Red Zone Math

About

Red Zone Math is an independent, non-commercial NFL analytics site: no ads, no paid features, no tracking.

What’s here

  • QBs, backs, receivers and defenders: season and career leaderboards, with EPA and our win probability added (WPA).
  • Teams: offense and defense per play, adjusted for the opponents faced.
  • Coaches: every 4th-down go, field goal and punt call compared with our model’s call in the same situation.
  • Games: every game since 1999, with a win-probability chart, the biggest swings and each 4th-down call.
  • 4th-down calculator: type in a situation and get the model’s call and how sure it is.
  • Kickers and punters: points and yards over what the season’s average kicker or punter would manage from the same spots, in the same weather.
  • Comebacks, collapses and upsets: every game ranked by how far the winner fell behind, how big an underdog it was, and how wild the ride was.
  • A page for every player since 1999 (season tables and game logs), every team and every head coach.

How the numbers work

  • Win probability comes from our own model, trained on past NFL play-by-play through the 2025 season. It handles overtime rules and end-of-game situations (kneel-downs, last-second kicks) explicitly. On the game charts, nflfastR’s win probability is drawn dashed for comparison.
  • 4th-down calls. For going for it, kicking a field goal and punting, the model estimates what can happen: the chance of converting, the kicker’s chance of making it (rated as of that game), and where a punt leaves the other team. It then works out the win probability after each outcome. The call is the option with the best average win probability over 101 versions of the model, each fitted to a different resample of past games.
  • Confidence is the share of those 101 versions in which the call is the best option. Confident is 83% or more, Lean is 67% to 83%, and anything lower is a Toss-up.
  • Coach ratings add up the win probability given up when a coach’s call differs from the model’s, and compare coaches who faced similar situations.
  • How sure are we? One season is a small sample: a QB’s EPA per dropback needs about 400 dropbacks before his own record outweighs what similar QBs usually do, and a kicker’s record needs years. The Stabilised switch on the leaderboards pulls each rate toward what similar players usually do by that much, shows a 90% range for it, and a range of ranks it could plausibly hold. On seasons the method never saw (2018–2025), stabilised rates predicted the rest of a player’s season 25–56% better than the raw ones.
  • EPA (expected points added) is nflfastR’s. WPA (win probability added) is the change in our win probability on a play, credited to the passer, the rusher and the target.
  • Updates: the pages refresh the morning after each game day, once the new games’ data is published.

Data and credits

  • Play-by-play, schedules, rosters and player stats: nflverse (nflfastR and nflreadr), licensed CC BY 4.0. Expected points and nflfastR’s win probability come from nflfastR. Thank you to the nflverse maintainers.
  • Who was on the field (2016 on) and play charting (2022 on): NFL Next Gen Stats via nflverse (to 2022) and FTN Data via nflverse (2023 on), licensed CC BY-SA 4.0. The tables built from them (playing time, pressure and play action, catchable balls and drops, team tendencies) are shared under CC BY-SA 4.0 too, and say so where they appear.
  • All of our models are trained and tested on nflverse data only.
  • This site isn’t affiliated with or endorsed by the NFL or any team. Team names and abbreviations are used only to identify teams.

Privacy

  • No ads, no cookies, no analytics or tracking scripts.
  • The 4th-down calculator remembers your last inputs in your own browser only. The situations you send are used to compute the answer and aren’t stored.
  • Like most web servers, ours keeps a standard access log (the address a request came from and the page asked for), used only to keep the site running.