xG Explained: What Expected Goals Can and Can't Tell You

A better-behaved version of the scoreline — and the five things it quietly ignores.

Expected goals has moved from analytics blogs to television graphics in under a decade, and with the mainstreaming has come a lot of confident misuse. Understanding what xG measures — and what it cannot measure — is one of the more valuable things a bettor can do.

What the number is

Expected goals assigns each shot a probability of being scored, based on a model trained on historical shots with similar characteristics: distance, angle, body part, type of assist, whether it followed a set piece, the pressure on the shooter. A tap-in from six yards might rate 0.7 xG; a speculative effort from thirty yards, 0.02.

A team's match xG is the sum of those values. An xG of 2.4 means that a typical side taking that exact set of chances would score 2.4 goals on average — not that they deserved 2.4 goals in any moral sense.

Why it predicts better than goals

Goals are rare, which makes them statistically noisy. A side can win 1-0 from 0.4 xG against 2.1, and nothing in the scoreline records how fortunate that was. Over the following ten matches, the underlying numbers reassert themselves far more often than the results do.

For a bettor this is directly usable. A team on a good run of results with poor underlying numbers is likely to be overpriced next week, because the market and the public are both anchored to the scorelines. The reverse — good numbers, bad results — is where value tends to accumulate.

The limits

• Small samples stay small. A single match's xG is only a marginally better guide than the scoreline. Six to ten matches is where it starts to say something.
• Finishing skill exists, faintly. Most sides regress to their xG, but a small number of players genuinely outperform it over long periods. Assuming all overperformance is luck will cost you on the exceptions.
• Models disagree. Different providers assign different values to the same shot. Do not mix sources within one analysis.
• It ignores everything that is not a shot. Defensive organisation that prevents chances from forming, game management, and a side that habitually shuts down a match at 1-0 are all invisible to it.
• Match state distorts it. A team that leads early concedes territory and accumulates xG against. That is a chosen trade, not a weakness, and raw season totals do not separate the two.

Using it properly

1. 1.Compare a side's xG difference to its actual goal difference over ten or more matches. A wide gap flags a likely correction in one direction.
2. 2.Split home and away. The difference is frequently large enough to change a selection.
3. 3.Filter for match state where your data allows it — xG accumulated while level tells you more about a side's true strength than xG accumulated while chasing.
4. 4.Use it as an input to a probability estimate, then check that estimate against the margin-adjusted price. xG is not a betting system; it is one component of one.

The bettors who get most from xG are the ones who treat it as a better-behaved version of the scoreline rather than a verdict on who should have won. It narrows the noise. It does not remove it.