shellgame

The Shell Game: Swapping the VAR without changing the name or the formula.

Overview

The shellgame package reveals how data quality silently degrades during geographic transformations while variable labels remain unchanged. Like a shell game where the ball gets swapped without anyone noticing, geographic transformations swap observed data for imputed data while keeping the same column name.

The key insight: This error is agnostic to:

• Variable: Population, median income, vehicle ownership - doesn't matter
• Tool: R, Python, Stata, ArcGIS - doesn't matter
• Geography: Hennepin County or anywhere else - doesn't matter

What matters: The transformation itself causes the error.

Installation

# Install from GitHub
# install.packages("devtools")
devtools::install_github("phinnmarkson/shellgame")

Quick Start

library(shellgame)

# Run audit on Hennepin County (example data included)
result <- audit_transformation(
  baseline_data = hennepin_zcta_baseline,
  zip_zcta_map = hennepin_zip_zcta_map,
  hud_crosswalk = hennepin_hud_crosswalk,
  county_fips = "27053",
  variable_name = "pop"
)

# View results
summary(result)
#> Baseline: 1,391,557 (74 ZCTAs, observed from ACS)
#> After transformation: 1,216,874 (recovered via HUD TOT_RATIO)
#> Loss: 174,683 (-12.6%)
#>
#> The shell game: Same column name, different underlying quantity.

The Shell Game Explained

What's happening:

1. Start: Real measured population (observed data)
2. Hop 1 (ZCTA→ZIP): Swap in imputed values via association
3. Hop 2 (ZIP→County): Swap in more assumptions via TOT_RATIO
4. End: Completely imputed values, still labeled "population"

The cups never announce the swap, so everyone thinks they're still tracking the original ball.

Example: Hennepin County

# Visualize the transformation
plot_transformation_loss(result)

Key Features

• Reproducible audit framework - Quantify error at each transformation hop
• Variable agnostic - Works with any ACS variable
• Tool agnostic - Demonstrates the transformation is the problem
• Publication-ready visualizations - Maps and plots included
• Pre-loaded example - Hennepin County data ready to use

Learn More

• vignette("hennepin-example") - Complete worked example
• vignette("understanding-shell-game") - Conceptual explanation
• vignette("data-preparation") - How to audit your own geography

Citation

If you use this package in your research, please cite:

Markson, P. (2026). shellgame: The Shell Game - Audit Geographic Data 
  Transformations. R package version 0.1.0.

Inspiration

https://doi.org/10.5281/zenodo.18634426 Package structure inspired by the excellent areal package by Christopher Prener.

License

MIT © Phinn Markson