## Beyond Independent Poisson
The standard Poisson model for football treats each team's goal scoring as independent. This is a simplification — the number of goals each team scores are not truly independent (a goal changes game dynamics, affecting subsequent goal probability).
## The Dixon-Coles Correction
Dixon and Coles (1997) introduced a correction factor τ for low-scoring outcomes (0-0, 1-0, 0-1, 1-1), which are under/over-represented in standard Poisson models:
P(score = (x, y)) = τ(x, y, λ_home, λ_away, ρ) × Poisson(x; λ_home) × Poisson(y; λ_away)
Where τ = 1 − ρλ_home λ_away (for x=0, y=0)
= 1 + ρ (for x=1, y=1)
= 1 + ρλ_away (for x=1, y=0)
= 1 + ρλ_home (for x=0, y=1)
= 1 (otherwise)
ρ is typically estimated at −0.13 to −0.10.
## Why This Matters for Bettors
The Dixon-Coles correction improves probability estimates for low-scoring games. If your model uses pure Poisson without this correction, it:
- Underestimates 0-0 probability by 15–20%
- Overestimates 1-1 probability by 5–10%
These distortions affect over/under markets and correct score markets systematically.
## Implementing the Correction
The correction is straightforward to implement in Excel or Python. The adjusted score probability matrix gives more accurate probability estimates for all market types derived from the scoreline distribution.
Most serious football modellers use Dixon-Coles as the standard. If you are not using it, you are working with a systematically biased probability distribution for low-scoring outcomes.
Create a free account to track your progress and save bookmarks.
