The home run in the ninth gets the headline, the photograph and the highlight package. It is also, quite often, the least interesting thing that happened. By the ninth the game may already have been decided, and the moment it was decided may have been a five-pitch at-bat in the sixth that ended in a groundout and produced no reaction at all.
There is a straightforward way to find that moment, and it does not require a subscription or a spreadsheet. It requires one concept, understood properly, and the discipline to look at a chart before deciding what a game was about.
The concept is leverage, and the discipline is refusing to read a chart backwards from the result. Both are learnable in an afternoon, and together they change what a recap can honestly claim.
What you need before you start
- The state, not the play. Inning, score, outs and runners. Everything follows from those four things.
- Leverage. A measure of how much is riding on the next at-bat, scaled so that an average situation sits at one.
- Win probability added. How much a play actually moved the chance of winning, which is a different question.
- The chart. A line showing win probability through the game. Useful for locating moments, dangerous for explaining them.
- The trap. Reading a chart with the result already known, which makes every earlier moment look inevitable.
- The output. One sentence you can defend, naming a moment and what it changed.
Why the headline play is usually not the turning point
A turning point is not the biggest event. It is the point at which the range of plausible outcomes narrows sharply, and those two things come apart more often than not.
A three-run home run that turns a four-run lead into a seven-run lead is a large event and a small turning point, because the game was already heavily decided in one direction. A one-out single that scores a runner from second in a tied game is a smaller event on the scoreboard and a much larger one in terms of what it settles.
Broadcast coverage is built around events rather than states, for perfectly good reasons: events are visible and states are not. The result is that the play everybody remembers and the play that decided the game are frequently different plays, and the gap between them is where most of the argument about a memorable postseason game eventually lands.
Leverage index, defined without the formula
Leverage index answers one question: how much could the next play swing this game, compared with an average moment.
The scale is built so that an average situation equals one. A moment with a value of two carries roughly twice the potential swing of an average moment. A moment with a value below one half is a situation where almost nothing that happens will change the outcome much. Nothing in the concept requires you to know how it is calculated; what matters is that the scale is anchored and comparable.
Two properties make it useful. It is known before the play happens, which means it measures the situation rather than the outcome, and it applies equally to both teams, because the potential swing is symmetric. By convention, values above roughly one and a half are described as high leverage and values above two as very high, though those labels are conventions rather than thresholds with any special meaning.
The three inputs: inning, score and base-out state
Leverage rises when a single play can produce a large change, and three things drive that.
The first is how late it is. Early in a game there is time for anything to be undone, so even a big swing narrows the outcome only a little. The same swing in the eighth narrows it a great deal. The second is the margin. Leverage is highest when the score is close, falls quickly as a lead grows, and approaches nothing once a lead becomes large enough that ordinary play cannot recover it.
The third is the base-out state, which is the input casual readers underrate. Runners on second and third with one out is a situation where a single play can produce two runs or none. Bases empty with two out is a situation where the next play can produce almost nothing. Those two states, in the same inning of the same game, carry completely different weights.
What a high-leverage at-bat looks like from the seat
You can identify these without a chart, and doing so is the fastest way to build the instinct. The signals are behavioural rather than numerical.
Both bullpens are active. The infield moves, either in to cut off a run or back to trade an out for a run. The manager visits the mound, or a pinch hitter appears who was clearly being saved. The catcher changes his sequence, and pitch selection narrows to whatever the pitcher trusts most.
The crowd is the least reliable signal and the loudest. Noise tracks events, and an enormous reaction to a routine fly ball in a decided game tells you nothing about leverage. Watching the fielders is better; they are the ones being repositioned according to how much the situation is worth.
The mirror image is just as informative. A low-leverage moment looks like a manager leaving a struggling pitcher in, an outfield playing at standard depth, and nobody warming. Those are the innings a broadcast fills with anecdotes, and recognising them stops you attaching significance to a stretch where the game had already stopped moving.
Win probability added, and how it differs
The two concepts are frequently confused and they answer different questions. Leverage is measured before the play and describes the situation. Win probability added is measured after and describes the result.
The relationship between them is simple to state. A play in a high-leverage situation has the capacity for a large win probability change; a play in a low-leverage situation does not. A single in the second inning of a tied game and a single in the ninth with two on may be identical acts and carry very different consequences, and the difference is entirely the situation they happened in.
This is why credit assignment using either measure alone goes wrong. Leverage credits situations, not performance. Win probability added credits outcomes, including outcomes that came from a fielder’s error, a bad bounce or a pitch that was hit hard directly at somebody. Neither is a measure of how well anybody played, and reading them as one is the most common misuse in this area of analysis.
Reading the chart without over-reading it
A win probability chart is a line that moves with every play. It is extremely good at one job, which is showing where the game changed shape, and it invites two errors that are worth naming.
The first is hindsight. Looking at a chart after the fact, the eventual winner’s line appears to be heading toward the result the whole time, and the moments where it dipped look like drama rather than genuine uncertainty. The correction is to cover the right-hand side of the chart with a hand and read it left to right, which restores the uncertainty the participants actually had.
The second is treating the model as a description of the specific teams playing. These charts are built from historical base-out-inning-score states across many games, so they describe an average team facing an average team. They do not know who is pitching, who is available in the bullpen, or that one lineup has been unable to hit anything for a week.
Finding the decisive at-bat in five steps
- Find the largest single move on the chart. Note the inning, the state and who was involved. This is a candidate, not an answer.
- Find the last time the line crossed the halfway mark. After that point the game had a favourite that never changed, which is a different and often better definition of decisive.
- List every at-bat with a high leverage value. These are the moments where the game could have gone differently, whether or not they did.
- Compare the two lists. A play that appears on both, a high-leverage situation that produced a large swing, is your turning point in almost every game.
- Check what the play actually was. If the swing came from an error or an unusual bounce, describe it that way rather than crediting a hitter for it.
Step two is the one that most often disagrees with the highlight package, and it is usually the one worth writing about. It also has the advantage of being unambiguous: there is exactly one last crossing in any game, so two people applying the method to the same chart will arrive at the same moment.
One refinement is worth adding once the basic pass is comfortable. Games occasionally have two candidates rather than one, where the line crosses the middle late and then moves decisively a second time. In those cases the honest description names both, because a single turning point is a convention imposed on the game rather than something the game always contains.
A worked example, built from scratch
Take a generic shape rather than a real game, so nothing here depends on remembering anything. Tie score, bottom of the seventh, runners on first and second, one out, and a hitter at the plate against a reliever who has just entered.
The leverage here is high and it is easy to see why. A double scores two and probably settles the game. A home run settles it outright. A double play ends the inning and hands the momentum of the situation back. The range of outcomes from one at-bat is unusually wide, which is what leverage measures.
Suppose the hitter grounds into a double play. Nothing appears on the scoreboard, nothing appears in a highlight reel, and the chart shows a substantial move toward the other team. If the game then ends with a solo home run in the ninth, the recap will be about the home run. The at-bat that decided it was the ground ball, because that was the moment the range of outcomes narrowed and never widened again.
What leverage cannot tell you
| Question | Leverage | Win probability added | What you actually need |
|---|---|---|---|
| Where did the game turn? | Identifies candidates | Identifies the biggest moves | Both, compared |
| Who played well? | No | No | Performance measures and video |
| Was a decision correct? | Partly, before the fact | No, it knows the result | The options available at the time |
| Did a team get lucky? | No | Hints at it | Contact quality and sequencing |
| Is a reliever clutch? | Shows where he was used | Accumulates over seasons | Far more games than one |
| Was the manager right to lift a pitcher? | Shows what was at stake | No | The alternatives and the state at the time |
The second row is the important one. Neither measure evaluates a player, and every argument that uses them to do so is borrowing authority the numbers do not have. That borrowing is behind a large share of the arguments that circulate after a big game.
Writing the sentence honestly
The claim a recap can support is narrower and more interesting than the one it usually makes. It is not that a particular player won the game. It is that at a particular moment the game stopped being genuinely uncertain, and here is what happened at that moment.
Three phrasings work. Naming the state: with two on and one out in a tied seventh, the game was still open, and it was not open again after that at-bat. Naming the swing: that single move accounted for more of the change in the result than anything else in the game. Naming the counterfactual carefully: had that ball found a gap, the chart says the game was close to even.
Three phrasings do not work. That he single-handedly won it, which no measure supports. That the ninth-inning home run was the turning point, when the chart shows the game was already settled. And that a hitter failed in the clutch, which converts a probabilistic situation into a character judgement and is the point at which this kind of analysis stops being analysis. The same discipline is what separates useful work from decoration across the whole field of performance measurement.
Frequently Asked Questions
Do I need the actual numbers to do this?
It helps and it is not essential. The inputs to leverage are inning, score, outs and runners, all of which are visible. With practice you can rank the situations in a game accurately from the seat, and the chart then serves as a check rather than a source.
Does leverage apply to sports other than baseball?
The idea does, and the implementation is harder. Baseball’s discrete states make it straightforward to define what a situation is worth. Continuous sports need possession models rather than base-out tables, and the equivalents exist but are noisier and less standardised.
Why do two sources give different win probabilities for the same moment?
Because they are built from different historical samples and make different adjustments, including for the teams and pitchers involved. A gap of a few percentage points between sources is normal and is not evidence that either is wrong.
Is a player who performs in high-leverage situations genuinely clutch?
The evidence on whether that is a stable, repeatable trait is weak, and the samples involved are small even over full careers. What is definitely true is that some players are used in those situations more than others, and usage is a managerial decision rather than a property of the player.
Can the decisive moment be a defensive play?
Frequently, and those are the moments this method is best at recovering. A diving stop with two runners on removes a large swing that would otherwise have happened, and because nothing changes on the scoreboard, it is exactly the kind of play a conventional recap loses.
Next time a postseason game ends, find the chart before you read anybody’s account of it. Cover the right-hand side, read left to right, and mark the last point at which the line crossed the middle. That is the game, and it will usually not be the play on the front page.
