insurance: government sponsored

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 has government-sponsored health insurance — primarily Medicare and Medicaid.

Trained on survey data, it uses income-level indicators, age, and neighborhood demographics to identify voters with public coverage.

Organizations can use these scores to focus outreach around government health program protections, funding, and eligibility.

approach: deep learning
training rows: 5,449
last updated: 2026-07-30 20:52:27

Validation

metric	value
area under ROC curve	0.84
area under precision-recall curve	0.77
accuracy	0.81
precision	0.79
recall	0.69
f1 score	0.73
mean log loss	0.47
gini coefficient	0.69
brier skill score	0.38

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	139	0.87	0.90	0.03	2.25
9	138	0.83	0.82	0.009	2.15
8	139	0.73	0.69	0.04	1.88
7	138	0.47	0.48	0.004	1.22
6	139	0.30	0.32	0.03	0.76
5	138	0.21	0.23	0.02	0.54
4	138	0.15	0.16	0.01	0.39
3	139	0.15	0.12	0.03	0.39
2	138	0.07	0.10	0.02	0.19
1	139	0.09	0.07	0.02	0.22

Architecture

This model is a deep neural network trained with TensorFlow.

Network structure

layer	units	activation
↓ dense	128	relu
↓ dropout (0.20 rate)
↓ dense	91	swish
↓ dropout (0.13 rate)
↓ dense	54

Hyperparameter configuration

loss: binary_focal_crossentropy
optimizer: nadam
batch size: 16
activation: sigmoid
label smoothing: 0.01
learning rate: 0.001

Features

Importance

Sourcing

Agency for Healthcare Research and Quality

received snap / food stamps (%)

Attom Property Data API

lot size (acres)

Revelio Labs

current position salary
max salary
number of jobs held

State voter files (via TargetSmart)

TargetSmart

double negative downballot gop score
double negative gop score
early vote / absentee vote score
ideology score (enhanced)
income rank
marriage score
midterm general turnout score
non-religious score
paid leave support score
presidential general turnout score
pro-choice score
recession sensitivity score

U.S. Census American Community Survey

block group % black
block group % income <$25k
tract median income

Verisk (via TargetSmart)

home value
medical occupation
net worth
spanish speaker in household

Implementation

Prediction distribution

statistic	value
mean	0.40
std dev	0.29
min	0.03
25th	0.14
median	0.30
75th	0.69
max	0.99

Segments

segment	lowest score	highest score	share of records
Most likely	0.54	0.99	32.8%
Somewhat likely	0.18	0.53	34.1%
Least likely	0.03	0.17	33.2%

External validity

Our model indicates that 40% of people with scored records have government-sponsored health insurance (Medicare or Medicaid). As of 2025, 36% of U.S. adults were receiving public health insurance.