arXiv paper: debiased ML single-imputation framework for survey item nonresponse
RexDouglass · x · 2026-11-09
A new arXiv paper (stat.ME) by Mehdi Dagdoug and David Haziza proposes a debiased machine learning single-imputation framework for survey item nonresponse. Key points:
- Plainly plugging ML predictions into imputed estimators fails to guarantee valid inference; regression estimation error breaks root-n consistency
- Uses a cross-fitted augmented inverse-probability-weighted (AIPW) estimator combining an imputation model, sampling design, and MAR mechanism
- Establishes double robustness; consistency of both nuisance estimators plus a product-rate condition yields oracle equivalence and asymptotic normality
- Constructs adjusted imputed values preserving observed responses whose weighted mean exactly reproduces the AIPW estimate, with consistent variance estimators
- Simulations: cross-fitting substantially reduces undercoverage but may raise MSE when nuisance predictions are inaccurate
More from Research
- New benchmarks DecoyCount and ScienceVision probe whether AI sees or reasons — liangchen5518 · 2026-11-09
- Video-guided machine translation paper lifts low-resource Indic languages by +1.75 BLEU — kalyan_kpl · 2026-11-09
- Peer-reviewed paper applies truth maintenance systems to LLM memory belief revision — letandrewcook · 2026-11-09
- A cathedral that sings: 1D acoustic sim shows voices fighting their own resonant body — Sauers_ · 2026-11-09
- Survey by 30+ authors maps modern NLP tokenization, from subword wins to its flaws — rdesh26 · 2026-11-09
- No 8 GPUs needed: six hands-on steps to actually understand transformers — cneuralnetwork · 2026-11-09