What is xG?
Expected goals, usually shortened to xG, measures the quality of a scoring chance. It estimates how likely a shot is to become a goal, based on information about that chance.
Each shot receives a value between 0 and 1. A shot worth 0.10 xG has an estimated 10% chance of being scored. Across many similar chances, a well-calibrated model would expect roughly one goal for every ten attempts.
That does not mean every tenth shot will go in. It is a probability, not a schedule.
Add together a team’s shot values and you get its total xG. That total helps describe the chances it took, rather than just how often it shot.
How is xG calculated?
An xG model learns from a large collection of past shots. Each shot has a recorded outcome, goal or no goal, plus details such as its location and how it was taken.
During training, a statistical or machine-learning model looks for relationships between those details and scoring outcomes. It then uses those relationships to estimate the scoring probability of another shot.
There is no single universal xG formula. Providers use different data, definitions and modelling methods, so they can assign different values to the same chance.
The basic process is:
- Collect shot data, including outcomes and relevant details.
- Train a model to estimate scoring probability from those details.
- Test it on separate data to check how well its predictions hold up.
- Assign each shot a probability and sum the values when needed.
A useful check is calibration: among many shots rated around 0.10 xG, does the actual scoring rate come out close to 10%?
Which factors affect a shot’s xG?
Shot location
Distance and angle are central inputs.
A close-range shot generally has a better chance of going in than a long-range attempt. A central position also usually gives the attacker more goal to aim at than a tight angle near the byline.
Distance alone is not enough. Two shots taken equally far from goal can present very different opportunities.
Shot type
Models commonly distinguish between headers and shots taken with the feet. Some also record other body parts or more detailed techniques.
A header and a right-footed shot from the same location are not necessarily equally difficult. The height of the ball and the control required can affect the chance of scoring.
Assist type
How the ball reaches the shooter matters too.
A model may distinguish between chances created by crosses, through balls, cutbacks or other passes. Some also identify rebounds and chances created without a pass.
A cutback to an attacker facing goal can produce a different opportunity from a high cross that must be headed under pressure, even when the attempts come from similar positions.
Other match details
Depending on the available data, models may include:
- Whether the chance came from open play, a corner or a free kick.
- Whether the shot was taken first time or after controlling the ball.
- The positions of defenders and the goalkeeper.
- The pressure on the shooter.
- Whether the attacker was in a one-on-one situation.
Penalties are commonly assigned a standard value within a provider’s model, based on historical conversion rates.
Not every model has access to all these details. Tracking data or images captured at the moment of the shot can add context that simpler event records miss.
A worked example: shots versus chance quality
Imagine two teams finish a match with these chances. The values below are illustrative, not figures from a real game.
| Team | Chances | Total shots | Total xG | |---|---|---:|---:| | Team A | Ten long-range shots, each worth 0.03 xG | 10 | 0.30 | | Team B | Two close-range shots, each worth 0.40 xG | 2 | 0.80 |
Team A wins the shot count comfortably. Team B creates the higher total chance quality.
The calculation is simple:
- Team A: 10 × 0.03 = 0.30 xG.
- Team B: 2 × 0.40 = 0.80 xG.
This is why xG is usually more informative than raw shots when assessing attacking chances. Shot counts treat every attempt equally. xG recognises that a hopeful effort from distance is not the same as a clear close-range opening.
Neither total tells you the actual score. Team A could score a spectacular goal while Team B misses both chances.
Does xG measure finishing?
Standard, pre-shot xG mainly describes the chance at the moment of the attempt. It generally does not use where the resulting shot travels within the goalmouth.
A poor finish from a good position still carries the original chance’s xG value. So does a brilliant finish from a difficult position.
Comparing goals with xG can help flag unusually strong or weak finishing results. But a short run above xG does not prove exceptional skill, and a short run below it does not prove poor finishing.
Post-shot xG models address a different question. They use information about the resulting shot, such as placement and sometimes speed, to assess how difficult it was to save.
What are xG’s limits?
It only values shots that happen. A dangerous pass that narrowly misses an attacker may produce no shot and therefore no standard shot-based xG.
It depends on the data. A model without defender positions may miss an important block or pressure situation.
Totals hide the sequence. Several rebound shots can add up to substantial xG, even though scoring the first would have prevented the later attempts. Total xG is not the probability of scoring at least once.
Match context matters. A leading team may defend deeper while its opponent accumulates shots. Read xG alongside the scoreline and the flow of the game.
The takeaway
Use xG to ask: how good were the chances? It improves on raw shot totals by weighting attempts according to scoring probability.
It is still an estimate, not a verdict. Combine it with match context, other statistics and what happened on the pitch.