Resolution Equity in NYC 311 Service Delivery
Overview
NYC's 311 system handles millions of service requests annually, from noise complaints to pothole repairs. But does every neighborhood get the same quality of response? This study uses the nyc311 equity analysis pipeline to quantify disparities in resolution times across community districts, decompose the sources of inequality, and test whether recent policy interventions have changed the pattern.
Data
The analysis examines 1,000,000 NYC 311 service requests covering May 2023 through December 2024. The balanced panel contains 76 community districts observed over 20 monthly periods (1,520 total observations). Mean monthly complaints per district: 657.9. Mean resolution rate: 21.3% (SD = 0.406).
Seasonal Patterns
STL decomposition reveals complaints peak in December and trough in April, with a seasonal amplitude of 269,484 complaints. The trend moved from -50,504 to 108,271 over the study period, indicating a steady increase in complaint volume. One anomalous observation was detected: July 2024 (z = +2.47).
Equity Analysis
A two-way fixed effects panel regression with entity and time fixed effects (clustered SE at the district level, N = 328, R-squared = 0.328) finds that higher complaint volume significantly predicts longer resolution times: a 1-unit increase in log complaints is associated with +4.0 additional days of median resolution time (p < 0.001).
Inequality Decomposition
The Theil T index is T = 0.2247, indicating moderate inequality. The decomposition is striking:
- Between-group component: 0.1657 (73.7% of total)
- Within-group component: 0.0590 (26.3% of total)
The majority of inequality is attributable to differences between boroughs rather than within them---the opposite pattern from rodent complaints, where within-borough variation dominates.
The largest contributors: Staten Island 01 (+0.156), Staten Island 03 (+0.089), Bronx 12 (+0.070), Staten Island 02 (+0.055), and Manhattan 12 (+0.020). Staten Island districts dominate the inequality index, likely reflecting their geographic isolation and different service delivery dynamics.
Oaxaca-Blinder Decomposition
Comparing low- vs. high-income districts: total gap = -0.77 days. Of this, 80% is explained by observable characteristics (complaint volume, composition) and 20% is unexplained. The small unexplained component suggests that overt discrimination is not the primary driver; instead, structural differences in complaint volume and type drive most of the disparity.
Spatial Analysis
Global Moran's I = 0.379 (z = 2.574, p = 0.007), indicating statistically significant positive spatial autocorrelation: districts with similar resolution times cluster together geographically. LISA identifies 3 HH clusters (high resolution time near other high areas), 9 HL outliers, and 7 LL clusters.
Policy Evaluation
PELT changepoint detection identifies one structural break at August 2024. Interrupted time series analysis around the March 2024 rat mandate finds a level change of -425,361 (p < 0.001) followed by a trend change of +34,886 per period (p < 0.001). The initial drop followed by an accelerating upward trend suggests a temporary disruption in the resolution pipeline followed by a rebound.
Limitations
Demographic covariates are from ACS 2022 PUMA-level estimates mapped to community districts; some CDs share PUMAs and have identical demographic values. Spatial weights use a 3 km distance threshold and results are sensitive to this choice. The reporting bias EM model assumes a Poisson-logistic structure that may not hold for all complaint types.
The full analysis pipeline is available in the nyc311 examples.
Built with nyc311 v0.3.0.
