Guide · updated 2026-10-05
Expected goals from a score model, and how it differs from xG
The expected goals on our match pages are a forecast: the average number of goals each team should score, worked out before kickoff from team ratings. Shot-based xG is a measurement of the chances a team actually created once the match is played.
Two different numbers with the same name
Pre-match expected goals come from the Dixon–Coles model: attack and defence ratings from past results, home advantage, and adjustments for missing players. Shot-based xG adds up the scoring chance of every shot in a match, based on things like distance and angle. One looks forward, the other looks back.
Example: Arsenal vs Leeds United
The model expects 1.93 goals for Arsenal and 0.77 for Leeds United, 2.70 in total. From those averages it gives Arsenal a 46.2% chance of a clean sheet and Leeds United 14.6%. Full probabilities are on the match page.
Reading them together
If a team keeps out-creating opponents on shot-based xG but its results lag behind, a results-only model can be slow to rate it up. That is one reason a model like ours can differ from the market, which sees chance quality, injuries and lineups as well as results.
Common questions
Is the expected goals figure on match pages xG?
No. It is the average number of goals the score model expects each team to score before the match, from team ratings. Shot-based xG measures the quality of chances a team actually created during a match.
Why can expected goals be a decimal?
It is an average over every possible result. 1.6 expected goals means that across many matches like this one the team would average 1.6 goals, not that 1.6 goals can be scored.
How does expected goals become a probability?
Each side's goals are treated as a Poisson count with that average, with a correction for low scores. That gives a probability for every scoreline, which add up to home, draw and away probabilities.
More guides
Probabilities are not guarantees and nothing here is betting advice. Responsible gambling.