Glossary

A team playing on consecutive days. Tired teams win a little less often, so our pregame win probability takes it into account.

See also: Pregame win probability

A shot attempt stopped by a skater before it reached the net. In a box score, BLK is the number of shots a player blocked.

See also: Unblocked shot attempt

Whether a model’s chances come true as often as it says. A well-calibrated model’s 60% calls happen about 60% of the time, its 20% calls about 20% of the time, and so on.

Example. Take 500 games where we gave the home team between 55% and 65%. If the home team won about 300 of them (60%), the model is calibrated there.

See also: Calibration slope, Reliability chart, Log loss. xG calibration →

One number for how far a model’s chances should be stretched or squeezed to match what happened. 1 is ideal. Under 1, the model was too sure of itself; over 1, not sure enough.

See also: Calibration. Win probability model →

Which goalie is credited with the win or loss: W, L or OT (an overtime or shootout loss). Only one goalie per team gets one.

The chance that a shot attempt becomes a goal, judged from where and how it was taken. Add up a team’s xG over a game and you get the number of goals its chances were worth, whoever was in net.

Example. A team takes three shots worth 0.05, 0.12 and 0.30 xG. Together they are worth 0.47 goals: about one goal every two games from chances like those.

See also: Unblocked shot attempt, Shot map, Calibration. How our xG model works →

Goals for minus goals against.

Example. A team that has scored 48 and allowed 41 has a goal differential of +7.

See also: Goals for and against

Goals a team has scored (GF) and allowed (GA). Shootout goals are not counted; a shootout win adds one goal to the winner’s total in the NHL standings, as the league publishes them.

See also: Goal differential

A season kept away from a model while it is built, then used to test it. The model never saw those games, so how it does on them is a fair guess at how it will do on games that have not been played.

Win probability model →

A team’s record over its last ten games, as wins, losses and overtime losses.

Example. 6-3-1 means six wins, three regulation losses and one overtime or shootout loss in the last ten.

A score for predicted chances: lower is better. Each game costs minus the natural log of the chance given to what actually happened, so a confident miss costs far more than a cautious one. We compare it with “no skill”: always predicting the usual home win rate.

Example. We gave the home team 60%. If it won, that game scores −ln(0.60) = 0.511. If it lost, it scores −ln(0.40) = 0.916. Had we said 90% and it lost, the cost would be −ln(0.10) = 2.303.

See also: Calibration. Win probability model →

A loss in overtime or a shootout. It still earns a point in the standings. In the PWHL column, OTL counts overtime and shootout losses together.

See also: Points

For a skater: +1 for each even-strength or shorthanded goal their team scores while they are on the ice, −1 for each one scored against. Power-play goals for and against do not count.

Standings points. NHL: 2 for a win, 1 for an overtime or shootout loss. PWHL: 3 for a regulation win, 2 for an overtime or shootout win, 1 for an overtime or shootout loss. In a box score, P is a skater’s goals plus assists.

Example. An NHL team at 11-4-1 has 11 × 2 + 1 = 23 points.

See also: Points percentage, Overtime loss, Regulation wins

Points earned as a share of the most a team could have earned: 2 a game in the NHL, 3 in the PWHL. It ranks teams fairly when they have played different numbers of games.

Example. An NHL team with 23 points from 16 games: 23 ÷ (16 × 2) = .719.

See also: Points

Each team’s chance to win before puck drop, overtime and shootout included. Ours comes from team ratings, home ice and rest. It is stored before the game and never changed afterwards. It knows nothing about injuries, lineups or the starting goalie, so read it as a baseline, not a pick.

Example. Home 58%, away 42% means that in many games like this one, the home team would win about 58 times in 100. The away team winning is not the model being “wrong”.

See also: Team rating, Back-to-back, Calibration. Win probability model →

Wins in 60 minutes, without overtime or a shootout. Used to break ties in the standings, and worth the full 3 points in the PWHL.

See also: Points

A picture of calibration. Predictions are grouped by size (for example, all the 40–50% calls), and each group’s average prediction is plotted against how often it came true. Points on the diagonal mean the chances came true as often as stated.

See also: Calibration. xG calibration →

Saves divided by shots on goal faced (SA).

Example. A goalie who stops 28 of 30 shots has a save percentage of 28 ÷ 30 = .933.

See also: Shots on goal

The rink diagram on each game page: every shot attempt with a recorded location, placed where the league logged it. On NHL games, goals, saves and misses are sized by their xG.

See also: Expected goals

Shots that would have gone in without a save: goals plus saves. Shots that miss the net or are blocked are not shots on goal.

See also: Unblocked shot attempt, Save percentage

A team’s current run of the same result.

Example. W3 is three wins in a row; L2 is two regulation losses; OT1 is one overtime or shootout loss.

Our running measure of team strength, built from results alone. Beating a strong team moves a rating more than beating a weak one; a regulation win counts more than an overtime or shootout win. Each summer every team is pulled part of the way back toward average.

See also: Pregame win probability. Win probability model →

Minutes and seconds a player was on the ice.

Example. 18:42 is 18 minutes, 42 seconds.

Any shot that was not blocked: goals, saves and shots that missed the net. These are the shots our xG model scores. Blocked shots are left out because the league logs where the block happened, not where the shot was taken from.

See also: Expected goals, Shots on goal. How our xG model works →