Soccer Score Simulator (Poisson) illustration

Soccer Score Simulator (Poisson)

Models both teams’ goals as independent Poisson distributions around their expected-goals inputs and computes every scoreline probability: win/draw/loss chances, the most likely exact scores, over/under 2.5 goals and both-teams-to-score. The classic baseline model behind football analytics, with an optional Dixon-Coles low-score correction that fixes plain Poisson’s known bias in 0-0, 1-0, 0-1 and 1-1 games.

Dixon-Coles (1997) adjusts only the 0-0, 1-0, 0-1 and 1-1 cells; every other scoreline and both goal distributions stay exactly Poisson.
-0.1
Negative ρ shifts probability into 0-0 and 1-1 and out of 1-0 and 0-1. League fits typically land between −0.05 and −0.15; Dixon-Coles’ own estimate was about −0.06.

Notes

  • Real scores are slightly correlated: large-sample studies (Dixon & Coles 1997, Karlis & Ntzoufras 2003) find plain Poisson underrates 0-0 and 1-1 draws a touch, which is exactly what the correction repairs.
  • The correction is zero-sum, so totals like over/under 2.5 goals and each team’s goal distribution are unchanged — only the draw/win split at low scores moves.
  • Get expected goals from shot-based xG models, or estimate them from average goals scored and conceded.
  • Educational estimate for sport and training purposes, not professional coaching or medical advice.