Intersectionality in Plumbing Insecurity: Educational Attainment and Ethnic Composition Are Stronger Predictors When Considered Independently

Author: Atharva Naik

Overview

This project analyzes the extent to which plumbing insecurity across U.S. counties is associated with educational attainment and ethnic composition, and whether an interaction between these variables improves predictive modeling. Using ACS data (via NHGIS) and generalized additive models (GAMs), the study finds that education and ethnicity contribute independently, while their interaction does not meaningfully enhance predictive performance.

Data

Data were obtained from the American Community Survey through NHGIS:

• B01003 — Total Population
• B03003 — Hispanic or Latino Origin
• B15003 — Educational Attainment (25+)
• B25049 — Tenure by Plumbing Facilities

All analyses are performed at the county level.

Methods

A series of GAM models were constructed:

• Baseline model with nonlinear smooth terms
• Models adding education × ethnicity interaction
• Ablation comparisons using R², RMSE, AIC, and GCV
• Partial Dependence Plots (PDPs) and Effective Degrees of Freedom (EDoF) used to assess feature influence and model complexity

Key Findings

• Interaction terms show minimal improvement across all metrics (|ΔR²| < 0.02; |ΔRMSE| < 0.001).
• PDPs indicate that individual variables exert stronger effects than their joint interaction.
• EDoF values for interaction terms (0–2) are far lower than those for individual smooth terms (4–7), suggesting weaker or inconsistent interaction patterns.
• Conclusion: Education and ethnicity are more informative when modeled independently.

Code, Data Access, & Presentation

• Code: https://github.com/eDoggo3779/Water_Insecurity
• NHGIS Data: https://www.nhgis.org/
• Oral Presentation: https://www.youtube.com/watch?v=AgllHcr-gtA

Acknowledgments

Thanks to The Coding School, Kayvan Jalali, and Dr. Tara Flaugher for technical and research guidance.