“Chop and Change”: Examining the occurrence of squad rotation and its effect on team performance in top European football leagues

Introduction

Professional football analysis is constantly getting more intricate and detailed with the availability of novel and complex data, all of which are aimed to increase the likelihood of success. ¹ There is considerable work done in developing new Key Performance Indicators (KPIs) that reflect on-field individual and group performance from a physical, tactical and mental perspective.^2,3 Given the need to ensure optimal player performance for team success, one perspective of value is that of squad rotation. This is particularly important with the expected increase in the number of fixtures in the near future, often congested in short time periods. ⁴ This may place high training and performance demands on players, potentially affecting player availability, injury and recovery.^5–8 Despite the potential effects of extended playing time on player availability and load, squad rotation and its effect on team performance have not received as much scientific attention as other performance measures.

One of the rare studies ⁹ that looked at squad rotation and team performance investigated one team's performance over a span of five seasons, evaluating team performance through collective technical aspects (forward passes, final third entries, goal attempts etc.) and tactical measures (final ranking, team points, goal difference, etc.). They found that a championship-winning season exhibited the least number of squad rotations and utilisation, while also experiencing the least number of injuries in that season. Another study that looked only at 31 most-used players from four best-ranked teams in the 2018 FIFA World Cup ¹⁰ found that within the international tournament setup, the most rotations were seen in the third group stage game, mostly after the respective team had won the first two matches and secured their position in the next round of the tournament. While this makes sense in the international competition setup, the same may not be applicable in the domestic league structure where the number of games is much higher, and the relative impact of each game is the same (3 points for a win, 1 point for a draw). Given how continuous involvement in high-intensity training and match play can lead to injuries or player unavailability, as well as changes in on-field performance such as high-intensity runs being replaced by low intensity runs^10,11 studying squad rotations becomes critical from a tactical perspective. Moreover, with some recent work focusing on fixture congestion as well as training load and its effect on injuries and team performance,^12–15 examining squad rotation and depth may be of high value in the ever-growing football calendar.

While the aforementioned studies give vital insights into squad rotations, there is still no body of work that establishes the general patterns of squad rotation and squad utilisation as observed in top European football leagues. Multiple studies find significant relations between league rankings and player availability based on injury rates in specific domestic leagues in France, ⁹ Iceland, ¹⁶ Qatar ⁷ as well as Germany ¹⁷ ; however, the extension of injury rates into squad rotation is still lacking. Long-term rotation and associated strategies have been studied in other sports, notably in basketball ¹⁸ and futsal, ¹⁹ with some evidence to suggest positive effects of rotation when considering the interaction of specific covariates on team performance or game outcomes. While there have been studies in football that have analysed league formats and match scheduling,^20,21 a general, league-wide or multi-season approach to studying squad rotation, interaction variables and team performance has not been examined yet. With this in mind, this study attempts to contribute to this gap in research by being the first study, in our knowledge, to look at the relationship between squad rotation and team performance from a big data perspective. The aim of this study is to establish a foundational understanding of how squads are rotated and utilised in top leagues worldwide, while considering injury rates as well as other contextual variables such as team quality and squad market value. It considers 44 full seasons of competitive football collected from 4 top European leagues that would help provide a general as well as a league-specific understanding about the relationship of squad rotations and team performance.

The section entitled “Methods” discusses data availability as well as data specifications that describe the seasons and variables included in the model, followed by descriptive statistics and results of the linear mixed-model. Finally, implications of general findings, league-specific findings as well as some specific cases of team performance are discussed.

Methods

A dataset of matchday lineups for each fixture in the English, French, Italian and Spanish league seasons was created from 2010–11 to 2021–22. It included a total of 16,720 matches played in each of the aforementioned leagues. The leagues were selected as the top European leagues with the same number of teams and an equivalent load of matches in the execution of the tournament (20 teams, two rounds of 19 matches per team each, with a total of 38 matches per team matches per season). Data was obtained from Transfermarkt. ^a For each match, contextual information about the competing teams, goal differences, and results was collected. The current market value of each starting player was also collected. From this initial dataset, the number of squad rotations was calculated for both teams participating in a match, i.e., the number of different players included in the starting lineup in comparison to the previous match. Since goalkeepers are usually not rotated because of load management, they were excluded from this analysis. Therefore, the number of rotations per team in a match could range from 0 (the exact same lineup from the previous match) to 10 (all players, excluding the goalkeeper, change).

Additionally, all injuries occurring from season 2010–11 to season 2021–22 were obtained, including information about players, their current teams, injury duration, start and end date, and type of injury. Injury types provided by Transfermarkt were further categorized according to Waldén et al.. ²² To focus on overuse-related injuries, contact related injuries (e.g., bone fractures) or injuries caused by other factors (e.g., viral infection) were removed from the dataset. Due to the unusual situation during the Covid-19 pandemic, season 2019–20 was removed from the dataset.

Finally, all data was aggregated at season level for each team, yielding a total sample size of 880 data points derived from 146 teams in 4 separate leagues over 11 full seasons. Team performance was calculated as end of season points (EOSP). Team strength was calculated as the average market value (MV) of the respective seasons’ lineup. Average rotations (AROT) was calculated as the mean of match-to-match number of rotations. Number of injuries was calculated as the total number of injuries, normalized by 1000 h of playing time (IPT).^17,22

Results

Descriptives

To illustrate the nature of the squad rotation metric, Figure 1 shows the distribution of average rotations per match for the teams of each league of the dataset. Figure 1 also documents the distribution of each team's market value over the seasons. Market values show clear growth, particularly in leagues such as England and Spain, with English football consistently showing higher values. The distribution of market values in leagues like Spain and France is skewed, characterized by a few teams — such as Paris Saint-Germain in France and Real Madrid and FC Barcelona in Spain — dominating in terms of market value. Italy and England however, have a more balanced distribution of their team market values.

OPEN IN VIEWER

Figure 1.

Distribution of average squad rotations and market value per team over the seasons. The seasons are depicted on the horizontal axis with the last two digits of their first (starting) year. Season 2019–20 is excluded due to COVID-19.

The season squad rotation average shows minor differences across leagues and disparities among teams, with values commonly ranging from 1 to 5. To further elaborate in this metric, Figures 2 and 3 show the match-by-match squad rotation of two different teams. Figure 2 shows the squad rotation executed by Leicester City in their well-known championship run in the 2015–16 season of the Premier League. In contrast, Figure 3 shows the squad rotation that Manchester City displayed during the 2021–22 season, where they won the title in the last round. Despite both teams managing to win the championship, this metric clearly differs between them. Leicester had long streaks of none squad rotations (i.e., playing with the same exact lineup in consecutive matches), and overall, they did not modify their lineup to more than 3 players until they won the title. Conversely, Manchester City displayed a more flexible rotation management, with a common squad rotation between 2 and 4 players, even in the late rounds while disputing the title against Liverpool.

OPEN IN VIEWER

Figure 2.

Leicester city squad rotation values per round in the 2015–16 season.

OPEN IN VIEWER

Figure 3.

Manchester city squad rotation values per round in the 2021–22 season.

Table 1 shows detailed descriptive statistics of dependent and independent variables included in the linear mixed-model.

OPEN IN VIEWER

Table 1.

Descriptive statistics (EOS points = end of season points; AROT = average rotation; IPT = number of injures per 1000 h playing time ; LogMVD = Logarithm of market value deflated by the seasons and leagues median market value (%)).

Effect	Min	Max	Mean	Median	SD
EOS points	16.00	102.00	52.15	48.00	16.91
AROT	0.68	4.86	2.63	2.61	0.62
IPT	0.00	74.96	12.81	9.57	10.84
LogMVD	−1.67	2.76	0.16	0.00	0.83

Model outputs

Table 2 shows the results of the linear mixed-model predicting the performance based on LogMVD, AROT, and IPT. The average team is estimated to get 56.29 points per season before considering market value, rotations, and injuries. An increase in 1 in LogMVD (100% increase in the seasons median log MV) increases EOSP by 14.57 points. One rotation more, on average, decreases EOSP by 2.54. The interaction between LogMVD and AROT shows no significant but tends towards a positive effect. The effect of injuries and all its’ interactions are not relevant from a predictive perspective.

OPEN IN VIEWER

Table 2.

Model parameters for the linear mixed effects model including all four leagues with 880 observations.

Fixed effect	Estimate	SE	t	CI (2.5%; 97.5%)
Intercept	56.29	2.33	24.12	51.71; 60.86*
LogMVD	14.57	2.93	4.98	8.84; 20.30*
AROT	−2.54	0.88	−2.88	−4.27; −0.81*
IPT	−0.21	0.17	−1.25	−0.53; 0.12
LogMVD:AROT	0.50	1.02	0.49	−1.49; 2.49
LogMVD:IPT	−0.02	0.17	−0.09	−0.34; 0.31
AROT:IPT	0.06	0.06	1.02	−0.05; 0.17
LogMVD:AROT:IPT	0.02	0.05	0.30	−0.09; 0.12

SE: Standard error, CI: Confidence interval, LogMVD: Deflated logarithm of market value, AROT: Average rotations, IPT: Injuries per thousand hours.

Random effect analysis shows that teams intercept vary by 3.66 standard deviations. Figure 4 shows the effect range and confidence intervals of the random intercepts. Each observation represents one team's average random intercept and its confidence interval. Solid markers indicate teams differing significantly from the average team's intercept. These teams are the following, in ascending order (mean ± sd): Aston Villa (−6.60 ± 2.35), Toulouse (−4.36 ± 2.21), Tottenham Hotspur (4.58 ± 2.05), Manchester United (4.63 ± 2.18), Atalanta (4.82 ± 2.14), Napoli (5.70 ± 2.20), Juventus (9.17 ± 2.45), and Manchester City (9.47 ± 2.20).

OPEN IN VIEWER

Figure 4.

Effect ranges of teams’ random intercepts.

Conditional R² = 0.73 for fixed and random effects. Marginal R² = 0.68 for the fixed effect. ICC = 0.16, indicating that a small amount of variance is explained by modelling the team identity as a random effect.

Table 3 shows the four individual models fitted to the data for the four leagues. Since IPT did not show significant fixed effects in the combined model, it was excluded from the league-based analysis. Splitting the data by league reveals a significant interaction between LogMVD and AROT for England and France. AROT and the interaction between LogMVD and AROT are not significant for Italy and Spain. However, for Italy, the confidence intervals for AROT clearly tend towards a negative estimate while the interaction between LogMVD and AROT tend towards a positive estimate. These effects would be in line with the results for England and France. England is the only league that does not show a significant effect of LogMVD. However, the confidence interval tends towards a positive effect.

OPEN IN VIEWER

Table 3.

Model parameters for the linear mixed effects model as observed in different premier division leagues in England, France, Italy and Spain with 220 observations each.

Fixed effect	Estimate	SE	t	CI (2.5%; 97.5%)
England
Intercept	58.99	3.03	19.47	53.05; 64.93*
LogMVD	5.84	4.02	1.45	−2.05; 13.72
AROT	−4.94	1.40	−3.52	−7.70; −2.19*
LogMVD:AROT	6.11	1.71	3.58	2.77; 9.45*
France
Intercept	61.98	3.29	18.82	55.52; 68.43*
LogMVD	9.15	3.35	2.73	2.58; 15.72*
AROT	−4.90	1.26	−3.88	−7.38; −2.43*
LogMVD:AROT	2.64	1.13	2.35	0.44; 4.86*
Italy
Intercept	56.36	3.52	16.01	49.47; 63.27*
LogMVD	10.67	3.64	2.92	3.53; 17.80*
AROT	−2.01	1.18	−1.73	−4.32; 0.30
LogMVD:AROT	2.08	1.18	1.76	−0.23; 4.40
Spain
Intercept	48.11	3.29	14.61	41.66; 54.57*
LogMVD	12.66	3.56	3.56	5.69; 19.63*
AROT	−0.01	1.14	−0.01	−2.26; 2.24
LogMVD:AROT	0.74	1.13	0.66	−1.48; 2.97

Discussion

The aim of this study was to investigate the influence of squad rotations, team strength and injury rate on performance in soccer. The main results show that end of season performance can be explained by the squad rotation and market value. Market value (LogMVD) was identified as a relevant predictor that increases the expected end of season points. Additionally, the results of this study show that an increase in average rotations (AROT) had a clear negative effect on end of season points. Despite not showing a significant effect, the interaction between market value (LogMVD) and average rotations (AROT) showed a positive trend. Therefore, while the overall effect of rotating players in the lineup might be negative for a team's performance, team quality might smooth this effect (e.g., teams with higher market value might have deeper rosters, better substitutes or alternatives in the lineup).

The model including all leagues suggests that performing squad rotations may not show a positive effect on team performance in terms of points, and has a relation to the team quality. This is indicated in the general model by the positive yet non-significant trend between end of season points and the interaction of team market value and average rotations (LogMVD:AROT), and is supported by the results observed when the model is split into different leagues. A positive and significant relation between team market value and average rotation is visible in the English and French leagues. The interaction effect yields 2.6 additional points in France, and over 6 points in England – where teams with the highest market values are found. This clearly indicates the importance of the quality in depth of a roster, and its effect on team performance as well as effectiveness of rotations associated with the roster. As the market value of teams increases, the potential effect of rotations on team performance also shows an increase. On the other hand, the interaction shows a positive yet non-significant effect for Italy and Spain. Differences between leagues are already visible in the development of market values over time (Figure 1). The interaction effects may be affected by the nature and volume of investments as well as policies such as Financial Fair Play (FFP). Therefore, these preconditions should be considered in the future when aggregating football data over multiple leagues and seasons.

This observed relationship may not, however, be prescriptive for all teams or indicate that all teams that win their respective domestic titles would exhibit the highest rotation. The same is exhibited in Figures 2 and 3 which show two title-winning English Premier League clubs: Leicester City's 2015–16 team and Manchester City's 2021–22 team. The average rotations per match show considerable differences, with Leicester City operating on minimal rotations (and waiting until title win to perform maximum rotations) while Manchester City on average shows almost double rotations per match. The pattern observed in the Leicester City title-winning campaign resonates with previous findings ⁹ that show how championship-winning teams prefer not to rotate much if they are not forced by injuries, indicating that other factors such as keeping a winning combination intact in certain contexts could also affect the way a team rotates their players. ³⁰ Manchester City, on the other hand, rotated players regardless of their league position, but also had to manage the additional load of performing in multiple domestic and international competitions, making it to the semi-finals of the domestic FA cup and international UEFA Champions League. This additional load would certainly have an effect on the team selection for specific matches. Further research is required to examine the effect of non-injury forced squad rotation amongst teams that are competing at the top of their respective leagues, while also striving to maintain high performance in multiple competitions at the same time.

Furthermore, for teams that are not operating at the highest budget, or do not have the material resources like their counterparts, there may be some implications worth considering given that there may be no unilateral benefit of rotating players. From a performance analytical perspective, it illustrates contextual factors specific to different teams that coaches, analysts and associated staff must consider when determining training and competitive strategies. This could have an effect on training strategies and injury prevention methods that look to protect starting players more than other squads that may have more expansive material provisions. It may be worth exploring this further from the perspective of training load and injury risk in general,^15,31 but can also be supplemented by research findings regarding position-specific load differences, with certain positions (centre-back) facing less physical load than others, ¹⁴ which could be used to inform training strategies.

In the present model, neither the main effect of injuries per thousand hours (IPT) nor its’ interaction effects were identified as relevant predictors. Table 1 shows an average of 12.81 injuries per 1000 h of match time. This is similar to previous research ¹⁷ that finds an average of 13.1 injuries per 1000 playing hours of match play. It must be noted that this injury occurrence differs from training injuries, and includes only overuse or trauma injuries, and estimates of injury occurrence would vary when including training and other injury types. The synthesis of the injury assessment of teams over a season through the metric injuries per thousand hours of match play did not show a statistically significant relationship with end of season points.

Including this variable in the analysis was motivated by being able to disseminate between teams with highly flexible lineups and tactics and teams that elevated their number of rotations due to injury reasons. Thus we decided to include an injury indicator at team level utilising the per 1000 h match time as determined by previous literature.^17,22 However, this metric does not account for the duration of the injuries (a player not available for one week or twelve weeks), it only measures the number of times a player got injured. Therefore, results must be critically analysed with regard to the definition of both rotation-based and injury-based metrics. The rotation variable was computed by average per match (between lineup of previous match and current match) while injuries are counted over an entire season and at load ratio (i.e., match time), which could affect the relationship observed in the model. It must also be noted that the Injuries per thousand hours (IPT) variable would impact performance in the same way for all teams regardless of country or league. In order to get a clearer picture of injuries, average rotations and end of season points, the measurement of injuries and rotations may require greater synchronisation and detail: for example, rotations could be further classified into types of rotations like forced (through injury), tactical (not forced by injury) and unrelated rotations (other reasons, e.g., disciplinary), such that injuries are integrated into the lineup changes every match. This would require considerable manual processing of the current dataset and subjective categorisation, but could be a valuable future direction of research.

Random effect analysis of teams shows that the teams differ significantly in their intercept. In this model, two teams (Aston Villa and Toulouse) reached significantly less points than the average team regardless of their market value, average rotations or injury rate (LogMVD, AROT, and IPT). On the other hand, six teams (Manchester City, Juventus, Napoli, Atalanta, Manchester United, and Tottenham Hotspur) reached significantly more points. Considering the overperforming teams are all well known for their high performance in the last decade, it seems unlikely that this is created by chance. Especially the high gap between Manchester City and Juventus indicate that such teams are able to even overperform their preconditions based on market value, squad depth, and injury rates. This may be explained by several factors on and off the pitch, like coaches’ decision, clubs’ infrastructure, or social environment not captured in this model, alongside psychological variables such as team reputation. Generally, more researchers should incorporate random factors, like teams, to uncover more detailed insights into the analysis of sports performance.

Conclusion

With the aim to establish a fundamental understanding of how squad rotations occur over the course of multiple seasons across top European football leagues, this study examined the relation between a team's average rotations, market value and injuries with team performance in terms of end of season points. Linear mixed-model results for all leagues (England, France, Italy and Spain) combined suggest that rotations in general did not have a positive effect on end season points, while team market value exhibits a significant positive relation. Notably, when looking at results for different leagues, a positive and significant relation with end season points is observed for the interaction variable team market value and average rotations in the English and French leagues. Overall, the results suggest that a positive effect of rotations on team performance is more likely to occur for those teams that operate at a higher market value. Different leagues exhibit different levels of impact, with the English Premier League showing the most pronounced effect in terms of end of season points. Additional factors such as involvement in international and domestic competition, causes of rotation such as tactical or injury-forced, as well as their relation to psychological and sociological variables like opponent reputation, club infrastructure and coaching philosophy must be studied in greater detail to supplement our current understanding of how and why squad rotations occur in top European football leagues.