FormalPara Key Points

There has been an increase in research examining performance in set plays (free kicks, corner kicks, penalty kicks) in the last few years that have provided valuable information on variables that influence their effectiveness.

Investigations into match activity profiles have evolved to include the characterization of effort during congested fixture periods, in competition with extra time periods and effective use of substitutions.

Specific collective system measures (e.g. a team’s geometric centre and dispersion) and analysis techniques (e.g. network, sequential and temporal-pattern analyses) have provided important information about how teammates and opponents interact during performance so that tactical behaviours can be better understood.

1 Introduction

A systematic review of research articles that were published before 2011 provided a timely overview of the most common research topics, methodologies and evolutionary tendencies of research in Association Football [1]. More recently, studies have discussed current approaches to tactical performance analysis in football [2,3,4].

In the last 5 years there has also been some progress in terms of books dedicated entirely or partially to football [5,6,7], with one peer-reviewed journal, Science and Medicine in Football (Taylor and Francis Group) emerging as a stand-alone journal after 3 years of publication as a regular supplement of the Journal of Sport Sciences (Taylor and Francis Group).

Additionally, match analysis as a methodological approach in sports science has progressively grown, based on proliferation of technological systems (e.g. global positioning system [GPS], Prozone—STATS, OPTA) to collect performance data. Interpretation of the data seeks to generate knowledge about team properties and the patterns that characterize their organization [8], with implications for coaches and sport analysts to design practice strategies and plan training sessions [9].

Progression of match analysis research in Association Football has increased exponentially since 2011, and recent literature since that date can offer new insights to the field if theoretically and systematically organized and interpreted. Systematically reviewing research published in refereed journals contributes in several ways, such as (1) informing researchers about the evolution of knowledge on match analysis; (2) the characterization of new techniques for gathering new information; and (3) offering an evolving theoretical organization of the key topics and concepts researched in performance analysis in football.

The purpose of this article was to conduct a systematic review of published articles on match analysis in adult male football, identify and organize common research topics, and synthesize the emerging patterns of work between 2012 and 2016, predicated on findings from the previous review by Sarmento et al. [1] of studies published up to 2011.

2 Methods

2.1 Search Strategy: Databases, Inclusion Criteria and Process of Selection

A systematic review of the available literature was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. The search strategy followed by Sarmento et al. [1] was adopted in the current study.

The Web of Science electronic database was searched on 6 January 2017 for relevant articles published between 1 January 2012 and 31 December 2016, using the keywords ‘football’ and ‘soccer’, each one associated with the terms ‘match analysis’, ‘notational analysis’, ‘game analysis’, ‘tactical analysis’ and ‘patterns of play’.

The inclusion criteria for these articles were (1) including relevant data regarding technical and tactical evaluation or statistical compilation, and time–motion analysis; (2) participants included amateur and/or professional adult male footballers; and (3) the articles were published in English. Studies were excluded if they (1) involved children or adolescents (under 18 years of age); (2) included females; (3) did not include relevant data for this study; and (4) were conference abstracts. If there was disagreement among authors regarding the inclusion of certain articles, a discussion was held until a consensus was found.

Two independent reviewers (HS, FC) independently screened citations and abstracts to identify articles potentially meeting the inclusion criteria. For those articles, full-text versions were retrieved and independently screened by two reviewers to determine whether they met the inclusion criteria. Disagreements about whether the inclusion criteria were met were resolved through discussion with the other authors, who then analyzed the full text of the papers that induced doubt in the two main reviewers who performed the initial screening process. In two cases, direct communication with the authors of the original articles helped in the final decision-making process. In this way, all final decisions resulted from a process of joint decision making.

2.2 Extraction of Data and Quality of the Studies

To evaluate the quality of the studies, a risk-of-bias quality form was adapted for the specific context of research developed in match analysis, from the original version developed by Law et al. [10], following an evaluation process by five senior researchers with substantial experience (including relevant publications) in soccer performance analysis. Some minor suggestions were introduced in the final critical review form (16 items) according to their evaluation (see Electronic Supplementary Table S1).

Articles were assessed based on purpose (item 1), relevance of background literature (item 2), appropriateness of study design (item 3), sample studied (items 4 and 5), use of informed consent procedure (item 6), outcome measures (item 7 and 8), method description (item 9), significance of results (item 10), analysis (item 11), practical importance (item 12), description of dropouts (item 13), conclusions (item 14), practical implications (item 15), and limitations (item 16). All 16 quality criteria were scored on a binary scale (0/1), wherein two of those criteria (items 6 and 13) presented the option: ‘If not applicable, assume 3’. The introduction of this option for items 6 ‘Was informed consent obtained?’ and 13 ‘Were any dropouts reported?’ has been included because, in some studies, the investigators were not required to obtain informed consent (item 6) or report dropouts (item 13). The introduction of the option ‘not applicable’ allowed an appropriate score for the article, eliminating the negative effect of assuming the value ‘0’ on a binary scale, when in fact that specific item was not applicable to that study. As in previous research [11, 12], to make a fair comparison between studies of different designs, the decision was taken to calculate a percentage score as a final measure of methodological quality. For this, the sum of the score of all items was divided by the number of relevant scored items for that specific research design. All articles were classified as (1) low methodological quality, with a score ≤ 50%; (2) good methodological quality, with a score of between 51 and 75%, and (3) excellent methodological quality, with a score > 75%.

A data extraction sheet (adapted from Cochrane Consumers and Communication Review Group’s data extraction template; available at http://cccrg.cochrane.org/author-resources) was developed and tested with 10 randomly selected studies. First, one researcher extracted the data from included studies and a second researcher then checked the extracted data. Disagreements were resolved by consensus.

3 Results

3.1 Search, Selection and Inclusion of Publications

The initial search identified 483 titles in the described database. These data were then exported to reference manager software (EndNote X8), and any duplicates (189 references) were eliminated automatically. The remaining 294 articles were then screened according to the title and abstract for relevance, resulting in another 156 studies being eliminated from the database. The full text of the remaining 138 articles was read and another 61 were rejected due to a lack of relevance for the specific purpose of the current study. At the end of the screening procedure, 77 articles received further in-depth reading and analysis for the systematic review (Fig. 1).

Fig. 1
figure 1

Preferred reporting items for systematic reviews flow diagram

Full size image

The main reason for exclusion was that a published study did not concern match analysis (n = 23). Other reasons for exclusion included (1) participants were youth players under 18 years of age (n = 10); (2) involvement of female players (n = 8); and (3) data were included from other team sports (n = 20), including rugby, futsal, handball, Australian Football, basketball, volleyball, field hockey, frisbee, floorball and waterpolo.

Sarmento et al. [1] reviewed 53 articles published up to the end of 2011. Interestingly, in a much shorter period (2012–2016), the data revealed an increase (n = 77) in the number of studies published on the selected topic.

3.2 Quality of the Studies

Sarmento et al. [1] justified the quality of the papers included in their revision due to the use of Web of Science as the search database. In contrast, in the current review, we evaluated the quality of the papers included for analysis. The results of the interobserver reliability analysis, calculated using the Kappa index, was 0.97 (95% confidence interval 0.97–0.98), indicating very good agreement between observers. The quality of indicators for the included papers was as follows: (1) the mean methodological quality score for the 77 selected articles was 89.8%; (2) 12 articles achieved the maximum score of 100%; (3) none of the articles scored below 50%; (4) two articles scored between 50 and 75% (good methodological quality); and (5) 75 articles achieved an overall rating of > 75% (excellent methodological quality).

Possible deficiencies identified in the 77 studies were mainly related to two items on the criteria list: (1) for criterion 16 (see Electronic Supplementary Appendix S1), some studies failed to clearly acknowledge the limitations of the study; and (2) some studies lacked information in relation to criterion 5, related to an explicit justification of the study sample size.

3.3 Data Organization

The previous systematic review by Sarmento et al. [1] categorized research according to the type of analyses performed (descriptive analysis, comparative analysis and predictive analysis), and type of variables analyzed. In contrast, the present review grouped studies according to major match analysis research topics (categories) that emerged from the detailed analysis (Fig. 2). This approach was adopted in order to contribute to a theoretical knowledge based on an ecological dynamics framework [7], without losing the bottom-up knowledge that emerges from the systematic analysis of studies review.

Fig. 2
figure 2

Scopes of match analysis

Full size image

Two independent reviewers (HS, FC) independently classified the papers according to the different major research topics. Disagreements were resolved through discussion with the other co-authors until a consensus was found. The aim was not to produce categories that were mutually exclusive since the same analysis can include topics that relate to different categories. Thus, an article included in a specific ‘category’, could also be classified in another ‘category’ whenever its content justified it.

3.4 Major Research Topics

The following subsections describe the studies identified in each of the specific research topics. Their findings are outlined and discussed in more detail in Sect. 4.

3.4.1 Set Plays

Set plays, such as free kicks, penalty kicks, corners and throw-ins, can provide match-winning situations. In the last few years, there has been an increase in research examining these match events [13] in different competitions (Table 1). Several studies have estimated that between 30 and 40% of goals are scored from set plays [14]. The importance of this type of situation is highlighted by professional coaches who identified an increased systematization of specific set-play training situations as an evolutionary trend of training/competition [15]. Given its importance to the match, dead-ball specialists work on the training ground to perfect their techniques; however, it is important that defenders also be prepared to face set plays. ***

Table 1 Studies with predominantly corner kicks, penalty kicks and free kicks analysis
Full size table

3.4.2 Activity Profile

3.4.2.1 Playing Roles

Modern professional soccer imposes more and more demanding requirements on players that relate to their pre-competition preparation [16], specifically to their roles on the field (i.e. goalkeepers, defenders, midfielders and forwards). There is a need to understand how roles performed in soccer can affect performance, as estimated by specific variables (Table 2).

Table 2 Studies with predominantly activity profile analysis—playing roles
Full size table
3.4.2.2 Fatigue Influence

Match activity and fatigue during football matches has been a topic of increased research in recent years (Table 3). Recent research has shown that physical performance during a match changes throughout the season and is related to players’ training status [17]. Previous research has also revealed the evolution of the game over the last few decades, especially the increasing intensity of play [18] that is directly related to physical performance decrements from the first to the second half of elite soccer match play [19]. These trends could result in an inability of players in certain roles to repeatedly cover distances during critical situations, and may also reduce technical capabilities that are related to match outcomes [20].

Table 3 Studies with predominantly activity profile analysis—fatigue influence
Full size table
3.4.2.3 Substitutions

Substitutions have enormous impact and importance in modern football contexts because coaches typically attempt to use well-timed substitutions to reduce the effects of fatigue across the team or in an effort to modify tactics. Consequently, some important research (Table 4) has analyzed this specific aspect of the match in the English Premier League [20], Spanish Professional Soccer League [21] and Union of European Football Associations (UEFA) Champions League [22]. However, little scientific evidence is available and has only recently been introduced in the scientific community [20,21,22].

Table 4 Studies with predominantly activity profile analysis—substitutions
Full size table
3.4.2.4 Altitude and Environmental Heat Stress Influence on Performance

With the aim to examine the effect of altitude and environmental heat stress on football performance, some interesting studies (Table 5) have been developed with the national teams that participated in the Fédération Internationale de Football Association (FIFA) 2010 [23] and 2014 World Cups [24].

Table 5 Studies with predominantly activity profile analysis—altitude and environmental heat stress influence on performance
Full size table

3.4.3 Variables Capturing Group Behaviours

3.4.3.1 Team Centre

The team centre represents the geometric centre of a set of points that represent the current positioning of soccer players on field during competition [25]. This measure has been used to analyze the collective spatial–temporal synchronization between competing teams and to identify instabilities or transitions that might lead to the emergence of critical moments in a competitive match, such as an opportunity to score a goal [26, 27]. A summary of the studies that have examined the dynamics of the collective variable ‘team centroid’ (geometrical centre of the players, excluding goalkeeper and not considering the position of the ball) can be found in Table 6.

Table 6 Studies of group behaviour analysis with the use of team centroid values
Full size table
3.4.3.2 Team Dispersion

The quantification of how far players are apart (dispersion) may help in understanding the nature of the space that emerges from the interactive dynamics of competitive matches, helping us to identify the natural expansion and contraction of soccer teams in attacking and defensive moments. Dispersion of the team can be estimated by calculating the area covered by all points of the team (surface area), identifying the dispersion of players from the geometrical centre (stretch index), calculating the Euclidean distance between each player and his/her teammates (a team’s spread) or analyzing the effective defensive triangulations (effective area of play and defensive play area) [25, 28, 29]. These measures have been used to observe the oscillations of the areas of the team during attacking and defensive moments and to compare their variability throughout competitive matches (Table 7).

Table 7 Studies of group behaviour analysis with the use of the team’s dispersion
Full size table
3.4.3.3 Team Interaction/Coordination Networks

The communication process may occur in different ways, but, in match analysis, these have been used to classify the interactions between teammates when in possession of the ball, during passing sequences [30]. Social network analysis based on graph and digraph theories has been used to classify general properties of the network and specific centrality levels of players (nodes) [31, 32]. Identifying networks on the field may help in understanding the specific relationships that emerge between teammates during attacking sequences and the general properties of collective team performance during passing sequences. A summary of the studies that analyzed the networks of soccer players can be found in Table 8.

Table 8 Studies of group behaviour analysis with the use of network
Full size table
3.4.3.4 Sequential Patterns

Sequential patterns analyze sequential combinations of interactions that emerge between players during a match or a set of matches [33]. Some studies (Table 9) have analyzed the attacking patterns of teams in different competitions, considering criteria such as duration of an attacking sequence of play, number and role of players involved in the attack, zone of pitch where the actions were performed, type of technical behaviours, and the number and co-location of players of both teams (interaction context) in the space adjacent to the ball on field [34, 35].

Table 9 Studies of group behaviour analysis with the use of sequential patterns
Full size table
3.4.3.5 Group Outcomes

Winning, drawing and losing may be constrained by, or constrain, some match-related statistics [36,37,38]. Based on these assumptions, some studies (Table 10) have been conducted to identify the variance that might exist between some playing actions and the final outcomes of a match.

Table 10 Studies of group outcomes
Full size table

4 Discussion

The aim of this paper was to systematically review the evolving patterns of match analysis in Association Football to organize research studies, published between 2012 and 2016, in a theoretically coherent way. After in-depth analysis, it was decided that the most appropriate way to discuss the results would be to categorize research topics according to similar themes (set plays, group behaviour and activity profile).

4.1 Set Plays

4.1.1 Corner Kicks

The reviewed studies that analyzed corner kicks were mainly focused on international competitions [14, 39] and the English Premier League [40, 41]. Corner kick effectiveness values of 2.6% [39], 2.2% [14], 4.1% [40] and 2.7% [41] were found, which means that, on average, between 24 and 45 corner kicks were needed to lead to a single goal scored. Casal et al. [14] reported that the likelihood of a shot on goal could be increased with the involvement of three or four attackers, a dynamic attacking move, and delivery of the ball to the far post. Pulling et al. [40] analyzed the importance of defensive strategy and concluded that the one-to-one marking set-up did not concede any attempts at goal from 95.7% of corner kicks, whereas a zonal marking system did not concede goal attempts from 97.7% of corner kicks. In addition, the percentage of corner kicks resulting in a goal or attempt at goal was higher when the defending team used a one-to-one marking system (31.3%), compared with a zonal marking system (30.2%). The investigators highlighted that, although this finding may suggest that a zonal marking set-up is better for defending corners, the percentage difference between these systems is very small. Variables such as the area (e.g. zones inside the penalty area) where the ball was delivered, the type of delivery (long- and short-corner kicks), and the influence of some situational variables (e.g. teams performed more short corners and took more short kicks and outswinging corner kicks when winning, but outstep and inswinging corner kicks when drawing and losing) [39] in the strategies used to perform the corner kick were also analyzed.

Data from the reviewed studies suggest that coaches should design training sessions that simulate the execution of more elaborate corner kicks that involve a short initial kick, followed by a dynamic interaction involving three or four players, before the ball is crossed to the far post. Regarding defensive strategies at corner kicks, there seem to be few differences in the effects of using zonal marking versus one-to-one marking systems. Additionally, coaches should be aware that the positioning of players on the goalpost(s), when defending corner kicks, does not significantly prevent goals from being scored. Rather this tactic actually increases the frequency of attempts on goal by the opposition. These findings suggest that the players positioned at goalposts could be ‘used’ to carrying out other defensive functions.

4.1.2 Penalty Kicks

The penalty kick is a peculiar event involving a direct confrontation between two opponents directly functioning in a dyadic system—the penalty taker and goalkeeper [13]. It is one of the most pressured and intense moments in a competitive match. In male professional football, approximately 70% of penalty kicks are scored [42].

The studies reviewed [13, 42,43,44] could help penalty takers and coaches improve their chances of successful outcomes as they provide information suggesting that (1) the areas of the goal to which the ball is aimed is significantly important for penalty effectiveness [13, 42]; (2) saves depend mainly on the goalkeeper’s reaction time but also on the ball speed in the penalty kick [42]; (3) situational factors (e.g. period in the match) may influence the success of penalty kicks [13]; (4) goalkeepers should wait longer in order to dive to the side of the goal to which the ball has been kicked [13]; and (5) penalty takers should use both a keeper-independent strategy and keeper-dependent strategy in order to increase their chances of success [43].

4.1.3 Free Kicks

Despite the importance of set-piece goals in modern football, free kicks have not been extensively studied [45, 46]. The study by Link et al. [46] revealed an average of 34.9 ± 7.6 free kicks per match, while Casal et al. [45] concluded that, on average, each team takes three indirect free kicks aimed at scoring a goal per match. Of these, 21.8% ended in a shot, 9.3% ended in a shot between the posts, and 2.9% ended in a goal. The type of attack and the number of players involved in the process has a direct influence on the outcome. Furthermore, Link et al. [46] analyzed variables such as position (2D-location of the free kick on the field) and zone (free-kick location in the attacking third on the field of play (35 m from goal), according to a specific categorization by the authors of the playing area. This included density (number of free kicks in each 1 m2 sector on the field), interruption time (time span between the foul that led to the free kick and the moment of ball contact when taking the free kick), distance to defensive wall (shortest distance between the ball and the defensive wall at the moment of ball contact), number of players participating in the wall, rule violation, type of play (shot on goal, cross, pass), and outcome shots (goal scored, header, save made by the goalkeeper, etc.). However, studies of this specific event remain rare and more research is needed to better understand the influence of different variables in the effectiveness of the free kick.

Nevertheless, the reviewed scientific evidence suggests that coaches could design specific training sessions that aim to improve the effectiveness of free kicks. They could facilitate players working on elaborate kicks, with the ball being played along the ground and involving interactions between three or four players (trying to reach the opposing team’s penalty area using short passes and dribbles).

4.2 Activity Profile

4.2.1 Playing Roles

The relationship between a player’s positional role and performance continues to be frequently studied [16, 47,48,49,50,51,52]. However, in their analysis, Sarmento et al. [1] concluded that previous investigations had grouped players according to different criteria, which made it difficult to accurately compare results regarding player roles.

In line with previous research findings (see Sarmento et al. [1] for a review), the results confirmed that midfielders covered the greatest average distance, followed by attackers, and then defenders. However, Clemente et al. [48] proposed an alternative way of conducting this analysis (distance that each player covered per minute) involving distance covered in possession of the ball and distance without the ball. This revision helped in understanding the running pace of players (m/min) during the match. Moreover, using such an approach, it was possible to compare players who played less frequently (minutes on field) with those who played more. This relative measure based on time allowed us to make comparisons between all players in the competition, independent of their playing time. The results showed that the greatest distances, in possession of the ball, were achieved by midfielders, followed by forwards. Significant differences were also observed between defenders who played wider (more distance covered in possession of the ball) and central defenders. Without the ball, midfielders covered the greatest distances during play.

Additionally, Andrzejewski et al. [50] found that the mean total sprint distance covered by professional soccer players in the UEFA Cup (≥ 24 km.h−1) amounted to 237 ± 123 m. The sprint distance covered by forwards was the highest (345 ± 129 m), 9% greater than midfielders (313 ± 119 m), and over 100% greater than the same value for central midfielders (167 ± 87 m). Elite footballers performed an average number of 11.2 ± 5.3 sprints per match, of which 90% were shorter than 5 s duration and only 10% were longer than 5 s. The results also revealed that forwards and wide-playing midfielders performed a far greater number of short duration sprints compared with central midfielders and central defenders. In addition, wide-playing defenders (e.g. wing backs) performed the highest number of long-duration sprints, differing significantly from central midfielders, who performed the fewest sprints.

A common mistake made by coaches preparing players for performance is application of the same workload to all players during training sessions. The reviewed studies identified specific physical load profiles for football players during a match, dependent on their specific playing positions, which can be used to design highly individualized training programmes for specific players.

4.2.2 Fatigue Influence

Interest in fatigue in football has mainly focused on the impact of a congested fixture list [53, 54], extra time periods [55, 56] on player performance, and variations in performance across a whole season [17]. Additionally, some researchers [57] investigated the fatigue rates and pacing strategies of players during matches by quantifying high-intensity running in rolling 5-min periods.

Studies of extra time clearly showed a greater decrement in physical performance markers during this period [55, 56]. Penas et al. [55] found that performance decrements affected players in all roles to a similar degree. Additionally, all of the physical markers under study showed a decline of 15–20% during the extra-time period in comparison with the first-half performance, and an increase in low-intensity activities in the second half. Russell et al. [56] verified that between 105 and 120 mins, acceleration and deceleration parameters reduced by > 10% compared with the opening 15 mins.

Dellal et al. [54] examined three different congested fixture periods (six matches in 21 days) and concluded that physical activities and technical performances were unaffected during these periods. Nevertheless, injury rate during match play was significantly higher during congested periods in the fixture list. This difference between training and match play can be explained by the low-intensity training promoted by coaches, the greater emphasis on recovery training programmes in the modern game, as well as players regulating their activity. The results reported by Soroka and Lago-Penas [53] are in line with data reported by Dellal et al. [54], although their study focused on a smaller congested period (three consecutive matches separated by 4 days).

It is noteworthy that the study by Dellal et al. [54] surpassed the limitations of previous research that investigated sporadic congested fixture schedules and only analyzed physical performance across two or three consecutive matches within a short time frame. They concluded that the overall distance covered was greater in the third period (October–November) of the season, whereas no differences were observed in the other speed thresholds (first period—August; second period—September). No studies have examined whether physical and technical activities decreased or varied according to stages of the season. However, a study by Silva et al. [17] examined match activity and the development of fatigue during competitive soccer matches in different periods across a whole season. They reported an association between muscle strength and power, and performance decrements in match-related physical parameters. Their results highlighted the importance of incorporating specific exercise programmes to improve the athletes’ strength and power during performance.

The reviewed studies highlighted different fatigue-related mechanisms related to physical performance decrements throughout the duration of a normal match (i.e. 90 + mins), the duration of extra time (i.e. 120 min) in match play, or during a congested fixture period (e.g. when three games may be played in 1 week). The available knowledge seems to be useful for technical and medical staff in their implementation of specific strategies to minimize performance decrements during a match or a congested fixture period. Such strategies would include (1) specific exercise programmes to improve the athletes’ aerobic capacity during the performance of soccer-specific activities; (2) nutritional supplementation protocols; (3) low-intensity activities during training sessions and adequate rotation of players in congested fixture periods; (4) use of objective markers of fatigue combined with subjective measures of performance; and (5) adaptation of tactical strategies.

4.2.3 Substitutions

As a limited resource for tactical interventions, substitutions are assumed to be important in football, although little scientific evidence is available on this issue [20,21,22]. Gomez et al. [21] concluded that most of the first and second substitutions occurred during the final third of the match (between 61 and 90 mins), while the third substitution occurred predominantly during the final quarter of the match (76–90 mins), in the Spanish first division. Conversely, in the English Premier League and UEFA Champions League, a large number of substitutions occurred at halftime and between 60 and 85 mins [20] and 57 and 78 mins [22], respectively.

The most substituted position is the central midfielder, followed by forwards, wide midfielders, fullbacks, and central defenders. Additionally, ‘like-for-like’ substitutions (the same playing position for player in and player out) were the most common, and the defensive and offensive substitutions showed similar distributions [21]. Substitutions became more attack-minded as the second half progressed [20].

Regarding match performance characteristics, Bradley et al. [20] found that the same players displayed more high-intensity running when they were introduced as substitutes compared with the equivalent period of the second half, but not the first-half period when tracked from the start of the match. The distances covered by high-intensity running were greater for attacking substitutes. These results were interpreted according to perceived tactical options and specific physical demands of playing roles.

The effects of situational variables on timing and tactics of substitutions have also been analyzed [21, 22]. Rey et al. [22] presented a decision-tree analysis that could be used to inform UEFA Champions League coaches by using the following heuristics: if losing a match, make the first substitution prior to the 53rd min, make the second substitution prior to the 71st min and make the third substitution prior to the 80th min; if the scores are tied or the team is ahead, make substitutions at will.

The results of these investigations can provide valuable information so that coaches could optimize player and team performance, but more work needs to be undertaken to investigate the overall impact of substitutes on physical and technical indicators, and their contribution to key moments in matches [20,21,22].

4.2.4 Altitude and Environmental Heat Stress Influence on Performance

Based on the assumption that exposure to altitude and environmental heat stress has a detrimental impact on exercise performance, some researchers have investigated the effects of these on football players [23, 24]. Nassis [23] conducted a study that examined the effects of altitude on soccer performance during the 2010 World Cup in South Africa. The study verified a 3.1% decrease in the total distance covered by teams during matches played at 1200–1400 m and 1401–1753 m, compared with sea level. Through the analysis of environmental heat stress in the 2014 FIFA World Cup (Brazil), Nassis et al. [24] concluded that there was no difference in playing time (average of both teams), total distance traveled (m/min/player), number of goals scored and number of cards issued by referees, compared with matches played under different environmental stress categories (e.g. the risk of heat stress at 50% relative humidity is ‘high’ for wet-bulb globe temperature (WBGT) 28–33 °C, ‘moderate’ for WBGT 24–28 °C and ‘low’ for WBGT < 24 °C). High-intensity activity was lower under high compared with low environmental stress, and the rate of successfully completed passes was greater in the former compared with the latter.

4.3 Group Behaviour

4.3.1 Team Centre

Different terminologies (centroid, i.e. the geometrical centre of players, excluding the goalkeeper and not considering the position of the ball; wcentroid, i.e. geometrical centre of the team that attributes weight to the teammates based on their proximity to the ball; team centre, i.e. geometrical centre of players, excluding the goalkeeper and not considering the position of the ball) have been used to describe the team centre, which has been defined as the geometric centre of a team, considering the positioning of all players on the pitch [25, 27, 58]. The present review identified three main approaches for centroid analysis in soccer: (1) centroid of the team, calculating the geometric centre without goalkeepers [27, 59]; (2) weighted centroid, considering the proximity of each player to the ball as the weight to move the centroid [25]; and (3) the centroid, considering the middle point between the two teammates furthest apart [60]. Team centre has been used to assess intra- and inter-team coordination in soccer in a temporal sequence [27].

In most cases, an in-phase relationship between the competing teams’ centre values during competitive performance has been investigated [26, 27, 61]. The study of elite European soccer suggested that team centroids moved synchronously both up and across the pitch [26]. Similar results in the variable relative phase were found in the final of the 2006 FIFA World Championship [62]. However, in the case of small-sided games (5 vs. 5), specific moments during performance may constrain the synchrony between teams, which can quickly turn into non-synchronization, and even a crossing of team centres that may relate to specific events in the match [63].

The idea of critical moments in competitive performance (e.g. goals, shots) was examined in competitive matches [26, 27]. The study of elite European soccer teams during official competitive 11-a-side matches [26] contradicted the evidence reported from observations of performance in small-sided games that some goals occurred at moments of non-synchronization or of crossing between centroids [63]. In a different approach, the investigators of the study of performance in small-sided games investigated the hypothesis that inter-team variability would indicate critical moments in a competitive match [27]. However, the results from that study suggested that inter-team distances (differences between teams’ centroids) were minimally related to emergence of critical moments in a match [27]. One possible explanation for this finding is that, during small-sided games, it is easier to remain close to other players and for the geometric midpoints of both teams to overlap.

The relationship between competing teams’ centroids may provide information about the synchronization of the teams and identify when non-synchronization of team centroids may lead to critical events in a match. The distance between teams may also be used to design small- or medium-sided games that better simulate specific subphases of the game (e.g. direct attacking, indirect attacking) based on the dimensions of the pitch.

4.3.2 Team Dispersion

The dispersion of a team can be defined by quantification of distances between teammates. Dispersion of the players on the pitch can be constrained by specific strategies and tactics that emerge during the match [25]. Regularly, players are more disposed to explore width and length of the pitch when attacking to exploit free space and to destabilize the opposition defense [64]. Conversely, distances between teammates tend to be smaller when defending, to optimize cover and to contract space [28]. To examine some of these suggestions, some measures have been used, including (1) stretch index; (2) weighted stretch index; (3) Frobenius norm (team’s spread); (4) surface area; (5) effective area of play; (6) playing area; (7) team length and width; and (8) defensive play area and triangulations.

Stretch index can be described as the mean dispersion of players from the team centre (non-weighted) [58, 59]. This measure can be quantified as the radius or only by the width or length axes. A similar concept was introduced by changing the weight of the centroid [8]. In a study of high-level European soccer teams, smaller stretch index values were reported for defending teams, compared with attacking teams [26]. Nevertheless, the data did not consistently associate stretch index values with goals scored [26]. Using the same measure in an analysis of a competitive match, the evidence suggested that the variability of approximate entropy values decreased progressively during the match, with the exception of the transition from the last 15 min of the first half to the first 15 min of the second half [59]. It was also reported that the stretch index was greater in home teams in most of the match-time observed [59].

The weighted stretch index was also used to analyze performance variance between halves of the match [25]. It was found that values of dispersion were smaller in the second half. In addition, the same group of researchers reported a greater weighted stretch index during drawn matches and no statistical differences between losing and winning situations [65].

The Frobenius norm was used as a measure of a team’s spread in four studies [26, 28, 66, 67]. In a study conducted on Brazilian teams, it was suggested that greater values of spread when defending were associated with the emergence of critical moments, such as shots on goal conceded. When attacking, this variable was greater when a team was closer to the opponent’s goal [28]. However, the evidence reported on spread in attacking phases of play was not confirmed in a study of elite European teams [26]. More recently, a predominant in-phase relationship (linear association of both teams’ spread over time) was observed between the spread of teams and anti-phase periods at critical moments of play, such as when shots at goal emerged or goals were scored [67].

Surface area has been used as an alternative dispersion measure that uses the convex hull to determine the polygon generated by all players [63]. In most cases, it was found that surface area values were greater when attacking than when defending [25, 68]. It was also found that surface area was greater when a team competed against weaker teams [68] and when the scores were level in a match [65]. In a single-match study, there was a progressive tendency of this measure to reduce in variability during the match [59].

The effective area of play was introduced as an alternative measure to the surface area, providing a notion of effectiveness in the defensive triangulation generated when defenders covered each other [25]. Using this measure, a significantly smaller space was covered in defence compared with attack, and there was a tendency for the effective area of play to be greater when the scores were level compared with when a team was losing or winning in matches [65].

In an alternative approach, playing area was introduced to verify the mean area covered per player and to identify the best value to use in designing small-sided games [69]. Individual playing area varied between 81.38 and 86.78 m2 in observed competitive matches, with the variability in these values influenced by the location of the ball in specific zones of the pitch. An individual playing area of 90 m2 in small-sided games was suggested to develop build-up play or attacking play in the finishing phase, whereas 80 m2 for small-sided games developed transition play [69].

Defensive play area was introduced as a tactical measure that determines the area covered by a group of players when defending [70]. It was found that the triangular positioning relationships generated when midfielders covered each other on the field were greater and, for that reason, the defensive playing area in the midfield area was significantly greater than upfront or at the back [70].

Dispersion measures can be directly used to evaluate the space required for defensive and attacking processes, and to adjust the playing dimensions and format for small- or medium-sided games designed in training. The right measure of dispersion can also be used to identify the width and length of attacking and defensive phases of play in teams. Such information can be used to characterize team performance and to help coaches plan better strategies to exploit opposition weaknesses or to reinforce playing patterns in their own team.

4.3.3 Team Interaction/Coordination Networks

The network process can quantify the centrality level of a player (individual values per player), dependence between players (meso-level of analysis) and the general properties of a graph (that quantify a value of a specific network property of a team). General network properties have been studied in association with team performance variables such as shots, goals and successful outcomes in competition [30, 71, 72]. High passing rates were related to an increase in team performance and greater centralization was associated with a decrease, defined by the number of goals scored in an analysis of 760 matches from the English Premier League [30]. It has been reported that winning teams display statistically greater levels of general network measures, such as density, homogeneity or number of total links, with small-to-moderate associations with goals scored, overall shots taken and shots on goal in 64 matches from the 2014 FIFA World Cup [71].

Differences between centrality levels of players (individual level of analysis) have also been investigated using the social network approach [31, 73]. Similar evidence was found in a specific analysis conducted on the Spanish national team during the same international competition [31]. In a study conducted on one team using the centroid measure, it was reported that the left back tended to be the dominant player during attacking sequences [74].

Variance of centrality measures between playing positions has also been analyzed [32, 72]. Midfielders were classified as the most prominent players after observations of 64 official matches from the 2014 FIFA World Cup, independent of the specific team and tactical format used [32]. The specific analysis conducted on the German team revealed that midfielders had greater levels of intermediation (capacity of a player to link two or more teammates to each other during play on field) and dominance (capacity to be the player who most often participates in team networks) [72].

Full passing sequences have been analyzed in most cases; however, in an alternative study, only the network interactions in passing sequences that resulted in scored or conceded goals were analyzed [75]. The results suggested that attacking midfielders and wing forwards were the most prominent players for receiving the ball and the right back was the dominant player for passing. Analysis of pitch zones revealed that central and wing regions closer to the goal being attacked were mainly influential in network interactions during attacking phases of play that led to goals being scored [75].

Identification of prominent players engaged in specific types of playing interactions may be used by coaches to adjust performance strategies. In understanding defensive behaviours, a coach can identify the most prominent opponent and which players are well linked to him. Based on the information gained, the coach may design a strategy to mark a key player or to prevent teammates from passing the ball to him/her. Moreover, knowledge of interactions between teammates can be used to identify how they cooperate and can be used to classify the main networks in the team.

4.3.4 Sequential Patterns

A temporal pattern can be described as a repeated temporal and sequential combination of the same order of events during a period, which are relatively invariant [33]. This kind of analysis, when focused on the sequence (and temporality) of events, supports the detection of patterns of play that have higher probabilities of occurrence than chance. The basis for any prediction model is that performance is repeatable, to some degree, suggesting that events that have previously occurred will occur again in some predictable manner.

Temporal analysis of attacking play has been adopted in many studies [34, 35, 76]. Performance criteria analyzed have included the lateral position (the pitch is split into three longitudinal areas—right, centre and left), zone (ultra-defensive, defensive, central, offensive and ultra-offensive), possession (ball in play), interaction contexts (specific regions of the pitch in which interactions between players emerge), recovery and loss of ball possession and time that the ball is out of play [34, 35].

In a study of performances of FC Barcelona in the Spanish national league and the UEFA Champions League, match-to-match and half-to-half patterns were reported [34]. Sixty-eight patterns were recognized during 10 matches observed. One of the identified patterns was an attacking structure that begins in the central defensive zone and progresses to the wings before entering the offensive zone. Another pattern was an attack that begins in the central defensive zone, progresses to the left side, moves back to the centre line and attacks again from the same side of the field [34]. A comparison of attacking patterns of play in a top Italian league team, when winning and losing matches, was analyzed [35]. Overall, 167 patterns emerged in 80% of the 19 matches studied. A greater volume of temporal patterns (n = 101) emerged when the team was losing matches, compared to when they were winning (n = 9). It seemed that, in winning matches, the team was more likely to continue using the same playing pattern [35]. The variance of temporal patterns between halves was investigated, again in an elite Italian team [76]. A greater number of temporal patterns were found in the second halves compared with first halves. More patterns were found [59] and the length and level of passing sequences and their patterning were greater. Five playing patterns were observed in the first half, and nine in the second. Temporal patterns also revealed at least one shot for each pattern (in the second half), whereas in the first half, no such evidence was observed [76].

The study by Sarmento et al. [33], using sequential analysis, is the only study in this updated review that involved the expert opinions of professional coaches to interpret the data. This close relationship between researchers and professionals can be very fruitful for interpreting data in match analysis.

Identification of temporal patterns of play may provide information about structural behaviours that are independent of opposition play. Such patterns can be used by coaches and analysts to identify strategies to negate opposition strengths or to verify congruence between performance behaviours worked on in training and their execution in competitive games.

4.3.5 Group Outcomes

Relationships between match-related statistics and match final scores have been analyzed by several investigators [36,37,38]. A study of 177 matches from the FIFA World Cups in 2002, 2006 and 2010 found that the total number of shots and shots on target were the main discriminatory variables to predict winning, losing and drawing matches [36]. Shots were also confirmed as the main discriminating variable for winning teams [77]. Moreover, a study conducted on 1900 Spanish league matches revealed that match winning teams displayed a greater number of ball recoveries and tended to perform longer passing sequences [78]. The patterns of ball recovery are also important to discriminate match-winning teams from those who lost and drew matches [77].

An interesting observation from a study of 12 Spanish league matches revealed that match outcome influenced match-related statistics [79]. Teams who drew and won matches showed a decrease in the probability of reaching the penalty area in possession of the ball, in comparison to when they were losing a match [79]. In terms of defensive playing patterns, it was found that losing teams tend to defend in more advanced pitch zones; however, more successful teams tend to be more efficient in defensive pressure and ball recovery [80].

Identification of specific key indicators or use of modelling methods may provide information for coaches to re-prioritize playing styles and to also help re-design training exercises to include indicators of successful performance.

4.4 Contextualizing Performance

Contextualizing performance has been a concern of researchers in this field of study. Although it is possible to categorize studies according to ‘major topics’ of research, in reviewed studies investigators analyzed different variables (e.g. work rate, technical behaviours, ball possession) according to contextual variables, including (1) match half [25, 57, 81, 82]; (2) quality of opposition [68]; (3) match location [21, 22, 38, 80, 83]; (4) scoring first [38, 83]; (5) group stage vs. knockout phase [22, 83]; (6), intervals of 5 [57] or 15 min used to record data [38, 82]; (7) timing and tactical nature of substitutions [21, 22]; (8) competitive level [51, 84, 85], and (9) different competitions (different leagues and cups) [86].

Interpretation of player behaviours and match outcomes in specific contexts may help identify specific strategies or training designs that coaches could incorporate to prepare the team for different opposition strategies, circumstances and game scenarios.

4.5 Limitations

A possible limitation of this systematic review is that it only includes studies from the Web of Science that were written in English, thereby potentially overlooking other relevant publications in other languages.

5 Conclusions

Research on match analysis in adult male football players has been the subject of growing interest in the past 5 years. Nonetheless, some limitations remain in the published studies between 2012 and 2017, namely the lack of operational definitions and conflicting classifications of activity or playing positions that make it difficult to compare similar groups of studies. Additionally, some potential weaknesses may be apparent in more recent published research, such as the small sample sizes used in some studies. Nevertheless, researchers have developed new methods in order to better contextualize the performance of players and teams, which is likely to be essential for planning and application of training loads in modern professional soccer.

A progressive increase in group analysis based on positional data is one of the main new insights compared with the previous systematic review by Sarmento et al. [1]. Positional data can be used to identify patterns of interaction between teammates and to explore the spatiotemporal patterns that emerge from a match [9, 87]. Team synchrony has been analyzed based on an in-phase relationship between teams using the measures of centre and dispersion, suggesting regularities in the dynamics of competing teams and some disturbances that emerge at specific critical points of the matches (e.g. goals, shots, counter attacks) [26, 66]. The new collective measures reviewed in this article could help identify the need for specific training conditions for the collective organization of a team, improving the efficacy of practice task design to augment the cognitions and perception of players regarding specific tactical behaviours [69]. Analysis of interactions between teammates may reveal collective properties that cannot be captured by players’ individual movements. For example, synergetic properties of the team can offer theoretical guidance to capture system properties such as dimensional compression, patterns of interpersonal linkage, reciprocal compensation or degeneracy [88]. In summary, the collective measures identified in this review provide different information than that typically gained from traditional notational analyses. Knowledge about the spatiotemporal relationships formed by players during competitive performance may explain some behaviours that notational analysis cannot quantify. The exact positioning of players, occupied space and values of interpersonal distance can be more easily and objectively measured by collective measures, using tracking or GPS systems. Moreover, t-patterns (defined as a particular set of event types recurring in the same order with significantly similar distance values between them) and network measures may be used to classify and rank players based on their importance in a competitive game, identifying specific interactions. Both of these methodologies can complement the use of notational analysis.

These novel measures require new measurement techniques, and the complexity engendered during soccer matches requires an integrated approach that considers multiple aspects of performance [89]. A big challenge for researchers is to align these new measures with the needs of the coaching staff, through a more interactive relationship between all practitioners, to produce practically relevant information that can improve performance through constant adaptations of training design. Reductionist methods and approaches should be avoided and multifactorial analyses must be conducted, integrating notational methods and computational collective measures to amplify knowledge and identify long-term patterns in performance dynamics during competition. The association between outcomes (notational analysis) and processes (spatiotemporal analyses) may also contribute to identify which patterns can be avoided or reinforced to increase possibilities for success. Additionally, future studies should promote real-world insights into optimal methodologies for player preparation through integrating sources of information about training requirements, periodization load, structure of the competition, and player fitness and fatigue. Collecting and measuring a large volume of data (e.g. positional, physiological, psychological, environmental conditions, etc.) in real time, and compressing it into a smaller set of variables, providing objective information for coaches that facilitates, to some extent, the prediction of performance outcomes, seems to be a useful path in this specific area. An augmented perception analysis framework for football (ARCANE) [90] represents an interesting first step that may explain how to achieve this significant goal.