Probability vs Prediction
These two words are often used interchangeably, but they mean very different things. A prediction claims to know what will happen: "Team A will win." A probability describes likelihood: "Team A wins in roughly 45% of matches with this profile."
The distinction matters because football is inherently random. Even when one team is clearly stronger, upsets happen regularly. A 70% favourite still loses 3 out of 10 times. Anyone who presents a football "prediction" as certain is being dishonest about the nature of the sport.
xGoal's philosophy: Probability ranges to inform, not predictions to persuade. Match balance, not outcome predictions.
How Match Probabilities Are Calculated
Match probability models work by combining multiple inputs — team strength metrics, form, venue, squad availability, and more — into a structured assessment. The general process follows three steps:
Gather signals
Collect data on both teams: recent form, defensive strength, chance creation, squad availability, venue history, pressing intensity, and more. Each signal provides a different lens on team quality.
Weight and combine
Different signals matter more in different contexts. Venue matters more when the home team has a strong record. Form matters more when the sample is recent. The model weights each signal based on its reliability and relevance.
Express as ranges
Rather than outputting a single number, honest models express uncertainty as a range. When data is strong and signals agree, the range is narrow (38–42%). When data is limited or signals conflict, the range widens (25–55%).
Why Probability Ranges, Not Single Numbers?
Most platforms show a single number: "Home Win: 52%." This creates a false sense of precision. The true probability depends on information that might not be available: unreported injuries, tactical changes, player confidence, weather conditions.
A probability range honestly communicates this uncertainty. "Home Win: 45–55%" tells you the home team has an advantage, but it's not overwhelming. "Home Win: 60–70%" tells you the advantage is clearer and the data is more reliable.
The width of the range is itself meaningful. A narrow range means the model is confident in its assessment. A wide range means there's genuine uncertainty — and that's important to know before forming an opinion.
What Affects Confidence?
High Confidence (narrow range)
- All signals have fresh data
- Signals mostly agree on direction
- Large sample of recent matches
- No major unknown factors
Low Confidence (wide range)
- Missing or stale data
- Signals contradicting each other
- Early season (small sample)
- International break gaps
xGoal's 29-Signal Approach
xGoal combines 29 distinct signals to generate probability ranges for every match. Each signal examines a different dimension of team quality, and the ranges honestly reflect how certain the data allows us to be.