enrolled in medicaid

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 enrolled in Medicaid — the joint federal-state program providing health coverage to low-income individuals and families.

Trained on survey data and consuming AHRQ-derived ecological-disaggregation outputs as features, it functions primarily as a ranking tool, enabling organizations to prioritize voters who would materially benefit from Medicaid preservation or expansion.

approach: deep learning
training rows: 15,541
last updated: 2026-08-29 19:49:51

Validation

metric	value
area under ROC curve	0.71
area under precision-recall curve	0.31
accuracy	0.56
precision	0.19
recall	0.73
f1 score	0.31
mean log loss	0.84
gini coefficient	0.41
brier skill score	0.07

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	391	0.39	0.39	0.007	2.92
9	390	0.23	0.18	0.05	1.77
8	390	0.15	0.14	0.01	1.11
7	391	0.12	0.12	0.0009	0.87
6	390	0.08	0.10	0.02	0.64
5	390	0.09	0.09	0.005	0.68
4	391	0.08	0.09	0.01	0.58
3	390	0.05	0.08	0.03	0.41
2	390	0.08	0.07	0.005	0.58
1	391	0.06	0.06	0.002	0.45

Architecture

This model is a deep neural network trained with TensorFlow.

Network structure

layer	units	activation
↓ dense	96	relu
↓ batch normalization
↓ dropout (0.30 rate)
↓ dense	70	swish
↓ batch normalization
↓ dropout (0.20 rate)
↓ dense	44
↓ batch normalization

Hyperparameter configuration

loss: binary_crossentropy
optimizer: nadam
batch size: 8
activation: sigmoid
label smoothing: 0.01
learning rate: 0.001

Features

Importance

Sourcing

Attom Property Data API

bedrooms
lot size (acres)

CDC WONDER

utility shut-off threat, past 12 mo (adults)

OpenSecrets

donations to democrats (%)
donations total (count)

Revelio Labs

avg salary differential
current position salary

State voter files (via TargetSmart)

TargetSmart

child tax credit support score
children present score
evangelical score
high school only score
homeowner score
ideology score
off-year general turnout score
race: black
working class score

U.S. Census American Community Survey

block group % high school diploma
block group % hispanic
block group % white

Verisk (via TargetSmart)

homeowner
medical occupation
political donations
residence year built
retired

Implementation

Prediction distribution

statistic	value
mean	0.19
std dev	0.16
min	0.03
25th	0.09
median	0.13
75th	0.24
max	0.99

Segments

segment	lowest score	highest score	share of records
Most likely	0.29	0.99	19.6%
Somewhat likely	0.17	0.28	18.6%
Least likely	0.03	0.16	61.9%

External validity

Our model indicates that 19% of people with scored records are enrolled in Medicaid. CBO 2024 Medicaid baseline reports ~34M nonelderly nondisabled adults plus ~10M disabled adults plus ~6M dual-eligibles enrolled in Medicaid (~50M adults total / ~266M U.S. adults ≈ 19%). See also the KFF Medicaid Enrollment and Unwinding Tracker.