In this guide
xG starts with a shot and its context
Expected goals, or xG, estimates how likely a shot is to become a goal. It uses a model to put a number on the chance, rather than judging it only by whether the ball went in. A shot assigned 0.30 xG has a modeled goal likelihood of 30% under that model. That is a shot estimate, not a team's chance of winning.
You can read xG more clearly by keeping three things separate: the information used to assess a shot, the sum of the shot values, and the goals actually scored. The example below follows all three without using real match data.
What information does the model use?
In Opta's explanation published in August 2023, xG uses a shot's context and similar historical shots to estimate its goal likelihood. The context includes distance and angle, shot type, goalkeeper and other player positions, pressure, and the play or assist preceding the shot.
Those details help describe the opportunity the player had. Counting shots alone leaves them out. When you read a value, ask which model produced it and what information that model used. The features above describe Opta's published 2023 model; they are not a checklist that every provider or later version must follow.
Read a shot log and its total
Imagine three shots assessed on the same model basis. We assign them the fictional values below. The outcome column records a goal as 1 and no goal as 0. These are invented values and results, not calculations from Opta or American Soccer Analysis.
| Shot | Assigned xG | Fictional observed goal |
|---|---|---|
| A | 0.10 | 0 |
| B | 0.30 | 1 |
| C | 0.50 | 0 |
| Total | 0.90 | 1 |
The assigned values add to 0.10 + 0.30 + 0.50 = 0.90 expected goals. The observed outcomes add to 0 + 1 + 0 = 1 goal. The two columns answer different questions: one summarizes the model's assessments; the other counts what happened.
The total of 0.90 does not mean that 90% of a goal happened, or that there was a 90% chance of scoring at least once. Nor does shot C's larger estimate require it to be the shot that scores. A fractional expectation and an integer result can differ without contradicting each other.
There is a useful mathematical distinction here. For a fixed set of goal indicators, the expectation of their sum equals the sum of their expectations. This is called linearity of expectation, and it does not require the shots to be independent. Using model estimates in that sum gives an estimated total; the identity does not establish that the estimates are accurate.
You would need more than this sum to describe the possible combinations of goals or calculate the chance of at least one goal. This selected shot log also does not turn into a forecast of a new match. For the broader distinction between an average and an outcome probability, see the guide to expected value and the chance of a result.
Check the model before comparing numbers
Two figures labeled xG may come from models built for different roles or with different inputs. American Soccer Analysis's explanation of Expected Goals 3.0, for example, distinguishes shooter/team and goalkeeper models. Its goalkeeper discussion includes shot placement and a calculation that assumes a geometrically optimal goalkeeper position. That assumed position is not a measurement of where the goalkeeper actually stood.
| Source or record | What it describes | What to keep separate |
|---|---|---|
| Opta explanation, August 2023 | Shot context before and up to the shot, including player and goalkeeper positions, pressure and preceding play. | This published model description and any later version's inputs or configuration. |
| ASA explanation, Expected Goals 3.0, undated | Distinct shooter/team and goalkeeper inputs; the goalkeeper discussion includes placement and an assumed-position calculation. | An assumed goalkeeper position and an observed one; the roles of the different models. |
| Our fictional three-shot log | Comparable assigned shot values and separate observed outcomes. | The estimated total, the actual goal count and any match-winning probability. |
Before comparing two readings, record the provider, model version, population it describes, what the shot model estimates, and which shots or period the figure covers. A shooter/team figure and a goalkeeper-model figure should not be treated as interchangeable just because both involve goals. These descriptions also give no basis for declaring one provider more accurate.
A total does not name the match winner
A team's xG total is a summary of modeled shot chances. A higher total does not establish that the team should have won, and it is not itself a match-winning percentage. The result and the model's assessment remain different things.
You also cannot identify finishing skill, bad luck or model accuracy from one unexplained gap between xG and goals. Start by checking what was counted and how it was assessed. A single total leaves much of that context out.
If you are reading other football numbers alongside xG, keep their meanings separate too. Implied probability from odds and probability inferred from a market price concern different objects and assumptions. Neither is supplied by adding the three shot values in this example.
Common questions about xG
What does 0.30 xG mean for a shot?
It represents the model's estimate of a 30% goal likelihood for that shot, given the information it uses. It does not establish the shot's true probability or guarantee a particular outcome.
How can a team score one goal from 0.90 xG?
The xG total adds estimates; the score counts goals. In our fictional log, the three assigned values sum to 0.90 while the observed outcomes sum to one. An expected count can be fractional even though the observed count is an integer.
Does higher xG mean a team should have won?
No. A larger xG total does not decide the result or give the team's winning probability. It also does not, by itself, explain why the result differed.
Can different xG models give different readings?
Yes. Check the provider, version, population, inputs and model target before comparing them. The named Opta and ASA explanations illustrate why you need that context; their differences alone do not tell you which model is more accurate.
Must the shots be independent to add their expected values?
No. Linearity of expectation applies to a fixed set of goal indicators without independence. That does not validate estimated shot values or provide the joint outcome probabilities needed to calculate the chance of at least one goal or a match win.
Explore HypaForecast markets
Browse the market list, or view the questions and round details on the BTC 5-minute and 15-minute pages. These links are separate from the football example and do not establish current availability or your eligibility.
Sources and model scope
This guide uses Opta’s explanation published on 8 August 2023 and the undated American Soccer Analysis page describing Expected Goals 3.0. Their explanatory text was read; linked studies, videos and images were not reviewed. These descriptions are not verified as the providers’ current model configurations. The three-shot log and its outcomes are invented for this guide.
HypaForecast publishes educational guides and operates a prediction market. This football-data explanation does not establish that the platform offers football markets or xG tools, or that you are eligible to trade.
Share this guide