enrolled in snap

WarningStaging

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 participates in the Supplemental Nutrition Assistance Program (SNAP).

Trained on survey data, it draws on working-class indicators, neighborhood income levels, and household characteristics to estimate enrollment likelihood.

These scores help organizations identify voters who depend on food assistance and target outreach around nutrition benefits and anti-hunger policy.

  • approach: deep learning
  • training rows: 5,026
  • last updated: 2026-05-02 16:17:23

Validation

metric value
area under ROC curve 0.75
area under precision-recall curve 0.17
accuracy 0.55
precision 0.12
recall 0.86
f1 score 0.21
mean log loss 0.93
gini coefficient 0.50
brier skill score 0.06

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
10 123 0.24 0.21 0.03 3.47
9 122 0.11 0.13 0.03 1.52
8 122 0.08 0.10 0.02 1.17
7 122 0.11 0.08 0.03 1.52
6 123 0.06 0.06 0.004 0.92
5 122 0.02 0.05 0.03 0.23
4 122 0.02 0.03 0.009 0.35
3 122 0.02 0.02 0.002 0.35
2 122 0.02 0.01 0.01 0.35
1 123 0.008 0.008 1e-05 0.12

Architecture

This model is a deep neural network trained with TensorFlow.

Network structure

layer units activation
↓ dense 128 swish
↓ batch normalization
↓ dropout (0.30 rate)
↓ dense 76 swish
↓ batch normalization

Hyperparameter configuration

  • loss: binary_focal_crossentropy
  • optimizer: rmsprop
  • batch size: 64
  • activation: sigmoid
  • label smoothing: 0
  • learning rate: 0.002

Features

Importance

Sourcing

Agency for Healthcare Research and Quality

  • received snap / food stamps (%)

Attom Property Data API

  • lot size (acres)
  • property tax billed

Revelio Labs

  • avg salary differential

State voter files (via TargetSmart)

  • age
  • registered republican

TargetSmart

  • abortion issue motivation score
  • catholic score
  • children present score
  • college graduate score
  • evangelical score
  • homeowner score
  • ideology score
  • local election voter score
  • midterm general turnout score
  • off-year general turnout score
  • partisan score
  • presidential general turnout score
  • pro-choice score
  • race: white
  • veteran score
  • working class score

U.S. Census American Community Survey

  • tract median rent

Verisk (via TargetSmart)

  • home value
  • number of adults in household

Implementation

Prediction distribution

statistic value
mean 0.07
std dev 0.08
min 0.01
25th 0.02
median 0.03
75th 0.08
max 0.77

Segments

segment lowest score highest score share of records
Most likely 0.22 0.77 6.3%
Somewhat likely 0.15 0.21 6.6%
Least likely 0.01 0.14 87.1%

External validity

Our model indicates that 7% of people with scored records receive SNAP benefits. USDA ERS SNAP Key Statistics reports 41.7M monthly SNAP recipients in FY2024. Children are roughly a third of recipients, so adult-level participation is closer to ~11% of U.S. adults.