Favourite does not mean winner
Across 48,148 club football matches in the model’s check, the most likely outcome came true in 50% of matches — roughly half. The model never promised more: its favourite was given 50% on average. When those two numbers match, the percentages are honest and favourites lose exactly as often as expected.
How often favourites of different strength fail to win
Here are the same matches split by how confident the model was in the favourite. “Under 40%” means near coin-flip games where all three outcomes had similar chances.
- Favourite won
- Model estimate
| Favourite’s chance | Matches | Model average | Won | Draw | Lost |
|---|---|---|---|---|---|
| under 40% | 10,377 | 38% | 37% | 30% | 32% |
| 40–50% | 18,307 | 45% | 45% | 29% | 26% |
| 50–60% | 11,085 | 54% | 55% | 26% | 19% |
| 60–70% | 5,235 | 64% | 65% | 22% | 13% |
| 70–80% | 2,209 | 74% | 75% | 16% | 9% |
| 80%+ | 935 | 85% | 86% | 10% | 4% |
Even a clear favourite at 70–80% fails to win about 25% of matches — one in 4.
At 80% and above it is 14%: of 935 such matches, the favourite failed to win 129.
Tennis and the NBA: no draws, and favourites still lose
Even a player given 90% or more loses 5% of matches: an injury mid-match, a bad day on serve, a dangerous qualifier.
| Favourite’s chance | Tennis: favourite lost | NBA: favourite lost |
|---|---|---|
| 50–60% | 45% | 43% |
| 60–70% | 34% | 36% |
| 70–80% | 26% | 24% |
| 80–90% | 18% | 15% |
| 90%+ | 5% | 6% |
Why it happens
- Goals are scarce: one lucky strike, a deflection or a red card can decide a match.
- The model does not know everything: motivation, weather, trouble inside a club, an injury during the game.
- Above all, 75% simply means 25 defeats in 100. If 75% favourites never lost, it would mean the model had been underrating them.
Upsets of recent weeks
We call it an upset when a side the model gave 30% or less wins (draws do not count). Over the last 3 weeks, across the 945 matches we cover, that happened 71 times. The model’s own probabilities expected about 90.
The biggest surprises:
Every upset week by week is in Upsets of the week.
What follows from this
- One upset says nothing about the model. What matters is how many there are over time and whether that matches the percentages.
- Check frequencies, not single matches: every check is on the accuracy page.
- How to read the percentages: how to read probabilities.
FAQ
How often do favourites lose in football?
Across 48,148 club football matches in the model’s check, the most likely outcome came true in 50% of matches. A favourite at 70–80% failed to win about 25% of the time, draws included.
Does an upset mean the model was wrong?
No, as long as there are as many upsets as the model expected. Over the last 3 weeks there were 71 against 90 expected.
What counts as an upset on Forepitch?
A win by a side the model gave 30% or less before the match. Draws do not count: in football they are often at 25–30%, and every other result would be an upset.