Methodology, measured

Model Calibration Lab

A record of picks tells you what won. Calibration tells you whether the probabilities meant anything: when we say 60%, does it happen 60% of the time? Below, every settled prediction, scored two ways — the sharp market (devigged) and our own model — with the uncomfortable comparison in plain sight.

Settled predictions 1,353
Brier score 0.588

Our model (Elo + form): 0.611

Log-loss 0.986

Our model (Elo + form): 1.020

Calibration error 1.4pp

Our model (Elo + form): 1.8pp

Predicted probability vs observed frequency

Each point is a 10-point probability band (min. 30 observations). The closer to the diagonal, the better calibrated. Every match contributes its three outcomes.

Sharp market (devigged) Our model (Elo + form) perfect calibration
0% 0% 25% 25% 50% 50% 75% 75% 100% 100% Our model (Elo + form) · 10–20%: 16.6% → 13.5% (n=303) Our model (Elo + form) · 20–30%: 26% → 24.9% (n=1857) Our model (Elo + form) · 30–40%: 34.6% → 33.8% (n=786) Our model (Elo + form) · 40–50%: 44.8% → 48.6% (n=642) Our model (Elo + form) · 50–60%: 54.2% → 57.3% (n=309) Our model (Elo + form) · 60–70%: 64% → 62.4% (n=101) Sharp market (devigged) · 0–10%: 7.4% → 6.3% (n=79) Sharp market (devigged) · 10–20%: 16.2% → 16.9% (n=508) Sharp market (devigged) · 20–30%: 25.2% → 23.8% (n=1528) Sharp market (devigged) · 30–40%: 34.3% → 33.5% (n=862) Sharp market (devigged) · 40–50%: 45% → 46.9% (n=480) Sharp market (devigged) · 50–60%: 54.8% → 57.8% (n=325) Sharp market (devigged) · 60–70%: 64.5% → 64.8% (n=182) Sharp market (devigged) · 70–80%: 74.4% → 80% (n=70)
Band Sharp market (devigged) — Predicted (avg) Observed Obs. Our model (Elo + form) — Predicted (avg) Observed Obs.
0–10% 7.4% 6.3% 79
10–20% 16.2% 16.9% 508 16.6% 13.5% 303
20–30% 25.2% 23.8% 1528 26% 24.9% 1857
30–40% 34.3% 33.5% 862 34.6% 33.8% 786
40–50% 45% 46.9% 480 44.8% 48.6% 642
50–60% 54.8% 57.8% 325 54.2% 57.3% 309
60–70% 64.5% 64.8% 182 64% 62.4% 101
70–80% 74.4% 80% 70

Lower Brier and log-loss are better; calibration error is the observation-weighted average gap between predicted and observed. The honest headline: the devigged sharp market is better calibrated than our model — which is precisely why this site prices everything against the market instead of selling you model predictions. Methodology: probabilities as displayed at logging time, no retro-fitting; outcomes from official results.

Full methodology →
มีประโยชน์ไหม? ส่งให้เพื่อนที่เล่นพนัน