clusterIV: Clustered Jackknife Instrumental Variables Estimation
Tools for instrumental variables estimation and inference under
clustered errors with many instruments. The current release provides the
cluster-jackknife IV estimator (CJIVE) of Frandsen, Leslie and McIntyre
(2025) <doi:10.1162/rest.a.263> for a single endogenous regressor in a
just-identified design, with cluster-robust inference: each observation's
first-stage value is fitted leaving out its entire cluster, which removes
the many-instrument bias that survives clustering. The leave-cluster-out
fits use an exact Woodbury block update – one factorisation of the
instrument Gram matrix plus a small solve per cluster – so the estimator
scales to large samples. A companion 'iv_compare()' reports ordinary least
squares, two-stage least squares, the observation-level jackknife and CJIVE
on a common cluster-robust standard error.
Documentation:
Downloads:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=clusterIV
to link to this page.