The Poisson distribution is the standard statistical tool for football prediction. Given how many goals a team is expected to score (its expected goals, or xG), Poisson tells you the probability of them scoring exactly 0, 1, 2, 3 … goals. Combine both teams and you can price every scoreline — and from there, every market.
The formula in one line
P(k goals) = (λ^k × e^(−λ)) / k!
where λ (lambda) is the team's expected goals and k is the exact number you want the probability of. It looks intimidating but it answers a simple question: if a team is expected to score λ goals, how likely is exactly k?
A worked example
Say Arsenal are expected to score λ = 1.8 goals at home. Poisson gives:
| Goals (k) | Probability |
|---|---|
| 0 | 16.5% |
| 1 | 29.7% |
| 2 | 26.8% |
| 3 | 16.1% |
| 4 | 7.2% |
Notice the most likely single outcome is 1 goal — even though the average is 1.8. That gap between the mean and the most likely value is exactly why intuition misleads bettors.
From one team to a full match
Model each team separately — say Arsenal λ = 1.8 and the away side λ = 1.0 — then multiply the independent scoreline probabilities to build a full grid:
- P(2–1) = P(home 2) × P(away 1)
- Sum every cell where home > away → home win probability
- Sum the diagonal (1–1, 2–2 …) → draw probability
From that single grid you can read off 1X2, Over/Under 2.5, BTTS and correct score probabilities at once. This grid is the heart of how our prediction model works.
The Dixon-Coles adjustment
Plain Poisson has a known flaw: it slightly underestimates low-scoring draws (0–0 and 1–1), which happen more often in real football than independence implies. The Dixon-Coles adjustment nudges those specific scorelines upward. We apply it because the draw is already the hardest outcome to price, and getting 0–0 and 1–1 right matters disproportionately.
Where Poisson breaks down
- Goals are not perfectly independent. A team that goes 2–0 up often eases off; a team chasing pushes harder.
- It needs good λ estimates. Poisson is only as accurate as the expected-goals figures you feed it — garbage in, garbage out.
- Red cards and injuries mid-match break the pre-match assumptions entirely.
This is why we never rely on the model alone: its output is cross-checked against bookmaker odds before any tip is published.
Frequently asked questions
Why use Poisson for football specifically?
Goals are rare, independent-ish events occurring over a fixed period — exactly the situation the Poisson distribution was designed to model. It fits football scoring patterns well, which is why it has been the academic and industry standard for decades.
What is lambda (λ) in a Poisson football model?
Lambda is the team's expected number of goals in the match — essentially its expected goals (xG) estimate for that specific fixture, adjusted for the opponent and home advantage.
Is a Poisson model accurate for predicting football?
It is accurate at estimating probabilities over large samples, not at calling individual results. A well-calibrated Poisson model produces realistic long-run frequencies; any single match still carries large irreducible uncertainty.
What is the difference between Poisson and Dixon-Coles?
Dixon-Coles is a refinement of the basic Poisson model that corrects its tendency to under-count low-scoring draws and adds a time-decay so recent matches weigh more heavily. It is more accurate for football than plain Poisson.
