Win probability and leverage

Step 01 · The idea

Win probability

At any moment of a game you can ask the question every fan asks with five minutes left: of all the teams that have ever been in exactly this spot, with this score, this much time left and this many skaters on each side, how many went on to win? That share is the team’s win probability. The model answers it from 2.2 million moments sampled every 15 seconds across sixteen seasons of NHL games.

The chart shows the home team’s chance through regulation for each score. A tie starts a little above 50% (home ice) and drifts toward a coin flip; a one-goal lead is worth about 77% at the start of the third and 96% with a minute left.

Being a skater up or down shifts these curves by a fixed fraction of a goal for as long as it lasts (in expected net goals):

Step 02 · Win Probability Added

What one goal did

Every event moves those odds a little or a lot, and the move is its Win Probability Added, or WPA. Here is the Oilers’ 1–0 home loss to the New York Islanders on January 15, 2026. At puck drop the Oilers’ chance was 53.4%. Nothing changed on the scoreboard until Anthony Duclair scored with 6 minutes and 18 seconds left in the 3rd period: the Oilers’ chance fell from 51.3% to 15.3%, a swing of 35.9 WP points. That goal’s WPA was +35.9 for the Islanders and −35.9 for the Oilers.

Add up the WPA of every event in a game, plus the slow drift of the clock, and you get exactly where the game ended (100% or 0%) minus where it started. WPA is the cleanest way to put a number on how much a moment mattered. Its unit is the percentage point of win probability, written WP points on this site: one hundred of them is one whole win.

Step 03 · Expected WPA

Pricing chances instead of goals

Goals are rare. That game had 1, so judging players by the WPA of goals alone would leave almost every shift at zero and hand the whole night to whoever happened to be on the ice for the bounce. Net goals uses expected WPA instead: every shot is worth its chance of scoring (expected goals) times the swing a goal would have caused at that moment.

Summed over the game, the Oilers’ 59 unblocked attempts (5.25 expected goals) were worth +121.6 WP points of expected WPA and the Islanders’ 28 attempts (2.04 expected goals) +45.4. The Oilers won the chance battle and lost the game, and net goals sides with the chances.

This is the raw material of the four on-ice components. EV offense takes it one step further: it compares each skater’s expected WPA with what an average lineup would have produced in the same seconds, and credits only the difference.

Step 04 · Leverage

When a goal matters most

Leverage is the size of that swing: how much a goal right now would move the game. It’s high when the game is close and late, low in a blowout or early on. You can feel it in the building before you see it in a table. This table gives the swing of one goal, in WP points, for the home team at even strength, with the leverage index beside it: the swing divided by the average shot’s swing in 2025-26 (17.2 WP points), so 1.0× is an ordinary moment.

A tie with a minute left is a 2.6× moment: a goal there is worth more than two and a half ordinary goals. Two goals up with two minutes left, the same goal is worth almost nothing. Because every shot is priced by its swing, a player who does his best work in close, late games is credited for it, and nobody can pad a rating in garbage time.

Step 05 · In net goals

Where these ideas show up

Each component page works one real season through from the raw events. Here is where win probability enters each of them:

One last conversion turns WP points into goals. A regression of team standings points on their players’ summed on-ice value says one win of on-ice expected WPA is worth about 1.28 goals to the player who shared it (five skaters share every second, so it is far below the six goals a whole win is worth). Step 6 of the EV offense page works through it.