evidoria

← Back to browse

Good practice Imported

The Allegheny Housing Assessment — Pittsburgh's Machine-Learning Tool Improves Housing Prioritization, But an Independent Study Finds Racial Disparities Persist

United States of America · Pittsburgh · See the United States of America profile · See the Pittsburgh profile

Evidence: Quasi-experimental Top 46% 60/100 · Ask Evidence Copilot about this practice

Allegheny County, Pennsylvania replaced a biased screening survey with a machine-learning score (AHA) to prioritize homelessness services. It predicts risk better, but a peer-reviewed study found racial disparities in who actually gets housed did not improve.

18,008
Historical assessments used to train AHA (2016–2024)
0.66–0.72
Model AUC range (2025 validation)
29–58 %
Positive predictive value range
The Allegheny Housing Assessment — Pittsburgh's Machine-Learning Tool Improves Housing Prioritization, But an Independent Study Finds Racial Disparities Persist

Details

Maturity
Established
Promoter
Allegheny County Department of Human Services
Period
2020–2025
Keywords
homelessness services, social services, housing, human services, public health

Context

Since August 2020, Allegheny County has used the Allegheny Housing Assessment (AHA), a predictive model trained on 18,008 historical housing assessments, to prioritise scarce supportive-housing access, replacing the VI-SPDAT survey shown to score white clients as lower-risk than Black clients with similar circumstances.

Results

A peer-reviewed 2024 study found that after deployment, actual service decisions became better aligned with the model's own risk scores and the score distribution became similar across racial groups — but the same study found the underlying racial disparity in who actually receives services did not shrink, tracing this to a poor-data-quality fallback survey tool and demographic differences in eligibility criteria.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
Allegheny County Department of Human Services

Conditions for success

  • Trained on the county's integrated data warehouse (18,008+ assessments)

Common failure modes

  • Racial disparity in service receipt persisted despite fairer risk scores, due to a poor-quality fallback survey tool and eligibility-criteria differences

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful