README

svytest provides a unified suite of diagnostic tests for assessing the informativeness of survey weights in regression models. It implements both classical parametric tests (Hausman–Pfeffermann, DuMouchel–Duncan, Pfeffermann–Sverchkov, Wu–Fuller, Estimating Equations) and novel non‑parametric permutation tests.

Why use svytest?

Survey weights are essential for unbiased descriptive statistics, but their role in regression modeling is debated. Using weights unnecessarily can inflate variance, while ignoring informative weights can bias estimates.

svytest helps you decide whether weights matter in your regression by offering:

• Difference‑in‑Coefficients tests (Hausman–Pfeffermann)
• Weight‑Association tests (DD, PS1/PS2, WF)
• Estimating Equations test (Pfeffermann–Sverchkov)
• Permutation tests (distribution‑free, robust alternatives)
• Simulation utilities to benchmark test performance

Installation

You can install the released version of svytest from CRAN with: install.packages("svytest") or install the development version from GitHub with: remotes::install_github("cnlubianski/svytest").

Usage

Load the package and your survey data.

library(svytest)
library(survey)

data("svytestCE", package = "svytest")
des <- svydesign(ids = ~1, weights = ~FINLWT21, data = svytestCE)
fit <- svyglm(TOTEXPCQ ~ ROOMSQ + BATHRMQ + BEDROOMQ + FAM_SIZE + AGE,
              design = des)

Example 1: Difference-in-Coefficients Test

res_dc <- diff_in_coef_test(fit, var_equal = TRUE)
summary(res_dc)
#> 
#> Difference-in-Coefficients Test
#> Call:
#> diff_in_coef_test(model = fit, var_equal = TRUE)
#> 
#> Method:
#>  Hausman-Pfeffermann Difference-in-Coefficients Test
#> 
#> Test Statistic:
#>  Chi-sq = 15.3278  on 6 df , p-value = 0.0179

This compares weighted vs. unweighted regression coefficients. A small p‑value suggests weights are informative.

Example 2: Weight-Association Test (DuMouchel–Duncan)

res_dd <- wa_test(fit, type = "DD")
summary(res_dd)
#> 
#> Weight-Association Test
#> Call:
#> wa_test(model = fit, type = "DD")
#> 
#> Method:
#>  DuMouchel-Duncan Weight-Association Test
#> 
#> Test Statistic:
#>  F = 2.5557  on 6 and 21914 df , p-value = 0.0178

This regresses residuals on weights. A significant slope indicates informativeness.

Example 3: Estimating Equations Test

res_perm <- perm_test(fit, stat = "pred_mean", B = 500, engine = "R")
summary(res_perm)

Getting Started

See the vignette for a comprehensive review of all tests, the statistical underpinnings, and references: vignette("svytest").