insurance: employer sponsored

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 health insurance through their employer.

Trained on survey data, it leverages age, income proxies, household composition, and employment signals to estimate employer-sponsored coverage.

These scores help organizations understand coverage patterns and target outreach around employer benefit standards, COBRA protections, and the relationship between employment and health care access.

  • approach: boosted tree classifier
  • training rows: 5,380
  • last updated: 2026-09-06 19:54:53

Validation

metric value
area under ROC curve 0.78
area under precision-recall curve 0.73
accuracy 0.72
precision 0.68
recall 0.80
f1 score 0.74
mean log loss 0.55
gini coefficient 0.57
brier skill score 0.26

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
10 138 0.78 0.80 0.01 1.64
9 137 0.76 0.75 0.005 1.59
8 137 0.72 0.71 0.02 1.52
7 138 0.62 0.66 0.04 1.31
6 137 0.58 0.60 0.02 1.21
5 137 0.58 0.51 0.06 1.21
4 138 0.32 0.34 0.02 0.67
3 137 0.10 0.17 0.06 0.21
2 137 0.19 0.13 0.06 0.40
1 138 0.12 0.10 0.02 0.24

Architecture

This model is a boosted-tree classifier trained with BigQuery ML.

Hyperparameter configuration

  • max iterations: 20
  • learn rate: 0.3033
  • l1 regularization: 2.72e-11
  • l2 regularization: 0.0488
  • min relative progress: 0.01
  • early stop: TRUE
  • input label columns: health_insurance_employer_sponsored
  • data split method: AUTO_SPLIT
  • max tree depth: 3
  • subsample: 0.8978
  • min split loss: 0
  • num trials: 12
  • max parallel trials: 4
  • hparam tuning objectives: ROC_AUC
  • enable global explain: TRUE
  • category encoding method: LABEL_ENCODING
  • booster type: GBTREE
  • num parallel tree: 1
  • tree method: AUTO
  • min tree child weight: 1
  • instance weight column: sample_weight
  • xgboost version: 0.9
  • auto class weights: FALSE

Features

Importance

Interaction terms are split evenly between the features they combine.

Sourcing

Attom Property Data API

  • property tax billed

CDC WONDER

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

Revelio Labs

  • current position salary

State voter files (via TargetSmart)

  • age
  • female

TargetSmart

  • children present score
  • college graduate score
  • cord cutter score
  • evangelical score
  • high school only score
  • income rank
  • marriage score
  • paid leave support score
  • recession sensitivity score
  • veteran score
  • working class score

U.S. Census American Community Survey

  • block group % college degree
  • block group median rent
  • tract median income

Verisk (via TargetSmart)

  • homeowner
  • newspaper reader
  • retired
  • x (twitter) user

Implementation

Prediction distribution

statistic value
mean 0.53
std dev 0.24
min 0.13
25th 0.21
median 0.62
75th 0.74
max 0.82

Segments

segment lowest score highest score share of records
Most likely 0.71 0.82 32.6%
Somewhat likely 0.49 0.70 33.8%
Least likely 0.13 0.48 33.6%

External validity

Our model indicates that 53% of people with scored records have employer-sponsored health insurance. KFF Health Insurance Coverage of the Total Population and Census ACS 2023 (S2701) place employer-sponsored coverage at roughly 50% of U.S. adults.