Poisson Football Calculator
Enter the expected home and away goal rates (λ) to build an independent-Poisson score distribution. The Poisson football calculator derives correct score (exact-score) probabilities, 1X2, BTTS, totals and fair odds from the same pair of goal rates.
How to use the Poisson football calculator
Enter the expected goals for the home and away teams. For example, Home λ = 1.56 and Away λ = 1.18. This Poisson calculator does not need betting odds — only the two goal rates. It builds a separate goal distribution for each team, combines them into correct score probabilities, and then derives 1X2, BTTS, goal totals and fair decimal odds.
Step 1 — Enter expected goals
Use one full-match goal-rate input for each team. λ = 0 is valid and means the model assigns 100% probability to that team scoring zero goals.
Step 2 — Build the score distribution
The independent-Poisson model calculates each team’s 0, 1, 2, 3… goal probabilities and multiplies the two distributions to form scoreline probabilities.
Step 3 — Read the derived markets
The score distribution produces the ranked scores and Home/Draw/Away probabilities. BTTS uses the independent-team formula; totals use Poisson with rate λ home + λ away. Fair decimal odds are 1 / probability.
Input contract: this page starts with λ values. If you instead have bookmaker 1X2 and Over/Under odds, use the market-implied goals workflow or the main ImpliedScore calculator to infer λ first.
Need the football-specific low-score dependence correction?
Compare with Dixon-ColesWhat is the Poisson distribution?
A Poisson distribution models a non-negative event count conditional on a rate parameter λ. In football, λ can represent a team’s expected full-match goals; once λ is specified, the distribution assigns a probability to 0, 1, 2, 3 and every higher goal count.
One parameter: λ (lambda) = expected goal rate. If a team's expected goals for this match are 1.5, λ = 1.5. From this single number:
Formula: P(k) = e^(−λ) × λ^k / k!. This is a full-match count model conditional on λ; it does not by itself specify when goals occur during the match.
Independent Poisson goal modelling in football goes back to Maher's 1982 football Poisson model, which used team-specific attack and defence parameters to produce fixture-level goal expectations.
For a less technical introduction to goal rates and the later low-score correction, read Maher and Dixon-Coles football models explained.
What assumptions does the football baseline make?
The independent-Poisson baseline is useful because it is transparent and converts two goal-rate inputs into a complete score distribution. Its main assumptions are:
1. Fixture-specific goal rates
Each team is summarized by its own pre-match λ rather than one universal football scoring rate.
2. Poisson marginal counts
Conditional on λ, each team’s full-time goal count is assumed to follow a Poisson distribution, so its conditional mean and variance are both λ.
3. Independent team totals
The simplest joint model multiplies the two marginal goal probabilities. Football-specific dependence can be added separately, for example with Dixon-Coles.
A homogeneous scoring process is not required for a full-time Poisson count: a deterministic time-varying intensity can still produce a Poisson total. The stronger limitation is that one full-match λ cannot describe game-state changes, red cards or other stochastic shifts unless they are modelled explicitly. For one-team Over/Under lines, exact goal counts and fair odds from a single λ, use the team-total probabilities from λ.
How Poisson calculates correct score probabilities
Two λ values — λ home and λ away — are enough to generate Poisson correct score probabilities under the independence assumption. The probability of each scoreline is the product of the two team goal probabilities:
Example with λ_home = 1.56 and λ_away = 1.18:
P(0:0) = 21.0% × 30.7% = 6.5%
P(2:1) = 25.6% × 36.3% = 9.3%
The visible heatmap shows 0–6 goals for each team. The calculator uses a larger adaptive computational grid for 1X2, BTTS and correct score probabilities, while totals use the exact Poisson distribution of total goals. Home win is the sum of cells where i>j, the draw is the diagonal, and BTTS is the probability mass outside row 0 and column 0. For the closed-form formula, Dixon-Coles effect and fair odds, use the BTTS Probability Calculator.
Where Poisson breaks down
1. Low-score dependence and local misfit
Independent Poisson can misfit the four lowest scorelines 0:0, 1:0, 0:1 and 1:1. The direction is not universally the same for every league, season or cell; the point of the Dixon-Coles correction is to alter this local dependence structure with an explicit ρ parameter.
| Score | Actual | Poisson | Gap |
|---|---|---|---|
| 0:0 (EPL) | 6.08% | 5.87% | +0.21pp |
| 1:1 (EPL) | 10.89% | 11.69% | −0.79pp |
| 0:0 (Bundesliga) | 5.61% | 4.63% | +0.98pp |
| 1:1 (Bundesliga) | 11.85% | 10.79% | +1.06pp |
The table illustrates that the residual pattern is local and league-dependent rather than a universal “all draws are underestimated” rule. Dixon-Coles is designed to redistribute probability among these four cells.
2. Non-constant scoring rate
Scoring intensity can vary with time, score state, substitutions and red cards. A full-match Poisson count can tolerate deterministic time variation, but one pre-match λ cannot represent stochastic game-state changes without an explicit dynamic model.
3. No team-specific behavioural model
Two fixtures with the same total expected goals can have different home/away splits and therefore different score distributions. Even the same λ pair cannot describe every tactical or game-state effect. Market-implied rates may reflect some contextual information through prices, but do not remove these structural limitations.
The Dixon-Coles correction
Dixon & Coles (1997) introduced τ(i,j) that adjusts four cells:
0:1 → ×(1 + λ_h·ρ) | 1:1 → ×(1 − ρ)
Negative ρ → 0:0 and 1:1 increase, 1:0 and 0:1 decrease. Shifts probability toward stalemates, away from narrow wins.
ρ varies by league
| League | Preset ρ | δ at league-average λ | Character |
|---|---|---|---|
| La Liga | −0.007 | +0.086pp | Near-zero aggregate preset |
| EPL | −0.010 | +0.117pp | Minimal correction |
| Serie A | −0.020 | +0.241pp | Moderate |
| Ligue 1 | −0.025 | +0.301pp | Moderate |
| Bundesliga | −0.096 | +1.036pp | Largest preset shift in this sample |
These are heuristic aggregate presets fitted on the ImpliedScore calibration sample, not universal league constants and not full team-level maximum-likelihood estimates. The probability moved in each affected cell is δ = −ρλhλae−(λh+λa), so raw ρ values are not directly comparable without the relevant goal rates. The table reports δ at the sample's league-average λ values; a higher-scoring environment does not by itself imply a larger correction. The historical market-calibration study describes the sample. These presets are not applied by the independent-Poisson calculator above.
Poisson vs xG: what's the difference?
Poisson requires a projected goal rate for the fixture. That estimate may come from betting odds or a model using historical xG, team strength and opponent information. A completed match's shot-based xG measures the chances already recorded; it is not automatically an appropriate pre-match λ for another fixture.
| Market-implied λ | Shot-based xG | |
|---|---|---|
| Source | Betting odds | On-pitch data |
| Timing | At the pre-match odds snapshot | Accumulates as shots occur; final after the match |
| Represents | Information reflected in market prices | Estimated quality of recorded shooting chances |
| Can miss | Anything the market misprices or has not incorporated | Non-shot information and model-specific contextual effects |
The two quantities answer different questions. Market-implied λ is a pre-match rate inferred from prices; shot-based xG is built from chance-quality estimates. Neither is automatically “more informative” for every purpose.
Is Poisson accurate enough?
Accuracy is an empirical question, not a property of the formula alone.
A Poisson matrix can reproduce a supplied λ pair exactly by construction, but that does not establish that the λ values or resulting probabilities predict future matches well. Predictive quality belongs in out-of-sample calibration, Brier and log-loss analysis.
For ImpliedScore, the practical value of the matrix is transparent probability decomposition: one pair of rates generates exact scores and related markets in a mutually consistent way. See the calibration methodology for evaluation.
Beyond Poisson: what else exists?
Dixon-Coles — changes the dependence structure at 0:0, 1:0, 0:1 and 1:1 while keeping Poisson marginals at fixed λ.
Bivariate Poisson — introduces a shared latent scoring component and broader positive dependence.
Negative binomial / other count models — change the marginal dispersion assumption when variance differs systematically from the Poisson mean.
Dynamic or event-time models — model changing team strength or within-match intensity explicitly rather than compressing the match into fixed pre-match rates.
No alternative fixes every failure mode automatically. See where Poisson football models fail for a diagnostic comparison.
Summary
| Question | Answer |
|---|---|
| What does Poisson model? | Probability of k goals given rate λ |
| Core inputs | Fixture-specific Home λ and Away λ |
| What can go wrong? | Rate misspecification, dependence, dispersion and game-state effects |
| What addresses low-score dependence? | Dixon-Coles τ-correction with ρ |
| Does ρ vary? | Yes — it is data- and method-dependent, not a universal constant |
| Does this prove an edge? | No — model structure and predictive validity are separate questions |
| What is it good for? | Transparent score distributions and internally consistent derived probabilities |
Full methodology: methodology →. To start from market prices, use the main score calculator linked above.
References
Dixon, M. J. & Coles, S. G. (1997). Modelling Association Football Scores and Inefficiencies in the Football Betting Market. JRSS: Series C, 46(2), 265–280.
Maher, M. J. (1982). Modelling Association Football Scores. Statistica Neerlandica, 36(3), 109–118.
Karlis, D. & Ntzoufras, I. (2003). Analysis of Sports Data by Using Bivariate Poisson Models. The Statistician, 52(3), 381–393.