Expected Goals (xG) Explained: What It Measures and How to Use It

How a shot gets its value

A model is trained on a large number of historical shots and learns how often similar shots became goals. The inputs typically include distance and angle to goal, body part (foot or head), the type of pass or situation that created the chance and whether it was a penalty or a set piece. Different providers use different models, so their values differ slightly.

Adding it up: a match example

Team A has five shots worth 0.05, 0.08, 0.12, 0.76 (a penalty) and 0.31. The match xG is 0.05 + 0.08 + 0.12 + 0.76 + 0.31 = 1.32. Team B creates 0.74. If the final score is 0-2 for B, the xG says A created more but did not convert, which can easily happen.

Under a Poisson model with an average of 1.32 goals, the chance of scoring zero is e^-1.32 ≈ 26.7%. Teams do fail to score from quality chances more often than intuition suggests.

Using xG over a season

One match is noisy, a whole season is much more informative. A team that scored 40 goals from 30 xG is over-performing by 10. Finishing is partly skill, but such a gap usually shrinks over time, so the team is more likely to cool off than to keep going.

• Compare goals scored with xG, and goals conceded with xG against, over at least 10-15 matches.
• Look at the trend: a team whose xG difference is positive but whose results are poor may be unlucky.
• Use xG as an input for expected goals per team, the lambda that drives Over/Under and BTTS probabilities.

Limits

• xG ignores who took the shot and who was in goal, unless the model is a post-shot variant.
• Game state matters: teams ahead often shoot less, so the xG of a match can be skewed by its flow.
• Providers disagree. Compare numbers only from the same source.
• It describes what happened, it does not predict a single match. A price must still be compared with a probability.