## When Poisson Is Not Enough
The Poisson distribution has one parameter (λ) and assumes that the mean equals the variance. In practice, sports goal scoring often has variance greater than the mean — a phenomenon called over-dispersion.
## Why Over-Dispersion Occurs
Over-dispersion in goals arises because:
1. Match quality varies significantly — some matches are high-intensity, others are tactical affairs
2. Weather and pitch conditions affect scoring rates
3. In-game state changes (red cards, early goals) shift the effective goal rate during the match
These factors make the true variance of goals scored across matches larger than the Poisson mean.
## The Negative Binomial Distribution
The negative binomial distribution extends Poisson by adding a second parameter (the dispersion parameter r) to capture over-dispersion:
P(X = k) = C(k+r−1, k) × (r/(r+μ))^r × (μ/(r+μ))^k
As r → ∞, the negative binomial approaches Poisson.
## When to Use Negative Binomial
If your data shows that the variance of goals across matches significantly exceeds the mean: use negative binomial. Compare:
- Mean goals per match: 2.7
- Variance of goals per match: 3.8 (vs 2.7 expected for Poisson)
Variance > mean: over-dispersion present → negative binomial more appropriate.
## The Practical Impact
Using negative binomial instead of Poisson in over-dispersed data:
- Assigns more probability mass to extreme totals (0 goals, 5+ goals)
- Assigns slightly less to the central range (2–3 goals)
- Produces better-calibrated over/under probabilities for extreme lines
For most applications, the difference is small. For bettors specifically interested in high-goals or low-goals markets, negative binomial is worth implementing.
Create a free account to track your progress and save bookmarks.
