Why Do Soccer Models Adjust the Odds of 0-0 and 1-1 Draws?
Dixon-Coles multiplies four cells of the scoreline matrix, 0-0, 1-0, 0-1 and 1-1. Here is what that does to a draw price, worked through one match.
Updated Sep 2026 · Part of the models series
A plain Poisson model treats the two teams’ goal counts as independent draws, and in low-scoring matches they are not. Dixon-Coles corrects it by multiplying exactly four cells of the scoreline matrix, 0-0, 1-0, 0-1 and 1-1, by a factor built from one fitted parameter called rho.
Why does independence break down at low scores?
Because both teams’ goal counts answer to the same match conditions. Two sides sitting deep on a wet pitch suppress each other at once, and independence has no way to represent that.
The effect only shows up where the scores are small. At 3-2 there have been enough scoring chances that the independence assumption does little damage. At 0-0 and 1-1 it does, and a plain Poisson model underprices draws as a result. Michael Maher’s 1982 paper set out the independent-Poisson model, and Mark Dixon and Stuart Coles’ 1997 paper added the correction that fixes its low-score corner.
What exactly does the correction multiply?
Four cells and nothing else. It raises 0-0 and 1-1, lowers 1-0 and 0-1, and leaves every other scoreline where plain Poisson put it.
Write the home team’s expected goals as lambda, the away team’s as mu, and the fitted parameter as rho. The four factors are 1 − lambda × mu × rho at 0-0, 1 + mu × rho at 1-0, 1 + lambda × rho at 0-1, and 1 − rho at 1-1. Rho is negative, so the first and last factors come out above 1 and the middle two come out below it. The matrix is renormalized afterwards, which is how the probability taken out of the narrow wins arrives in the draws.
How much does it move a draw price?
Between two and three percentage points, which is worth about 24 cents on a fair draw price. The example below runs one match through the model twice, once with the correction switched off.
A 1.7-point shift on a 25% outcome sounds small until it is priced. Twenty-four cents of movement on a three-way market is larger than the margin many books post on the same match.
What this does not tell you
The correction touches four cells, so a 2-2 draw gets nothing from it. Rho is refitted from finished matches, and it moves. The 2026-05-16 refit took it from -0.13 to -0.07 across 690 matches. Getting draws right is calibration, not an edge, and Fairline has run no forward preregistered comparison of its Premier League model against a sharp closing line. The correction also says nothing about which side to back. It changes the shape of the distribution, and the current parameter value is rendered live on the EPL model page.