[Paper Review] Archetypal Athletes

Manuel J. A. Eugster|arXiv (Cornell University)|Jan 1, 2011

Sports Analytics and Performance7 references4 citations

TL;DR

This paper proposes archetypal analysis to identify extreme, outstanding athletes—both positively and negatively—by modeling them as convex combinations of data-driven archetypes on the boundary of the convex hull of sports statistics. Applied to NBA and soccer skill ratings, the method reveals distinct archetypal player types (e.g., offensive, defensive, all-around), and identifies Wayne Rooney and Leo Messi as top performers under a defined composite archetype of offensive excellence.

ABSTRACT

Discussions on outstanding---positively and/ornegatively---athletes are common practice. The rapidly grown amount ofcollected sports data now allow to support such discussions withstate of the art statistical methodology. Given a (multivariate) dataset with collected data of athletes within a specific sport,outstanding athletes are values on the data set boundary. In thepresent paper we propose archetypal analysis to compute these extremevalues. The so-called archetypes, i.e., archetypal athletes,approximate the observations as convex combinations. We interpret thearchetypal athletes and their characteristics, and, furthermore, thecomposition of all athletes based on the archetypal athletes. Theapplication of archetypal analysis is demonstrated on basketballstatistics and soccer skill ratings.

Motivation & Objective

• To provide a data-driven, reproducible method for identifying outstanding athletes based on multivariate sports statistics.
• To address the limitation of one-dimensional rankings in multidimensional sports data by preserving information through convex combinations.
• To interpret athletes not just by rank, but by their composition relative to archetypal player types (e.g., offensive, defensive, all-around).
• To demonstrate the method’s consistency with expert opinion using real-world NBA and soccer datasets.
• To offer a transparent framework for evaluating player performance that supports managerial and tactical decisions in sports.

Proposed method

• Uses archetypal analysis to compute k archetypes as convex combinations of observed athletes, minimizing residual sum of squares (RSS) between original data and reconstructed data.
• Solves the optimization problem via alternating constrained least squares: iteratively updates coefficient matrices α (for reconstruction) and β (for archetype formation).
• Archetypes are constrained to lie on the convex hull of the data, ensuring they represent extreme, boundary cases in the multivariate space.
• Employs the elbow criterion to select the optimal number of archetypes k, balancing model fit and interpretability.
• Represents each athlete as a convex combination of archetypes using α coefficients, enabling interpretation of their 'type' and relative position.
• Applies the method to two datasets: 2009/2010 NBA player statistics and 2011 soccer skill ratings from major European leagues.

Experimental results

Research questions

• RQ1 How can we objectively identify the most extreme athletes in multidimensional sports data without reducing to a single dimension?
• RQ2 Which statistical method best preserves interpretability and information while identifying 'best' or 'worst' athletes in a multivariate context?
• RQ3 Can archetypal analysis reveal distinct, interpretable player types (e.g., offensive, defensive) that align with expert perception?
• RQ4 How do real-world athletes compare to archetypal types in terms of composition, and which players best approximate idealized player profiles?
• RQ5 Can the method support decision-making in sports by providing a transparent, reproducible framework for player evaluation?

Key findings

• Archetypal analysis successfully identifies four distinct player archetypes in soccer: offensive all-rounder (Archetype 1), center forward (Archetype 2), weak player (Archetype 3), and defender (Archetype 4), each with distinct skill profiles.
• The α coefficients show that defenders are predominantly composed of Archetype 4, confirming the method’s interpretability and alignment with positional roles.
• Under a defined composite archetype favoring offensive excellence, Wayne Rooney (0.82 α₁) is identified as the top player, followed by Leo Messi (0.79 α₁), Cristiano Ronaldo (0.68 α₁), and others.
• The method’s results are consistent with general expert consensus, validating its use for objective, data-driven player evaluation.
• The model achieves a balance between fit and interpretability, with k=4 selected via the elbow criterion for the soccer dataset.
• All data and R code for replication are publicly available via the SportsAnalytics package in R, ensuring reproducibility and transparency.