NJ Car Crash Data

Analysis of NJ State Police fatal crash data (2008-present) and NJDOT crash data (2001-2023)

• Plots: neighbor-ryan.org/nj-crashes
• Hudson County map: neighbor-ryan.org/nj-crashes/hudson/map

Fatal crash data updates daily (via GitHub Action), plots below are current through either the most recent month or year.

NJ traffic deaths w/ 12mo rolling avg

2021 and 2022 set 17-year highs for traffic deaths in NJ

(NJSP only includes victim-type data since 2020)

In NJ, traffic deaths are 1.5x-2x homicides

(Uniform Crime Reports 1990-2020, Disaster Center 1960-2019)

Traffic deaths, by month

---

Setup

This project uses uv for Python dependency management:

# Install dependencies
uv sync

# Run scripts
uv run njsp refresh_data
uv run njdot compute db

# Or activate the virtual environment
source .venv/bin/activate
njsp refresh_data

Data Management

This repository tracks two distinct crash datasets with different update patterns:

NJSP (Daily Updates, Git-tracked)

• Source: NJ State Police fatal crash XMLs
• Coverage: Fatal crashes only, 2008-present
• Update Frequency: Daily (via GitHub Action)
• Size: Small (~1-2MB databases)
• Storage: XMLs committed to git, databases regenerated locally
• Regenerate:

  njsp refresh_data  # By default, current year and 2 preceding
  njsp update_data   # Generate parquets from XMLs
  njsp update_pqts   # Update SQLite databases

NJDOT (Annual Updates, DVC-tracked)

• Source: NJDOT crash data (zip files)
• Coverage: All crashes (fatal, injury, property damage), 2001-present
• Update Frequency: Annual
• Size: Large (2.4GB crashes.db, ~280MB parquets)
• Storage: DVC-tracked, stored in S3
• Pipeline: See njdot/README.md
• Pull databases:

  dvc pull www/public/njdot/crashes.db.dvc
  # Or regenerate (slow, requires geo-processing):
  env -u PYTHONPATH njdot compute pqt -f
  env -u PYTHONPATH njdot compute db -f

Analysis

• parse-njsp-xmls.ipynb
• nj-crime-stats.ipynb
• Hudson Crashes.ipynb

NJDOT crash data parsing / analysis in njdot/.