Win probability is a model's estimate, at any moment in a game, of the chance the team with the ball ends up winning. It comes from score, time remaining, down, distance, field position, timeouts and, in the better versions, the pregame point spread. Win probability added (WPA) is the change on a single play, which is how the biggest plays of a game are identified and how fourth-down decisions are graded.
Updated Sept. 16, 2026.
Week 1's biggest swing came in New Orleans Saints at Detroit Lions: one snap moved the New Orleans Saints' win probability by 57 points (OT, 01:38: T.Shough pass short right to N.Fant for 8 yards, TOUCHDOWN.). That is the 174th-largest swing on any play since 1999, out of 1,078,012.
Swing is the size of the change in win probability from before the snap to after the play, in percentage points.
| Game | Play | Swing |
|---|---|---|
| New Orleans Saints at Detroit Lions | OT 01:38: T.Shough pass short right to N.Fant for 8 yards, TOUCHDOWN. | 57 pts |
| New England Patriots at Seattle Seahawks | Q4 00:26: D.Maye pass deep right intended for M.Hollins INTERCEPTED by J.Jobe [D.Lawrence] at SEA -3. Touchback. | 42 pts |
| New Orleans Saints at Detroit Lions | OT 06:19: J.Goff pass deep left to J.Williams ran ob at NO 5 for 22 yards. | 35 pts |
| New England Patriots at Seattle Seahawks | Q4 11:38: D.Lock pass short left to J.Smith-Njigba for 45 yards, TOUCHDOWN. | 34 pts |
| Buffalo Bills at Houston Texans | Q4 01:41: J.Allen pass deep middle to J.Palmer for 34 yards, TOUCHDOWN. | 32 pts |
| New Orleans Saints at Detroit Lions | OT 03:13: T.Shough pass deep right to C.Olave to NO 47 for 20 yards (A.Maddox). | 32 pts |
| New Orleans Saints at Detroit Lions | OT 01:34: TWO-POINT CONVERSION ATTEMPT. T.Shough pass to B.Lance is incomplete. ATTEMPT FAILS. | 30 pts |
| Buffalo Bills at Houston Texans | Q4 00:16: C.Stroud sacked at BUF 28 for -10 yards (G.Rousseau). FUMBLES (G.Rousseau) [G.Rousseau], RECOVERED by BUF-T.Bernard at BUF 24. | 30 pts |
| Atlanta Falcons at Pittsburgh Steelers | Q4 14:16: C.Rush pass short right intended for Bi.Robinson INTERCEPTED by T.Watt at ATL 35. T.Watt for 35 yards, TOUCHDOWN. | 24 pts |
| New Orleans Saints at Detroit Lions | OT 01:53: T.Shough pass deep right to J.Johnson pushed ob at DET 8 for 27 yards (T.Harper). | 22 pts |
Every fan carries a rough version in their head: up 10 with four minutes left is safe, down 3 with 90 seconds and no timeouts is grim. Win probability puts a number on it, built from every game since 1999. From your own 25, first-and-10, tied, with eight minutes left, a team wins about 50 percent of the time; add a 7-point lead and it is near 85 percent.
The model behind the public numbers (nflfastR's) includes the pregame spread, which matters early: a 10-point favorite trailing by 3 in the first quarter is still the likelier winner. As the clock runs down the spread fades and the scoreboard takes over.
WPA is the play-level version. A game-winning touchdown with no time left is +50 points of WPA or more; a first down on third-and-2 in the second quarter is +2. Summing a player's WPA over a season gives a clutch-weighted version of value, and the fourth-down decision charts that coaches now use are win-probability comparisons of going for it, kicking and punting.
In the 2022 season the Vikings trailed the Colts 33-0 at halftime with a win probability under 1 percent and won in overtime, the largest comeback in NFL history. Every play of the second half was a big WPA play for Minnesota; the winning field goal alone was worth about +40 points.
Win probability is the output of a model (nflfastR uses gradient-boosted trees) fit on historical plays, with inputs of score differential, seconds remaining, down, yards to go, yard line, timeouts, and the pregame spread. WPA = win probability after the play minus win probability before it.
Well calibrated over large samples: situations the model calls 70 percent are won about 70 percent of the time. It is not a prediction of any single game, and it says nothing about who is playing well, only about the scoreboard and clock. That is why a team can be outplayed and still sit at 80 percent.
Win probability added: the change in win probability on one play. It is the stat behind 'biggest play of the game' and the basis for grading fourth-down decisions, where the coach's choice is compared to the win probability of the alternatives.
Because the teams are not equal. Early in a game the score tells you little and the spread tells you a lot; a 14-point favorite down 7 in the first quarter is still the favorite. Models without the spread treat every game as a coin flip at kickoff.
Category: Stats & Analytics. Part of the StickToTheModel NFL Encyclopedia.