manual occupation

Overview

This model predicts whether a voter is employed in manual labor, a trade, or manufacturing — including construction, mechanic work, production, and W-2 driving.

Trained on survey data, it uses salary levels, working-class indicators, and economic sentiment features to identify likely manual workers.

Organizations can use these scores for outreach around workplace safety, trade apprenticeships, and labor policy.

  • approach: deep learning
  • training rows: 13,402
  • last updated: 2026-05-06 04:54:33

Validation

metric value
area under ROC curve 0.76
area under precision-recall curve 0.19
accuracy 0.54
precision 0.10
recall 0.81
f1 score 0.17
mean log loss 0.90
gini coefficient 0.52
brier skill score 0.07

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
10 332 0.21 0.22 0.004 3.60
9 331 0.10 0.11 0.01 1.70
8 332 0.05 0.07 0.02 0.93
7 331 0.07 0.05 0.02 1.19
6 332 0.04 0.04 0.003 0.72
5 331 0.04 0.03 0.01 0.72
4 331 0.03 0.03 0.002 0.46
3 332 0.03 0.02 0.007 0.46
2 331 0.009 0.01 0.006 0.15
1 332 0.003 0.008 0.006 0.05

Architecture

This model is a deep neural network trained with TensorFlow.

Network structure

layer units activation
↓ dense 128 swish
↓ dropout (0.10 rate)
↓ dense 100 swish
↓ dropout (0.08 rate)
↓ dense 72 swish
↓ dropout (0.05 rate)
↓ dense 44 swish

Hyperparameter configuration

  • loss: binary_focal_crossentropy
  • optimizer: nadam
  • batch size: 8
  • activation: sigmoid
  • label smoothing: 0
  • learning rate: 5e-04

Features

Importance

Sourcing

CDC WONDER

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

State voter files (via TargetSmart)

  • age
  • female
  • registered democrat
  • registered republican

TargetSmart

  • college graduate score
  • double negative gop score
  • midterm general turnout score
  • partisan score
  • presidential primary turnout score
  • pro-choice score
  • progressive tax support score
  • race: latino
  • trump support score
  • working class score

U.S. Census American Community Survey

  • block group % hispanic

Verisk (via TargetSmart)

  • service occupation

Implementation

Prediction distribution

statistic value
mean 0.11
std dev 0.08
min 0
25th 0.04
median 0.08
75th 0.18
max 0.40

Segments

segment lowest score highest score share of records
Most likely 0.24 0.40 11.5%
Somewhat likely 0.20 0.23 10.3%
Least likely 0 0.19 78.3%

External validity

Our model indicates that 11% of people with scored records work in manual labor, a trade, or manufacturing (construction, mechanic, production, W-2 driving). According to the 2024 ACS 1-year PUMS, ~15% of U.S. adults (~40M) are employed in a role that we consider to be ‘manual.’ (To aggregate up to this category, we combined counts for BLS SOC major groups 37 (building & grounds), 45 (farming), 47 (construction & extraction), 49 (installation/maintenance/repair), 51 (production), and 53 (transportation & material moving).)