When an estimate says a team has a 45% chance of winning, what exactly has been claimed? Not that the team will win, and not that it will lose. The statement is about how things would play out if the same situation could be repeated many times.
A frequency, not a promise
Imagine the same match, with exactly the same conditions, being played 100 times. A 45% estimate means the team would be expected to win in roughly 45 of those. In the other 55 it would draw or lose. No single match is ever replayed, of course, so we never see the proportion directly; we see just one outcome.
That is why a single result cannot prove a probability right or wrong. If a die shows a six, nobody concludes that "one in six" was a mistake.
An event at 30% is not rare
People often treat 30% as "almost impossible". It is not. Roll a die and ask how often it lands on one or two: that is about 33%, and it happens all the time. In a typical weekend with many matches, several outcomes that were given 20% to 30% will occur. If none ever did, the estimates would be badly calibrated.
"The model got it wrong"
This phrase usually mixes up two different things. A probability model can be wrong, but you can only tell by looking at many predictions together and checking whether, say, events given 30% happen about 30% of the time. One surprising result tells you almost nothing. When a team given a 15% chance wins, the likeliest explanation is not a broken model; it is that 15% events do occur.
Reading three numbers together
Predicta shows three probabilities: team A wins, draw, team B wins. They always sum to 100%, so they should be read as a set.
- Roughly 34 / 33 / 33 describes an almost evenly balanced match.
- Something like 55 / 25 / 20 shows a clear lean, yet the other two outcomes together still hold 45%.
- Watch the draw figure; it is rarely the top number, but it is rarely tiny either.
The aim is to appreciate the shape of the uncertainty, not to extract a single pick.
Key takeaways
- A probability describes how often something would happen over many repetitions.
- Events at 20% to 30% occur regularly and are not shocking.
- A single surprising result does not show that a model failed.
- Read the three outcome probabilities as one picture.