How Football Metrics Read Differently in the Women's Game

Football metrics are not universal constants. Expected goals, expected assists, and most advanced statistics are models fitted to a specific population of matches, and their outputs are only as valid as the resemblance between that population and the football you are applying them to. This is the central issue when reading data from the women's game.

The point is frequently misunderstood in both directions. It is not that women's football requires different concepts — a pass is a pass, and pressing is pressing. It is that the numerical calibration underneath several familiar metrics was estimated somewhere else, and transplanting it without adjustment produces quiet, systematic error.

A Metric Is a Model, and a Model Has a Population

Consider what an expected goals figure actually is. A model observes a very large number of historical shots, records their characteristics — distance, angle, body part, assist type, defensive pressure — and learns what proportion of shots with those characteristics were scored. When a new shot arrives, the model reports the historical conversion rate of shots that resembled it.

Everything in that description depends on the phrase "historical conversion rate", and that rate comes from whichever matches were used to train the model. If a model is fitted on men's professional football, it is reporting how often men's professional footballers converted comparable chances.

Apply that model unchanged to a different population and it will still produce a number. The number will look reasonable. It will simply be answering a question about a different set of players. This is not a flaw in the concept of expected goals; it is a flaw in how expected goals is often deployed.

Where Calibration Actually Bites

The practical consequence is concentrated in the parts of a model most sensitive to physical and technical differences between populations.

  • Shot distance. The conversion probability curve as distance increases is one of the most population-sensitive relationships in any shot model. If it differs at all, long-range chances are the first place the mis-calibration appears.
  • Goalkeeping baselines. Post-shot models measure shot quality against an implicit standard of what a goalkeeper is expected to save. That standard is estimated from a specific set of goalkeepers.
  • Set-piece value. The value of a corner or a free-kick delivery depends on aerial contest outcomes and defensive organisation, both of which are population-specific.
  • Pressure adjustments. Models that account for defender proximity are estimating how much a given amount of pressure suppresses conversion.

None of this implies that any of these values is higher or lower in one game than the other. The honest statement is narrower and more useful: these parameters are estimated, they are not guaranteed to be identical across populations, and a model that has not been refitted is assuming they are. A properly built women's-game model estimates them from women's-game matches. The difference between a refitted model and a borrowed one is not cosmetic.

The Data History Is Shorter, and That Compounds

Detailed event data collection in women's football began meaningfully later than in the men's game, and coverage remains uneven across competitions. This creates a second-order problem on top of calibration.

Models need data to be fitted. A shot model needs a very large number of shots before its estimates for uncommon situations — headers from wide positions, shots after a specific type of build-up — become stable. Where fewer seasons of tracked matches exist, models are fitted on thinner evidence, which widens their uncertainty even when they are correctly specified for the population.

Coverage also varies sharply between competitions. The most-watched leagues and major international tournaments receive detailed collection; other professional and semi-professional leagues receive far less. A player's statistical record can therefore be dense for part of their career and sparse for the rest, which makes career-long comparisons awkward in ways that rarely arise for men's players in the top five leagues.

Competitive Dispersion Changes What Per-90 Means

This is the factor most often ignored, and it distorts player comparisons more than calibration does.

Leagues at earlier stages of professionalisation tend to have wider gaps between their strongest and weakest clubs. Resources, full-time training environments, and squad depth arrived at different times for different clubs, and the effects persist for years. Where dispersion is wide, matches between the top and the bottom are less balanced, and the players at dominant clubs accumulate volume statistics against opposition that a mid-table player never faces on equal terms.

Raw per-90 figures absorb all of this. A forward at a dominant club and a forward at a mid-table club in the same division are not playing the same league, in any meaningful statistical sense. Opponent-adjusted or possession-adjusted figures partially correct for it; raw totals do not correct for it at all.

The same logic applies to defensive metrics, which are famously possession-dependent. A defender in a team that dominates the ball has fewer defensive actions to perform, and that says nothing about their ability. Where the gap between clubs is wide, that effect is amplified.

Structural Differences Between Competitions

The competitions themselves are organised differently, and the structure feeds back into the data.

Europe's leading domestic leagues — the Women's Super League in England, Liga F in Spain, the Frauen-Bundesliga in Germany, and the Première Ligue in France — use the familiar European model: a league table, promotion and relegation, and continental qualification into the UEFA Women's Champions League. The WSL's transition to full professionalism in 2018 is a useful reference point for how recent this landscape is, and season-on-season comparisons that straddle a professionalisation step are comparing different competitive environments.

North America runs on a different logic. Leagues there have historically used mechanisms drawn from the American sports model — salary caps, centralised player allocation, a closed league without relegation, and a playoff to decide the championship — with the details evolving through collective bargaining. Roster construction responds to those rules rather than to a transfer market, which changes squad turnover patterns and makes cross-continental player comparison a matter of adjusting for two different economies.

Calendars differ too. A league running spring to autumn and one running autumn to spring do not align, so international tournaments interrupt them at different points in their competitive cycles.

What Transfers Across Unchanged

The list of things that need no adjustment is longer than the list that does, and it is worth stating clearly.

  • Tactical concepts. Build-up phases, pressing structures, defensive blocks, half-space occupation, and rest defence are structural ideas, not population parameters.
  • Positional and role frameworks. A deep-lying playmaker is defined by function and position, and the definition is identical.
  • Relative and rank-based metrics. Percentile ranks within a league, PPDA compared to league average, and field tilt are all relative measures that calibrate themselves against their own competition.
  • Descriptive event counts. Passes, carries, duels, and touches in the box are observations rather than model outputs.
  • Visualisation. Shot maps, pass networks, and heat maps display what happened without estimating a probability, so they carry no borrowed calibration.

The pattern is consistent: descriptive and relative statistics travel freely, while probabilistic model outputs need refitting. That distinction is the single most useful thing to hold onto.

How to Read Women's Football Numbers Well

Start by asking where the model came from. If a source publishes expected goals for women's matches, it is worth knowing whether the underlying model was estimated on women's football or borrowed. Good data providers state this; the question is fair to ask of any that do not.

Compare within a league before comparing across leagues, and prefer percentile ranks to raw per-90 values wherever dispersion is wide. Check how many matches sit behind a figure, remembering that shorter league seasons in some competitions mean a full campaign supplies a smaller sample than the same season length elsewhere. Treat comparisons that cross a professionalisation threshold with the caution you would give any structural break in a time series. Platforms such as RubiScore publish women's competitions alongside men's with the same entity structure — competition, club, player, fixture — which makes the within-league comparisons straightforward once the calibration question is settled.

The underlying discipline is the same one that good analysis always requires: know what your number was estimated from, and do not ask it questions about a population it never observed. Women's competition data, fixtures, and tables are published on rubiscore.com.

2100 Clubhouse Drive, Granbury, TX  76048