RPIV: Residual Prediction Tests for Well-Specification of Instrumental Variable Models

Two tests for the well-specification of the linear instrumental variable model. The first test is based on trying to predict the residuals of a two-stage least-squares regression using a random forest. The second test is robust to weak-identification and based on trying to predict the residuals for a particular candidate parameter and can also be used to construct confidence sets with an Anderson-Rubin-type inversion. Details can be found in Scheidegger, Londschien and Bühlmann (2025) "Machine-learning-powered specification testing in linear instrumental variable models" <doi:10.48550/arXiv.2506.12771>.

Version:	1.1.0
Imports:	ranger, stats
Suggests:	testthat (≥ 3.0.0)
Published:	2026-03-24
DOI:	10.32614/CRAN.package.RPIV
Author:	Cyrill Scheidegger [aut, cre, cph]
Maintainer:	Cyrill Scheidegger <cyrill.scheidegger at stat.math.ethz.ch>
License:	GPL (≥ 3)
URL:	https://github.com/cyrillsch/RPIV
NeedsCompilation:	no
Materials:	README, NEWS
CRAN checks:	RPIV results

Documentation:

Reference manual:

RPIV.html , RPIV.pdf

Downloads:

Package source:	RPIV_1.1.0.tar.gz
Windows binaries:	r-devel: RPIV_1.1.0.zip, r-release: RPIV_1.1.0.zip, r-oldrel: RPIV_1.1.0.zip
macOS binaries:	r-release (arm64): RPIV_1.1.0.tgz, r-oldrel (arm64): RPIV_1.0.0.tgz, r-release (x86_64): RPIV_1.1.0.tgz, r-oldrel (x86_64): RPIV_1.1.0.tgz
Old sources:	RPIV archive

Linking:

Please use the canonical form https://CRAN.R-project.org/package=RPIV to link to this page.