What Is Expected Goals (xG)? Reading Football Statistics Correctly

Even when a match ends 0-0, you can often feel that one side created far more danger — xG turns exactly that intuition into a number. Expected Goals is a modern statistical measure that calculates the probability of each shot becoming a goal from variables such as location, distance, shot type and defensive pressure. In this article we explain step by step how xG is calculated, where it can mislead, what the small-sample problem is and how to use it properly in match analysis — a guide for data literacy, not betting advice.

What Is xG? The Basic Logic

Expected Goals starts from a simple question: "How often does a chance like this typically turn into a goal?" The answer is built from historical data. By examining the real outcome (goal / no goal) of tens of thousands of shots, every shot is assigned a probability value.

For example, if a volley from the middle of the penalty area with the goal wide open has historically turned into a goal about 35% of the time, that chance carries an xG of 0.35. A one-on-one with the goalkeeper from within six yards deserves an xG of roughly 70–80%. By contrast, hard shots from 30 metres score only around 2–4% of the time.

A match's xG total is the simple sum of the probability values of all shots produced in it. If a team took 10 shots and those shots add up to 1.8 xG, the model is saying "with that performance, this team would be expected to score 1.8 goals on average."

How Is xG Calculated?

Different data providers build their own models, but the core variables are largely shared:

Location and Distance

Where on the pitch the shot is taken from is the strongest determinant of xG. A shot from the penalty spot has an xG of about 0.76, because roughly 76% of penalties are scored. The average for shots from outside the box falls below 0.05.

Shot Type

Headers, side-foot shots, volleys and first-time shots have clearly different historical scoring rates. The general tendency: headers get a lower xG than shots with the foot — they are harder to place and to generate power with.

Type of Previous Action

It matters whether the shot came from an assist pass, a cross, a corner, or after an individual dribble. For example, headers from direct corner deliveries get a lower xG than headers from open-play passes in a similar position — because the goalkeeper and the defence are ready and organised.

Opponent Pressure

More advanced models also take into account the number and distance of opposing players at the moment of the shot. A shot taken under pressure gets a clearly lower xG than one taken from an open position.

Defensive Situation

The goalkeeper's position, the height of the defensive line and the number of defenders bypassed by a pass are added as extra variables in some advanced models. Such "post-shot xG" models produce much more precise estimates.

The Strength of xG: Reading Pressure Instead of the Scoreline

The practical value of xG shows up here: when you compare a team's xG total over the last 10 matches with its actual goal total, the gap between the two numbers gives an important signal.

If a team has produced 14 xG in 10 matches but scored only 8 goals, its finishing has fallen short of xG. There are two explanations: either the quality of the forwards is low (persistently weak finishing), or a luck component is involved (a temporary deviation). Sample size is critical for telling the two apart — we return to this below.

On the other hand, a team that lost the match but won the xG fits the "lost despite controlling the game" profile. In the long run this picture tends to reverse; a match analysis that ignores it reads the future wrongly.

Where xG Can Mislead

xG is a powerful tool, but presenting it as if it had no limits is as misleading as the measurement itself.

The Small-Sample Problem

The xG of a single match carries high variance. A team can produce 2.5 xG and score 0 goals — the probability of that happening is, in theory, not low. The event is not rare; variance in football is large.

The minimum sample recommended for meaningful inference is usually 8–10 matches. A single-match xG is a snapshot, not a picture of the whole. Pundits saying "this team produced 3.2 xG today, so they are very good" are making the mistake of turning one match into a season-wide generalisation.

Practical rule: when interpreting xG, don't go on without asking "how many matches is the window?"

Differences in Data Quality

Not all xG models are of the same quality. Some sources use only location and distance (simple xG), some add shot type and pressure, and some also process head and body-part positions from camera data. The xG figures that different providers give for the same match can diverge.

Comparing xG numbers directly without knowing which source they come from is as misleading as using different scales.

Goalkeeper Quality

Standard xG models do not account for the individual performance of the goalkeeper at the moment of the shot. Post-shot xG (PSxG) fixes this to some extent: using the placement and speed of the ball within the frame of the goal, the "difficulty" of the save the goalkeeper had to make is calculated separately. However, this data is not available from every provider.

If goalkeeper quality is not taken into account, high-performing goalkeepers can be wrongly counted as having "saved" too little xG, or the opposite mistake can occur.

The Distinction Between a Creative Chance and an Obvious One

A shot can carry 0.9 xG, but the player expected to score may lose their balance, a defender may make a last-moment challenge, or the ball may hit the post. An xG model does not see this fine detail; it can't go beyond saying "this shot has historically turned into a goal 90% of the time." It does not model the quality of an individual's split-second decision.

Using xG Correctly in Match Analysis

Once you understand the limits, xG becomes a powerful analysis tool when paired with the right questions:

"Does this team consistently score below its xG across the season?" — If so, it points to a finishing problem. Asking the same question about a single match is meaningless.

"Did the losing team win the xG today?" — If so, the profile is dominant control but inefficient finishing. Whether this match pattern continues should be tested in the following matches.

"How does the goals − xG difference evolve over the season?" — If a positive difference persists, either there is an exceptional forward or a luck component is at work. Statistically, a positive difference is expected to converge toward zero in the long run (regression to the mean).

"Is there a gap between xGA (xG conceded) and actual goals conceded?" — If the gap is large, goalkeeper performance or the defensive line may be benefiting from, or suffering from, short-term luck.

Models That Use xG, and +90 Futbol

September 2026 update: In the top-5 European leagues (using Understat data), xG is now the backbone of the model. We combine the xG each team has produced and conceded in its last 38 league matches (75%) with actual goals (25%) to derive an attack/defence rating, and we calculate the expected goals (λ) from those. We made this change after measuring, in a test that predicts every match using only the data before it, that it gave more accurate results than the old goal-based model in all three markets: match result, over/under 2.5 and both teams to score.

Because we have no access to xG data for the Süper Lig, the same rating method is applied with actual goals (with an Elo adjustment); in the Champions League, since teams come from different leagues, a cross-league Elo rating is used. We don't hide this on our methodology page.

The strength of xG is that it reduces variance: while the score of a single match depends heavily on luck, the quality of the chances a team creates is a more stable signal. Even so, xG does not capture factors such as injuries, shifts in motivation or referee effects. It isn't perfect; but the methodology is open.

Glossary

The xG derivatives you will often meet in discussions:

• xG (Expected Goals): The sum of the expected-goal values per shot.
• xGA (Expected Goals Against): The total xG allowed to the opponent; measures defensive quality.
• xGD (Expected Goals Difference): xG − xGA; the net balance of pressure.
• PSxG (Post-Shot xG): xG recalculated after the shot is taken, based on where the ball is heading in the frame of the goal; used to measure goalkeeper performance.
• npxG (Non-Penalty xG): xG calculated with penalty shots left out; used to separate winning penalties as a skill.
• xA (Expected Assists): The xG value of the shot that results from a pass that becomes an assist; measures the quality of creative players.

Data Literacy: Questions Worth Asking

When you see an xG number, ask yourself these questions:

1. How many matches is this number drawn from? A single match, 10 matches, a season?
2. Which model is being used — simple location-based, or including pressure and shot type?
3. How big is the gap with the actual goal count? Is the gap consistent, or a one-match deviation?
4. Is xGA shared too — is defensive pressure in the table as well, not just attack?

An "xG analysis" that can't answer these four questions is most likely using the number to confirm a feeling.

Conclusion

Expected Goals is one of the most important metrics for moving football analysis out of arguments about the referee and into data. It puts a number on that "if only it had been a goal" moment that intuition keeps counting, and over the long run it makes visible which team controls the pitch. But like every powerful tool, it has limits: it is unreliable in small samples, it skips the individuality of the goalkeeper, and it calls for care when comparing across different providers.

Used correctly, xG answers a much more valuable question than "did they score?": "How good are the chances this team creates, and how efficiently does it turn those chances into results?" Being able to answer these two questions separately fundamentally changes how you read a match.

This article is not betting advice. xG and its derivatives are probabilistic tools; no statistic promises a certain outcome. Our aim is to convey clearly what the data says and what it does not say — the decision belongs to the reader.

You can follow the xG-based probabilities of our model and its Brier performance transparently on the past results page.

Frequently Asked Questions

What is Expected Goals (xG), in short?

A statistical metric that calculates the historical probability of each shot becoming a goal from variables such as location, distance, shot type and pressure. It takes a value between 0 and 1; an xG of 1.0 means "historically, this chance has turned into a goal every single time."

Is xG reliable for a single match?

No. Variance is high in a single match; a team can produce 2.5 xG and score zero goals. For meaningful interpretation a window of at least 8–10 matches is recommended. A single-match xG is a snapshot, not a season table.

What does it mean if xG and the goal count keep differing?

A persistent positive difference (goals > xG) indicates either exceptional finishing quality or a luck component. Statistically, this difference is expected to converge toward zero in the long run — regression to the mean. A persistent negative difference points to inefficient finishing or a spell of bad luck.

What is PSxG (Post-Shot xG) used for?

It is used to measure how difficult the saves a goalkeeper has actually made were. While standard xG models the situation before the shot, PSxG processes the placement and speed of the ball within the frame of the goal. The difference between PSxG and goals actually conceded tells you whether the goalkeeper saved more or fewer than expected.

Where can I see xG data?

On +90, xG is the main input to the team rating for top-5 European league matches; our data source is stated on the methodology page. When comparing xG values from different sources, check which model was used and whether the methodology is open.