Weak-Instrument Inference and Model Diagnostics in Double Machine Learning for Linear Instrumental Variable Models
En palabras de los autores
We study weak-instrument-robust inference in partially linear instrumental-variables models with learned nuisances. Building on an i.i.d. uniform-size baseline, we establish uniform asymptotic validity of clustered cross-fitted Anderson-Rubin inference and cross-fitted score inference under the stated sampling and nuisance-rate conditions. Extensions establish Anderson-Rubin validity for survey-weighted target laws and fold-nested generated instruments. A supporting population benchmark shows that prediction-optimal instrument residualization maximizes concentration when the treatment nuisance is correctly specified. A five-design diagnostic portfolio, a known-truth Monte Carlo, and an exploratory selector examine how nuisance specification and estimation affect realized strength and inference.
Apareció: martes, 10 de noviembre. arXiv. Preprint, todavía sin revisión por pares.
Comentario de los autores: 25 pages, 4 figures, 2 tables. Main text only; the supplementary material is available from the author on request