Expected goals (xG)

Each unblocked shot is scored for quality (how far out, at what angle, with what kind of shot, at what strength, and whether it came off a rebound) and those probabilities are added up. A team's expected goals is what an average finisher would have scored on the chances it created.

The model here is binned rather than a regression: shots are grouped by those five features and each bin uses its own observed goal rate, shrunk toward coarser bins where the sample is thin. Every number in it is a goal rate you can look up, which matters more for a figure that gets published than a fraction of a point of accuracy.

It is calibrated within each season, so the league's expected goals equal the goals the league actually scored that year. Scoring environments move, and without that a team from a low-scoring season carries an era effect before it plays a game.