Simons Institute workshop, 2018)
• 11mCorrelation Adjusted Debiasing (CAD): Debiasing the Lasso with Inaccurate Covariate ModelMichael Celentano · Meet the Fellows Welcome Event
• 54mDebiasing Random Forests For Treatment Effect EstimationJasjeet Sekhon · Statistics in the Big Data Era 2022
• 26mPanel DiscussionVanessa Didelez, Luke Miratrix, Georgia Papadogeorgou · Learning from Interventions 2022
• 11mCausal Inference With Corrupted Data: Measurement Error, Missing Values, Discretization, And...Rahul Singh · Meet the Fellows Welcome Event
• 45mOrthogonal Statistical LearningVasilis Syrgkanis · Algorithmic Aspects of Causal Inference 2022