| editor | ||||
|---|---|---|---|---|
|
Authors: Angel Rutherford, Ixchel Ramirez, Tess Vu
Affiliation: University of Pennsylvania | MUSA 5080: Public Policy Analytics
Date: December 8, 2025
Eviction is both a cause and consequence of poverty that destabilizes entire neighborhoods. Currently, city responses to eviction are reactive, with resources like legal aid and rental assistance deployed after filing volumes become a crisis. This project develops a Real-Time Operational Tool for the Philadelphia Office of Homeless Services and the Fair Housing Commission. By shifting from reactive to predictive analysis, we enable the city to allocate limited staff to specific census tracts predicted to experience elevated eviction filings in the coming month.
Our Negative Binomial regression model leverages temporal momentum, spatial spillover effects, policy intervention effects, property tax delinquency stress, and American Community Survey socioeconomic indicators to forecast monthly eviction filing counts at the census tract level.
The model demonstrates strong performance with meaningful improvement, and sets the foundation for building up to a practical and usable tool down the line. Using a robust temporal validation strategy (training through 2023, testing on 2024-2025), the model generalizes well to future periods without overfitting. Also stark racial disparities in eviction burden was identified, with Black-majority tracts accounting for disproportionate shares of filings. These findings emphasize the need for equity-centered implementation safeguards to prevent perpetuating existing disparities through algorithmic resource allocation.
"Where should renter's assistance programs be targeted in Philadelphia?"
Target Variable: Monthly Count of Eviction Filings per Census Tract.
ACS 2023 Data via tidycensus API
Eviction data is zero-inflated (37% of tracts have 0 filings) and overdispersed (Variance > Mean). Standard OLS or Poisson models fail to capture the unpredictable, volatile nature of eviction spikes.
We trained a series of Negative Binomial Models, progressively adding complexity:
- Model 1 (Baseline): Time + Seasonality.
- Model 2 (+ Actionable): Adds Tax Delinquency + Spatial Lag.
- Model 3 (+ Structural): Adds ACS Demographics.
- Model 4 (+ Interaction): Tests Racial Disparities in Policy Impact.
To handle extreme "mass displacement events" (e.g., 694 filings in a single tract), we developed a robust capping strategy where the target variable capped at the 99th percentile (20 filings) to stabilize coefficients, created an is_extreme_spike flag to explicitly model crisis propensity, and evaluated on raw, uncapped data to prove real-world usefulness.
Our final model (Model 3 + ACS) achieved the best balance of accuracy and stability.
-
Test MAE: 1.48 (The model is accurate to within < 2 filings per tract).
-
Stability: The model performed better on the 2024-2025 Test Set than the Training Set, proving it is not overfit and can generalize well to unseen, real-world data.
-
Disparate Impact: Black-majority tracts face a structurally higher baseline risk that standard economic variables cannot fully explain.
-
Model Fairness: The model's error rate (MAE) is slightly higher in Black tracts, indicating higher unmeasured volatility in these neighborhoods that are historically and currently most vulnerable.
This model's future would be best used as a monthly Triage Dashboard:
- Input: Load previous month's filing data on 1st of month.
- Output: Generate list of the Top 50 "Critical Risk" Tracts.
- Potential Action:
- Deploy Canvassers: Aid 50 tracts before any major court dates in the system.
- Direct Mail: Send "Know Your Rights" flyers to all rental units within zip codes.
- Legal Aid Pop-Ups: Establish temporary clinics in these specific zones.
Ethical Safeguard: This tool must be used strictly for providing resources, never for automated decision-making or punitive enforcement.