Football Stat Definitions

Glossary of all football statistics used across the site

Football > Definitions

All stat categories, model concepts, and definitions used across Football pages.

Panna Model

The Panna rating system is a predictive player value model for association football, analogous to TORP for AFL.

Term Description
Panna The headline career-trait rating: a recency-weighted, multi-season adjusted plus-minus (xG/90). “How good is this player” — the best guess of their next-match impact. Offense − Defense; higher = more valuable.
Piero A composite player rating — one number blending the three trait ratings Panna, EPR and PSR (panna-led, 0.5 / 0.3 / 0.2), z-scored across the pool and rescaled onto the Panna (xG/90) scale. The player-level cousin of the team Tiento; named after Alessandro Del Piero.
xRAPM The per-season version of the same impact metric — this season’s adjusted plus-minus contribution, an empirical-Bayes posterior that shrinks the raw evidence below toward a statistical prior. Panna is the career-weighted version of it (xRAPM is the building block).
Raw RAPM The unshrunk prior-free adjusted plus-minus estimate straight off the ridge regression — the raw evidence xRAPM is built from, before Bayesian shrinkage pulls thin-minutes players toward the prior. Noisier below ~900 minutes; shown for transparency alongside xRAPM, not as a replacement for it.
Offense Offensive component of Panna — expected goals (xG) created per 90, adjusted for teammate and opponent quality.
Defense Defensive component of Panna — xG prevented per 90. Negative values are better (more xG prevented).
SPM Statistical Plus-Minus — what a player’s box-score stat line alone predicts their impact to be, season-to-date. The Bayesian prior xRAPM/Panna shrink toward is a separate, windowed role-aware variant of this same idea (not the SPM number shown here).
Percentile Player’s rank as a percentile among all rated players (100 = top).

World Cup 2026

Term Description
Tiento The headline national-team strength rating for the World Cup: goals better (or worse) than the average qualified side on a neutral pitch. A blend of four independent systems — squad Panna (40%), Elo (30%), EPR (20%), PSR (10%) — z-scored across the 48 teams, with each squad metric the minutes-weighted sum of the players actually shown on the squad page. Named for the ball Argentina supplied for the first half of the first-ever World Cup final (Montevideo, 1930). Drives the Simulator’s knockout ties.
BT Bradley-Terry tournament strength — the match model’s predictions for all 1,128 possible team pairings compressed into one number per team. Host advantage (USA, Canada, Mexico) is baked in, so it reads as expected strength at this tournament.
EPR (squad) Squad aggregate of player Expected Possession Ratings — per-90 expected possession value summed across the squad’s projected contributors.
PSR (squad) Squad aggregate of player box-score skill ratings.
Champ % Probability of winning the tournament, from the 10,000-run Monte Carlo simulation.

Possession Value & Win Probability

Per-match value breakdowns, live on the match and match events pages:

Term Description
EPV Expected Possession Value — the probability the team in possession scores next, given the current game state.
EPV credit / Equity Per-action credit assigned to each player based on how their action moved EPV. Positive = value added, negative = value lost. Summed per player per match on the events pages.
WPA Win Probability Added — per-action change in the team’s chance of winning, credited to the actor (and receiver where relevant). Late-game actions in close matches swing hardest — that’s the point: it measures this match’s stakes, not repeatable skill.
EPR Expected Possession Rating — a player’s predictive rating derived from historical EPV contributions, per 90, adjusted for opponent strength and league scoring environment.
PSR Player Skill Rating — predictive rating built from box-score skill estimates, centered within league and position, with a cross-league adjustment estimated from players who play in multiple competitions (league play, continental cups, internationals).

Stat Value (PSV)

The per-match box-score value model — football’s cousin of basketball’s Box Plus-Minus. Shown on match pages (live during matches) and in game logs:

Term Description
PSV Player Stat Value — everything a player did in the box score (shots, key passes, duels won above expectation, tackles…) weighted by how much each stat contributes to winning, expressed vs the average player in that league and role. Finished matches show the exact production value; in-progress matches show a live estimate from the same model.
OSV / DSV The offensive / defensive split of PSV — they always sum exactly to PSV.
Piero Value The per-match combined rating: 50/50 blend of adjusted EPV credit and PSV — “value added this game” in one number. The production twin of Piero: EPR’s per-match expression is EPV credit and PSR’s is PSV, so both of Piero’s stat components appear here in match form. (Panna has no per-match twin — a single game’s adjusted plus-minus is pure noise, which is exactly why the match number is built from EPV+PSV instead.)
xG Expected Goals — the probability an average player scores this chance, from location, body part, assist type. Penalties count a fixed 0.80.
xGOT Post-shot xG — chance of scoring given where the shot went in the goal mouth. Only defined for on-target shots.
Placement xGOT − xG summed over a player’s shots: finding corners. The most repeatable finishing skill we measure.
G − xGOT Goals minus post-shot xG: keeper quality plus luck. It is not shooter skill — don’t rate finishers on it.
GSAA Goals Saved Above Average (keepers) — the xGOT of on-target shots faced minus goals conceded: shot-stopping vs the placement actually faced.
WOE duels “Wins over expected” for aerials, take-ons, tackles and containment — duel wins above what an average player would manage from the same positions, so volume counts (winning 8 of 12 beats winning 2 of 2).

Ratings vs Values — the naming convention

Every metric here is one of two kinds, and the last letter usually tells you which:

  • Ratings (…R) answer “how good is this player?” — a smoothed, predictive estimate of their next match: Panna, Piero, EPR, PSR (and the OSR/DSR splits).
  • Values (…V) answer “what did they actually do in this match?” — production, which bounces around game to game: EPV credit, WPA, PSV (and OSV/DSV), and Piero Value.

Each rating has a value twin: EPR↔︎EPV, PSR↔︎PSV, Piero↔︎Piero Value. Panna is the deliberate exception — trait-only, because it comes from on/off regression rather than counting match events, and no single match contains enough signal to compute it.

What each rating is good at — and its limits

Honest guidance for reading the numbers. Every metric answers a different question; none answers all of them.

Rating Best at Limits
Piero The single best answer to “how good is this player right now” — three independent estimators blended, so one model’s quirks wash out. Units are abstract (rescaled xG/90); inherits a little of each component’s blind spots.
Panna (career) True impact with teammates and opponents controlled — the most rigorous skill estimate here. Needs no cross-league adjustment (opposition is controlled directly). Slow to move: multi-season weighting means breakout seasons take time to fully register; wide error bars under ~a season of minutes.
xRAPM (season) This season’s impact, same methodology. Noisy below ~900 minutes; a season aggregate, not current form.
EPR On-ball value from every touch — updates weekly, opponent-adjusted. On-ball only: off-ball defending and structure are under-credited.
PSR Which skills drive a player’s value — granular, stable, works for thin samples via shrinkage. Box-score rates barely differ by league, so it leans entirely on the cross-league adjustment (which is deliberately conservative — strong players in weak leagues still read slightly high).
PSV What a player produced this match, available live. One match is mostly noise; box-score only; live GK values are rough until save-detail features ship.
EPV credit Valuing every single action in goal units — the best “how did they shape this match” view. On-ball bias; opponent adjustment applies to the total only, not the off/def split.
WPA Clutch context — a 90th-minute winner counts like the moment it was. Deliberately spiky; not a skill measure. Never rank a season by WPA alone.
xG Chance quality; the best single predictor of future scoring. Shooter-agnostic — a Messi chance and a centre-back chance score the same.
xGOT / Placement The most repeatable finishing signal (placement skill persists season to season far better than raw goals-minus-xG). On-target shots only; reliable goal-mouth data starts 2021-22.
GSAA Keeper shot-stopping against the placements actually faced. Shooter-agnostic (an unstoppable placement counts the same regardless of who hit it); shot-stopping only — sweeping and distribution live in PSV/EPV.

Box-Score Stats