Methodology

Leverage

Not all possessions in an NBA game carry equal weight. A possession in a tie game with 30 seconds left has far more impact on the outcome than a possession in the first quarter of a blowout. Leverage quantifies this by measuring how much a single possession could swing the win probability of the game.

How leverage is calculated

Leverage is built directly on top of our win probability model. For each possession, we ask: "how much could win probability change depending on what happens here?"

To answer this, we consider the five possible scoring outcomes of any possession (0, 1, 2, 3, or 4 points) weighted by how often each actually occurs in the NBA. These frequencies are derived from our six seasons of training data. For example, roughly 59% of possessions result in 0 points (no score), about 20% result in a 2-point field goal, 9% in a 3-pointer, and so on.

For each of these hypothetical outcomes, we compute what the win probability would be after the possession ends (accounting for the new score and slightly less time remaining). If scoring a 3-pointer would push win probability to 85% but failing to score would drop it to 55%, that's a gap of 30% between the best and worst outcomes, and the possession clearly matters a lot. If the win probability barely changes regardless of the outcome (say, moving between 98% and 99% in a blowout), the possession doesn't carry much significance. The leverage score captures this difference between the possible outcomes.

Our leverage numbers aren't displayed in the UI but they play an important role behind the scenes in determining which plays qualify as high leverage.

The leverage scale

Leverage is normalised so that the average possession across all games has a leverage score of approximately 1.0. This makes the numbers intuitive:

A leverage score of 2.0 means the possession is twice as consequential as an average possession. A score of 0.5 means it's half as consequential. In practice, leverage scores typically range from near 0 (deep into a blowout) up to around 5 or 6 (a tie game in the final seconds).

What drives leverage up

Two factors push leverage higher:

1. A close score

When the game is tight, any scoring play can meaningfully shift win probability. In a blowout, even a 3-pointer barely moves the needle because the trailing team's win probability is already near zero.

2. Less time remaining

Late in a game, each possession represents a larger proportion of the remaining opportunities. A 3-point swing with 30 seconds left is far more impactful than the same swing with 10 minutes remaining, because there's less time for the other team to respond.

These two factors compound. A tie game with 5 minutes left is high leverage. A tie game with 30 seconds left is extremely high leverage.

High leverage vs clutch

High leverage and clutch are related but distinct concepts. Clutch is a binary label where a possession either is or isn't clutch based on fixed time and margin thresholds (final 5 minutes, margin within 5 points). Leverage is a continuous score that captures the full spectrum of game importance.

Clutch possessions are almost always high leverage, but high leverage possessions aren't necessarily clutch. A tie game with 8 minutes left in the fourth quarter is very high leverage but not clutch (it's outside the 5-minute window). Leverage captures these important moments that fall outside the clutch definition.

This distinction between clutch and high leverage plays is precisely why we offer both as separate stat categories. They answer different questions.

Attribution

Our leverage implementation is based on the methodology described by Darryl Blackport of PBPStats.com and is adapted for our win probability model and possession data.

Up next: High leverage