Live win probability
Saturday, June 13
Nothing is live right now. Here is how the win chance moved in the latest finished games, with each game's biggest swing.
Latest finals
- NYK 94SAS 90FinalNYK 100%0% SAS
Biggest swing: Knicks 98% after the 6-2 run in Q4 (from 49%), Q4 0:08.
Each chart runs from tip-off (left) to the final buzzer (right). Above the middle line the home team is favored, below it the road team. The shaded band is the biggest swing.
How this works
The method
At any moment, the home team's win chance comes from a logistic regression with three ingredients: the score margin, the time left, and the pregame win chance from our win probability model. The margin counts more as the clock runs down (we divide it by the square root of the minutes left), and the pregame edge fades as the game goes on. Overtime runs on its own five-minute clock. When time runs out, a lead means 100%.
We fit it on 1,322 games from 2025-26 (regular season and playoffs), using the score at every basket plus a check every 30 seconds of game clock, so long scoring droughts count too (258,820 moments in all). The biggest swing is the largest change in win chance inside any three minutes of game time, and the biggest single play is the basket that moved it most.
How we tested it
We held out a random 264 games (20%), fit on the rest, and scored every moment of the held-out games (51,751 moments). Log loss was 0.453 and the Brier score 0.150 (lower is better for both). A version that only knows the score and the clock scored 0.487 and 0.166, so the pregame number helps, most of all early in games. The side the model favored went on to win 78% of the time.
| Quarter | Moments | Log loss | Score and clock only | Brier | Favorite won |
|---|---|---|---|---|---|
| Q1 | 12,862 | 0.570 | 0.644 | 0.194 | 70% |
| Q2 | 12,891 | 0.521 | 0.570 | 0.174 | 75% |
| Q3 | 13,057 | 0.443 | 0.459 | 0.145 | 79% |
| Q4 | 12,689 | 0.270 | 0.269 | 0.086 | 88% |
| OT | 252 | 0.628 | 0.574 | 0.206 | 68% |
Late in games the pregame number no longer adds anything, and in overtime (252 moments from a handful of games) the simpler version did a bit better. There are too few overtime games to read much into that.
Does 70% mean 70%?
| Home win chance | Moments | Average forecast | Home team won |
|---|---|---|---|
| 0 to 10% | 6,626 | 3.0% | 5.1% |
| 10 to 20% | 3,496 | 15.0% | 18.8% |
| 20 to 30% | 3,632 | 25.0% | 24.7% |
| 30 to 40% | 3,830 | 35.0% | 28.4% |
| 40 to 50% | 4,105 | 45.0% | 37.3% |
| 50 to 60% | 4,432 | 55.1% | 48.5% |
| 60 to 70% | 4,948 | 65.2% | 62.5% |
| 70 to 80% | 5,409 | 74.9% | 71.7% |
| 80 to 90% | 5,465 | 85.1% | 82.9% |
| 90 to 100% | 9,808 | 96.6% | 96.6% |
Moments from the same game move together, so 264 test games is a small sample. 5 of 10 buckets landed within 3 points of the forecast. The biggest miss was the 40 to 50% bucket, where the home team won 7.7 points less often than forecast.
What it can't see
- Who has the ball, fouls, timeouts, the bonus or who is on the floor. Up 2 with 10 seconds left reads the same whoever has possession.
- Injuries or ejections during the game, beyond what the score shows.
- In the last seconds it is cautious: a 2-point lead with a second left shows about 89%, when the real chance is higher. The fit has few moments that close to the buzzer.
- Finished games use ESPN play-by-play timing, which can be a few seconds off, and a few games miss some late baskets (the chart then jumps to the final).
- Live charts add a point only when the score or the quarter changes, from a feed that updates about every 15 seconds.