Probability, decimal odds and sample size in football predictions
A projection, a probability and decimal odds describe different things. Reading them separately helps avoid false certainty.
A projection estimates a quantity
A projection of 4.2 shots on target is an estimated quantity. It does not promise exactly four attempts and does not by itself give the probability of exceeding a line.
Different distributions can share the same average while having different variability. A team projection and a match-total projection also answer different questions.
A probability is uncertain
An estimated 75% probability is the model’s assessment of a specific condition, not a confirmed success rate. Evaluating it requires many predictions recorded before kickoff and compared with later outcomes.
Three correct predictions from four examples do not establish that future selections will succeed at that rate.
Read decimal odds separately
For decimal odds, the reciprocal gives the implied probability before accounting for the market’s margin. Illustrative example: odds of 2.00 imply 50%; this is arithmetic, not evidence that the event truly has a 50% chance.
Compare definitions, settlement rules and the current offered price before comparing probabilities.
Observed and effective sample sizes
The number of recorded matches is not necessarily the effective sample after weighting. Recent matches, competition context and other enabled similarity factors can receive different weights.
Check which metric has data in each match. Missing observations are not zeros, and combining different metrics does not create a larger valid sample.
Minimum odds and evaluation
The model’s minimum odds are a comparison threshold based on its estimate and margin. They are not a bookmaker offer or a guarantee of profit. Model error and losses remain possible above that threshold.
Judge a method using recorded pre-match selections, actual available prices and later outcomes. Accuracy alone does not demonstrate profitability.