About Numbers for Goalposts

CANTONA FPM (v3.3)

CANTONA (Contribution Above Normalised Team-adjusted Offensive-context Aggregate) is our in-house Football Player Metric — a Box-Plus-Minus style rating estimating how many non-penalty expected goal difference (npxGD) per match a player contributes above the average player at their position.

• ·Trained on 231 EPL team-seasons (2014/15–2025/26) of Understat data
• ·Features are league-season normalised: each per-90 stat is expressed as the player's deviation from that season's league mean, removing team-quality inflation from the inputs
• ·New in v3.3: finishing residual feature (npG minus npxG per 90, league-normalised) isolates shot conversion ability from shot volume; xG-chain involvement removed after negative LOSO coefficient
• ·Team constant adjustment (from CPM/BPM): each team's minutes-weighted average CANTONA is forced to equal the team's actual npxGD, redistributing residual credit for pressing, defensive shape, and set pieces not captured by individual stats
• ·Per-season position centring: 0.0 = average for that position in that season; elite seasons typically range +0.5 to +1.5
• ·Continuous √(minutes/900) shrinkage: players below 900 minutes are progressively regressed toward zero rather than hard-excluded, converging to full weight at a sustained season
• ·Separate O-FPM (offensive contribution) and D-FPM (defensive contribution) components; composite is their sum
• ·Scores are EPL-referenced: a player in any future added league will have their score expressed in EPL-equivalent terms via a league quality offset

Interpretation: a CANTONA of +0.30 means that, per match played, the team's npxGD is estimated to be 0.30 goals better with this player than with an average player at the same position. A score of 0.00 is a competent starter. Negative scores indicate a drag on team performance relative to position average.

Data sources

Kaggle (1992–2022) · Understat (2014/15–present)

Made for analysts & fans

Built by statisticians and football fans, this site attempts to showcase the intersection of data and the beautiful game, in order to make clean, stable, and interpretable player rankings. A leaderboard for all single season performances; the peak page allows for rolling-windows and a more focused view; scatter graph to visualise two variables (bivariate data); comparison page to compare up to four players across various stats; CANTONA page to explore our impact metric without distractions; player and team pages providing further breakdowns.