Expected goals explained: what xG is, how to read it, and why the better chances often lose

Open a match page on ScoreTable and, under the statistics, you will sometimes find a row marked Expected Goals (xG) with a number like 1.17 next to a number like 0.32. It is the least self-explanatory figure on the page, and the most useful once you know what it is doing.

What the number actually is

Expected goals assigns every shot a probability that it becomes a goal, and a team's xG for a match is those probabilities added up.

The scale runs from 0 to 1 for a single shot. A shot valued at 0.3 is one that gets scored about 30 per cent of the time, going on what has happened to thousands of comparable shots before it. A tap-in from two yards with the goalkeeper beaten might be 0.9 or higher. A speculative effort from thirty-five yards might be 0.01, or one goal in every hundred tries.

What decides the value is the circumstances of the shot rather than what happened next. Models differ, but they generally use the distance and the angle to goal, the part of the body that struck the ball, what created the chance and what phase of play it came from, and where the defenders and the goalkeeper were standing.

So a team's xG answers a narrow question precisely: given the chances this side worked, how many goals would an average team have scored? It says nothing about whether they actually scored them.

Four matches where it disagreed with the result

The gap between xG and goals is the whole point of looking at it, and the Europa League's opening night produced four clean illustrations.

Omonia Nicosia 1–0 Celta Vigo is the classic shape. Celta had nineteen shots to Omonia's eleven and an xG of 1.17 against 0.32, so by the chances they created they were more than three times the better side. Omonia scored on 88 minutes and won. Nothing about that is unjust, and it is exactly the outcome a 0.32 side gets some of the time.

Olympiakos 2–1 Jagiellonia is the opposite, a match where the numbers and the result agree entirely. Olympiakos took twenty-five shots to Jagiellonia's four and built 2.53 xG to 0.52. They won by one goal, which is a narrower margin than the chances suggest, but they were the better side on both counts.

Sunderland 1–0 AZ Alkmaar shows why the scoreline can flatter a loser. Sunderland's 2.90 xG against 1.49 is close to a goal and a half of difference in chance quality, and they won it with a single penalty on 66 minutes. A 1–0 reads like a tight night. It was not.

AC Milan 0–2 Benfica is the one that most annoys people. Milan created the better chances, 0.96 to 0.64, had more of the ball and more shots, and lost by two at home. Benfica's two goals came from a set of chances that would usually yield well under one. That happens, often, and it is not evidence that xG is broken.

What it cannot tell you

Three limitations are worth holding on to, because they are where most arguments about xG come from.

There is no single xG. It is not a standardised statistic. Different providers use different data, define events differently and feed their models different features, so the same match can carry different xG values depending on who calculated it. Numbers from two sources are not directly comparable, and a value can be revised after the match as the underlying data is corrected.

One match is a tiny sample. A single match contains perhaps twenty or thirty shots between the two sides. Over a sample that small, the difference between goals and xG is driven substantially by chance. Calling a team unlucky on the basis of ninety minutes is not a strong claim.

Over and under performance is noisy. Comparing goals scored against xG across a season is the standard way to talk about clinical or wasteful finishing, and over enough matches it carries real signal. Over four or five matches it mostly does not.

The idea is older than the current argument about it, incidentally. Vic Barnett and Sarah Hilditch used the term "expected goals" in a 1993 paper on the effect of artificial pitches in English football, decades before it reached television coverage.

Where to find it on ScoreTable

xG appears in the statistics table on a match page, between the shot counts and the passing numbers, labelled Expected Goals (xG).

It is not on every match. The figure depends on the data provider supplying it for that fixture, and coverage is thinner outside the biggest competitions. All nine of the Europa League matches played on the opening Thursday came through without an xG value, for instance, while six of the nine played the night before had one. When it is missing, the shot counts and the shots on target are the next best guide to who created what.

Sources: Expected goals, Wikipedia, American Soccer Analysis on xG.