does not visit pcp

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 has not visited a primary care provider in the past year — the policy-relevant minority disconnected from preventive care.

Trained on survey data, it uses age, insurance coverage signals, income proxies, and area-level health-access indicators to identify voters who skip routine primary care.

Organizations can use these scores for preventive-health outreach, community-clinic enrollment campaigns, and primary-care-access advocacy.

  • approach: dnn classifier
  • training rows: 5,382
  • last updated: 2026-09-06 19:43:53

Validation

metric value
area under ROC curve 0.68
area under precision-recall curve 0.18
accuracy 0.54
precision 0.14
recall 0.70
f1 score 0.23
mean log loss 0.92
gini coefficient 0.37
brier skill score 0.04

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 0.24 0.24 0.002 2.40
9 137 0.18 0.17 0.01 1.83
8 137 0.08 0.14 0.06 0.81
7 138 0.11 0.11 0.004 1.09
6 137 0.09 0.09 0.004 0.88
5 137 0.12 0.07 0.05 1.24
4 138 0.07 0.06 0.01 0.73
3 137 0.04 0.05 0.007 0.44
2 137 0.03 0.04 0.01 0.29
1 138 0.03 0.03 0.0002 0.29

Architecture

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

Hyperparameter configuration

  • max iterations: 20
  • learn rate: 0.0027
  • l1 regularization: 0
  • l2 regularization: 0
  • min relative progress: 0.01
  • warm start: FALSE
  • early stop: TRUE
  • input label columns: health_does_not_visit_pcp
  • data split method: AUTO_SPLIT
  • hidden units: 64, 32, 16
  • batch size: 32
  • dropout: 0.2231
  • 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

State voter files (via TargetSmart)

  • age

TargetSmart

  • early vote timing score

Implementation

Prediction distribution

statistic value
mean 0.16
std dev 0.10
min 0
25th 0.06
median 0.14
75th 0.24
max 0.45

Segments

segment lowest score highest score share of records
Most likely 0.28 0.45 16.1%
Somewhat likely 0.22 0.27 15.0%
Least likely 0 0.21 68.9%

External validity

Our model indicates that 16% of people with scored records did not see a primary care provider in the past year. SHADAC analysis of National Health Interview Survey (NHIS) data, 2022–2023 reports that ~86% of U.S. adults had a general doctor or provider visit in the past year and ~14% did not.