How our Poisson prediction model actually works
Most football prediction models start the same way: estimate how many goals each team is likely to score against a league-average opponent, then combine those two numbers into a probability distribution over final scorelines using the Poisson distribution.
We layer a Dixon-Coles correction on top, which nudges the probabilities for very low-scoring results (0-0, 1-0, 0-1, 1-1) because raw Poisson models tend to underrate how often tight, cagey matches actually finish level.
The output is a full probability grid across scorelines, from which we pick the single most likely one as our headline prediction, and report our confidence as the probability mass behind the winning outcome (home, draw, or away) rather than the exact score, since exact scores are inherently harder to call.
Back to blog