9
Conversion
The Unwritten Manual formulas put to the test with an expected goals (xG) model on about 500k team matches.
Set piecesIFKIndirect free kickDFKDirect free kickPKPenalty kick
The model tested
P(goal) = 0.92 · att3.5 / (att3.5 + def3.5)
Example: A has 50 midfield and 50 left attack; B has 40 midfield and 40 right defence.
Possession 66.1% → chances on the left = 10 × 66.1% × 25% = 1.65 → P(goal) = 63.1% → 1.04 expected goals on the left.
Assumed fixed values: DFK 50%, non-tactical LS 60%, PK 80%.
Actual vs expected goals (total)
- L/C/R attacks score about 4% more than expected.
- DFKs, PKs and non-tactical shots score much less than the fixed values, especially with little possession (PK −75% below 25% possession).
⚠ Beware of CHPP data
In match reports many missed chances are not shown. Conversions calculated from match reports are therefore inflated: 54% overall, IFK 67.5%, PK 85.6%. Studies since 2022 use the Live dataset, which contains every event.
From the manual's rule: how to use the formula
- The attack vs defence formula needs the real ratings (CHPP), i.e. those in the match report minus 0.75. For example, attack 12 and defence 13.5 become 11.25 and 12.75.
- The scorer is chosen at random, but players with more scoring are more likely to end up on the scoresheet.
- Possession and chances: in the a3/(a3+b3) formula you can use either the ratings or the possession percentages, because only the ratio counts.