Rat Containerization Policy Evaluation

Background

New York City's containerization mandate required businesses and residences to set out trash in sealed containers rather than loose bags. The pilot launched in lower Manhattan community districts in June 2024, with citywide enforcement beginning November 12, 2024.

This case study applies the full nyc311 causal inference pipeline---synthetic control, staggered difference-in-differences, event study, and regression discontinuity---to evaluate the policy's effect on rodent complaints.

Data

The analysis draws on 81,467 real Rodent-type 311 complaints (January 2023 through December 2024), structured as a balanced panel: 70 community districts across 24 monthly periods (1,680 observations). Nine Manhattan districts (CDs 01--09) received treatment beginning June 2024; 61 remaining districts serve as controls. ACS 2022 five-year demographics provide population, race, income, and tenure covariates.

Descriptive Findings

Brooklyn dominates rodent complaints at 37.9% of the total (30,861 complaints), followed by Manhattan (26.3%) and Queens (17.8%). The worst-hit districts are Manhattan 11 (East Harlem, 188 complaints/month), Brooklyn 03 (Bedford-Stuyvesant, 151/month), and Brooklyn 01 (Williamsburg, 150/month).

The overall resolution rate is 97.3%, with "Rat Sighting" (62%) and "Condition Attracting Rodents" (18%) as the dominant descriptors. Topic concentration is very high (HHI = 0.88), meaning most complaints fall into just a few categories. The recurrence rate of 13.1% indicates that roughly 1 in 8 complaints comes from a repeat location.

Temporal Patterns

STL decomposition reveals a strong seasonal cycle peaking in August and troughing in December, with a seasonal amplitude of 2,464 complaints. One anomalous month was detected: August 2023 (z = +2.21). PELT changepoint detection found no structural breaks in the 24-month series---the containerization mandate did not produce a visible shift in the aggregate city-wide trend.

Per-borough trends show modest declines everywhere (Manhattan -9.0%, Brooklyn -3.8%, Bronx -3.7%, Staten Island -3.9%, Queens -0.3%), consistent with a city-wide downward drift rather than a treatment-specific effect.

Inequality Structure

The Theil T index is T = 0.165, with 67.6% attributable to within-borough differences and only 32.4% between boroughs. This is a key insight: the rodent burden varies dramatically within boroughs. Manhattan 11 has 5x the city median while Manhattan 01 has near zero.

Moran's I is 0.19 (p = 0.064), marginally insignificant---rodent complaints show weak spatial clustering. LISA identifies 2 HH hotspots (Brooklyn 04, Queens 05) and 11 LL coldspots.

Causal Evidence

The causal results are mixed and instructive:

Synthetic control (Manhattan 03 vs. counterfactual): ATT = -2.6 complaints/month, suggesting a modest reduction. The pre-treatment MSPE of 24.4 indicates an imperfect donor match.

Staggered DiD (9 treated Manhattan CDs vs. control): ATT = +6.8 complaints/month (p = 0.011, 95% CI [+1.6, +12.0]). Contrary to expectations, treated districts saw a rise relative to controls. The pre-trend F-test (F = 0.48, p = 0.78) strongly supports parallel trends, so the increase appears real. This likely reflects a reporting-awareness surge from media coverage rather than actual increases in the rat population.

Event study: Post-treatment coefficients range from -2.1 to +10.1, largest immediately post-mandate, consistent with a surge that attenuates over time.

RDD (haversine distance from treated-zone centroid): estimate = -91.2, p = 0.88---no significant discontinuity at the spatial boundary.

Power Analysis

The MDE is 17.2 complaints at 80% power. Both the SCM ATT (-2.6) and DiD ATT (+6.8) fall below this threshold. A definitive evaluation would require longer post-treatment data, rodent activity measures beyond 311 complaints, and explicit modeling of the reporting-awareness channel.

Limitations

Treatment assignment is based on reported pilot rollout dates; exact enforcement timing may vary. The RDD running variable uses haversine distance from the treated-zone centroid rather than the precise mandate boundary polygon. Rodent complaints reflect reporting behavior, not underlying rodent populations---and reporting propensity may shift with policy awareness.

The full analysis pipeline and all scripts are available in the nyc311 examples.

Built with nyc311 v0.3.0.