I mapped almost every USA traffic death in the 21st century

Original Article ↗ Hacker News Discussion ↗

Data source & scope

  • Map uses NHTSA’s Fatality Analysis Reporting System (FARS), aggregated from state and local law-enforcement crash reports.
  • Coverage is 2001‑01‑01 to 2023‑01‑01; only fatal crashes are included, not all crashes or injuries.
  • Several commenters cross‑checked local incidents: many matched well, some were missing or had incorrect attributes (e.g., ages, seatbelt use, locations).

Site performance & implementation

  • The site was repeatedly “hugged to death” by HN traffic: slow loads, 500s, partial data loading, and mobile issues.
  • Backend is PostgreSQL/PostGIS serving GeoJSON; performance concerns around generating large responses on the fly.
  • Suggestions:
    • Pre‑generate vector tiles (Tippecanoe, Planetiler, FlatGeobuf, MBTiles) and host on S3 or similar.
    • Consider MapLibre/Mapbox instead of pure Leaflet.
    • Move some filtering to the frontend; use expressions instead of regenerating GeoJSON each change.
    • Use caching, CDN (e.g., Cloudflare), and proper Postgres tuning.

Desired features & UI feedback

  • Common requests:
    • Filters (time of day, season, DUI, speeding, pedestrian/cyclist, multi‑vehicle, medical events, vehicle type).
    • Heatmap or density visualization instead of (or in addition to) individual pins.
    • Normalization by traffic volume or population to distinguish “busy” vs “dangerous” roads.
    • High‑level statistics and rankings (dangerous corridors, intersections).
    • Clearer entry into the map (many users didn’t realize the title image is a link).
  • Some users want routing or “risk scores” by road segment; others want a per‑place search to find specific incidents.

Data quality & interpretation

  • Data is shaped by a multi‑stage pipeline (local → state → federal); quality varies widely by jurisdiction.
  • Problems mentioned: inaccurate coordinates, mis‑classified locations (e.g., Manhattan as Flushing), inconsistent cause coding, missing incident types, and non‑standardized narratives.
  • Self‑reported and human‑entered fields (e.g., speeding, substance use, some questionnaires) are especially unreliable.
  • Commenters note that fatality locations do not always align with “highest crash” locations; low‑frequency but severe sites differ from everyday fender‑bender hotspots.

Traffic safety debates

  • Many use the map to argue that US road design is unusually dangerous versus other rich countries, citing:
    • High speeds, wide “stroads,” car‑centric planning, driveways and frequent access points, and lack of traffic calming.
  • Others emphasize trade‑offs:
    • Desire for fast, wide roads and shorter commutes vs. increased fatalities and injuries.
    • Environmental impacts of slower, stop‑and‑go traffic vs. benefits of reduced crashes and mode shift.
  • Strong debate over:
    • Road design vs enforcement (automated speed cameras, tech‑based limits).
    • Metrics: per‑capita deaths vs per‑mile/per‑vehicle vs absolute counts.
    • Feasibility and timescale of redesigning cities, adding transit, or building Dutch‑style bike networks.