Food Insecurity in America

Overview

SNAP (formerly Food Stamps) is a critical lifeline for households struggling to cover grocery costs. Economic downturns hit certain communities harder — and those same communities tend to recover more slowly. This project assessed demographics vulnerable to food insecurity using SNAP Quality Control data from the USDA, conducting a 10-year gap analysis comparing 2007 and 2017 datasets to examine lasting impacts on vulnerable communities.

Using an ArcGIS spatial analysis MOOC, two contrasting counties were identified as study areas: San Juan County, New Mexico (an emerging hot spot) and Cherry County, Nebraska (an emerging cold spot). The target variable was CAT_ELIG — whether a household was eligible (1) or not eligible (0) for SNAP benefits.

GIS hot and cold spot analysis — ↑ Click to enlarge — Hot and cold spots identified via ArcGIS spatial analysis

Methodology

Each dataset started with 400,000+ records and 800 features. Through nullity assessment, correlation analysis, and domain knowledge, the data was reduced to 32 features and ~4,000 final records.

Downloaded SNAP Quality Control data from the USDA for 2007 and 2017
Removed columns with all null values, then dropped columns with more than 50% nulls
Imputed remaining nulls with the mean using sklearn SimpleImputer
Used correlation analysis on 2007 data to select top 5 features per section — 32 final features
Extracted and compared county-level data for San Juan County, NM and Cherry County, NE
Built a Vote Ensemble model combining Random Forest, Gradient Boost, and Bagging Classifier
Achieved a 95% cross-validation score on the final model
Visualized feature importances and state-level predictions using Matplotlib and Seaborn

↑ Click to enlarge — Vote Ensemble model performance (95% cross-validation score)

Feature importance chart — ↑ Click to enlarge — Feature importance from Random Forest model

Key Findings

Several patterns emerged from the analysis:

SNAP requesters typically had small households — one to two children maximum
Approximately 50% were single mothers heading households
The top 4 predictors were all related to shelter and housing costs: FSSLTDED, SHELDED, FSTOTDED, and FSSTDDE2
New Mexico saw ~100 fewer working poor on SNAP in 2017; Nebraska saw ~50 more
In 2007, SNAP recipients received less assistance from other welfare programs than in 2017
San Juan County, NM showed greater economic vulnerability than Cherry County, NE across both years

New Mexico vs Nebraska comparison — ↑ Click to enlarge — State-level comparison: New Mexico vs Nebraska

Model coefficients and shelter variables — ↑ Click to enlarge — Model coefficients: top predictors related to shelter and housing

Predicted SNAP Households by State (2018)

The original project identified a next step: building an interactive geographic dashboard to pinpoint high-need areas. The map below realizes that goal — showing all three layers: predicted SNAP households, shelter deductions, and self-employment deductions, visualized at the state level for 2018.

↑ Hover over a state to see values. Toggle layers using the control in the top left. Layers were created in QGIS, exported to GeoJSON, and displayed here using Leaflet.