20 Aug 2026
How to turn a football prediction into realistic 1X2 probabilities
A good prediction does not end with “home win” or “evenly matched”. It should provide probabilities for a home win, draw and away win, acknowledge uncertainty, and be tested against actual results and market prices.
Start with three probabilities that add up to 100%
In the 1X2 market, there are three mutually exclusive outcomes: a home win (1), a draw (X) and an away win (2). A useful prediction might assign 48% to the home team, 29% to the draw and 23% to the away team. The total must be 100%. Avoid converting vague statements directly into numbers: “favourite” could mean 55%, but it could also mean 70%. Set your criteria in advance, such as adjusted historical strength, recent form in context, confirmed absences, home advantage and the fixture schedule.
Use a baseline estimate before adjusting for the match
The baseline estimate should come from comparable data. Strength ratings such as Elo, results from multiple seasons with greater weight given to recent matches, and expected goals (xG) are usually more informative than a short run of scorelines. Then adjust only for factors with a measurable effect: the absence of a key player, long-distance travel, less rest or a clear change in squad strength. If a team has a higher rating, is playing at home and has retained its squad, its win probability increases. Do not count similar arguments twice: a strong campaign, a high points total and a positive goal difference may all describe the same performance.
Include the draw and address uncertainty explicitly
The draw is not simply whatever remains after selecting the two winners. In football leagues, it occurs frequently and tends to become more likely in evenly matched games, low-scoring contests or meetings between cautious teams. Even a well-built model makes mistakes because of red cards, in-game injuries, penalties and normal sporting variation. Small differences therefore do not justify strong conclusions. A projection of 48%-29%-23% says that a home win is the most likely outcome, not that the home team will win. If the data is incomplete, move the probabilities closer together instead of inventing precision.
Calibration shows whether the percentages match reality
Calibration answers a simple question: do 60% predictions win about 60% of the time? To test it, record each prediction before the match and group results into ranges such as 50%-55%, 55%-60% and 60%-65%. In a group of 100 matches predicted at 60%, around 60 wins would be consistent; 45 or 75 would indicate miscalibration. Also use metrics such as the Brier Score, which penalises the gap between the predicted probability and the observed result. Picking the winner is useful, but it is not enough: a model that always assigns 51% may get plenty right and still be poorly calibrated.
Sample size limits confidence
Ten or 20 matches are almost never enough to conclude that a method works. A team may win six of 10 matches even when its true probability of winning each game was close to 50%. To assess calibration, hundreds of predictions provide a more stable picture; thousands improve analysis by probability range. Splitting the data by competition, season and type of favourite further reduces the available sample. Record the date, the information known before kick-off and the result. Never assess a prediction using news or lineups that only emerged afterwards.
Compare with market odds without confusing price and probability
Decimal odds can be converted into implied probability using the formula one divided by the odds. Odds of 2.00 imply 50% before market costs; odds of 4.00 imply 25%. In the 1X2 market, the sum of these probabilities usually exceeds 100% because of the bookmaker's margin. For a simple comparison, divide each implied probability by the sum of the three. If the odds imply 52%, 30% and 24%, the total is 106%; the normalised probabilities are approximately 49.1%, 28.3% and 22.6%. The market aggregates public information and reacts quickly to lineups and injuries, but it is not the final truth. The comparison is useful for identifying discrepancies and reviewing assumptions, not for automatically validating a prediction.
Related pre-match analysis
Keep reading
Analysis: PK Sport · our methodology
Analysis based on public data and market signals. For analysis only — not betting advice.