What is xG in football? Expected goals and key stats explained

Every shot gets a value between 0 and 1. The model looks at features such as:

• distance from goal and angle to the goal,
• body part (foot or head),
• type of assist (a cross, a through ball, a cut-back),
• situation (open play, a corner, a direct free kick, a penalty),
• in more advanced models, the position of defenders and the goalkeeper.

The model then asks: of all past shots like this one, what share went in? That share is the xG.

Exact values differ between data providers, because each uses its own model. Compare xG figures from the same source.

A 0.76 chance (about the value of a penalty in many models) isn't "a goal that should have been scored". It means that, from that position and situation, players score about 76% of the time and miss about 24% of the time. If a striker misses one, it's disappointing, not impossible. If a team misses three in a match, that's bad luck or bad finishing, and xG will show it.

So a team with 1.23 xG "deserved" a bit more than one goal on average. If it scored three, it finished well or was lucky; if it scored none, the reverse. Over one match, luck is huge. Over a season, xG is a strong guide to how good a team's attack really is.

A team with a high xGD but a poor league position has often been unlucky and tends to climb; a team high in the table with a low xGD often slips back. That's the main reason analysts follow xG rather than results alone. For how the table itself works, see league tables explained.

• Shots and shots on target: simple and widely available, but they treat every shot the same. Five shots from 30 metres are worth less than one tap-in.
• Possession: the share of the ball a team has. It describes a style more than quality. Plenty of teams win with 35% possession by defending deep and counter-attacking.
• PPDA (passes allowed per defensive action): how many passes the opponent makes in its own part of the pitch before a defensive action (a tackle, interception or foul). Lower PPDA = more intense pressing. A pressing team might allow 8–10 passes; a deep-defending team 15 or more.
• Big chances, touches in the box, field tilt: other ways of measuring territory and threat.

xG is most useful for goals markets, because it estimates how many goals each team should score and concede. A simple approach is to take each team's expected goals for a match and use a Poisson model, the standard way to turn an average number of goals into probabilities.

The example shows something important: bookmakers use the same kind of data, so the market price is usually close to what a simple model says. A model only helps if you have better inputs than the market (team news, a change of coach, a style mismatch). For how these markets work, read over/under goals and BTTS; for the maths of comparing your probability with the price, see value betting.

• Treating one match's xG as the truth. It's an estimate built on averages, and a single game is a tiny sample. Use 10+ matches.
• Comparing numbers from different providers. Models differ; stick to one source.
• Ignoring game state. A team that leads early often sits back, which lowers its xG and raises the opponent's.
• Reading possession as dominance. Look at the chances created, not the share of the ball.
• Forgetting penalties. One penalty adds about 0.76–0.79 xG; npxG gives a cleaner view of open play.

You can see team and league numbers on our stats page, and scoring patterns such as over 2.5 and BTTS rates on the trends page.