Attack and defence strength: goals for and against, read properly
TL;DR. Goals for and against are the most basic data in football and, raw, among the most misleading: they depend entirely on who they were scored against. Turning them into an opponent-adjusted attack and defence rating is the step that separates a table from a prediction tool, and it is the foundation of any serious goals model.
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.
The problem with raw goals
A team with 30 goals in 15 rounds looks like an attacking side. But if those goals came against the league's five worst defences, its real attack is far more modest than the number suggests. And conversely: conceding 20 goals having faced the five best attacks does not describe a bad defence.
The fixture list distributes opponents unevenly, and by mid-season the differences are enormous. That is why raw goals are, especially early on, more a reflection of the schedule than of the team.
How the rating is built
The idea is simple: compare what a team does with what the league average would do against that same opponent. If the league average scores 1.2 goals against an opponent and this team scores 2.4, its attacking strength against that defence is twice the average.
Repeating that comparison across every match gives two numbers per team: an attack strength and a defence strength, both relative to their competition's average. A value of 1.00 is exactly average; above it is better than average in attack (or worse in defence, depending on how the scale is oriented).
| Rating | Read |
|---|---|
| Attack > 1 | Scores more than an average side would against those same opponents |
| Attack ≈ 1 | Average attacking output |
| Defence better than average | Concedes less than an average side would against those opponents |
From ratings to probability
Combining the home side's attack strength, the away side's defence strength and the league's scoring average gives the home team's expected goals; and symmetrically the away team's. Those two numbers are the input to the Poisson model with the Dixon-Coles correction, which spreads probability across every possible scoreline.
From there come, at once, the 1X2 probability, the Over/Under goals probability and the both teams to score probability: they are not three different models, they are the same scoreline distribution read three ways. We explain it in full in how our model works.
Its limits
- It needs sample. After five rounds the opponent adjustment is still weak; early-season ratings should be taken with care.
- It is a snapshot of the recent past. It knows nothing about signings, injuries or a change of manager until results reflect them.
- It assumes a closed league. Ratings are only comparable within the same competition: a 1.30 attack in one league is not a 1.30 in another.
- Own goals and penalties distort it. They are infrequent events that weigh the same as any other goal.
Frequently asked questions
Why is looking at goals for and against not enough?
Because they depend entirely on the fixture list. Thirty goals against the league's worst defences do not describe the same attack as thirty against the best. The opponent adjustment is what turns raw data into comparable information.
What does an attack strength of 1.3 mean?
That the team scores roughly 30% more than an average side in its league would against those same opponents. The scale is relative to the competition's average, not absolute.
Can ratings from two different leagues be compared?
Not directly. Each rating is normalised against its own competition's average, so a 1.30 in one league is not equivalent to a 1.30 in another.