NickCH-K/MagnifiedIV: Magnified Instrumental Variables Estimator

This package performs a Magnified Instrumental Variables estimate, as in Huntington-Klein (2019) "Instruments with Heterogeneous Effects: Monotonicity, Bias, and Localness." Magnified Instrumental Variables is an IV estimator that is modified only slightly to allow the effect of the instruments to vary, so as to reduce noise in the small-sample IV bias term. These methods produce a super-local average treatment effect (SLATE) which weights high-instrument-effect observations more heavily than does regular IV. This can be done by either bringing in a group identifier to interact with the instrumental variable(s), or by estimating heterogeneity in the instrumental variables effect and using that to form a sample weight. The MagnifiedIV package includes support for the GroupSearch and causal forest methods for estimating effect heterogeneity, as described in the original paper. Causal forest estimates use the grf package.

install.packages("remotes")
remotes::install_github("NickCH-K/MagnifiedIV")