excessive costs: groceries

Staging

Built from the latest training run on 2026-09-30 17:45 UTC, not from a released model set. The released documentation is at https://matter.indigo.engineering/docs/.

Overview

This model predicts whether a voter is experiencing excessive grocery costs — a felt-burden indicator combining direct reports of difficulty paying for groceries, an elevated grocery share of household expenses, and food-assistance program enrollment.

Trained on survey data, it uses income proxies, consumer indicators, and area-level cost-of-living features to identify voters feeling acute grocery cost pressure under current inflation conditions.

Organizations can apply these scores to target SNAP outreach, food affordability messaging, and anti-hunger campaigns.

approach: deep learning
training rows: 18,295
last updated: 2026-08-27 21:17:43

Validation

metric	value
area under ROC curve	0.67
area under precision-recall curve	0.76
accuracy	0.68
precision	0.70
recall	0.86
f1 score	0.77
mean log loss	0.61
gini coefficient	0.34
brier skill score	0.09

Lift Chart

Each bar shows how much more common the outcome is within a score decile than in the population overall. Bars taller than the dashed line (lift of 1) mean the model’s top-ranked groups concentrate the outcome; a steady decline from decile 1 to 10 indicates the model is ranking effectively.

Calibration Plot

Bars of similar height within each decile indicate well-calibrated predictions. Systematic differences reveal where the model over- or under-predicts.

Cumulative Gains

The further the model curve bows above the diagonal, the better the model concentrates the outcome in its top-ranked predictions.

Decile Analysis Table

decile	count	mean actual	mean predicted	mae	lift
10	459	0.81	0.81	0.003	1.27
9	458	0.76	0.77	0.02	1.18
8	458	0.79	0.75	0.04	1.23
7	458	0.74	0.73	0.007	1.15
6	458	0.68	0.71	0.02	1.07
5	458	0.64	0.68	0.04	0.99
4	458	0.62	0.62	0.002	0.97
3	458	0.55	0.53	0.01	0.85
2	458	0.47	0.46	0.01	0.73
1	459	0.36	0.35	0.007	0.56

Architecture

This model is a deep neural network trained with TensorFlow.

Network structure

layer	units	activation
↓ dense	96	relu
↓ dropout (0.50 rate)
↓ dense	84	relu
↓ dropout (0.50 rate)
↓ dense	72	relu
↓ dropout (0.50 rate)
↓ dense	60	relu
↓ dropout (0.50 rate)
↓ dense	48	relu
↓ dropout (0.50 rate)
↓ dense	36	relu

Hyperparameter configuration

loss: binary_focal_crossentropy
optimizer: rmsprop
batch size: 32
activation: hard_sigmoid
label smoothing: 0
learning rate: 0.001

Features

Importance

Sourcing

Agency for Healthcare Research and Quality

received snap / food stamps (%)

Attom Property Data API

land market value

Revelio Labs

avg tenure (days)
max salary

State voter files (via TargetSmart)

years registered to vote

TargetSmart

catholic score
children present score
climate change concern score
cord cutter score
double negative downballot gop score
high school only score
ideology score
ideology score (enhanced)
presidential general turnout score

U.S. Census American Community Survey

block group % college degree
block group % female
block group % high school diploma
block group % income <$25k
block group % income $25–50k
block group % income $50–100k
block group households
block group median rent

Verisk (via TargetSmart)

homeowner
medical occupation
senior present in household

Implementation

Prediction distribution

statistic	value
mean	0.59
std dev	0.10
min	0.14
25th	0.53
median	0.60
75th	0.67
max	0.91

Segments

segment	lowest score	highest score	share of records
Most likely	0.65	0.91	32.9%
Somewhat likely	0.56	0.64	33.7%
Least likely	0.14	0.55	33.4%

External validity

Our model indicates that 59% of people with scored records report excessive grocery costs. For comparison, a 2026 survey by LendingTree found that 49% of Americans reported difficulty affording food, while a 2024 report by the USDA Economic Research Service indicated that 14% of US households had “low or very low food security.”