A solo home run in the second inning of an 8–1 laugher and a solo home run to walk it off in the bottom of the ninth are, to most hitting metrics, the identical event: one run, one big swing, the same dent in wOBA. To everyone actually watching the game, they could not be more different. One was housekeeping. The other was the whole season compressed into a single pitch. Win Probability Added — WPA — is the stat built to tell those two homers apart.
WPA, and its companion Leverage Index, measure something the context-neutral stats deliberately ignore: the drama. They answer not “how good was this play in a vacuum?” but “how much did this play actually swing this game?” That makes them wonderful for storytelling and honest accounting of what happened — and, as we’ll see, nearly useless for predicting what happens next. Knowing which job each metric is for is the whole game.
What WPA actually measures
WPA is built directly on top of a win-probability model. As we covered in how win probability models work, every game state — the inning, the score, the outs, the runners on base — maps to a probability that each team eventually wins, estimated from decades of historical games in similar spots. Before a pitch, the home team might have a 62 percent chance to win. The batter doubles home a run; now they have a 78 percent chance. That swing of sixteen percentage points — 0.16 of a win — is the WPA of the play, credited to the hitter and debited from the pitcher.
Add up every play in a game and the WPA always sums to a clean ±1.0: the winning team gains exactly one win of probability over the course of nine innings, distributed across its players, and the losing team sheds exactly one. The hitter who delivers the decisive blow banks a fat positive number; the reliever who coughs up the lead in a tie game can post a deeply negative one. Sum WPA across a season and you get a running tally of how much each player added to or subtracted from his team’s actual win column, moment by moment, exactly as it unfolded.
Context-dependent, by design
This is the line that separates WPA from almost every other value stat. wOBA and FIP are context-neutral on purpose — they strip out the situation so that a double is a double whether the bases are empty or loaded, whether it’s April or October. That neutrality is a feature: it isolates the player’s own contribution from the luck of when his hits happened to fall.
WPA does the exact opposite. It is context-dependent to its core. The same double is worth wildly different WPA depending on when it lands. With the bases empty in a blowout, a double barely moves the win probability and earns a sliver of WPA. With two on in a tie game in the ninth, the same double can swing the win probability by thirty or forty points and earn a massive one. WPA doesn’t care how hard the ball was hit or whether the hitter “deserved” it; it cares only about how much the scoreboard’s win-odds moved. That is its entire point and, as we’ll see, its entire limitation.
Leverage Index: how much a moment can swing
If WPA measures how much the win probability did move on a given play, Leverage Index measures how much it could move — the stakes of the situation before the pitch is even thrown. LI is scaled so that an average situation equals 1.0. A bases-loaded, two-out, one-run-game situation in the ninth might carry an LI of 3 or 4 or higher — several times the swing potential of a typical moment. A leadoff at-bat in the fourth inning of a blowout might sit at 0.1, a situation where almost nothing that happens will change who wins.
Leverage is what makes WPA legible. A reliever who enters a tie game with the bases loaded in the ninth is working in a furnace; a reliever protecting a six-run lead in a finished game is not, even though the box score may credit both with an appearance. To capture this, analysts use gmLI — the average Leverage Index a reliever faces at the moment he enters the game — as a clean readout of how a manager actually deploys his bullpen. A high gmLI marks the genuine fireman, the arm summoned for the biggest spots. A low gmLI marks the mop-up man, regardless of how shiny his ERA looks. Leverage, in other words, is the context that tells you whether a reliever’s innings were the ones that decided games.
Clutch, and the trouble with it
Put WPA and Leverage Index together and you can build a “clutch” measure: did a player perform better or worse than his own baseline specifically when the leverage was high? The standard construction compares a hitter’s context-dependent production to what his context-neutral rate would have produced in those same spots. Outperform your own norm in high-leverage moments and you grade out as clutch; wilt and you grade out as a choker.
Two honest cautions come attached. First, clutch as measured this way is descriptive — it tells you what already happened, not a stable trait a player carries from year to year. The research consensus is that year-to-year clutch performance is mostly noise; this season’s clutch hero is not meaningfully more likely to be next season’s. Second, “clutch” is defined relative to the player’s own level, so a superstar who is merely excellent in the clutch can post a negative clutch score while still being far more valuable in big spots than an average hitter having a career October. The metric measures deviation from self, not raw quality. Both facts are easy to forget the moment a narrative wants a hero.
A clearly-illustrative example
Suppose two hitters each finish a season with an identical, excellent wOBA — they produced the same quality and quantity of offense by the context-neutral books. Hitter A piled up his damage in the late innings of tight one-run games: three-run doubles in the eighth, a walk-off here and there, hits that arrived when the Leverage Index was screaming. Hitter B did all his work early, in games already decided one way or the other, his big swings landing when the win probability had nowhere to move.
Their wOBA is identical. Their WPA is not remotely close — Hitter A’s could be several wins of added probability while Hitter B’s hovers near zero. Has Hitter A proven he is a clutch performer who will keep doing this? No. He has proven that this season his production happened to fall in the highest-leverage moments, which is genuinely worth more to his team’s actual win total this year. Whether the timing repeats is a separate question, and the honest answer is: probably not in any reliable way. (Both hitters here are hypothetical, chosen to isolate the one variable — when the production happened.)
What WPA is for — and what it isn’t
Use WPA as a storytelling and after-the-fact value metric. It is unmatched at answering “who actually won that game?” A WPA game graph — the jagged line tracing win probability pitch by pitch — is the single best one-picture summary of how a game felt, and a season-long WPA leaderboard is a fair record of who delivered when the lights were brightest.
Do not use WPA to project the future. Because it is soaked in the timing and sequencing luck that context-neutral stats are specifically built to remove, WPA is a poor forecaster of a player’s underlying skill. The hits that happened to land in high-leverage spots this year carry no promise of doing so next year. When you want to know how good a player is — the input to a projection — reach for the context-neutral stats. When you want to know how much a player’s performance moved this season’s games, reach for WPA. They are different questions, and the cardinal sin is using the answer to one as the answer to the other.
The bottom line
WPA and Leverage Index are the metrics of drama. WPA credits players for how much each play swung the win probability, summing to a tidy one win per game; Leverage Index scales the stakes of every moment around an average of 1.0 and tells you whether a reliever’s innings were the ones that mattered. Together they let you say, with numbers, that the ninth-inning walk-off was worth ten times the second-inning solo shot — something wOBA will never concede. Just hold the line on what they’re for. They are a beautiful record of what happened and a treacherous guide to what comes next. The walk-off was real. It was also, in all likelihood, not a promise.
Sources & Further Reading
- FanGraphs — WPA, Leverage Index, gmLI, and Clutch leaderboards and live win-probability graphs.
- FanGraphs Library — definitions of WPA, Leverage Index, and the Clutch statistic.
- Baseball-Reference — Win Probability Added and championship-leverage tools.