The Poisson distribution gives football analysis a repeatable way to describe the probability of scoring a specific number of goals. It is a model of count data, not a machine that predicts a final score. The standard definition is documented in NIST’s Engineering Statistics Handbook, and the assumptions matter as much as the formula.
Start with an expected-goal value, usually written as lambda. The model then assigns probabilities to zero goals, one goal, two goals, and so on. A second team receives its own lambda. Combining the two distributions creates a scoreline matrix from which broader outcomes can be summed.
The Poisson formula
The probability of exactly k goals when the expected-goal value is lambda is:
P(X = k) = (e-lambda x lambdak) / k!
The formula has three moving parts. e is the natural exponential constant, lambda is the expected number of goals, and k is the goal count being tested. The factorial prevents the count from being treated like a continuous measurement. The output is a probability for one count, not a promise that the count will occur.
Worked example with two expected-goal values
Suppose the home team is assigned lambda = 1.60 and the away team lambda = 0.90. Rounded to two decimal places, the individual goal probabilities look like this:
| Goals | Home, lambda 1.60 | Away, lambda 0.90 |
|---|---|---|
| 0 | 20.19% | 40.66% |
| 1 | 32.30% | 36.59% |
| 2 | 25.84% | 16.47% |
| 3 | 13.78% | 4.94% |
To estimate a 1-0 scoreline, multiply the home team’s probability of one goal by the away team’s probability of zero goals. Repeating that multiplication for every plausible pair creates the scoreline matrix. Home-win, draw, away-win, and total-goal estimates then come from sums of cells, not from selecting the single most likely scoreline.
How to choose the inputs
A simple starting point combines attacking and defensive scoring rates with a league average. Those rates should be adjusted only for evidence that can be described, such as venue, opponent strength, availability, schedule, and likely game state. A small sample can be distorted by a red card, a new manager, or an unusually strong sequence of opponents.
Expected goals are inputs, not facts waiting to be discovered. Keep the data window, calculation method, competition, and adjustments beside the output. If those choices change, the resulting probability changes too. The implied probability guide explains how to express the output as percentages, and the Asian handicap guide covers the goal-margin convention.
Where the simple model breaks
Goals are not independent in every match. A team that falls behind may attack more, while a team protecting a lead may create fewer chances. Scoreline correlation, referee decisions, tactical changes, substitutions, and injuries can therefore make a basic Poisson model overstate or understate parts of the matrix. Some analysts use extensions to account for those effects, but the extra complexity should be justified by better evidence.
How to use the output responsibly
Record the assumptions, preserve the unrounded calculations, and revisit the model after the match without changing the original inputs. A useful review asks which assumption failed and what evidence would have changed it. The model is an auditable starting point for discussion, not a certainty engine or a promise of profit. Any market comparison should remain informational and should never encourage chasing losses.
