Conventional player rating systems have a structural problem: they are predominantly reward-based, crediting outcomes – goals, assists, clean sheets – rather than the positional and relational behaviours that generate those outcomes. A recent analytical project shared in the football tactics community attempts to sidestep that limitation by applying a PageRank-style network algorithm to all 1,435 players across Europe’s top five leagues, producing a single comparative score grounded in influence rather than accumulated event data.
The premise is borrowed directly from web-graph theory. In Google’s original PageRank model, a page’s authority is determined not merely by how many links point to it, but by the authority of those linking pages. Transposed onto a football passing network, the logic holds: a player’s influence is shaped by the quality and centrality of the interactions flowing through them, not just the volume of touches or progressive actions they complete in isolation.
The Methodological Premise
Network centrality as an analytical lens is not a new arrival in football analytics. Academic work has applied betweenness and closeness centrality to passing graphs for well over a decade, and the football analytics literature has documented how density, clustering coefficients, and eigenvector-adjacent measures can map how a team’s structure changes in and out of possession. What distinguishes a PageRank-specific approach is the recursive weighting: connections to central nodes carry more value than connections to peripheral ones, which creates a non-linear amplification effect for players operating at the structural core of high-quality build-up networks.
The practical implication is significant for evaluation. A deep-lying midfielder who connects a centre-back pairing to a technically dominant number eight will accumulate a different kind of influence than a striker whose primary interactions are with wide forwards rather than build-up players. Standard metrics – passes per 90, progressive carries, key passes – do not fully capture that distinction. A network-weighted score, in theory, can.
The dataset scope – 1,435 players across five leagues – is large enough to generate meaningful cross-position and cross-league comparisons, which is where single-number rating systems historically earn or lose their credibility. Earlier unified models, including the EA Sports Player Performance Index, attempted full cross-position comparability – rating all players with one score regardless of specialty – though available documentation does not confirm whether that system extended beyond a single league’s data. Extending network analysis across what are understood to be Europe’s top five leagues simultaneously introduces a legitimate comparative dimension, provided the underlying pass-map data is consistent in its event-coding standards across competitions.
Why Outliers Are the Analytical Point
In any large-scale rating exercise, the middle of the distribution is largely uncontroversial. Players who rank highly on both conventional metrics and network scores offer validation but not insight. The analytically productive zone is at the edges – players whose network rank diverges materially from their reputation, market value, or event-based statistical profile.
Positive outliers in a PageRank model tend to share a structural characteristic: they operate as bridges between high-authority nodes. In a 4-3-3 with a high press, the player connecting the defensive line to the central midfield triangle under pressure is absorbing and redistributing influence from both directions. If that player is a relatively unheralded defensive midfielder at a well-organised mid-table club, their network score may significantly exceed their market valuation – precisely because their contribution is relational rather than event-driven.
Negative outliers are equally instructive. A technically gifted forward at a club with a disorganised build-up structure may generate high xG and shot volume while registering a modest network score, because the interactions flowing through them originate from low-authority nodes – long balls, second balls, isolated wide play – rather than from a connected, high-density passing network. The score is not saying the player is poor; it is saying their influence is structurally constrained by the system around them.
This is where the network approach intersects most meaningfully with recruitment analysis. Advanced efficiency metrics applied to large player cohorts have, in the view of analytics practitioners, shown that surface-level output can mask the underlying quality of chance creation and defensive contribution. A PageRank layer on top of xG efficiency data would, in principle, distinguish between a player generating high-quality chances within a high-authority passing structure and one achieving similar output numbers through volume and low-quality interaction chains.
Cross-League Comparability and Its Limits
Applying a single algorithm across five leagues simultaneously forces a confrontation with a structural issue that single-league models avoid: the passing networks themselves differ materially in density, tempo, and spatial organisation. It has been suggested in analytical commentary that some leagues may be associated with higher pass-network densities reflecting pressing intensity and short combinational build-up, while others produce sparser networks with greater reliance on direct transition play. Whether these characteristics hold consistently across seasons and clubs is itself a live methodological question.
A player operating as a central node in a dense network is not necessarily more influential in a football sense than a player performing the equivalent structural role in a leaner environment – but a raw cross-league PageRank comparison might produce that reading if the algorithm does not account for baseline network density by league or by club. Whether the project addresses this normalisation question is the key methodological variable its findings would need to answer.
A 2025 comparison of major player rating systems across 2,190 players found systematic differences between platforms – WhoScored ratings trending materially lower overall, for instance – confirming that methodological choices upstream of the final score produce non-trivial divergences at the player level. The same dynamic applies here: the comparability of a PageRank score between a Bundesliga defensive midfielder and a Serie A playmaker depends entirely on how the underlying graph is constructed and whether league-level normalisation has been applied.
Positional Profiles and Network Centrality
Not all positions sit equally in a passing network’s authority hierarchy, and a well-constructed PageRank application to football should reflect that structurally rather than treating it as noise. Centre-backs at possession-dominant clubs will naturally accumulate high interaction counts with other authoritative nodes – the goalkeeper, the double pivot, the wide centre-backs – producing elevated baseline scores that may not distinguish elite from competent performers at that position.
The more diagnostically powerful applications tend to cluster around the middle third of the pitch, where positional authority is genuinely contested. An interior midfielder in a 4-3-3 who acts as the primary link between build-up and final-third creation sits at a genuine network junction; their PageRank score should meaningfully separate elite performers from functional ones, because the quality of connections flowing through them varies considerably. The same logic applies to deep-lying playmakers operating between defensive and progressive phases, and to inverted wingers whose interaction patterns differ structurally from traditional wide forwards.
Strikers, conversely, present the inverse problem to centre-backs: their network positions are naturally peripheral relative to build-up authority, meaning a high-scoring centre-forward at a direct, transition-heavy club may register a low PageRank score while still being the most impactful player on the pitch in terms of match outcomes. This is not a flaw in the methodology so much as a known limitation of interaction-graph models when applied to players whose primary contribution comes at the terminal end of attacking sequences.
The analytical framework here has parallels with cross-domain thinking about decision-making under constraint – the kind of structural reasoning applied when game-theoretic frameworks are used to model football match dynamics. In both cases, the analytical value lies not in the headline number but in what the structural position within the network or game tree reveals about a participant’s actual influence relative to their apparent role.
Validation and the Scouting Question
The most important test for any player rating system is predictive validity: does a high score today correlate with strong performance, or strong market outcomes, at a later point? Player-strength models developed in the academic literature – including work that aimed to rank both players and teams, predict future matches, and identify emerging talent – have been explicit about this requirement. A score that is mathematically elegant but disconnected from out-of-sample prediction is a classification exercise, not an analytical tool.
The football analytics literature is broadly enthusiastic about network-based approaches while remaining cautious about validation. A systematic review of passing-network studies found that the field is heavily concentrated on offensive phases of elite men’s football, leaving defensive contributions and women’s football significantly underrepresented. A PageRank model applied to 1,435 players covering both offensive and defensive positions across five leagues would represent a meaningful step toward fuller coverage, but only if the scoring framework accounts for the asymmetry between how defensive and offensive interactions register in a directed passing graph.
Defensive actions – interceptions, pressures, aerial duels – do not appear as edges in a pass-completion network, which means a centre-back or defensive midfielder whose primary value is disruptive rather than distributive will be systematically underrated unless the graph incorporates ball-recovery events as directed interactions. Whether this project’s methodology extends to defensive interaction data is a variable that would materially change the interpretation of every ranking it produces.
What the Network Approach Adds to Existing Frameworks
The case for network-based player rating does not rest on replacing conventional metrics but on adding a dimension that event-data systems structurally cannot capture. Goals, assists, progressive carries, and xG contribution measure what a player does with the ball at specific moments. Network centrality measures where a player sits in the flow of a team’s collective decision-making – a distinction that matters most for positional players whose contribution is invisible in a box-score but decisive in a team’s ability to build through pressure.
The players most likely to be misevaluated by purely event-based systems are those who create the structural conditions for others to produce events: the half-space midfielder who drags pressing lines out of shape before releasing to a forward, the wide centre-back who initiates build-up sequences under high press, or the false nine who drops to generate numerical superiority in midfield before the ball moves to overlapping runners. None of these contributions generates a measurable event in conventional terms; all of them generate authority in a network model that tracks interaction quality.
The project’s value, ultimately, depends on the transparency of its methodology and the honesty of its validation. A rating system covering 1,435 players across Europe’s top five leagues with a network-based algorithm is a genuinely ambitious analytical undertaking. Whether it produces actionable scouting intelligence or a compelling but unvalidated ranking will be determined by the choices made in graph construction, league normalisation, and the inclusion or exclusion of defensive interaction data – none of which a headline number alone can answer.
