Professional Documents
Culture Documents
Journal of Sports Sciences
Journal of Sports Sciences
To cite this article: Carlos Lago & Rafael Martn (2007) Determinants of possession of the ball in soccer, Journal of Sports
Sciences, 25:9, 969-974, DOI: 10.1080/02640410600944626
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Journal of Sports Sciences, July 2007; 25(9): 969 – 974
Abstract
In research on the importance of the possession of the ball in soccer, little attention has been paid to its determinants. Using
data from 170 matches of the 2003 – 2004 Spanish Soccer League, we explain why differences in the possession of the ball
among teams are so great. In particular, four variables are examined: evolving match status (i.e. whether the team is winning,
Downloaded by [Universite De Paris 1] at 09:11 25 July 2013
losing or drawing), venue (i.e. playing at home or away), and the identities of the team and the opponent in each match.
Results of linear regression analysis show that these four variables are statistically significant and together explain most of the
variance in possession. In short, home teams have more possession than away teams, teams have more possession when they
are losing matches than when winning or drawing, and the identity of the opponent matters – the worse the opponent, the
greater the possession of the ball. Combinations of these variables could be used to develop a model that predicts possession
in soccer.
Correspondence: C. Lago, Facultade de CC da Educación e o Deporte, Av. Buenos Aires s/n, 36002 Pontevedra, Spain. E-mail: clagop@uvigo.es
ISSN 0264-0414 print/ISSN 1466-447X online Ó 2007 Taylor & Francis
DOI: 10.1080/02640410600944626
970 C. Lago & R. Martı́n
Table I. Categorization of different performance indicators that have been used in notational analysis of soccer.
Dependent variable
status i.e. whether the team was wining, losing or
drawing). They found that successful and unsuccess- The dependent variable was the percentage of time
ful teams had longer periods of possession in in minutes during a match in which one team had
Downloaded by [Universite De Paris 1] at 09:11 25 July 2013
matches when they were losing than when they were possession of the ball when in play (POSSESSION:
winning. Similarly, Pollard (1986) and Thomas, PO). For example, a possession of 50% means that a
Reeves, and Davies (2004) propose that playing at team had possession of the ball for half of the time it
home or away is important. Many coaches believe was in play. Possession was deemed to have been
that the style of play of each team (‘‘direct play’’ or gained when a player had sufficient control of the ball
‘‘possession play’’) is an important factor in explain- to effect a deliberate influence on its subsequent
ing performance. path. Possession was lost when the ball went out of
The main aim of this paper is to examine causal play, an opposing player touched the ball or the
mechanisms behind teams’ possession of the ball in referee blew the whistle for an infringement (Jones
soccer. Why does one team have more possession et al., 2004). When interpreting the results, positive
than the other in a match? Why does one team have or negative coefficients indicate a higher or lower
more or less possession in two games? The goals of propensity respectively to have possession. Teams’
the research are two-fold. The first is to develop a possession data used in the research were supplied by
theoretical model to explain the variability of this GECA SPORT, a private sector company dedicated
performance indicator. In particular, four variables to assessing the performance of teams in the Spanish
are considered: evolving match status (i.e. whether Soccer League (information on the data set can be
the team was wining, losing or drawing), venue (i.e. found at www.gecasport.es).
playing at home or away), and the identities of the
team and the opponent in each match. Our
Independent variables
hypotheses were as follows: (1) home teams should
have more possession than away teams; (2) teams Five independent variables were included in the
should have more possession when they are losing research. The first is ‘‘playing at home or away’’:
than when they are winning or drawing; and (3) the PLAYING AT HOME (PH). It is a dichotomous
poorer the opponent in a match, the greater the variable: 1 ¼ the team is playing at home, 0 ¼ the
possession. The second aim is to demonstrate how a team is playing away. The anticipated effect of this
combination of these variables can be used to predict variable is positive: home teams should have greater
teams’ future possession in soccer. possession than away teams. Research into home
advantage across different sports has received a lot of
attention in the sport psychology literature in the
last 20 years (Agnew & Carron, 1994; Bray, 1999;
Methods Bray & Widmeyer, 2000; Moore & Brylinsky, 1995;
Wright, Voyer, Wright, & Roney, 1995). Pollard
Sample
(1986) defined home advantage as ‘‘the number of
The sample consisted of 170 matches played in the points won at home [ . . . ] expressed as a percentage
first 17 days of the 2003 – 2004 Spanish Soccer of all points gained’’ (p. 239). For example, for
League. The number of observations was 340, seasons 1888 – 1990 in the First Division of Football
because data from both teams in each match were League in England in which 2630 games were
recorded. This strategy allowed us to double the played, 1536 were home wins, 592 were home
number of observations. defeats, and 502 were draws, giving a home
Determinants of possession of the ball in soccer 971
(ML). Similarly, the third variable measures the time þ b5 ðMDi HTi Þ þ b6 TEi þ b7 OPi þ ei
each team was drawing in the match: MINUTES
ð2Þ
DRAWING (MD). For example, if team A scores a
goal in the tenth minute and team B equalizes in
the final minute, team A has been loosing for 0 When estimating both models, no evidence of
minutes and drawing for 10 minutes; while team B heteroscedasticity in residuals or multicollinearity
has been losing for 80 minutes and drawing for 10 among regressors was found (King, Keohane, &
minutes. Verba, 2000). Moreover, the RESET test of Ramsey
Fourth, possession is also a function of the identity (1969) did not reveal specification problems. RE-
of teams. Nineteen dummy variables were created to SET stands for Regression Specification Error Test.
identify each of the 20 teams in the Spanish Soccer The classical normal linear regression model is
League: TEAM (TE). The reference category was specified as:
Real Madrid because when the data were collected it
led the championship. Therefore, the interpretation y ¼ Xb þ e
of the dummies for the rest of the teams is easier.
The choice of the reference or omitted category has The error vector e is presumed to follow the
no affect on the substance of the regression results. multivariate normal distribution N(0,s2I). Specifi-
The only components that change are the standard cation error is an omnibus term that covers any
errors and the interpretation of the t-tests. The departure from the assumptions of the maintained
goodness of fit, the coefficients of the other variables, model. Serial correlation, heteroscedasticity and
and the t-statistics for the other variables do no alter non-normality all violate the assumption that the
(Greene, 1997). errors are distributed [N(0,s2I)]. Tests for these
Finally, teams’ possession also depends on the specification errors have been described above. In
opponent. Consequently, the variable OPPONENT contrast, RESET is a general test for the following
(OP) was been created to identify each of the types of specification errors:
possible opponents in a match. The reference cate-
gory was Real Madrid again. Table II provides . Omitted variables: X does not include all relevant
descriptive statistics for the variables. variables.
Standard
Mean deviation Minimum Maximum Observations
. Incorrect functional form: some or all of the Table III. Determinants of possession in the Spanish Soccer
variables in y and X should be transformed League.
to logs, powers, reciprocals or in some other Model
way.
. Correlation between X and e, which could be Dependent variable 1 2
caused by measurement error in X, simultaneity Playing at home (PH) 5.672* 3.813*
or the presence of lagged y values and serially (0.487) (0.899)
correlated disturbances. Minutes losing (ML) 0.091* 0.089*
(0.011) (0.010)
Under such specification errors, Ordinary Least Minutes drawing (MD) 0.045* 0.023***
Squares estimators will be biased and inconsistent, (0.011) (0.014)
and conventional inference procedures will be PH * MD – 0.042**
invalidated. (0.017)
Intercept 43.02* 44.09*
(1.941) (1.645)
R2 0.65 0.66
Results and discussion
Number of observations 340 340
The results are displayed in Table III.
Downloaded by [Universite De Paris 1] at 09:11 25 July 2013
These results provide evidence of home advantage Table IV. Simulated possessions for Real Madrid in the match
in soccer. This is in line with the findings of Pollard Real Madrid vs. Barcelona.
(1986) and Thomas and colleagues (2004). Minutes losing
Minutes
drawing 0 15 30 45 60 75 90
The impact of evolving match status
on teams’ possession 0 48.69 50.06 51.43 52.81 54.19 55.56 56.83
15 49.36 50.73 52.11 53.49 54.86 56.23
The variable time losing was significant in both 30 50.04 51.41 52.78 54.16 55.53
models at the 0.01 level and had the anticipated sign: 45 50.71 52.08 53.46 54.83
teams have greater possession of the ball when they 60 51.38 52.75 54.13
75 52.05 53.43
are losing than when they are winning or drawing. In 90 52.73
line with the results of the two regression models,
every 11 min losing increases possession by 1%. For
example, if team A was losing for 90 minutes the
predicted possession would be 8% higher than that of
the opponent’s possession (90 minutes losing 0.09). Table V. Simulated possessions for Real Madrid in the match
These results are similar to the findings of James Barcelona vs. Real Madrid.
et al. (2004). Teams (successful or unsuccessful)
Downloaded by [Universite De Paris 1] at 09:11 25 July 2013
Minutes losing
have greater possession when they are losing than Minutes
when they are wining. In support of this finding, drawing 0 15 30 45 60 75 90
Blommfield et al. (2005) report that possession is 0 43.02 44.39 45.76 47.14 48.51 49.89 51.26
influenced by the score. 15 43.69 45.06 46.44 47.81 49.18 50.56
30 44.37 45.74 47.11 48.48 49.86
45 45.04 46.41 47.79 49.16
The impact of the team and the opponent 60 45.71 47.09 48.46
on teams’ possession 75 46.38 47.76
90 47.06
Regarding the variables team and opponent, 14 teams
differed at the 0.01 level when compared with the
reference team, Real Madrid. For the other teams,
differences can be rejected. It is likely that differences
in teams’ possession are due to the tactics and the Finally, to illustrate the findings, estimates of actual
style of play adopted. Possession is affected by the and simulated possession for a match between Real
score, but different teams appear to follow different Madrid and Barcelona under different scenarios are
strategies (retaining more or less possession) that displayed. What possession would be predicted for
reflect the individual style of coaching and manage- Real Madrid when the evolving match status differs?
ment, the characteristics of the players, team forma- Is it similar when Real Madrid plays at home or away?
tion, and philosophy of play based on tradition of the In Tables IV and V, different possibilities for each
clubs (Hughes & Franks, 2005). Blommfield et al. variable of the regression model are included. For
(2005), for example, showed that the top three teams example, in the match Barcelona vs. Real Madrid, the
in the Premier League in the 2003 – 2004 season predicted possession for Real Madrid differs signifi-
(Chelsea, Arsenal, and Manchester United) domi- cantly according to match status (by 8.3%). If the
nated possession against their opponents whether final result were 1 – 0 to Real Madrid and they scored
winning, drawing or losing. Hughes and Franks the goal in the first minute (0 min losing and 0 min
(2005) suggested that because successful (League drawing), their possession would be 43.0%. If
champions, World champions, European cham- Barcelona won 1 – 0 and scored the goal in the first
pions) teams do not resort to ‘‘direct play’’, there minute, the possession of Real Madrid would be
are patterns of play for successful and unsuccessful 51.3%. Which of these scenarios is realistic? Nobody
teams. For the 1986 World Cup Finals, Hughes, can know. But, if the variables that affect teams’
Robertson, and Nicholson (1988) found that suc- possession are known, coaches and analysts could
cessful teams had more touches of the ball during implement more effective practice.
each period of possession than unsuccessful teams.
It would be useful in the future if teams
Conclusions
were grouped according to their style of play (‘‘direct
play’’ or ‘‘possession play’’) and if the homogeneity of The main findings of the empirical analysis reported
estimated coefficients was checked. Our hypothesis is here are as follows. First, teams’ possession depends
that teams’ possession would differ depending on the on the evolving match status – that is, whether the
style of play of the two teams in each match. team is winning, losing (P 5 0.01) or drawing
974 C. Lago & R. Martı́n
(P 5 0.01). Teams have greater possession of the ball Hadley, L., Poitras, M., Ruggiero, J., & Knowles, S (2000).
when they are losing than when they are winning or Performance evaluation of National Football League teams.
Managerial and Decision Economics, 21, 45 – 56.
drawing. Second, playing at home increases posses- Hughes, M. D. (1993). Notational analysis of football. In
sion by 6% compared with playing away (P 5 0.01). T. Reilly, J. Clarys, & A. Stibbe (Eds.), Science and football II
These results support home advantage in soccer. (pp. 151 – 159). London: E & FN Spon.
Third, there are differences in possession depending Hughes, M. D. (2003). Notational analysis. In T. Reilly &
M. Williams (Eds.), Science and soccer (pp. 245 – 264). London:
on the identities of the team and the opponent. This
Routledge.
study has not considered the reasons for this Hughes, M. D., & Bartlett, R. (2002). The use of performance
observation. It is likely that the style of play is the indicators in performance analysis. Journal of Sports Sciences, 20,
reason for teams’ differences in possession. Our 739 – 754.
results highlight a number of variables that explain Hughes, M. D., & Franks, I. (2005). Analysis of passing
possession of the ball in soccer. Combinations of sequences, shots and goals in soccer. Journal of Sports Sciences,
23, 509 – 514.
these variables could be used to develop a model that Hughes, M. D., Langridge, C., & Dawkin, N. (2001). Perturba-
predicts possession in soccer. tion leading to shooting in soccer. In M. D. Hughes &
F. Tavares (Eds.), Notational analysis of sport IV (pp. 23 – 32).
Porto: University of Porto.
Acknowledgements Hughes, M. D., Robertson, K., & Nicholson, A. (1988). An
analysis of 1986 World Cup of Association Football. In
The authors wish to express their thanks to GECA
Downloaded by [Universite De Paris 1] at 09:11 25 July 2013