Start with an expected scoring rate
A Poisson score model assigns a mean goal rate to each team. For one team with mean lambda, the probability of scoring k goals is exp(-lambda) multiplied by lambda to the power k, divided by k factorial. The mean is an expectation across comparable matches, not a score the team must reach.
An independent model multiplies the home and away goal probabilities to form a score grid. Every cell describes one exact score. Add the cells above, on or below the diagonal to obtain home-win, draw and away-win probabilities. Do not mistake the most likely exact score for the full outcome probability.
A small worked score grid
Use illustrative goal rates of 1.4 for the home team and 0.9 for the away team. The independent probability of 0–0 is exp(-2.3), approximately 10.03 percent. The probability of 1–0 is 1.4 times exp(-2.3), approximately 14.04 percent. The probability of 1–1 is 1.26 times exp(-2.3), approximately 12.63 percent.
These are three cells, not a complete forecast. A finite grid also leaves probability outside its maximum goal count. A careful implementation measures that omitted tail and either extends the grid or explains its normalization. Quietly dropping the tail changes the reported probabilities.
Where the model can fail
The independence assumption is a simplification. Teams respond to the score, substitutions and dismissals. Dixon and Coles introduced a correction involving low-scoring outcomes rather than treating every score pair as independent. A correction still needs a documented fitting procedure and evaluation on data not used for fitting.
Before using a model output, check whether its rates come from recorded team results, shot-based expected goals or another input. These are different measurements. Match Insights preview estimates must retain their source limitations. A familiar model name does not make stale fixtures, missing lineups or an untested forecast reliable.
Common question
Is a model goal rate the same as shot-based xG?
No. A fitted pre-match scoring rate and shot-based expected goals describe different inputs and must be labelled separately.
Sources and further reading
Continue your research
- How to read football match probabilities
- Why sample size matters in football research
- Expected goals: what xG can and cannot explain