How the numbers are made.
Where the data comes from, what the model does with it, and how we check whether it was right. GoalMetrix is informational: we don't take bets and we don't give staking advice.
- competitions covered
- 56
- competitions covered
- past matches replayed to check us
- 8,121
- past matches replayed to check us
- stats per side, each half too
- 5
- stats per side, each half too
The short version
From raw data to every number you see
- Step 1CollectResults, team and player stats, lineups, injuries, prices.
- Step 2RateEach side's attack and defence, home and away.
- Step 3ExpectExpected goals, corners, shots and cards for this match.
- Step 4PriceEvery score and count turned into chances.
- Step 5CheckReplayed daily against what actually happened.
Step 1
Where the data comes from
- Match statisticsResults, team and player stats from a licensed data provider, every minute while games are live.
- Lineups and team newsConfirmed lineups, injuries and suspensions from the same provider.
- Market pricesPrices from major sportsbooks, one input to expected goals where available.
- WeatherConditions at the stadium for kickoff time.
Step 2 and 3
Expected goals: the number everything else is built on
Each team gets an attacking rate (the chances and goals it creates) and a defensive rate (what it allows), home and away, built mostly from expected goals and partly from real goals.
This season is blended with each team's level last season, worth eight games, so a hot start over a few games can't run away with the numbers. Rest and fixture congestion can nudge it, never more than 10% either way.
A side that creates more than average, facing one that allows more than average, gets more than either alone would suggest.
When a sharp sportsbook has priced the match, we blend in the expected goals its prices imply. The market knows things a season of stats can't, such as late team news.
Step 4 · Goals
From expected goals to every score
Expected goals become a table of every possible score using a Dixon-Coles model, the standard football model that corrects for how often low-scoring draws happen. Win, draw, both-teams-to-score and over/under all come from that one table, so they always agree.
| Home goals down, away goals across | 0 | 1 | 2 | 3 | 4 |
|---|---|---|---|---|---|
| 0 | 7.7 | 6.4 | 4.1 | 1.5 | 0.4 |
| 1 | 10 | 13 | 6.5 | 2.4 | 0.7 |
| 2 | 8.6 | 9.5 | 5.2 | 1.9 | 0.5 |
| 3 | 4.6 | 5.0 | 2.8 | 1.0 | 0.3 |
| 4 | 1.8 | 2.0 | 1.1 | 0.4 | 0.1 |
Home goals down, away goals across, chance of each score in %
That's why the predicted score is so often 1-1 or 2-1. It's the single most likely cell, even when one side is clearly better, so we always show expected goals next to it.
Step 4 · Counts
Corners, shots and cards
Each side's count combines what it usually makes, what the opponent usually allows, and home or away, all compared with the league. Cards also use the referee's record.
These counts swing more than goals, so we use a negative binomial model, and every number comes with a range.
7 to 13 corners covers about 68% of games like this. On match pages, a result inside the range counts as in range.
Europe and cups
Few games? We use the league games too
A team plays only a handful of Champions League or cup games, so we also use its domestic league games, rescaled to the competition and adjusted for how strong each league is.
- Premier League100
- La Liga95
- Bundesliga93
- Primeira Liga84
- Eredivisie83
- Liga MX80
- MLS75
A side scoring two a game in a weaker league isn't treated like one doing it in the Premier League.
Before kickoff
Lineups and key players
- Expected lineupsEach XI predicted from recent starts, minutes, injuries, suspensions and the next game. Official lineups replace it.
- With and withoutHow corners, shots and goals change without each regular, after allowing for the opponents. Shown now; built in only once it proves itself.
Step 5
How we check ourselves
- Daily replaysRecent matches replayed with only what was known before kickoff. 60% should happen about 6 times in 10.
- Every matchAfter full time, each match page puts what we said next to what happened, misses included.
- Tested changesChanges to the model go through the same replays before they're switched on.
Limits
What no model can know
- Red cardsOne decision changes a match.
- Early injuriesA key player off after ten minutes.
- Surprise lineupsRotation nobody saw coming.
- MotivationA side with nothing to play for.
A 70% chance still fails 3 times in 10. Use the numbers as context for your own judgment, never as a promise.
Questions
Common questions
Is this betting advice?
How often do the numbers update?
Why do your chances differ from a sportsbook's?
Why is the predicted score often 1‑1 or 2‑1?
Do you show the times you were wrong?
See it on a real match
Today's free game has every number open, with the reasons behind each one.