Chance quality
Expected goals (xG) explained
Expected goals assigns each shot a probability of becoming a goal, then adds those probabilities to describe the quality of chances created. It is evidence about shots—not a verdict on which team deserved to win.
· 8 min read
What does xG mean in football?
Expected goals, usually shortened to xG, estimates the probability that a shot becomes a goal. A shot assigned 0.20 xG is one that a particular model expects to be scored about 20% of the time across many comparable attempts.
How an xG model estimates a shot
Models learn from historical shots and their outcomes. Location and angle are common inputs. Richer models may also consider the body part used, the type of assist, whether the attempt followed a set piece, defensive pressure and the position of the goalkeeper.
- Closer shots usually receive more xG than distant shots.
- A clear central angle is generally more favourable than a narrow one.
- Headers and shots with the foot can have different scoring patterns.
- Different data and feature choices produce different xG values.
Team xG is the sum of its shot probabilities
Adding the individual shot probabilities gives the expected number of goals across repeated comparable sets of chances.
The total is an expectation, so it can be a decimal and can exceed the number of goals that were actually possible in one realised sequence. It does not mean the team should literally have scored that exact number in the match.
Worked example: two chances worth 0.45 xG
The 0.45 total is not automatically a 45% chance of scoring at least once. Under a simple independence assumption, that probability would be 1 − (0.65 × 0.90) = 41.5%. The distinction matters when xG is used inside a wider match model.
How xG can help a football prediction model
Goals are rare and noisy. Chance-quality measures can add context to a run of results by separating the opportunities a team created from the finishing that happened afterward. A prediction model may use recent xG for and against alongside opponent strength, venue and other pre-match evidence.
Why two xG providers can disagree
There is no single universal xG model. Providers can observe different features, clean events differently, train on different competitions and choose different algorithms. Small differences are therefore expected rather than proof that one number is fabricated.
- Compare values from the same provider when tracking a trend.
- Check whether penalties and own goals are treated consistently.
- Look for model documentation and validation, not only attractive graphics.
- Avoid mixing pre-shot xG with post-shot measures such as xG on target.
What xG cannot tell you
xG measures recorded shots, so it does not fully describe attacks that never became attempts, tactical control, game-state choices or every defensive action. It also cannot remove finishing variation or guarantee the next result.
- It cannot say which team deserved to win from one total alone.
- It cannot turn a single match into a reliable sample.
- It cannot make providers with different models directly interchangeable.
- It cannot replace a complete, timestamped evaluation of forecast probabilities.