METHODS FOR STUDYING COLLECTIVE PERFORMANCE IN SPORTS: A SYSTEMATIC LITERATURE REVIEW

Video and Statistical Data

Many studies were based on on-the-ground observations from which the researchers extracted video or statistical data. Their goal was then to conceive instruments aimed at analyzing the different elements of the game, and to assess the reliability of these instruments. In all the articles taken into account, Jones et al. (2008) were the first to apply this methodology. In order to analyze the performances of a professional team of rugby union over the course of several seasons, they designed and tested what they called “standardized game indicators” (e.g., goal kick success, handling errors). Similar studies have been conducted in rugby (Kempton & Coutts, 2015; Kempton et al., 2015), football (Tenga et al., 2009), volleyball (De Alcaraz et al., 2017), goalball (Morato et al., 2017), netball (Bruce et al., 2018), and Australian rules football (Clarke et al., 2018). Some studies focused on specific aspects of the game like, in rugby, ball possession (Villarejo et al., 2014) or, in football, offensive phases (Ortega-Toro et al., 2019). This close-up insight is particularly useful to define which skills are required in some less-known team sports. For instance, Lupo et al. (2012) shed light not only on the physical, but also on the technical and tactical demands of water basketball.

Some studies were more comprehensive, notably when researchers looked after similarities or recurrences in the game of a series of teams. That's what Carling et al. (2015) did when they monitored the physical, tactical and technical parameters of a single professional football team for 5 consecutive years. By the same token, Woods et al. (2018) created 15 performance indicators that they used to assess the evolutionary dynamics of gameplay in every single team of a same national league over the course of 10 seasons. This kind of long-term study can lead, in the case of Cambre Añon et al. (2014), to the characterization of a game model. We tagged another method that consisted of comparing two teams' indicators of performance. For instance, Vaz et al. (2015) listed the differences in scoring profiles and game-related statistics between two national teams in rugby union. However, the aforementioned studies were mainly based on statistics and therefore reliant on observational methodologies (Arias-Estero, 2013). They were very descriptive and their results only reflected individual moves such as dribbling, passing, receptions, shoots, or kicks.

Positional Data

Monitoring and collecting data about players' positions throughout the game in team sports has become commonplace thanks to recent technological progress. Thus, it is nowadays possible to accurately measure time-motion variables. For example, Sampaio et al. (2016) studied the effect of defensive pressure in basketball under several scopes, both individual (e.g., distance covered by each player) and collective (e.g., total distance covered by the team at specific speeds). A lot of studies also collected GPS data to compute tactical metrics or collective variables (e.g., surface area, centroids, distance to goal). It is therefore easier nowadays to detect collective behaviors in team sports, a concept known as a “pattern.” A group of researchers even used GPS data to study space–time coordination dynamics in basketball, both through intra- and intercouplings among player dyads (Bourbousson, Sève, & McGarry, 2010a), and the interactions between two teams (Bourbousson, Sève, & McGarry, 2010b). As a result, they showed that there was some reciprocity between teams when they switched from the expansion to contraction phase after ball possession had changed sides. Such a method is called “multi-level approach of game constraints” (Bourbousson et al., 2014). García-Rubio et al. (2015) demonstrated for their part how computer science could help to decipher the interactions between two opposing teams, as well as movement synchronization between players. In their wake, Folgado et al. (2018) coupled elements of positional synchronization and tactical variables to explain why a team's performances improved. Other studies, focused on space occupation in specific areas of a basketball court, came to the conclusion that the teams who are best able to adapt their collective behavior to changes in the environment are more likely to improve (e.g., Esteves et al., 2016). Broadly speaking, positional data allowed researchers to identify key collective patterns of behavior or, in other words, to create variables that captured the emergent functional behaviors of players. Travassos et al. (2012) showed that in futsal, players from the defending team manage to anticipate ball trajectories simply by looking at their opponent's positions on the pitch. In rugby union, thanks to a specific variable called “distance gained,” which is calculated on the basis of the ball's trajectory, Correia et al. (2011) were able to better interpret a team's display. In general terms, the amplitude of ball movements appeared to be a metric related to attack effectiveness in team sports. Moreover, there was a tendency among researchers to detect collective behaviors in spatiotemporal variables such as the required movement velocity (Vilar et al., 2013). Consequently, researchers paid particular attention to small-sided games when they wanted to identify collective variables (e.g., length/width ratio of a block of players) that are crucial to assess collective tactical behaviors and their evolution according to players' ages or game format (Folgado et al., 2012). Carvalho et al. (2014) launched and validated an empirical function that captures the spatiotemporal relationship between tennis players during exchanges, and then proved that such an approach can be applied to an individual sport. Notably, they created an index that successfully describes the players' interaction patterns, as well as disruptions in their playing patterns provoked by changes in their relative positions on the court. Studies based on positional data were useful to discover what skills are most frequently found in high-performing collectives of athletes, as well as some specific characteristics in interactions between athletes.

Interaction-Based Data

The studies classified in this subcategory were devoted to the interactions within teams, between players themselves, or between players and coaching staff. This section also displayed a broad range of methodologies.

Part of these works were dedicated to the interactions between players and often referred to the “social network perspective.” This approach has proven efficient to study collective performance in sports since it aims to explore, represent, and explain individual relationships and their influence on group dynamics (Warner et al., 2012), and even enables graphic representation of several variables (e.g., efficacy, trust, friendship). Actually, the “social network perspective” is part of a wider category of scientific approaches, called “network methods,” commonly used to interpret several types of interactions between players (e.g., passing, position exchanges) and their impact on a team's performance (Passos et al., 2011). This task has recently been made easier by new software conceived to record interactions between teammates and the translation of these interactions into matrices that reflect network structures (F. M. Clemente et al., 2016). Such software has been applied to rugby union where Sasaki et al. (2017) studied defensive strategies (e.g., tackles involving multiple players, turnovers) through the lens of social network centrality analysis in order to detect who holds tactical leadership in a team.

Another way followed by researchers to exploit the interactions within teams was to examine how players share contextual information. Using course-of-action analysis, Poizat et al. (2009) asked athletes to comment on video footage of their performances. Interviews revealed that players alternated between different forms of information-sharing (e.g., symmetrical, asymmetrical, or no sharing at all) and different processes of information-sharing regulation (e.g., inquiring, monitoring, displaying, masking, and focusing). The same researchers (Poizat et al., 2012) led a similar study on table tennis players and demonstrated that in doubles, pauses between exchanges are a crucial moment to share information. Overall, this methodological approach emphasizes how teammates' behaviors are key to their understanding of game situations. This is particularly relevant in team sports where game situations follow a pattern of constant construction/deconstruction. In basketball, for instance, Bourbousson et al. (2010) described the cognitive links between players during matches. Slightly different and mostly found in French-speaking articles, the “enactivist approach of social pairings” consists of asking players to watch their own games and then to comment their moves. Feigean et al. (2018) confirmed that, in football, information collected during the game improves coordination between teammates. The study of interpersonal coordination can also be enriched by phenomenological and mechanical data. When combined, these data helped researchers to explore the salience and accuracy of players' experience sharing (R'Kiouak et al., 2016). The latter studies provide precious information on how athletes can improve the effectiveness of their collective endeavors.

The remaining articles exploiting interaction data were specifically focused on interactions between players and coaching staff. Usually, these articles were case studies in which researchers described the keys coaches like to push to enhance their team's performances. As early as 2012, Zucchermaglio and Alby (2012) had had the idea to record an Italian professional football team's technical staff's meetings at three particular moments of the week (after a victory, after a defeat, and before a game). It gave them a glimpse of how coaches are in a constant process of team building in order to steer their players forward. Other studies followed that path, this time through interviews (unstructured or semistructured) with athletes and coaches. From this material, Collins and Durand-Bush (2016) extracted a range of best practices to optimize teamwork and soundbites that required further questioning. Their approach contributed to the launching of themed toolboxes (e.g., “team chemistry creation in sport”) destined at sport psychologists and coaches (Gershgoren et al., 2016). Hemmestad and Jones (2017) refreshed these findings by using ethnographic tools to detect the core characteristics of a high-performing sports team. For this purpose, they led a 2-year immersive study into the day-to-day life of a Norwegian female elite handball team. Bringing into light the interactions between players and staff helped them to unveil the common culture within a team, and how every single player contributes to it. In 2018, Middlemas et al. made the same discovery, this time by using the whole gamut of ethnographic techniques (observation, video recordings, formal and informal interviews, field notes, statistics, and document analysis). As a conclusion, they gave clues for a better use by rugby union coaches of pre- and postmatch talks, which oftentimes are no more than quick meetings focused on the latest results, but with little hindsight.

Game Results Data

There were studies in which researchers began by classifying the collectives of athletes according to their results (whether it be a ranking, a score, or other specific parameters). In these, “the results of a sequence and the transitory or final score of a match were the most widely used parameters” (Ávila-Moreno et al., 2018, p. 17). Only after having analyzed these parameters, researchers combed through the collective's dynamics and characteristics to find performance indicators. Such studies sought to build predictive models of success in sport.

With rankings as one of the most common methods to classify and reward performance, a first and obvious step was to compare the routines of high- and low-ranked teams, and to diagnose the differences between them. Porath et al. (2016) did just that when they connected the technical and tactical mastery of young volleyball players to their competitive ranking. They then discovered that teams in the highest tier of the table were also the ones that fared best in the technical and tactical aspects of the game. The same method was used in many studies (e.g., Filho et al., 2015; Kempton et al., 2017; Marszałek et al., 2018; Yang et al., 2018), sometimes after having divided the tables into sections, in order to compare teams with similar results (Gabbett & Hulin, 2018; Yu et al., 2018). Some researchers were even more precise: For instance, they compared the first and last four teams of a table (Migdalski & Stone, 2019), or teams from different competitive tiers (Castellano & Casamichana, 2015). H. Liu et al. (2015) went even further: They attributed different tags to football teams according to their level of competition (high, intermediate, or low) and listed their general characteristics (number of shots on goal, passes, or fouls) in order to know why their performances varied so widely.

Paying attention to match results is another option to compare high- and low-performing teams (e.g., Alves et al., 2019; Csataljay et al., 2009; Harrop & Nevill, 2014; Palao & Ortega, 2015; Vinson & Peters, 2016). In 2010, Turner and Sayers brought to light a correlation between, on the one hand, defense to offense transition speed in football, and on the other hand, the number of positive attacking outcomes. Other researchers enriched this approach by introducing distinctions between age groups (Medeiros et al., 2017) or by taking into account the match result in itself, balanced or unbalanced (Gómez et al., 2014). However, Watson et al. (2017) warned that key factors that differentiate overachieving from underachieving teams could vary widely from one competition to another. An alternative method is to focus solely on losing teams. This is why Kerr and Males (2010) met the members of a lacrosse national team that had lost four games in a row during a world championship and, through semistructured interviews, asked them about their motivation and psychological state. For their part, Wergin et al. (2018) explored the causes of collective collapse in sports by observing several German professional sports teams who suffered several consecutive defeats.

Some researchers tried to find correlations between game sequences results and game characteristics such as scoring rate in volleyball (Nikos & Elissavet, 2011) or inside pass success in basketball (Courel-Ibáñez et al., 2016). This method may help to predict the impact of some performance indicators at key moments of the game. In volleyball, Sánchez-Moreno et al. (2015) identified a correlation between the average duration of rallies and a game's final result. A similar study led in football (Paixão et al., 2015) tended to prove that teams adapt the duration of their offensive endeavors to the scoreline, which suggests that every team presents a distinctive playing style.

The compilation of all these collective performance indicators enables predictive models (e.g., Bennett et al., 2019; Bremner et al., 2013; Parmar et al., 2017; Robertson et al., 2016; Scanlan et al., 2016; Woods et al., 2017). George and Panagiotis (2008) showed that in beach-volley, 2-0 match winners displayed better performances than losers in almost all technical skills. Nevertheless, there is still room to improve these predictive models, namely by introducing further indicators—for instance, a spatial component in a mathematical model of self-organization in football (Chassy et al., 2018). In other cases, researchers commonly chose to sharpen their predictive models of the teams' performances by adding new ratios to the statistics they had already collected (Drikos et al., 2009).

Factors Related to the Collective's Environment

Among these factors, we selected all of those which were directly related to the competitions or tournaments themselves. First, the organizational factors, which can weigh on statistics and performance in various ways, such as playing at home or away (Campos et al., 2015; Lago-Peñas et al., 2013; T. Liu et al., 2019). Ponzo and Scoppa (2018) showed that crowd support contributes to home advantage in football, notwithstanding other variables such as the players' familiarity with the place or travel fatigue. The impact on performance of other fatigue-inducing factors was also studied, like jetlag (Nutting & Price, 2017), the number of days off in a competition (Scoppa, 2015), fixture congestion (Folgado et al., 2015), weather conditions, travel distances, or the quantity of rest days (Watanabe et al., 2017). Such works can provide technical staff some clues to adapt their training programs, and help international sporting bodies to appoint the right host and pick the right dates for major tournaments or competitions.

Economic factors also have an impact on collective performance. How much money a club or organization can spend on its activities is considered an important variable, albeit not the main one. Lago-Peñas and Sampaio's article (2015) is a case in point: They demonstrated how crucial it is in football to have a good start of season, especially if you have one of the league's lowest budgets. Indeed, “the better the team performance at the beginning of the season, the better the ranking at the end of the season” (p. 4). It is worth mentioning that in the articles surveyed, economic data was always considered in relation to other economic variables that are specific to the sport, such as: investments in new facilities (Rockerbie & Easton, 2019), regulatory loopholes that make losing more profitable than winning (Price et al., 2010), recruitment and wage policies (Rossi et al., 2018), salary gaps between players of the same team (Annala & Winfree, 2011; Yamamura, 2015), technical staff's wages (Colbert & Eckard, 2015), managerial changes during the season (Balduck et al., 2010; Martinez & Caudill, 2013), and recruitment (Bergman & Logan, 2016). And while the articles aforementioned often linked motivation to money, Maier et al. (2016) revealed for their part that job satisfaction in sports is also strongly related to organizational support, a factor that may become relevant to optimize performances. Organizational support in sport is determined by the general managers' ability to implement culture change and best practices (Cruickshank et al., 2014), and by their technical experience and education (Juravich et al., 2017).

Factors Related to the Collective's Characteristics

Several studies took into account the collective's composition to analyze collective performances. Longitudinal studies measured, for instance, how much performances are correlated to the experience acquired by players (e.g., number of matches played at the international level, Kalén et al., 2017) or coaches (Roach, 2016). Other studies found a statistical link between, on one side, the performances of a national team, and on the other side, the total number of international games played by its members (Sedeaud et al., 2017). Montanari et al. (2008), for their part, related the results of a club team with players' turnover and the intensity of relations between them. Unsurprisingly, these studies revealed that maintaining squad stability improves results. However, stability doesn't mean homogeneity. To know if double pairings in tennis table would gel independently of the players' respective skills, Van Opstal et al. (2018) submitted players to experimental protocols based on virtual tasks and concluded that skill gaps within pairings did not affect how players share the responsibility to intercept the ball.

Among the collective characteristics under study, psychosocial factors (e.g., cohesion, leadership, role distribution, motivational climate) were a major item. Callow et al. conducted in 2009 a study focused on the relations between transformational leadership, cohesion, and team performance. This seminal article paved the way to other studies, more restricted in scope, that modelled the relations between performance on the one hand and, on the other hand, managerial leadership (Bormann et al., 2016), trust in leaders (Mach & Lvina, 2017), or the team skipper's leadership abilities (Fransen, Vanbeselaere, De Cuyper, Vande Broek, et al., 2014). In team sports specifically, the weight of collective emotions on performances has been the object of intense scrutiny (e.g., Campo et al., 2018). The interest for this question stemmed from an article by Crombie et al. (2009), who insisted that emotions in a sports team are collective by nature. Following that path, Uphill et al. (2012) tried to correlate a team's performances to their perceived level of happiness, and Slater et al. (2018), to the passion visible on players' faces when they sing their national anthems, under the principle that intense feelings drive behavioral changes on the playing field. Unity or division within the team was another psychosocial factor used to predict performances and/or collective efficacy (Leo et al., 2015). And if predictive models do not prove accurate, researchers can turn the problem the other way around. According to Benson et al., 2016), results at the halfway point of the season faithfully predict team cohesion in the second half of the season.

Eventually, researchers paid attention to the factors related to the individual characteristics of team members, and assessed their impact on performances. Hill et al. (2014) pointed out that an individual trait like team-oriented perfectionism can boost performances. Taking a bird's eye view, Kim et al. (2018) noted a correlation between specific traits and specific roles within a team or group. However, more recent studies, built on the principle that stress management and stress coping are collective processes more than individual ones, analyzed stress activation and stress coping as collective phenomena (Leprince et al., 2018). A few years earlier, previous articles focused on individual perceptions had unveiled a correlation between performance and the level of cohesion within a team, at least according to its coach (Eys et al., 2015). In 2014, Fransen, Vanbeselaere, De Cuyper, Coffee et al. emphasized the impact of team leaders on mutual trust.

Factors Related to the Game Components

To study the impact of several game parameters on collective performance, researchers paid particular attention to small-sided games. When there are only a handful of players in each team, collective tactical behavior is closely correlated to parameters such as the number of players involved, scoring rules (Clemente et al., 2014), power play or short-handed play (Travassos et al., 2014), pitch size and pitch sectorization (Coutinho et al., 2017; Olthof et al., 2018), scoring rules (Almeida et al., 2016), room available to each player relative to pitch size (P. Silva et al., 2015), or constantly-changing rules and limitations of space (Tan et al., 2017). Other researchers opted to study the influence of constraints and configurations proper to a sporting discipline, but in the context of competition: effect on performance of the six regulatory rotations in volleyball (M. Silva et al., 2016) and correlation, in the North American basketball league (aka NBA), between the results obtained by a franchise and the tally of games missed on injury by its players (Podlog et al., 2015).

Competition is a fertile ground to study the impact of game events on collective performance. More precisely, parameters as distinct as performance gaps between the two halves of a football game (M. F. Clemente et al., 2013; Ric et al., 2016), sending-offs (Lago-Peñas et al., 2016; Mechtel et al., 2011; Prieto et al., 2015), or skill balance/imbalance between opponents (García-de-Alcaraz & Marcelino, 2017; Marcelino et al., 2011; Ramos, Coutinho, Silva, Davids, Guimarães, et al., 2017), were submitted to thorough experiments in which performance was measured with a set of indicators, both technical (e.g., ball possession, passes completed) and tactical (e.g., interteam distance variability; Frencken et al., 2012). Some studies were dedicated to individual factors, for instance, unsportsmanlike conduct (Gómez, Lago-Peñas, & Owen, 2016; Gómez, Ortega, & Jones, 2016), or energy expenditure during a game, calculated from the distance covered by each player (Weimar & Wicker, 2017).

Among the other factors directly related to the game, the characteristics of the opposing sides were especially relevant when it came to tactics. Ten years ago, Carling (2011) analyzed how tactical schemes in professional football impacted physical and technical performances, two parameters that in turn impact the display of both teams. For instance, defensive pressure in basketball makes a dent into the opposing side's shooting percentage (Csataljay et al., 2013). Overall performance is also influenced in volleyball by players' location and attacking pace (Ramos, Coutinho, Silva, Davids, & Mesquita, 2017) and, in Australian rules football, by the game's momentum and players' position on the field (Alexander et al., 2019).

However, it's the coaches' behavior that caught researchers' attention the most. Smittick et al. (2018) demonstrated quite clearly that when a coach is rude with his flock, results tend to decline. Conversely, Zavertiaeva et al. (2018) proved that when a football coach is overconfident about his players' skills, they tend to overperform. It comes as no surprise then that when a coach expresses happiness and satisfaction instead of anger or rebuke, their players fare better (Van Kleef et al., 2019). A large chunk of the studies focused on coaches' methods, the efficacy of which is reflected in their teams' results. In a recent article, it was stated that in football, players' confidence in their coach strategy is a good omen (Keatlholetswe & Malete, 2019). That very same trend, found in a lot of different sports, is reflected in several aspects of coaching such as squad turnover (Clay & Clay, 2014; Gómez, Lago-Peñas, & Owen, 2016; Gómez et al., 2017) or requests for timeouts (Sampaio et al., 2013). Although the object of only one study, refereeing decisions may also influence game results (Rodenberg & Lim, 2009).

Finally, we flagged a pair of articles that were noteworthy for their reliance on multilevel analyses (e.g., Filho & Rettig, 2018; Lazarus et al., 2018). In them, researchers resorted to statistics to detect factors that are key to performance and therefore might act as potential predictors. These factors ranged from team characteristics (e.g., average age, weight/height, number of capped players, competitive experience) to competition settings (e.g., number of days off between games, match venue).

Intervention Protocols Acting on Several Factors

This first subcategory includes studies aimed at assessing an intervention protocol applied to a specific group, compared with a control group. Usually, these interventions are one-shots. It was the case in Son et al. (2011), who tested if three different types of self-talk improved self-belief and motor skills: The first type of self-talk was focused upon one's personal abilities; the second one emphasized collective skills; and the third type was a simple series of neutral statements (control group). Having practiced one of the three kinds of self-talk, 80 students then took part in a team event in darts after which their confidence in their own abilities (self-efficacy), and in their teammates (collective efficiency), was assessed. Results showed that individuals who had practiced the self-talk focused upon the group's abilities had higher levels of self-efficacy, collective efficiency, and motor skills. Recently, Campo et al. (2019) produced a similar study in which they acted on the self-abstraction skills of 30 elite rugby players by inducing in them a self- or a team-oriented goal. Comparison between both groups indicated that the individual ability to self-abstract maximized collective and individual performances. This study—the only one that openly manipulated athletes' social identity in order to analyze collective emotions in sports—suggested that social identity, when coupled with team-oriented emotions, could be a key factor of the emotion/performance mix in team sports. However, there was a caveat: The performance indicators presented by Campo et al. were self-reported, therefore subjective.

To get out of that quandary, Fortes et al. (2018) designed a crossover intervention that submitted young volleyball players to an 8-week cognitive imagery training program aimed at improving their decision-making skills. After having asked a group of players to picture themselves executing passes in a context of competition, researchers assessed their decision-making skills during an exhibition game. The experimental group showed signs of improvement, whereas the control group did not.

Intervention Protocols Collecting Athletes' Experience

In other studies, athletes were invited to give their opinion about the intervention protocol they were submitted to. For example, Collins and Durand-Bush (2010) engaged an elite female curling team (four players and their coach) in a 16-week-long self-regulation intervention protocol revolving around cohesion and performance. Eight collective sessions were held, after which the players' opinions were collected in interviews, in a questionnaire, and through participant observation. The results showed that self-control, cohesion, and performance levels all improved during the program, and that their progression could even be traced back in time. Both coach and athletes also reported better performances after the intervention. Athletes' testimonies also provided a benchmark against which the efficiency of new digital training tools could be measured. Vinson et al. (2017) watched how players and coaches in rugby union and field hockey made use of a video-based, online-training platform to monitor teamwork and individual performances. After a round of individual and collective interviews, positive outcomes were reported, namely an upbeat atmosphere and better interpersonal work or self-teaching, which brought the four researchers to the conclusion that their platforms had encouraged all the members of the squad to take part in the analysis of their performances.

To sum up, Figure 5 synthesizes all the methodologies used by sports science researchers to study collective performance.

View Figure

• Download as Powerpoint
• Download Figure