Ask a fan to predict a match and they will usually give a scoreline: "2-1, easily." It sounds precise and satisfying. Statistically, it is one of the hardest things you can ask of any model.
Three outcomes versus dozens of scores
There are only three possible results in a match: home win, draw, away win. Even a rough estimate has a decent chance of landing on the right one, and the three probabilities always add up to 100%.
Scorelines are a different story. In practice a match can finish 0-0, 1-0, 2-1, 3-2, 4-0 and so on. Consider just the scores where each side scores between zero and five goals: that is already 36 distinct possibilities, and rarer scores exist beyond that. The probability has to be shared out among all of them, so each individual score receives a small slice.
Why the most likely score is still unlikely
This is the point that surprises people most. Suppose a model says the most probable score is 1-1. It may give that scoreline a probability of around 11%. Nothing is wrong with the model: 1-1 is simply the biggest slice of a pie that was cut into many pieces. It also means that roughly nine times out of ten, the match will not finish 1-1.
Predicta displays the most probable score because it is a useful summary of the model's picture, but it should be read as "the single likeliest option among many unlikely ones", not as a promise.
Which measure deserves more attention
For this reason, the result is the more meaningful target. If a model repeatedly points to the correct outcome over a large number of matches, that suggests it is capturing something real. Exact scores will be hit far less often, even by an excellent model, simply because of the arithmetic of spreading probability across so many outcomes.
It also helps to look at all three result probabilities rather than only the top one. A match with 40% / 30% / 30% is genuinely open; a match with 70% / 20% / 10% leans clearly one way. The headline pick hides that difference.
When you read an estimate, a good question is not "what will the score be?" but "how is the probability distributed, and how confident is the model really?"
Key takeaways
- A match has three possible results but dozens of plausible scorelines.
- Probability spreads thinly across scores, so even the likeliest one usually has a modest chance.
- The correct result is a far more realistic measure of a model than the exact score.
- Read all three result probabilities, not just the leading one.