RobustIV and controlfunctionIV: Causal Inference for Linear and Nonlinear Models with Invalid Instrumental Variables

TL;DR

This paper introduces R packages RobustIV and controlfunctionIV for causal inference using instrumental variables, capable of handling invalid instruments in both linear and nonlinear models with robust inference methods.

Contribution

The paper presents novel software tools that implement advanced methods for valid causal inference with potentially invalid instrumental variables in linear and nonlinear models.

Findings

RobustIV effectively selects valid instruments in high-dimensional settings.

The packages provide robust inference methods that tolerate errors in instrument selection.

Applications demonstrate practical utility on economic and health data sets.

Abstract

We present R software packages RobustIV and controlfunctionIV for causal inference with possibly invalid instrumental variables. RobustIV focuses on the linear outcome model. It implements the two-stage hard thresholding method to select valid instrumental variables from a set of candidate instrumental variables and make inferences for the causal effect in both low- and high-dimensional settings. Furthermore, RobustIV implements the high-dimensional endogeneity test and the searching and sampling method, a uniformly valid inference method robust to errors in instrumental variable selection. controlfunctionIV considers the nonlinear outcome model and makes inferences about the causal effect based on the control function method. Our packages are demonstrated using two publicly available economic data sets together with applications to the Framingham Heart Study.