Every probability on WinnScore comes from one model. Here is exactly what it does, what it cannot do, and why you should treat its output as an estimate rather than a forecast you can rely on.
1. Team ratings from this season
For every league we collect each finished match this season. A team’s attacking rating compares the goals it scores with what an average side scores at the same venue; its defensive rating does the same for goals conceded. Recent games count more: a result loses half its weight every sixty days. Early in the season ratings are pulled towards average, because five games say much less than twenty.
2. Expected goals
Home expected goals = the league’s average home goals × home attack × away defence. Away expected goals are built the same way. In European cups we use each club’s domestic rating.
3. Bookmaker prices
Prices carry information a results table cannot see, such as injuries and rotation. Where a match has prices, we remove the bookmaker’s margin, work out the goal expectancy the prices imply and blend it with our own figure, with slightly more weight on the market. Matches with no usable results use the market alone, and the lean strength is reduced.
4. Every scoreline
Expected goals feed a Poisson model of each side’s goals, with the Dixon-Coles adjustment for 0–0, 1–0, 0–1 and 1–1, results a plain Poisson model gets wrong. That gives a probability for every score up to 9–9. Win, draw, goal lines, both teams to score, clean sheets, double chance and draw no bet are all added up from the same grid, so they are consistent with each other.
5. Halves
We split expected goals between the halves using the share of goals each league has actually produced before half-time this season, shown on our goal timing page.
6. Value
Edge = our probability × decimal odds − 1. A positive edge means the price is longer than our estimate says it should be. Big edges usually mean we are missing news, not that the bookmaker has made a mistake.
7. Lean strength
The one-to-five pips show how clearly one result stands out, reduced when the data behind it is thin.
Limits
A model is a simplification. Ours does not know about a late injury, a manager’s team selection or a dead rubber unless the market has priced it in. A 70% chance still fails three times in ten. Read our disclaimer before using any number on this site.
Data
Fixtures, scores, tables, head-to-head records and reference prices come from publicly available sports data and refresh up to every minute. Data can be delayed or wrong; tell us at [email protected] if you spot an error.