nflverse Play-by-Play and Player Data
nflverse (open-source community)
Every NFL play since 1999 with expected points and win probability already computed.
Automated releases from the nflverse project (nflfastR, nflreadr) covering NFL play-by-play, player and team stats, rosters, schedules, draft picks, and more. Each play-by-play row is one snap with down, distance, yard line, play type, and model outputs such as EPA and win probability. It is the default public data source for NFL analytics.
Access
- Cost
- FreeFree via GitHub releases or the nflreadr (R) and nflreadpy (Python) packages.
- License
- Open licenseCC BY 4.0
- Sign-up
- No sign-up
- How you get it
- Bulk downloadParquet, CSV, RDS
What's in it
- One row is
- One row per play (play-by-play release); other releases per player-week or per game
- Covers
- National Football League
- Years
- 1999-present (play-by-play)
- Updated
- Daily
- Size
- Roughly 45,000-50,000 plays per season
Quality
Actively maintained, documented with a data dictionary, and updated automatically during the season.
- EPA and win probability are model outputs, not raw observations
- Underlying play data is scraped from NFL feeds and can change after corrections
Connects to
- NFL Big Data Bowl Tracking Data: Player-tracking frames for a subset of the same plays, joinable on game and play IDs
- CollegeFootballData.com API: College play-by-play and recruiting data to follow players before the draft
Questions it could answer
- How much has fourth-down aggressiveness changed since 2018?
- Which quarterbacks add the most expected points per dropback after adjusting for opponent?