Score probabilities

Poisson models for football predictions

A Poisson football model starts with an expected goal rate for each team, then estimates the probability of every possible goal count. Combining the two distributions creates scoreline and match-result probabilities.

· 9 min read

What is a Poisson model in football?

The Poisson distribution models a count of events in a fixed interval. In football, the event is a goal and the interval is one match. A model supplies a mean goal rate, written as lambda (λ), and the distribution assigns a probability to scoring 0, 1, 2 and more goals.

The Poisson goal formula

P(X = k) = e^(−λ) × λ^k ÷ k!

X is the number of goals, k is a specific non-negative goal count, λ is the expected goal rate, e is Euler’s number and k! is the factorial of k.

If a team has λ = 1.60, the model expects 1.60 goals on average across many comparable matches. That does not mean the team will score 1.60 goals in this match; actual goals remain whole-number outcomes.

How the expected goal rates are estimated

A basic football model can estimate separate attack and defence strengths from historical scores, then combine the home team’s attack with the away team’s defence and a home-advantage term. Modern systems may also use expected-goals data, opponent strength, recency and availability evidence.

  • Use only information available before kickoff.
  • Adjust for the strength of the opponents faced.
  • Weight recent evidence without discarding useful older matches.
  • Validate the resulting probabilities on matches not used to fit the model.

Worked example: the probability of a 2–1 score

The same calculation can be repeated across a score matrix. Add the cells where home goals exceed away goals for the home-win probability, equal scores for the draw, and lower home scores for the away win.

From scorelines to match probabilities

A score matrix supports more than the match result. Summing the relevant cells can estimate totals, both-teams-to-score and exact-score probabilities. The matrix must cover enough goal counts that the omitted tail probability is negligible, rather than silently treating high scores as impossible.

Where a basic Poisson model can fail

The simplest version assumes a constant scoring rate during the match and independence between the teams’ goal counts. Football can violate both assumptions because red cards, score effects, tactics and shared match conditions change how the teams behave.

  • Low-score outcomes can show dependence that an independent model misses.
  • A fixed rate cannot represent a match whose scoring conditions change sharply.
  • Old results can distort current team strength after transfers or tactical changes.
  • Good fit to score frequencies does not guarantee calibrated future probabilities.

Sources and further reading