NFL Win Probability Added: 3 Steps to Calculate a Play’s WPA
Win probability added (WPA) is the change in a team’s chance of winning caused by a single play: WPA = WP_after − WP_before. Modern win probability models read the game state, such as score, clock, field position and down, and some adjust for the pre-game spread. Tools like nflfastR compute these fields for you, and we’ll show you how to use them.
TL;DR:
- For reproducible estimates, nflfastR uses score, clock, field position, down and distance, timeouts, second half kickoff possession, and the pregame spread.
- Use vegas_wp or vegas_wpa across teams of different strength; three point favorites win about 61% of games, compared with 75% for seven point favorites.
- WPA attribution can give a quarterback and receiver full credit for the same play, so season totals should not be treated as isolated measures of skill.
- Pair WPA with EPA and success rate, report game context and sample size, and avoid assigning one player credit for outcomes produced by the whole team.
Table of Contents
- What win probability and WPA mean for NFL fans
- Which inputs and tools power NFL win probability models
- How WPA is calculated with a worked play example
- Where WPA earns its keep: leverage, players and teams
- How to compute WP and WPA from play-by-play data
- Limits of WPA and how to read it responsibly
- Why WPA is worth learning as a fan or student analyst
- Follow the swings live with Athlo
- FAQ
- Sources
What win probability and WPA mean for NFL fans
Win probability (WP) is the estimated chance a team wins from the current game state. Modern models feed on score, time left, field position, down and distance. WPA is simply how much one play moves that number.
Here’s the plain-language version:
- WP is a snapshot: “Right now, this team wins X% of the time.”
- WPA is the movement: “That play pushed the number up or down by this much.”
- Sign matters: positive helps the offense, negative helps the defense.
Brian Burke’s explainer shows how a routine play might add about +0.03, while a late turnover can swing far more. Analysts treat WPA as a narrative stat: it tells you which plays decided the game, not who has the most talent.
Which inputs and tools power NFL win probability models
A win probability model is only as good as the game state you feed it. The nflfastR documentation lists the fields its calculate_win_probability function needs:
- receive_2h_ko: who gets the second-half kickoff
- score_differential: the current margin
- half_seconds_remaining and game_seconds_remaining: the clock
- spread_line: the pre-game point spread
- down, ydstogo and yardline_100: situation and field position
- posteam_timeouts_remaining and defteam_timeouts_remaining: timeouts for each side
The spread matters because a model that ignores team strength treats every matchup as a coin flip at kickoff. Spread-adjusted versions, exposed as vegas_wp and vegas_wpa, account for expected strength. Public calculators show why: a 3-point favorite wins about 61% of games, and a 7-point favorite about 75%.
For reproducible methods, look at nflfastR itself and the nflWAR paper, which builds on nflscrapR data.
How WPA is calculated with a worked play example
The formula is WPA = WP_after − WP_before, using the offense’s perspective. Here’s an illustrative example, with made-up numbers for teaching:
- Your team trails by four with the ball on the opponent’s 30 in the fourth quarter, facing 3rd and 6. Say WP_before is 0.38.
- The QB completes a 24-yard pass to the 6-yard line for a first down. Say WP_after is 0.52.
- WPA = 0.52 − 0.38 = +0.14 for the offense, and −0.14 for the defense.
A positive WPA means the play increased the win probability of the team whose perspective the model reports; a negative value means it decreased that probability. The nflWAR paper works through a similar real example with a WPA of +0.1266.
Crediting is where implementations differ. Burke notes that full credit often goes to every player directly involved, so a QB and his receiver can both get the entire +0.14. Add up each play’s WPA and you get player, unit or team totals for a game or season.
Where WPA earns its keep: leverage, players and teams
WPA does a few jobs well:
- Flagging high-leverage plays: sort a game by absolute WPA and the swing plays rise to the top.
- Ranking player contributions: season WPA totals feed award debates and highlight reels.
- Summarizing teams and units: compare how passing and rushing plays moved the needle.
- Judging coaching calls: a fourth-down decision or clock choice shows up in the numbers.
Close games are why this matters. Historical summaries show that 3-point and 7-point margins occur in a notable share of NFL games, so plenty of games hinge on a handful of plays.
Don’t use WPA alone. Pair it with efficiency measures like EPA and success rate so team context doesn’t get mistaken for individual skill. Our football EPA explainer covers that side.
How to compute WP and WPA from play-by-play data
Here’s the short roadmap, following the nflfastR reference:
- Load play-by-play data for a season or a single game.
- Confirm the required columns from the inputs list exist, including spread_line and both timeout counts.
- Run calculate_win_probability to get wp and vegas_wp for each play.
- Group by game_id and take the difference in wp from one play to the next to get wpa.
- Filter for large absolute WPA to surface the biggest plays.
Many datasets already include these results. The nflfastR data fields include wp, vegas_wp, wpa and vegas_wpa. Check those columns first.
Pro Tip: Run both the raw and spread-adjusted versions when comparing plays across teams, and for academic-style work stick to reproducible packages and cite your method.
Limits of WPA and how to read it responsibly
The main limit: WPA is context-dependent and descriptive. The same turnover carries a very different WPA in the first quarter than in the fourth with the score tight. A high season total can also mean a player was on the field for many high-leverage snaps, not that he is “clutch.”
Credit is messy too. A receiver’s drop can lower a QB’s number even when the throw was accurate.
Use this checklist:
- Pair WPA with EPA and success rate.
- Report the game context and sample size.
- Prefer spread-adjusted models for cross-team comparisons.
- Don’t hand one player credit for an outcome the whole team produced.
Pro Tip: Before quoting a WPA total, ask how many of those plays came in close games, then say so.
Why WPA is worth learning as a fan or student analyst
WPA is the fastest way to turn “that was the whole game” into a number you can check. It trains you to think about leverage, context and model choices, which carries into every other stat. If you want the next step, our NFL advanced stats guide covers weekly edges, and our live win probability piece shows the same idea in basketball.
— Abdulazeez
Follow the swings live with Athlo
WPA is most fun when you watch the swings as they happen. Athlo gives you live scores, lineups and stats so you can follow games in real time, including on the local scene:
- Real-time scores on our live sports page
- Lineups and stats for major sports, plus live scores for youth and high-school games
Download Athlo from the App Store or Google Play and start tracking the games you care about.
FAQ
What does win probability added mean in the NFL?
Win probability added is the change in a team’s chance of winning produced by one play. You get it by subtracting the pre-play win probability from the post-play value, and a positive result helps the team with the ball.
How is NFL win probability calculated?
Models estimate it from game-state inputs such as score difference, time remaining, field position, down, distance and timeouts. Modern versions like nflfastR’s can also use the pre-game spread to account for team strength.
Is WPA a good measure of player skill?
Not on its own. Analysts describe it as a narrative statistic that shows which plays mattered, so pair it with efficiency metrics before judging talent.
What is the difference between wp and vegas_wp?
wp comes from the game state alone, while vegas_wp also factors in the pre-game spread. Use the spread-adjusted version when comparing teams of different strength.
Sources
- Win probability added
- Advanced Football Analytics (formerly Advanced NFL Stats): Win Probability Added (WPA) Explained
- nflfastR: Functions to Efficiently Access NFL Play by Play Data
