excessive costs: child care

Overview

This model predicts whether a parent is experiencing excessive child care costs — a felt-burden indicator combining direct survey reports of difficulty paying for child care with the category share of household expenses.

Trained on survey data, it uses household composition, income proxies, child-care-aged-population features, and area-level cost indicators to identify families for whom child-care costs represent a meaningful burden under current macroeconomic conditions.

Organizations can use these scores to target outreach around universal pre-K, child care subsidies, Child Tax Credit expansion, and family economic policy.

approach: deep learning
training rows: 5,493
last updated: 2026-05-02 15:49:55

Validation

metric	value
area under ROC curve	0.90
area under precision-recall curve	0.15
accuracy	0.53
precision	0.06
recall	1
f1 score	0.11
mean log loss	0.95
gini coefficient	0.79
brier skill score	0.09

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	141	0.13	0.16	0.02	4.84
9	140	0.09	0.08	0.02	3.34
8	140	0.04	0.03	0.007	1.28
7	140	0.01	0.01	0.005	0.51
6	140	0	0.003	0.003	0.00
5	140	0	0.001	0.001	0.00
4	140	0	0.0007	0.0007	0.00
3	140	0	0.0003	0.0003	0.00
2	140	0	0.0001	0.0001	0.00
1	141	0	0	0	0.00

Architecture

This model is a deep neural network trained with TensorFlow.

Network structure

layer	units	activation
↓ dense	128	relu
↓ batch normalization
↓ dense	91	swish
↓ batch normalization
↓ dense	54
↓ batch normalization

Hyperparameter configuration

loss: binary_focal_crossentropy
optimizer: nadam
batch size: 8
activation: hard_sigmoid
label smoothing: 0
learning rate: 0.001

Features

Importance

Sourcing

Revelio Labs

current position tenure (days)

State voter files (via TargetSmart)

TargetSmart

children present score
cord cutter score
non-presidential primary turnout score

Implementation

Prediction distribution

statistic	value
mean	0.02
std dev	0.05
min	0
25th	0
median	0
75th	0.02
max	0.21

Segments

segment	lowest score	highest score	share of records
Most likely	0.19	0.21	2.5%
Somewhat likely	0.15	0.18	2.2%
Least likely	0	0.14	95.4%

External validity

Our model indicates that 2% of people with scored records report excessive child care costs. About 10% of U.S. adults have a child under age 6 at home (2024 BLS), and roughly 43% of those families spend more than HHS’s formal affordability benchmark for child care (2025 Center for American Progress).