What 62% means
Imagine the same match could be played a hundred times under identical conditions. 62% means the favourite wins about 62 of them and fails to win the other 38 — a draw or a defeat. The real match is just one of those hundred, and nobody knows in advance which one.
That is why a single match says nothing about the model. If a 62% outcome does not happen, that is expected 38 times out of 100. The model is wrong when, over many matches, its 62% outcomes come true clearly more or less often than 62 times in 100.
Why a 62% favourite fails to win 4 times in 10
Our own data shows it. Across 48,148 club football matches in the model’s check, there were 5,235 where the favourite was given between 60 and 70% (64% on average). It won 65% of them and failed to win 35%: 22% were draws and 13% defeats.
Football has three outcomes, and “failed to win” includes the draw, so even a confident favourite leaves its opponent plenty of room. For every level of confidence, see why favourites lose.
Most likely does not mean “will probably happen”
The three outcomes add up to 100%. If the model gives the home side 45%, the draw 28% and the away side 27%, a home win is the most likely result — yet the chance that the home side does not win is 55%, more than half. In football the most likely outcome often fails to happen more often than it happens.
Calibration: are the percentages honest?
You cannot check a probability on one match, but you can on thousands. Group every estimate the model made — everything it gave 20–30%, 30–40% and so on — and see how often those outcomes happened. If the dots sit on the diagonal, the percentages are honest: 60% comes true 60 times in 100, not 45 and not 80.
Since 28 September 2026 the site has published estimates for 1,282 matches that have now been played. The model’s most likely outcome came true in 59% of them; its own probabilities expected 58%.
Tables for every sport and competition are on the accuracy page.
Probability versus odds
Bookmakers quote odds rather than percentages. Decimal odds turn into a probability by dividing one by them: 1.60 is 62.5%. But those “probabilities” across all outcomes add up to more than 100% — in football by 5.8% on average: that is the margin built into the price. So odds cannot be compared directly with the model’s percentage. More in our look at the bookmaker’s margin.
Forepitch only publishes the model’s estimates: we do not take bets, sell betting tips or encourage anyone to act. A gap between the model’s percentage and the odds is not a signal to do anything.
In short
- A probability is about frequency over many matches, not about one game.
- 70% does not mean “certain”: it means 30 misses in 100.
- Judge a model by its calibration over thousands of matches, not by single misses.
- The numbers change when new information arrives: line-ups, injuries, other results.
FAQ
If the model gave 70% and it did not happen, was the model wrong?
Not necessarily: a 70% outcome fails to happen in about 30 matches out of 100. Errors only show over many matches — that is what the calibration on the accuracy page shows.
How can a favourite be below 50%?
Football and hockey have three outcomes. If the home side gets 45% and the draw and the away side 27–28% each, the home side is the favourite even though the chance it fails to win is over half.
Why do the percentages change before kick-off?
The model re-estimates every 15 minutes as line-ups, injury news or other results come in.