insurance: government subsidized

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 receives government-subsidized health insurance, such as ACA marketplace plans with premium tax credits.

Trained on survey data, it incorporates age, employment sector, education, and residence characteristics to identify likely subsidy recipients.

These scores help organizations target outreach around subsidy awareness, enrollment assistance, and coverage affordability.

  • approach: dnn classifier
  • training rows: 5,465
  • last updated: 2026-09-14 02:43:40

Validation

metric value
area under ROC curve 0.62
area under precision-recall curve 0.08
accuracy 0.52
precision 0.07
recall 0.67
f1 score 0.13
mean log loss 0.98
gini coefficient 0.25
brier skill score 0.01

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
10 140 0.09 0.11 0.02 1.60
9 140 0.09 0.08 0.007 1.60
8 139 0.09 0.07 0.02 1.74
7 140 0.03 0.06 0.04 0.53
6 140 0.06 0.06 0.009 1.20
5 139 0.06 0.05 0.009 1.07
4 140 0.03 0.04 0.01 0.53
3 139 0.04 0.03 0.009 0.80
2 140 0.03 0.03 0.002 0.53
1 140 0.02 0.02 0.006 0.40

Architecture

This model is a deep neural network classifier trained with BigQuery ML.

Hyperparameter configuration

  • max iterations: 20
  • learn rate: 0.0028
  • l1 regularization: 0
  • l2 regularization: 0
  • min relative progress: 0.01
  • warm start: FALSE
  • early stop: TRUE
  • input label columns: health_insurance_government_subsidized
  • data split method: AUTO_SPLIT
  • hidden units: 32, 16
  • batch size: 128
  • dropout: 0.2398
  • num trials: 18
  • max parallel trials: 5
  • hparam tuning objectives: ROC_AUC
  • enable global explain: FALSE
  • tf version: 1.15
  • auto class weights: FALSE
  • activation fn: elu
  • optimizer: sgd

Features

Importance

Sourcing

Agency for Healthcare Research and Quality

  • tract % direct-purchase insurance only
  • tract % uninsured

Revelio Labs

  • min salary

State voter files (via TargetSmart)

  • age

TargetSmart

  • double negative democrat score
  • double negative downballot dem score
  • ideology score
  • non-presidential primary turnout score
  • pro-choice score
  • race: aapi
  • race: latino

U.S. Census American Community Survey

  • block group % some college

U.S. Census CPS Annual Social and Economic Supplement

  • current medicare coverage
  • private insurance policyholder

Verisk (via TargetSmart)

  • started college

Implementation

Prediction distribution

statistic value
mean 0.04
std dev 0.02
min 0
25th 0.03
median 0.04
75th 0.05
max 0.08

Segments

segment lowest score highest score share of records
Most likely 0.08 0.08 1.7%
Somewhat likely 0.07 0.07 5.2%
Least likely 0 0.06 93.0%

External validity

Our model indicates that 4% of people with scored records receive subsidized ACA marketplace coverage. In 2025, ~22M marketplace enrollees qualified for premium tax credits (2025 CMS) — roughly 8% of U.S. adults. With the expiration of the expanded tax credits, enrollment is expected to decrease by up to 42% in the coming years (2024 Urban Institute).