In the present study, 153 matches of the 2020/21 German Bundesliga season were analyzed. The data basis was official tracking and event data. Due to existing restrictions regarding the accessibility of the match data, only the second half of the season could be analyzed in this study. The tracking data consisted of X and Y data of all 22 players on the pitch and the ball and were recorded by a semi-automated optical tracking system (TRACAB, ChyronHego, Melville, NY, United States). This system was recently considered valid [validity (spatial precision of position measurement): 0.07–0.18 m RMSE (root mean square errors); between device reliability of total distance: −0.15 m ± 0.37%] (Linke et al., 2020). The event data were raised manually by Sportec Solutions (Sportec Solutions AG, Ismaning, Germany) and the definitions of the events were based on an official checklist (Deutsche Fußball Liga (DFL), 2019). Tracking and event data were synchronized by matching the respective time-point of the tracking data for every event using the algorithm of Forcher et al. (2022b). All data processing and analysis were executed using Python 3.9 with the NumPy, Pandas, and Matplotlib libraries.

The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the local ethics committee (Human and Business Sciences Institute, Saarland University, Germany, identification number: 22–02, 10 January 2022).

To quantify the offensive playing style (i.e., ball possession or counter-attacking style) of each team in every match, we used and further developed a playing-style formula of Kempe et al. (2014). Therefore, we conducted a principle component analysis to weigh the physical and technical parameters within this already existing formula. This procedure allows a weighting of the parameters according to their importance concerning the classification into an offensive style of play. This new weighted formula is subsequently referred to as the playing style coefficient [PSC]. The PSC elevates the offensive style of play on an offensive playing style continuum between ball possession and counter-attacking style. While high PSC values are associated with a focus on ball possession, low PSC values are associated with a focus on counter-attacking.

The study by Yi et al. revealed that the offensive playing style significantly influences the ball possession rate of a team (Yi et al., 2019). Since the ball possession rate influences match performance (e.g., high-intensity running profile, number of passes), the present study examines all dependent variables solely during the ball possession of the respective team (Bradley et al., 2013; Mota et al., 2015). Therefore, similar to Goes et al. (2018) and Forcher et al. (2022b), ball possessions were defined as a phase where one team is controlling the ball. A possession ended with either the opponent gaining ball control or a stoppage of play (i.e., foul, offside, goal, final whistle, ball out of bounds). The dependent variables were examined at a team level and were categorized into three performance parts [physical (accelerations, decelerations, high-intensity distance, sprinting distance, and all four physical variables in relation to attacking time), technical (percentage of short passes, percentage of medium passes, percentage of long passes, percentage of horizontal passes, percentage of backward passes, dribblings, the success rate of all six technical parameters, and passing velocity), and success (xGoals, goals, and points)].

Firstly, the physical variables acceleration, deceleration, high-intensity distance, and sprint distance per attack were collected. Similar to Rhodes et al. predetermined thresholds for accelerations (> 3 m/s²) and decelerations (< −3 m/s²) were used (Rhodes et al., 2021). The high-intensity distance was defined as the distance where running speeds between 19.8–25.0 km/h are reached and the sprint distance with speeds above 25.0 km/h (Bradley et al., 2011; Tierney et al., 2016; Aquino et al., 2019; Borghi et al., 2020; Arjol-Serrano et al., 2021). Moreover, since it can be assumed that ball possession-oriented teams have longer ball possessions per attack, all physical variables were not only used as absolute values but also normalized based on attacking time and subsequently included as additional parameters.

Secondly, the technical variables passes and dribblings were raised. Additionally, for each technical parameter, the success rate was determined. Dribbling was recorded if a player in safe ball control tried to dribble past an opponent. Dribblings were considered successful if the respective player managed to dribble past the opponent. Furthermore, based on their distance, passes were categorized into short (<10 m), medium (10-30 m), and long (>30 m) (Forcher et al., 2022d). In addition, passes were classified backward or horizontal according to their playing angle (see Supplementary Figure S1). Since the results of Yi et al. suggest that ball possession oriented teams play more passes per attack, all passes were analyzed in relation to the total number of passes (Yi et al., 2019). Moreover, the average velocity of a pass was quantified. Passes were rated as successful when the ball reached a teammate.

Thirdly, as success-related variables points per match (0 = loss, 1 = draw, 3 = win), goals scored, and expected goals [xGoals] were recorded. xGoals were estimated after the definition of the German football league (Deutsche Fußball Liga (DFL), 2019).

Furthermore, the offensive tactical formation was captured by deploying the formation description algorithm by Forcher et al. (2022b). It clusters the average positions of all players into three formation lines (e.g., 4–4-2). The offensive formation represents the tactical distribution of all players on the pitch and is only measured for the team in possession (Forcher et al., 2022c).

PSC ( p l a y i n g s t y l e c o e f f i c i e n t ) = ( PA ∗ 0.35 ) + ( FP ∗ − 0.03 ) + ( TP ∗ 0.23 ) + ( PS ∗ 0.32 ) + ( FPS ∗ 0.32 ) + ( BP ∗ 0.32 ) + ( DPA ∗ 0.34 ) + ( RAT ∗ − 0.32 ) + ( MAT ∗ 0.35 ) + ( RD ∗ − 0.24 ) + ( MPA ∗ 0.35 ) .

PA ≙ Number of passes of one offensive action.

FP ≙ Number of passes forward in relation to the overall number of passes subtracted from 1.

TP ≙ Number of passes to a target player in relation to number of overall and non-target player passes.

PS ≙ Number of successful passes in relation to the overall number of passes.

FPS ≙ Number of successful passes forward in relation to the overall number of passes forward.

BP ≙ Sum of all periods of possession of one team in relation to the sum of the periods of possession of both teams.

DPA ≙ Distance covered during all attacks in relation to the total number of attacks.

RAT ≙ Mean time of the attack of the opponent subtracted by the own mean time of the attack.

MAT ≙ Relation of the total time of all attacks to the number of attacks.

RD ≙ Relation of the distance covered within one attack to the time with ball possession.

MPA ≙ Relation of the total number of passes to the total number of attacks.

Since the PSC is a new formula, it was examined a priori for its validity. To evaluate the PSC, the results of the PSC for every match performance, the results of the formula of Kempe et al. (2014), and the results of a formula based on an expert rating were compared. The formula based on an expert rating was developed by weighting the individual parameters based on the rating of three licensed and experienced coaches of a professional club. All three raters independently rated the parameters according to their importance for the quantification of the offensive playing style with the help of a questionnaire (i.e., each variable could be classified as important, neutral, or unimportant). To compare the results of the three calculations, all 18 included teams were sorted in a table (i.e., from ball possession to counter-attacking focused) based on their average values (i.e., average score over all 17 matches). Before executing the three alternating calculations, the values were transformed into z-scores. To compare the table results of the three calculations, a Spearman rank correlation was calculated between the tables based on the results of all three formulas (see Supplementary Table S1). As the results between the PSC, the previously evaluated formula by Kempe et al., and the formula based on the expert rating showed a high degree of agreement (ρ = 0.93–0.97; 95% CI = 0.77–0.99; p < 0.01), the PSC was assessed as valid.

For each dependent physical (e.g., sprinting distance), technical (e.g., dribblings), and success (e.g., xGoals) variable a single repeated measures linear mixed model was conducted using the statsmodels library in Python 3.9. The value of the PSC served as the fixed effect for each model. Hence, the PSC is the independent variable used to predict the respective dependent variable (i.e., physical, technical, or success variable). Each physical variable was examined in absolute form and in relation to the attacking time.

A hierarchical modeling strategy was implemented, following the example of Fernandez-Navarro et al. (2018). Therefore, random effects (i.e., team, offensive formation) were added step by step for each model independently. Hence, depending on the model, a different number of random effects were the consequence. The data structure was hierarchical, as, for example, all teams are ranked higher than one single team (Heck et al., 2014).

To evaluate the model performance, the Akaike criterion [AIC] was used (i.e., lower AIC values = better model). Furthermore, restricted maximum likelihood (REML) estimation was implemented for model fitting. The statistical significance level was set a priori at p < 0.05.

All physical match performance parameters examined in this study were significantly influenced by the offensive style of play. In detail, with increasing PSC values (i.e., emphasis on ball possession) the number of accelerations and decelerations, as well as the distance in high-intensity and sprinting speeds per match increased. Moreover, Yi et al. found similar results, indicating higher high-intensity and sprinting distances for ball-possession style teams (Yi et al., 2019). However, it should be noted that Yi et al. investigated the physical match performance of a whole match (i.e., also during opposing ball possession). Accordingly, the offensive playing style of ball possession is associated with an additional physical effort for the players. It is important to highlight, that since physical match performance increases with effective playing time (Altmann et al., 2023), it can be assumed that the increased attacking time regarding the ball possession style (see Supplementary Table S3, e.g., team 17) is the reason for the higher physical match performance of ball possession-oriented teams.

Furthermore, high-intensity and sprint distances decreased with increasing PSC value (i.e., emphasis on ball possession style), when analyzing the distances normalized at the attacking time. One could conclude that after gaining the ball, teams with an emphasis on counter-attacking style have to cover a large distance at high speeds in transition to get in front of the opponent’s goal. This has to happen as quickly as possible to use the short time when the opposing defense is disorganized. In contrast, teams with a focus on ball possession have more time since they face an orderly opponent and try to disorganize him with several successive passes. Therefore, the distance to the opponent’s goal can be covered with lower speeds. However, the number of accelerations and decelerations normalized by attacking time remains the same comparing both ends of the playing style continuum. This relationship reveals that irrespective of the offensive playing style, short high-intensity actions (e.g., accelerations and decelerations) are necessary to get in goal-threatening situations.

In terms of the technical match performance, the offensive playing style influenced the technical parameters to varying degrees. On the one hand, the percentage of short passes, the percentage of medium passes, the average passing velocity, the number of dribblings, and the success rate of dribblings are influenced by the offensive playing style to only a small extent. Accordingly, it can be stated that teams of both extreme ends on the playing style continuum play a similarly high percentage of short and medium passes. One possible explanation for this finding is that short and medium passes are used very frequently regardless of the style of play (see Supplementary Table S4). Both styles of play also go along with a similar amount of dribblings. There seems to be no difference in the amount of dribbling, as both playing styles use dribblings only to a small extent (see Supplementary Table S4).

On the other hand, the percentage of long passes is strongly influenced by the offensive playing style. With a growing focus on counter-attacking (=PSC values decreased), the proportion of long passes increased. As explained above, counter-attacking teams try to quickly bridge the space to the opponent’s goal in transition play. To optimally achieve this objective, counter-attacking teams play more long passes. In contrast, ball possession teams try to control the ball throughout longer periods (see Supplementary Table S3, e.g., team 1). Since long passes increase the risk of losing the ball (see passing success rates), ball possession-style teams play fewer long passes to reduce the risk of losing ball control.

The abovementioned conclusion is also supported by the results for the passing success rates. For almost all passes (e.g., medium, backward), teams with a focus on ball possession revealed better success rates than counter-attacking style teams. This finding is supported by the results of Yi et al., who similarly found better passing success for ball-possession style teams (Yi et al., 2019). As already explained, a poorer passing rate leads to more ball losses and consequently less control of the match. Consequently, ball possession-oriented teams need to have high-quality passing success rates to control the match by possession and thus allow little possession time for the opponent.

Moreover, with a rising focus on a ball possession style (= higher PSC values), the percentage of horizontal passes increased. As counter-attacking style teams try to exploit the disorganized opponent directly after gaining ball control, they have to cover the long distance to the opponent’s goal with not only long but also vertical passes (see Supplementary Table S3, e.g., team 15). Teams with a focus on ball possession pursue the approach of destabilizing a defensively organized opponent through targeted passing (Forcher et al., 2021). Therefore, a larger percentage of the passes needs to be played horizontally, for example, to enable lateral shifts to destabilize the opponent and hence receive scoring opportunities.

In contrast to physical and technical match performance, the influence of offensive playing style on success-related parameters remained small. Concluding, values that are strongly influenced by chance, such as goals and points (Brechot and Flepp, 2020) (i.e., points awarded for the match outcome, e.g., three, two, or zero), are not influenced by the style of play. In contrast, findings by Yi et al. suggested that a focus on a ball-possession style is associated with an increased probability of success (Yi et al., 2019). In the context of the present study, the only success-related parameter significantly interacting with the PSC was xGoals. In detail, there was a slight tendency for teams with a greater focus on ball possession to achieve more xGoals than teams with an accent on counter-attacking. A possible explanation for this result could be related to the two different playing styles and their objectives. A common observation in professional soccer suggests a higher focus on defensive play for teams playing against a stronger opponent. Therefore, these weaker teams (i.e., in relation to the opponent) strongly focus on the objective of not conceding a goal. Consequently, their offensive play is limited to a few counter-attacking opportunities, which decreases the chance of realizing a large number of scoring opportunities (i.e., potentially low xGoals value). The opposite scenario can be observed with teams playing against a weaker opponent (i.e., focus on offensive play potentially leads to more scoring opportunities). Thus, these stronger teams (i.e., in relation to the opponent) could be associated with ball-possession and weaker teams with a counter-attacking style which possibly leads to the observed difference in xGoals (Kempe et al., 2014).

To obtain a complete picture of the present study, the limitations should be considered in the following. Since the used data is from the Bundesliga and the dependence of match performance on country and league is confirmed, the transfer of results and conclusions is limited (Rampinini et al., 2007; Dellal et al., 2011). In addition, only a continuum of two offensive playing styles was considered in the survey of different styles of play (Fernandez-Navarro et al., 2016). This may represent a simplification of reality. Furthermore, to strengthen the significance of the findings in this study an already validated formula was used to determine the style of play. However, in this calculation, various technical and physical parameters were included and, therefore, it cannot be precluded that those variables are independent of the dependent physical, technical, and success variables used in this study. Moreover, the opponent was not considered in the present study. Since performance in soccer arises from the interaction between the two teams, the opponent should be considered in future studies. This leads directly to the topics for future research.

Fruitful avenues for future studies could be to examine the parameters collected on an individual level and, for example, consider other contextual factors influencing soccer match performance (e.g., playing position, quality of the teams) (Forcher et al., 2022d). Furthermore, an investigation of the influence of defensive playing style on match performance during defensive play could complete the picture of the current study (Forcher et al., 2022a).