Terminal Models

Fairness Audit

Model AModel B

Overlay Model B

Within-league signed gap · Model A

+0.093

95% CI +0.063 to +0.124 · about +10%

Model A+0.093

Model B+0.164

A − B−0.071

Model A − Model B: −0.071
Accuracy gap (MAE): +0.011
Held-out players: 5,820

01 / 07 · The gap

Are non-UEFA players' errors different from UEFA players'?

Residuals by groupmean residual and gaps, log scale, 95% CI

UEFA playersmean residual−0.019
Non-UEFA playersmean residual+0.078
Gap, rawnon-UEFA minus UEFA+0.097
Gap, within leaguesprimary measure+0.093

Residual = actual − predicted. Right of zero: under-predicted

Show as a table

Mean residuals and gaps
Row	Estimate	95% CI low	95% CI high
UEFA players (mean residual)	−0.0188	−0.0341	−0.0028
Non-UEFA players (mean residual)	0.0780	0.0516	0.1043
Gap, raw (non-UEFA minus UEFA)	0.0968	0.0666	0.1264
Gap, within leagues (primary measure)	0.0933	0.0629	0.1240

Same accuracy, opposite directionheld-out, by group

Mean absolute error

UEFA0.403
Non-UEFA0.414

Mean signed residual

UEFA−0.019
Non-UEFA+0.078

Both groups are missed by about the same amount (MAE gap +0.011, interval includes zero), but in opposite directions: UEFA players slightly over-predicted, non-UEFA players under-predicted.

What moves the groups' predictions apartmean SHAP, non-UEFA minus UEFA, served Model A

sums to +0.253 log units

Age+0.064
League+0.064
Days since last match+0.024
Minutes (last 365 days)+0.021
Club points per game+0.020
Goal contributions per 90+0.020
Goals per 90+0.013
Minutes per appearance+0.005
Role+0.005
Minutes at current club+0.005

The served model predicts non-UEFA players +0.253 log units higher on average; these are the inputs that account for that difference. They describe the model's arithmetic, not causes, and they are about predictions, not about the gap in errors above.

Under Model A, non-UEFA players' market values sit about +10% above the model's estimate relative to UEFA players in the same leagues (about €362k per player, a figure dominated by expensive players). Removing league and club moves the gap from +0.093 to +0.164: context explains part of it; a clear gap remains.

Row

Estimate

95% CI low

95% CI high

UEFA players (mean residual)

−0.0188

−0.0341

−0.0028

Non-UEFA players (mean residual)

0.0780

0.0516

0.1043

Gap, raw (non-UEFA minus UEFA)

0.0968

0.0666

0.1264

Gap, within leagues (primary measure)

0.0933

0.0629

0.1240

Fairness Audit

Are non-UEFA players' errors different from UEFA players'?

Residuals by groupmean residual and gaps, log scale, 95% CI

Same accuracy, opposite directionheld-out, by group

What moves the groups' predictions apartmean SHAP, non-UEFA minus UEFA, served Model A

How to read the observatory

Fairness Audit

Are non-UEFA players' errors different from UEFA players'?

Residuals by groupmean residual and gaps, log scale, 95% CI

Same accuracy, opposite directionheld-out, by group

What moves the groups' predictions apartmean SHAP, non-UEFA minus UEFA, served Model A

How to read the observatory