npiv_regression: Nonparametric Instrumental Variables Regression

Overview

The npiv_regression function performs the non-parametric regression procedure as described by Chen et al. (2023)](https://academic.oup.com/restud/advance-article-abstract/doi/10.1093/restud/rdae025/7624036?redirectedFrom=fulltext) in a Difference-in-Differences (DiD) context, as detailed in: Callaway et al. (2024)

This implementation was part of a Master's dissertation project. I have since realised that a similar implementation is available in R at Jeffrey Racine's GitHub repository.

Installation

Before using the npiv_regression function, make sure you have installed and loaded the required package. You can install it from CRAN or GitHub if it's part of a custom package:

# Download the package
devtools::install_github("oddish3/contdid")

# Load the package
library(contdid)

Usage

The npiv_regression function is straightforward to use. You need to provide the function with the names of the treatment and outcome columns, along with the dataset that contains these columns.

Function Signature

npiv_regression(treatment_col, outcome_col, data)

Parameters

• treatment_col: A string representing the name of the dosage variable in the dataset. This is the variable for which you want to estimate the causal effect, make sure all values are between 0 and 1.
• outcome_col: A string representing the name of the outcome variable in the dataset. This is the dependent variable whose relationship with the treatment you want to study.
• data: A data frame containing the variables specified in treatment_col and outcome_col.

Example

Suppose you have a dataset final_data that includes a column killed_w_transformed as your treatment variable and change_gb_tot as your outcome variable. You can perform the nonparametric instrumental variables regression as follows:

# Perform NPIV regression
res <- npiv_regression(treatment_col = "killed_w_transformed",
                       outcome_col = "change_gb_tot", 
                       data = final_data)

# View the results
print(res)

Example Analysis (to be implemented)

Here's a brief example of how you might interpret the results:

# Example of interpreting the results
summary(res)

The output of summary(res) will provide you with the details of the regression, allowing you to make inferences about the causal impact of killed_w_transformed on change_gb_tot.

Replicating replication 1 & 2.

Script 1 in the repository replicates the analysis using data from \url{https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/NXFB5R}. Script 2 in the repository replicate the analysis using data from \url{https://www.openicpsr.org/openicpsr/project/113746/version/V1/view}.