Statistical Models
Learn to build statistical models that systematically estimate match probabilities — from Elo ratings to Poisson distributions.
Duration
14m
XP Reward
+80
Status
Active
Learning Objectives
- Understand the main types of predictive models (Elo, xG, Poisson, regression)
- Learn how the Poisson distribution models goal-scoring
- Recognize model limitations and the danger of overfitting
- Build a simple xG-based model as a starting point
Quick Summary
- → Statistical models combine data into systematic probability estimates
- → Poisson models goal-scoring; Elo rates team strength; regression combines factors
- → Models have limitations: correlation, context blindness, overfitting
- → Calibration is essential — predicted probabilities should match actual outcomes
Building Predictive Models
A statistical model is a systematic framework for estimating match probabilities. It combines multiple data inputs into a single probability estimate.
Model Types
- Elo ratings: Originally for chess, adapted for football. Teams gain/lose rating points based on results. The rating difference predicts win probability.
- xG-based models: Use expected goals created and conceded to estimate team strength and match probabilities.
- Poisson models: Model goal-scoring as a Poisson process. Predict the probability of specific scorelines.
- Regression models: Combine multiple factors (xG, form, venue, rest) using logistic regression.
The Poisson Model
The Poisson distribution models the probability of a given number of events (goals) occurring in a fixed interval. If a team averages 1.5 goals per game, the Poisson model gives:
- P(0 goals) = 22.3%
- P(1 goal) = 33.5%
- P(2 goals) = 25.1%
- P(3 goals) = 12.6%
By modeling both teams' goal expectations, you can calculate the probability of every scoreline and derive win/draw/loss probabilities.
Model Limitations
- Correlation: Poisson assumes teams' scoring is independent — but in reality, if one team scores more, the other often scores less (defensive reaction)
- Context blindness: Models don't account for motivation, fatigue, tactical changes
- Data quality: Models are only as good as the data feeding them
- Overfitting: Complex models can fit past data perfectly but fail on new data
Model Calibration
A well-calibrated model produces probabilities that match reality:
- When your model says 60%, the actual win rate should be ~60%
- Track calibration over 100+ predictions
- If your model says 60% but actual win rate is 45%, your model is miscalibrated (overconfident)
Building Your First Model
Start simple:
- Collect xG data for both teams (last 10 games)
- Calculate attacking and defensive ratings
- Estimate expected goals for each team in the matchup
- Apply Poisson to derive scoreline probabilities
- Compare to market odds — identify value
Your model won't be perfect. But it provides a systematic baseline that removes emotional bias from your estimates.
Key Takeaways
Statistical models combine data into systematic probability estimates
Poisson models goal-scoring; Elo rates team strength; regression combines factors
Models have limitations: correlation, context blindness, overfitting
Calibration is essential — predicted probabilities should match actual outcomes
Knowledge Check
Question 1 of 2
What is overfitting?
Glossary
Elo Rating
A rating system where teams gain/lose points based on results, with rating differences predicting win probability
Poisson Distribution
A probability distribution modeling the number of events (goals) in a fixed interval, widely used in football modeling
Model Calibration
When a model's predicted probabilities match actual outcomes — e.g., 60% predictions win ~60% of the time
Overfitting
When a model fits historical data too closely, capturing noise rather than signal, and fails on new data
Lesson Summary
Statistical models combine data into probability estimates. Types include Elo, Poisson, and regression. Calibration and avoiding overfitting are essential.
Related Lessons
This lesson is for educational purposes only. Sports markets involve financial risk and outcomes are never guaranteed. Never wager money you cannot afford to lose.