Expected goals per team: the number everything else comes from
TL;DR. Two very different things go by the name «expected goals». One measures what has already happened: the xG of the shots in a played match. The other estimates what is going to happen: the average number of goals a model assigns each team before kick-off. This article is about the second, the number our whole prediction comes from.
Note. Informational and statistical content. It is not betting advice nor a promise of profit. No statistic predicts a single match: they are there to estimate probabilities over the long run. 18+ only. Play responsibly.
Two things with the same name
| xG of a played match | Model expected goals | |
|---|---|---|
| When it is computed | Afterwards, on real shots | Before the match |
| What it depends on | The location of each shot | Both teams' ratings |
| What it is for | Evaluating a match's performance | Generating the scoreline distribution |
| What it is NOT | A prediction | A measure of what happened |
The confusion is understandable because they are related: an xG history is a good basis for estimating the ratings that produce pre-match expected goals. But they are not interchangeable, and mixing them produces comparisons that mean nothing.
How the model's are calculated
Three things are combined: the attack strength of the attacking side, the defence strength of the defending side, and the league's scoring average as a reference. The result is one number per team: the goals expected from them in this specific match against this specific opponent.
In our engine that estimate is further adjusted by anchoring supremacy — the expected gap between the two — to what the 1X2 market reflects. It is a deliberate decision: the market aggregates a great deal of information a statistical model cannot see (line-ups, late absences, context), and refusing to use it would mean discarding good information out of model pride.
From two numbers to every market
With home and away expected goals, the full scoreline distribution is built using Poisson with the Dixon-Coles correction — which adjusts the frequency of low scorelines, where pure Poisson falls short.
And from that single table come all the probabilities: 1X2 by summing home win, draw and away win cells; Over/Under by summing those above the line; both teams to score by summing those where neither side is kept at zero; handicap by summing by margin. That is why all our probabilities stay consistent with each other: there are not four models, there is one distribution read four ways.
The limits
- It is an average, not a scoreline forecast. A value of 1.7 expected goals does not mean the team will score one or two: it means that is the mean of a distribution including 0 and 4.
- It depends on the quality of the ratings. With few matches played, ratings are unstable and expected goals inherit that instability.
- It does not know the line-ups. A significant late absence is not in the number, although the market anchor captures it partially.
- A good estimate is not an edge. Goals markets are among the most efficient there are: getting the probability right and finding room against the price are different things.
Frequently asked questions
Are the model's expected goals the same as xG?
No. xG measures the quality of shots in a match already played; the model's expected goals estimate before kick-off how many goals each team will score. One describes the past, the other predicts the future.
What does it mean for a team to have 1.7 expected goals?
That this is the mean of a distribution of possible outcomes that includes scoring zero and scoring four. It is not a forecast that they will score one or two.
Why does the model anchor to the market?
Because the market aggregates information a statistical model cannot see: line-ups, late absences and context. Ignoring it on principle would mean discarding good information.