owns home

Overview

This model predicts whether a voter owns their home.

Trained on survey data, it combines property records, consumer residence indicators, lot size, and household composition to estimate homeownership status.

Organizations can use these scores to segment outreach around housing policy — from property tax relief for homeowners to renter protections and affordable housing advocacy.

  • approach: boosted tree classifier
  • training rows: 4,921
  • last updated: 2026-09-06 18:57:24

Validation

metric value
area under ROC curve 0.87
area under precision-recall curve 0.92
accuracy 0.84
precision 0.86
recall 0.91
f1 score 0.89
mean log loss 0.39
gini coefficient 0.74
brier skill score 0.41

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
10 121 0.95 0.94 0.01 1.33
9 120 0.95 0.93 0.02 1.32
8 120 0.93 0.92 0.01 1.30
7 120 0.93 0.91 0.02 1.30
6 120 0.88 0.90 0.02 1.22
5 120 0.91 0.88 0.03 1.27
4 120 0.72 0.80 0.08 1.01
3 120 0.50 0.54 0.04 0.70
2 120 0.28 0.24 0.03 0.38
1 121 0.12 0.13 0.002 0.17

Architecture

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

Hyperparameter configuration

  • max iterations: 20
  • learn rate: 0.3565
  • l1 regularization: 1.367e-11
  • l2 regularization: 7e-04
  • min relative progress: 0.01
  • early stop: TRUE
  • input label columns: family_owns_home
  • data split method: AUTO_SPLIT
  • num trials: 12
  • max parallel trials: 4
  • hparam tuning objectives: ROC_AUC
  • enable global explain: TRUE
  • auto class weights: FALSE
  • max tree depth: 5
  • subsample: 0.8112
  • min split loss: 0
  • 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

Features

Importance

Interaction terms are split evenly between the features they combine.

Sourcing

Revelio Labs

  • avg salary differential
  • industry sector

State voter files (via TargetSmart)

  • age
  • registered democrat

TargetSmart

  • abortion issue motivation score
  • catholic score
  • climate change concern score
  • college graduate score
  • double negative downballot gop score
  • early vote timing score
  • evangelical score
  • homeowner score
  • marriage score
  • midterm general turnout score
  • presidential general turnout score
  • progressive tax support score
  • race: latino
  • race: white
  • recession sensitivity score

U.S. Census American Community Survey

  • tract median income

Verisk (via TargetSmart)

  • children age 0–5 in household
  • homeowner
  • number of adults in household
  • residence tenure
  • spanish speaker

Implementation

Prediction distribution

statistic value
mean 0.55
std dev 0.33
min 0.06
25th 0.18
median 0.60
75th 0.90
max 0.95

Segments

segment lowest score highest score share of records
Most likely 0.86 0.95 32.9%
Somewhat likely 0.29 0.85 32.9%
Least likely 0.06 0.28 34.2%

External validity

Our model indicates that 55% of people with scored records own their home. In 2024, 63% of U.S. adults owned or co-owned their home (2025 Federal Reserve, Economic Well-Being of U.S. Households).