Density-Free RIF Decompositions for Unconditional Expectiles

Implements a density-free recentered influence function (RIF) regression framework for unconditional expectiles, and embeds it in a two-sample Oaxaca-Blinder decomposition indexed continuously by the expectile level. Unlike quantile-based RIF decompositions, which require estimating an inverse density term at each quantile, the expectile RIF depends only on primitive moments of the outcome distribution and requires no density estimation, no bandwidth selection, and no kernel smoothing. The package provides expectile estimation by iteratively reweighted least squares, closed-form RIF construction, two-sample composition/structure decomposition across a grid of expectile levels, bootstrap-based inference, and plotting methods. The underlying methodology is described in Ndoye (2025), "Semi-Nonparametric Expectile RIF Regression for Distributional Decomposition," presented at the 2025 World Congress of the Econometric Society, Seoul, Korea, < https://www.econometricsociety.org/regional-activities/conference-papers/view/282/943>.

rifexpectile

Density-free RIF (recentered influence function) regressions for unconditional expectiles, embedded in a two-sample Oaxaca-Blinder decomposition indexed continuously by the expectile level. Unlike quantile-based RIF decompositions, which require estimating an inverse density term at each quantile, the expectile RIF depends only on primitive moments of the outcome and requires no density estimation, no bandwidth choice, and no kernel smoothing.

Installation

install.packages("rifexpectile")

Development version from GitHub:

# install.packages("remotes")
remotes::install_github("aandoye/rifexpectile", build_vignettes = TRUE)

Example

library(rifexpectile)
data(wage_gap_sim)

fit <- rif_decompose(
  logy ~ x1 + x2,
  data = wage_gap_sim,
  group = "group",
  ref_group = "2000",
  alpha = seq(0.1, 0.9, by = 0.2),
  boot_reps = 200,
  seed = 1
)
fit
plot(fit, which = "structure")

See vignette("rifexpectile-intro") for a full walkthrough, including an example with the publicly available wage1 data set from the wooldridge package.

Methodology

The closed-form expectile influence function, the two-sample asymptotic theory, and an empirical application are described in a companion manuscript (Ndoye, 2026, unpublished). This package focuses on the software; see ?rif_decompose for the exact algebra of the decomposition implemented here.