International Journal of Performance Analysis in Sport

ISSN: 2474-8668 (Print) 1474-8185 (Online) Journal homepage: http://www.tandfonline.com/loi/rpan20

Application of entropy measures to analysis of

performance in team sports

Pedro Silva, Ricardo Duarte, Pedro Esteves, Bruno Travassos & Luís Vilar

To cite this article: Pedro Silva, Ricardo Duarte, Pedro Esteves, Bruno Travassos & Luís Vilar
(2016) Application of entropy measures to analysis of performance in team sports, International
Journal of Performance Analysis in Sport, 16:2, 753-768, DOI: 10.1080/24748668.2016.11868921

To link to this article: https://doi.org/10.1080/24748668.2016.11868921

Published online: 03 Apr 2017.

Submit your article to this journal

Article views: 26

View related articles

View Crossmark data

Citing articles: 2 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=rpan20
 International Journal of Performance Analysis in Sport
2016, 16, 753-768.

45-344.

Application of entropy measures to analysis of

performance in team sports

Pedro Silva1,2 , Ricardo Duarte3, Pedro Esteves4,5, Bruno Travassos5,6 and Luís Vilar3,7

1 FC Zenit, St. Petersburg, Russia

2
Universidade do Porto, Faculdade de Desporto, Centre for Research, Education,
Innovation and Intervention in Sport, Portugal
3
CIPER, Faculdade de Motricidade Humana, Universidade de Lisboa, Lisboa, Portugal
4
Polytechnic Institute of Guarda, Guarda, Portugal
5
CIDESD – Sports Sciences, Health Sciences and Human Development, Portugal
6
Universidade da Beira Interior, Department of Sport Sciences, Covilhã, Portugal
7
Escola de Turismo, Desporto e Hospitalidade, Universidade Europeia, Lisboa,
Portugal

Abstract

Over the last years, several researchers have been claiming that team ball
sports may be viewed as dynamical systems and, thus, they should be
thoroughly investigated using congruent concepts and tools. The study of
variability in the sport performance domain has shown potential to
contribute with valuable information about tactical behaviours related
with space and time management within ever changing task constraints
featuring team sports contests. Here we detail how different entropy
measures have been applied to the study of performance variability to
uncover the interactions underlying players and teams’ performances.
With that purpose, urging issues related with information entropy,
approximate entropy and sample entropy applications are discussed as a
mean to enrich the state of the art in team sport performance. In sum,
measurements of entropy in team sports have shown great potential to
assess the uncertainty of players’ spatial distributions and dominant
regions areas and of several collective team behaviours (e.g., team
synchrony and team dispersion) throughout the course of a match.
Entropy can also be used as a potential tool to identify expert
performances and differentiate skilled from novice athletes. Future holds
many other applications of this statistic in the context of performance
analysis in sports, and the inclusion of new and more sophisticated
entropy algorithms.

Keywords: variability, sports teams as complex systems, Shannon entropy,

approximate entropy, sample entropy.

753
1. Variability as a measure of performance

Performance analysis as a scientific area of research has been under debate over the last
years, in an attempt to open new paths to better synthesize players and teams’
performance and consequently improve the information provided to coaches (Carling et
al., 2013; Glazier, 2010; McGarry, 2009; Travassos et al., 2013; Wright et al., 2014).
Traditionally, performance analysis methods drive the analysis of unidimensional and
discrete performance indicators towards probabilistic and correlational approaches to
predict future performance (Wright et al., 2014). However, this discrete dimension
associated with the lack of understanding about the interactions that sustain players’
actions and team behaviours have generated somewhat limited results from which is
difficult to obtain functional information (McGarry, 2009; Wright et al., 2014).
According to that, it has been advocated that performance analysis in team sports must
also focus on the interactions between performers that sustain team behaviours (Glazier,
2010; Travassos et al., 2013) so that performance can be explained and not just
described (Vilar et al., 2012). From a dynamical systems viewpoint, understanding how
movement coordination emerges from the interaction between the system components
(e.g., players in a team) is a key question (Davids, 2001).

In the last years, several researchers have been claiming that team sports may be
conceived as dynamical systems unfolding in time (e.g., Davids et al., 2005; McGarry,
2005). This position raises the importance of using appropriate concepts and tools,
within a performance analysis approach, to adequately uncover the mechanisms
underlying teams’ performances. In dynamical systems theory (DST), movement
coordination emerges through processes of self-organization tendencies between body
components that typically evolve for functionally preferred coordination modes (or
attractors), supporting goal-directed behaviours. This theoretical approach encompasses
many implications for performance-oriented sports biomechanics and performance
analysis research in team sports (Glazier et al., 2003). The inherent complexity of the
movement system implies that variability and nonlinear changes of skills and movement
characteristics over time (Harbourne and Stergiou, 2009; Stergiou et al., 2006) are
deemed as essential elements to understand its dynamics (Hamill et al., 2000; Oullier et
al., 2006).

Variability results from the mastering of a large number of degrees of freedom in the
human body and it is conceived as a paramount feature for movement adaptation across
all biological systems (Newell & Corcos, 1993). The redundancy of the degrees of
freedom within the system allows for the use of multiple strategies to accomplish any
given task that naturally affords variability (Harbourne & Stergiou, 2009). In clinical
research, an optimal amount of variability has been associated with healthy system
functioning (Stergiou et al., 2006). Heart rate variability, for instance, has been
extensively studied in the past two decades as a measure of cardiovascular health. Heart
rate depends of various interactive regulatory systems inputs. In this case, the more
regulatory inputs a system has, the higher its irregularity. Hence, a more regular
heartbeat has been interpreted to reflect a dysfunction in one of the hearts’ regulatory
systems (Mayer et al., 2014). Periodic (high regular) or random (high variable) heart
rate has been associated with ventricular fibrillation (Goldberg et al.), cardiac

754
arrhythmia (Goldberger and West) and low survival rate in elderly individuals (Tsuji et
al., 1994).

Less than optimal variability characterizes rigid biological systems, whereas greater
than optimal variability characterizes typically noisy and unstable systems (Stergiou et
al., 2006). Other organismic mechanisms have also been associated to unhealthy
behaviours when displaying lower or higher levels of optimal system variability. For
instance, individuals with chronic ankle instability demonstrate less stride-to-stride
variability of the frontal plane ankle kinematics compared to healthy controls (Terada et
al., 2015). This is characteristic of a less adaptable sensorimotor system to
environmental changes (see Glass, 2001, for more examples in human physiology).

The same principles apply to phenomena that occur at the social level in humans (Riley
et al., 2011) and previous research has already reported these tendencies in different
contexts such as traffic jam (Helbing & Molnar, 1998), audience applause (Néda et al.,
2000) or walking in a busy street (Helbing & Molnar, 1995). Collective behaviour has
also been well documented across a variety of animal species, has an expression of the
emergent properties of the group that go beyond individual agency. Observations of
such phenomenon has revealed some advantages in achieving shared objectives, such as
feeding and member security. For instance, the labour of thousands of bees in a colony
depends on extensive interactions between members of the group so that surrounding
areas are surveyed most efficiently for food sources of nectar and pollen (Kesebir,
2012). At such a social level of interaction, in which individual agents spontaneously
forge different types of coordination patterns to produce group level patterns, variability
is also of central importance to ensure adaptive and innovative solutions to achieve
certain goal within environmental circumstances (Sumpter, 2006).

In sports, many practitioners and performers have been considering stability and
consistency in movement to be essential characteristics of skilled performance.
However, some behaviours, which appear to be stable, may occur in quite variable
ways, particularly in expert movement (Handford, 2006). This explains the skilful
flexibility of expert players to adapt to the ever-changing task constraints of dynamic
environments like in sports contests (Davids et al., 2003; Hamill et al., 1999; Stergiou
et al., 2006). Contrariwise, initial coordination patterns of learners for new tasks begin
as fixed and rigid linkages between body parts (Davids et al., 2008).

The study of players’ movement variability can also provide important information
about their tactical behaviours in team ball sports. It may reveal how they manage space
and time as a function of specific performance constraints. Conventional time-discrete
statistical parameters such as standard deviation or coefficient of variation quantify the
magnitude of variability but are ineffective measures in analysing its structure when
data comes from a both deterministic and stochastic origin (Preatoni et al., 2010; Slifkin
and Newell, 1999). In order to reveal the time-evolving dynamics of these space-time
interactions the use of specific non-linear statistical tools is mandatory (Harbourne and
Stergiou, 2009). The use of non-linear methods for estimating variability is paramount
to access the underlying nature of the process and unmask the hidden structure of
variability (Harbourne and Stergiou, 2009; Pukėnas et al., 2012).

755
Over the last two decades, various measures of entropy have been used to examine the
variability features of human movement and other biological signals like its periodicity
or regularity (Rhea et al., 2011). Entropy is a measure of the uncertainty of a random
variable (Cover and Thomas, 2006). A decrease in entropy reflects a decrease on
unpredictability because the minimum quantity of information needed to describe the
system was also reduced. Thus, the system behaviour becomes more predictable,
presenting less variability. Inversely, high entropy means that the minimum information
necessary to describe the system increased with system variability and its behaviour is
more unpredictable and, thus, more variable. In this paper, we outline how different
entropy measures can be useful as a measure of variability governing the emergence of
key aspects of sportive performance that can be extended from the individual to
interpersonal coordination underlying collective tactical behaviours in team sports.

2. Information entropy as a measure of spatial uncertainty in team ball sports

When Claude Shannon developed information theory in 1948, he was far from
imagining the potential use of his information entropy (a fundamental cornerstone of
this theory) for human movement and sports performance analysis purposes, as well as
to other scientific domains (e.g., cryptography, neurobiology, linguistics, plagiarism
detection, et.). Shannon originally developed this concept to quantify the statistical
nature of lost information in phone-line signals during the II World War (Gleick, 2011).
Information entropy (or Shannon entropy) is a measure of unpredictability of
information content that estimates the average amount of information contained in each
message received. The Shannon entropy (ShannEn) of H is given by equation 1
(Shannon, 1948):

𝐻 = − ∑𝑁
𝑖=1 𝑝𝑖 𝑙𝑜𝑔𝑏 𝑝𝑖 (1)

where pi is the probability of character number i showing up in a stream of characters of

a given script. This means that the amount of information gained depends on some
function of probability.

The ShannEn has been recently applied in team sports by Silva et al. (2014a) to
individual spatial distribution maps (usually known as heat maps or action zones) to
infer about the underlying variability of the players’ spatial distributions during small-
sided and conditioned games. The authors aimed to scrutinize for differences in players’
movement trajectories between games played in pitches of different dimensions and
players of different expertise levels. They have analysed the players’ movements
uncertainty (by estimating the entropy of individual spatial distribution maps – the field
was divided in 1 m2 bins and the number of time points that each player spent inside
each bin was accounted) to provide important information about the probability of
players showing up in broader or stricter areas of the field. Higher entropy means that
there is more uncertainty in predicting the location of players in the field because they
are constrained to move through more varied zones of the pitch and assume broader
tactical roles (e.g., moving both backwards and forward or playing both on left and right
side positions).

756
On the other hand, if players restrict their action zones to specific on-field locations
(lower entropy), the probability of finding a player in one specific zone increases, and
thus, players’ tactical roles could be interpreted as being more structured and played
according to specific positions and roles (e.g., left defender, striker, etc.). Silva et al.
(2014a) showed that both pitch dimension and players’ skill had an impact on the
uncertainty related with their positioning during play. Their action zones became more
restricted (and thus, more predictable) as the available playing area increased (depicted
by lower ShannEn values), suggesting a more structured style of play in larger pitch
areas, according to specific positioning and playing roles (e.g., attacker, defender, right
wing, left wing – see Figure 1). Skilled players were also more sensitive to pitch size
manipulations than the less skilled players as they presented more demarcated
differences across pitches, with significantly higher variability in the small and
intermediate pitches than in the larger pitch. These data revealed that the same task
constraints could yield different effects according to the players’ skill level suggesting a
potential application of entropy to talent identification in young football players.

Figure 1. Spatial distribution maps and Shannon entropy (ShannEn) values for one
single player across distinct field dimensions during small-sided football games
(arrows represent attacking direction). As field dimension increases, the uncertainty
associated to finding that player in a specific field location (measured by ShannEn)
decreases, suggesting a more positional style of play in larger fields – in the example
above the central zone of the field was considered. The authors have used normalized
ShannEn (obtained by dividing equation 1 by log N) to allow comparisons between
distinct field dimensions. Adapted from Silva et al. (2014a).

A recent publication by Vilar et al. (2013) applied ShannEn to the study of local
numerical relations across an official football match. Authors divided the effective area
of play into seven functional sub-areas of play: right-back (RB), center-back (CB), and
left-back (LB); center- middle (CM); right-front (RF), center-front (CF), and left-front

757
(LF) and calculated the difference between the number of players that each team
allocated in each sub-areas of play, throughout the course of the match. The ShannEn
measure was applied to assess the uncertainty of team numerical relations occurring in
each sub-area of play. Results showed that the highest uncertainty in numerical relations
was in the CM sub-area of play, depicting the dynamic shift of players into and out of
the central sub-area, from the adjacent sub-areas. Interestingly, the team that won the
game was the one that induced higher uncertainty on the opponent’s CB sub-area of
play.

Another example of the application of ShannEn to assess spatial uncertainty in sportive

performance can be found in the study of penalty kick performances in Association
Football performed by Vilar et al. (2011). Assuming that the ability to determine ball
direction in advance is a key constraint on goalkeeping performance, the authors
examined whether the positioning of the goalkeeper on the goal line could reduce the
uncertainty of the shooting direction of the penalty kicker (Figure 2). Results showed
that when goalkeepers were positioned at 0.33 m to the right side of the goal,
uncertainty seemed to be lower (1.5) than when they were exactly in the middle or
0.33m to the left of the goal (1.8). These studies might be seen as useful applications of
ShannEn to understand how different constraints promote specific patterns of
performance behaviour that can be harnessed by coaches to design more effective
practice tasks.

2.2

2
Uncertainty of shooting angle

1.8

1.6

1.4

1.2

0.8

0.6
−3.3 −2.64 −1.98 −1.32 −0.66 0 0.66 1.32 1.98 2.64 3.3
Goalkeeper position in goaline (meters)

Figure 2. Uncertainty of shooting angle in penalty kicks accordingly to the goalkeeper

position in the goal line (0 m indicates that goalkeeper is the middle, positive values
that the goalkeeper is towards the right side of the goal, and negative values that the
goalkeepers is towards the left side of the goal). Results showed that within a range of
1m from the centre of the goal, it is when the goalkeeper moves 0.66 m to the right
that the uncertainty of shooting angle is lower. Adapted from Vilar et al. (2011).

758
3. Approximate entropy and sample entropy as measures of structural regularity
of performance

Other measures of entropy have been developed to assess the variability of systems
composed of interacting components. The examples described earlier with ShannEn,
despite being very useful for analysing spatial interactions, did not capture the
uncertainty of time-evolving behaviours in which time-dependency of events is a central
issue. For some theoretical and practical purposes, it might be useful to analyse the
temporal structure of the data series to reveal the time-evolving dynamics underlying
relevant behaviours usually masked when using time-discrete measures (Harbourne &
Stergiou, 2009). For an illustrative purpose, Figure 3 presents two time-series
possessing the same mean (M=15) and standard deviation (SD=5.01) but completely
different in terms of their time-evolving structures. Time-discrete statistical parameters,
such as mean and standard deviation will not distinguish these two time-series.

The proposition of approximate entropy (ApEn) was firstly made by Pincus in 1991 to
medical purposes (Pincus et al., 1991). More precisely, Pincus quantified the amount of
regularity on time-dependent data of an electrocardiogram through heart rate analyses
and endocrine hormone release pulsatility. This non-linear statistics was used to
quantify the level of regularity within a time-series (Pincus, 1991) by calculating the
logarithmic likelihood that a series of data points a certain distance apart will exhibit
similar relative characteristics on the next incremental comparison within the state space
(Pincus, 1991; Pincus and Goldberger, 1994; Stergiou, 2004). That is, a sequence of
points is more regular if a subsequence and an expansion of that subsequence are similar
(Pincus, 1991). Time series with a greater likelihood of remaining the same distance
apart will display low ApEn values, while time series that exhibit large differences
between data points will result in higher ApEn values.

Figure 3. Illustration of two time-series with the same mean (M=15) and standard
deviation (SD=5.01), however with different temporal structures. Periodicity of both
series measured through ApEn yield values of 0.46 and 0.14, for A and B,
respectively, depicting distinct temporal structures.

759
To calculate ApEn, one must divide the time-series in vector lengths of m and m+1
points and count the number of similar vector lengths that fall within a tolerance of ±r.
Then, the conditional probabilities (𝜙) of each vector (m and m+1) are calculated
dividing the number of matches of each vector by the total number of vectors. The
natural logarithm of each conditional probability (m and m+1) are summed up and
averaged. Finally, their difference defines the ApEn value:

𝐴𝑝𝐸𝑛(𝑚, 𝑟, 𝑁) = 𝜙 𝑚(𝑟 ) − 𝜙 𝑚+1 (𝑟) (2)

Typical experiments using kinematic variables and biological signals set m input
parameter at 2 and r-value between 0.1 and 0.25 of the standard deviation of data. Time-
series length (N) may be as 1000, even as short as 75 data points (Stergiou, 2004).
Nonetheless, a consensus on parameter choice has been difficult to reach and parameter
estimation is currently an important challenge for researchers.

The ApEn values typically range from 0 to 2, where 0 represents a perfectly predictable
time series (i.e., a repeated patterns time-series where the number of similar vectors is
the same for both m length vectors and m+1 length vectors) and 2 is a totally
unpredictable time series (i.e., a time series where vectors are similar just by chance).
Duarte et al. (2013b) have measured the regularity of stretch and contraction patterns of
collective behaviours displayed by professional football teams over the natural course of
the match with ApEn. Variables included were the teams’ surface area, stretch index,
length and width. Authors found that ApEn values tended to decrease significantly
during each half of the match, while the percentage of coefficient of variation increased.
It was concluded that, despite the decrease in collective stability (depicted by large
values of coefficient of variation) teams tended to become more predictable in their
organizational shape over the natural course of the match (probably due to combined
effects of accumulated fatigue and co-adaptation processes).

Also, Fonseca et al. (2012) have measured the variability of individual players’
dominant regions and the minimum interpersonal distance between teammates of
attacking and defending teams in futsal using ApEn. The results revealed that the
variability of both variables was significantly different between opponent players and
similar between players of the same team. The players’ dominant regions revealed
lower variability (or uncertainty provided by lower values of ApEn) for the attacking
team and higher stability for the defending team (depicted by smaller values of standard
deviation). The authors justified the higher unpredictability (larger values of ApEn) of
the defending team players’ dominant regions areas as the need to constantly re-adjust
their spatial organization according to the moves of the attacking team. More examples
of the application of ApEn can be found in the work of Sampaio and Maçãs (2012) on
an intervention study addressing inter-player coordination during small-sided games and
of Gonçalves et al. (2013) that have used ApEn to measure the variability in players’
movement displacements while considering position-specific centroids in association
football.

Despite the usefulness of the ApEn algorithm in capturing variability among various
scientific domains, it has been criticized due to the inherently included bias towards
regularity as it counts self-matches of vectors. Also, the reliability of ApEn for short

760
time-series data is low (Richman and Moorman, 2000). Moreover, the algorithm fails to
provide relative consistency when the input parameters (m, r and N) are changed
(Yentes et al., 2013). In order to overcome these limitations, Richman and Moorman
(2000) developed the sample Entropy (SampEn) algorithm, especially suitable to the
study of clinical cardiovascular experiments and other biological works involving time
series analysis. In their seminal paper, SampEn revealed greater level of consistency
than ApEn, across a wide range of conditions, even for short record lengths. SampEn(m,
r, N), much like ApEn, is defined as the negative natural logarithm of the conditional
probability that two sequences similar for m points (length of the vector to be
compared) remain similar at the next point m + 1.

However, this algorithm reduces the bias towards regularity by eliminating the counting
of self-matches and by taking the logarithm of the sum of conditional probabilities,
rather than the logarithm of each individual conditional property as ApEn does (Yentes
et al., 2013). Values of SampEn range from 0 towards infinity, where 0 represents a
perfectly predictable (periodic) time series and infinity is a totally unpredictable time
series. Parameter estimation for the SampEn algorithm is exemplified in the study of
Lake et al. (2002) and Alcaraz et al. (2010) (see Appendix A for a brief explanation).

An example of the application of this quantity to performance analysis in team sports

can be found in the work of Duarte et al. (2013a). In this study, the authors investigated
the movement synchronization of players through an innovative multivariate time-series
method – the cluster phase analysis. This method provides a continuous measurement of
the whole team synchronization across the match. Based on this parameter, the authors
inspected, in a functional sense, the regularity of team synchronization across the match
using SampEn. Findings revealed team synchronization levels to be significantly more
irregular (i.e., higher values of SampEn) in the longitudinal displacements rather than in
the lateral displacements in the playing field (see Figure 4). Irregularity was also higher
in the second half of the match compared with the first half. This means that the way
team players tend to increase and decrease team synchronization is more stable and
regular in the side-to-side displacements, mainly in the first half.

Another interesting application of SampEn was provided by Silva et al. (2014b). The
authors calculated the SampEn of distances to nearest opponents for national- and
regional level youth football players during 4-a-side small-sided games played in
distinct field dimensions. They found that, regardless of mean distances being identical
for both groups of players, the unpredictability of each player’s distance to his nearest
opponent (measured through SampEn) was larger for the national-level players. The
authors related this result with the superior ability of national-level players to stay
unmarked by performing “off-the-ball” movements more often in an attempt to create
free space, while, at the same time, defenders tried to constantly restrict the space
created.

761
 Figure 4. Regularity of team synchronization captured using sample entropy
(SampEn) analysis. Upper panel illustrates the cluster phase method capturing the
movement synchronization of all team players in a single parameter (lateral
displacements); Middle panel shows an example of this parameter on a team for the
entire first half of a match; Bottom panel displays mean and standard deviation
SampEn values measuring the regularity of team synchronization of several
sequences of play (n=45) in a professional match-play. Variability of team movement
coordination was lower in the lateral direction of the field, compared with the
longitudinal direction. Adapted from Duarte et al. (2013a).

762
4. Urging issues relating to approximate entropy and sample entropy applications

After Richman and Moorman algorithm release in 2000, there was a boost in the
number of studies using SampEn to quantify the regularity of time-series (see Figure 5).
However, while in some cases it appears that SampEn is more reliable, sometimes
ApEn is suggested as a better alternative, depending on the type of data being analysed.
For instance, SampEn showed higher reliability in short gait data series, being less
sensitive to changes in data length and demonstrated fewer problems with relative
consistency (Yentes et al., 2013). Contrariwise, with data sets containing body
temperature time-series in patients with multiple organ failure, ApEn discriminated
better between two different groups when compared to SampEn (Cuesta-Frau et al.,
2009). SampEn seems also to be more sensitive to the frequency of the signal. Biased
estimations of entropy can be obtained if the experimenter fails to estimate the spectral
density of the time-series (a fast Fourier transform can be used to set an appropriate
frequency for the analysis of a data set).

The choice for which algorithm to use is not yet completely clear (Liu et al., 2013) and
the variation of several parameters in order to optimize the reliability of the results is a
very time consuming process (Mayer et al., 2014). A complementary measurement of
both ApEn and SampEn can be conceivable and, perhaps, useful. For time-series with
short data length (N), SampEn is a more reliable entropy measure (Richman &
Moorman, 2000). After comparing several input parameters of ApEn and SampEn with
short data sets, Yentes et al. (2013) recommended to use SampEn instead of ApEn if (i)
N > 200 data points, (iii) m = 2 and (iv) different r variations maintain relative
consistency.

Figure 5. Total number of publications using approximate entropy and sample entropy
listed on PubMed from 1991 to 2015.

763
5. Concluding remarks

In this paper we have highlighted the use of entropy measures to analyse performance in
team sports and how such methods have been applied in experimental studies to
measure the spatial uncertainty and regularity of relevant tactical performance variables
in team sports. The use of ShannEn, ApEn and SampEn measures allow pushing
traditional statistical methods forward by adding knowledge about the regularity,
variability and/or unpredictability of performance behaviours. Researchers and
practitioners in the sports domain are provided with new insights but mainly with
important methodological challenges for future work. Still, there are several open-
questions to debate especially in what regards to the selection of the entropy measures
in function of data properties such as time series length or signal frequency.

Future applications of entropy may include the use of fuzzy measures of entropy as
recent studies have shown as it provides a more accurate entropy definition than ApEn
and SampEn, with stronger relative consistency and less dependency on data length (Liu
et al., 2013). Despite the relative novelty of these entropy measures in the sports
performance domain, the reported studies open a new window into the exploration of
dynamical systems properties of players’ behaviours. As it was illustrated, the
possibility to quantify the uncertainty associated to some coaching concepts is a
promising and challenging approach to team performance analysis and to the
identification of skilled behaviours in sports.

6. Acknowledgements

The Authors would like to thank Dr. Jennifer Yentes and Dr. Nick Stergiou for having
provided part of the data displayed on Figure 4 (period from 1991 to 2011). The authors
would also like to thank Dr. Ana Paula Rocha for her insights in earlier drafts of this
manuscript.

7. References

Alcaraz, R., Abásolo, D., Hornero, R., & Rieta, J.J. (2010). Study of sample entropy
ideal computational parameters in the estimation of atrial fibrillation
organization from the ecg. Paper presented at the Institute of Electrical and
Electronics Engineers Conference, Belfast.
http://www.scopus.com/inward/record.url?eid=2-s2.0-
79953816271&partnerID=40&md5=3d4d84f1a35a5b8fa3e9dc3dc26db75c
Carling, C., Wright, C., Nelson, L., & Bradley, P. (2013), Comment on ‘performance
analysis in football: A critical review and implications for future research’,
Journal of Sports Sciences, 1-6.
Cover, T., & Thomas, J. (2006), Elements of information theory (second edition ed).
New Jersey: John Wiley & Sons, Inc.
Cuesta-Frau, D., Miro-Martinez, P., Oltra-Crespo, S., Varela-Entrecanales, M., Aboy,
M., Novak, D., & Austin, D. (2009), Measuring body temperature time series
regularity using approximate entropy and sample entropy, Conference

764
 proceedings : Annual International Conference of the IEEE Engineering in
Medicine and Biology Society. IEEE Engineering in Medicine and Biology
Society. Conference, 2009, 3461-3464.
Davids, K. (2001). Genes, training, and other constraints on individual performance: A
role for dynamical systems theory? Sportscience, 5(2).
sportsci.org/jour/0102/kd.htm
Davids, K., Araújo, D., & Shuttleworth, R. (2005), Applications of dynamical system
theory to football. In T. Reilly, J. Cabri & D. Araújo (Eds.), Science & football
v (pp. 556-569). Oxon: Routledge.
Davids, K., Button, C., & Bennett, S. (2008), Dynamics of skill acquisition: A
constraints-led approach. Champaign (IL): Human kinetics.
Davids, K., Glazier, P., Araújo, D., & Bartlett, R. (2003), Movement systems as
dynamical systems: The functional role of variability and its implications for
sports medicine, Sports Medicine, 33(4), 245-260.
Duarte, R., Araújo, D., Correia, V., Davids, K., Marques, P., & Richardson, M. (2013a),
Competing together: Assessing the dynamics of team-team and player-team
synchrony in professional association football, Human Movement Science, 32,
555-566.
Duarte, R., Araújo, D., Folgado, H., Esteves, P., Marques, P., & Davids, K. (2013b),
Capturing complex, non-linear team behaviours during competitive football
performance, Journal of Systems Science and Complexity, 26, 62-72.
Fonseca, S., Milho, J., Travassos, B., & Araújo, D. (2012), Spatial dynamics of team
sports exposed by voronoi diagrams, Human Movement Science, 31(6), 1652-
1659.
Glass, L. (2001), Synchronization and rhythmic processes in physiology, Nature,
410(6825), 277-284.
Glazier, P., Davids, K., & Bartlett, R.M. (2003). Dynamical systems theory: A relevant
framework for performance-oriented sports biomechanics research. Sportscience
. sportsci.org. Retrieved 11/03/ 2011, from
http://www.sportsci.org/jour/03/psg.htm
Glazier, P.S. (2010), Game, set and match? Substantive issues and future directions in
performance analysis, Sports Medicine, 40(8), 625-634.
Gleick, J. (2011), The information: A history, a theory, a flood. London: Fourth
Estate.
Goldberg, R. J., Gore, J.M., Alpert, J.S., Dalen, J.E. (1986). Recent changes in attack
and survival rates of acute myocardial infarction (1975 through 1981). The
worcester heart attack study, Journal of the American Medical Association,
255(20), 2774-2779.
Goldberger, A.L., & West, B.J. (1987). Fractals in physiology and medicine, Yale
Journal of Biology and Medicine, 60(5), 421-435.
Gonçalves, B.V., Figueira, B.E., Maçãs, V., & Sampaio, J. (2013), Effect of player
position on movement behaviour, physical and physiological performances
during an 11-a-side football game, Journal of Sports Sciences, 32(2), 191-199.
Hamill, J., Haddad, J., & McDermott, W. (2000), Issues in quantifying variability from
a dynamical systems perspective, Journal of Applied Biomechanics, 16(4),
407-418.

765
Hamill, J., van Emmerik, R.E.A., Heiderscheit, B.C., & Li, L. (1999), A dynamical
systems approach to lower extremity running injuries, Clinical Biomechanics,
14(5), 297-308.
Handford, C. (2006), Serving up variability and stability. In K. Davids, S. Bennett & K.
Newell (Eds.), Movement system variability (pp. 73-83). Champaign, IL:
Human Kinetics.
Harbourne, R.T., & Stergiou, N. (2009), Movement variability and the use of nonlinear
tools: Principles to guide physical therapist practice, Physical Therapy, 89(3),
267-282.
Helbing, D. & Huberman, B. A. 1998 Coherent moving states in highway traffic.
Nature, 396, 738–740.
Helbing, D. & Molnar, P. 1995 Social force model for pedestrian dynamics. Phys. Rev.
E, 51, 4282–4286.
Kesebir, S. (2012), The Superorganism Account of Human Sociality: How and When
Human Groups Are Like Beehives. Personality and Social Psychology Review
16, 233–261.
Lake, D., Richman, J., Griffin, P., & Moorman, R. (2002), Sample entropy analysis of
neonatal heart rate variability, American Journal of Physiology: Regulatory,
Integrative and Comparative Physiology, 283, R789-R797.
Liu, C., Li, K., Zhao, L., Liu, F., Zheng, D., Liu, C., & Liu, S. (2013), Analysis of heart
rate variability using fuzzy measure entropy, Computers in Biology and
Medicine, 43(2), 100-108.
Mayer, C., Martin, Bachler M Fau - Hortenhuber, M., Hortenhuber M Fau - Stocker, C.,
Stocker C Fau - Holzinger, A., Holzinger A Fau - Wassertheurer, S., &
Wassertheurer, S. (2014), Selection of entropy-measure parameters for
knowledge discovery in heart rate variability data, BMC Bioinformatics,
15(Suppl 6)(S2), 1471-2105.
McGarry, T. (2005), Soccer as a dynamical system: Some theoretical considerations. In
T. Reilly, J. Cabri & D. Araújo (Eds.), Science & football v (pp. 570-579).
Oxon: Routledge.
McGarry, T. (2009), Applied and theoretical perspectives of performance analysis in
sport: Scientific issues and challenges, International Journal of Performance
Analysis in Sport, 9(1), 128-140.
Néda, Z., Ravasz, E., Brechet, Y., Vicsek, T., and Barabási, A. L. (2000). Physics of
rhythmic applause. Physical Review E, 61, 6987–6992.
Newell, K., & Corcos, D.M. (1993), Variability and motor control. Champaign:
Human Kinetics.
Oullier, O., Marin, L., Stoffregen, T., Boostma, R., & Bardy, B. (2006), Variability in
postural coordination dynamics. In K. Davids, S. Bennett & K. Newell (Eds.),
Movement system variability (pp. 25-47). Champaign, IL: Human kinetics.
Pincus, S. (1991), Approximate entropy as a measure of system complexity, Proc. Natl.
Acad. Sci. USA, 88, 2297-2301.
Pincus, S., Gladstone, I.M., & Ehrenkranz, R.A. (1991), A regularity statistic for
medical data analysis, Journal of Clinical Monitoring and Computing, 7(4),
335–345.
Pincus, S., & Goldberger, A.L. (1994), Physiological time-series analysis: What does
regularity quantify?, American Journal of Physiology - Heart and
Circulatory Physiology, 266(4 35-4), H1643-H1656.

766
Preatoni, E., Ferrario, M., Dona, G., Hamill, J., & Rodano, R. (2010), Motor variability
in sports: A non-linear analysis of race walking, Journal of Sports Sciences,
28(12), 1327-1336.
Pukėnas, K., Poderys, J., & Gulbinas, R. (2012), Measuring the complexity of a
physiological time series: A review, Education. Physical Training. Sport(84),
48-54.
Rhea, C.K., Silver, T.A., Hong, S.L., Ryu, J.H., Studenka, B.E., Hughes, C.M.L., &
Haddad, J.M. (2011), Noise and complexity in human postural control:
Interpreting the different estimations of entropy, PLoS ONE, 6(3), e17696.
Riley, M., Richardson, M., Shockley, K., & Ramenzoni, V. (2011). Interpersonal
synergies. Frontiers in Psychology, 2(38), 1-7.
Richman, J., & Moorman, R. (2000), Physiological time-series analysis using
approximate entropy and sample entropy, Am J Physiol Heart Circ Physiol,
278, H2039-H2049.
Sampaio, J., & Maçãs, V. (2012), Measuring tactical behaviour in football,
International Journal of Sports Medicine, 33(5), 395-401.
Shannon, C. (1948), A mathematical theory of communication, The Bell System
Technical Journal, 27(3), 379-423.
Silva, P., Aguiar, P., Duarte, R., Davids, K., Araújo, D., & Garganta, J. (2014a), Effects
of pitch size and skill level on tactical behaviours of association football players
during small-sided and conditioned games, International Journal of Sports
Science & Coaching, 9(5), 993-1006.
Silva, P., Duarte, R., Sampaio, J., Aguiar, P., Davids, K., Araújo, D., & Garganta, J.
(2014b), Field dimension and skill level constrain team tactical behaviours in
small-sided and conditioned games in football, Journal of Sports Sciences,
32(20), 1888-1896.
Slifkin, A., & Newell, K.M. (1999), Noise, information transmission, and force
variability, Journal of Experimental Psychology-Human Perception and
Performance, 25(3), 837-851.
Stergiou, N. (2004), Innovative analyses of human movement. Champaign, IL:
Human Kinetics.
Stergiou, N., Harbourne, R.T., & Cavanaugh, J.T. (2006), Optimal movement
variability: A new theoretical perspective for neurologic physical therapy,
Journal of Neurologic Physical Therapy, 30(3), 120-129.
Sumpter, D. (2006). The principles of collective animal behaviour. Philosophical
Transactions of the Royal Society Biological Sciences, 361, 5-22.
Terada, M., Bowker, S., Thomas, A.C., Pietrosimone, B., Hiller, C.E., Rice, M.S., &
Gribble, P.A. (2015), Alterations in stride-to-stride variability during walking in
individuals with chronic ankle instability, Human Movement Science, 40(0),
154-162.
Travassos, B., Davids, K., Araujo, D., & Esteves, P.T. (2013), Performance analysis in
team sports: Advances from an ecological dynamics approach, International
Journal of Performance Analysis in Sport, 13(1), 83-95.
Tsuji, H., Venditti Fj Jr Fau - Manders, E.S., Manders Es Fau - Evans, J.C., Evans Jc
Fau - Larson, M.G., Larson Mg Fau - Feldman, C.L., Feldman Cl Fau - Levy,
D., & Levy, D. (1994), Reduced heart rate variability and mortality risk in an
elderly cohort. The framingham heart study, (0009-7322 (Print)).

767
Vilar, L., Araújo, D., Davids, K., & Bar-Yam, Y. (2013), Science of winning soccer:
Emergent pattern-forming dynamics in association football, Journal of Systems
Science and Complexity, 26(1), 73-84.
Vilar, L., Araújo, D., Davids, K., & Button, C. (2012), The role of ecological dynamics
in analysing performance in team sports, Sports Medicine, 42(1), 1-10.
Vilar, L., Santos, L., Araújo, D., & Davids, K. (2011), Goalkeepers’ positioning can
reduce the uncertain of penalty shot direction in association football, British
Journal of Sports Medicine, 45(15), A12-A13.
Wright, C., Carling, C., & Collins, D. (2014), The wider context of performance
analysis and it application in the football coaching process, International
Journal of Performance Analysis in Sport, 14(3), 709-733.
Yentes, J., Hunt, N., Schnid, K., Kaipust, J., McGrath, D., & Stergiou, N. (2013), The
appropriate use of approximate entropy and sample entropy with short data sets,
Annals of Biomedical Engineering, 41(2), 349-365.

Appendix A

Several auto-regressive (AR) models of various orders (1 to 10) have been fit to the data
with the argument that if data come from an AR(p) process then m ≥ p. The r parameter
was picked to minimize the efficiency metric, which represents the maximum relative
error of SampEn and of the conditional probability estimate, respectively. This criterion
reflects the efficiency of the entropy estimate because it favours estimates with low
variance and, simultaneously, penalizes conditional probabilities near 0 and near 1 (for
more detailed explanations of these procedures see Lake et al. (2002), page R791).

768