Ever wondered how prediction sites calculate win probabilities? Here's the math.
Originally designed for chess by Arpad Elo. Each team has a rating (average ~1500). Win against a strong team = big rating boost. Lose to a weak team = big drop. FiveThirtyEight used Elo for their famous NBA and football predictions. Simple, elegant, and surprisingly accurate.
Every shot in football is assigned a probability based on: distance from goal, angle, body part, assist type, game state, and defensive pressure. Machine learning models trained on 300,000+ shots. The best models (StatsBomb, Opta) achieve ~0.12 log-loss. xG is now the standard for evaluating team and player performance.
Run the season 10,000 times with random variation. Each simulation uses team strength ratings + home advantage + schedule difficulty. The percentage of simulations where a team wins the title = their title probability. This is how FiveThirtyEight, The Athletic, and our prediction calculator work.
Advanced plus-minus models that estimate how many points a player adds per 100 possessions. They use box score stats + tracking data + on/off court splits. RAPTOR (FiveThirtyEight) and EPM (Dunks & Threes) are the gold standard for NBA player evaluation.
Football scores follow a Poisson distribution — the probability of scoring X goals given an expected rate. If a team's xG is 1.8, the probability of scoring exactly 2 goals is 26.8%. This is the mathematical foundation of match prediction and is used by every major prediction model.