Weighted Descriptive Statistics.

wstats

Base R provides weighted.mean() but nothing else. wstats fills the gap with weighted versions of the other common descriptive statistics.

Installation

# install.packages("devtools")
devtools::install_github("jgaeb/wstats")

Usage

library(wstats)

x <- c(1, 2, 3, 4, 5)
w <- c(0.5, 1.0, 2.0, 1.0, 0.5)   # unnormalised importance weights

weighted_var(x, w)
weighted_sd(x, w)
weighted_quantile(x, w, probs = c(0.25, 0.5, 0.75))
weighted_median(x, w)
weighted_mad(x, w)
weighted_skewness(x, w)
weighted_kurtosis(x, w)

y <- c(2, 3, 1, 5, 4)
weighted_cov(x, y, w)
weighted_cor(x, y, w)

Convention

All functions currently use a population formula — weights are treated as probability masses of a discrete distribution, not survey sampling weights. Concretely, variance is Σ(ŵᵢ (xᵢ − μ)²) where ŵᵢ = wᵢ / Σwⱼ, with no Bessel correction. This is appropriate for importance sampling and the Bayesian bootstrap; support for survey/frequency weights may be added in a future release.

Computationally intensive routines are implemented in C++ via cpp11.