AI & ML
Causal Effect Estimation Under Uncertain Statistical Surrogacy
arXiv stat.TH0 views
arXiv:2611.26289v1 Announce Type: cross Abstract: Surrogate endpoints can improve the efficiency of causal effect estimation, but such gains may depend on structural assumptions. One such assumption is statistical surrogacy, which requires the primary outcome to be conditionally independent of treatment given the surrogate and baseline covariates. We study causal effect estimation with multiple candidate surrogates of uncertain validity under statistical surrogacy, considering settings where the primary outcome is either fully observed or subject to missingness. We characterize the asymptotic efficiency differences between surrogate-assisted and benchmark estimators, examine the bias induced by violations of statistical surrogacy, and identify conditions under which incorporating surrogate information leads to efficiency gains or losses. To address uncertainty in surrogate validity, we propose Surrogate-assisted Adaptive Causal effect EStimation (SACES), a procedure that anchors estimation to a robust benchmark while adaptively selecting among a finite set of surrogate-assisted estimators using an estimated mean squared error criterion. SACES aims to leverage efficiency gains when statistical surrogacy holds while preserving consistency when the assumption is violated. Simulations demonstrate that SACES achieves a favorable bias-variance trade-off across varying degrees of surrogacy violation. We illustrate the proposed methods using data from the ACTG 175 clinical trial to demonstrate how surrogate information can be adaptively incorporated into causal effect estimation.
Source:https://arxiv.org/abs/2611.26289PreviousA note on the learning coefficient of exceptional singularitiesNextAdjusting SPRT for an Asymptotically Efficient Procedure with Finite Number of Applications of Less Effective Treatment
Related posts
Reading project health off a download page — faster than star count
0dev.to
Best time trackers for developers, for code and for AI
0dev.to

Serving Qwen3.5-0.8B with a 0.016% weight edit: measure the loops, pin the sampling on the server
0dev.to
Design Tokens for AI: Extract a Website's Real Colours, Fonts, Shadows and Icons So an LLM Builds On Brand
0dev.to
Trying to get our Arms around AI efficiency
0Stack Overflow Blog
Why an Embedded Game Fails on Some Phones, and How to Debug It
0dev.to