Football Analytics

Win Probability

Every game state in football has a historical win rate attached to it. Win probability models track that number in real time, showing you which team is ahead and by how much, play by play.

How Win Probability Models Work

A win probability model looks at the current state of a football game and estimates the likelihood of each team winning based on historical data. The inputs are straightforward: score differential, time remaining, which team has the ball, field position, down and distance, and timeouts remaining. Using decades of NFL play-by-play data, the model calculates how often teams in that exact situation went on to win the game.

At kickoff, both teams start at roughly 50 percent, with a small advantage for the home team. As the game progresses, the win probability shifts with every play. A touchdown swings it toward the scoring team. A turnover swings it toward the defense. Late in the game, the swings become larger because there is less time to recover from negative plays. A fumble in the first quarter might shift win probability by 5 to 8 percentage points. The same fumble with two minutes left in a tie game might shift it by 30 or more.

Most public win probability models assume both teams are equally talented and play at average levels going forward. This is a simplification. The model does not know that one team has Patrick Mahomes and the other has a backup quarterback. Some advanced models do incorporate pre-game talent ratings, but the standard models rely purely on game state. This means they are best used for understanding leverage and context rather than as precise predictions of the final outcome.

Win Probability Added (WPA)

WPA measures how much a single play changed the win probability. If a team's win probability was 55 percent before a play and 72 percent after, that play generated 17 percentage points of WPA. This is an incredibly intuitive way to identify the most impactful plays in a game. The plays with the highest WPA are the ones that most dramatically shifted the outcome, and they usually match what you felt watching the game in real time.

WPA can be accumulated over a game or a season to measure a player's total impact on winning. A quarterback's cumulative WPA tells you how many percentage points of win probability he added through his play. This naturally weights clutch performance more heavily than garbage time because plays in close games with time running short have much higher leverage. A 10-yard completion in the fourth quarter of a one-score game is worth far more WPA than a 10-yard completion in the first quarter of a blowout.

The limitation of WPA is that it is heavily context-dependent. A mediocre quarterback on a bad team will rarely have high-leverage opportunities because his team is often trailing by too much for any single play to matter. A good quarterback on a competitive team will face more high-leverage situations and have more chances to accumulate WPA. This makes WPA better for measuring game impact than player quality. For player evaluation, EPA-based metrics are more stable and reliable.

Leverage and Why Some Plays Matter More

Leverage index measures how much a particular game state matters relative to the average game state. A 1.0 leverage index is average. A 3.0 means the play is three times as important as a normal play in terms of its potential to swing the game's outcome. Leverage spikes in close games late in the fourth quarter, on critical third downs, and on red zone plays where a score would significantly alter the win probability.

Understanding leverage helps explain why certain plays feel so much bigger when you are watching. A third-and-goal conversion in the fourth quarter of a one-point game carries enormous leverage because the outcome of that single play might determine the winner. The same conversion in the second quarter of a 21-point blowout carries almost no leverage because the game is effectively decided regardless. Fans instinctively feel this difference. Win probability models quantify it.

Coaches can also use leverage concepts in decision-making. Going for it on 4th and 1 near midfield in a tie game is a high-leverage decision because the swing between converting and turning it over is enormous. Punting in that situation reduces the variance but also caps the upside. Analytics-minded coaches use win probability models to calculate whether the expected gain from going for it exceeds the expected cost of failing. The math does not always agree with traditional coaching instincts, which is part of what makes the intersection of analytics and coaching so fascinating.

Model Inputs

Score differential, time remaining, field position, down and distance, timeouts remaining, and whether a team has the ball. Every game state produces a probability.

WPA (Win Prob Added)

The change in win probability caused by a single play. A pick-six might swing win probability by 40 points. A 3-yard run in the first quarter might move it by 1 point.

Leverage Index

How much a play matters relative to the average play. Late in close games, leverage is sky-high. In blowouts, leverage is near zero. WPA captures this naturally.

Every play has a price.

Win probability puts a number on what your gut already knows. Some plays change everything. Most plays barely move the needle. WPA tells you which is which.

Go beyond the box score.