qqkrls — Quantile-on-Quantile Kernel Regularized Least Squares

Author / maintainer: Dr Merwan Roudane  ·  merwanroudane920@gmail.com  ·  Repo: https://github.com/merwanroudane/qqkrlsr

qqkrls implements Quantile-on-Quantile KRLS as in Adebayo et al. (2024). For each quantile theta of the independent variable, the response is fit by KRLS (Hainmueller and Hazlett 2014) on the corresponding sub-sample. The tau-quantile of the resulting pointwise marginal effects yields the QQKRLS coefficient beta(theta, tau). Standard errors come from a paired bootstrap.

All 3D surfaces, heatmaps and contour plots default to MATLAB Parula.

Installation

# from CRAN (once accepted)
install.packages("qqkrls")

# development version
# install.packages("remotes")
remotes::install_github("merwanroudane/qqkrlsr")

# from a local source tarball
install.packages("qqkrls_1.0.0.tar.gz", repos = NULL, type = "source")

Quick start

library(qqkrls)

set.seed(1)
n <- 200
x <- rnorm(n)
y <- 0.5 * sin(x) + rnorm(n, sd = 0.3)

fit <- qqkrls(y, x, n_boot = 100)
print(fit)

# 3D surface with MATLAB Parula
plot_qqkrls_3d(fit, value = "coefficient", colorscale = "Parula")

# Heatmap with significance stars
plot_qqkrls_heatmap(fit, value = "coefficient", show_stars = TRUE)

References

License

GPL-3