Expected Goals

Six inputs decide a shot's number

A model trained on hundreds of thousands of historical shots scores every new one from 0 to 1: the share of near-identical attempts that went in.

• Distance to goal. The single biggest factor.
• Angle. How much of the goal frame the shooter can actually aim at. A shot from the penalty spot beats one from the same distance on the byline.
• Body part. Strong foot, weak foot and header convert at different rates from identical positions.
• Shot type. Open play, set pieces, free kicks and penalties each carry their own base rate.
• Assist type. A cut-back to the penalty spot leaves a defence more exposed than a cross from the same position.
• Defensive pressure. Where the defenders and the goalkeeper stood when the shot was struck.

The model never sees who took the shot. Every player hitting the same chance gets the same number. That is deliberate. It turns the gap between a player's goals and their xG into a clean read on finishing, not a number that already has finishing baked in.

Read one shot at face value, then stop

A shot worth 0.5 xG goes in about half the time and misses about half the time. Missing one proves nothing. Getting into the position to take it does.

Match totals carry the real signal. Add up every shot a team took and you get a picture of the chances created, independent of finishing. A side that creates 2.4 xG and loses 1-0 played well and lost. That is not consolation: teams in that position score more and win more over the following weeks than the scoreline suggests.

Season totals settle further still. Goals and xG converge across a full campaign, because outscoring or underscoring your chance quality is hard to sustain over several hundred shots. The gap between the two columns is one of the clearest signals on a league table.

Sample size bites hardest at player level. A single penalty is worth roughly eight speculative efforts from outside the box. Five goals from 2.1 xG is a good month, not proof of a gift for finishing, and it takes hundreds of shots to tell the two apart. That is why a player's xG predicts their next ten matches better than their goals do.

xG answers three different questions

Judging whether a team is good. Cut through the results. A side seventh on a run of tight wins and a side fourteenth after four one-goal defeats can carry near-identical xG numbers. The table tends to move towards both of them over time.

Judging which player produces next. xG leads goals as an indicator, not the other way round. A forward whose goals dried up while chance quality held is a different case from one whose chances dried up too. Only one of those is a problem worth worrying about.

Explaining what actually happened. "They were unlucky" becomes "they created four clear chances and scored none of them", a claim you can check against the shot list rather than argue about.

The objections, and why they do not hold

• "It ignores finishing skill." It isolates finishing. Comparing goals against xG is the only way to measure it, and that comparison needs the second number to exist.
• "A high xG miss proves the player is poor." Reaching a 0.8 xG position is the hard, repeatable part. Converting any single chance mostly is not.
• "It cannot handle set pieces or penalties." Every shot carries its build-up type, and a penalty carries its own base rate near 0.78. Use non-penalty xG when you need open play alone.
• "Different sites disagree, so the number is unreliable." Providers train different models on different event data, and values for one shot can move by around 0.1 between them. They agree on direction almost every time. xG Stat runs on Wyscout data throughout, so every figure on this site is at least consistent with every other figure on it. That matters more for comparison than matching somebody else's number.