Propagate Uncertainty for ACS Tabular Estimates

Utilities for propagating uncertainty in American Community Survey tabular workflows that use published estimates and margins of error, following U.S. Census Bureau derived-estimate guidance and complementing 'tidycensus' margin-of-error workflows. Includes covariance-aware derived estimates, simulation helpers, geographic aggregation, confidence-interval conversion, and reliability diagnostics.

acsmoe

acsmoe propagates uncertainty in American Community Survey tabular workflows. It is scoped to estimate + MOE workflows, not ACS data fetching or microdata variance estimation.

It provides:

covariance-aware sums, differences, ratios, proportions, products, and linear combinations;
simulation helpers for derived quantities;
grouped geographic-style aggregation of paired estimate/MOE columns;
MOE/SE/CI conversion utilities;
coefficient-of-variation reliability diagnostics.

With no covariance supplied, results match the standard zero-covariance ACS approximation formulas used by tidycensus, so it drops into existing estimate/MOE pipelines without changing the baseline numbers.

Installation

# install.packages("acsmoe")  # once on CRAN

# Development version
# install.packages("pak")
pak::pak("dshkol/acsmoe")

References and attribution

The zero-covariance formulas match the U.S. Census Bureau's ACS guidance for derived estimates, especially Chapter 8 of Understanding and Using American Community Survey Data: What All Data Users Need to Know.

The standard R workflow and baseline MOE helpers are provided by tidycensus; see Walker and Herman's package and Walker's MOE vignette: https://walker-data.com/tidycensus/articles/margins-of-error.html.

The broader motivation for ACS uncertainty handling and regionalization comes from Spielman, Folch, Nagle, Arribas-Bel, and Koschinsky's ACS uncertainty papers. The old geoss/censumander repository is cited as historical reference material only; regionalization is not part of this package.