service occupation

Overview

This model predicts whether a voter is employed in a service occupation — including food service, retail, teaching, nursing, and social work. The matter survey’s “service” bucket bundles consumer-facing service work with caring-economy professions, so the prediction lands somewhere between a narrow consumer-service definition and a broad service-plus-caring-economy one.

Trained on survey data, it incorporates wage levels, demographic features, and economic indicators.

Organizations can use these scores to reach service and caring-economy workers with outreach around wages, scheduling, health benefits, and workplace protections.

approach: dnn classifier
training rows: 12,020
last updated: 2026-09-14 03:46:18

Validation

metric	value
area under ROC curve	0.71
area under precision-recall curve	0.21
accuracy	0.55
precision	0.15
recall	0.77
f1 score	0.26
mean log loss	0.88
gini coefficient	0.42
brier skill score	0.06

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	294	0.25	0.25	0.003	2.51
9	293	0.18	0.19	0.01	1.79
8	293	0.12	0.14	0.03	1.17
7	293	0.13	0.11	0.02	1.31
6	294	0.09	0.09	0.006	0.93
5	293	0.05	0.06	0.01	0.52
4	293	0.09	0.05	0.04	0.93
3	293	0.04	0.04	0.0008	0.41
2	293	0.02	0.03	0.02	0.17
1	294	0.02	0.03	0.002	0.24

Architecture

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

Hyperparameter configuration

optimizer: adagrad
max iterations: 20
learn rate: 0.0042
l1 regularization: 0
l2 regularization: 0
min relative progress: 0.01
warm start: FALSE
early stop: TRUE
input label columns: employment_type_service
data split method: AUTO_SPLIT
hidden units: 128, 64, 32
batch size: 32
dropout: 0.3641
num trials: 18
max parallel trials: 5
hparam tuning objectives: ROC_AUC
enable global explain: FALSE
tf version: 1.15
activation fn: relu
auto class weights: FALSE

Features

Importance

Sourcing

Attom Property Data API

gross building area
property tax billed

Federal Reserve

county debt-to-income decile

Revelio Labs

avg tenure (days)
broad job category
current position salary
current position tenure (days)
occupational prestige score

State voter files (via TargetSmart)

TargetSmart

children present score
non-religious score
paid leave support score

U.S. Census American Community Survey

block group % aapi
block group % graduate degree

Implementation

Prediction distribution

statistic	value
mean	0.14
std dev	0.08
min	0.03
25th	0.06
median	0.13
75th	0.20
max	0.34

Segments

segment	lowest score	highest score	share of records
Most likely	0.24	0.34	14.0%
Somewhat likely	0.20	0.23	14.4%
Least likely	0.03	0.19	71.7%

External validity

Our model indicates that 14% of people with scored records work in a service occupation (food service, retail, teaching, nursing, social work). According to the 2024 ACS 1-year PUMS, ~13% of U.S. adults (~36M) are employed in a role that we consider to be ‘service.’ (To aggregate up to this category, we combined counts for BLS SOC major groups 31 (healthcare support), 35 (food prep & serving), and 39 (personal care), plus security and other protective service (group 33 excl. supervisors/fire/law enforcement), cashiers and retail salespersons (41-2) and retail supervisors (41-1011), and remaining sales.