Analytic method for evaluating players’ decisions in team sports: Applications to the soccer goalkeeper

Abstract

The aim of this study was to define a method for evaluating a player’s decisions during a game based on the success probability of his actions and for analyzing the player strategy inferred from game actions. There were developed formal definitions of i) the stochastic process of player decisions in game situations and ii) the inference process of player strategy based on his game decisions. The method was applied to the context of soccer goalkeepers. A model of goalkeeper positioning, with geometric parameters and solutions to optimize his position based on the ball position and trajectory, was developed. The model was tested with a sample of 65 professional goalkeepers (28.8 ± 4.1 years old) playing for their national teams in 2010 and 2014 World Cups. The goalkeeper’s decisions were compared to decisions from a large dataset of other goalkeepers, defining the probability of success in each game circumstance. There were assessed i) performance in a defined set of classes of game plays; ii) entropy of goalkeepers’ decisions; and iii) the effect of goalkeepers’ positioning updates on the outcome (save or goal). Goalkeepers’ decisions were similar to the ones with the lowest probability of goal on the dataset. Goalkeepers’ entropy varied between 24% and 71% of the maximum possible entropy. Positioning dynamics in the instants that preceded the shot indicated that, in goals and saves, goalkeepers optimized their position before the shot in 21.87% and 83.33% of the situations, respectively. These results validate a method to discriminate successful performance. In conclusion, this method enables a more precise assessment of a player’s decision-making ability by consulting a representative dataset of equivalent actions to define the probability of his success. Therefore, it supports the evaluation of the player’s decision separately from his technical skill execution, which overcomes the scientific challenge of discriminating the evaluation of a player’s decision performance from the action result.

A review of the interaction between the striker and the goalkeeper at the individual tactical level in football

“…Tables 2 and 3 present the extracted data that pertains to study design and quality. One study 43 investigated football at a league level with the majority covering the World cup 44,45 and the European Championships. 45–47 These studies cover all competitions from 2002–2014 (with competition held every four years), so a good account could be surmised.…”

“…One study 43 investigated football at a league level with the majority covering the World cup 44,45 and the European Championships. 45–47 These studies cover all competitions from 2002–2014 (with competition held every four years), so a good account could be surmised. Alternatively, Villemain and Hauw 48 used players from the French National Academy as participants in a qualitative study that used thematic analysis to identify 23 critical game situations; however, these goalkeepers were considered future professionals.…”

“…44 analysed 88% of 2002 World Cup matches and as such had a sufficient number of matches and different goalkeepers in their sample, Lamas et al. 45 limited their sample to all goals and hard saves from all matches of three World Cups, and Villemain and Hauw 48 collected 11 goalkeepers of similar age and skill level (i.e. elite) for their qualitative analysis.…”

“…Lamas et al. 45 employed a novel method, where an ‘oracle dataset’ of goals and ‘hard saves’ was compared to a smaller, limited sample of shots on target for their test groups. As the name suggests, an oracle data set is designed to provide a medium through which insight can be benchmarked, forecasted or prophesised.…”

“…The rationale for the choice of outcome variables presented in each study was shaped by expert knowledge, 44,45 constrained by the data provider, 43 guided by the participants 48 or not presented. 46,47 Whilst all articles included the goalkeeper in their analyses, two 43,46 included variables that were not unique to nor especially relevant for goalkeepers (i.e.…”