Experiment: reproduce Kansas City COVID-311 paper (Tran et al. 2023) on SF data

[alexstorer]

Hypothesis / question

Do the findings of Tran, Bani-Yaghoub & DeLisle (2023), "Non-emergency responses in the 311 system during the early stage of the COVID-19 pandemic: a case study of Kansas City" (DOI: 10.20517/dpr.2022.08, CC-BY 4.0) reproduce on San Francisco 311 data — and where does SF fall in the cross-city picture the paper describes (NYC noise ↑ vs Dallas noise ↓)?

Paper's core claims to test (Mar–Aug 2019 vs 2020)

1. Aggregate decline in total 311 volume (KC: −13%).
2. Decline is not universal — a handful of categories surge while street-condition categories fall with reduced mobility.
3. Text-mine the free-text description to tag COVID-related requests (KC: 20 keywords → 2,379 = 4.3%; mask & social-distancing dominate).
4. Channel shift — COVID requests skew to phone/email, away from web.

Method

• Pull all-category service_name counts for both Mar–Aug windows from Socrata (vw6y-z8j6) via aggregation queries (no bulk download); build SF's equivalent of the paper's Table 1.
• Place graffiti and noise within that ranked table (ties into this repo's existing noise-vs-graffiti work).
• Probe service_details for COVID keywords to test whether claim 3's method is even applicable to SF.
• Compare the source distribution across the two windows (claim 4).

Tasks

• Reproduction script (scripts/covid_category_reproduction.py) with validation assertions
• Processed outputs: category-change table + summary CSV
• Figure: category movers + noise/graffiti divergence
• README section documenting the cross-city reproduction & novelty verdict
• PR linking this issue

Expected value

Adds SF as a data point to the paper's spatial-heterogeneity thesis, and stress-tests whether its text-mining method generalizes.