Complex Systems in Sport

Complex systems in nature are those with many interacting parts, all capable
of influencing global system outcomes. There is a growing body of research
that has modelled sport performance from a complexity sciences perspective,
studying the behaviour of individual athletes and sports teams as emergent
phenomena which self-organize under interacting constraints.
This book is the first to bring together experts studying complex systems in
the context of sport from across the world to collate core theoretical ideas,
current methodologies and existing data into one comprehensive resource. It
offers new methods of analysis for investigating representative complex sport
movements and actions at an individual and team level, exploring the
application of methodologies from the complexity sciences in the context of
sports performance and the organization of sport practice.
Complex Systems in Sport is important reading for any advanced student
or researcher working in sport and exercise science, sports coaching,
kinesiology or human movement.
Keith Davids is Professor of Motor Control at the Centre for Sports
Engineering Research, Sheffield Hallam University, UK.
Robert Hristovski is Professor in the Faculty of Physical Education at the
University of Ss. Cyril and Methodius, Republic of Macedonia.
Duarte Araújo is Associate Professor in the Faculty of Human Kinetics at
University of Lisbon, Portugal.
Natàlia Balagué Serre is Professor of Exercise Physiology in the INEFC at
the University of Barcelona, Spain.
Chris Button is Associate Professor of Motor Learning at the School of
Physical Education, Sport and Exercise Sciences, University of Otago,
Dunedin, New Zealand.
Pedro Passos is Assistant Professor of Motor Control in the Faculty of
Human Kinetics at the University of Lisbon, Portugal.
Routledge Research in Sport and Exercise Science
The Routledge Research in Sport and Exercise Science series is a showcase
for cutting-edge research from across sport and exercise sciences, including
physiology, psychology, biomechanics, motor control, physical activity and
health, and every core subdiscipline. Featuring the work of established and
emerging scientists and practitioners from around the world, and covering the
theoretical, investigative and applied dimensions of sport and exercise, this
series is an important channel for new and groundbreaking research in the
human movement sciences.
Also available in this series:
1. Mental Toughness in Sport
Developments in Theory and Research
Daniel Gucciardi and Sandy Gordon
2. Paediatric Biomechanics and Motor Control
Theory and Application
Mark De Ste Croix and Thomas Korff
3. Attachment in Sport, Exercise and Wellness
Sam Carr
4. Psychoneuroendocrinology of Sport and Exercise
Foundations, Markers, Trends
Felix Ehrlenspiel and Katharina Strahler
5. Mixed Methods Research in the Movement Sciences
Case Studies in Sport, Physical Education and Dance
Oleguer Camerino, Marta Castaner and Teresa M. Anguera
6. Complexity and Control in Team Sports
Dialectics in Contesting Human Systems
Felix Lebed and Michael Bar-Eli
7. Complex Systems in Sport
Edited by Keith Davids, Robert Hristovski, Duarte Araújo,
Natàlia Balagué Serre, Chris Button and Pedro Passos
Complex Systems in Sport

Edited by
Keith Davids, Robert Hristovski, Duarte Araújo,
Natàlia Balagué Serre, Chris Button and Pedro
Passos
First published 2014
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN

and by Routledge
711 Third Avenue, New York, NY 10017

Routledge is an imprint of the Taylor and Francis Group, an informa business

© 2014 Keith Davids, Robert Hristovski, Duarte Araújo, Natàlia Balagué Serre, Chris Button and
Pedro Passos

The right of the editors to be identified as the authors of the editorial material, and of the authors for
their individual chapters, has been asserted in accordance with sections 77 and 78 of the Copyright,
Designs and Patents Act 1988.

All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by
any electronic, mechanical, or other means, now known or hereafter invented, including photocopying
and recording, or in any information storage or retrieval system, without permission in writing from the
publishers.

Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

Library of Congress Cataloging-in-Publication Data

Complex systems in sport / edited by Keith Davids, Robert Hristovski, Duarte
Araújo, Natàlia Balagué Serre, Chris Button and Pedro Passos. – First edition.
pages cm. – (Routledge research in sports and exercise science)
1. Sports sciences. 2. Exercise. 3. Human mechanics. 4. Complexity
(Philosophy) I. Davids, K. (Keith), 1953-
GV558.C65 2014
796.01'5–dc23
2013014582

ISBN: 978-0-415-80970-2 (hbk)

ISBN: 978-0-203-13461-0 (ebk)

Typeset in Times New Roman

by FiSH Books Ltd, Enfield
Contents
List of illustrations
Contributors
Acknowledgements
Preface
KEITH DAVIDS, ROBERT HRISTOVSKI, DUARTE ARAÚJO AND
NATÀLIA BALAGUÉ SERRE

PART 1
Theoretical bases for understanding complex systems in sport

1 Basic notions in the science of complex systems and nonlinear dynamics

ROBERT HRISTOVSKI, NATÀLIA BALAGUÉ SERRE AND WOLFGANG SCHÖLLHORN

2 Coordination dynamics and cognition

J. A. SCOTT KELSO

3 Why coordination dynamics is relevant for studying sport performance

CHRIS BUTTON, JON WHEAT AND PETER LAMB

4 Psychobiological integration during exercise

NATÀLIA BALAGUÉ SERRE, ROBERT HRISTOVSKI, ALFONSAS VAINORAS,
PABLO VÁZQUEZ AND DANIEL ARAGONÉS

PART 2
Methodologies and techniques for data analyses in investigating complex
systems in sport

5 Nonlinear time series methods for analyzing behavioural sequences

NIKITA KUZNETSOV, SCOTT BONNETTE AND MICHAEL A. RILEY

6 Interpersonal coordination tendencies induce functional synergies through

co-adaptation processes in team sports
PEDRO PASSOS, DUARTE ARAÚJO, BRUNO TRAVASSOS, LUIS VILAR AND RICARDO
DUARTE
7 The measurement of space and time in evolving sport phenomena
SOFIA FONSECA, ANA DINIZ AND DUARTE ARAÚJO

8 Self-organizing maps and cluster analysis in elite and sub-elite athletic

performance
WOLFGANG SCHÖLLHORN, JIA YI CHOW, PAUL GLAZIER AND CHRIS BUTTON

9 Single camera analyses in studying pattern-forming dynamics of player

interactions in team sports
RICARDO DUARTE, ORLANDO FERNANDES, HUGO FOLGADO AND DUARTE
ARAÚJO

10 Using virtual environments to study interactions in sport performance

VANDA CORREIA, DUARTE ARAÚJO, GARETH WATSON AND CATHY CRAIG

11 Methods of measurement in studying team sports as dynamical systems

DANIEL MEMMERT AND JÜRGEN PERL

12 Team sports as dynamical systems

TIM MCGARRY, JÜRGEN PERL AND MARTIN LAMES

PART 3
Complexity sciences and sport performance

13 Ecological dynamics as an alternative framework to notational

performance analysis
LUIS VILAR, CARLOTA TORRENTS, DUARTE ARAÚJO AND KEITH DAVIDS

14 Talent development and expertise in sport

ELISSA PHILLIPS, KEITH DAVIDS, DUARTE ARAÚJO AND IAN RENSHAW

15 Creativity in sport and dance: ecological dynamics on a hierarchically

soft-assembled perception–action landscape
ROBERT HRISTOVSKI, KEITH DAVIDS, DUARTE ARAÚJO, PEDRO PASSOS, CARLOTA
TORRENTS, ALEXANDAR ACESKI AND ALEXANDAR TUFEKCIEVSKI

PART 4
Complexity sciences and training for sport

16 Variability in neurobiological systems and training

 CHRIS BUTTON, LUDOVIC SEIFERT, DAVID O'DONOVAN AND KEITH DAVIDS

17 Individual pathways of change in motor learning and development

YEOU-TEH LIU AND KARL M. NEWELL

18 A constraints-based approach to the acquisition of expertise in outdoor

adventure sports
KEITH DAVIDS, ERIC BRYMER, LUDOVIC SEIFERT AND DOMINIC ORTH

19 Skill acquisition and representative task design

ROSS A. PINDER, IAN RENSHAW, JONATHON HEADRICK AND KEITH DAVIDS

Index
Illustrations
1.1 Top: a stable linear system with one attractor and one basin of
attraction (see converging arrows). Bottom: unstable linear system with
one repeller and diverging flows toward infinity. There are no
alternative stable states (attractors) for the behavioural variable Y
1.2 Nonlinear systems can possess repeller(s), but also two or more stable
states of the behavioural variable Y and associated basins of attraction
shown by converging arrows toward the attractors
1.3 Left panel: the increase of the inter-strike time interval for increasing
of D control parameter (right-to-left direction). The whole cycle of
leaning forward to strike – upright posture restoration – leaning
forward to strike, increases, showing increase in the relaxation time
toward the upright parallel-foot stance. For D0 − D < 0.05, the
relaxation time towards parallel stance tends to become infinite; that is,
no relaxation toward the parallel stance exists anymore. The parallel
stance loses its stability and transits to a more stable diagonal stance;
see right panel
3.1 The interpretation of coupling angles; arrows represent the direction of
a vector connecting data points on an angle-angle diagram
4.1 Feedback (A) and feed-forward (B) control of voluntary movement. A
sign of correction is created in both mechanisms to change the action
of the muscles according to the difference between the desired and
achieved states
4.2 Quasi-static arm-curl exercise holding an Olympic bar with an elbow
flexion of 90° at 80% of the one-repetition maximum until the
fatigueinduced spontaneous termination point
4.3 A) Time series of the elbow-angle data for participant 1; B) A typical
difference in the power spectral density values for the online
fluctuations of the elbow angle in the first (a) and the last (b) third of
the quasi-static exercise. The difference of the elbow angle variability
between the first and the third phase spans over sub-second and
 seconds time scale. This signifies a correlated instability of the system
under fatigue
4.4 Differences in means for the spectral degrees of freedom; horizontal
axis: the first and third part of exercise; vertical axis: spectral degrees
of freedom
4.5 Volition-state dynamics in six participants during the quasi-static
elbow angle exercise
4.6 Cycle ergometer exercise
4.7 (Upper panel): standardized fluctuations from the data on revolutions
per minute; (middle panel): power spectrum for the first half of the
exercise with its slope of −1.5 showing anti-persistent fractional
Brownian motion (fBm); (lower panel): power spectrum for the second
half of the exercise with a slope of −2.2 showing persistent fBm
4.8 A: Treadmill exercise with a participant reporting through signs; B:
sample of 11 individual time series of task-unrelated–task-related
thought dynamics. Starting with the task-unrelated thoughts (TUT)
state, one can observe the switches between TUT and task-related
thoughts (TRT). Eventually, the TRT state becomes the one that
precedes the exhaustion point. Numbers on the left signify participants
5.1 Simulated time series to illustrate the distinction between measures of
central tendency and measures sensitive to the sequential properties of
the time series. Panel A depicts random uncorrelated noise, while Panel
B shows ‘pink’ or 1/f noise where each successive observation is
positively correlated with the others
5.2 The phase space plot of the Lorenz attractor; we used the parameters Ρ
= 10, σ = 28 and β = 8/3 and initial conditions of x = 1, y = 1, z = 1
5.3 Panel A shows the simulated time series used in the tutorial calculation
of SampEn; panel B shows the same time series with added patterns of
[1, 2, 3] values (shown in squares) to increase the regularity of the time
series
5.4 Illustration of the digitization (quantization) effect on the SampEn
calculation
6.1 The calculation of the coordinative variable
6.2 Angle between geometrical centre of each team and the goal position
6.3 Exemplar data of a network in water polo
7.1 Exemplar data of the distance from a sailing boat to a fixed starting
position before the start of the regatta; the length of this time series is
579 data points
7.2 Simulated time series of length n = 200 of a stationary process (top),
non-stationary process with non-constant mean (middle) and a
nonstationary process with non-constant variance (bottom)
7.3 Simulated time series of length n = 200 of a white noise (top) and
respective sample autocorrelation function at lags k = 0, …, 50
(bottom) with critical bounds (dashed lines)
7.4 Exemplar data of length n = 579 of the distance from a sailing boat to a
fixed starting position before the regatta starting (top) and respective
sample autocorrelation function at lags k = 0, …, 350 (bottom) with
critical bounds (dashed lines)
7.5 Simulated time series of length n = 200 of a white noise (top) and
respective normalized periodogram at frequencies λr = 2πr with r = 0,
…, 0.5 (bottom)
7.6 Exemplar data of length n = 579 of the distance from a sailing boat to a
fixed starting position before the regatta starting (top) and respective
normalized periodogram at frequencies λr = 2πr with r = 0, …, 0.5
(bottom)
7.7 Simulated time series of length n = 200 of a short-range correlation
process (top), its sample autocorrelation function at lags k = 0, …, 50
(middle) and normalized periodogram at frequencies λr = 2πr with r =
0, …, 0.5 (bottom)
7.8 Simulated times series of length n = 400 of a long-range correlation
process (top), its sample autocorrelation function at lags k = 0, …, 50
(middle) and normalized periodogram at frequencies λr = 2πr with r =
0, …, 0.5 (bottom)
7.9 Example of a set of points in a plane (A) and corresponding Voronoi
diagram (B)
7.10 Spatial configurations: (A) attacker (grey areas) and defender (white
areas) teams and; (B) attacker player breaking defence organization;
the arrow indicates the direction of the attack
7.11 Example (one play) of the mean Voronoi area (VA) across time for
each team; error bars represent the standard deviation
7.12 Comparison of the mean entropy of Voronoi area (VA) between teams
in the same trial; error bars represent the standard deviation
7.13 Construction of the superimposed Voronoi diagram (bottom) from
considering, separately, the Voronoi diagrams for team A (black dots)
and team B (white dots)
7.14 Construction of the superimposed Voronoi diagram for (A) exclusively
paired opponents and (B) randomly located individuals
7.15 Measures from the superimposed Voronoi diagram: (A) maximum
percentage of overlapped area for each individual of the group marked
with black dots; and (B) percentage of free area (in black)
7.16 Example (one play) of the observed percentage of free area (%FA)
across time (solid line) and the 95% confidence interval for spatial
random distribution (dashed lines)
8.1 Clustering dendrogram resulting from an average linkage algorithm;
horizontal lines indicate the level of the rescaled distance at which the
respective movements are grouped into one cluster
8.2 Linear versus nonlinear separation
8.3 Variations of data processing by means of self-organizing maps
9.1 Schematic representation of single-camera video motion capture
9.2 The TACTO 8.0 device window; manual tracking of a selected
working point with a computer mouse allows virtual coordinates of the
tracked player/object to be obtained
9.3 The direct linear transformation (2D-DLT) method for camera
calibration and bi-dimensional reconstruction
9.4 Converted pitch coordinates (metres) allow the reproduction of
movement displacement trajectories of players in the space of action
9.5 The two behavioural states of the order parameter; top panel shows the
distance of the single centroid to the defensive boundary line of all the
analyzed trials, synchronized by the assistance pass instant; bottom
panel shows the order parameter subtracted by the own mean to
highlight the qualitative changes associated with perturbations of the
initial stability of teams
9.6 Identification of the nonlinear qualitative changes between the two
behavioural states; top panel displays an exemplar play with increased
slope in the transition between the two system states (down-sampled to
1 Hz); bottom panel shows the average and standard deviation bands of
the first derivative of the order parameter in all plays, highlighting with
an ellipse the high rate of change of the order parameter; the slope or
high rate of change indicates the order−order transition
9.7 Moving average and standard deviation values of the inter-centroid
distances (top panel) and relative stretch index (bottom panel); vertical
dashed lines highlight the instant of the instabilities corresponding to
the zero crossing of the order parameter previously identified in Figure
9.5
10.1 Illustrative immersive interactive virtual reality apparatus in the
Movement Innovation Laboratory at Queen's University, Belfast. It
shows a participant wearing a pair of gloves with attached hand
trackers, a head-mounted display with attached head tracker and a
back-pack housing the control unit
10.2 A schematic representation of what the ball-carrying participant could
see in front of him/her (i.e. the defensive line) in a virtual environment
simulated three-versus-three rugby task
11.1 (left) Self-organizing map (SOM: neurons are grouped to clusters
which form the output; (right) feed-forward network (FFN):
comparison of expected output Oexp and computed output Ocomp is
feedback and so changes the neuron connections to minimize the
difference between expected and computed output
11.2 Network with a process trajectory and the corresponding type profile
11.3 (left): TeSSy input interface with video control (bottom), animation
interface (left), attribute selection and input panel (top); (right):
 examples of a stroke frequency matrix and a stroke success matrix
representing tactical concepts and technical skills, respectively
11.4 Squash court with game process BR–FR–BL (left); trained net
representing the most frequent 4-position-sequences (right); BL =
backhand left side, BR = backhand right side, FR = forehand right side
11.5 Tactical patterns of players depending on their opponents; A, B and D
are the players. The squares represent the corresponding networks,
where the circles represent the players' tactical concepts. Each circle
corresponds to a stroke sequence, the frequency of which is encoded by
the diameter of the circle
11.6 The network was trained with game processes from volleyball,
presenting the most important sequences as circles, whose diameters
represent the frequencies of the corresponding sequences, three of them
being explained in more detail
11.7 (left): Trained net with a trajectory representing a sequence of player
formations starting at the‘ O' and ending at the ‘X'; (right): scheme of a
configuration prototype of a team formation, which could, for example,
correspond to the marked cluster
11.8 Trajectories showing the preparation of the defence against the
opponent's service
11.9 A two-dimensional replication of a match situation by means of
position data
11.10 Net-based recognition of formation types and the recombination with
position and time information
11.11 Example of the user interface of a tool for the combined quantitative
and qualitative analysis of formations in football
11.12 Trained neural network with grey shaded areas that illustrate different
quality levels (top left) and a representation of the trajectories of
hockey training. The learning process begins in the dark grey square
and ends in the light grey square; the colours of the neurons correspond
to those in the large net graphic (top left)
12.1 Illustration of football as a complex dynamical system
12.2 Phase space for two players in a tennis rally; Serena Williams (left)
Justine Henin (right); ▵ , □ = strokes of Williams; ▪, ♦ = strokes of
Henin; going for the ball to strike and returning to a neutral position
results in cyclical structures in a speed/position phase space
12.3 Longitudinal team centres, differences and relative phase for Italy and
France during the first half of the 2006 World Cup final game
12.4 Separating a constellation of players on the playground into its
formation and position
12.5 A trajectory of formations on the net and its reduction to a formation
type trajectory
12.6 Frequencies of formations and their correlations
12.7 Distribution of a typical pair of formations between minutes 21 and 30
12.8 Example of a typical tactical pattern produced between the two teams
12.9 Prototype of the tactical pattern from Figure 12.5, together with
success values
13.1 The constraint of goal location on coordination processes in dyadic
systems presented in decomposed format: (left column) distances of
each player to the centre of the goal; (right column) angles of each
player to the centre of the goal; (a) and (b) exemplar data from attacker
five [A5] and nearest defender [Def]; (A) and (B) exemplar data from
attacker five [A5] and nearest defender [D]; (C) and (D) dynamics of
the relative phase of the exemplar data from A5 and nearest D; (e) and
(f) frequency histograms of the relative phases of all A–D dyadic
systems (n = 52)
13.2 Mean values and standard error of the required velocity for intercepting
the ball of (A) defender and (B) goalkeeper in shots that ended in a
defender's interception, in a goalkeeper's save and in a goal. The
represented levels of statistical significance are P < 0.05 (*), P < 0.01
(**) and P < 0.001 (***). Note that the required velocity of the
goalkeeper was not measured when the defender intercepted the ball,
since it is impossible to compute the goalkeeper's interception point
15.1 Schematic (i.e. one-dimensional) presentation of the corrugated
hierarchically soft-assembled potential landscape with two confining
 barriers on both sides
15.2 (Top) snapshots of the four metastable states of dancers for the first 35
seconds of improvisation; (middle): overlaps q of the four metastable
body configurations with principal components (axis Y) and pathway of
their dynamics. Overlap values are given in the legend on right;
(bottom): reconfigurations are given as nonzero Hamming distances of
the dancer's action system; after some reconfigurations take place, the
action system relaxes and dwells in a state where no further
reconfigurations occur for some time; i.e. zero reconfiguration; this
process represents the metastable attractor configurations
(movement/posture pattern)
15.3 The profile of the average dynamic overlap qd(t) for different time lags;
its dynamics proceed on three timescales (from seconds to several
minutes) and does not converge to zero during the observation time
scale
16.1 Continuous relative phase (CRP) between elbow and knee through a
complete cycle for 24 beginners (left panel) and for 24 expert
swimmers (right panel), showing lower inter-individual variability for
experts
16.2 Angle between horizontal, left limb and right limb (left panel); modes
of limb coordination as regards the angle value between horizontal, left
limb and right limb (right panel). The angle between the horizontal line
and the left and right limbs was positive when the right limb was above
the left limb and negative when the right limb was below the left limb
16.3 Trajectory plots from O'Donovan et al. (2011)
17.1 Waddington's (1957) schematic of the epigenetic landscape
17.2 The landscape dynamics of the basic equation of the HKB (Haken-
Kelso-Bunz) model, with the same parameters as in Kelso (1995,
Figure 2.7)
17.3 (a) Landscape of learning the 90-degree phase task of the HKB model;
at the beginning of practice (C = 0.4) only temporary stabilization of
the target phase x0 = 0.25 can be achieved when starting from special
initial conditions close to C; (b) right at the transition (C = 0.425) the
 target phase x0 = 0.25 shows one-sided stability: initial conditions
close to C will be attracted to the new attractor. Note that, in this
situation, the system is very sensitive to noise perturbations; (c) after
sufficient practice (C = 0.525), all initial conditions close to the target
attractor x0 = 0.25 will converge to the fixed point
17.4 A two-timescale landscape model associated with Snoddy's (1926)
score data (black dots) as elevation levels. The four clusters correspond
to the four practice sessions. The x behavioural variable corresponds to
the slow timescale (shallow dimension), whereas the y variable
corresponds to the fast timescale (steep dimension)
17.5 Initial and final performance of the mirror tracing task; (A) breakdown
of one-dimensional performance score measure into movement time
and spatial error (movement time) components. Filled and open
symbols indicate outcome score on trial 1 and trial 50, respectively.
Triangle, circle and square symbols reflect movement time (MT),
mixed (MIXED) and spatial error (SE) group assignment (see text for
detailed explanation); (B) performance score on trial 1 and trial 50 as a
function of group; error bars indicate standard error
18.1 (left) Angles identified for the horizontal planes of the left and right
limbs in the upper and lower body of ice climbers; (right) modes of
limb coordination as regards the angle value between horizontal, left
limb and right limb
19.1 The role of Brunswik's lens model in understanding informational
variables for complex systems in sport – analysis of a tennis serve
19.2 Principles for the assessment of representative learning design
Tables
3.1 Categories of coordination and their associated coupling angle (ã)
ranges
4.1 Summary of the four experiments performed until the fatigue-induced
spontaneous termination point
11.1 Summary of the results of all trajectories of all three groups (hockey,
soccer, control)
Contributors
Alexandar Aceski is currently Assistant Professor at the Faculty of Physical Education, University of
Ss. Cyril and Methodius, Skopje, Republic of Macedonia. He has research interests in methods of
qualitative analysis and modelling used in biomechanics, particularly as applied to transfer in motor
learning.
Daniel Aragonés is a PhD student at the National Institute of Physical Education of Catalonia,
University of Barcelona, Spain. His research is currently focused on the psychobiological integration
of exercise-induced fatigue.
Duarte Araújo is Associate Professor at the Faculty of Human Kinetics of the University of Lisbon,
Portugal. He is the Director of the Laboratory of Expertise in Sport. He is the President of the
Portuguese Society of Sport Psychology and a member of the National Council of Sports. His
research on ecological dynamical approaches to expertise and decision making in sports has been
funded by the Fundação para a Ciência e Tecnologia.
Natàlia Balagué Serre is Professor of Exercise Physiology at INEFC University of Barcelona, Spain.
Her field of research is complex systems in sport, with special focus on dynamic integrative
approaches to exercise-induced fatigue and the nonlinear psychobiological integration during
exercise. In 2003, she organized the first Complex Systems in Sport Congress, which was held in
Barcelona. Co-author of Complejidad y Deporte and papers relating to the effects of exercise-
induced fatigue on attention focus, perceived exertion and exercise termination.
Scott Bonnette received a BA in psychology from Wheeling Jesuit University, USA, in 2009 and a
MA in experimental psychology from the University of Cincinnati, USA, in 2013. He is currently a
graduate student at the University of Cincinnati. He has been involved with research and publications
concerning the nonlinear analysis of postural control and with research on how exploratory
movements facilitate perception.
Eric Brymer is a psychologist and Senior Lecturer in the Faculty of Health at Queensland University
of Technology, Australia. His research focuses on investigating nature-based activities, adventure
and extreme sports. Eric is interested in the broad psychological understanding of the experience, the
development of skill and how such activities enhance positive health and wellbeing. He is
particularly interested in the role of the physical environment and how to design and facilitate nature-
based experiences so that positive outcomes for the environment and people are optimised.
Chris Button is Associate Professor of Motor Learning at the University of Otago, New Zealand. His
research interests include the ecological dynamics approach to motor learning and human behaviours
in relation to water safety.
Jia Yi Chow is Assistant Professor at the Physical Education and Sports Science Department, National
Institute of Education, Nanyang Technological University, Singapore. His area of specialization is in
motor control and learning. Jia Yi's key research work includes nonlinear pedagogy, investigation of
multiarticular coordination changes, analysis of team dynamics from an ecological psychology
perspective and examining visual–perceptual skills in sports expertise.
Vanda Correia is Assistant Professor at the University of Algarve in Faro, Portugal. She did her PhD
in Sport Sciences at the Faculty of Human Kinetics of the University of Lisbon, Portugal.
Specializing in decision making in team sports, Vanda is particularly concerned with understanding
how the dynamics of players' interactions express adaptive behaviours to performance constraints
 and are coupled with key information sources. She has been conducting research both in the field and
in virtual reality settings.
Cathy Craig is Head of the School of Psychology of Queens University Belfast, Northern Ireland, and
Director of the Movement Innovation Laboratory of that institution. She completed her PhD at the
University of Edinburgh under the supervision of Professor Dave Lee in the Perception in Action
Laboratories.
Keith Davids is Professor of Motor Learning at the Centre for Sports Engineering Research at
Sheffield Hallam University in the UK. He currently holds additional appointments at the University
of Jyväskylä in Finland (FiDiPro) and at the Queensland University of Technology in Australia. He
is a graduate of the University of London and gained a PhD at the University of Leeds in 1986.
Between 1993 and 2001, he led the Motor Control group at the Department of Exercise and Sport
Science at Manchester Metropolitan University, UK. In 2002 he moved to the University of Otago in
New Zealand before taking up an appointment at the Queensland University of Technology in
Australia. Currently he supervises doctoral students from Portugal, UK, Australia and New Zealand.
His major research interest involves the study of movement coordination and skill acquisition in
sport. He is particularly focused on understanding how to design representative learning and
performance evaluation environments in sport.
Ana Diniz is a teacher and researcher at the Department of Mathematics of the Faculty of Human
Kinetics, University of Lisbon, Portugal. Her investigation includes mathematical methods and
models related to motor control and interpersonal coordination processes.
Ricardo Duarte is Lecturer in Training Methods of Soccer at Faculdade de Motricidade Humana,
Portugal. His academic career involves mentoring young football coaches, developing research on
tactical behaviours in soccer with applications to training and performance analysis and presenting
his perspective in applied soccer training courses, congresses and seminars around the world.
Orlando Fernandes is Assistant Professor at the Department of Sport and Health of the University of
Évora, Portugal. He is an expert in biomechanics of human movement, with particular interest in
signal processing and nonlinear analysis of time-series data. His expertise has been also transferred
to the preparation of some elite athletes in track and field sports.
Hugo Folgado is Lecturer at the Department of Sport and Health, University of Évora, Portugal, and
collaborator at the Research Center for Sport Sciences, Health and Human Development, Portugal.
He is currently working on his PhD studies about football players' movement synchronization. His
research interests are performance analysis and expertise in team sports.
Sofia Fonseca has a degree in statistics from the Faculty of Sciences of Lisbon (1999), a PhD in
statistics from the University of Aberdeen (2004) and a PhD in sports science from the University of
Lisbon (2012). Sofia has been Assistant Professor at the Faculty of Physical Education and Sports,
Lusofona University, Lisbon, Portugal, since 2006. Her research interests are team sports behavior,
particularly modelling players' and teams' spatial organization.
Paul Glazier is a Research Fellow at the Institute of Sport, Exercise and Active Living, Victoria
University, Melbourne, Australia. He has expertise in sports biomechanics, motor control, skill
acquisition and performance analysis of sport. He has authored or co-authored over 40 peer-reviewed
journal articles, invited book chapters and published conference papers in these areas. Paul also has a
wealth of practical experience, having provided sports biomechanics and performance analysis
services to a wide range of athletes and teams, from regional juniors to Olympic and World
Champions, in a variety of sports.
Jonathon Headrick is a PhD scholar at the School of Exercise and Nutrition Sciences, Queensland
University of Technology, Australia. His research interests include the application of an ecological
 dynamics approach for studying the role of emotion in learning and skill acquisition in sport.
Robert Hristovski is currently Professor at the Faculty of Physical Education, University of Ss. Cyril
and Methodius, Skopje, Republic of Macedonia. He obtained his MSc degree in 1994 and a PhD
degree in 1997 at the Faculty of Physical Education. In 2001 he had a five-month research visit at the
Institute of Nonlinear Science at the University of California, San Diego. He has research interests in
methods of analysis and modelling in nonlinear dynamics, particularly as applied to human action
selection and adaptation during training. He is also an invited lecturer on masters and soctoral
courses at several European universities.
J. A. Scott Kelso is a neuroscientist and Glenwood and Martha Creech Chair in Science, Professor of
Complex Systems and Brain Sciences, Florida Atlantic University, Boca Raton, Florida, USA, and
Professor of Psychology, Biological Sciences and Biomedical Science at the University of Ulster
(Magee Campus), Derry, Northern Ireland. He has worked on coordination dynamics, the science of
coordination and on fundamental mechanisms underlying voluntary movements and their relation to
the large-scale coordination dynamics of the human brain. His experimental research in the late
1970s and early 1980s led to the HKB model (Haken–Kelso–Bunz), a mathematical formulation that
quantitatively describes and predicts how elementary forms of coordinated behaviour arise and
change adaptively as a result of nonlinear interactions among components.
Nikita Kuznetsov received a BA in psychology from California State University, Northridge, in 2008,
and a PhD in experimental psychology from the University of Cincinnati in 2013. He is currently a
post-doctoral associate at Northeastern University, USA. His research focuses on perception–action
from the complex/dynamical systems perspective and the application of nonlinear methods in motor
control.
Peter Lamb was born in Canada and obtained a PhD in biomechanics from the University of Otago,
New Zealand, in 2010. Currently, Peter is Associate Researcher at the Technische Universität
München (TUM), Germany. As head of research and diagnostics at the TUM Golf Laboratory, Peter
works with national-level golf teams and their coaches, as well as private golfers. Part of his work
includes applying the constraints-led perspective to coordination of the golf swing in search of
critical boundaries of stability. The implications are both for theoretical aspects of human movement,
as well as golf-specific applications to training and on-course strategies. Peter is an avid golfer, skier,
cyclist and ice-hockey player.
Martin Lames is full Professor at the Faculty of Sports and Health Science at the Technical University
of Munich, Germany. His main research interests are modelling of sports performances, talent
research and top-level sports, with special focus on information technology support. He has served as
President of the International Association of Computer Science in Sport (IACSS) since 2013.
Yeou-Teh Liu was born and raised in Taipei. She received her PhD in kinesiology from the University
of Illinois at Urbana-Champaign, USA, and is currently Professor in the Department of Athletic
Performance, National Taiwan Normal University. Her research focuses on the dynamics of motor
skill acquisition, movement adaptation and motor control. Her other research interest is in the
performance analyses in competitive sports including both team and individual sport events.
Tim McGarry is Associate Professor in the Faculty of Kinesiology, University of New Brunswick,
Canada. He has published many journal articles and book chapters on various aspects of movement
control and sports performance. He serves as an advisory editorial board member on the Journal of
Sports Sciences and the International Journal of Performance Analysis in Sport and is coeditor of the
2013 Routledge Handbook of Sports Performance Analysis.
Daniel Memmert is Professor and Head of the Institute of Cognitive and Team/Racket Sport Research,
German Sport University of Cologne. His research interests are cognitive science, human movement
science, computer science and sport psychology. He has 15 years of teaching and coaching
 experience, has published more than 100 publications, 20 books or book chapters and is a recognised
figure through his multi-keynotes in football. He is a reviewer for several international journals and
has transferred his expertise to business and to several professional soccer clubs within the
Bundesliga.
Karl M. Newell PhD is the Marie Underhill Noll Chair of Human Performance and Professor,
Department of Kinesiology, Pennsylvania State University, State College, USA. His research focuses
on the coordination, control and skill of normal and abnormal human movement across the life span;
developmental disabilities and motor skills; and the influence of drug and exercise on movement
control. One of the specific themes of his research is the study of variability in human movement and
posture, with specific reference to the onset of aging and Parkinson's disease. His other major
research theme is processes of change in motor learning and development that is the focus of his
chapter contribution to this book.
David O'Donovan is a postgraduate student who currently works for High Performance New Zealand
as a knowledge editor. He completed his Masters in 2011 in which he examined the throwing
kinematics of elite athletes with cerebral palsy. He has worked as a coach and sport science provider
for Boccia New Zealand and he has supported athletes participating in World Championships,
Commonwealth and Paralympic Games.
Dominic Orth is a PhD student at the University of Rouen, France, and Queensland University of
Technology (QUT), Australia. He completed his undergraduate and Masters degree by research at
the School of Exercise and Nutrition Science at QUT. His research programme examines the role of
adaptive movement variability in skilled climbers.
Pedro Passos is Assistant Professor at the Faculty of Human Kinetics, University of Lisbon, Portugal.
He gained his PhD in sport sciences in 2008. His research involves the study of the dynamics of
interpersonal coordination in team sports, which led him to produce several papers accepted for
publication in scientific journals, book chapters, as well as communications in scientific meetings.
He currently maintains his research work regarding interpersonal coordination in social systems as
team sports and extending the paradigm of analysis to video games cooperative tasks, searching for
new methods of analysis and extending his collaboration with researchers in Portugal, across Europe,
Singapore, Australia and New Zealand. He supervises masters' and doctoral students from Portugal.
Parallel with his research activity, he is also technical coordinator of a rugby union club. In his
leisure time, he practices mountain biking, surfing and alpine skiing.
Jürgen Perl is Professor Emeritus of Computer Science, University of Mainz, Germany. His main
research interests using modelling and simulation methods include pattern recognition of game
behaviours and player movements, as well as physiological load performance dynamics. He is a
founding member of the International Association of Computer Science in Sport, serving as
President from 2003–2007 and as Honorary President thereafter.
Elissa Phillips is Senior Biomechanist at the Australian Institute of Sport. Elissa's responsibility
includes implementing a programme of biomechanical services and research for athletes and coaches
to enhance performance. Recent research has focussed on feedback technology and coordination
profiling in expert performance.
Ross Pinder is Lecturer in Sport and Exercise Sciences at the University of the Sunshine Coast,
Australia. He is primarily interested in maximising skill learning in sport through the design of
representative experimental and practice environments. He currently works as a skill acquisition and
high-performance consultant for the Australian Paralympic Committee.
Ian Renshaw is Senior Lecturer, Queensland University of Technology, Brisbane, Australia. Ian's
teaching and research interests are centred on applications of ecological dynamics to sport settings.
Given Ian's background in physical education and coaching, he is particularly interested in enhancing
 pedagogical practice. Recent research has focussed on developing the links between sport
psychology and skill acquisition; implementing constraints-led approaches in physical education;
emotions and learning in sport; talent development; developing expertise in cricket and visual
regulation of run-ups. Ian has worked with numerous sports providing coach education and skill
acquisition advice.
Michael A. Riley received a BA in psychology from the University of Louisiana- Monroe, USA, in
1994 and a PhD in experimental psychology from the University of Connecticut, USA, in 1999. He
is currently Professor of Psychology and Director of the Center for Cognition, Action and
Perception, University of Cincinnati, USA, where he has been on the faculty since 2000. His
research on ecological and complex dynamical systems approaches to perception–action has been
funded by the National Science Foundation and the US Army Medical Research and Materiel
Command.
Wolfgang I Schöllhorn is Professor for Movement and Training Science and Director of the Institute
of Sport Science at the University of Mainz, Germany. With a background in physics, sports,
pedagogy and neurophysiology his research areas include dynamic systems, adaptive behaviour,
learning and brain states, biomechanics, signal analysis and pattern recognition.
Ludovic Seifert is Associate Professor at the Faculty of Sport Sciences, University of Rouen, France.
He conducts his research in the field of motor learning and motor control regarding expertise in
sport, movement variability and temporal dynamics of learning. He gained a PhD in expertise and
coordination dynamics in swimming at the University of Rouen in 2003, then a certification to direct
research in 2010 entitled ‘Motor coordination and expertise: A complex and dynamical system
approach of sport and physical education', for which he exhibited numerous publications in this field.
He is also a mountain guide certified by the International Federation of Mountain Guides
Associations and now investigates expertise and motor learning in climbing.
Carlota Torrents is a teacher of expressive movement and dance and a researcher at the Human Motor
Behavior and Sport Laboratory, National Institute of Physical Education of Catalonia, University of
Lleida, Spain. She completed her PhD in complex systems applied to training methods and has
published international books and papers related to complex systems, dance and sport.
Bruno Travassos is Assistant Professor at the Department of Sport Sciences, University of Beira
Interior, Portugal, and member of the group of performance analysis at CIDESD – Research Centre
in Sports, Health Sciences and Human Development, Portugal. His research interests are in the area
of game analysis and also in the learning processes and decision-making behaviour of players in
team sports with special emphasis in futsal and soccer.
Alexandar Tufekcievski is Professor at the Faculty of Physical Education, University of Ss. Cyril and
Methodius, Skopje, Republic of Macedonia. He has research interests in methods of qualitative
analysis and modelling used in biomechanics, particularly as applied to transfer in motor learning.
Alfonsas Vainoras is Professor at the Lithuanian University of Health Sciences. He investigates novel
methods of analysis of the electrocardiograph using complex systems approach and analysis tools.
Has participated in the development of the E-Health program in Lithuania and Europe.
Pablo Vázquez is a PhD student at the National Institute of Physical Education of Catalonia,
University of Barcelona, Spain. The focus of his research is the application of complex systems
principles on processes related to motor and performance changes under fatigue.
Luis Vilar completed his PhD in sports sciences, investigating the informational constraints on attacker
and defender performance in futsal. Currently, he is Assistant Professor at the Faculty of Human
Kinetics, University of Lisbon, Portugal, and at the Faculty of Physical Education and Sports,
Lusófona University of Humanities and Technologies, Portugal. He teaches UEFA-pro courses for
 coaches. He is also head of youth football department and coach at Colégio Pedro Arrupe. He was a
football and a futsal player.
Gareth Watson gained a BSc in psychology from Queens University Belfast, Northern Ireland, and
concluded his PhD in mechanical and aeronautical engineering and psychology at the same
university.
Jon Wheat is a Principal Research Fellow in Biomechanics, Centre for Sports Engineering Research
(CSER), Sheffield Hallam University, UK. He gained his undergraduate degree in sport and exercise
science from Manchester Metropolitan University before completing his PhD at Sheffield Hallam
University. Jon works on biomechanics research and consultancy projects in CSER and teaches on
the MSc sports engineering and MSc sport and exercise science degrees. His work is influenced by
the ecological approach to motor control and dynamical systems theory and he has a keen interest in
the development and application of biomechanics measurement systems for use outside the
laboratory, in more representative settings. He leads the Biomechanics Research Group in CSER
which has several research and consultancy projects in this area.
Acknowledgements
We give our profound thanks to all the chapter authors for their willingness to
share their comprehensive knowledge of the complexity sciences. We also
thank all the individuals who helped with reviewing the text and compiling
the index, especially Dominic Orth, José Pedro Silva, Rens Meerhoff, Pablo
Vázquez Justes, Sergi Garcia Retortillo.
Keith Davids: I acknowledge the efforts of my fellow co-editors for their
wonderful professionalism in seeing this project through from
conceptualisation to completion. They showed exemplary patience in dealing
with my endless requests. Finally, as always, I dedicate this book to my
family (Anna, Mike, Jake, Charlie and India) for their love and support.
Robert Hristovski: I dedicate this book to my family who supported me
each step of the way.
Duarte Araújo: This book is dedicated to those students who taught me the
meaning of inter-independence: Bruno Travassos, Vanda Correia, Ricardo
Duarte, Luis Vilar, Pedro Esteves and João Carvalho. I also acknowledge the
students of SpertLab for stretching their autonomy to offer me more time for
the book.
Natàlia Balagué Serre: I dedicate to this book to Dani, Gerard and Pau.
Chris Button: This book has been an amazing act of teamwork. As such I
give my thanks to my offsiders, the chapter reviewers, and the authors.
You've made my editorial role a pleasure. And to the most important dyad in
my life: Ange and Melanie, thanks for your support.
Preface
Complex systems in nature are those with many interacting parts, all capable
of influencing global system outcomes. There is a growing body of research
that has modelled sport performance from a complexity sciences perspective,
studying the behaviour of individual athletes and sports teams as emergent
phenomena which self-organize under interacting constraints. This literature
has been published in journals covering physical education, sport science and
sports medicine, coaching science, psychology and human movement
science.
This book was conceived over many years of discussion and research,
when it became apparent that there was a need to bring together the
conceptual creativity and innovative ideas of many experts studying complex
systems in the context of sport performance from across the world. The
intention is to provide a coherent summary of where we currently stand with
regards to theory, current methodologies and empirical data concerning the
understanding of sport performance from a complexity sciences perspective.
Our rationale is to complete a comprehensive overview of complex systems
in sport for advanced undergraduate students, postgraduate students and
academics in a range of disciplinary areas. The authors contributing chapters
to this edited textbook have undertaken comprehensive overviews to
summarize the key ideas that have appeared in many excellent empirical
reports, reviews and theoretical position papers in the extant literature. The
aims of the various chapters in this book include the presentation of key ideas
from the complexity sciences and a summary of how these ideas might be
adopted in the organization of sport practice. In this way, this textbook builds
on existing material to provide a comprehensive foundation for students of
complex systems in sport.
This is a timely endeavour, since the study of complex systems in sport has
gained increasing prominence in recent years, leading to a deep interest from
students from a range of different disciplinary backgrounds. For example,
there is an increasing number of physicists and mathematicians involved in
this field of work and this book showcases research in sport science and
performance analysis that will enhance their understanding of the
applications of methodologies from the complexity sciences. Conversely,
within the sport sciences, there is a need to document the range of new
methods of analysis used for investigating representative complex sport
movements and actions at an individual and team level. These requirements
cannot be adequately captured by experimental designs and methods of
analysis that already exist in the basic movement sciences.
The proposed structure of the book has been carefully designed to equip
readers with the basic theoretical knowledge of complex, nonlinear
dynamical systems (Part 1: Theoretical bases for understanding complex
systems in sport) prior to delivering an understanding of current methods for
studying such systems in sport performance environments (Part 2:
Methodologies and techniques for data analyses in investigating complex
systems in sport). In the final parts of the book (Part 3: Complexity sciences
and sport performance and 4: Complexity sciences and training for sport), the
emphasis is on expanding knowledge of practical applications of ideas and
methods in the study of training and performance in individual and team
sports.
Many of the chapters deliberately explore common themes, although each
chapter attempts to provide a unique and detailed contribution to the topic of
complexity in sports. The interaction of many authors across multiple
chapters allows the overall book to develop collective themes, which emerge
throughout the whole text. This collective approach has been a deliberate
strategy of the editors, meaning that comprehension of the whole book
provides much more than the sum of each chapter in isolation. Typical
complexity science thinking!
This book supports a burgeoning area of academic interest. There is now a
biennial international scientific congress dedicated to this area of study,
which attracts around 250 delegates to each meeting. The proposed book
neatly complements the research presented at this meeting. There are also
academic journals newly emerging to support this field of work. The editors
acknowledge the powerful and influential role that every one of the chapter
authors of this book have played in bringing together theoretical material and
practical applications which will further develop our understanding of
complex systems in sport.
Keith Davids
 Robert Hristovski
Duarte Araújo
Natàlia Balagué Serre
Chris Button
Pedro Passos
Part 1
Theoretical bases for
understanding complex
systems in sport
1 Basic notions in the science of
complex systems and nonlinear
dynamics
Robert Hristovski, Natàlia Balaguél Serre
and Wolfgang Schöllhorn
The historical roots of the complexity sciences can be traced back to ancient
philosophers such as Aristotle (384–322 BC), whose famous saying, ‘The
whole is more than the sum of its parts', indicated the duality of holism versus
reductionism in science. The beginning of modern Western science is mostly
associated with the development of a mechanistic world view, originating in
contributions from Galileo, Kepler and Newton in the seventeenth century.
The mathematicoexperimental method became trend setting and in the same
period Newton created the mathematical basis of dynamical systems theory.
By showing explicitly that, celestial mechanics, Earthly tides and falling
bodies were governed by the same law of universal gravity, he actually paved
the way to what later became a foundation of general systems theory and
particularly synergetics: the search for the same principles acting at different
levels in the organization of matter. This world view may be conceived as a
special kind of holism where general principles manifest themselves through
different contexts, i.e. levels of organization. The whole manifests itself
through different partial phenomena, owing to different contexts in which
these phenomena are embedded.
These ideas have been influential in the movement sciences and, during the
1970s and 1980s, concepts of dissipative structures and self-organization
were incorporated in explanations of movement coordination (e.g. Kugler et
al. 1980). General predictions of this theoretical approach, such as non-
equilibrium phase transitions and critical fluctuations enhancement in cyclic
movements, were corroborated (Kelso 1984), and modelled (Haken et al.
1985; Schoner et al. 1986) with great success. These papers became
milestones in the search for principles of motor behaviour from a complexity
sciences and dynamical systems perspective, which made direct contact with
theory in sport science. Principles of self-organization were successfully
applied to multi-limb cyclic movements before they were experimentally
corroborated (Fuchs et al. 1992) and mathematically modelled (Jirsa et al.
1994) at the level of the central nervous system, as well as in learning
processes (Zanone and Kelso 1992). Self-organizing phenomena were also
discovered in studies of social coordination (Schmidt et al. 1990). In 2005, a
unified model of rhythmic and discrete movements was published (Jirsa and
Kelso 2005), predicting as a generic consequence the possibility of the
emergence of false starts. In the past two decades, the complex dynamic
systems paradigm became a fruitful experimental and theoretical approach in
capturing and explaining many phenomena of motor behaviour that are
closely related, although not equal, to problems in sports science. This
relatedness and prospects of the dynamical systems approach to sports
science problems were advocated in the works of Davids and colleagues (see
e.g. Davids et al. 1994, 2003).
In the following sections, we define complex systems and point to some
main differences between non-living and living systems. We then discuss in
more detail the differences between linear and nonlinear dynamical systems
and point to some necessary concepts important for understanding why
nonlinear dynamics is important in explaining sports phenomena. The
material is presented in a way suitable for unfamiliar readers to be acquainted
with basic terms and meanings from the complex dynamical systems
approach to sports.
What are complex systems?
Complex systems consist of many components which interact among
themselves and, as a whole, interact with their environments. Complex
systems may be homogenous or heterogeneous. For example, a piece of ice
contains innumerable interacting components, i.e. water molecules. These are
complex but homogenous systems. Living complex systems, besides having
many interacting elements, consist of structurally and functionally
heterogeneous (neural, muscle, tendinous, etc.) components, so they belong
to the class of heterogeneous complex systems. Biological systems also
contain parts existing in different physical phases: fluid (e.g. blood), semi-
rigid (muscles have properties of liquid crystals) and rigid (e.g. bones). Social
systems, as well, consist of interactions between heterogeneous agents. Thus,
whereas between water molecules there is one kind of interaction, i.e.
hydrogen bonds, between heterogeneous components there may be different
kinds of interactions (generally informational or/and mechanical). These may
have varying intensities, and span different spatiotemporal scales, which
immensely increases the level of complexity of description of such systems.
In such systems, each single component can ‘perceive’ a different
environment. There is another important difference between non-living and
living complex systems. In non-living systems one can isolate a large portion
of the larger system and study it because the behaviour of the system will be
the same. This is one of the main advantages that make statistical physics
feasible, and in living systems this is not possible. One cannot isolate an
organ that will function independently of the organism. Living complex
systems are also adaptive and goal directed, while one cannot find an
argument to claim the same for the non-living systems. Adaptive systems are
those which evolve, develop and learn to negotiate with their environments
by changing and fitting their behaviour to emerging constraints.
Besides these qualitative differences, there are universal features that are
valid for either living or non-living complex systems. Both kinds of systems
possess mutual interactions and interdependence between constituent
components. It seems that interactions are largely responsible for the
possibility of capturing both kinds of systems within similar formal
frameworks because mutual interactions, and recursive self-interactions that
result from these, form the nonlinear character of such systems. As a
consequence, complex systems, living or non-living, exhibit nonlinear
dynamical properties and form the class of complex nonlinear dynamical
systems. How these interactions change depends largely on the constraints
embedded within the complex systems. Under some constraints, new forms
of behaviour emerge spontaneously, without being previously designed and
imposed on the system's behaviour, and this is a property of all complex
systems, regardless whether they are living or non-living. Complex systems
may exhibit complex or simple behaviour. An athlete may perform simple
arm-curl rhythmic movements but also may be able to perform complex
sequences of dribbling actions. On the other hand, simple systems like a
single-component nonlinear pendulum may produce simple oscillatory
behaviours but also a very complex pattern of chaotic behaviour. Hence, the
complexity of behaviour should not be confused with the complexity of the
system. Complex systems may behave in a simple fashion because their
interacting components, under certain constraints, may form large coalitions
of cooperative elements, which reduces the dimensionality of the behaviour.
In this way, a complex system attains simple behaviour and may be treated as
a simple system on a macroscopic level. We get simplicity from complexity.
There are unifying principles that make possible to treat complex systems in
a relatively simple fashion.
Linear and nonlinear complex dynamical systems
Dynamical systems are systems that change over time. Because all systems
change over time, although on different timescales, it follows that all systems
are dynamical. They are usually represented by differential or difference
equations but also by cellular automata and networks, or even by a mixture of
some of these. Dynamical or behavioural variables converge in their
evolution to a stable state in which they can dwell infinitely under given sets
of constraints. This stable state is called an attractor, because it attracts all
nearby initial states of the system (Figure 1.1).
If the system is placed into different initial positions, it will converge to
one state that is stable, i.e. the attractor. The set of initial states that converge
toward the attractor form its basin of attraction. The attractor can be
conceived as a source of forces that pull all initial states toward it. Its
antipode is the unstable state called a repeller. The repeller repels all the
nearby initial states further away from it. If the system is placed into different
initial positions close to the repeller, they will diverge far from it (see Figure
1.1).
Dynamical systems consist of two broad classes: linear and nonlinear.
Linear dynamical systems are those whose rate of change of the relevant
behavioural variable is a linear function of that same variable. These systems
are proportional, in the sense that a small change of the constraints
influencing them brings about a small change in the behavioural variable. A
large change in constraints is needed to produce a large change in the
behavioural variable. In a sense, linear systems are overly flexible because of
their proportional response to changing constraints. However, they are
monostable (see Figure 1.1); i.e. under any set of constraints they either
converge to a well-defined attractor or diverge to infinity if they become
unstable. In this sense, linear systems are too rigid.
Figure 1.1 Top: a stable linear system with one attractor and one basin of attraction (see converging
arrows). Bottom: unstable linear system with one repeller and diverging flows toward infinity. There
are no alternative stable states (attractors) for the behavioural variable Y

Nonlinear systems, although containing unstable states, i.e. repellers, do

not send nearby states to infinity but enable other alternative, behavioural
stable finite states (Figure 1.2). Moreover, nonlinear dynamical systems
enable more than one stationary behavioural solution for one and the same set
of constraints supporting multi-stability of behaviour. Nonlinear systems are
safe from exploding to infinity and their behaviour is confined to a set of
finite behavioural modes.
Because of this stabilizing property, nonlinear systems enable multiple
stable behavioural solutions. In such a way, they are multifunctional and such
multifunctionality may be reached and changed by changing system's
influential parameters. Hence, multi-functionality of nonlinear system may be
harnessed by parametric control.

Figure 1.2 Nonlinear systems can possess repeller(s), but also two or more stable states of the
behavioural variable Y and associated basins of attraction shown by converging arrows toward the
attractors
Collective variables, instability and bifurcations (phase
transitions)
Consider a complex system comprising many, say thousands, of components
and their connections, enabling a vast set of interactions among them. If we
seek to capture the dynamics of that system we have to formulate thousands
of equations describing the dynamical laws governing their behaviour and
then solve them to deduce the behaviour of each of those components. This
kind of microscopic approach to capturing complex systems behaviour seems
quite unreasonable. Think of the complexity of description if we are to
deduce the macroscopic behaviour of a biological system, say running,
starting from microscopic biochemical processes in each cell of the organism.
Fortunately, large masses of cells in living systems perform in coherent and
cooperative ways so that they create much smaller numbers of mesoscopic
and macroscopic behavioural variables, which render their comprehension
easier. These macroscopic variables are those which are essential for
describing the coordinated behaviour of the system as a whole and, because
they emerge from the cooperative behaviour of collectives of components,
they are called collective variables. Since they arise from the collective task
dependent cooperation among components, they capture the order, i.e.
coordination, present within the system and hence they are called order
parameters.
Now, how are these collective variables or order parameters connected to
stability and instability properties of the system? It happens that these
variables are best detected in the vicinity of the instability points of the
system. In this region, the system, after a perturbation, incrementally returns
(relaxes) back to the attractor, as some parameter is varied, a property known
as a critical slowing down. The increase of the local relaxation time shows
that components of the system behave less cooperatively, i.e. they are losing
their coherent synergic action and attain a larger degree of independence. As
the control parameter nears a critical value, any initial perturbation grows and
leaves the previous stable state. This point is called a critical point. At this
point, the system suffers a loss of stability and the local relaxation time
becomes infinite, since the system never relaxes back to the previous
attractor. This growth is due to the self-enhancing, positive feedback process.
Positive feedback exists when the subsequent influences enhance the initial
change. At critical points, a qualitative discontinuous change in a system's
behaviour occurs – a bifurcation or a phase transition – and the values of
influential parameters at those points are called bifurcation or critical values.
An example of this critical phenomenon in the sports domain concerns the
interstrike time intervals of phase-free boxing actions used to strike a target
(Chow et al. 2009). Consider when performers initiate strikes in a parallel
stance from different scaled distances to the target. Scaled distance D may be
measured as a ratio between the physical distance of a performer's tip of the
toes from the target and their arm's length. The forward performer's strikes
toward the target act as unidirectional perturbations on their centre of mass,
tentatively considered as a collective variable. For increasing scaled distances
D from the target, the forward lean is increasingly less quickly restored, so
the inter-strike time intervals increase too. The restoration time of the
forward-leaned trunk position back to an upright two-foot parallel stance
slows down and tends to infinity for critical scaled performer–target distances
region D > 1.35, i.e. D0– D < 0.05, with D0 = 1.4, i.e. exhibits a critical
slowing down effect (Figure 1.3, left panel). In fact, the curve has a typical
critical behaviour form: <###> = A(D0 – D)–α + B, where <###> is the
average inter-strike time interval; A = 0.63; B = 0.19 and the critical
exponent α = 0.456.
In other words, below D0 = 1.35, a qualitative coordination change from
the parallel to the more stable diagonal stance, by stepping forward, takes
place (Figure 1.3, right panel), settling the centre of mass to a more stable
state. This posture-to-posture transition, which places the centre of mass in a
more stable state, is obviously preceded by increasing of the relaxation, (i.e.
restoration) time of the forward lean toward backwards as the scaled distance
D was increased. In the newly formed coordination, not only does the centre
of mass become more stable but the new stance is also more functional and
affords a more stable position adjacent to the target-striking position.
In general, it has been shown (Haken 1987) that, at a transition point, only
one or few collective modes of the system become unstable and grow (like
the centre of mass). The other system degrees of freedom, such as the leg
components, become dependent on these collective modes and start to be
governed by them. In other words, collective modes enslave the rest of the
components and force them to organize in a certain way (e.g. a step forward).
This is the well-known slaving principle introduced by Herman Haken (e.g.
Haken 1987). These enslaved components stabilize the value of the collective
variables by nonlinear interactions to a finite stable value (the newly formed
body position). That is why, instead of growing infinitely, the behavioural
variable converges to a finite value, which is the new stable state of
organization, i.e. the attractor, of the system. A temporal hierarchy is settled
in a spontaneous way. The collective variable (e.g. the centre of mass) takes
the role of a slowly varying top-down influence and the enslaved elements,
i.e. the leg components, adopt the role of fast variables that follow the
behaviour of the collective variable. Enslaved components, by their
cooperative behaviour, maintain the collective variable and the collective
variable governs the components. The system spontaneously splits into a two-
level hierarchy; i.e. on variable(s) that govern and those that are being
governed. This is the meaning of the circular causality present in complex
systems. In this way, a pattern emerges from the interaction of components
that is greater than and different from the individual components themselves.
Emergence means that the macroscopic pattern has properties that cannot be
found in the components that form it. A parallel or diagonal stance cannot be
reduced to properties of individual motor units, metabolic processes or single
neural firings. Motor units, metabolic processes and neurons do not possess a
stance themselves. This is how synergetics solves the problem of the part and
the whole, which we have already discussed briefly.
Figure 1.3 Left panel: the increase of the inter-strike time interval for increasing of D control parameter
(right-to-left direction). The whole cycle of leaning forward to strike – upright posture restoration –
leaning forward to strike, increases, showing increase in the relaxation time toward the upright
parallelfoot stance. For D0 – D < 0.05, the relaxation time towards parallel stance tends to become
infinite; that is, no relaxation toward the parallel stance exists anymore. The parallel stance loses its
stability and transits to a more stable diagonal stance; see right panel (modified from Chow et al. 2009,
with kind permission from Nonlinear Dynamics, Psychology and Life Sciences)

Because the components of a complex system are mutually interacting, any

influence of one component affects a large set of other components.
However, this influence is recurrently fed back to the same component
because it is itself under the influence of the rest of the components in the
system. This self-interaction is the root of nonlinearity and circular causality.
Hence, each component self-interacts through its influence on other
components. An illustrative example of a self-interaction is the co-adaptive
behaviour of two players. Player A reacts to the actions of the player B and
player B adapts his actions with respect to player A's actions (see, e.g. Passos
et al. 2008). In this sense, each of them changes their behaviour by changing
the behaviour of the other.
Now, let us assume that, at some critical value of the influential parameter,
a small initial seed of behaviour has been formed by a few components.
These affect other components in such a way that makes them behave in the
same way or some kind of cooperative fashion. The bottom-up cooperation of
the components forms a macroscopic collective order. As more and more
components are recruited in this recurrent way, the top-down influence of the
macroscopic collective order becomes increasingly stronger. New
components are recruited in a bottom-up fashion in an ever increasing rate.
These newly recruited components add to the strength of the collective
macroscopic order that has emerged and, in turn, recruit more components. It
is important to note how top-down and bottomup processes cooperate.
However, depending on the value of the constraints impinging on the system,
this recruitment process will be stabilized around some value. Then, the same
entangled recurrent interactions of components will act as a negative
feedback, stabilizing the newly formed collective order. If some small subset
of components for some reason tries to change its behaviour, the collective
interaction field will force it to behave in a collective fashion. The collective
field governs the components that form macroscopic patterns of behaviour.
This is why self-organization is proclaimed to be a spontaneous pattern-
forming process.
In this way, an enormous amount of information contained in the large
number of degrees of freedom becomes reduced and the system deals with
one or only few of them. This is the reason for calling these events self-
organization. The complex system self-organizes by reducing its large
dimensionality to only one or a few variables without the information bearer,
a designer, imposing order from outside. Such variables are of great
importance for researchers and coaches because they contain the compressed
information about a system's component behaviours. In this sense, collective
variables are essential because they are relevant for guiding the innumerable
microscopic components which depend on them. That is why the search for
such macroscopic variables is important in sports science research. Their
behaviour informs us about the behaviour of the microscopic components;
i.e. about their macroscopic task-dependent functional role. Examples of self-
organizing systems within biological systems and their levels and processes
are many: human movement (Kelso 1984), the human brain (Fuchs et al.
1992), central pattern generators (Wojcik et al. 2011), intracellular
organization (Daga et al. 2006), cell aggregates and tissue formation
(Garfinkel et al. 2004) to name a few.
We have seen that the stability properties of the system depend on the
values of influential constraints of the system. They are the generators of
stability and instability in complex systems, being able to control a system's
stability properties. That is why these influential parameters are known as
control parameters (e.g. scaled distance D in the previous example). Such
parameters are nonspecific, meaning that they do not refer to the specific
properties of the system's collective variables. In complex dynamical
systems, a large set of control parameters may be present, each one acting on
different spatial or time scales. In sport settings, such control parameters are
usually formed by confluences of informational task, personal and
environmental constraints. For example, hormonal, neuromodulator and
neurotransmitter concentrations, synaptic strengths, network topologies,
morpho-anatomical properties, psychological states and motor abilities may
play a role as personal constraints
A vast set of variables may play the role of collective variables, i.e. order
parameters, owing to their context dependence, task specificity or level of
analysis. Collective variables may take the form of, for example, the
amplitude of oscillations (Haken 1987), concentrations in reaction–diffusion
processes (Meinhardt and Gierer 2000) and phase relations between
neurobiological system components (Kelso 1984; Haken et al. 1985). In
complex networks and actions, these may include local (e.g. Arenas et al.
2006; Hristovski et al. 2011,) or nonlocal correlation quantities (e.g. Endres
et al. 2011), orientations and directions (Hristovski et al. 2006), density
differences (e.g. in gas–liquid transitions). Generally, angle variables (of real
or formal spaces) are good candidates for order parameters. In particular,
useful collective variables may be eigenvectors extracted by principal
component analysis (Jirsa et al. 1994; Maisuradze et al. 2009;
Balasubramaniam and Turvey 2004), which may be subsequently
dynamically modelled.
Bifurcations are discontinuities in functional dependence
One of the properties of bifurcation or phase-transition points is the onset of
discontinuity of a behavioural state variable with respect to some control
parameter (or set of control parameters). This discontinuity, i.e. singularity,
may be manifested as a change of the functional dependence of the
behavioural variable for a certain value of the control parameter. This is why
bifurcations are known as points of qualitative system change. In linear
systems, knowing the law of change of the representative curve at one point,
we can extrapolate its value at any other point. This provides the feasibility of
extrapolating our knowledge about one region of a system's behaviour to any
other. This is not always the case for the behaviour of nonlinear systems. For
certain values of control parameter(s), a point of discontinuity arises. By
knowing the curve of change in one parameter region, we are not able to infer
anything about its behaviour in the other region and vice versa, unless we
know the basic dynamic characteristics of the system. This is called a non-
analyticity of the function, i.e. behaviour. In the sport context, it points to the
non-proportionality of stimulus–response (or dose–effect) relationship
present in nonlinear biological systems such as athletes. The property of non-
analyticity may often lead coaches wrongly to extrapolate the behaviour of
athletes, knowing their behaviours only in a certain control parameter range.
This kind of extrapolation may be a frequent cause of the emergence of
overtraining states in athletes, where the role of a control parameter is usually
some workload quantity (e.g. Hristovski et al. 2010). Moreover, we may
wonder how these discontinuities arise in the first place when, e.g.
physiological functions within athletes or the workload change more or less
continuously during exercise. Such questions bring us to the second
important characteristic of bifurcation points. In such points, old stable
behaviours die and new ones occur and the number of possible behaviours
may change.
Notice here the difference between linear and nonlinear systems. We have
already noted that linear systems respond to any minute change of the control
parameter with a minute change in their behaviour. In nonlinear systems,
such minute changes may accumulate and no detectable or only a small effect
may be noticed. Then, with a further small change, a qualitative dynamic
change may emerge. It is as if the system slowly accumulates a residual
potential for change and then releases it in the form of discontinuous
reorganization of behaviour. This behaviour cannot be captured by linear or
nonlinear regression procedures.
Rearrangement and breaking of symmetry
At bifurcation points, the symmetry of the state of the system may change.
Before the bifurcation, the system may be in a monostable state of symmetry.
For example, consider a team sport situation in which all your teammates are
well marked by opponents and you have ball possession. Each teammate's
action is fully compensated by an opponent's actions. You see the same
environment wherever you look. This is a symmetrical state of the system.
No passing opportunity emerges. Hence, no game starts. The situation is
stable. Then, one of your teammates succeeds in avoiding a marker. You pass
the ball to her/him. The system symmetry is broken and the game starts, all
over again. The game dwells on broken symmetries.
But the previous symmetry may simply rearrange itself into two (or more)
equally attractive symmetrical states. Two players simultaneously lose their
markers and your decision may be again stuck, at least for a moment – an
instance of the ‘Buridan's donkey’ problem in sport. In symmetrical cases, the
system usually switches to one of the newly emerging states because of any
random event. So, although there are two equally attractive states, the system
selects the behavioural solution spontaneously, as if by chance, relaxing into
one of the equally attractive states. This process is an instance of spontaneous
symmetry breaking because there is no clearly discernible constraint that
forces the system to prefer one of the solutions more than the other. This kind
of bifurcation is connected to what is referred to as a second-order phase
transition or a pitchfork bifurcation.
However, usually in any performance context there exists a bias in
informational constraints. Such events are connected to what is referred to as
a first-order phase transition or a saddle-node bifurcation. In this case,
asymmetry is induced into the system by some asymmetric control parameter.
This kind of symmetry breaking is called forced symmetry breaking, because
some kind of bias forces the system to contain asymmetric behavioural
solutions. In sports settings, an example of this kind of symmetry breaking
emerges when one of the teammates has a better position to score a goal or
point than the other. The difference in the positions of the two teammates
induces a bias which constrains the decision of the player in possession of the
ball, so s/he passes the ball to the one who is better positioned. The diagonal
stance in martial arts also may play a role of a symmetry breaking parameter,
i.e. a bias-inducing constraint, forcing either left- or right-handed punches to
be more used, respectively.
In forced symmetry breaking systems, an interesting effect can arise –
hysteresis. Hysteresis could arise in the previous example if the player who is
in possession of the ball continues to pass the ball to one of the teammates,
although the other teammate attains a better position to score. At one
moment, the player who has ball possession decides to pass it to the player
who is better positioned for a score. But, now he becomes stuck by passing to
this teammate, even though the other teammate becomes better positioned for
scoring. This type of inertia or system ‘memory’ is typical for systems with
forced symmetry breaking. As we stated earlier, this phenomenon cannot be
found in pitchfork bifurcations, i.e. the second-order phase transitions.
When initial symmetry is broken and the system becomes unstable owing
to combinations of control parameters, dynamic systems become weakly
unstable or metastable and dwell for a long time close to the remnants of the
attractor points. This happens when, under the influence of control
parameter(s), the stability of the system is lost but the dynamics are trapped
in the vicinity of the previously stable point, i.e. attractor. The system
eventually escapes to be temporarily trapped in the vicinity of another
attractor remnant. This semi-transient behaviour is a result of purely
deterministic dynamics and is a case of relative coordination (Kelso 1995),
where components are not tightly integrated but only weakly so, allowing
them to flexibly decouple and form new transient relations. More information
on this extraordinary mechanism of system flexibility may be found in
previous published works (e.g. Bressler and Kelso 2001 and the references
therein and, particularly in sport contexts, Chow et al. 2011).
The overall message is that bifurcations, i.e. thresholds in nonlinear
systems, may be of different kinds and they cannot be thought of as fixed set
points given once for all time but as points that depend on the combinations
of control parameter values and symmetries of the system. Because these
parameters are formed by combinations of natural constraints impinging on
the system, this would mean that the dynamics of such complex systems and
threshold values are guided by those constraints.
Fluctuations, exploration and rugged potential landscapes
In the deterministic approach to dynamical systems when a system reaches a
stable state, it keeps that value for an infinitely long time if control
parameters are not changed. However, all complex systems in nature possess
intrinsic noise, that is, random-like processes that live on timescales faster
than the one observed or researched. Such processes do not allow the system,
described by some behavioural variable, to stay in an attractor state but push
it on the both sides of this state. They counteract the stabilizing, negative
feedback loops, ‘forces’ which try to pull the system into the attractor state.
Close to instability points such fluctuations are enhanced, owing to the larger
decoupling tendencies among the system components. This phenomenon is
called critical fluctuation enhancement and together with critical slowing
down are hallmarks of phase-transition phenomena.
When noise is present, the system exhibits fluctuations around the
attractor, i.e. around the steady value or trajectory. In this case, the attractor
can be defined as the most probable point (or trajectory) of the dynamics.
Since the different values of the behavioural variable actually refer to
different configurations, i.e. patterns, of the components that form them, it
follows that the attractor is the most probable configuration of the system.
Depending on the shape of the attractor basin, other configurations take less
probable values. The larger the deviation from the attractor, the larger is the
amount of component reconfiguration that occurred. It seems that
fluctuations make the system explore its own capacity for pattern production.
At the beginning of exploration, the system has not well explored all its
available configurations and is in a non-equilibrium state. After some longer
time period, it has produced all of the available patterns many times and it
has reached some stationary probability distribution. The system has
equilibrated. However, in nonlinear systems, there may be two or more
attractor basins separated by hills or barriers, on top of which the repeller is
located (see Figure 1.2). So, while the system has explored the basin of
attraction where it was first located and equilibrated there, it may still not
reach other available attractor basins. The system has locally equilibrated but
not globally. The rate of this global equilibration depends on the height of the
barriers and on the strength of fluctuations. It is proportional to the strength
of fluctuations and inversely proportional to the height of the barriers. This
kind of dynamics of exploration is also called metastability (e.g. Baker 1998).
If the system is weakly stable, i.e. there exist large fluctuations and low
barriers between stable attractors, the rate of exploration may be very large
and vice versa, high barriers and low fluctuations would lead to small
exploratory rates. These ideas highlight the functionality of system
fluctuations for adaptive behaviours, for example, motor learning (Schollhorn
et al. 2009). In complex systems, control parameters may be themselves
subject to noise. Interesting phenomena may arise around critical points,
which may smear the critical point (i.e. the threshold) of the system and
produce two- instead of one-phase transitions, i.e. noise-induced transitions.
For example, lactate thresholds are sometimes difficult to locate with respect
to the load intensity taken as control parameter.
Complex systems often have a complex structure of attractor basins. There
may be basins within basins within basins of attraction. In this case, we speak
of rugged potential landscapes (e.g. Hristovski et al. 2011). These systems
have a large spectrum of relaxation times and the equilibration process may
be very slow. They may never explore the whole space of possible state
configurations, i.e. patterns. For example, a non-gymnast would never
reconfigure his/her action into a Tsukahara vault or a gymnast may never
engage the fat system of energy production during workouts. These systems
are non-ergodic. Ergodicity signifies that, if the system starts with any initial
configuration, during the equilibration, i.e. exploration, process, it will visit
all other configurations many times in some large time limit. For non-ergodic
systems, some or even many possible configurations are unavailable. Task
and other constraints acting on the system do not allow this to happen and
break the ergodicity, allowing the system to equilibrate only locally within
relatively small regions. Specific constraints combine to arrest the system
within limited configuration space and, if noise strength is small compared
with barrier heights, make the system explore for a longer time only a limited
set of patterns before it transits to another basin of attraction, where it will
face even larger barriers. Over larger time scales, the relaxation becomes
increasingly slower. Thus, the trapping role of the constraints may produce
correlated fluctuations. Depending on the distribution of the barrier height
hierarchy, the global relaxation (equilibration) times of patterns or time
correlations may take a form of a stretched exponential, power law, show
plateau or even logarithmic relaxation. In this sense, athletes may be viewed
as non-ergodic, out of equilibrium systems, exploring larger and larger
regions of the state space but eventually getting trapped within some
relatively small set of the whole state space by the constraints of their sport
discipline. Within such regions, some local equilibration may be attained in
the long run.
From a practical point of view, these noise phenomena may explain why
precise values of thresholds or threshold points in complex biological
systems, such as athletes, are hard or sometimes impossible to define. They
may be changed by different combinations of control parameters values, as
we have noted earlier, but also they may be smeared around some mean value
by the influence of intrinsic biological fluctuations. They may have narrower
or wider threshold regions instead of threshold points, pointing to the
intertwined nature of stochastic and deterministic processes in nonlinear
complex dynamical systems. Finding an abrupt change of the probability
distribution function as a consequence of a small control parameter change is
one possible way to test the critical behaviour of nonlinear systems.
One can also see how the unpredictable behaviours of complex systems
arise. The multistability of the system, which is a product of nonlinearity,
enables the system to be in more than one state for the same value of
parameter. A random event, i.e. a fluctuation, may push the system into a less
expected state of organization, perhaps indicative of a lower performance
level. Hence, nonlinearity and noise in weakly stable systems might
synthesize a behaviour that does not always have to have ‘a good reason’ to
emerge. The only way to reduce the probability of such erratic behaviours is
to make attractors more stable by changing the control parameter(s) in an
adequate direction. But sometimes such weakly stable behaviours are
desirable because they may produce novel functional performance behaviours
which no one knew existed. In this way, complex systems are also creative, a
valuable resource in sport performance.
References
Arenas, A., Guilera, A. D., and Perez-Vicente, C. (2006) Synchronization
reveals topological scales in complex networks. Physical Review Letters, 96
(11): 114102.
Baker, D. (1998) Metastable states and folding free energy barriers. Nature
Structural Biology, 5: 1021–4.
Balasubramaniam, R. and Turvey, M. T. (2004) Coordination modes in the
multisegmental dynamics of hula hooping. Biological Cybernetics, 90: 176–
90.
Bressler, S. L. and and Kelso, J. A. S. (2001) Cortical coordination dynamics
and cognition. Trends in Cognitive Sciences, 5 (1): 26–36.
Chow, J.-Y., Davids, K., Button, C., Rein, R., Hristovski, R. and Koh, M.
(2009) Dynamics of multi-articular coordination in neurobiological systems.
Nonlinear Dynamics, Psychology and Life Sciences, 13 (1): 27–55.
Chow, J.-Y., Davids, K., Hristovski, R., Araújo, D. and Passos, P. (2011)
Nonlinear pedagogy: learning design for self-organizing neurobiological
systems, New Ideas in Psychology, 29 (2): 189–200.
Daga, R., Lee, K.-G., Bratman, S., Salas-Pino, S. and Chang, F. (2006) Self-
organization of microtubule bundles in anucleate fission yeast cells. Nature
Cell Biology, 1108–13.
Davids, K., Handford, C. and Williams, M. (1994) The natural physical
alternative to cognitive theories of motor behaviour: an invitation for
interdisciplinary research in sports science? Journal of Sport Sciences, 12 (6):
495–528.
Davids, K., Glazier, P., Araújo, D. and Bartlett, R. M. (2003) Movement
systems as dynamical systems: the role of functional variability and its
implications for sports medicine. Sports Medicine, 33: 245–60.
Endres, M., Chenau, M., Fukuhara,T., Weitenberg, C., Schaus, P., Gross, C.,
Mazza, L., Banuls, M. C., Pollet, L., Bloch, I. and Kuhr, S. (2011)
Observation of correlated particlehole pairs and string order in low-
dimensional mott insulators. Science, 334 (6053): 200–3.
Fuchs, A., Kelso, J. A. S. and Haken, H. (1992) Phase transitions in the
human brain: spatial mode dynamics. International Journal of Bifurcation
and Chaos, 2: 917–39.
Garfinkel, A., Tintut, Y., Petrasek, D., Bostrom, K. and Demer, L. L. (2004)
Pattern formation by vascular mesenchymal cells. PNAS, 101 (25): 9247–50.
Haken, H. (1987) Synergetics: an approach to self-organization in self-
organizing systems, in Self-organizing Systems: The Emergence of Order, F.
Eugene Yates (ed.) Life Science Monographs 21. New York: Plenum, pp.
417–34.
Haken, H., Kelso, J. A. S. and Bunz, H. (1985) A theoretical model of phase
transitions in human hand movements. Biological Cybernetics, 51: 347–56.
Hristovski, R., Davids, K. and Araújo, D. (2006) Affordance – controlled
bifurcations of action patterns in martial arts. Nonlinear Dynamics,
Psychology and Life Sciences, 4: 409–44.
Hristovski, R., Venskaityte, E., Vainoras, A., Balagué, N. and Vazquez, P.
(2010) Constraints-controlled metastable dynamics of exercise-induced
psychobiological adaptation. Medicina, 46: 447–53.
Hristovski, R., Davids, K., Araújo, D. and Passos, P. (2011) Constraints-
induced emergence of functional novelty in complex neurobiological
systems: a basis for creativity in sport. Nonlinear Dynamics, Psychology and
Life Sciences, 15 (2): 175–206.
Jirsa, V. K. and Kelso, J. A. S. (2005) The Excitator as a Minimal Model for
Coordination Dynamics of Discrete and Rhythmic Movement Generation.
Journal of Motor Behaviour, 37 (1): 35–51.
Jirsa, V. K., Friedrich, R., Haken, H. and Kelso, J. A. S. (1994) A theoretical
model of phase transitions in the human brain. Biological Cybernetics, 71:
27–35.
Kelso, J. A. S. (1984) Phase transitions and critical behavior in human
bimanual coordination. American Journal of Physiology: Regulatory,
Integrative and Comparative, 15: R1000–4.
Kelso, J. A. S. (1995) Dynamic Patterns. Cambridge, MA: MIT Press.
Kugler, P. N., Kelso, J. A. S. and Turvey, M. T. (1980) On the concept of
coordinative structures as dissipative structures: I. Theoretical lines of
convergence. In G. E.
Stelmach and J. Requin (eds) Tutorials in Motor Behaviour. Amsterdam:
North Holland, pp. 49–70.
Maisuradze, G. G., Liwo, A. and Scheraga, H. S. (2009) Principal component
analysis for protein folding dynamics. Journal of Molecular Biology, 385 (1):
312–29.
Meinhardt, H. and Gierer, A. (2000) Pattern formation by local self-activation
and lateral inhibition. BioEssays, 22: 753–60.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Milho, J. and Serpa, S.
(2008) Informationgoverning dynamics of attacker–defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Schmidt, R. C., Carello, C. and Turvey, M. T. (1990) Phase transitions and
critical fluctuations in the visual coordination of rhythmic movements
between people. Journal of Experimental Psychology: Human Perception
and Performance, 16 (2): 227–47.
Schollhorn, W. I., Mayer-Kress, G., Newell, K. M. and Michelbrink M.
(2009) Time scales of adaptive behavior and motor learning in the presence
of stochastic perturbations. Human Movement Science, 28 (3): 319–33.
Schoner, G., Haken, H. and Kelso, J. A. S. (1986) A stochastic theory of
phase transitions in human hand movement. Biological Cybernetics, 53: 247–
57.
Wojcik, J., Clewley, R. and Shilnikov, A. (2011) Order parameter for
bursting polyrhythms in multifunctional central pattern generators. Physical
Review E, Statistical, Nonlinear, and Soft Matter Physics, 83 (5 Pt 2):
056209.
Zanone, P. G. and Scott Kelso, J.A. (1992) Evolution of behavioral attractors
with learning: nonequilibrium phase transitions. Journal of Experimental
Psychology: Human Perception and Performance, 18 (2): 403–21.
2 Coordination dynamics and
cognition1
J. A. Scott Kelso
Could he whose rules the rapid comet bind, describe or fix one
movement of his mind?
Alexander Pope
In his preface to the Principia, Isaac Newton speculated that not just the
motions of the planets, the Moon and the tides could be explained by the
forces of attraction and repulsion but all other natural phenomena as well
(Pope's eulogy to Newton, quoted above, was rather more sceptical).
Although Newton himself was cautious about these as yet unknown forces,
others in many fields set out to become the Newtons of their time. After all, if
Newton could explain celestial events through a simple gravitational force,
more mysterious areas of inquiry should be open to the same approach.
For the eighteenth century Scottish philosopher David Hume, all the
sciences bore a relation to the human mind. In his Treatise of Human Nature
(1738), Hume first divided the mind into its contents: ideas and impressions.
Then he added dynamics, noting the impossibility that simple ideas could not
form more complex ones without some bond of union between them. Hume's
three dynamical laws for the association of ideas – resemblance, contiguity
and cause and effect – were thought to be responsible for controlling all
mental operations. A kind of attraction, Hume thought, existed in the mental
world: the motion of the mind was conceived as analogous to the motion of a
body. Mental ‘stuff’ was governed (somehow) by dynamics.
It is practically a cliché that human cognition and behaviour, not to speak
of human social interactions, unfold in time. Human behaviour is emergent:
patterns of behaviour arise from the way that individual parts or processes are
coordinated. Context is crucial. Nonlinearity rules: small changes sometimes
produce large effects and large changes no effect at all. Interactions,
nonlinearity, emergence and meaningful contextual information, although
ubiquitous in the cognitive, brain and behavioural sciences have proven
remarkably resistant to understanding. Yet such understanding is obviously
central to the development of a truly dynamic cognition (Kelso 2000, 2008,
2012a).
In this chapter, the path towards a dynamic cognition is deemed to lie in
coordination dynamics, a line of scientific enquiry that aims to understand,
through theory, analysis and experiment, how patterns of coordinated
behaviour emerge, persist, adapt and change in living things in general and
human beings (and brains) in particular (Kelso 1995). Cognition arises from
the coevolution of brains, bodies and the environment in which they are
immersed; tempered, of course, by developmental and learning processes.
Specific cognitive functions require coordination within and between
specialized regions of the brain. The dual nature of this coordination, how the
numerous parts of the brain retain their local individualized functions while
interacting to form global context-dependent spatiotemporal patterns of
neural activity may be understood through coordination dynamics, especially,
as we shall see, in its multi- and metastable regimens. This theme of duality
or coexistence permeates the entire chapter. Complementary pairing is
fundamental to coordination dynamics (Kelso and Engstrom 2006).
Conceptual distinctions such as ‘programmes’ versus ‘selforganization’,
‘representations’ versus ‘dynamics’ evaporate in very much the same way as
the wave/particle duality in quantum mechanics. Within coordination
dynamics itself, basic dynamical concepts like attractor and repeller, stability
and instability, etc., go together like bread and cheese: one does not make any
sense without the other. The coordination dynamics of complex systems in
biology, psychology and the social sciences thus embraces a fundamental
complementarity (Kelso 1995). In complex systems and the science of
coordination, dichotomy is seldom if ever the path to understanding.
In the next section, this dual, complementary nature of coordination
dynamics is described, since it is crucial to the development of a truly
(spatiotemporal) dynamic cognition. Self-organizing dynamics creates
information and information modifies and directs dynamics. The following
section stresses the need to identify relevant variables, parameters and their
dynamics on a given level of description and articulates a research strategy
(‘constructive reduction’) for doing just that, and for relating levels. A brief
and non-inclusive summary of evidence is then presented for the first
conceptual foundation of coordination dynamics, spontaneous self-organizing
processes. Following this, the second, coexistent foundation of coordination
dynamics is addressed: meaningful, often directed, information. Emphasis is
placed on metastable coordination dynamics. This is the regimen in which
the tendency of individual component parts and processes to function
independently (sometimes called segregation) coexists with the tendency to
coordinate together (sometimes called integration). The metastable regimen is
also where new information is created without any need for amplification. In
coordination dynamics, information may take the form of specific perceptual,
cognitive, linguistic and emotional constraints on behaviour. The concepts
and tools of coordination dynamics are elaborated further in a section on the
context of human skill learning, especially the particularly thorny problem of
identifying pre-existing biases that influence and are influenced by the
learning process. A possible neural coordination dynamics for handling the
joint influences of learning and attention is proposed. In the brain, changes in
synaptic weights during learning are accompanied by attention-related
neuromodulatory changes that take place in cell bodies, the soma. It may be
that cell bodies, including astrocytes and synapses function as a bi-
directionally coupled unit (Wade, et al. 2011). In the final section of the
chapter, some of the main messages of the coordination dynamics of learning
are briefly presented, in the hope that these may be useful to practitioners and
education policy makers, together with a short summary and some
concluding remarks.
The two coexistent foundations of coordination dynamics:
spontaneous tendencies and meaningful information
Coordination is not just matter in motion. It refers to different kinds and
degrees of functional (dis)ordering among interacting parts and processes in
space and time. Such functional ordering comes in many guises, from genes
to cells to neural ensembles to behaviour (both individual and group) and
cognition. Coordination dynamics seeks to identify the laws, principles and
mechanisms underlying the coordinated behaviour of a variety of different
systems at multiple levels of description (for a comprehensive set of
encyclopaedia articles, see Fuchs and Kelso 2009; Oullier and Kelso 2009;
Kelso 2009a). Coordination dynamics contains two coexistent aspects, a self-
organizing, ‘undirected’ aspect and a ‘directed’ aspect. Coexistence is
inherent to the coordination dynamics of cognition because cognitive
function is inherently informational: information, as we shall see later in this
chapter, is created by self-organizing dynamical processes and, in turn,
modifies or directs the dynamics (Kelso 2002).
The first conceptual foundation of coordination dynamics, synergetic
selforganization, refers originally to collective or cooperative effects that
arise spontaneously when ordinary matter takes on novel properties, as in
lasers and superconductors, or when new forms of organization among water
molecules arise, as in the weather (e.g. Haken 1977; Nicolis and Prigogine
1977). In such pattern forming systems, cooperation among the many
microscopic components emerges, not because of some special coordinating
agent or instruction set residing inside the system but rather as a result of the
system's ability to organize itself under specific boundary constraints. The
second, coexistent root of coordination dynamics, complementing
spontaneous self-organizing processes, concerns the special nature of the
boundary constraints (parameters, initial conditions) in human cognition and
behaviour. This ‘directed’ aspect of coordination dynamics deals with how
information is created de novo and how information guides, directs,
modulates and is modulated by self-organizing processes (Kelso 1995).
‘Directed’ informational terms, such as ‘plans', ‘programmes', ‘intentions',
and so forth, common in cognitive science, are embraced rather than reified
in coordination dynamics. Unlike the behaviour of inanimate things, the self-
organizing dynamics of human brains and the behaviour they produce is
fundamentally informational, although not in the standard sense of data
communicated across a channel (Shannon and Weaver 1949). Rather, in
coordination dynamics, the coordination variables are themselves context
dependent and intrinsically meaningful. Context dependence does not imply
lack of reproducibility, nor does it mean that every new context requires new
coordination variables and dynamics. What it does mean is that specific
information (situational context, perception of a task to be learned, intention
to change behaviour, attention devoted to particular task components, a
template or memory of the pattern to be produced, etc.) can be expressed in
terms of parameters acting on system-relevant coordination variables. Thus,
the full coordination dynamics contains a term defining ‘informational
forcing’ and a term expressing spontaneous, self-organizing processes (see
Kelso 1995, p. 163). In recent theoretical developments, such as so-called
parametric stabilization (Jirsa et al. 2000), parameters and variables may act
multiplicatively, meaning that a parameter can directly affect a relevant
variable and vice versa. It is also possible that the coordination variable or
‘order parameter’ can act back on to parameters and modify them; a kind of
nonlinear feedback process. In terms of scientific understanding, the benefit
of identifying relevant variables is that one knows what to parametrically
modify. Likewise, the structure of the self-organizing dynamics – prior to the
introduction of new information – influences how that information is used.
Thus, information is not lying out there as mere data: information is
meaningful and specific to the extent that it modifies and is modified by the
underlying dynamics.
Identifying coordination dynamics in complex systems
In the brain, cognitive, behavioural and social sciences, neither the key
coordination quantities nor their coordination dynamics are known a priori
and have to be identified. Brain structure and function, for example, appears
to be hierarchically organized; the component elements on multiple levels of
description are heterogeneous and the relevant coordination variables
characterizing ongoing neural and cognitive processes are invariably context,
task or function dependent.
How do we discover relevant coordination variables in such complex
systems? And which, if any, can we reasonably discard as not relevant? In the
following, the six-step research strategy of coordination dynamics is
presented, which can be (and has been) applied to different fields of study. A
possible limitation of this strategy is that it appears to be restricted to
laboratory settings. However, many of the same steps – or at least the
questions raised by these steps – may be adopted in other contexts, such as
observational or field studies, including sports (see, e.g. Duarte, et al. 2012;
Passos et al. 2008).

Step 0: Choose a level of description

In coordination dynamics, the key is to choose a level of description and
understand the (coupling) relation between adjacent levels, not (or not only)
reducing to some ‘fundamental’ lower level. The choice of a level of
description is the scientist's and can only be made with informed insight.
Choice of a level of description is not an issue of a ‘top-down’ versus
‘bottom-up’ approach. In coordination dynamics, the first step at any level of
description, whether one is studying biomolecular processes or social
behaviour, requires the identification of relevant coordination variables and
their dynamics.

Step 1: Prune away complications

Science always needs special entry points; places where irrelevant details
may be pruned away while retaining the essential aspects of the phenomena
that one is trying to understand. One may call this the Galilean strategy. By
studying balls rolling down an inclined plane and abstracting away friction,
Galileo was able to provide deep insights into the nature of falling bodies and
planetary motion. There is no formula or recipe that may be applied to this
step. One has to find or even invent experimental model systems or situations
that distil away irrelevant aspects, leaving the ones that really matter.

Step 2: Use qualitative change

Unique to the perspective of self-organizing coordination dynamics is its
focus on qualitative change, places where behaviour bifurcates or a phase
transition occurs (Haken et al. 1985; Kelso 1984). Qualitative change is
important for both theoretical and empirical reasons. Theoretically, dynamic
instability is the generic mechanism underlying spontaneous self-organized
pattern formation and change in all systems coupled to their internal or
external environment (Haken 1983; Nicolis and Prigogine 1977).
Experimentally, qualitative change affords a clear distinction between one
pattern and another, thereby enabling one to identify the key coordination
variables that characterize patterned states. In a typical situation, many
variables may be changing some smoothly and linearly. However, the one(s)
that change(s) abruptly are likely to be the most informationally meaningful,
both for the system itself and for the scientist trying to understand them
(Kelso 1994b). Qualitative change may also be used to infer relevant
quantities in more naturalistic settings (for a lovely example, see Beekman et
al. 2001). Instead of driving the system toward instabilities, for example, one
can examine time series of complex interacting systems for places where
abrupt changes occur and use these to identify the underlying dynamics – a
kind of reverse engineering approach.

Step 3: The coordinative level

Finding collective or coordination variables on a chosen level of description
(neural, behavioural, cognitive) is the ‘yin’ of the coordination dynamics
research strategy. The ‘yang’ is identifying the control parameter(s) that
cause coordinative change. Naturally occurring environmental conditions or
intrinsic, endogenous factors may qualify as control parameters.
Technically speaking, a parameter is a control parameter if, when it
changes smoothly and continuously, the coordinative behaviour of the system
changes qualitatively and abruptly. Thus, when a control parameter crosses a
critical value, instability occurs, leading to the formation of new (or different)
patterns. The payoff from knowing coordination variables and control
parameters is high: they enable one to obtain the dynamical rules, i.e., the
equations of motion that describe the stability and change of patterned
coordinative states on a given level of description. Coordination variables and
control parameters (whether specific or nonspecific) are complementary: you
do not have one without the other and vice versa. Moreover, they may be
interchangeable, depending on level of description (Kelso 1995). For
example, in studies of phase transitions in bimanual coordination movement,
frequency is a control parameter at the coordinative level and a coordination
variable at the component level.

Step 4: The component level

By adopting the same strategy at the next level down, the individual
component dynamics may be studied and identified. In general, in complex
systems, it is difficult to isolate the components and study their dynamics.
The reason is that the individual components seldom exist outside the context
of the functioning whole and have to be studied as such. Just as an example,
recent work on interpersonal coordination (Oullier et al. 2008) shows that
individuals retain a memory of the interaction with a partner, even when they
are no longer coupled.

Step 5: Establish relations among levels

A ‘final’ but nontrivial step is to derive the coordinative level dynamics from
the, in general, nonlinear coupling between the individual components. This
is what some people call ‘emergence’ and it allows for a bridge to be built
across different levels of description in a systematic fashion. It should be
clear that no level is any more or less ‘fundamental’ than any other. Likewise,
there are no absolute ‘macro’ or ‘micro’ levels. The complete coordination
dynamics on a given level of description always requires three adjacent tiers:
the specific boundary conditions and control parameters that establish the
context for a particular phenomenon to arise; the coordinative level and its
dynamics; and the component level and its dynamics. Notice that the
descriptors at each level are always different from each other. However, the
language of coordination dynamics serves to bridge different domains. I call
this constructive reductionism: by starting with a coherent description of
behaviour on one level and by focusing on adjacent levels ‘above’ (boundary
constraints) and ‘below’ (individual component parts and processes), the
behaviour of the whole may be seen as ‘emerging’ from the nonlinear
interactions among component subsystems. The strategy of constructive
reductionism can be implemented at least ‘one level down’. In recent times, a
complementary picture has been forming – namely that laws of coordination
in neurobehavioural systems deal with collective properties that emerge from
microscopic dynamics but are not deducible from them. Such ‘mesoscopic
protectorates’ (after Laughlin and Pines 2000) are generic, emergent
behaviours that are reliably the same from one system to another, regardless
of details, and are repeatable within a system on multiple levels and scales
(Kelso 2012b; see also Turvey and Carello 2012). The notion that laws of
coordination are truly emergent and sui generis suggests that it may not be
possible – even in principle – to deduce psychological-level descriptions
from (more microscopic) neural- or molecularlevel descriptions. As stressed
here in step 5, this does not mean that we should not try to understand the
relationships between different levels. Rather, the task is to come up with
lawful descriptions that allow us to understand emergent behaviour at all
levels and to respect the autonomy of each.
The self-organizing nature of coordination dynamics
Are the foundational concepts of self-organization in physical, chemical and
biochemical systems relevant to cognitive and behavioural function? Over the
last three decades, starting with experiments on people (Kelso 1981, 1984)
and theoretical modelling of such (Haken et al. 1985), it has been shown that
the same coordination dynamics (equations of motion whose parameters alter
the stability and change of coordination patterns over time, together with the
nonlinear coupling among components that gives rise to them) apply to the
functional coordination among anatomically different parts and processes,
including but not limited to: intentional movements of two or more fingers
and limbs (Carson et al. 1995; Fuchs and Kelso 1994; Haken et al. 1985;
Kelso 1984; Kelso and Jeka 1992; Peper et al. 1995; Schöner et al. 1990;
Swinnen et al. 1997; Treffner and Turvey 1996); coupling among the joints
of a single, multi-jointed limb (Buchanan and Kelso 1993; Carson et al.
1999; Kelso et al. 1991); perceptionaction coupling between visual auditory
and tactile stimuli and motor responses (Kelso et al. 1990, 1998; Lagarde and
Kelso 2006; Stins and Michaels 1999; Wimmers et al. 1992); postural sway
(Bardy et al. 1999; Dijkstra et al. 1994; Jeka et al. 1997); visually mediated
coordination between two people (Amazeen et al. 1995; Dumas et al. 2010;
Naeem et al. 2012; Riley et al. 2011; Schmidt et al. 1990; Tognoli et al.
2007); between humans and computers (Kelso et al. 2009); and even between
humans and other species (Lagarde et al. 2005). In numerous situations, the
coordination dynamics – at both the coordinative level and at the level of the
nonlinear interactions among the components – explicitly incorporates the
role of specific neuromuscular–skeletal constraints. For instance,
eigenfrequency differences between coupled limb movements (Fuchs and
Kelso 1994; Kelso and Jeka 1992; Sternad et al. 1995) as well as neurally
based informational couplings between auditory, tactile or visual stimuli and
movement are known to shape or sculpt the form of coordination observed
(Kelso et al. 1990; Kelso 2010; Riley et al. 2011).
Neural correlates of the stability and change of behavioural coordination
have been revealed using a high-density superconducting quantum
interference device and electroencephalograph arrays (Banerjee et al. 2012;
Daffertshofer et al. 2000; Fuchs et al. 1992, 2000a,b; Kelso et al. 1991, 1992;
Mayville et al. 1999, 2001; Frank et al. 2000; Wallenstein et al. 1995), as
well as functional magnetic resonance imaging and positron emission
tomography (DeLuca et al. 2010; Jantzen et al. 2008, 2009; Meyer-
Lindenberg et al. 2002; Swinnen 2002; Swinnen and Wenderoth 2004; Ullen
et al. 2000; Fuchs et al. 2000c). Theoretical work at the neural level has
progressed from phenomenological modelling at behavioural (Beek et al.
1995; Fuchs and Jirsa 2001; Haken et al. 1985; Jirsa et al. 2000; Kelso et al.
1990, 1993; Schöner et al. 1986, 1990; Treffner and Turvey 1996) and brain
levels (Jirsa et al. 1994; Uhl et al. 1995) to neurobiologically grounded
accounts of both unimanual (Jirsa and Haken 1997; Fuchs et al. 2000; Frank
et al. 2000) and bimanual coordination (Jirsa et al. 1998) that are based on
known cellular and neural ensemble properties of the cerebral cortex. Recent
work has extended this neural theory to include the heterogeneous
connectivity between neural ensembles in the cortex (See ‘Neural
coordination dynamics of learning and attention’ below; Jirsa and Kelso
2000). In all these circumstances, once general laws at behavioural and brain
levels have been identified, it has proved possible to derive them from a
deeper theory grounded in neuroanatomical and neurophysiological data,
thereby causally connecting different levels of description (Kelso et al. 1999
for review). In showing that stability and change of coordination at both
behavioural and neural levels is due to nonlinear interactions among
interacting components some of the mysticism behind the contemporary
terms ‘emergence’ and ‘self-organization’ has been removed.
The informational nature of coordination dynamics
It may be said that the self-organizing nature of coordination dynamics
possesses ‘universal’ properties (Haken 1996; Kelso 1995), seemingly
unrelated systems behaving in essentially the same way. Individual
components or features of complex coordinative systems may be coupled by
material forces, by light, by sound, by touch and by intention (Kelso 2009b;
Kelso et al. 2001 for review and experimental evidence). How then does
information arise? And what does it do? How are the central concepts of
cognitive science terms such as ‘plans’, ‘programmes’, ‘intentions’ and so
forth, handled in coordination dynamics? Let us first consider the creation of
information question.
Elsewhere (Kelso 1995, 2002), I have proposed that the origins of
information lies in the metastable regimen of the coordination dynamics. This
is the regimen in between the idealized states of complete interdependence
between interacting components (such as patterns of phase and frequency
synchronization between neural regions of the brain) and total independence
of the component parts from each other (i.e., each local region of the brain
expresses its own dynamic properties without any interaction with other local
regions).
In the metastable regimen, there are no attractors or repellers in the
coordination dynamics but there is still attraction to where the attractors used
to be. The reason is that intrinsic differences between the individual
components are sufficiently large that they do their own thing, while still
retaining a tendency to cooperate. Thus, the relative phase between the
components may drift in time, but is occasionally trapped near ‘remnants’ or
‘ghosts’ of patterned coordination states. This, I propose, is how global
integration, in which component parts are locked together, is reconciled with
the tendency of the parts to function as locally specialized autonomous units.
Because of metastable coordination dynamics, the brain is able to exhibit a
far more variable, plastic and fluid form of coordination in which tendencies
for integration and segregation coexist. Metastable coordination dynamics is
characteristic of successful organizations (let the parts remain semi-
autonomous while still cooperating loosely) and is especially evident in the
functional organization of the human brain (e.g. Bressler and Kelso 2001;
Friston 2000; Kelso 1992, 1995, 2002, 2012a; Kelso and Tognoli 2007;
Tognoli and Kelso 2009) and cognition (e.g. Chen et al. 1997, 2001; Ding et
al. 2002; Kello et al. 2010; Tuller et al. 1997; van Orden et al. 2005).
How does metastable coordination dynamics relate to the issue of the
origins of information? If the brain is metastable (and the evidence is
compelling at both behavioural and neural levels) it may be conceived as a
measuring device that is poised to create information. Analogous to effective
devices for measuring quantal events (Green 2000), in order to be sensitive to
the world, the brain must exist in the metastable regimen, where the slightest
fluctuation or change in parameters will nudge it into and out of coordinative
states. Such coordinative states are intrinsically meaningful. What could be
more relevant to the brain than information that communicates the
relationship between its component parts? Information can never be known
fully in advance. It must be discovered. The remarkable aspect of metastable
coordination dynamics is that no amplification of microscopic events (as in
quantal emission) is warranted or required. The essentially nonlinear nature
of the coupling between the individual parts and the intrinsic nature of the
parts themselves gives rise to metastability in a natural way. Indeed, in the
Kelso et al. (1990) elaboration of the HKB (Haken–Kelso–Bunz) model,
metastable coordination dynamics is caused by the interplay of two factors,
the intrinsic properties of the component features themselves and the
nonlinear coupling between them. Akin to quantum theory, after the
measurement of information the metastable brain is in a coordinated (read
pure, idealized) state that (importantly) is quantified by the values of
collective states). Selection among meaningful coordination states may occur
in several ways. First, an extremum principle may govern the selection or
choice between equally available collective or coordinated states (Blekhman
1988). Second, selection may occur via parametrically induced instability, the
selection via instability principle enunciated earlier (Kelso 2000). (Still
another selection mechanism – selection via matching – has been revealed in
studies of sensorimotor learning, see Kostrubiec et al. 2012.)
Once created due to the metastability of self-organizing coordination
dynamics and expressed in terms of system-relevant coordination states, what
does information do? Again, cognitive information can play the dual role of
stabilizing and destabilizing behaviour depending on context (initial
conditions). Early and more recent research shows that information is capable
of stabilizing the coordination dynamics even under conditions in which
patterns of coordinated activity typically become unstable and switch. For
example, internally generated intentions (Kelso et al. 1988; Lee et al. 1996)
and cognitive strategies (Kelso et al. 1990; see also Chen et al. 2001) enable
the cognitive system to stay longer in a pattern of behaviour than it normally
would or could. Intention to stay in a pattern can simultaneously recruit new
biomechanical degrees of freedom and annihilate ones that were formerly
engaged (Buchanan and Kelso 1999; Buchanan et al. 1997; Fink et al. 2000a;
Kelso et al. 1993). Perceptual information also serves to stabilize coordinated
behaviour, as in the well-known ‘anchoring’ effect (Byblow et al. 1994; Fink
et al. 2000b) in which specific, attended to perceptual inputs are selectively
coupled to specific aspects of activity. Such so-called ‘parametric
stabilization’ effects have been modelled mathematically within the current
framework of coordination dynamics (Fink et al. 2000b; Jirsa et al. 2000).
Multimodal information from auditory, haptic and intentional sources is
bound together in time in a coherent way that serves to stabilize coordinative
actions (Kelso et al. 2001; Lagarde and Kelso 2006). In more recent work by
Kovacs et al. (2010), a so-called Lissajous template may serve to stabilize
apparently arbitrary phasing patterns between the limbs.
Likewise, perceptual, cognitive, emotive and linguistic information can
also destabilize behaviour. Depending on task context and parameters, such
as stimulation rate, perceptual information can destabilize the cognitive
system, causing it to switch from one pattern of behaviour to another (Kelso
et al. 2001). For example, haptic input that is counterphase to a rhythmic
auditory stimulus can cause a switch in behaviour such that haptic and
auditory inputs coincide with intended movement, forming a
spatiotemporally coherent bond. In a similar vein, intentions can destabilize
behaviour and cause it to switch (DeLuca et al. 2010; Kelso et al. 1988; Lee
et al. 1996; Scholz and Kelso 1990). Perceptually available spatial
information can readily overcome apparent neuromuscular biases such as
preferences for coactivation of homologous muscles (e.g. Kelso and Jeka
1992; Kelso et al. 1991; Meschner et al. 2001; but see Salter et al. 2004) or
one muscle group over another (Kelso et al. 2001), at least up to a point.
In short, the parameters of cognitive coordination dynamics emerge from a
variety of sources that mutually constrain behaviour and behavioural choice.
These informational sources contribute to what we typically refer to under the
umbrella term, context. They include the perceptual requirements of the task,
information that arises during performance of the task, intention to perform a
particular behavioural pattern, allocation of attentional resources (Temprado
et al. 1999), memory of previous experiences and even ‘low-level’
neuromuscular–skeletal factors. All of these constraints have been connected
to the concept of stability of the coordination dynamics in one form or
another (including metastability, multistability and instability; see Kelso
2012a) and measurements thereof. Many of these constraints are prevalent in
the context of learning, a topic to which we now turn.
The coordination dynamics of skill learning
The organism, so goes the truism, is not a blank slate. Every individual enters
the learning situation with a history of pre-existing capacities. A key idea of
the coordination dynamics of the learning process is that learning involves a
modification of the learner's pre-existing capacities in the direction of the
skill to be learned. However, although other theorists also stress that learning
and development proceed in the context of pre-existing biases (e.g. Sporns
and Edelman 1993), ways to evaluate this pre-existing movement repertoire
prior to learning are lacking or, more usually, totally ignored in theories of
skill acquisition. Because discovering the nature of pre-existing capabilities is
so difficult, investigators have tried to use tasks that are as novel as possible
and hence they are completely unrelated to any existing coordination
tendencies that the learner might possess. Ironically, this strategy may
prevent us from understanding the features of the learned representation that
are shared across tasks and the level at which they are specified. In addition,
rather than studying a group of people performing a novel task and averaging
across them, coordination dynamics studies the individual learner, searching
instead for learning mechanisms that are common across individuals.
Starting in 1987, Pier-Giorgio Zanone and I developed methods to
overcome the difficult problem of evaluating the pre-existing capabilities of
the learner by scanning – before learning begins; throughout the learning
process; and after practice is over – the space of the coordination variable
proven to be valid for visuomotor coordination tasks, namely, the relative
phase between the interacting components (Haken et al. 1985; Kelso 1984;
Schmidt and Turvey 1995; Swinnen et al. 1997). This technique allows us to
set the learning task on an individual basis such that it does or does not
correspond to pre-existing coordination tendencies, which, whether innate or
acquired, may already exist in the individual learner's repertoire. According
to coordination dynamics, new task requirements may cooperate or compete
with pre-existing tendencies, thereby influencing the nature and rate of the
learning process. As the task is learned, the stability of the performed pattern
increases (indexed by shifts in the mean relative phase toward the learned
pattern, a sharpening of the distribution of phasing fluctuations and faster
relaxation times to the pattern). At the same time, the memorized relative
phase evolves on a slower timescale, biasing the performed pattern toward
the tobe- learned relative phase (for formal details, see Schöner and Kelso
1988; Schöner et al. 1992).
In our initial work (Zanone and Kelso 1992; see also Fontaine et al. 1997;
Kelso 1990, 1995 chapter 6; Magne and Kelso 2008; Zanone and Kostrubiec
2004), we showed not only that a new attractive state is established as
memorized information gains strength but also that the learning process may
take the form of a non-equilibrium phase transition or bifurcation:
stabilization of the learned pattern increased the number of coordination
patterns available to the learner and destabilized others (at least temporarily).
Learning, in other words, not only altered behaviour in the direction of the
to-be-learned pattern but also changed the entire layout of the coordination
dynamics (see Schöner et al. 1992, for the formal details). More recently, we
have provided evidence that learning may delay or even eliminate instabilities
along with neurophysiological correlates of such (Jantzen et al. 2001, 2002).
An important issue in theories of skill acquisition and learning concerns
the nature of what is learned. Coordination dynamics constitutes a lawful
representation that governs how the central nervous system assembles
coordinated patterns of activity on different levels of description. How
abstract this representation is may be ascertained by determining the
effectiveness of transfer or generalization from one (trained) effector system
to another (untrained) effector system. In previous work, we established that
transfer of timing relations occurs spontaneously between two components
within the same effector system (Zanone and Kelso 1994, 1997). We then
tested whether transfer of learning also occurs across different effector
systems, causing similar alterations in their respective coordination dynamics
(Kelso and Zanone 2002). The task was to learn a specific phase relationship
through practice with the arms or the legs. In order to assess modifications
induced by learning and transfer, the coordination dynamics of both effector
systems were evaluated before and after practice through scanning probes
aimed at revealing underlying attractive states of the coordination dynamics.
We predicted that transfer and generalization of learning should occur
across different effector systems to the extent that they share comparable
coordination dynamics. Thus, if this hypothesis is correct, both the trained
and the untrained effector combination should simultaneously exhibit
stabilization of the to-belearned phasing pattern. Moreover, other phasing
relations are predicted to be biased toward the to-be-learned pattern, if as our
theory predicts, the entire coordination dynamics is altered by the learning
process. Such a result is not typically examined in work from other traditions
because the full range of task-related coordination tendencies and how these
may change with learning is not explored. That is, traditional approaches
seldom measure how other timing relations, beyond the task to be learned,
are influenced by the learning process (for discussion, see Schmidt and Lee
1998 pp. 382–3).
Whether complete transfer occurs, of course, may depend on whether the
learning task is accomplished by components that are biomechanically
similar (e.g. the two arms or the two legs and their neural activation) or
different (e.g. an arm and a leg). The extended form of the HKB dynamics
(Kelso et al. 1990) contains a term (δω) that respects asymmetries, such as
those caused by biophysical differences between limbs (Jeka and Kelso 1995;
Kelso and Jeka 1992; Sternad et al. 1996) or differences between stimulus
and response components (Kelso et al. 1990; Wimmers et al. 1992), whereas
the original form (Haken et al. 1985) does not. Given the many differences of
neural and biomechanical origin between arms and legs, no one, of course,
expects perfect transfer. Nevertheless, evidence that the unpractised pattern is
learned and stabilized would bolster the view that coordination dynamics
constitutes a single abstract representation for an entire equivalence class of
coordinated actions, specifically those dealing with the relative timing
between coordinating components.
Results of the Kelso and Zanone (2002) study indicated that learning a
novel relative phase with one effector system spontaneously transferred to the
other untrained effector system. Not only was transfer seen as performance
improvements in both systems when the to-be-learned phasing was required
but it was also revealed by qualitative modifications of their underlying
coordination dynamics. That is, the dynamics of both the trained and
untrained limb pairs exhibited either comparable phase transitions,
themselves a signature of learning (Kelso 1990; Zanone and Kelso 1992) or
similar shifts in pre-existing attractive states, a further, parametric sign of
learning (Zanone and Kelso 1997). Irrespective of which form the learning
process actually took, the perceptually required phasing pattern was learned
and remembered, creating a new attractive state in both the practised and
unpractised coordination dynamics.
An important provision taken in the present approach is that the pattern
selected as a learning task does not coincide with already existing attractive
states (or preferences) of the underlying coordination dynamics.
Theoretically, before learning, any contribution to the coordination dynamics
owing to the novel task requirement should compete with any pre-existing,
so-called ‘intrinsic’ coordination tendencies. Such competitive interaction
between behavioural task demands and individual coordination tendencies
leads to observed biases and increases in variability of the performed pattern.
Our results strongly suggest that a common mechanism underlies learning
and transfer, namely reduction of the competition that initially arises between
task requirements and intrinsic coordination tendencies. How such
competition is instantiated in the central nervous system is an interesting
question. It is now well-established that different behavioural phasing
patterns have their expression in spatiotemporal patterns of brain activity,
quantified (using time-averaging techniques) in terms of spatial modes and
their time-dependent amplitudes (e.g. Fuchs et al. 1992; Fuchs et al. 2000;
Jirsa et al. 1998; Kelso et al. 1992, 1998) or (in the frequency domain) as
patterns of power and coherence, particularly in the beta (15–30 Hz) range
(Chen et al. 1999, 2003; Jantzen et al. 2001; Mayville et al. 2001). Still other
evidence using positron emission tomography has found that neural areas
such as the sensorimotor cortex are quite effector-specific, whereas activity in
parietal cortex remains high after transfer has occurred from fingers to arms
in a sequencing task (Grafton et al. 1998). Our findings that similar
alterations occur in the coordination dynamics of both the trained and
untrained effector systems confirm the hypothesis that learning and transfer
occur at a rather abstract level of system function. It may well be that parietal
areas are involved in generating action sequences at this abstract level, quite
independent of the effectors used. The latter statement is not meant to
minimize specific neuromuscular–skeletal factors that have been shown to
sculpt the coordination dynamics (Carson and Riek 1998; Kelso et al. 2001;
Kelso and Jeka 1992; Jeka and Kelso 1995).
How does the present work relate to more classical views of skill
acquisition and learning? In most, if not all previous views, the outcome of
learning is addressed in terms of abstract, task-specific entities such as
schemas, images of achievement and generalized programs (e.g. Bartlett
1932; Bernstein 1967; Schmidt 1975). For example, the aim of schema theory
was to explain how variable experiences with a skill allow the learner to
parameterize it in the form of a generalized motor programme. This
generative, rule-like feature of schema theory is intrinsic to even the most
elementary form of the coordination dynamics. The HKB (1985) equation,
for example, incorporates not only the so-called ‘invariance’ properties of the
generalized motor programme but also the crucially important dynamic
features of multistability, phase transitions and hysteresis. For
neurobehavioral dynamical systems (Kelso 1991, 2012a), the latter
correspond respectively to multi-functionality (different behavioural patterns
for the same parameter values), switching or ‘decision making’ (one
behavioural pattern is selected over another at critical parameter values) and a
primitive kind of memory (the history of system behaviour affects the current
state), respectively. Whereas data suggesting that the temporal structure of
movement is preserved across various kinds of parameterizations are used as
prima facie evidence for a generalized motor programme (as it was for the
earlier notion of coordinative structure, e.g. Kelso et al. 1979; Turvey et al.
1978), coordination dynamics rationalizes why this is so in terms of the
concept of stability. For example, in coordination dynamics loss of stability
provides a selection mechanism in the form of bifurcations or phase
transitions for the emergence of novel behavioural patterns. Fluctuations or
variability are not ‘errors’ or ‘noise’ in the output of the motor programme
but rather a fundamental way for the system to test its own stability under the
current circumstances. Thus, in coordination dynamics, fluctuations are an
essential part of the ‘decision-making’ mechanism determining whether the
system switches behaviour or not (e.g. Kelso et al. 1986). These theoretical
differences notwithstanding, a persistent issue in cognitive science has been
to define equivalence classes of processes in order to understand how two
different processes may be accomplished by the same higher-level
mechanism or ‘algorithm’ (Marr 1982). Viewed in the context of
coordination dynamics, this problem reduces to identifying the ensemble of
behaviours that share the same task- or function-specific coordination
dynamics (Kelso 2009a,b). By showing task-level transfer, the Kelso and
Zanone (2002) study provided an indication of just how abstract and
generalizable the coordination dynamics is and what form it takes. A more
recent summary of this entire research programme, including a specific
theoretical model that accommodates old as well as newer experimental
observations on learning, memory and attention may be found in Kostrubiec
et al. (2012).
Neural coordination dynamics of learning and attention
Coordination dynamics and its theoretical extensions, e.g. which model
observations of human brain and behavioural activity in terms of dynamic
neural fields (Fuchs et al. 2000; Jirsa et al. 1994, 1998; Jirsa and Haken
1996; Jirsa and Kelso 2000; Kelso et al. 1999) may mark the genuine arrival
of a new kinematics and dynamics for psychological states and cognitive
processes (Churchland 1988). Once the laws are known at the behavioural
level, it has proved possible to derive them from the excitatory and inhibitory
dynamics of neural populations and their long- and short-range connections.
The neural theory, in turn, poses a number of challenges to experiment, such
as how synaptic and cellular properties are influenced by learning, arousal
and attention (Kelso 2000). Let us pursue that challenge a bit further and
delve into what it entails.
It is well-known that spontaneous macroscopic reorganization of activity
occurs in both behavioural and brain dynamics (e.g. Chialvo 2010;
Daffertshofer et al. 2000; Frank et al. 2000; Fuchs et al. 1992; Jantzen et al.
2001; Jirsa et al. 1998; Kelso 2010; Kelso et al. 1991, 1992; Mayville et al.
1999, 2001; Meyer- Lindenberg et al. 2002; Plenz and Thiagarian 2007;
Wallenstein, Kelso and Bressler 1995). Such phase transitions are described
by the destabilization of a macroscopic activity pattern when neural
populations are forced to reorganize their spatiotemporal behaviour.
Destabilization is typically controlled via unspecific control parameters. In
contrast, traditional neuroscience describes reorganization of neural activity
as changes of spatial and timing relations among neural populations. Both
views are tied together by neural field theory; here, the spatiotemporal
evolution of neural activity is described by a nonlinear retarded integral
equation, which has a heterogeneous integral kernel. The latter describes the
connectivity of the neocortical sheet and incorporates both continuous
properties of the neural network as well as discrete long-distance projections
between neural populations. Mathematical analysis (Jirsa and Kelso 2000) of
such heterogeneously connected systems shows that local changes in
connectivity alter the timing relationships between neuron populations. These
changes enter the equations as a topological control parameter and can
destabilize neural activity patterns globally giving rise to macroscopic phase
transitions. Heterogeneous connectivity also addresses the stability–plasticity
dilemma: a stable transmission of directed activity flow may be achieved by
projecting directly from area A to area B and from there to area C (stability).
However, if necessary, area A may recruit neighbouring populations of
neurons (plasticity). See also Banerjee et al. (2012) for a postulated
recruitment mechanism for the case of bimanual interactions between the
limbs.
On the basis of our integral formulation of neural coordination dynamics,
we are currently addressing the following two aspects of brain functioning,
namely learning and attention/intention. Learning is approached via local
changes of synaptic weights resulting in a temporal dependence of the
connectivity function; i.e., the integral kernel. Two dynamics may be
distinguished: 1) the learning dynamics in which we employ established
methods usually based on unsupervised modified Hebbian learning; and 2)
the neural dynamics itself.
Reorganization of spatiotemporal neural activity can be controlled via
learning (Jirsa and Kelso 2000). Attention and intention are approached via
local changes in the sigmoidal response curves of neural ensembles, the so-
called conversion operations. These conversion operations have been
investigated in quantitative detail as a function of attention (e.g. Freeman
1975). The main result is that the slope and the height of the sigmoid vary by
a factor of 2.5 between minimal and maximal attention. The sigmoidal
variation of the ensemble response is realized biochemically by different
concentrations of neuromodulators such as dopamine and norepinephrine.
Mathematically, the neural dynamics described by the spatiotemporal integral
equations can be coupled to a one-dimensional concentration field in which
elevated values designate increased values of slope and height of the
sigmoidal response curve of neural ensembles. An increased slope and height
of the sigmoid typically causes increased amplitude and excitability of the
neural sheet.
The following line of thought allows an operationalization of the above
concepts, both experimentally and theoretically: a novel (or more difficult)
task requires more attention/arousal than an automated behavioral pattern
(Jantzen et al. 2001; Temprado et al. 1999). Increased attention demand is
realized in the cortical sheet via a larger concentration of neurotransmitter
substance resulting in a steeper slope and larger height of the sigmoid
response function. As a consequence, excitability and amplitude of neural
activity are increased in task conditions requiring more attention. During
learning, changes in connectivity occur. After learning, the task condition is
no longer novel but rather more ‘automated’, reflected in decreased
concentrations of neurotransmitters and thus decreased excitability and
amplitude of ongoing neural activity (see also Jantzen et al. 2001).
Implications and applications of coordination dynamics
Since this volume deals with applications of complex, self-organizing
dynamical systems to sport and since sport involves highly skilled activity, it
may be useful to summarize the ‘bottom line’ or message of coordination
dynamics in the context of skill learning. Although much remains to be
understood about the learning process and the brain mechanisms that underlie
it, the following summary statements are based on results obtained thus far
within the context of coordination dynamics. Several propositions are still
under detailed experimental scrutiny.
• The individual, not the group is the significant unit of analysis (this
statement contradicts practically all of the twentieth century
experimental psychology of learning).
• The individual is not a blank slate. Every individual enters a new
learning situation with an existing set of capabilities (‘repertoire’ or
‘signature’) unique to her/him.
• In order to understand the nature of learning, these pre-existing
capabilities need to be identified before the introduction of a novel
task/new material. The reason is that they influence the way new skills
are learned and remembered. As all good teachers know, ‘you have to
have hooks'.
• Learning, fundamentally, means the modification
(expansion/elaboration) of pre-existing capabilities. It is not (or not
only) a reinforcement/repetition/ associative process.
• New information (e.g. from task demands, the environment) either
cooperates or competes with the existing capabilities (the current
coordination dynamics) of the learner. Cooperative and competitive
mechanisms determine the rate of learning: fast for the former, slow for
the latter.
• Learning is not necessarily a smooth gradual process. Rather, depending
on the strength of cooperative and competitive mechanisms it may
involve smooth shifts in parameter space (the shift route) or highly
nonlinear, abrupt transitions (‘Eureka!’) – the bifurcation route.
• Remembering refers to the stability of learned states, a process that can
 be and has been precisely quantified. Notice, remembering is a process,
memory is a thing.
• When the human brain is learning a skill, activity in local neural
populations and the coordination among distant neural areas is
dramatically reorganized. Moreover, the individual brain, after it has
learned, is functioning far more economically than one that has not.
• The degree of brain plasticity (changes in the size and distribution of
active regions in the cerebral cortex following learning) is remarkable,
unexpected and warrants serious investigation in children and adults.
The above brief and incomplete list amounts to (at least a partial) re-
evaluation of the psychological and neuronal basis of learning. The focus on
individual preexisting capabilities as constraints on the learning process and
the need to structure the learning environment in light of them has significant
consequences for education and educational policy, as well as implications
for sports training, therapy and rehabilitation (e.g. in learning disabilities and
recovery of function following stroke or brain disease).
Summary and conclusions
Most theories in cognitive psychology and cognitive neuroscience tend to
view the mind in terms of fairly static representations. Symbolic
representation, by definition, is discrete and time independent. Yet many
perceptual and cognitive processes unfold in time. Indeed, in many processes
– such as perceiving, learning, remembering, forgetting, decision-making and
moving – time and timing are essential. Over the last few decades, a new
foundation for understanding spatiotemporally organized behaviour on
multiple levels has emerged, called coordination dynamics. Coordination
dynamics is both a theory and a research programme. In its mature form, it
contains two complementary conceptual themes. One concerns the
cooperative and competitive mechanisms that give rise to the spontaneous
formation of patterns and pattern change in complex cognitive and social
systems. The other deals with how information is created de novo in such
complex systems and how it modulates and is modulated by spontaneous,
pattern forming processes.
Dynamics is the language of understanding and transcends levels bridging
individual and collective processes. Dynamics, however, is not a panacea. In
every case, dynamical tools must be filled with conceptual content. It is not
enough to talk about the ‘dynamical systems approach’ to motor control,
development, cognition, and so forth, without the crucial concepts and
methods of self-organization, identifying meaningful pattern variables, and
informational specificity in the form of parameters of the coordination
dynamics. Although every system is different, what we learn about one may
aid in understanding another. Indeed, such an approach led to the recognition
of universality and mesoscopic protectorates in coordination dynamics.
Although the theoretical concepts of self-organizing dynamical systems
now enjoy some popularity in the social, behavioural, cognitive and brain
sciences, their usage is still quite restricted and still largely metaphorical –
though times, it must be said, are changing. One reason for the inertia was
that the tools are difficult to learn and require a degree of mathematical
sophistication. Their implementation in real systems is nontrivial, requiring a
different approach to experimentation and observation. Another reason is that
coordination dynamics (and the dynamical perspective in general) are often
cast (or cast themselves) in opposition to more conventional theoretical
approaches, instead of as an aid to understanding. On a personal note, the
author has been invited on many occasions over the years to write papers on
‘programmes’ versus ‘dynamics'. Selforganizing dynamics tends to
emphasize decentralization, collective decision making, spontaneous and
cooperative behaviour among many interacting elements. Conventional
cognitive psychology tends to focus on individual psychological processes
such as intention, perception, attention, memory, action, and so on, as if they
were clearly separable aspects of the goal-directed coordination of living
things. Yet, as evidence and theory now show, processes that we associate
with meaningful information such as intending, perceiving, attending,
deciding and remembering – as well as spontaneous self-organizing processes
– prove to be essential, coexisting aspects of the coordination dynamics of
cognition. Representation and dynamics are complementary: two sides of the
same coin.
Acknowledgements
Much of the research described herein was originally supported by NIMH
(Neurosciences Research Branch) grants MH42900, MH01386 and the
Human Frontier Sciences Program and, more recently, by grants from the
National Institute of Mental Health (MH080838), the National Science
Foundation (BCS0826897), the US Office of Naval Research
(N000140510117) and the Chaire d'Excellence Pierre de Fermat. I am
extremely grateful to all of these agencies for their support over many years
and to my colleagues and students for their tireless work and freely given
talents which are an inspiration.
References
Amazeen, P. G., Schmidt, R. C. and Turvey, M. T. (1995) Frequency
detuning of the phase entrainment dynamics of visually coupled rhythmic
movements. Biological Cybernetics, 72: 511–18.
Amazeen, E. L., Amazeen, P. G., Treffner, P. J. and Turvey, M. T. (1995)
Attention and handedness in bimanual coordination dynamics. Journal of
Experimental Psychology: Human Perception and Performance, 23: 1552–
60.
Banerjee, A., Tognoli, E., Kelso, J. A. S. and Jirsa, V. K. (2012)
Spatiotemporal reorganization of large-scale neural assemblies mediates
bimanual coordination. NeuroImage, 62 (3): 1582–92.
Bardy, B. G., Marin, L., Stoffregen, T. A. and Boutsma, R. J. (1999) Postural
coordination modes considered as emergent phenomena. Journal of
Experimental Psychology: Human Perception and Performance, 25: 1284–
96.
Bartlett, F. C. (1932) Remembering: A Study in Experimental and Social
Psychology. New York: Cambridge University Press.
Beek, P. J., Peper, C. E. and Stegeman, D. F. (1995) Dynamical models of
movement coordination. Human Movement Science, 14: 573–628.
Beekman, M., Sumpter, D. J. and Ratnieks, F. L. W. (2001) Phase transition
between disordered and ordered foraging in pharaohs’ ants. Proceedings of
the National Academy of Sciences of the United States of America, 98 (17):
9703–6.
Bernstein, N. (1967) The Coordination and Regulation of Movements.
Oxford: Pergamon.
Blekhman, I. I. (1988) Synchronization in Science and Technology. New
York: ASME Press.
Bressler, S. L. and Kelso, J. A. S. (2001) Cortical coordination dynamics and
cognition. Trends in Cognitive Sciences, 5: 26–36.
Buchanan, J. J. and Kelso, J. A. S. (1993) Posturally induced transitions in
rhythmic multijoint limb movements. Experimental Brain Research, 94 (1):
131–42.
Buchanan, J. J. and Kelso, J. A. S. (1997) To switch or not to switch:
recruitment of degrees of freedom stabilizes biological coordination, Journal
of Motor Behavior, 31 (2): 126–44.
Buchanan, J. J., Kelso, J. A. S., DeGuzman, G. C. and Ding, M. (1997) The
spontaneous recruitment and suppression of degrees of freedom in rhythmic
hand movements. Human Movement Science, 16: 1–32.
Byblow, W. D., Chua, R. and Goodman, D. (1995) Asymmetries in coupling
dynamics of perception and action, Journal of Motor Behavior, 27 (2): 123–
37.
Carson, R. G. and Riek, S. (1998) The influence of joint position on the
dynamics of perception-action coupling. Experimental Brain Research, 121:
103–14.
Carson, R. G., Goodman, D., Kelso, J. A. S. and Elliot, D. (1995) Phase
transitions and critical fluctuations in rhythmic coordination of ipsilateral
hand and foot. Journal of Motor Behavior, 27 (3): 211–24.
Carson, R. G., Chua, R., Byblow, W. D., Poon, P. and Smethurst, C. S.
(1999) Changes in posture alter the attentional demands of voluntary
movement. Proceedings of the Royal Society of London B Biological
Sciences, 266: 853–7.
Chen, Y., Ding, M. and Kelso, J. A. S. (1997) Long term memory processes
(1/f type) in human coordination. Physics Review Letters, 79: 4501–4.
Chen, Y., Ding, M. and Kelso, J. A. S. (1999) Alpha (10 Hz), Beta (20 Hz)
and Gamma (40 Hz) networks in the human brain and their functions in a
visuomotor coordination task revealed by MEG. Society for Neuroscience,
25: 1893.
Chen, Y., Ding, M. Z. and Kelso, J. A. S. (2001) Origins of human timing
errors. Journal of Motor Behavior, 33: 3–8.
Chen, Y., Ding, M. and Kelso, J. A. S. (2003) Task-related power and
coherence changes in neuromagnetic activity during visuomotor
coordination. Experimental Brain Research, 148: 105–16.
Chialvo, D. (2010) Emergent complex neural dynamics. Nature Physics, 6:
744–50.
Churchland, P. M. (1988) Matter and Consciousness. Cambridge, MA: MIT
Press.
Daffertshofer, A., Peper, C. E. and Beek, P. J. (2000) Power analysis of
event-related encephalographic signals. Physics Letters A, 266: 290–302.
DeLuca, C., Jantzen, K. J., Comani, S., Bertollo, M. and Kelso, J. A. S.
(2010) Striatal activity during intentional switching depends on pattern
stability. Journal of Neuroscience, 30 (9): 3167–74.
Ding, M., Chen, Y. and Kelso, J. A. S. (2002) Statistical analysis of timing
errors. Brain and Cognition, 48: 98–106.
Dijkstra, T. M. H., Schöner, G., Geise, M. A. and Gielen, C. C. A. M. (1994)
Frequencydependence of action-perception cycle for postural control in a
moving visual room: relative phase dynamics. Biological Cybernetics. 71:
489–501.
Duarte, R., Araujo, D., Freire, L., Folgado, H., Fernandes, O. and Davids, K.
(2012) Intraand inter-group coordination patterns reveal collective behaviours
of football players near the scoring zone. Human Movement Science, 31 (6):
1639–51.
Dumas, G., Nadel, J., Soussignan, R., Martinerie, J. and Garnero, L. (2010)
Inter-brain synchronization during social interaction. PloS One 5 (8): e12166;
doi:10.1371/journal.pone.0012166.
Edelman, G. M. (1987) Neural Darwinism. New York: Basic Books.
Fink, P., Kelso, J. A. S., Jirsa, V. K. and DeGuzman, G. C. (2000a)
Recruitment of degrees of freedom stabilizes coordination. Journal of
Experimental Psychology: Human Perception and Performance, 26: 671–92.
Fink, P. W., Kelso, J. A. S., Foo, P. and Jirsa, V. K. (2000b) Local and global
stabilization of coordination by sensory information. Experimental Brain
Research, 134: 9–20.
Fontaine, R. B., Lee, T. D. and Swinnen, S. P. (1997) Learning a new
bimanual coordination pattern: Reciprocal influences of intrinsic and to-be-
learned patterns. Canadian Journal of Experimental Psychology, 51 (1): 1–9.
Frank, T. D., Daffertshofer, A., Peper, C. E., Beek, P. J. and Haken, H.
(2000) Towards a comprehensive theory of brain activity: Coupled oscillator
systems under external forces. Physica D, 144: 62–86.
Freeman, W. J. (1975) Mass Action in the Nervous System. New York:
Academic Press.
Friston, K. (2000) The labile brain. III. Transients and spatio-temporal
receptive fields. Philosophical Transactions of the Royal Society of London
Series B, Biological Sciences, 355 (1394): 253–65.
Fuchs, A. and Jirsa, V. K. (2001) The HKB model revisited: how varying the
degree of symmetry controls dynamics. Human Movement Science, 19: 425–
49.
Fuchs, A. and Kelso, J. A. S. (1994) A theoretical note on models of
interlimb coordination. Journal of Experimental Psychology: Human
Perception and Performance, 20 (5): 1088–97.
Fuchs, A. and Kelso, J. A. S. (2009) Movement coordination, in R. A.
Meyers (ed.) Encyclopedia of Complexity and System Science, Heidelberg:
Springer.
Fuchs, A., Kelso, J. A. S. and Haken, H. (1992) Phase transitions in the
human brain: spatial mode dynamics. International Journal of Bifurcation
and Chaos, 2: 917–39.
Fuchs, A. Mayville, J., Cheyne, D., Weinberg, H., Deecke, L. and Kelso, J.
A. S. (2000a) Spatiotemporal analysis of neuromagnetic events underlying
the emergence of coordinative instabilities. NeuroImage, 12: 71–84.
Fuchs, A., Deecke, L. and Kelso, J. A. S. (2000b) Phase transitions in the
human brain revealed by large SQuID arrays. Physics Letters A, 266, 303–8.
Fuchs, A., Jirsa, V. K. and Kelso, J. A. S. (2000) Theory of the relation
between human brain activity (MEG) and hand movements. NeuroImage, 11:
359–69.
Grafton, S. T., Hazeltine, E. and Ivry, R. B. (1998) Abstract and effector-
specific representations of motor sequences identified with PET. Journal of
Neuroscience, 18: 9420–8.
Green, H. S. (2000) Information theory and Quantum Physics. Berlin:
Springer.
Haken, H. (1983) Synergetics, an Introduction: Non-equilibrium Phase
Transitions and Self-organization in Physics, Chemistry and Biology. Berlin:
Springer.
Haken, H. (1996) Principles of Brain Functioning. Berlin: Springer.
Haken, H., Kelso, J. A. S. and Bunz, H. (1985) A theoretical model of phase
transitions in human hand movements. Biological Cybernetics, 51: 347–56.
Jantzen, K. J., Fuchs, A. Mayville, J. M. and Kelso, J. A. S. (2001)
Neuromagnetic activity in alpha and beta bands reflects learning-induced
increases in coordinative stability. Clinical Neurophysiology, 112: 1685–97.
Jantzen, K. J., Steinberg, F. L. and Kelso, J. A. S. (2002) Practice-dependent
modulation of neural activity during human sensorimotor coordination: A
Functional Magnetic Resonance Imaging study. Neuroscience Letters, 332:
205–9.
Jantzen, K. J., Oullier, O. and Kelso, J. A. S. (2008) Neuroimaging
coordination dynamics in the sports sciences. Methods, 45: 325–35.
Jantzen, K. J., Steinberg, F. L. and Kelso, J. A. S. (2009) Coordination
dynamics of largescale neural circuitry underlying sensorimotor behavior.
Journal of Cognitive Neuroscience, 21: 2420–33.
Jeka, J. J. and Kelso, J. A. S. (1995) Manipulating symmetry in the
coordination dynamics of human movement. Journal of Experimental
Psychology Human Perception and Performance. 21 (2): 360–74.
Jeka, J. J. and Lackner, J. R. (1994) Fingertip contact influences human
postural control, Experimental Brain Research, 100 (3): 495–502.
Jirsa, V. K. and Haken, H. (1996) Field theory of electromagnetic brain
activity. Physical Review Letters, 77: 960–3.
Jirsa, V. K. and Haken H. (1997) A derivation of a macroscopic field theory
of the brain from the quasi-microscopic neural dynamics, Physica D
(Nonlinear Phenomena), 99: 503–26.
Jirsa, V. K. and Kelso, J. A. S. (2000) Spatiotemporal pattern formation in
neural systems with heterogeneous connection topologies. Physical Review
E, 62: 8462–5.
Jirsa, V. K., Fuchs, A. and Kelso, J. A. S. (1998) Neural field theory
connecting cortical and behavioral dynamics: bimanual coordination. Neural
Computation, 10: 2019–45.
Jirsa, V. K., Friedrich, R., Haken, H. and Kelso, J. A. S. (1994) A theoretical
model of phase transitions in the human brain. Biological Cybernetics, 71:
27–35.
Jirsa, V. K., Fink, P. W., Foo, P. and Kelso, J. A. S. (2000) Parametric
stabilization of biological coordination: A theoretical model. Journal of
Biological Physics, 26: 85–112.
Kello, C. T., Brown, G. D. A., Ferrer-i-Cancho, R., Holden, J., Linkenkaer-
Hansen, K., Rhodes, T. and Van Orden, G. C. (2010) Scaling laws in
cognitive sciences. Trends in Cognitive Sciences, 14: 223–32.
Kelso, J. A. S. (1981) On the oscillatory basis of movement. Bulletin of the
Psychonomic Society, 18: 63.
Kelso, J. A. S. (1984) Phase transitions and critical behavior in human
bimanual coordination. American Journal of Physiology: Regulatory,
Integrative and Comparative, 15: R1000–4.
Kelso, J. A. S. (1990) Phase transitions: Foundations of behaviour, in H.
Haken (ed.) Synergetics of Cognition. Berlin: Springer, pp. 249–68.
Kelso, J. A. S. (1991) Behavioral and neural pattern generation: the concept
of neurobehavioral dynamical system (NBDS), in H. P. Koepchen (ed.)
Cardiorespiratory and Motor Coordination. Berlin: Springer, pp. 224–38.
Kelso, J. A. S. (1992) Theoretical concepts and strategies for understanding
perceptualmotor skill: from information capacity in closed systems to self-
organization in open, nonequilibrium systems. Journal of Experimental
Psychology General, 121 (3): 260–1.
Kelso, J. A. S. (1994a) Elementary coordination dynamics, in S. Swinnen, H.
Heuer, J. Massion and P. Casaer (eds) Interlimb Coordination: Neural,
Dynamical and Cognitive Constraints. Academic Press: San Diego, pp. 301–
18.
Kelso, J. A. S. (1994b) The informational character of self-organized
coordination dynamics. Human Movement Science, 13: 393–413.
Kelso, J. A. S. (1995) Dynamic Patterns: The Self Organization of Brain and
Behavior. Cambridge: MIT Press.
Kelso, J. A. S. (1997) Relative timing in brain and behavior: Some
observations about the generalized motor program and self-organized
coordination dynamics. Human Movement Science, 16: 453–60.
Kelso, J. A. S. (2000) Principles of dynamic pattern formation and change for
a science of human behaviour, in L. R. Bergman, R. B. Cairns, L.-G. Nilsson,
L. and Nystedt, Developmental Science and the Holistic Approach. Mahwah,
NJ: Erlbaum, pp. 63–83.
Kelso. J.A.S. (2001) Metastable coordination dynamics of brain and
behavior. Brain and Neural Networks (Japan) 8: 125–30.
Kelso, J. A. S. (2002) The complementary nature of coordination dynamics:
self-organization and the origins of agency. Journal of Nonlinear Phenomena
in Complex Systems, 5: 364–71.
Kelso, J. A. S. (2008) An essay on understanding the mind. Ecological
Psychology, 20: 180–208.
Kelso, J. A. S. (2009a) Coordination dynamics, in R.A. Meyers (ed.)
Encyclopedia of Complexity and System Science. Heidelberg: Springer, pp.
1537–64.
Kelso, J. A. S. (2009b) Synergies: Atoms of brain and behavior. Advances in
Experimental Medicine and Biology, 629: 83–91.
Kelso, J. A. S. (2010). Instabilities and phase transitions in human brain and
behavior. Frontiers in Human Neuroscience, 4: 23.
doi:10.3389/fnhum.2010.00023
Kelso, J. A. S. (2012a) Multistability and metastability: Understanding
dynamic coordination in the brain. Philosophical Transactions of the Royal
Society B Biological Sciences, 367: 906–18.
Kelso, J. A. S. (2012b) Criticality reveals emergence (‘mesoscopic
protectorates’) in neural and behavioral systems, in Criticality in the Nervous
System, D. Plenz and E. Niebur (eds). Bethesda, MD: National Institutes of
Health.
Kelso, J.A.S. and Engstrom, D. A. (2006) The Complementary Nature.
Cambridge, MA: MIT Press.
Kelso J. A. S. and Jeka, J. J. (1992) Symmetry breaking dynamics of human
multilimb coordination. Journal of Experimental Psychology: Human
Perception and Performance, 18 (3): 645–68. Kelso, J. A. S. and Tognoli, E.
(2007) Toward a complementary neuroscience: metastable coordination
dynamics of the brain, in R. Kozma and L. Perlovsky (eds) Neurodynamics of
Cognition and Consciousness. Heidelberg: Springer, pp. 39–60.
Kelso, J. A. S. and Zanone, P. G. (2002) Coordination dynamics of learning
and generalization across different effector systems. Journal of Experimental
Psychology: Human Perception and Performance, 28: 776–97.
Kelso, J. A. S., Southard, D. and Goodman, D. (1979) On the nature of
human interlimb coordination. Science, 203: 1029–31.
Kelso, J. A. S., Scholz, J.P. and Schöner, G. (1986) Non-equilibrium phase
transitions in coordinated biological motion: Critical fluctuations. Physics
Letters A, 118: 279–84.
Kelso, J. A. S., Scholz, J. P. and Schöner, G. (1988) Dynamics governs
switching among patterns of coordination in biological movement. Physics
Letters A, 134: 8–12.
Kelso, J. A. S., DelColle, J. and Schöner, G. (1990) Action perception as a
pattern formation process, in M. Jeanerod (ed.) Attention and Performance
XIII. Hillsdale, NJ: Erlbaum, pp. 139–69.
Kelso, J. A. S., Buchanan, J. J. and Wallace, S. A. (1991) Order parameters
for the neural organization of single, multijoint limb movement patterns.
Experimental Brain Research, 85: 432–44.
Kelso, J. A. S., Bressler, S. L., Buchanan, S., DeGuzman, G. C., Ding, M.,
Fuchs, A. and Holroyd, T. (1992) A phase transition in human brain and
behavior. Physics Letters A, 169: 134–44.
Kelso, J. A. S., Buchanan, J. J., DeGuzman, G. C. and Ding, M. (1993)
Spontaneous recruitment and annihilation of degrees of freedom in biological
coordination. Physics Letters A, 179: 364–8.
Kelso, J. A. S., Fuchs, A., Holroyd, T., Lancaster, R., Cheyne, D. and
Weinberg, H. (1998) Dynamic cortical activity in the human brain reveals
motor equivalence. Nature, 392: 814–18.
Kelso, J. A. S., Fuchs, A. and Jirsa, V. K. (1999) Traversing scales of brain
and behavioral organization. I–III, in C. Uhl (ed.) Analysis of
Neurophysiological Brain Functioning. Berlin: Springer, pp. 73–125.
Kelso, J. A. S., Fink, P., DeLaplain, C. R. and Carson, R. G. (2001) Haptic
information stabilizes and destabilizes coordination dynamics. Proceedings
of the Royal Society of London B: biological Sciences, 268: 1207–13.
Kelso, J. A. S., DeGuzman, G. C., Reveley, C. and Tognoli, E. (2009) Virtual
partner interaction (VPI): exploring novel behaviors via coordination
dynamics. PLoS ONE, 4 (6): e5749; doi: 10.1371/journal.pone.0005749
Kostrubiec, V., Zanone, P.-G., Fuchs, A. and Kelso, J. A. S. (2012) Beyond
the blank slate: routes to learning new coordination patterns depend on the
intrinsic dynamics of the learner: experimental evidence and theoretical
model. Frontiers in Human Neuroscience, 6: 222.
Kovacs, A. J., Buchanan, J. J. and Shea, C. H. (2010) Impossible is nothing:
5:3 and 4:3 multi-frequency bimanual coordination. Experimental Brain
Research, 201: 249?59.
Lagarde, J. and Kelso, J. A. S. (2006) Binding of movement, sound and
touch: multimodal coordination dynamics. Experimental Brain Research,
173: 673–88.
Lagarde, J., Peham, C., Licke, T. and Kelso, J. A. S. (2005) Coordination
dynamics of the horse–rider system. Journal of Motor Behavior, 37: 419–24.
Laughlin, R. B. and Pines, D. (2000) The theory of everything. Proceedings
of the National Academy of Sciences of the USA, 97: 28–31.
Lee, T. D., Blandin, Y. and Proteau, L. (1996) Effects of task instruction and
oscillation frequency on bimanual coordination. Psychological Research, 59:
100–6.
Magne, C. and Kelso, J. A. S. (2008) A dynamical framework for human skill
learning. Advances in Psychology, 139: 189–203.
Marr, D. (1982) Vision. San Francisco: Freeman.
Mayville, J. M., Bressler, S. L., Fuchs, A. and Kelso, J. A. S. (1999)
Spatiotemporal reorganization of electrical activity in the human brain
associated with a phase transition in rhythmic auditory-motor coordination.
Experimental Brain Research, 127: 371–81.
Mayville, J. M., Fuchs, A., Ding, M., Cheyne, D., Deecke, L. and Kelso, J. A.
S. (2001) Event-related changes in neuromagnetic activity associated with
syncopation and synchronization tasks. Human Brain Mapping, 14: 65–80.
Mechsner, F., Kerzel, D., Knoblich, G. and Prinz, W. (2001) Perceptual basis
of bimanual coordination, Nature, 414: 69–73.
Meyer-Lindenberg, A., Zieman, U., Hajak, G., Cohen, L. and Faith Berman,
K. (2002) Transition between dynamical states of differing stability in the
human brain. Proceedings of the National Academy of Sciences of the USA,
99: 10948–53.
Naeem, M., Prasad, G., Watson, D. R. and Kelso, J. A. S. (2012)
Electrophysiological signatures of intentional social coordination in the 10–
12Hz range. NeuroImage 59: 1795–803.
Nicolis, G. and Prigogine, I. (1977) Self-organization in Nonequilibrium
Systems. New York: Wiley.
Oullier, O. and Kelso, J. A. S. (2009) Social coordination from the
perspective of coordination dynamics, in R. A. Meyers (ed.) Encyclopedia of
Complexity and Systems Science, Heidelberg: Springer, pp. 8198–212.
Oullier, O., DeGuzman, G. C., Jantzen, K. J., Lagarde, J. and Kelso, J. A. S.
(2008) Social coordination dynamics: measuring human bonding. Social
Neuroscience, 3: 178–92.
Passos, P., Araujo, D., Davids, K., Gouveia, L., Milho, J. and Serpa, S.
(2008) Informationgoverning dynamics of attacker–defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Peper, C. E., Beek, P. and van Wieringen P. C. (1995) Frequency-induced
phase transitions in bimanual tapping. Biological Cybernetics, 73 (4): 301–9.
Plenz, D. and Thiagarian, T. (2007) The organizing principles of neuronal
avalanche activity: cell assemblies and cortex. Trends in Neurosciences, 30:
101–10.
Riley, M. A., Richardson, M. C., Shockley, K. and Ramenzoni, V. C. (2011)
Interpersonal synergies. Frontiers in Psychology (Movement Science and
Sports Psychology), 2: 38; doi: 10.3389/fpsyg.2011.00038
Salter, J, Wishart, L. R., Lee, T. D. and Simon, D. (2004) The perceptual and
motor basis for bimanual coordination. Neuroscience Letters, 363: 102–7.
Schmidt, R. A. (1975) A schema theory of discrete motor skill learning.
Psychological Review, 82: 225–60.
Schmidt, R.A. and Lee, T. D. (1998) Motor Control and Learning: A
Behavioral Emphasis (3rd ed.). Champaign IL: Human Kinetics.
Schmidt, R. C., Carello, C. and Turvey, M. T. (1990) Phase transitions and
critical fluctuations in the visual coordination of rhythmic movements
between people. Journal of Experimental Psychology Human Perception and
Performance, 16 (2): 227–47.
Scholz, J. P. and Kelso, J. A. S. (1990) Intentional switching between
patterns of bimanual coordination depends on the intrinsic dynamics of the
patterns. Journal of Motor Behaviour, 22 (1): 98–124.
Schöner, G. and Kelso, J. A. S. (1988) Dynamic pattern generation in
behavioral and neural systems. Science, 239: 1513–20.
Schöner, G., Haken, H., Kelso, J. A. S. (1986) A stochastic theory of phase
transitions in human hand movement. Biological Cybernetics, 53: 442–52.
Schöner, G., Jiang, W. Y. and Kelso, J. A. S. (1990) A synergetic theory of
quadrupedal gaits and gait transitions. Journal of Theoretical Biology, 142:
359–91. Schöner, G., Zanone, P. G. and Kelso, J. A. S. (1992) Learning as
change of coordination dynamics: theory and experiment. Journal of Motor
Behavior, 24: 29–48.
Shannon, C. E. and Weaver, W. (1949) A Mathematical Model of
Communication. Urbana, IL: University of Illinois Press.
Sporns, O. and Edelman, G. M. (1993) Solving Bernstein's problem: a
proposal for the development of coordinated movement by selection. Child
Development, 64: 960–81.
Sternad, D., Collins, D. and Turvey, M. T. (1995) The detuning factor in the
dynamics of interlimb rhythmic coordination. Biological Cybernetics, 73:
27–35.
Sternad, D., Amazeen, E. L. and Turvey, M. T. (1996) Diffusive, synaptic
and synergetic coupling: an evaluation through in-phase and antiphase
rhythmic movements, Journal of Motor Behavior, 28 (3): 255–69.
Stins, J. F. and Michaels, C. F. (1999) Strategy differences in oscillatory
tracking: stimulus-hand versus stimulus–manipulandum coupling. Journal of
Experimental Psychology: Human Perception and Performance, 25: 1793–
812. Swinnen, S. P. (2002) Intermanual crosstalk: From behavioural models
to neural-network interactions. Nature Neuroscience, 3: 350–61. Swinnen, S.
P. and Wenderoth, N. (2004) Two hands, one brain: cognitive neuroscience
of bimanual skill. Trends in Cognitive Sciences, 8: 18–25.
Swinnen, S. P., Lee,T. D., Verschueren, S., Serrien, D. J. and Bogaerds, H.
(1997) Interlimb coordination: learning and transfer under different feedback
conditions. Human Movement Science, 16 (6): 749–85.
Temprado, J. J., Zanone, P. G., Monno, A. and Laurent, M. (1999)
Attentional load associated with performing and stabilizing preferred
bimanual patterns. Journal of Experimental Psychology: Human Perception
and Performance, 25: 1595–608.
Tognoli, E. and Kelso, J. A. S. (2009) Brain coordination dynamics: true and
false faces of phase synchrony and metastability. Progress in Neurobiology,
87: 31–40.
Tognoli, E., Lagarde, J., DeGuzman, G. C. and Kelso, J. A. S. (2007) The phi
complex as a neuromarker of human social coordination. Proceedings of the
National Academy of Sciences, 104: 8190–5 [from the cover; see also
Scientific American Mind, August, 2007].
Treffner, P. J. and Turvey, M. T. (1996) Symmetry, broken symmetry and
handedness in bimanual coordination dynamics. Experimental Brain
Research, 107 (3): 463–78.
Tuller, B., Ding, M. and Kelso, J. A. S. (1997) Fractal timing of phonemic
transforms. Perception, 26: 913–28.
Turvey, M. T. and Carello, C. (2012) On intelligence from first principles:
guidelines for inquiry into the hypothesis of physical intelligence (PI).
Ecological Psychology, 24: 3–32.
Turvey, M. T., Shaw, R. E. and Mace, W. M. (1978) Issues in the theory of
action: degrees of freedom, coordinative structures and coalitions, in J.
Requin (ed.) Attention and Performance VII. Hillsdale, NJ: Erlbaum, pp.
557–95.
Uhl, C., Friedrich, R. and Haken, H. (1995) Analysis of spatiotemporal
signals of complex systems. Physical review E, Statistical Physics, Plasmas,
Fluids and Related Interdisciplinary Topics, 51 (5): 3890–900.
Ullen, F., Ehrsson, H. H. and Forssberg, H. (2000) Brain areas activated
during bimanual tapping of different rhythmical patterns in humans. Society
for Neuroscience Abstracts, 26: 458.
van Orden, G. C., Holden, J. G. and Turvey, M. T. (2005) Human cognition
and 1/f scaling. Journal of Experimental Psychology General, 134 (1): 117–
23.
Wade, J. J., McDaid, L. J., Harkin, J. G., Crunelli, V. and Kelso, J. A. S.
(2011) Bidirectional coupling between astrocytes and neurons mediates
learning and dynamic coordination in the brain: a multiple modeling
approach. PLoS ONE, 6: e29445; doi:10.1371/journal.pone.0029445
Wallenstein, G. V., Kelso, J. A. S. and Bressler, S. L. (1995) Phase
transitions in spatiotemporal patterns of brain activity and behavior. Physica
D, 84, 626–34.
Wimmers, R. H., Beek, P. J. and van Wieringen, P. C. W. (1992) Phase
transitions in rhythmic tracking movements: a case of unilateral coupling.
Human Movement Science, 11: 217–26. Zanone, P. G. and Kelso, J. A. S.
(1992) The evolution of behavioral attractors with learning: Nonequilibrium
phase transitions. Journal of Experimental Psychology: Human Perception
and Performance, 18/2: 403–21.
Zanone, P.G. and Kelso, J. A. S. (1994) The coordination dynamics of
learning: theoretical structure and experimental agenda, in S. P. Swinnen, H.
Heuer, J. Massion and P. Casaer (eds) Interlimb Coordination: Neural,
Dynamical and Cognitive Constraints. San Diego. CA: Academic Press, pp.
461–90.
Zanone, P. G. and Kelso, J. A. S. (1997) The coordination dynamics of
learning and transfer: collective and component levels. Journal of
Experimental Psychology: Human Perception and Performance, 23: 1454–
80.
Zanone, P. G. and Kostrubiec, V. (2004) Searching for (dynamic) principles
of learning, in V. Jirsa and J. A. S. Kelso (eds) Coordination Dynamics:
Issues and Trends. Berlin: Springer, pp. 57–89.
1 The present chapter is a slightly revised and updated version of Kelso, J. A. S. (2003). Cognitive
coordination dynamics, in W. Tschacher and J. P. Dauwalder (eds) The Dynamical Systems
Approach to Cognition: Concepts and Empirical Paradigms Based on Self-Organization,
Embodiment and Coordination Dynamics. Singapore: World Scientific, pp. 45–71.
3 Why coordination dynamics is
relevant for studying sport
performance
Chris Button, Jon Wheat and Peter Lamb
The role of coordination dynamics in understanding movement behaviour in
sport is focused on identifying key parameters that describe system
organization (order parameters), as well as variables that act as information to
constrain the way that systems adapt to key environmental objects and events
(control parameters). Coordination dynamics has been used to study human
movement behaviour over the past three decades and more recently it has
been applied to the study of performance in team games. This chapter
reviews and summarizes current research in this area.
Sports performers coordinate their actions in many different ways to
achieve their goals. Coordination is a universal feature of sport common to
the graceful, precise actions of an ice dancer through to the explosive,
physical power of a shot putter. When watching elite sport performers, their
coordination patterns simultaneously entertain us and also remind us of the
incredible dexterity of the human body. Coordination dynamics offers a new
and potentially valuable lens through which our understanding of sports
performance can be developed. Indeed, movement scientists are only just
beginning to appreciate the complex, coalition of constraints that engulfs and
enables a sports performer. In this chapter, we explain what coordination
dynamics represents, with recourse to research that has applied a dynamical
systems approach to human movement behaviour. We then discuss why
coordination dynamics are relevant for sports performance in relation to
optimization, adaptation to constraints and rehabilitation from injury and
learning.
In writing this chapter, we conducted a critical review of relevant literature
before developing some practical implications based on this information.
Combining general search terms including ‘coordination dynamics', ‘sport
performance’ and ‘movement’ in several online search engines (ProQuest
Central [3,438], SportsDiscus [2,617] and Science Direct [2,582]) it was
possible to identify a relatively large body of literature. In fact, using the
same search terms, the open access search engine, Google Scholar, returned a
staggering over 25,000 research citations (admittedly not all quality-assured
outputs). A high percentage of this body of research appears to have been
published since 2000; for instance, in the case of ProQuest Central, this
proportion was 93%. Clearly, a significant amount of research using the term
‘coordination dynamics’ has been published in the new millennium,
reflecting the growing interest in this topic. Incidentally, ScienceDirect lists
the most-cited article statistics of this body of literature, identifying an
impressive 78 citations for the coordination dynamics review article of
Davids et al. (2003), with most other articles being cited less than 30 times.
To explain why coordination dynamics is relevant for studying movement
underlying sport performance, one must first describe the challenges that this
field of study has historically faced. In the field of skill acquisition, scientists
spent most of the twentieth century preoccupied with addressing theoretically
driven issues in laboratory controlled environments with applied sport
research very much on the fringes (Williams and Ford 2009). Simple, closed
tasks have proven to be popular vehicles through which to test specific
theoretical predictions. Admittedly, limitations in technology and restrictions
in relation to athlete tracking in competition have curtailed the extent to
which movement analysis could be conducted in naturalistic sports
environments.
Human movement scientists have been typically inclined to adopt a
reductionist (or positivist) philosophy in terms of analyzing sport
performance from the relative comfort of their laboratories. Notational
analysis of video footage is an excellent example of how the scientific
method has permeated sport performance research (Glazier 2010). It is likely
that the underlying assumption that sport performance can be optimized for a
given set of constraints perpetuates this traditional philosophy (Brisson and
Alain 1996; Glazier and Davids 2009). However, in setting out to identify the
best way to move, scientists have conveniently put to one side the more
fundamental issue of understanding how people move. Glazier and Davids
(2010) suggest that sports biomechanists working to optimize human
movement need to reconceptualize the athlete as a nonlinear, stochastic,
biological system.
Important issues such as these have meant that the subdisciplines of
biomechanics and motor control have historically rested in the shadows of
their more prominent, peer disciplines (i.e. physiology and psychology) when
it comes to supporting sports performance (Williams and Hodges 2005).
However, Button and Farrow (2012) point out that skill acquisition has made
considerable strides in recent times with notable developments in research
techniques, theory, practical implications and technology. Indeed, increasing
recognition of the importance of experimental design that is representative of
the context to which the results are to be generalized has demonstrated that
sport can provide an excellent vehicle for the progression of theory (Davids
et al. 2006; Pinder et al. 2011). As we aim to demonstrate in this chapter, the
advent of coordination dynamics may herald a new era in sports performance
research.
What is coordination dynamics?
Defined broadly as the science of coordination (Kelso 2009), coordination
dynamics can describe, explain and predict the patterns of coordination that
form, adapt, persist and dissolve in complex systems. Using nonlinear
structural equations to model behaviour, scientists have demonstrated how
seemingly complex, disordered systems adhere to simple, elegant principles.
Mutual information exchange between system components is required for the
system to organize itself. The field of coordination dynamics is founded upon
several important concepts which merit further discussion here, namely
synergies, self-organization, collective variables and control parameters, and
metastability.

Synergies
Synergies are functional groupings of components which are temporarily
assembled as a single unit. Synergies reflect nature's propensity for animate
objects to reduce their organizational complexity. For example, brain
synergies can be identified in the presence of a perturbation to one part of the
synergy (or network) and the subsequent reorganization of putatively linked
brain areas (Jantzen et al. 2008). Bernstein (1967) recognized the existence of
synergies in all forms of human movement, preferring the term ‘coordinative
structure’ in explaining how humans solve the degrees of freedom problem.
That being said, given the complexity of the movement system, how can the
human movement system arrive at a single solution for a task, given the
infinite possible solutions?

Self-organization
Complex systems have many independent parts which communicate in
different ways but primarily in terms of information exchange. Such systems
have a tendency to form patterned behaviour (synergies) which is not
prescribed externally nor indeed controlled by a central manager. Instead,
behaviours are said to spontaneously organize at a macroscopic level as a
result of the microscopic fluctuations of individual components. Whilst sports
teams are influenced to varying extents by the instructions of key figures
such as a coach or captain, the spatiotemporal trajectories of each player can
be thought of as a product of self-sorganization (Duarte et al. 2012a).

Collective variables and control parameters

Coordination dynamics is expressed through the interrelated behaviour of
collective variables and control parameters. Collective variables are high-
order, relational properties that capture or characterize the system at different
levels of analysis. As we shall discover in the next section, the relative phase
between two oscillating components (e.g. legs in gait) has been a popular
choice for a collective variable. Control parameters are naturally occurring
environmental conditions or intrinsic factors that can move the system
through its repertoire of patterns. For example, the gradient of a hill or an
intention to run can qualitatively change the value of the collective variable
(i.e. interlimb phasing), resulting in a phase transition to a new stable state.
Hence, control parameters act to influence the macroscopic (self-
)organization of a system which result from the collective variables that
characterize the microscopic levels. The reciprocity of this nested relationship
allows the system to settle in stable states, while still adapting functionally to
external factors when required.

Metastability
Complex systems are typically composed of multiple stable states.
Consequently, they exhibit periods of stability and instability, as they transit
between them in response to changing control parameter dynamics. A
particularly important property arises when pre-existing stable states dissolve
(i.e. bifurcations) to create remnants or ‘ghosts’ of these stable states. In a
metastable performance region, one or several movement patterns are weakly
stable (when there are multiple attractors) or weakly unstable (when there are
only attractor remnants) and switching between two or more movement
patterns occurs according to interacting constraints. A metastable system can
simultaneously realize a number of different competing patterns and thus has
the potential to exhibit novel and independent solutions (i.e. creativity) as
well as stable, coordinated behaviour. The neural dynamics of the brain
capitalize upon the metastability of the system, to flexibly reorganize
thoughts, memories and intentions on a moment-to-moment basis. This
allows appropriate online guidance of our actions.
 Particular impetus to the field of coordination dynamics has been provided
by studies of bimanual coordination in identifying the role of key constructs
of selforganization, collective variables and control parameters, as well as
transitions between stable states of neurobiological organization (see Kelso
1984; Schöner and Kelso 1988). The construction and adaptation of
movement patterns has been successfully modelled and investigated by
means of synergetic theoretical concepts since Haken, Kelso and Bunz
(HKB; 1985) applied them in investigations of brain and behaviour. In their
pioneering HKB model and its subsequent development (e.g. Schöner et al.
1986), abrupt changes in bimanual and multi - limb oscillatory movement
patterns (Jeka and Kelso 1995) were explained by a ‘loss of stability’
mechanism, which produced spontaneous phase transitions from less stable to
more stable states of motor organization with changes in critical control
parameters.
Together, these theoretical and empirical advances have provided a sound
rationale for a coordination dynamics-based explanation of how processes of
perception, cognition, decision making and action underpin intentional
movement behaviours in dynamic environments (e.g. van Orden et al. 2003).
This framework proposes that the most relevant information for decision
making and regulating action in dynamic environments is emergent during
performer–environment interactions.
Traditional investigations of limited-degree-freedom actions have provided
some useful models for understanding how control systems may operate
during neurobiological action. But they have shed fewer insights on
understanding how many biomechanical degrees of freedom are managed in
complex actions prevalent in dynamic performance environments common to
sport (Davids et al. 2006). Although many initial studies of coordination
dynamics tended to favour analysis of actions involving a limited number of
degrees of freedom, over multi-articular movement patterns (for a review of
that body of work see Davids et al. 1999), investigation of complex multi-
articular movements has proceeded rapidly in the last two decades (see for
example: Chen et al. 2005 on learning a pedalo task; Forner-Cordero et al.
2007 on postural control). Interesting issues in neurobiological coordination
and control concern the specific order parameter/collective variable dynamics
that have been studied in this body of work and how coordination training
shapes its manifestation over time.
How has coordination dynamics been analyzed in movement
science?
Coordination dynamics have been studied in the movement sciences at
several scales of analysis. These scales range from the study of intralimb
coordination to examine interactions between players in sports. Various
methods and techniques have been used to analyze coordination dynamics at
these different scales. In addition to providing a brief account of the details of
the methods and techniques in this section, pertinent issues related to their
application in different contexts are highlighted.

Cross-correlations
Similar to commonly used correlation techniques such as Pearson's
productmoment, cross-correlations assume that a linear relationship exists
between the two time series under analysis – for example hip and knee
flexion–extension angles during gait. However, unlike other correlation
techniques, crosscorrelations do not assume that the variables change in
synchrony during motion (Mullineaux et al. 2001). Rather, by time shifting
one relative to the other, the ‘time lag’ at which the correlation between two
time series is greatest can be identified. As such, in addition to identifying the
strength of the relationship or degree of linkage between the time series,
cross-correlation analyses reveal the type of relationship (the degree to which
the time series are in-phase or anti-phase). Indeed, if expressed relative to the
period of the motion, the time lag associated with the greatest correlation
coefficient indicates the phase relationship between the two segments
(Temprado et al. 1997) and is analogous to the measure of discrete relative
phase – discussed later in this section.
As Mullineaux et al. (2001) highlighted, cross-correlations have been
suggested as being particularly suited to the study of human movement, as
the coordinated actions of body segments and joints are often time shifted
relative to each other. However, several issues need to be considered before
conducting cross-correlation analyses in the study of coordination dynamics.
Firstly, crosscorrelations are not suited to the analysis of two-time series that
have a non-linear relationship (Sidaway et al. 1995). It is prudent that cross-
correlations are interpreted alongside more qualitative indications of the
relationship between the time series, such as variable–variable plots (known
as angle–angle plots if the time series concerns body segment/joint angles).
Secondly, as Pohl and Buckley (2008) highlighted in their study of foot and
shank motion during running, crosscorrelation techniques provide
information about only the temporal similarity of, but not the ratio of the
coupling between, two time series. In other words, crosscorrelation
techniques do not take account of excursion magnitudes (Pohl and Buckley
2008). Thirdly, there is disagreement regarding the recommended maximum
number of time lags that should be applied when estimating cross-
correlations. Mullineaux et al. (2001) highlighted that the probability of type-
I statistical errors inflates with an increasing number of lags. To reduce the
risk of a type-I error, they cited a recommendation that a maximum of plus or
minus seven lags be used. Alternatively, as a general rule, Derrick and
Thomas (2004) recommended a maximum time lag of n/2 (where n is the
number data points in the time series) but acknowledge that factors specific to
the time series under investigation should be considered when defining a
maximum number of time lags. A final point to consider before using cross-
correlations in the study of coordination dynamics is that they provide only
one, discrete, measure of coordination per movement cycle.

Vector coding
Angle–angle diagrams provide a convenient means for qualitatively
analyzing coordination dynamics. The shape of the angle–angle trace reveals
important information about the interaction and coupling between the body
joint/segment angles of interest. Several techniques have been developed to
provide a quantitative measure of the shape of the angle–angle trace; the
techniques are collectively referred to here as vector coding methods.
An early vector coding technique was presented by Freeman (1961). By
superimposing a grid on to the angle–angle curve and defining an eight-
element direction convention, a chain of integers (0–7) is established to
represent the shape of the curve. Although the integer chains capture the
shape of the trace and this technique has been used to study human
movement (e.g. Hershler and Milner 1980; Whiting and Zernicke 1982), a
limitation of the approach is that ratio data (joint/segment angles) are reduced
to the nominal scale (Tepevac and Field-Fote 2001). More recent vector
coding methods have addressed this issue. Hamill et al. (2000) reported a
modification of a technique presented by Sparrow et al. (1987), in which the
shape of the angle–angle trace is quantified by calculating a ‘coupling angle’;
the angle formed by the vector connecting two adjacent data points on an
angle–angle trace and the right horizontal.
The coupling angle provides information about the shape of the angle–
angle trace and the interaction/coupling between body segments. Figure 3.1
illustrates how coupling angles can be interpreted. Coupling angles of 0° and
180° indicate movement solely in the body segment/joint angle represented
on the x axis of the angle–angle diagram – where 0° indicates positive and
180° indicates negative, angular motion. Similarly, coupling angles of 90°
and 270° indicates movement solely in the body/segment angle represented
on the y axis of the angle–angle diagram (90° indicates positive and 270°
indicates negative, angular motion). Coupling angles of 45° and 225° indicate
that both body segment/joints are moving at the same rate in the same
direction (in-phase). Finally, coupling angles of 135° and 315° indicate that
the body segments/joints are moving at the same rate but in opposite
directions (anti-phase).

Figure 3.1 The interpretation of coupling angles; arrows represent the direction of a vector connecting
data points on an angle-angle diagram
 Coupling angles have been used to analyze coordination in a variety of
contexts. For example, Wilson et al. (2009) used vector coding methods to
determine the degree to which specific training drills represented lower-
extremity coordination patterns seen during triple jumping. Also, many
studies have used coupling angles to investigate coordination in gait (e.g.
Pohl and Buckley 2008; Ferber et al. 2005; Ong et al. 2011; Chang et al.
2008). For example, Ferber et al. (2005) investigated the effect of foot
orthotics on the coupling between the rearfoot and tibia during running. In
many studies, interpretations of coupling angles regarding coordination are
made directly using the approach outlined in Figure 3.1. However, it is rare
for coupling angles to be exactly equal to those highlighted in Figure 3.1,
making interpretation more complex. Recently, Chang et al. (2008)
introduced an approach to aid the interpretation of coordination, whereby the
range of possible coupling angles (0–360°) represented in Figure 3.1 is split
into four quadrants. The quadrant within which a particular coupling angle
lies is used to categorize coordination patterns and indicate the type of
coordination present (Table 3.1).
Table 3.1 Categories of coordination and their associated coupling angle (γ)
ranges; x-axis phase and y-axis phase denote the phases in which
movement is predominantly associated with the joint/segment
rotations represented on the on the x and y axis of the angle-angle
diagram, respectively
Type of coordination Coupling angle ranges

Anti-phase 112.5° ≤γ < 157.5°, 292.5° ≤γ < 337.5°

In-phase 22.5° ≤γ < 67.5°, 202.5° ≤γ < 247.5°

x-axis phase 0° ≤γ < 22.5°, 157.5° ≤γ < 202.5°, 337.5° ≤γ

< 360°

y-axis phase 67.5° ≤γ < 112.5°, 247.5° ≤γ < 292.5°

Source: adapted from Chang et al. (2008)

Vector coding techniques have the advantage that they are more intuitive and
easier to relate to than more complex techniques such as relative phase
(Field-Fote and Tepavac 2002). However, it is important to note that the most
commonly used vector coding methods involve calculating a coupling angle,
which is a circular variable, requiring directional statistics to be used when
further analysing the data (Batschelet 1981).

Relative phase
Coordination dynamics have also been investigated by calculating the relative
phase between oscillating system components. Both discrete relative phase
(DRP) and continuous relative phase (CRP) methods have been used. DRP
estimates the latency of the motion of one system component relative to
another (Kelso 1995). It is calculated by estimating the time difference
between an event common to both system components, relative to the period
of oscillation. For example, where the system components of interest are joint
angles, the common event might be the time at which the maximum joint
angles occur and the period would be the time to complete a joint rotation
cycle. DRP has been used to investigate the coordination between, for
example, respiration and stride rate during gait (O’Halloran et al. 2012),
pelvis and thorax rotations during treadmill walking (Lamoth et al. 2002) and
upper-arm segments during field hockey drives (Brétigny et al. 2011). DRP is
simple to calculate and, generally, does not require data normalization
(Hamill et al. 2000; Wheat and Glazier 2006; Krasovsky and Levin 2010).
However, similar to the coupling angle, DRP is a circular variable and should
be analyzed using directional statistics (Batshelet 1981). Finally, a
disadvantage of DRP is that it provides only one measurement of
coordination per movement cycle.
A continuous measure of relative phase has also been used in the study of
coordination dynamics: CRP. CRP indicates the phase relation between two
oscillating system components at each time point of a cycle of movement.
Based on phase plane plots – a plot of velocity against position – phase
angles are calculated by obtaining the arctangent of the ratio between velocity
and position. In other words, the angles between vectors connecting the
origin and each data point on the phase plane and the right horizontal identify
phase angles time series for each system component. The CRP between two
system components is then calculated as the difference between their phase
angles (0° represents in-phase and 180° represents anti-phase coupling).
 In addition to providing a continuous measure of coordination, as velocity
is included in the calculation of phase angles, CRP offers the potential for a
rich and more detailed analysis of coordination (Hamill et al. 1999). Many
studies have used CRP to study coordination dynamics. For example, Silfies
et al. (2009) used CRP to investigate differences in movement strategies
during reaching in participants with and without lower-back pain. Irwin and
Kerwin (2007) used CRP to identify effective skill progressions for
developing the long swing on the high bar in men's gymnastics. Also,
numerous studies have used CRP in the study of bimanual coordination and
nonlinear phase transitions originating from the seminal studies of Kelso et
al. (1981, 1984). In addition to studying intra- and intersegmental
coordination, CRP has also been used on a macro scale to investigate
interactions between players in sports such as squash (McGarry et al. 1999)
and tennis (Palut and Zanone 2005).
There are important factors to consider before CRP is used. First, position
and velocity data used in the calculation of phase angles can require
normalization and the resulting CRP values vary dependent on the
normalization used – see Peters et al. (2003) for more information. Related to
this, an assumption of CRP is that the time series used in the calculation are
sinusoidal. Data that violate the sinusoidal assumption can make
interpretation of CRP difficult (Peters et al. 2003). Alternative methods have
been developed to address this limitation, including those based on the
Hilbert transform (Rosenblum and Kurths 1998) and relative Fourier phase
(Lamoth et al. 2002).

Principal components analysis

Much of the coordination dynamics literature has focused on studying the
coordination between pairs of joints. When movements involving several
degrees of freedom are analyzed, multiple pairwise analyses of coordination
are performed. However, this approach can fail to grasp the global
perspective on the coordination task as a whole (Forner-Cordero et al. 2005).
The analysis of more complex coordination patterns, involving many degrees
of freedom, is becoming increasingly possible given advances in data
acquisition technology. Daffertshofer et al. (2004) suggested that multivariate
analysis techniques can be useful for analyzing these complex coordination
patterns. Principal components analysis (PCA) is one such technique that can
be used to reduce the dimensionality of high-dimensional multivariate
datasets. Based on the covariance between variables, PCA identifies mutual
information. In an n-dimensional space (where n is the number of variables in
a dataset), PCA first identifies the orientation of an axis that explains the
most variability in the data – the first principal axis. Subsequently, by
removing the variability in the dataset associated with the first principal axis,
the axis that explains the next greatest variability in the data is identified
–variability associated with the first principal axis is removed by ensuring
that the second principal axis is orthogonal to the first. This process
continues, with each succeeding principal axis accounting for as much of the
remaining variability in the dataset as possible. In other words, PCA converts
the input data variables into new, uncorrelated, principal components; where
principal components are the linear combinations of the input variables with
maximal variance (Forner- Cordero et al. 2005).
For many human movements, much of the variability in the dataset can be
explained by only a few principal components. For example, in a study of
lower body kinematics during walking, 90% of the variance in the data was
represented by the first two principal components (Lee et al. 2009). By
analyzing the relatively small number of principal components that explain
maximal variance, the dimensionality of the original input data set is reduced.
Daffertshofer et al. (2004) presented a tutorial which provided a detailed
account of PCA calculations and examples of how PCA can be applied to
study coordination. They highlighted how PCA can be used to study
coordination during gait; examples with kinematic and electromyographic
data are presented. Daffertshofer et al. (2004) also demonstrated how PCA
can be used as a ‘datadriven filter’ to help isolate deterministic and random
components of a dataset. There are many further examples of the use of PCA
to study coordination dynamics. For example, Forner-Cordero and colleagues
have used PCA to study the global coordination of the arms during complex
multi-joint coordinative movements (Forner-Cordero et al. 2005) and
examine the interaction between multi-joint coordinative movements and
postural control (Forner-Cordero et al. 2007). PCA has also been used to
study coordination during gait in, for example, people with stroke (Olney et
al. 1998) and lower back pain (Lamoth et al. 2006) and participants carrying
external loads (Lee et al. 2009).
PCA enables the analysis of global coordination in high-dimensional
systems. However, several issues related to PCA must be considered. Firstly,
PCA assumes that linear relationships between variables and the relationships
between biological signals can be nonlinear (Krasovsky and Levin 2010).
Secondly, as variations in the amplitude between variables in the input
dataset can have a large effect on the calculated principal components, the
issue of normalizing the input variables should be considered before
conducting PCA (Lamoth et al. 2006). The interpretation of PCA can also be
difficult. When the predominant mode of coordination is either inor anti-
phase, interpretation can be relatively straightforward (Forner-Cordero et al.
2005). For example, Forner-Cordero et al. (2005) prescribed experimental
motions that required either in- or anti-phase movements of the arm joints. In
this situation, the first principal component can be used to study overall
global coordination. However, when coordination is not predominantly in- or
anti-phase, interpretation/ classification of principal components becomes
more difficult. In their study of gait in participants with stroke, Olney et al.
(1998) related the first principal component to locomotion speed (this
accounted for 41% of the variance in the data), the second to interlimb
symmetry (accounting for 13% of the variance) and the third to postural
flexion bias. The remaining principal components accounted for minimal
variability and were not interpretable. Although interpretation/classification
of principal components is possible, the process is often based on prior
knowledge and expertise (Krasovsky and Levin 2010). Furthermore,
sometimes interpretation/ classification of principal components is not
possible.

Self-organizing maps
Self-organizing maps (SOMs) are a specific type of artificial neural network
most commonly used for dimensionality reduction and pattern classification
(Kohonen 2001). A SOM is commonly visualized as two layers of
information: an input layer and an output layer. The input layer consists of a
series of nodes, each of which represents an input vector (e.g. a single
multivariate time sample of a movement trial). The output layer consists of a
grid of nodes, which are each associated with a respective weight vector. The
weight vectors in the output layer compete to best represent nodes in the
input layer and, as a result, collectively model the input layer. Because of the
competitive learning algorithm, SOMs are able to model high-dimensional,
time-series data with a simple low-dimensional, visualizable mapping, while
maintaining the original topology of the input. Therefore, similar to PCA,
SOMs provide an opportunity to study complex coordinated movement.
Additionally, SOMs are also able to model nonlinear relationships in the
input – a prevalent feature of neurobiological systems. The nonlinear input-
output mapping is possible, owing to the competitive learning strategy and
the neighbourhood function (see Kohonen 2001), which together tend to
cluster similar data to similar map regions.
In the two-layer SOMs, the output layer represents postures or
coordination states during the movement. By connecting the sequence of
best-matching nodes for a single movement trial and superimposing the
trajectory on a mapping of the output layer, the time-series change in
coordination can be visualized. SOMs have been used to identify changes in
coordination for trials of gait (e.g. Barton et al. 2006; Lamb et al. 2011a) and
various sports actions, including discus throwing (Bauer and Schöllhorn
1997), football kicking (Lees and Barton 2005) and golf chipping (Lamb et
al. 2011b).
Several authors have employed a second SOM or a third layer, which is
trained on the output of the original SOM (e.g. Barton 1999; Lamb et al.
2011b). The sequence of the best-matching nodes is used as input for the
second SOM, which represents the movement pattern as a whole, rather than
the various states of coordination during the movement. Classifying the
movement as a whole can often be used to complement the findings of the
original SOM, especially when the coordination patterns underlying the
classification are of interest. In particular, and relevant to the coordination
dynamics framework, Lamb et al. (2011b) used the SOM trajectory of best-
matching nodes to study the coordination dynamics of the golf chip. Since the
SOM trajectory represents the evolution of coordination throughout the
movement, the authors used the trajectory as a collective variable, which they
then used to train a second SOM and subsequently to identify coordination
stability and transitions to new stable patterns.
SOMs represent a powerful tool for studying human movement,
particularly from a coordination dynamics perspective. Some issues that
remain contentious are whether to train one SOM on the data of several
subjects, at the risk of masking intra-individual changes in coordination, or to
introduce several SOMs unique to each subject, at the risk of generalizing
between subjects; how to determine the training parameters and which
normalization procedures are most appropriate. General solutions to these
issues are difficult and should instead be considered with respect to the
specific research question. For example, data normalization should be treated
as Hamill et al. (2000) outline for the continuous relative phase. Training
parameters should be data driven, the principal components of the input data
give an objective method for determining the number of nodes, the
dimensions of the map and the training length (both rough and fine tuning;
Vesanto et al. 2000). With continued use by researchers, SOM analyses
should become more familiar and their use more uniform across different
working groups.
Why coordination dynamics are relevant for sports
performance
Having introduced some of the key concepts underpinning coordination
dynamics and the various techniques used to examine these characteristics,
we finally consider what this field of study can offer for our understanding of
sport performance. Athletes in sports exemplify self-organizing, complex
systems, using specific information sources to coordinate their actions with
respect to important environmental objects, surfaces, and events (Turvey
1990). As they train, athletes educate their attention by becoming better at
detecting key information variables that specify movements from the myriad
variables that do not (see Chapter 4). In addition, learners calibrate actions by
tuning existing coordination patterns to critical information sources and,
through practice, establish and sustain functional information–movement
couplings to regulate coordinated activity with the environment (see Chapter
18). Coordination dynamics may be conceived as the manifestation of these
processes, whereby athletes match their intrinsic dynamics to the
requirements of a task (i.e. behavioural information).

Performance enhancement
With knowledge of the coordination dynamics of a system, it is theoretically
possible to model a task space and to determine potential solution spaces
(‘hot spots') within that area in which high levels of performance are most
likely to surface (Cohen and Sternad 2009). Furthermore, manipulation of
control parameters that move a system towards these hot spots would enable
sports practitioners to systematically improve an athlete's performance
(McGinnis and Newell 1982). In other words, understanding the stability
attributes of the athlete–environment system allows one to identify potential
strengths and weaknesses and, hence, these can be used to develop strategies
and tactics in an objective fashion (as opposed to relying on the intuition of a
coach). For instance, in team sports such as soccer and basketball, the
dynamics of the team centroid (including its position and stretch index) can
be used to predict and potentially influence critical moments in a game such
as shots or turnovers in possession (see Chapter 19).
A number of applied studies exemplify these practical suggestions. In
association football (soccer), Duarte et al. (2012b) demonstrate how the
stability of a defending group of players is upset by the collective movements
of attacking players as they converge toward the goal. Moreover, a series of
studies in futsal have recently shown how key parameters such as
interpersonal distance, phase angle between attacker and defender relative to
goal and the distance from goal influence the likelihood of success in
attacking scenarios (for an overview see Button et al. 2012). Switching to
individual sports, Barbosa et al. (2010) identified a number of critical factors
related to optimizing swimming performance. They noted that a swimmer's
segmental mechanics and centre of mass kinematics are strongly related to
energetics and ultimately to optimal performance. Cignetti et al. (2009)
examined how the coordination dynamics of cross-country skiing were
adapted under varying degrees of slope steepness (e.g. 0–7°). Common to
many other studies of bimanual coordination, a number of stable modes of
coordination were revealed and transitions between them were marked by
temporary losses in stability.

Technique modification, rehabilitation and injury prevention

During training, coordinative structures form a kind of ‘task-specific device’
suited for specific performance circumstances. Neurobiological systems ‘soft
assemble’ the coordination solutions into movement problems. These ideas
suggest that the inclusion of movement pattern variability is a useful strategy
for training of coordination and control processes in sport. Movement system
variability is an important part of learning design in sport and rehabilitation.
The abundance of mechanical degrees of freedom available in the degenerate
human movement system can be configured in different ways by athletes to
perform a diverse range of tasks. In appropriately designed training
environments, athletes can learn how to utilize different system degrees of
freedom to stabilize a functional coordination pattern. The implication is that
coordination training should not be aimed at reproduction of a putative
common optimal movement pattern by all learners but that each individual
performer needs to learn how to exploit variability within the movement
system in different ways to adapt to changing task constraints over time (see
Chapter 19).
The functional role of coordination variability in helping to prevent injury
and also rehabilitating from disease or injury has been recognized for over a
decade now (Hamill et al. 1999; Davids et al. 2003). It has been proposed
that movement variability is necessary to encourage injured and rehabilitating
individuals to adapt to the new, changing constraints imposed upon them, in
contrast to the traditional medical model, in which deviation from ideal
patterns is viewed as abnormal. For example, athletes recovering from
anterior cruciate ligament injuries show reduced centre of pressure variability
in comparison to healthy, control athletes (Davids et al. 1999). Hence, higher
variability in sway should not be associated with increased instability but
instead is indicative of normal exploratory behaviour.
The key question for practitioners is how to use this information in
diagnosing and treating injuries and/or disease. It is possible that knowledge
of the coordination dynamics of a system can be used to identify, monitor and
treat for the influence of pain amongst injured athletes. For example, van
Ryckeghem et al. (2012) propose that a multi-task switching paradigm can be
used to investigate how pain interferes with the ability of an individual to
carry out multiple tasks in complex environments. Neilsen and Cohen (2008)
remind us that corticospinal plasticity plays an important role in the
acquisition of athletic skills to a high standard. In particular, they suggest that
neurorehabilitation clinics should explore the use of emerging therapies that
simultaneously stimulate sensory inputs whilst also activating the motor
cortex, which may lead to enhanced cortical representations of motor skills,
as well as the excitability of the specific muscles engaged in those skills.
Finally, as Tessitore et al. (2011) demonstrate it is possible that monitoring
the coordination of athletes at various phases in their periodization cycle
(such as during preseason training in soccer) can lead to more effective,
targeted injury prevention strategies.
In summary, we have briefly explained some of the key concepts
underpinning the emerging field of coordination dynamics. This chapter
overviewed a number of studies using sports performance as vehicles to
reveal the coordination dynamics of athletes. Coordination dynamics exist
across multiple timescales and as such a range of analysis tools may be
required to uncover their characteristics. Sports scientists have begun to
reveal that coordination dynamics has the potential to unlock new
enhancements in performance in a range of different sports. Modifications to
athletic techniques can be brought about through coordination training as
performers develop metastable characteristics. Finally, coordination
dynamics may also play an important role in sports medicine, as clinicians
seek to prevent injury and rehabilitate athletes who must balance heavy
training demands with the need to optimize performance in competition.
References
Barbosa, T. M., Bragada, J. A., Reis, V. M., Marinho, D. A., Carvalho, C.
and Silva, A. J. (2010) Energetics and biomechanics as determining factors of
swimming performance: updating the state of the art. Journal of Science and
Medicine in Sport, 13: 262–9.
Barton, G. (1999) Interpretation of gait data using Kohonen neural networks.
Gait and Posture, 10: 85–6.
Barton, G., Lees, A., Lisboa, P. and Attfield, S. (2006) Visualisation of gait
data with Kohonen self-organising neural maps. Gait and Posture, 24: 46–53.
Batschelet, E. (1981) Circular Statistics in Biology. London: Academic Press.
Bauer, H. and Schollhorn, W. I. (1997) Self-organizing maps for the analysis
of complex movement patterns. Neural Processing Letters, 5: 193–7.
Bernstein, N. A. (1967) The Coordination and Regulation of Movements.
London: Pergamon Press.
Brétigny, P., Leroy, D., Button, C., Chollet, D. and Seifert L. (2011)
Coordination profiles of the expert field hockey drive according to field roles.
Sports Biomechanics, 10 (4): 339–50.
Brisson, T. A. and Alain, C. (1996) Should common optimal movement
patterns be identified as the criterion to be achieved? Journal of Motor
Behavior, 28: 211–23.
Button, C. and Farrow, D. (2012) Working in the field (Southern
Hemisphere), in N. Hodges and M. Williams. (eds) Skill Acquisition in Sport,
2nd edn. Abingdon: Routledge, pp. 367–80.
Button, C., Chow, J-Y., Travassos, B., Vilar, L., 32., Passos, P., Araújo, D.
and Davids, K. (2012) A nonlinear pedagogy for sports teams as social
neurobiological systems: how teams can harness self-organization tendencies,
in A. Ovens, T. Hopper and J. Butler. (eds) Complexity Thinking in Physical
Education: Reframing Curriculum, Pedagogy and Research. Abingdon:
Routledge, pp. 135–50.
Chang, R., Emmerik, R. Van and Hamill, J. (2008) Quantifying rearfoot-
forefoot coordination in human walking. Journal of Biomechanics, 41 (14):
3101–5.
Chen, H. H., Liu, Y. T., Mayer-Kress, G. and Newell, K. M. 2005. Learning
the pedalo locomotion task. Journal of Motor Behavior, 37: 247–56.
Cignetti, F., Schena, F., Zanone, P. G. and Rouard, A. (2009) Dynamics of
coordination in cross-country skiing. Human Movement Science, 28: 204–17.
Cohen, R. and Sternad, D. (2009) Variability in motor learning: relocating,
channeling and reducing noise. Experimental Brain Research, 193 (1): 69–
83.
Daffertshofer, A., Lamoth, J. C., Meijer, O. and Beek, P. J. (2004) PCA in
studying coordination and variability: A tutorial. Clinical Biomechanics, 19:
415–28.
Davids, K., Kingsbury, D., George, K., O’Connell, M., and Stock, D. (1999)
Interacting constraints and the emergence of postural behavior in ACL-
deficient subjects. Journal of Motor Behavior, 31: 358–66.
Davids, K., Glazier, P., Araújo, D. and Bartlett, R. M. (2003) Movement
systems as dynamical systems: the role of functional variability and its
implications for sports medicine. Sports Medicine, 33: 245–60.
Davids, K., Button, C., Araújo, D., Renshaw, I. and Hristovski, R. (2006)
Movement models from sports provide representative task constraints for
studying adaptive behavior in human movement systems. Adaptive Behavior,
14: 73–95.
Derrick, T. R. and Thomas, J. M. (2004) Time series analysis: the cross-
correlation function, in N. Stergious (ed.) Innovative Analyses of Human
Movement. Champaign, IL: Human Kinetics, pp. 189–205.
Duarte, R., Araújo, D., Correia, V. and Davids, K. (2012a) Sports teams as
superorganisms: implications of sociobiological models of behaviour for
research and practice in team sports performance analysis. Sports Medicine,
42 (8): 633–42.
Duarte, R., Araújo, D., Freire, L., Folgado, H., Fernandes, O. and Davids, K.
(2012b) Intraand inter-group coordination patterns reveal collective behaviors
of football players near the scoring zone. Human Movement Science, 31 (6):
1639–51.
Ferber, R., Davis, I. M. and Williams, D. S. (2005) Effect of foot orthotics on
rearfoot and tibia joint coupling patterns and variability. Journal of
Biomechanics, 38 (3): 477–83.
Field-Fote, E. C and Tepavac, D. (2002) Improved intralimb coordination in
people with incomplete spinal cord injury following training with body
weight support and electrical stimulation. Physical Therapy, 82: 707–15.
Forner-Cordero, A., Levin, O., Li, Y. and Swinnen, S. P. (2005) Principal
component analysis of complex multijoint coordinative movements.
Biological Cybernetics, 93 (1): 63–78.
Forner-Cordero, A., Levin, O., Li, Y. and Swinnen, S. P. (2007) Posture
control and complex arm coordination: analysis of multijoint coordinative
movements. Journal of Motor Behavior, 39: 215–26.
Freeman, H. (1961) On the encoding of arbitrary geometric configurations.
IEEE Transactions on Electronic Computers, EC-10: 321–31.
Glazier, P. S. (2010) Game, set and match? Substantive issues and future
directions in performance analysis. Sports Medicine, 40: 625–34.
Glazier, P. S. and Davids, K. (2009) Constraints on the complete
optimization of human motion. Sports Medicine, 39: 15–28.
Haken, H., Kelso, J. A. S. and Bunz, H. (1985) A theoretical model of phase
transitions in human hand movements. Biological Cybernetics, 51: 347–56.
Hamill, J., van Emmerik, R. E. A., Heidersheit, B. C. and Li, L. (1999) A
dynamical systems approach to lower extremity injury. Clinical
Biomechanics, 14: 297–308.
Hamill, J., Haddad, J. and McDermott, W. (2000) Issues in quantifying
variability from a dynamical systems perspective. Journal of Applied
Biomechanics, 16: 407–18.
Haykin, S. (1999) Neural Networks: A Comprehensive Foundation, 2nd edn.
Englewood Cliffs, NJ: Prentice-Hall. Hershler, C. and Milner, M. (1980)
Angle-angle diagrams in the assessment of locomotion. American Journal of
Physical Medicine, 59: 109–25.
Irwin, G. and Kerwin, D. G. (2007) Inter-segmental coordination in
progressions for the longswing on high bar. Sports Biomechanics, 6 (2): 131–
44.
Jantzen, K. J., Oullier, O. and Kelso, J. A. S. (2008) Neuroimaging
coordination dynamics in the sport sciences. Methods, 45: 325–35.
Jeka, J. J. and Kelso, J. A. (1995) Manipulating symmetry in the coordination
dynamics of human movement. Journal of Experimental Psychology Human
Perception and Performance, 21: 360–74.
Kelso, J. A. S. (1981) On the oscillatory basis of movement. Bulletin of the
Psychonomic Society, 18: 63.
Kelso, J. A. S. (1984) Phase transitions and critical behavior in human
bimanual coordination. American Journal of Physiology: Regulatory,
Integrative and Comparative Physiology, 15: R1000–4.
Kelso, J. A. S. (1995) Dynamic Patterns: The self-organisation of Brain and
Behaviour. Cambridge, MA: MIT Press.
Kelso, J. A. S. (2009) Coordination dynamics, in R. A. Meyers (ed.)
Encyclopedia of Complexity and Systems Science. New York: Springer, pp.
1537–64.
Kohonen, T. 2001. Self-organizing Maps, 3rd edn. Berlin: Springer.
Krasovsky, T. and Levin, M. F. (2010) Review: toward a better
understanding of coordination in healthy and poststroke gait.
Neurorehabilitation and Neural Repair, 24 (3): 213–24.
Lamb, P. F., Mundermann, A., Bartlett, R. M. and Robins, A. (2011a)
Visualizing changes in lower body coordination with different types of foot
orthoses using self-organizing maps (SOM). Gait and Posture, 34: 485–9.
Lamb, P. F., Bartlett, R. M. and Robins, A. (2011b) Artificial neural
networks for analyzing inter-limb coordination: the golf chip shot. Human
Movement Science, 30: 1129–43.
Lamoth, C. J. C., Beek, P. J. and Meijer, O. G. (2002) Pelvis-thorax
coordination in the transverse plane during gait. Gait and posture, 16 (2):
101–14.
Lamoth, C.J., Daffertshofer, A., Meijer, O. G. and Beek, P. J. (2006) How do
persons with chronic low back pain speed up and slow down? Trunk-pelvis
coordination and lumbar erector spinae activity during gait. Gait and Posture,
23 (2): 230–9.
Lee, M., Roan, M. and Smith, B. (2009) An application of principal
component analysis for lower body kinematics between loaded and unloaded
walking. Journal of Biomechanics, 42 (14): 2226–30.
Lees, A. and Barton, G. (2005) A characterisation of technique in the soccer
kick using a Kohonen neural network analysis, in T. Reilly, J. Cabri, and D.
Araújo (eds) Science and Football V: Proceedings of the Fifth World
Congress on Science and Football. London: Routledge, pp. 83–8. McGarry,
T. I. M., Khan, M. A. and Franks, I. A. N. M. (1999) On the presence and
absence of behavioral traits in sport: an example from championship squash
matchplay. Journal of Sports Sciences, 17: 297–311.
McGinnis, P. and Newell, K. (1982) Topological dynamics: a framework for
describing movement and its constraints. Human Movement Science, 1 (4):
289–305.
Mullineaux, D. R., Bartlett, R. M. and Bennett, S. (2001) Research design
and statistics in biomechanics and motor control: research design and
statistics in biomechanics and motor control. Journal of Sports Sciences, 19:
37–41.
Nielsen, J. B. and Cohen, L. G. (2008) The Olympic brain: does corticospinal
plasticity play a role in acquisition of skills required for high□performance
sports? Journal of Physiology, 586 (1): 65–70.
O’Halloran, J., Hamill, J., McDermott, W. J., Remelius, J. G. and Van
Emmerik, R. E. (2012) Locomotor-respiratory coupling patterns and oxygen
consumption during walking above and below preferred stride frequency.
European Journal of Applied Physiology, 112 (3): 929–40.
Olney, S. J., Griffin, M. P., and McBride, I. D. (1998) Multivariate
examination of data from gait analysis of persons with stroke. Physical
Therapy, 78 (8), 814–28.
Ong, A., Koh, M. and Hamill, J. (2011) Quantifying lower limb gait
coordination in of-theshelf orthotic shoes. Footwear Science, 3: 83–90.
Palut, Y. and Zanone, P.-G. (2005) A dynamical analysis of tennis: concepts
and data. Journal of Sports Sciences, 23 (10): 1021–32.
Peters, B. T., Haddad, J. M., Heiderscheit, B. C., Van Emmerik, R. E. and
Hamill J. (2003) Limitations in the use and interpretation of continuous
relative phase. Journal of Biomechanics, 36 (2): 271–4.
Pinder, R. A., Davids, K., Renshaw, I. and Araújo, D. (2011) Representative
learning design and functionality of research and practice in sport. Journal of
Sport and Exercise Psychology, 33: 146–55.
Pohl, M. B. and Buckley, J. G. (2008) Changes in foot and shank coupling
due to alterations in foot strike pattern during running. Clinical
Biomechanics, 23 (3): 334–41.
Rosenblum, M. and Kurths, J. (1998) Analysing synchronization phenomena
from bivariate data by means of the Hilbert transform, in H. Kantz, J. Kurths
and G. Mayer-Kress (eds) Nonlinear Analysis of Physiological Data. Berlin:
Springer, pp. 91–9.
Schöner, G. and Kelso, J. A. S. (1988) A dynamic pattern theory of
behavioral change. Journal of Theoretical Biology, 135: 501–24.
Schöner, G., Haken, H. and Kelso, J. A. S. (1986) A stochastic theory of
phase transitions in human hand movement. Biological Cybernetics, 53: 247–
57.
Sidaway, B., Heise, G. and Schonfelder-Zohdi, B. (1995) Quantifying the
variability of angle-angle plots. Journal of Human Movement Studies, 29:
181–97.
Silfies, S. P., Bhattacharya, A., Biely, S., Smith, S. S. and Giszter, S. (2009)
Trunk control during standing reach: a dynamical system analysis of
movement strategies in patients with mechanical low back pain. Gait and
Posture, 29 (3): 370–6.
Sparrow, W. A., Donovan, E., Van Emmerik, R. E. A. and Barry, E. B.
(1987) Using relative motion plots to measure changes in intra-limb and
inter-limb coordination. Journal of Motor Behavior, 19: 115–29.
Temprado, J., Della-Grasta, M., Farrell, M., and Laurent, M. (1997) A
novice-expert comparison of (intra-limb) coordination subserving the
volleyball serve. Human Movement Science, 16 (5): 653–76.
Tepavac, D. and Field-Fote, E. C., (2001) Vector coding : a technique for
quantification of intersegmental coupling in multicyclic behaviors. Journal of
Applied Biomechanics, 17: 259–70.
Tessitore, A., Perroni, F., Cortis, C., Meeusen, R., Lupo, C. and Capranica, L.
(2011) Coordination of soccer players during preseason training. Journal of
Strength and Conditioning Research, 25: 3059–69.
Turvey, M. T. (1990) Coordination. American Psychologist, 45 (8): 938–53.
van Orden, G. C., Holden, J. G. and Turvey, M. T. (2003) Self-organization
of cognitive performance. Journal of Experimental Psychology: General,
132: 331–50.
van Ryckeghem, D. M. L., Crombez, G., Eccleston, C., Liefooghe, B. and
Van Damme, S. (2012) The interruptive effect of pain in a multitask
environment: an experimental investigation. Journal of Pain, 13: 131–8.
Vesanto, J., Himberg, J., Alhoniemi, E. and Parkankangas, J. (2000) SOM
Toolbox for MATLAB 5, Technical Report No. A57). Espoo, Finland: Neural
Networks Centre, Helsinki University of Technology. Available from
www.cis.hut.fi/projects/somtoolbox/ (accessed 15 July 2013).
Wheat, J. S. and Glazier, P. (2005) Measuring coordination and coordination
variability, in K. Davids, S. Bennett and K.Newell (eds). Movements Systems
Variability. Champaign, IL: Human Kinetics, pp. 167–81.
Whiting, W. C. and Zernicke, R. F. (1982) Correlation of movement patterns
via pattern recognition. Journal of Motor Behavior, 14: 135–42.
Williams, A. M. and Ford, P. R. (2009) Promoting a skills-based agenda in
Olympic sports: the role of skill-acquisition specialists. Journal of Sports
Sciences, 27: 1381–92.
Williams, A. M. and Hodges, N. J. (2005) Practice, instruction and skill
acquisition in soccer: challenging tradition. Journal of Sports Sciences, 23:
637–50.
Wilson, C., Simpson, S. and Hamill, J. (2009) Movement coordination
patterns in triple jump training drills. Journal of Sports Sciences, 27 (3): 277–
82.
4 Psychobiological integration
during exercise
Natàlia Balagué Serre, Robert Hristovski,
Alfonsas Vainoras, Pablo Vázquez and
Daniel Aragonés
For over a century, physiologists have tried to find the limits of performance
following a reductionist approach. Although this approach has provided a
wealth of descriptive knowledge about acute and chronic changes of systemic
components with different types of exercise, it has failed to supply a unified
and clear explanation about the limits of exercise tolerance. Trying to find
such limits in specific sites or processes, the initial and major focus of
research has been the muscle and its metabolism, followed by the brain
(McKenna and Hargreaves 2008). Owing to controversial findings (Cairns
2006; Enoka and Duchateau 2008; McKenna and Hargreaves 2008; St Clair
Gibson and Noakes 2004; Nybo 2008; Weir et al. 2006), more recently, an
increasing attention has been paid to integrative approaches (Lambert et al.
2005), with renewed emphasis being placed on the role of the brain in
establishing the limits of exercise tolerance (Marcora and Staiano 2010,
Noakes, St Clair Gibson and Lambert 2005; Taylor et al. 2006).
The extant integrative models highlight the function of variables such as
perceived exertion in relation to the decision to terminate (Marcora 2008;
Shephard 2009; Weir et al. 2006). With regard to the way in which this
decision is made, there is now a lively debate about the respective roles of the
brain and the muscle (Marcora and Staiano 2010). However, two basic
questions need to be answered before engaging with this debate about the
integrative aspects of effort tolerance. Firstly, what kind of integration is
there between the different system components of the human organism: is it
linear or nonlinear? Secondly, and linked to the first question: is the
integration based on fixed, time-invariant and well-defined encapsulated
modules or is it task dependent, context sensitive and, therefore, flexible?
This chapter discusses these possibilities and proposes ways of resolving
these questions in light of recent empirical findings.
From linear to nonlinear models of psychobiological
integration
In exercise biology, what is known as the ‘systems approach’ still treats the
human organism as a machine or technical device and therefore its integrative
functions are studied within the framework of traditional control theory.
Concepts such as homeostasis and explicit feedback loops, controllers and
plants (i.e. controlled subsystems) are usually evoked to describe the ongoing
regulation and control of biological systems. This ‘engineering’ approach
aims to apply adequate corrective solutions in order to obtain the desired
stability and predictability of the subsystems under control. In doing so, the
notion of proportionality between inputs and outputs is usually used to
explain how the system adapts to internal (temperature, pH, gas
concentration, etc.) or external changes (workload, type of muscle
contraction, etc.; Kenney et al. 2011).
Descriptive block diagrams are commonly used to represent the way in
which organic structures and processes interact to achieve and regulate
different functions during exercise (such as voluntary movement; Figure 4.1).
The basic assumption of these diagrams (formed by control loops and
controllers) is that of time-invariant encapsulated modules, processes and
regulation profiles.
As long as one deals with conceptual (i.e. verbal) descriptive modelling,
this approach based on explicit feedback loops seems fine. Problems arise,
however, when one tries to model mathematically more than a couple of
interlinked components together (Kelso 1995). Then, the system rapidly
becomes impossible to treat in terms of explicit feedback circuits.
Figure 4.1 Feedback (A) and feed-forward (B) control of voluntary movement. A sign of correction is
created in both mechanisms to change the action of the muscles according to the difference between the
desired and achieved states

Another important problem is that, in contrast to what is claimed by

models based on control theory, complex biological dynamical systems do
not have simple and constant reference states with which feedback can be
compared and no single locus where comparison operations are performed
(see, for example, Thompson and Swanson 2010). Rather, steady states
emerge from the nonlinear interactions between the system's components but
without explicit and simple feedbackregulated set points or reference values,
as in, for example, an engineered thermostatic device (Fox et al. 2005;
Izhikevich et al. 2004). Linear models have also problems in predicting
certain observations, such as critical behaviour. In fact, they have serious
prediction problems because linearity and additivity are barely evident or
nonexistent in the human organism (Van Orden and Paap 1997).
 Two different models of exercise tolerance are currently being debated in
the exercise physiology literature: the ‘central governor’ model (St Clair
Gibson and Noakes 2004) and the psychobiological model of exercise
tolerance (Marcora 2008; Marcora et al. 2008, 2009) based on motivational
intensity theory (Wright 1996). The former uses the notion of a regulatory
device or ‘governor’ that is able to integrate algorithmically the large number
of changing variables during effort (Lambert et al. 2005; Noakes et al. 2005).
Through a kind of activation threshold (i.e. a set point), this central device
may produce cessation of activity in order to avoid systems failure (St Clair
Gibson and Noakes 2004). The efficiency of such a programmer or
regulatory unit is thought to be unconstrained, i.e. it functions as an
encapsulated module that is able to control at any given time any peripheral
change and it is unaffected by the fatigue process (as noted by Balagué and
Hristovski 2010). Later authors mention two problems in relation to this
psychobiological model: 1) an infinite regress problem, i.e. who programs the
programmer?; and 2) is it possible for the programmer to remain unaffected
by the changes occurring at all levels?
The psychobiological model of exercise tolerance does not consider the
need for a subconscious ‘entity’ such as the central governor, because the
decision as to when to terminate exercise is deemed to be made by the
conscious brain (Marcora 2008). The end of exercise therefore arrives when
exercise continuation is perceived as impossible. Perceived exertion,
considered as the awareness of central motor commands to the locomotor and
respiratory muscles, seems to be generated from efferent rather than afferent
sensory inputs and it would fix the limits of exercise tolerance. Thus, the
increase in central motor command required to exercise at the same workload
with muscles weakened by locomotor muscle fatigue is perceived as
increased effort.
Both models may find it difficult to explain the qualitative changes
occurring during exercise in varied contexts (such as the above mentioned
fatigue-induced termination point) unless they resort to specific ad hoc
explanations. In fact, neither of them specifically explains the onset of
exercise termination. Even if performers are conscious of exercise
termination, this does not necessarily mean that the termination is
consciously produced. Moreover, the processes that underpin this event are
not explained at all by saying that the termination is consciously produced.
 The fact that every new context (such as a different type of exercise)
requires a different mechanism or adaptation process leads, in general, to
fragmented knowledge (this being the current state of affairs in the majority
of sciences dealing with the biological order). This, in turn, brings about the
abovementioned controversial findings and fosters the idea that there may not
be general principles for the psychological and biological domains. At first
glance, these considerations seem to suggest that complex biological systems
are unable to satisfy the aims of general scientific theories. However, before
rejecting this possibility, it is worth considering that any macroscopic
behaviour of a complex adaptive system, such as performance in sport, is the
result of an immense number of highly coordinated spatiotemporal processes.
In other words, macroscopic behaviour is a collective effect of sets of highly
interdependent components within the system, i.e. a result of their synergy in
space and time. Research showed some time ago (Haken 1983) that these
collective (or cooperative) effects can be successfully studied by searching
for collective variables which capture the coherent, coordinated behaviour of
a system's component processes. These collective variables or order
parameters (since they represent the macroscopic state of biological order)
are the most adequate for studying the behaviour of complex systems because
they capture the approximately linear and also nonlinear regimen of operation
of such systems. These variables are best determined close to the points of
qualitative, discontinuous change, where a large set of other variables become
subservient to them and the behaviour of the system becomes low
dimensional. As they contain compressed information about all subservient
variables, the collective variables also become the most informative
quantities, i.e. to external observers (e.g. researchers) they are ‘informators’
about the macroscopic behaviour of complex systems (Haken 2000).
Testing the behaviour of such collective variables under the change of
constraints, i.e. applying a coordination dynamics approach (Kelso 1995),
would seem to be a viable way of investigating the type of integration shown
by a complex system. Complex adaptive systems may exhibit different kinds
of collective behaviour, such as stationary or nonstationary, i.e. metastable,
periodic or chaotic behaviour. The mode of behaviour depends basically on
the configuration of constraints or control parameters, i.e. on variables that do
not specifically prescribe or impose the behaviour of the system but which
constrain it. In short, the control of dynamical systems is constraints based.
For a certain configuration of constraints, nonlinear systems undergo a
qualitative change in their behaviour, a partial or complete rearrangement of
their component interactions and, hence, a discontinuous change in the order
parameter. These events are referred to as bifurcation phenomena. One reason
why these phenomena arise is because there is more than one possible stable
state, and this property, i.e. multistability, stems from the nonlinear
interactions between the system's components.
Exercise-induced fatigue experiments
In order to study the psychobiological integration during exercise performed
until failure from the perspective of coordination dynamics we performed a
set of experiments (Table 4.1). The accumulated effort was the control
parameter in all experiments.
The behaviour of four different potential collective variables providing
information about the state of the system as a whole during constant static
and dynamic exercises performed until the fatigue-induced spontaneous
termination point (FISTP) were continuously monitored and analyzed
(Hristovski and Balagué 2010). Two variables were kinematic (elbow angle
and pedalling frequency) and two psychological (attention focus and volition
state). Changes in these variables over time provide information about the
state of performer/task interactions. In complex systems, these states may
exist in different modes as pointed out before. These modes reveal the type of
interactions that occur between the different components and processes in the
system. Thus, they may help to get a clearer picture of the kind of system we
are dealing with, i.e. the type of integration between components in the
system.
Variables that capture the biological or psychological state of collective
order are classically studied either separately or through a hierarchy of
encapsulated general purpose modules, as illustrated by the recent debate
regarding ‘mind over muscle’ (Marcora and Staiano 2010). However, the
interaction between variables and, more importantly, their task dependence
are almost completely ignored when trying to separate their effect from the
general context.
As is well known, accumulated effort is accompanied by continuous
changes in both peripheral constraints (lactic acid accumulates, muscle
substrates change their concentration, etc.) and central constraints (rate of
perceived exertion increases, attention focus change, etc.). In our research,
we hypothesized that, for a certain configuration of constraints, nonlinear or
qualitative change in the behaviour might occur and, therefore, there would
be a discontinuous change in the order parameter (as described in the fourth
column of Table 4.1). As the termination point is understood here, as a result
of a bifurcation phenomenon (FISTP), it must be manifested as an abrupt
shift in the activity towards lower energy expenditure levels or rest
(Hristovski and Balagué 2010).
Table 4.1 Summary of the four experiments performed until the fatigue-
induced spontaneous termination point
Experiment Exercise Order parameter Nonlinear change

1. How fatigue Isometric arm- Elbow angle (0° From 90° to 0°

constrains quasi- curl flexion 90° (gravity
sometric exercise alignment)

2. How fatigue Volition state From metastable

constrains (UP/DOWN) to stable ‘down’
volition states state

3. How fatigue Cycling at 70 Cycling From 70 RPM to

constrains RPM frequency (RPM) close to 0 RPM
dynamic exercise

4. Attention focus Running while Attention focus From imposed

with effort having dissociative
accumulation dissociative thought to
thoughts associative

RPM = revolutions per minute

A second and equally important aspect of our research was that we sought
to find the dynamical hallmarks of these transitions, such as the enhancement
of fluctuations and change in the dimensionality of fluctuation dynamics as
the termination point approached. These effects would corroborate the
hypothesis that the termination point is a dynamical product, an effect of self-
organizing processes. Note that such effects are not at all predicted by extant
theories of fatigue. Conversely, they are a generic prediction of formal
theories of selforganizing systems (e.g. Haken 1983). Therefore, the finding
of these effects provided strong support for the hypothesis that the
termination point is a selforganizing dynamic event, which in turn poses a
serious challenge to theories which, assuming a linear integration, do not
predict these effects.

How fatigue constrains quasi-isometric exercise performance

On five alternating days over a period of two weeks, six well-trained
participants, who were familiar with the task, performed a quasi-isometric
arm-curl exercise holding an Olympic bar (weight: 80% of the one-repetition
maximum) with an initial elbow flexion of 90° until the FISTP. Participants
were encouraged to persist, even if the initial 90°-angle was lost finalizing
when the spontaneous disengagement from the task was produced. Changes
in both elbow angles during the trial were registered at a rate of 50 Hz by an
electrogoniometer (Biometrics, software by Ebiom; Figure 4.2).
As shown in Figure 4.3A (for details, see Hristovski and Balagué 2010) the
elbow angle starts fluctuating weakly and continuously around 90°, owing to
the initial fine adjustments. These adjustments are produced by the
intentionally sustained cooperation among the higher control loops
(presumably responsible for task specific perception, attention, motivation),
down to spinal reflexes and muscular processes. As fatigue develops, the
continuous changes occurring in the neuromuscular system progressively
destabilize the elbow angle, producing an increase in its variability. The
competition between the intention to sustain the task and the progressive loss
of neuromuscular tension is illustrated by the sudden increases in angle
values (above 90°) during the second third of the exercise. Finally, the
enhancement of elbow angle (i.e. order parameter) fluctuations precedes the
sudden reduction in the angle which coincides with the FISTP. It is
interesting to note that the participants' loss of ability to return to and remain
at the initial elbow-angle values after some accumulated effort may be
interpreted as a loss of stability of that initial attractor. The system was
unable to relax back to the previous attractor, which means that the relaxation
time became infinite for that state. The fact that performers were able to
sustain other smaller angles for some time suggests that the intentionally
sustained state of exertion may dwell in a metastable, rugged energy
landscape, trapping the system in transient basins of attraction far from the
final termination point.
Figure 4.2 Quasi-static arm-curl exercise holding an Olympic bar with an elbow flexion of 90° at 80%
of the one-repetition maximum until the fatigue-induced spontaneous termination point

An additional analysis using spectral degrees of freedom (Blackman and

Tuckey 1958; Vaillancourt and Newell 2003) revealed a highly significant
reduction in the degrees of freedom as the FISTP approached (Figure 4.4).
In other words, the dynamics within the system became increasingly low
dimensional, which points to the enhanced cooperative, i.e. mutually aligned,
as well as competitive behaviour of component processes within the
neuromuscular axis of performers. In the first third of the exercise, the
combination of weak fluctuations and the higher value of spectral degrees of
freedom signify a potential for more flexible control of the task goal. By
contrast, the combination of enhanced, bursting fluctuations (Figure 4.3A)
and the low value of spectral degrees of freedom signifies an increasingly
coherent and, therefore, more rigid control of the activity. Similar results are
routinely found at the level of the timing coordination dynamics of
electrocardiographic signals (see Hristovski et al. 2010) where an enhanced
coherence (lost of complexity) is emerging as the termination point
approaches.

Figure 4.3 A) Time series of the elbow-angle data for participant 1; B) A typical difference in the
power spectral density values for the online fluctuations of the elbow angle in the first (a) and the last
(b) third of the quasi-static exercise. The difference of the elbow angle variability between the first and
the third phase spans over sub-second and seconds time scale. This signifies a correlated instability of
the system under fatigue (Hristovski et al. 2010)
 At the behavioural level, these dynamics of order parameter variability
may be explained in terms of fatigue-induced dynamic competition between
two global processes at the neuromuscular level: the increasingly cooperative
protective inhibition and the goal-directed, intermittent bursting excitation,
the aim of which is to match more closely the task constraints, i.e. to keep
elbow flexion closer to the task-goal value of 90°. Under task constraints, the
increasingly cooperative protective inhibitory component processes, acting in
compliance with the pull of gravity, have to be counteracted by increasingly
cooperative excitatory component processes. Whereas in the initial phase of
exercise these processes compete over shorter time scales, resulting in a
stabilizing effect (small fluctuations) on the goal variable (elbow angle) as
exercise proceeds, they begin to compete over longer time scales, i.e. seconds
(Figure 4.3A), leading to larger fluctuations.

Figure 4.4 Differences in means for the spectral degrees of freedom; horizontal axis: the first and third
part of exercise; vertical axis: spectral degrees of freedom

These increasingly coherent yet competitive processes seem to be

responsible for the impending low dimensionality of order parameter
variability close to the termination point. Such processes closely resemble the
nucleation event of newphase formation in first-order phase transitions (e.g.
water phase and newly formed ice nuclei phase coexist close to the
transition). The first small nuclei of the new inhibition phase, i.e. small areas
within the neuromuscular system, may emerge simultaneously or with a short
lapse of time in several distant places of the system (some muscle metabolic
pathways, some synapses, etc.). The metabolic inhibition might be reflected,
for example, by the lower contractile ability of some muscle fibres and
provoke a larger inhibition effect. The increased GABA levels (Yakovlev
1979) in some central nervous system synapses can also enlarge the
inhibitory effects at this level. The accumulated effort enhances the growth of
this phase and it becomes increasingly macroscopic. On the other hand,
owing to taskgoal constraints the excitatory phase also segregates, becoming
more coherent in order to counteract the increasing inhibition. Thus, what
emerges is a formation of two macroscopic competing coalitions. However,
the increasing accumulated effort makes the excitatory coalition increasingly
unstable. Eventually, at the behavioural level, only the downwards movement
mode survives as a result of the cooperation between the macroscopic
neuromuscular inhibition and the gravitational pull under anatomical
constraints, stabilizing as the new global minimum of dynamics is reached
(alignment of the arm with gravity).
The power spectrum data (Figure 4.3B) show a globally correlated
enhancement of variability in the elbow angle. This enhanced variability was
simultaneously present on a sub-second scale to a tens-of-seconds scale,
indicating that all control loops along the neuromuscular axis were
destabilized by the accumulated effort. One can also see that the linear slope
of the spectrum differs, with the one derived from the third part of the effort
being steeper than the one derived from the first part. Together with the
results of the spectral degrees of freedom analysis, this points to the increased
rigidity of control under increasing accumulated effort. Hence, from the
coordination dynamics point of view, the exercise-induced fatigue represents
an ever-increasing destabilization of the previous configurations of the
psychobiological network and their continual reconfiguration under
immediate organismic, task and environmental constraints. In other words,
fatigue, seen dynamically, may be viewed as a typical example of constraints-
induced self-organization of metastable, soft-assembled configurations of
action system components, which eventually finds its global energetic
minimum aligning with the gravitational potential well.
 Note how the reductionist approach, i.e. focusing only on the muscle or
central processes and their isolated changes, would barely be able to give an
account of how exercise termination emerges. By contrast, even a simple
macroscopic approach within the framework of coordination dynamics is able
to integrate gross psychological and physiological processes into a single
language and capture the dynamic features of the impending exercise
termination. In other words, exercise termination seems to be a dynamical
event and it is increasingly clear that it should be studied and modelled as
such. In the future, more elaborate models may be developed, taking into
account well-defined control and dynamical variables.

How fatigue constrains volition states (quasi-static exercise)

A common general experience during a constant exercise (whether static or
dynamic) that is performed until the termination point is struggling with the
urge to cancel in the final moments. The aim of this experiment was to
investigate the dynamics of conscious states of volition during the same arm-
curl quasi static exercise described before.
On five alternating days (over a period of two weeks), six student
volunteers who were familiar with resistance training performed an isometric
arm-curl exercise – holding an Olympic bar (25 kg men, 17 kg women) with
an initial 90° elbow flexion. Motivational strategies were applied to ensure
that the participants continued exercising until their FISTP. During the effort
they were asked to verbalize their state of volition, simplifying its content to
‘up’ (continue) and ‘down’ (urge to cancel). Each one of the five trials was
divided into ten nonoverlapping windows and the probabilities of ‘up’ and
‘down’ volition states were calculated for each window. Probabilities of these
states were interpreted as signs of their relative attractiveness, i.e. stability.
The evolution of the probabilities of experienced and reported ‘up/down’
volition states across the trials showed the existence of three phases: the first
was dominated by an ‘up’ state, the second by a meta-stable ‘up-down’ state,
indicating competition between the two volition states, and the third was
dominated by a ‘down’ state (see Figure 4.5).
These results indicate that fatigue-induced conscious states of volition are
subject to nonlinear dynamical effects and give support to the hypothesis that
the state of will is a dynamical product of complex body-brain interactions.
 It is tempting to associate the increased probability of finding the ‘down’
state, as termination was approaching, with the increasing macroscopic
inhibition revealed in the first experiment. Indeed, it can be hypothesized that
the enhanced ‘down’ urge is a conscious manifestation of this growing
inhibition under the accumulated effort. Hence, the termination of effort
could be explained as a spontaneous dissolution, i.e. a loss of stability of the
intention to act in a certain way. From this perspective the role of the ‘up’
intentions is to keep the action stable. In this sense, every ‘up’ collaborates on
finding a new coordination to continue the task. The final ‘down’ urge
stabilizes as a consequence of the loss of stability of excitatory ‘up’ intention.
This leads to a spontaneous dissolution of the conscious intention that
emerges as a dynamical product and terminates the exertion. Note the
difference between the current proposals of extant psychobiological models
supporting that termination is a conscious decision (Marcora 2008) or
mediated by a central programmer (Lambert et al. 2005) and the claim of
nonlinear models that the termination is a spontaneous dissolution of the
conscious intention to act (Balagué et al. 2011; Hristovski and Balagué
2010). In this experiment, it was the ‘up’ intention as part of the task goal
which stabilized the order parameter (elbow angle), creating an attracting
basin around the goal value (elbow flexion under 90°) other than the resting
state. The dissolution of this attractor through a dynamic loss of stability
mechanism implied the dissolution of the intention to act. Hence, within this
framework the termination is not consciously produced but, rather,
spontaneously emerges at a psychological and action level as dissolution of
the intentional act.
Figure 4.5 Volition-state dynamics in six participants during the quasi-static elbow angle exercise

How fatigue constrains dynamic exercise performance

To test for the correlated properties of action variability during a continuous
dynamic task performed until exhaustion the following experiment was
conducted (Hristovski et al. 2010). Twelve triathletes performed a continuous
cycle ergometer exercise at 80% of their maximum workload, the task goal
being to maintain a pace of 70 revolutions per minute (RPM) until their
FISTP. The cycling frequency was treated as a potential collective variable
and its values were recorded continuously by a cycle ergometer system (Sport
Excalibur 925900; Figure 4.6).
Figure 4.6 Cycle ergometer exercise

The time series of the RPM variable were analyzed by time and frequency
domain methods (autocorrelation and spectral analysis). The spectral indexes
were calculated by estimating the linear fit slope of the power spectrum with
respect to frequency in logarithmic coordinates (Figure 4.7).
Figure 4.7 (Upper panel): standardized fluctuations from the data on revolutions per minute; (middle
panel): power spectrum for the first half of the exercise with its slope of −1.5 showing anti-persistent
fractional Brownian motion (fBm); (lower panel): power spectrum for the second half of the exercise
with a slope of −2.2 showing persistent fBm
 A scale-invariant relationship was found between the spectral power of
RPM variability and the frequency. The values of the spectral indexes were in
the intervals between –1 and –2.5, pointing to the presence of a fractal time
structure for RPM variability (from anti-persistent to persistent fractional
Brownian motion (fBm) as fatigue develops. The results of this experiment
corroborate the results presented in experiment one for a dynamic type of
exercise. The scale invariance in the power spectra suggests that there may be
no specific site and associated timescale of their dynamics that would
dominate the cycling frequency variability. In other words, the system seems
to be dominated not by the components but rather by interactions between
processes, e.g. control loops dwelling in different time scales.
The power spectra slopes showed a dominantly anti-persistent profile
(between –2 and –1) for the RPM variable in the first half of the exercise
(right upper panel). In the second half, there was a clearly persistent or super-
diffusive fBm profile (spectral slope between –2 and –3) in participants who
performed to exhaustion and whose time series at the end were dominated by
high-amplitude non-stationary fluctuations of the RPM variable (lower
panel). Performers who cancelled at accumulated effort values which did not
produce such a fluctuation profile attained fBm values of around –2 and less.
These results are consistent with those of the previously discussed quasi-
static exercise study from a different perspective. Anti-persistent fBm is
characterized by dynamics in which increments are anticorrelated, meaning
that the present trend is more likely to be followed by an opposite trend. This
characteristic points to a stabilizing synergy for constant continuous tasks,
such as maintaining cycling frequency at 70 RPM. On the other hand,
persistent fBm is characterized by increments which are positively correlated
in time; in other words, the present trend is more likely to be followed by the
same trend rather than the opposite. This tendency puts the system in a state
that is dominated by inappropriately low frequency variability at the expense
of high-frequency, shorter timescale corrections, as well as by the inability to
maintain the mean value constant around 70 RPM. Such a profile clearly
points to a system whose stability is disrupted. These findings show that
when highly motivated performers are close to the termination point the
fluctuation profile closely resembles the one discussed earlier and points to a
change in the neuromuscular cooperative processes prior to stop.
Attention focus during a dynamic accumulated effort
To investigate the emergent nature and dynamics of task-related thoughts
(TRT) during accumulated effort, 11 participants ran twice on a treadmill at
an intensity of 80% of their maximum heart rate until voluntary exhaustion,
while selfmonitoring and reporting through signs the changes in their
thoughts (Figure 4.8). During the first run, the intrinsic dynamics of their
thought processes was established. As no participant reported an emergence
of task-unrelated thoughts (TUT), only TRT, they were asked during the
second run to intentionally maintain TUT and to report back about
spontaneous switches from TUT to TRT and vice versa (for more details, see
Balagué et al. 2012). As can be seen in Figure 4.8B, the results revealed that
the intentionally imposed TUT was stable at the beginning of the exercise but
switched spontaneously to TRT with accumulated effort. Close to voluntary
exhaustion the TUT and TRT competed, showing a fully developed
metastability until the final TRT state prevails. In summary, a nonlinear
dynamic effect of thought processes (loss of stability of TUT, spontaneous
emergence of TRT, spontaneous switches from TUT to TRT (a metastable
dynamical regimen) and, finally, an absolute destabilization of TUT and
spontaneous transition to TRT) during the dynamic exercise was noted until
the termination of effort. This is a further demonstration that intentional
systems are subject to different constellations of peripheral and central
constraints (like attention focus) as exertion and fatigue accumulates.
Figure 4.8 A: Treadmill exercise with a participant reporting through signs; B: sample of 11 individual
time series of task-unrelated-task-related thought dynamics. Starting with the task-unrelated thoughts
(TUT) state, one can observe the switches between TUT and task-related thoughts (TRT). Eventually,
the TRT state becomes the one that precedes the exhaustion point. Numbers on the left signify
participants

This results illustrate that performers were not able to impose TUT
deliberately with equal efficiency during the exercise. Rather, the thought
states were constrained by the accumulated effort. The intention and attention
focus spontaneously self-organized into a different, more stable solution, i.e.
the TRT state of mind.

Phases of psychobiological integration

In summary, some general traits of nonlinear psychobiological integration
during exercise performed until exhaustion may be discerned from the
findings discussed above. These traits give raise to three effort phases:
• The initial part of the effort is characterized by greater flexibility and
stability of the psychobiological integration, as revealed through the
dynamics of the kinematic and psychological studied variables. Smaller
values of the elbow angle and the RPM fluctuations and their
dominantly anti-persistent profile, as well as the higher values of
spectral degrees of freedom are noticed on the kinematic level. The
ability of performers to maintain the intentionally imposed TUT, which
 are intrinsically unstable under high exertion rates, and the low
probability of finding urges to terminate the exercise illustrate the
stability and flexibility that are present on the psychological level.
• The second effort phase is characterized by relative stability at the
kinematic level, although the spontaneous, i.e. involuntary, emergence
of TRT (such as body monitoring) constitutes the first sign of
destabilizing effects of the accumulated effort. Within this interval there
is a metastable regimen characterized by switching from TRT to TUT
and vice versa, as well as balanced probabilities of urges to terminate
and volition to continue the exercise.
• The third phase is characterized on the kinematic level by a reduction in
the spectral degrees of freedom of the collective variable (elbow angle in
the quasi-static exercise) or by persistent or super-diffusive fBm (RPM
in the dynamic exercise) and enhanced fluctuations in both. This profile
signifies the formation of low-dimensional competition between two
increasingly coherent processes of inhibition and excitation, correlated
across the whole neuromuscular axis of performers. On a psychological
level these processes are associated with dominance of the urge to
terminate, the loss of stability of TUT and stabilization of TRT. The
stabilization of TRT points to the loss of flexibility and a lowering of the
dimensionality of attentional/thought processes, which corresponds to
the lowering of the spectral degrees of freedom on the kinematic level.
Eventually, the urge to cancel, being itself a TRT, becomes stabilized
close to and at the termination point. Taken together, all this suggests a
psychobiological integration that is not fixed, and only at some points
creates an association between psychological and biological spaces.
Especially when close to the termination point their mutual coupling
results in lower dimensionality and a dominant more rigid dynamics.
Implications for the future
As has been shown in the different experiments, the psychobiological
integration is highly likely to be nonlinear, soft-assembled and metastable. In
this context, exercise-induced effects and control may be explained through
the ‘selforganization under constraints’ paradigm. This generic mechanism
would enable biological systems, through their immense behavioural
flexibility and constant striving, to adapt to task and environmental demands.
The experimental results suggest that a viable way of investigating
psychobiological adaptation during exercise would be to study collective
variables, which are products of the cooperative, coordinated interactions
among component processes. As has been shown, these potential collective
variables may be observed at different levels of the human psychobiological
continuum. Thus, it would be especially important to study the ways in which
these coordinated dynamics are reconfigured on different time scales and also
to carry out more elaborate studies of key control parameters, i.e.
configurations of constraints that act upon the stability properties of
coordinated states.
In this regard, the findings described in this chapter present a challenge for
future research and might have important implications for cognitive and
physical interventions used to improve performance. Dynamical concepts
such as stability, metastability and loss of stability may prove to be important
in resolving the extant controversies concerning such interventions and could
help to identify suitable strategies for improving performance. While
intentions may change the peripheral states (e.g. pacing, etc.), the periphery
also seems to constrain the intentions and stability properties of the mind
(Balagué et al. 2011). Hence, the idea of circular causality, which captures
not only interaction but also interdependence, rather than a simple, linear top-
down cause–effect relationship between the mind and peripheral systems,
seems more plausible and provides further evidence of nonlinearity.
From the perspective developed here it would seem unfruitful to pose the
debate in terms of the ‘mind over muscle’ hypothesis (Marcora and Staiano
2010) or ‘muscle over mind’. As has been shown, there appears to be no
unique site or process that is responsible for exercise termination. Rather, this
event seems to be produced by the destabilization of interactions between a
number of components belonging to both central and peripheral subsystems.
This interaction is not fixed (as occurs when invariant set points are in
charge) but, rather, is task-specific, soft-assembled and therefore flexible,
thereby enabling adaptation to the different conditions created in the
organism and the environment during the development of fatigue.
A further point of note is that, since the system prior to termination dwells
close to the instability point, many contingent and also emergent accidental
events (a small increment of discomfort or pain, onset of nausea, dizziness,
and so forth), may sufficiently perturb the organization of the already
destabilized action system. This, in turn, could trigger exercise termination,
i.e. the switch toward the low-activity, resting state, a global minimum of the
rugged metastable energy landscape. In this sense, exercise termination is an
emergent phenomenon, a consequence of fatigue-induced
instability/dissolution of the couplings within the distributed control loops
that are responsible for the maintenance of the intended activity. This means
that the system flows through its dynamical states controlled by immediate
constraints and there is no need for specific peripheral site impairment or a
specialized exercise-termination module within the brain that would be fully
responsible for controlling and switching off the activity by issuing strict
commands to the periphery. Rather, it is sufficient for there to be a distributed
neuromuscular network of components that self-organize under constraints
into local, transient or, eventually, global energy minimum states. This
nonlinear, constraints-based control of exercise flow and termination is also
experimentally demonstrable in the hysteresis behaviour of the collective
variable with respect to certain physiological constraints (Balagué and
Hristovski 2010). Future research emphasizing the task-dependent dynamic
formation and dissolution of functional structures within physiological
variables may prove to be a viable way of capturing soft-assembling
coordination dynamics on that level. Integration of these findings with
psychological processes, such as those mentioned in this chapter, could lead
to more detailed, formal models of psychobiological integration during
exercise.

Questions for students

1. Identify and compare the main characteristics of the linear and
nonlinear psychobiological integration models that are mentioned
 in the text.
2. Name some of the variables included in each of the order
parameters studied in the different experiments.
3. Describe, using four different graphs, the nonlinear change
produced by the control parameter in the four different order
parameters as fatigue develops (Table 4.1).
4. Explain why there is a change in the slope of the spectral density of
elbow angle between the first and third part of the quasi-static
exercise (Figure 4.3B) and a change in the slope of the spectral
density of the RPM between the first and second half of the
dynamic exercise (Figure 4.5).
5. Describe and define using the same graph (with exertion time on
the horizontal axis), the three phases found in each of the four sets
of experimental results.
6. Explain the dynamics of order-parameter variability in the first two
experiments (static and dynamic exercise) in terms of
psychobiological inhibition/excitation processes. How gravitational
pull cooperates with inhibition?
7. Describe the differences between the currently debated models of
psychobiological integration and the nonlinear approach to explain
fatigue-induced exercise termination.
References
Balagué, N. and Hristovski, R. (2010) Modeling physiological complexity:
dynamic integration of the neuromuscular system during quasi-static exercise
performed until failure, in J. Wiemayer, A. Baca and M. Lames (eds)
Sportinformatik Gestern, Heute, Morgen: Festschrift zu Ehren von Prof. Dr
Jürgen Perl. Hamburg: Feldhaus, pp. 163–71.
Balagué, N., Hristovski, R. and Aragones, D. (2011) Rol de la intencion en la
terminacion del ejercicio inducida por la fatiga. Aproximacion no-lineal
[Role of intention in the fatigue induced exercise termination; nonlinear
approach]. Revista de Psicologia del Deporte, 20 (2): 505–21 [Spanish].
Balagué, N., Hristovski, R., Aragones, D. and Tenenbaum, G. (2012)
Nonlinear model of attention focus during accumulated effort. Psychology of
Sport and Exercise, 13: 591–7.
Blackman, R. B. and Tukey, J. W. (1958) The Measurement of Power
Spectra: From the Point of View of Communications Engineering. New
York: Dover.
Cairns, S. P. (2006) Lactic acid and exercise performance. Culprit or friend?
Sports Medicine, 36: 279–91.
Enoka, R. M. and Duchateau, J. (2008) Muscle fatigue: what, why and how it
influences muscle function. Journal of Physiology, 586: 11–23.
Fox, M. D., Snyder, A. Z., Vincent, J. L., Corbetta, M., Van Essen, D. C. and
Raichle, M. E. (2005) The human brain is intrinsically organized into
dynamic, anticorrelated functional networks. PNAS: Proceedings of the
National Academy of Sciences of the USA, 27: 9673–8.
Gandevia, S. C. (2001) Spinal and supraspinal factors in human muscle
fatigue. Physiological Reviews, 81: 1725–89.
Haken, H. (1983) Synergetics. An Introduction. Heidelberg: Springer.
Haken, H. (2000) Information and Self-organization. A Macroscopic
Approach to Complex Systems. Heidelberg: Springer.
Hristovski, R. and Balagué, N. (2010) Fatigue-induced spontaneous
termination point: nonequilibrium phase transitions and critical behavior in
quasi-isometric exertion. Human Movement Science, 29: 483–93.
Hristovski, R., Venskaityte, E., Vainoras, A., Balagué, N. and Vazquez, P.
(2010) Constraints controlled metastable dynamics of exercise-induced
psychobiological adaptation. Medicina, 46: 447–53.
Izhikevich, E., Gally, J. A. and Edelman, G. (2004) Spike-timing dynamics of
neuronal groups. Cerebral Cortex, 14: 933–44.
Kelso, J. A. S. (1995) Dynamic Patterns. The Self-organisation of Brain and
Behavior. Cambridge, MA: MIT Press.
Kenney, W. L. Wilmore, J. H. and Costill, D. L. (2011) Physiology of Sport
and Exercise (5th ed.). Champaign, IL: Human Kinetics.
Lambert, E. V., St Clair Gibson, A. and Noakes, T. D. (2005) Complex
systems model of fatigue: integrative homeostatic control of peripheral
physiological systems during exercise in humans. British Journal of Sports
Medicine, 39: 52–62.
McKenna, M. J. and Hargreaves, M. (2008) Resolving fatigue mechanisms
determining exercise performance: integrative physiology at its finest!
Journal of Applied Physiology, 104: 286–7.
Marcora, S. (2008) Do we really need a central governor to explain brain
regulation of exercise performance? European Journal of Applied
Physiology, 104: 929–31.
Marcora, S., Staiano, W. and Manning, V. (2009) Mental fatigue impairs
physical performance in humans. Journal of Applied Phisiology, 106: 857–
64.
Marcora, S. M. and Staiano, W. (2010) The limit to exercise tolerance in
humans: mind over muscle? European Journal of Applied Physiology, 109:
763–70.
Marcora, S. M., Bosio, A. and de Morree, H. M. (2008) Locomotor muscle
fatigue increases cardiorespiratory responses and reduces performance during
intense cycling exercise independently from metabolic stress. American
Journal of Physiology Regulatory Integrative and Comparative Physiology,
294: R874–83.
Noakes, T. D., St Clair Gibson, A. and Lambert, E. V. (2005) From
catastrophe to complexity: a novel model of integrative central neural
regulation of effort and fatigue during exercise in humans. British Journal of
Sports Medicine, 38: 511–14.
Nybo, L. (2008) Hyperthermia and fatigue. Journal of Applied Physiology,
104: 871–8. Shephard, R. J. (2009) Is it time to retire the ‘central governor’?
Sports Medicine, 39: 709–21.
St Clair Gibson, A. and Noakes, T. D. (2004) Evidence for complex system
integration and dynamic neural regulation of skeletal muscle recruitment
during exercise in humans. British Journal of Sports Medicine, 38, 797–806.
Taylor, J. L., Todd, G. and Gandevia, C. (2006) Evidence for a supraspinal
contribution to human muscle fatigue. Clinical and Experimental
Pharmacology and Physiology, 33: 400–5.
Thompson, R. H., and Swanson, L. W. (2010) Hypothesis-driven structural
connectivity analysis supports network over hierarchical model of brain
architecture. Proceedings of the National Academy of Sciences, 107: 15235–
9.
Vaillancourt, D. E. and Newell, K. M. (2003) Aging and the time and
frequency structure of force output variability. Journal of Applied
Physiology, 94: 903–12.
Van Orden, G. C. and Paap, K. R. (1997) Functional neuroimages fail to
discover pieces of mind in the parts of the brain. Philosophy of Science, 64:
85–94.
Weir, J. P., Beck, T. W., Cramer, J. T. and Housh, T. J. (2006) Is fatigue all
in your head? A critical review of the central governor model. British Journal
of Sports Medicine, 40: 573–86.
Westerblad, H., Allen, D. G. and Lannergren, D. J. (2002) Muscle fatigue:
lactic acid or inorganic phosphate the major cause? News in Physiological
Sciences, 17: 17–21.
Wright R. A. (1996) Brehm's theory of motivation as a model of effort and
cardiovascular response, in P. M. Gollwitzer and J. A. Bargh (eds) The
Psychology of Action: Linking Cognition and Motivation to Behavior. New
York: Guilford, pp. 424–53.
Yakovlev, N. N. (1979) Biochemistry of Sport. Moscow: Fiskultura i Sport
[Russian].
Part 2
Methodologies and techniques
for data analyses in
investigating complex systems
in sport
5 Nonlinear time series methods
for analyzing behavioural
sequences
Nikita Kuznetsov, Scott Bonnette and
Michael A. Riley
This chapter provides an overview of nonlinear analysis methods that
quantify the time-dependent characteristics of behavioural sequences. We
review the fundamental notions useful for an understanding of these analyses
and briefly summarize several frequently used methods. We then describe
sample entropy (SampEn) in more detail and provide a tutorial example of
calculating it. Finally, some relevant factors for the interpretation and design
of experiments employing sample entropy are discussed.
The analyses discussed in this chapter quantify the structure of variability
in time series. While the utility of using a time series approach is determined
by the motivating theoretical questions of a field of study, practically any
observable phenomenon can be recorded as a time series. Many disciplines
use time series to gain insights into their phenomena of interest. Stock price
fluctuations have for a long time puzzled economists (Scheinkman and
LeBaron 1989). In psychiatry and psychology, variations in mood over time
can provide useful clinical insights into mood disorders (Pincus 2006) and
time series methods have led to a conceptualization of self-perception as an
emergent property (Delignières et al. 2004). Similar analyses provide clues
about the structure of the cognitive system by quantifying response time
variability (Van Orden et al. 2003). Perceptual-motor control has also
benefited from the application of time series analyses (Riley and Turvey
2002). We believe that the methods described in this chapter (together with
Chapter 7) provide similar insights for sport science.
Dynamics and time series

Sequential dependence
In sport science, the time series we capture will usually have sequential
dependence – the order of the data points matters. This occurs when the
underlying process that generated the data is not random. The position of a
player on the pitch at one point in time is influenced by that player's position
at some earlier time or by the position of the ball, for example. These
dependencies can be strong or weak. Sometimes, when the dependency is
strong, we can discern a highly specific, deterministic rule that describes how
our measured quantity (the ‘output’ of the behavioural system under study)
varies over time as a function of some inputs and some parameters. This is
often not possible, however, and more often we can identify some general
properties about how the measurements change over time but not an exact
rule. In either case, though, pinpointing the nature of sequential dependence
in the data helps us to understand what kinds of laws or constraints shape the
behaviour of the system we are studying.
Figure 5.1 shows two artificial time series. Both have means of about zero
but they clearly differ visually. The bottom series is more structured, with
upward and downward trends, while the top is more erratic and
unpredictable. Typical measures of variability, like the standard deviation,
just measure the spread of observations around the mean, and do not measure
how the time sequence impacts the data. The series' means or standard
deviations would not differ.
The more interesting differences between the time series in Figure 5.1 lie
in the patterns of change over time (i.e. the dynamics of the behaviour).
These dynamics are a consequence of the sequential dependence (or lack
thereof) of the time series. If you randomly rearranged the order of data
points in a time series, the mean and variance would not change but doing
this would destroy any dependence among the data points and thus would
alter the dynamics. Time-series analyses quantify the dynamics of behaviour.
Using a method described later (detrended fluctuation analysis; DFA), it can
be shown that the two series evolve over time very differently. The bottom
series exhibits a type of sequential dependence termed 1/f scaling and the
data points are correlated with each other over time, whereas the top series
lacks this temporal structure and evolves over time randomly (it lacks
sequential dependence altogether).

Figure 5.1 Simulated time series to illustrate the distinction between measures of central tendency and
measures sensitive to the sequential properties of the time series. Panel A depicts random uncorrelated
noise, while Panel B shows ‘pink’ or 1/f noise where each successive observation is positively
correlated with the others

Nonlinearity
There are many kinds of sequential dependence. As noted, one distinction is
strength. A system with weak dependence is impacted more strongly by
random factors than one with stronger sequential dependence. Another basic
distinction is whether the sequential dependence is linear or nonlinear. This is
a fundamentally important distinction.1
Nonlinear is a term used to describe a type of physical system or
mathematical equation for which the ‘output’ is not directly proportional to
the ‘input’. Any system, whether it is the movement of an entire football
(soccer) team or repeated finger-tapping movements, is assumed to have
components or mechanisms responsible for the process (see Carello and
Moreno 2005). Linear and nonlinear analyses assume that the components of
a system interact in fundamentally different ways. For a linear system, the
components interact additively, so their behaviour adds up to the system's
behaviour – the whole is the sum of its parts. For nonlinear systems, the
components interact multiplicatively rather than just additively, which is why
the output is disproportional to the input and the whole can differ from the
sum of its parts. Nonlinearity is more general than linearity. A practical
implication of this is that nonlinear time-series analyses can work if the
system under study is linear but linear time series analyses may not work for
studying nonlinear systems. This is a reason to strongly prefer nonlinear time
series analyses over linear. (This does not imply, however, that linear
methods are not useful or that they are never preferred. Many linear methods,
such as spectral analysis, have well-developed mathematical foundations that
are not yet available for some nonlinear methods, for example.)

Dynamical systems
Nonlinear time-series analyses stem from a field of mathematics known as
nonlinear dynamical systems. For good introductions to the topic see Kaplan
and Glass (1995) and Kantz and Schreiber (2004). Dynamical systems are
simply those systems whose states change over time. To better appreciate
nonlinear time-series methods, it is useful to understand a few basic concepts
of dynamical systems.
The mathematical study of dynamical systems usually employs well-
known equations rather than empirical data. An example is the Lorenz
system, which is used to model processes of heat transfer known as thermal
convection, given by:

The Lorenz system has three state variables (x, y and z) and three parameters
(σ, ρ and β). The values of the state variables change over time according to
equation 1. The dot over the variables on the left side of the equal sign
signifies change over time in the variable (i.e. the derivative of the variable)
and the terms on the right-hand side specify the exact rule according to which
this change occurs. Although we have written three separate equations (one
for each state variable), the Lorenz system is considered to be a unified
system because the state variables are coupled to each other – the equation
for each state variable contains at least one of the other state variables,
meaning that the evolution of each state variable depends on that of the
others. We can simulate these equations and plot time series of each state
variable (Figure 5.2) but none of those individual time series fully
characterizes the total behaviour of the system. The proper way to visualize
the system is a plot of the phase space – a three-dimensional (in this case)
space whose coordinates are the state variables x, y and z.
A graph of the phase space (Figure 5.2, left) shows that the Lorenz system
gravitates toward certain regions and never enters others; the trajectory is
drawn to a subset of the phase space. This subset is called an attractor and it
is the solution to the underlying equations. The attractor is termed such
because whatever the initial values of the state variables, the trajectory will
eventually be drawn to the attractor. The Lorenz attractor is called a ‘strange’
attractor because the Lorenz system exhibits deterministic chaos – a small
change in initial conditions or a small perturbation will become amplified so
that the long-term behaviour of the system is unpredictable, even though
there is no element of actual randomness (the system is completely
deterministic but unstable). Trajectories in the phase space never actually
cross each other, because of a mathematical rule known as the uniqueness
theorem (differential equations must have unique solutions; if the trajectories
crossed, it would mean two solutions existed at once).
How does this relate to using nonlinear time series analyses on behavioural
sequences? Usually, we do not know the underlying equations that govern the
behavioural sequences that we analyze. Nor do we usually even know what
the state variables are, or what the attractor may look like. Presumably, the
variables that we can measure are relevant state variables but, even then, we
are left with a one-dimensional view of a potentially high-dimensional
system. This onedimensional signal may contain distortions that result from
projecting the data from a higher-dimensional space to the single dimension
of the measured variable (just like a two-dimensional map of the three-
dimensional Earth contains distortions in the shape and size of land masses).
However, thanks to another mathematical theorem called the embedding
theorem (Takens 1981; also see Webber and Zbilut 2005), it turns out that
measuring a single-state variable (i.e. a single time series) is sufficient to
allow us to understand the underlying dynamics of the system. Essentially,
this theorem permits us to reconstruct a phase space that preserves the
dynamical properties of the system and that is free of projection distortions,
provided that the underlying system is nonlinear (a safe assumption in
practice) and the state variables are coupled. Phase-space reconstruction
(Abarbanel 1996) is a first step for several nonlinear methods, such as
recurrence quantification analysis (RQA). Webber and Zbilut (2005) and
Pellecchia and Shockley (2005) discuss phase-space reconstruction
extensively in the context of RQA, so we do not go into more detail here.

Figure 5.2 The phase space plot of the Lorenz attractor; we used the parameters ρ = 10, σ = 28 and β =
8/3 and initial conditions of x = 1, y = 1, z = 1

Analyzing real-world data is very different than analyzing idealized

mathematical signals. Real data usually contain measurement noise, can be
non-stationary and are frequently measured over a brief time. Some
consequences of these properties are discussed below.

Important properties of time series

There are several properties of a time series that influence how it should be
analyzed. We discuss three of these in the following paragraphs. The
experimenter has direct control over the first two, while the other sometimes
can be manipulated by statistical procedures. All three properties entail
special nuances in data treatment and analysis, because they can influence the
results of nonlinear time-series analysis, sometimes considerably, depending
on both the system and the particular analysis.
The first property is the sampling rate. This is simply the rate at which
observations are recorded by the researcher. The sampling rate can be as
many as thousands of observations per second or as few as a single
observation per day or less. For most of the analyses discussed in this
chapter, choosing an appropriate sample rate is critical for obtaining valid
results. A sampling rate that is too low will not provide enough information
about a behaviour. However, a sampling rate that is too high can lead to an
oversaturation of data points (Abarbanel 1996); rather than providing new
information about the system under study, an oversampled time series gives
the illusion of greater sequential dependence than might be present. There are
clear criteria for sampling frequency selection for linear systems (e.g. the
Nyquist sampling theorem) but nonlinear systems do not necessarily follow
the same guidelines. One should be careful to choose a sampling rate that
does not oversample the phenomenon of interest and should tailor sampling
frequency to the guiding questions of the study and pragmatic considerations
(see Chapter 2 of Kantz and Schreiber 2004). For example, it is often not
necessary to sample the position of a player in the game every 100
milliseconds but it may be necessary if a very detailed analysis of a single
event is desired (e.g. attacker defender interaction during an attack in
football).
A second property is the time series length. Generally, results are more
accurate with a large number of data points (N). Typically the minimum N
ranges from about 100 to 1,000, depending on the analysis. This requirement
exists because many techniques are derived from methods that assume
mathematically perfect time series of infinite length. Obviously, the challenge
for the experimenter lies in balancing real-world practicality with the need to
obtain as large an N as possible. A temptation is to simply increase the
sampling rate to increase N; however, because of the oversaturation problem
noted above, this is often not a valid solution.
A third property is stationarity. Stationarity is present when the moments
of the distribution of the data are independent of time (Kaplan and Glass
1995). A more practical definition of stationarity is that the parameters of the
measured system remain constant over the duration of the measurement
(Kantz and Schreiber 2004). Most time-series analysis methods assume
stationarity and will yield erroneous results if used on non-stationary data. It
is often practically impossible to positively establish that a given time series
is stationary with absolute certainty but, for most purposes, it is sufficient to
check for stationarity of the mean (first moment) and variance (second
moment) because these are assumed by many analyses. A stationary time
series has a stable mean and variance, while those quantities change over
time for a non-stationary time series (see Figure 7.2 in Chapter 7). An easy
method to investigate a time series’ stationarity is to divide the series into
several non-overlapping windows. If the means and variances of each
window are statistically different, it is likely the series is nonstationary.
Sometimes it is possible to make a time series stationary by differencing
(subtracting each data point from the subsequent value) or by fitting a line or
polynomial function and subtracting the fitted trends from the data (Kantz
and Schreiber 2004). However, this is often not advisable – you may discard
important information. In the case of non-stationary data, it is advisable to
use a method that does not assume stationarity, such as RQA (Riley et al.
1999; Webber and Zbilut 1994, 1996, 2005).
The variety of nonlinear measures
In the following sections we briefly review several methods that have been
developed for examining the time-dependent properties of behavioural time
series. Many other methods exist and the studies presented in each category
merely scratch the surface of the current literature and do not constitute the
full scope of the work done using these techniques.

Recurrence quantification analysis

Recurrence plots were developed by Eckmann et al. (1987). Recurrence plots
capitalize on the property of dynamical systems to repeat over time (i.e. to
recur). They allow you to visualize how the states of dynamical systems
evolve using a two-dimensional plot and can reveal hidden patterns in
complex and irregularlooking data sets (Marwan et al. 2007). Webber and
Zbilut (Webber 1991; Webber and Zbilut 1994, 1996) developed RQA as a
way of objectively quantifying the visual features of recurrence plots. RQA
allows one to measure (with corresponding RQA measures in parentheses):
how often the system visits the same state (recurrence rate); how often the
same sequences of states repeat (determinism); how long are the repeating
segments (meanline and maxline); how many different patterns of repeated
sequences are there (entropy); how long the time series remains in the same
state (laminarity and trapping time); and whether the time series is stationary
(trend). RQA is often suitable for noisy, brief, and non-stationary time series,
but presents the challenge of having to identify values of a number of input
variables. Webber and Zbilut (2005) provided an excellent tutorial on RQA.
RQA has been successfully applied in a variety of settings from
astrophysics to psychology. There are few applications in sport science so far.
One study by Cotuk and Yavuz (2007) used RQA to examine the changes in
the dynamics of successive passes in football games from the 2006 World
Championship as an indication of play organization of the teams.

Fractal measures
Fractal methods have been widely used for investigating dynamical systems
in physiology, movement science, psychology and other disciplines (e.g.
Bassingthwaighte et al. 1994; Holden 2005; Liebovitch 1998). Most of these
methods describe how a measure of variability scales with sample size (i.e.
the amount of data over which the measure is computed). Many people are
familiar with visual images of fractals that repeat geometric patterns (i.e. they
are selfsimilar) across different spatial scales. Fractal signals are also self-
similar across scales – this is what is meant by ‘variability scales with sample
size’ – but, in this case, we refer to time scales rather than spatial scales and
the self-similarity is statistical rather than exact. The various fractal analyses
usually provide a single measure, the scaling exponent, which describes the
relation between the measure of variability and time scale. Brown and
Liebovitch (2010) provide a good introduction to practical uses of fractal
methods and Chapter 7 discusses the closely related concept of long-range
correlations.
There are many kinds of fractal analyses, each with unique procedures and
assumptions about the data being analyzed (Eke et al. 2000; Gao et al. 2006).
Spectral analysis (see Chapter 7) – a linear method – can be used to measure
fractal scaling in a time series because fractals exhibit what is termed a 1/f
power spectrum wherein spectral power scales inversely with frequency, f. In
this case, the scaling exponent is the slope of a linear fit (in log–log
coordinates) of the power spectrum. For certain types of data, spectral
analysis can be a preferred method but it places limiting assumptions on the
data (particularly stationarity).
A robust (especially with regard to non-stationarity) and widely used
fractal method is DFA (Peng et al. 1994). DFA yields a scaling exponent, α,
which describes how a variability measure called the detrended fluctuation
function scales with the size of a time window over which it is computed.
Like scaling exponents derived from other fractal analyses, α can be used to
classify the type of sequential dependence in a time series. For α = 0.5, the
time series lacks sequential dependence; the data are random white noise and
each point is independent of the others. For α = 1.0, the time series possesses
sequential dependence; the data are correlated. The particular sequential
dependence indicated by this α value is pink noise, which is another term for
1/f noise. For α = 1.5, the data are more strongly structured brown noise.
DFA and related fractal methods have been frequently used to study motor
control, where changes in α may indicate changes in neuromuscular control,
such as those that result from learning. One recent study, for example,
showed that people's hand movements become more pink (closer to ideal 1/f
noise) with practice on a Fitts’ law task (Wijnants et al. 2009).

Approximate entropy
Entropy is a measure of the amount of disorder in a system. To get an
intuitive feel for the meaning of the original usage of the concept of entropy
(from statistical mechanics), imagine a room with a bottle of perfume in the
middle. There is no draft or exchange of air with the outside world – the
room is a closed system. In the initial state, the molecules of the perfume are
all concentrated in the bottle and therefore are in a state of low entropy (high
order) with respect to the total positions that the room allows them to take.
Low entropy reflects the fact that the distribution of perfume molecules is not
uniform across the room such that one particular location contains a
disproportionally large quantity of molecules. When the bottle is opened, the
perfume molecules spread in the room and will fill the whole room uniformly
given enough time. In this process, the perfume molecules reach a state of
high entropy (high disorder), wherein all parts of the room have the same
probability of housing perfume molecules.2
There are similar notions about the amount of ‘disorder’ in observed
measurements. This sense of entropy is rooted in information theory
(Schneider and Sagan 2005). The information-theoretical definition of
entropy (Kolmogorov– Sinai entropy; KS) indexes the predictability of a time
series (Gao et al. 2007). Entropy in this case is related to the following
question: if we measure a value of a system (a value in a time series) at a
particular moment in time, how much can we predict about the next state of
the system (Kantz and Schreiber 1994) or how much information is generated
about the system with each measurement? For deterministic periodic systems,
prediction of future states is easy – we only need to know a few points to be
able to perfectly predict their evolution; these systems have low entropy. For
example, knowing only one cycle of a sine wave is enough to fully describe
the underlying system that produced it (Gao et al. 2007). Things become
more interesting when irregularity and randomness appear because the rate of
information obtained about the system underlying these time series increases
with each measurement. Such systems have higher entropy.
Pincus (1991) introduced approximate entropy (ApEn) with the intention
of providing a practical method of calculating the regularity or repeatability
of relatively short and noisy empirical time series. This measure is
conceptually related to KS entropy but KS entropy requires really large N to
be accurate. The logic of ApEn is simple: what is the average probability that
a sequence of m + 1 data points finds a match in the time series given that it
has already found a match for m data points? (see Pincus 1991, for a
mathematical definition). Matches do not have to be exact; matching
sequences are identified within a tolerance defined by r. The probability of
finding matches is expressed as a negative logarithm to yield the ApEn value.
If ApEn is closer to zero, the signal is very regular, predictable, and less
complex – the next observation can be readily predicted from the previous m
observations. If ApEn is high (closer to two), the signal is more
unpredictable, random and, consequently, more complex. We suggest Pincus
and Goldberger (1994) for a detailed step-by-step visual introduction to
ApEn.
ApEn has been applied to cardiovascular dynamics (Pincus and Viscarello
1992; Tulppo et al. 2001), postural control (Cavanaugh et al. 2005), isometric
force production (Slifkin and Newell 1999) and psychological time series
(Bauer et al. 2011; Yeragani et al. 2003). A general finding from the
application of this and other measures of complexity to the cardiovascular
system is that less-healthy systems show more regular, less-complex
dynamics (lower ApEn). In the context of cardiac physiology, Pincus (1994,
2006) hypothesized that a decrease in complexity corresponds to a
breakdown of communication between the subsystems participating in
cardiovascular control. Conversely, increased complexity (higher ApEn)
suggests greater coupling between the subsystems and fast and efficient
communication. This hypothesis has been supported using numerical
simulations from a variety of different types of mathematical models (Pincus
1994). These and similar observations using DFA have led to a general
perspective (Goldberger et al. 2002) that associates health with complex
fluctuations in physiological systems and pathology with a loss of
complexity. However, the pattern of changes in complexity with disease does
not always follow this trend (Vaillancourt and Newell 2002).
ApEn has also been applied to behavioural time series. Cavanaugh et al.
(2005) found that the regularity of centre of pressure (COP) fluctuations
exhibited by healthy young adults decreased with the removal of sensory
information. Following Pincus (1994), they interpreted the decrease in
complexity as an indication of restriction of the interactions among
components of the postural system. Cavanaugh et al. (2005) also reported
that athletes, following concussion, have more regular sway compared with
their own preinjury baseline values. In terms of recovery rates from
concussion, Cavanaugh et al. (2006) found that, while the traditional measure
of postural stability (equilibrium score) returned to preinjury levels within
three to four days, ApEn remained lower than the baseline level beyond this
period. The major implication is that the recovery is not as fast as has been
previously thought and that these athletes should not be allowed to return to
sport participation so soon.
Despite its success, ApEn is susceptible to some shortcomings related to
the specifics of the calculation of the conditional probabilities of repeating
patterns (Richman and Moorman 2000). The original ApEn algorithm
requires that each pattern of length m and m + 1 (called template vectors) find
at least one match in the time series, because the algorithm involves taking a
logarithm and the log of 0 is undefined. The algorithm therefore counts self-
matches in the estimates of conditional probability to make sure that at least
one match is present. This means that ApEn is a biased statistic and does not
estimate the population value of entropy well, especially for short time series.
To alleviate the bias, Richman and Moorman introduced an improved
algorithm called sample entropy, SampEn (described in detail below), which
does not count self-matches and uses a slightly different procedure to
quantify regularity of the time series. Additionally, ApEn is more affected by
measurement noise and sampling frequency than SampEn (Rhea et al. 2011).
Practically, this means that SampEn may be more reliable than ApEn when
comparing across studies that use different equipment or sampling rates.
Because of these limitations of ApEn, we focus on SampEn as a measure of
complexity.
Sample entropy
SampEn also quantifies signal regularity and is conceptually similar to ApEn.
Here, we demonstrate the SampEn calculations based on the description
provided by Richman and Moorman (2000). It is possible to do these
calculations using Microsoft Excel® but Matlab® (MathWorks Inc.) is better
because it is well suited for working with vectors. The Excel and Matlab files
used in the example are posted online at
http://homepages.uc.edu/~rileym/pmdlab/nonlinear/index.html. The sample
code is for illustration purposes only and we encourage you to use the more
robust code by Lake, Moorman and Hanqing available on PhysioNet
(www.physionet.org) for data analysis.
Assume that we continuously sampled a time series from a system of
interest. A plot of these data (Figure 5.3A) shows that they are relatively
stationary (the mean is about 5 and the variance does not change drastically
over the measurement period). The presence of stationarity is an important
requirement for SampEn (Govindan et al. 2007). Our example data series is
[0, 4, 8, 0, 4, 2, 0, 10, 8, 10, 7, 2, 3, 6, 1, 6, 7, 9, 0, 10, 8, 7, 4, 1, 8, 5, 9, 8, 2].
We first define all possible vectors of length m = 2 from the original N =
30 time series. A vector in this case is simply an array of numbers taken from
the original time series. We combine two consecutive measurements into a
vector and then move ahead one point to define the next two-element vector
until the last data point is reached. These vectors are fundamental to the
algorithm because all other calculations are based on them. Following the
algorithm for SampEn introduced by Richman and Moorman (2000), the total
number of vectors will be equal to N–m; in this case, it is 28. The vectors are
(where i stands for the vector number):
Figure 5.3 Panel A shows the simulated time series used in the tutorial calculation of SampEn; panel B
shows the same time series with added patterns of [1, 2, 3] values (shown in squares) to increase the
regularity of the time series (see text for details)

Xm = 2 (i = 1) = [0, 4],
Xm = 2 (i = 2) = [4, 8],
Xm = 2 (i = 3) = [8, 0],
Xm = 2 (i = 4) = [0, 4],
Xm = 2 (i = 5) = [4, 2],
Xm = 2 (i = 6) = [2, 0],
Xm = 2 (i = 7) = [0, 10],
…
Xm = 2 (i = 28) = [9, 8].
Now we designate one vector as a template with which all other vectors are
compared. We will take Xm = 2 (i = 1) as a template and find the number of
other vectors whose respective elements (i.e. values in the time series) differ
from the template by an amount less than the matching threshold parameter r.
For this example, we set r = 1. We exclude self-matches; they do not add any
new information about the regularity of the vectors extracted from the time
series (self-matches are counted in ApEn). The calculation below applies to
the first vector Xm = 2 (i = 1) but the procedure is exactly identical for all other
vectors:
Xm = 2 (i = 1) – Xm = 2 (i = 2) = [–4, –4],
Xm = 2 (i = 1) – Xm = 2 (i = 3) = [–8, 4],
Xm = 2 (i = 1) – Xm = 2 (i = 4) = [0, 0],
Xm = 2 (i = 1) – Xm = 2 (i = 5) = [–4, 2],
Xm = 2 (i = 1) – Xm = 2 (i = 6) = [–2, 4],
Xm = 2 (i = 1) – Xm = 2 (i = 7) = [0, –6],
…
Xm = 2 (i = 1) – Xm = 2 (i = 28) = [–9, –4].
Out of all possible 27 comparisons, only one vector Xm = 2 (i = 4) is within r =
1 from the template vector Xm = 2 (i = 1). We record that fact as a match for
this particular vector i of length m = 2:MATCHi=1 m=2 = 1. We then calculate
the probability of having a matching vector for our template by dividing the
number of observed matches by the number of possible matches:
Bi=1 m=2(r) = MATCHi=1 m=2/(N–m–1) = 1/27 = 0.037.
We repeat these steps for all vectors (i from 1 to 28) and find the average
probability of finding a matching vector Bm=2(r) for vectors of length m = 2
in the whole time series:
Bm(r) = sum(Bi=1:27 m=2(r))/(N–m).
For this particular time series, Bm=2(r) = 0.00793651 represents the
probability of finding a vector of length m = 2 that matches the template
vector within the radius r in the time series.
 We then repeat these steps for vector length m+1 to find the probability of
finding matching vectors of length 3. This quantity is exactly equal to Bm=3 as
defined above but, in the literature, it is sometimes referred to as Am(r). In the
case of this time series, Am(r) = 0.00284903.
We then convert the probabilities of observing recurrences of vectors of
length m and m+1 into the numbers of actual recurrences denoted by B and A
for Bm(r) and Am(r), respectively. Using the formula provided by Richman
and Moorman (2000):
B = [((N–m–1) × (N–m))/2]×Bm(r)
A = [((N–m–1) × (N–m))/2]×Am(r).
This calculation is also warranted because simply calculating all possible
matches of between the vectors overestimates the number of real matches.
For example, vector 1 may be recurrent with vector 2 and the algorithm
would automatically count the match of vector 2 with vector 1, as well thus
introducing redundancy. This formula removes the redundancy – it
effectively forces only forward matches to be counted.
We then find the ratio between the number of matches of m+1 length
(Am(r)) and the number of matches of length m (Bm(r)). The ratio of A to B is
the conditional probability that two sequences within a tolerance r for m
points remain within r of each other at the next point (Richman and Moorman
2000, p. 2042). We take the negative natural logarithm (ln) of this ratio to
make the final value positive, since number of matches of length m+1 will
always be less than the number of matches of length m. If the A is exactly the
same as B, then we have a limiting case in which the SampEn of the system is
0. As the time series becomes less predictable, the number of m+1 matches
becomes smaller, making the ratio closer to 0 while the –ln of the ratio
increases. The SampEn of our sample data time series is thus:
SampEn(m,r,N) = –ln(A/B) = –ln(1/15) = 1.0986
This rather high SampEn value suggests the time series has few repeatable
patterns that remain close to one another. This is not surprising because these
data were actually generated using uncorrelated samples from a Gaussian
distribution. The lowest non-zero SampEn value is [(N–m–1)*(N–m)]/2 and
the maximum value is ln(N–m)+ln(N–m–1)–ln(2) (Richman and Moorman
2000).
To gain additional intuitions about what SampEn measures, we
deliberately increased the degree of predictability of the measurements by
introducing sequences of values [1, 2, 3] in the early and late parts of the time
series (Figure 5.3B). After performing the calculation described above, we
find that SampEn is 0.559 – a lower value than one for the original time
series.
It is always possible to get a number for SampEn. It is up to the researcher
to make sense of the results. In general, the interpretation is easier if one has
a good intuitive feel for the structure in the data, which can be enhanced by
always inspecting a plot of the data

Empirical considerations for using SampEn

Parameter selection
Typical parameter settings are m = 1 or 2 and r between 0.1 and 0.25 of the
standard deviation of the time series (Pincus and Goldberger 1994).
However, there are more involved selection criteria that rely on the
calculation of autoregression parameters of the time series (Lake et al. 2002).
Alternatively, Ramdani et al. (2009) proposed to estimate m by plotting
median SampEn values for the time series as a function of different values of
r. The m value at which the SampEn-r curves become similar should be
selected. They also recommended selecting r based on estimates of SampEn
relative error in entropy estimation (Richman and Moorman 2000). The r
value at which relative error is minimal should be chosen for the analysis.
One thing to keep in mind is that these parameters should be kept constant
between all compared conditions once they are selected (Pincus 1991). A
practical piece of advice is to perform the analysis with a range of different r
values to make sure that the results are not just an artefact of parameter
selection.

Data length
The appropriate length of the time series for regularity classification depends
on the quality of the measurements. In some cases, signals as short as 60
points may work (Pincus 2006). Richman and Moorman (2000) suggested
using time series of the order of 100 to 20,000 data points.
There are situations when the experimenter expects that the lengths of time
series will naturally differ across conditions because of natural differences in
the durations of measured behaviour. For example, one may be interested in
the complexity of the attacker's movement trajectory between successful and
unsuccessful attacker–defender situations. The duration of the measured time
series will differ from one behavioural sequence to another, owing to a host
of factors. Despite the fact that length of the time series may distort the
results of SampEn, it is still possible to compare the regularity across these
qualitatively similar conditions. But to minimize the effects of data length it
is advisable to standardize the length of the time series to some reasonable
value (e.g. average length of all recorded time series).3 In other situations, it
may not be possible to use the same strategy, especially when the behaviour
of interest qualitatively changes as a function of measurement length. For
example, comparing the complexity of heart rate between a four-minute and a
ten-minute practice period is not appropriate because the players may be
using different strategies for conserving energy between the two.

Outliers
SampEn is not very susceptible to singular outliers. However, one way
outliers may affect estimation of SampEn is by biasing the value of r. A
typical suggestion is to set r as a percentage of the standard deviation of the
series (e.g. 10–25%). However, the standard deviation (SD) is likely to be
inflated owing to outliers, leading to an increase in r and a consequent
decrease in SampEn. One can remove outliers if appropriate or use the
median instead of the mean for the SD calculation.

Non-stationarity
SampEn is designed to work with stationary data and it is likely to
malfunction when strong non-stationarity is present. Positively establishing
stationarity is tricky in nonlinear systems (Kantz and Schreiber 2004). One
practical solution is to use RQA to check for stationarity prior to SampEn
analysis using the trend parameter. As an example of poor applicability of
SampEn to non-stationary signals, we take the results of Rhea et al. (2011),
who showed that SampEn of non-differenced COP position data decreased
with faster sampling rate whereas differenced, stationary COP signals were
not subject to this artefact. Ramdani et al. (2009) showed that SampEn
discriminated between standing with eyes open compared with closed only
when analyzing the differenced (stationary) COP. Of course, differencing
removes non-stationarity but it should only be done if the differenced signal
is theoretically meaningful as in the case of postural control (Delignières et
al. 2011).

Long-range correlations
A time series has long-range correlations when the autocorrelation function
decays exponentially slowly as a function of time lag (Diniz et al. 2011). This
is typical of fractal processes described earlier and in Chapter 7. The presence
of long-range correlations reduces estimates of system complexity provided
by SampEn, potentially biasing the estimate (Govindan et al. 2007; Richman
and Moorman 2000). SampEn assumes that there are only significant lag-1
autocorrelations because the template vectors are created from consecutive
values from the time series. If there are correlations, then the template vectors
need to be defined from non-consecutive values (e.g. take every fifth value)
to minimize the correlations between the template vectors (Govindan et al.
2007). However, for truly long-range correlated signals it is impossible to
find an appropriate lag for creating the vectors, so Govindan et al. (2007) also
suggested differencing the time series and conducting the analysis on the
increments.

Periodic data
The method proposed by Govindan et al. (2007) will be useful for periodic,
continuously sampled data such as breathing and gait kinematics. In such
cases, it is still possible to use SampEn but now we need to introduce a delay
into the definition of the template vectors. Another possibility is to difference
the data as suggested above (Bruce 1996) or define the events of interest and
do the analysis on the differences between these events along the time or
magnitude dimensions. One apparent disadvantage of increasing the delay
time for vector embedding is a reduction in the number of template vectors.

Sampling rate
Time series collected using A/D converters are digitized versions of the
recorded continuous processes. When the resolution of the A/D converter (or
any other measurement procedure) is not fine enough to capture the
continuous dynamics of the phenomenon of interest, SampEn of the system
will become an ordinal variable (Stevens 1946). Consider the time series of
COP velocity presented in the left panel of Figure 5.4. This time series was
obtained by differencing the COP recorded from a person standing on a force
plate sampled at 100 Hz for three seconds. There is a clear discretization
effect such that the values of the velocity vary between a limited number of
states (especially –0.006, 0.003, 0.004, and 0) – this is also known as
quantization error. In such cases, entropy does not change continuously with
r. To illustrate this, we calculated SampEn as a function of r varying between
0.05 and 1 in steps of 0.05 and plotted the SampEn–r curve in the right-hand
panel of Figure 5.4. As expected, increasing the radius decreases SampEn but
there are also apparent plateaus of constant entropy. The effect of
discretization seems to be more apparent in short time series. Plateaus and
discontinuity in the entropy estimates on SampEn–r plots make SampEn an
ordinal-dependent variable, with the implication that only non-parametric
statistics should be used for data analysis. Therefore, we recommend
examining how SampEn changes as a function of r in a representative subset
of the data before doing the statistical analysis to establish whether
parametric or non-parametric statistics should be used. If SampEn changes
continuously with r for all experimental time series, then parametric statistics
are appropriate.

Figure 5.4 Illustration of the digitization (quantization) effect on the SampEn calculation
Filtering
Filtering usually makes the time series more regular so it will lower SampEn.
As long as all conditions are filtered similarly, this should not affect the
overall pattern of results within a study. But differences in the filtering
procedures need to be considered when making comparisons across studies
(Rhea et al. 2011). In general, a good practice is to filter the signal as little as
possible because excessive filtering may remove the legitimate aspects of the
dynamics in the time series instead of measurement artefacts (Abarbanel
1996).
Implications for sport science and conclusions
The dynamical systems approach and the time-series analyses described in
this chapter can potentially provide useful information about the phenomena
of interest to sport scientists beyond that of the standard measures of central
tendency and variability. Time-series measures take into consideration time-
dependent properties of the data and therefore are sensitive to changes in
evolution of the time series. Our hope is that this will allow researchers to
develop more sensitive measures of performance, more efficient methods of
learning new skills and more effective treatment evaluation protocols for
injuries. Measures of signal complexity such as SampEn are especially
promising because they can be readily applied to short and noisy sequences
of data that are typical of sport science.

Questions for students

1. How can nonlinear measures be applied in such a way so as
provide supplemental information to the questions previously
addressed by traditional measures (mean, SD)?
2. What kinds of questions do you wish to ask about your system of
interest? This will constrain the choice of analysis method.
3. Does your measurement protocol permit collection of a sufficiently
long time series?
4. Can you ground the results of nonlinear time series analysis in
some mechanistic understanding or model of your system of
interest?
5. Do your data meet the assumptions of your preferred analysis
method?
6. How do you make empirical predictions based on dynamical
system models or time series methods?
Acknowledgements
Supported by NSF grant BCS 0926662.
Notes
1 Standard statistical quantities such as mean and variance derive from a linear perspective on how
the natural world works. For this reason these are sometimes (e.g. Harbourne and Stergiou 2009)
referred to as ‘linear measures’ and this term is used to distinguish them from other methods of
analysis. This is a valid distinction when comparing those quantities to the kinds of nonlinear
methods featured later in this chapter. A more accurate and fundamental distinction is that only
time series methods are sensitive to sequential dependence. Moreover, there are both linear (e.g.
autocorrelation and spectral analysis; see Chapter 7) and nonlinear (e.g. DFA and many others
described later) time-series analyses; thus, labelling those statistical quantities as ‘linear’ does
not accurately distinguish them.
2 Note that ‘order’ and ‘disorder’ have somewhat counterintuitive meanings in this context
provided by statistical mechanics. Disorder is tantamount to uniformity and order to non-
uniformity. When the perfume is in the bottle, the molecules are all concentrated in that location
rather than dispersed evenly across the room, as is the case after the bottle has been opened and
the perfume has uniformly filled the room. In the latter case, individual perfume molecules could
swap locations in the room without creating a change in the overall system, while that is not true
when the molecules are enclosed in the bottle.
3 For example, in Matlab ‘normData = spline(1:length(data),data,linspace(1,length (data),100))’;
will turn a time series of any N into one with a length of 100 points while preserving the overall
pattern of the observations.
References
Abarbanel, H. D. I. (1996) Analysis of Observed Chaotic Data. New York:
Springer.
Bassingthwaighte, J. B., Liebovitch, L. S. and West, B. J. (1994) Fractal
Physiology. New York: Oxford University Press.
Bauer, M., Glenn, T., Alda, M., Grof, P., Sagduyu, K., Bauer, R., Lewitzka,
U. and Whybrow, P. C. (2011) Comparison of pre-episode and pre-remission
states using mood ratings from patients with bipolar disorder.
Pharmacopsychiatry, 44: S49–53.
Brown, C. and Liebovitch, L. (2010) Fractal Analysis. Los Angeles, CA:
Sage.
Bruce, E. N. (1996) Measures of respiratory pattern variability, in M. C. K.
Khoo (ed.) Bioengineering Approaches to Pulmonary Physiology and
Medicine. New York: Plenum, pp. 149–59.
Carello, C. and Moreno, M. (2005) Why nonlinear methods? in M. A. Riley
and G. C. Van Orden (eds) Tutorials in Contemporary Nonlinear Methods for
Behavioral Scientists Web Book. Arlington, VA: National Science
Foundation, pp. 1–25. Available online at
www.nsf.gov/sbe/bcs/pac/nmbs/nmbs.jsp (accessed 6 June 2013).
Cavanaugh, J. T., Guskiewicz, K. M. and Stergiou, N. (2005) A nonlinear
dynamic approach for evaluating postural control: new directions for the
management of sportrelated cerebral concussion. Sports Medicine, 35: 935–
50.
Cavanaugh, J. T., Guskiewicz, K. M., Giuliani, C., Marshall, S., Mercer, V.
S. and Stergiou, N. (2006) Recovery of postural control after cerebral
concussion: new insights using approximate entropy. Journal of Athletic
Training, 41: 305–13.
Cotuk, B. and Yavuz, E. (2007) Recurrence plot analysis of successive
passing sequences in 2006 World Championship. Journal of Sports Science
and Medicine, 6 (Suppl 10): 4. Available online at
www.aemef.org/turquia_2007.pdf (accessed 6 June 2013).
Delignières, D., Fortes, M. and Ninot, G. (2004) The fractal dynamics of self-
esteem and physical self. Nonlinear Dynamics in Psychology and Life
Sciences, 8: 479–510.
Delignières, D., Torre, K. and Bernard, P. L. (2011) Interest of velocity
variability and maximal velocity for characterizing center-of-pressure
fluctuations. Science and Motricité, 74: 31–7.
Diniz, A., Wijnants, M. L., Torre, K., Barreiros, J., Crato, N., Bosman, A. M.
T., Hasselman, F., Cox, R. F. A., Van Orden, G. C. and Delignières, D.
(2011) Contemporary theories of 1/f noise in motor control. Human
Movement Science, 30: 889–905.
Eckmann, J.-P., Kamphorst, S. O., and Ruelle, D. (1987) Recurrence plots of
dynamical systems. Europhysics Letters, 4: 973–7.
Eke, A., Herman, P., Bassingthwaighte, J. B., Raymound, G. M., Percival, D.
B., Cannon, M., Balla, I. and Ikrenyi, C. (2000) Physiological time series:
distinguishing fractal noises from motions. European Journal of Physiology,
439: 403–15.
Gao, J., Hu, J., Tung, W.-W., Cao, Y., Sarshar, N., and Roychowdhury, V. P.
(2006) Assessment of long-range correlation in time series: how to avoid
pitfalls. Physical Review E, Statistical, Nonlinear, and Soft Matter Physics,
73 (1 Pt 2): 016117.
Gao, J., Cao, Y., Tung, W-.W. and Hu, J. (2007) Multiscale Analysis of
Complex Time Series. Hoboken, NJ: John Wiley and Sons.
Goldberger, A. L., Peng, C.-K., and Lipsitz, L. A. (2002) What is physiologic
complexity and how does it change with aging and disease? Neurobiology of
Aging, 23: 23–6.
Govindan, R. B., Wilson, J. D., Eswaran, H., Lowery, C. L. and Preisl, H.
(2007) Revisiting sample entropy analysis. Physica A Statistical Mechanics
and its Applications, 376: 158–64.
Harbourne, R.T. and Stergiou, N. (2009) Movement variability and the use of
nonlinear tools: principles to guide physical therapist practice. Physical
Therapy, 89: 267–82.
Holden, J. G. (2005) Gauging the fractal dimension of response times from
cognitive tasks, in M. A. Riley and G. C. Van Orden (eds) Tutorials in
Contemporary Nonlinear Methods for Behavioral Scientists Web Book.
Arlington, VA: National Science Foundation, pp. 267–318. Available online
at www.nsf.gov/sbe/bcs/pac/nmbs/nmbs.jsp (accessed 6 June 2013).
Kantz, H. and Schreiber, T. (2004) Nonlinear Time Series Analysis.
Cambridge: Cambridge University Press.
Kaplan, D. and Glass, L. (1995) Understanding Nonlinear Dynamics. New
York: Springer.
Lake, D. E., Richman, J. S., Griffin, M. P. and Moorman, J. R. (2002)
Sample entropy analysis of neonatal heart rate variability. American Journal
of Physiology Regulatory, Integrative and Comparative Physiology, 283:
R789–97.
Liebovich, L. S. (1998) Fractals and Chaos Simplified for the Life Sciences.
New York: Oxford University Press.
Marwan. N., Romano, M.C., Thiel, M. and Kurths, J. (2007) Recurrence
plots for the analysis of complex systems. Physics Reports, 438: 237–9.
Pellecchia, G. L. and Shockley, K. (2005) Application of recurrence
quantification analysis: Influence of cognitive activity on postural
fluctuations, in M. A. Riley and G. C. Van Orden (eds) Tutorials in
Contemporary Nonlinear Methods for Behavioral Scientists Web Book.
Arlington, VA: National Science Foundation, pp. 95–141. Available online at
www.nsf.gov/sbe/bcs/pac/nmbs/nmbs.jsp (accessed 6 June 2013).
Peng, C.-K., Buldyrev, S. V., Havlin, S., Simons, M., Stanley, H. E. and
Goldberger, A. L. (1994) Mosaic organization of DNA nucleotides. Physical
Review E, 49: 1685–9.
Pincus, S. and Viscarello, R. (1992) Approximate entropy: a regularity
measure for fetal heart rate analysis. Obstetrics and Gynecology, 79: 249–55.
Pincus, S. M. (1991) Approximate entropy as a measure of system
complexity. Proceedings of the National Academy of Sciences, 88: 2297–301.
Pincus, S. M. (1994) Greater signal regularity may indicate increased system
isolation. Mathematical Biosciences, 122: 161–81.
Pincus, S. M. (2006) Approximate entropy as a measure of irregularity for
psychiatric serial metrics. Bipolar Disorders, 8: 430–40.
Pincus, S. M. and Goldberger, A. L. (1994) Physiological time-series
analysis: what does regularity quantify? American Journal of Physiology
Heart and Circulatory Physiology, 266: H1643–56.
Ramdani, S., Seigle, B., Lagarde, J., Bouchara, F. and Bernard, P. L. (2009)
On the use of sample entropy to analyze human postural sway data. Medical
Engineering and Physics, 31: 1023–31.
Rhea, C. K., Silver, T. A., Hong, S. L., Ryu, J. H., Studenka, B. E., Hughes,
C. M. L. and Haddad, J. M. (2011) Noise and complexity in human postural
control: interpreting the different estimations of entropy. PLoS One, 6 (3):
e17696; doi:10.1371/journal.pone.0017696
Richman, J. S. and Moorman, J. R. (2000) Physiological time-series analysis
using approximate entropy and sample entropy. American Journal of
Physiology Heart and Circulatory Physiology, 278: H2039–49.
Riley, M. A. and Turvey, M. T. (2002) Variability and determinism in motor
behavior. Journal of Motor Behavior, 34: 99–125.
Riley, M. A., Balasubramaniam, R. and Turvey, M. T. (1999) Recurrence
quantification analysis of postural fluctuations. Gait and Posture, 9: 65–78.
Scheinkman, J. A. and LeBaron, B. (1989) Nonlinear dynamics and stock
returns. Journal of Business, 62: 311–37.
Schneider, E. D. and Sagan, D. (2005) Into the Cool: Energy Flow,
Thermodynamics, and Life. Chicago, IL: University of Chicago Press.
Slifkin, A. B. and Newell, K. M. (1999) Noise, information transmission, and
force variability. Journal of Experimental Psychology: Human Perception
and Performance, 25: 837–51.
Stevens, S. S. (1946) On the theory of scales of measurement. Science, 103:
677–80.
Takens, F. (1981) Detecting strange attractors in turbulence, in D. Rand, and
L.-S. Young (eds) Lecture Notes in Mathematics, Vol. 898, Dynamical
Systems and Turbulence, Warwick 1980. Berlin: Springer, pp. 366–81.
Tulppo, M. P., Hughson, R. L., Makikallio, T. H., Airaksinen, K. E. J.,
Seppanen, T. and Huikuri, H. V. (2001) Effects of exercise and passive head-
up tilt on fractal and complexity properties of heart rate dynamics. American
Journal of Physiology Heart and Circulatory Physiology, 280: 1081–7.
Vaillancourt, D. E. and Newell, K. M. (2002) Changing complexity in human
behavior and physiology through aging and disease. Neurobiology of Aging,
23: 1–11.
Van Orden, G. C., Holden, J. G. and Turvey, M. T. (2003) Self-organization
of cognitive performance. Journal of Experimental Psychology: General,
132: 331–50.
Webber, C.L. (1991) Rhythmogenesis of deterministic breathing patterns. In
H. P. Koepchen and H. Haken (eds) Rhythms in Physiological Systems, pp.
177–91. Berlin: Springer-Verlag.
Webber, C. L. and Zbilut, J. P. (1994) Dynamical assessment of
physiological systems and states using recurrence plot strategies. Journal of
Applies Physiology, 76: 965–73.
Webber, C. L. and Zbilut, J. P. (1996) Assessing deterministic structures in
physiological systems using recurrence plot strategies, in M. C. K. Khoo (ed.)
Bioengineering Approaches to Pulmonary Physiology and Medicine.
NewYork: Plenum Press, pp. 137–48.
Webber, C. L. and Zbilut, J. P. (2005) Recurrence quantification analysis of
nonlinear dynamical systems, in M. A. Riley and G. C. Van Orden (eds)
Tutorials in Contemporary Nonlinear Methods for Behavioral Scientists Web
Book. Arlington, VA: National Science Foundation, pp. 26–94. Available
online at www.nsf.gov/sbe/bcs/pac/nmbs/nmbs.jsp (accessed 6 June 2013).
Wijnants, M. L., Bosman, A. M. T., Hasselman, F., Cox, R. F. A. and Van
Orden, G. C. (2009) 1/f scaling in movement time changes with practice in
precision aiming. Nonlinear Dynamics, Psychology, and Life Sciences, 13:
75–94.
Yeragani, V. K., Pohl, R., Mallavarapu, M. and Balon, R. (2003)
Approximate entropy of symptoms of mood: an effective technique to
quantify regularity of mood. Bipolar Disorders, 5: 279–86.
6 Interpersonal coordination
tendencies induce functional
synergies through co-
adaptation processes in team
sports
Pedro Passos, Duarte Araújo, Bruno
Travassos, Luis Vilar and Ricardo
Duarte
The use of measures that capture coordination in different levels of team
sports, such as geometrical centres, inter- and intrateam interpersonal
distances, relative velocities and angles between opponent players, as well as
the network analysis, have been increasingly used to describe and explain
coordination between performers in sport. This chapter summarizes the
developments in this area of work.
During the last decade, there was an increase in research on interpersonal
coordination in team sports. At the first stage, the main issue of researchers
was to found variables that accurately describe the behaviour of a set of
players in interaction, in other words, how to capture how several players
coordinate with each other. These interactions prompt the emergence of
collective behaviours that attain levels of performance that are different from
the sum of individual performances (Duarte et al. 2012; Sumpter 2010). This
led us to the notion of functional synergies (i.e. coordinative structures),
which are groupings of structural elements (e.g. players) that temporally are
constrained to function as a single coherent unit (Kelso 2009).
Coordinated collective behaviours demand the existence of co-adaptation
(Kauffman 1993) among system components, a concept that was originally
used to characterize how agents cooperate within complex biological
systems. The existence of co-adaptation means that each agent within a
system acts accordingly with the nearest neighbour. For instance, when the
ball carrier of a sport team change his running line trajectory the players in
his neighbourhood aiming to maintain the support to the ball carrier also
adjust their running line trajectories accordingly. Thus, a continuous co-
adaptation is a key issue for intrateam coordination in team sports. The same
interpretation is lawful for interteam coordination since defenders
continuously co-adapt to the attackers' actions which in turn co-adapt to
defenders' behaviours. But do the players have the need (even if that was
possible) to continuously co-adapt to all other players within the pitch?
Co-adaptation demands that players within a performance field must be
informationally coupled. Some data have been supporting the expectations
that the strength of coupling increases with decreasing of interpersonal
distances among opponent players (Passos et al. 2011a). Players’ behaviours
are constrained by their intentions, goals and roles that decrease the
randomness in their movement displacement trajectories in the field.
However, the decreasing of interpersonal distances increases players’
contextual dependency, meaning that each player's behaviour is mutually
dependent from the behaviour of the other players in the neighbourhood.
When this behavioural dependency occurs the players enter in a critical
region where candidates to control parameters (e.g. players' relative
velocities) might increase or decrease values moving the system towards an
eventual performance outcome. When these candidates to control parameters
values equalized transition values the attacker–defender system enter in a
state of criticality (Jensen 1998), where something is about to occur and one
player (or team) will gain advantage over the other.
In this chapter, we focus on the most frequently used methods for
behavioural data analysis, mainly the tools and variables employed. Special
attention will be directed to the coordinative variables that were used in
different levels of analysis of team sports. Some studies go further and
present candidates to control parameters, which aim to explain how an
attacker–defender system moves from one stable state to another.
The last decade assisted to a considerable increase in the number of studies
dedicated to interpersonal coordination in team games. Sèveral researchers all
over the world invest a lot of work aiming to describe how individuals
interact on ongoing team sports performance. From Australia, New Zealand,
Singapore to Portugal, France, Canada and Netherlands, issues regarding how
players interact within a competitive performance environment were under
analysis with a common feature: the focus of analysis was on the interaction
that a performer has within a given context (e.g. the interaction between two
or more players) rather than on the player as a single unit. As a result
researchers found or create new variables to describe players' interactions
with opponents and teammates and also candidates to control parameters that
explain how the attacker–defender balance was broken. Associated with these
investigations, some innovative methods of analysis appeared to better help
sports scientists and practitioners to enhance their analyses, some of them
being adopted from other areas of research. This is what we describe in the
following sections of this chapter.
Uncovering interpersonal coordination in team sports
Players' behaviours are continuously constrained by a huge number of
variables, which implies that attacker–defender systems are constrained by
multiple causes that produce multiple effects, a general feature of complex
systems.
In order to increase the understanding concerning the complex nature of
interactive behaviours in team sports, researchers have been making an effort
to identify relevant coordinative variables, which are single variables that
capture and synthesise the interactive behaviours between the individual parts
of a system. These compound variables aim to accurately describe the system
states of coordination and its time-evolving dynamics (Duarte et al. 2012;
Kelso 2009). To explain why the system transit from one state of
coordination to another, researchers need also to explore the role of potential
candidates to control parameters.
Interpersonal distances and player velocities have been considered relevant
variables to analyze interpersonal coordination in team sports. The work of
Araújo et al. (2002) in basketball investigated whether the distance of the
attacker–defender dyad from the basket became less stable as the
interpersonal distance between the attacker and defender decreased. For that
purpose, the authors examined one-on-one attacker–defender dyadic
behaviour proposing the medium point between attacker and defender to the
basket as a coordinative variable that accurately describes dyadic system
behaviours. Aiming to explain when the attacker–defender balance was
broken, the authors proposed as candidates to control parameters the
interpersonal distance between an attacker and his immediate defender, which
when achieving a critical value, drove the dyadic system towards a specific
performance outcome, such as a defender interception or a shot to the basket
by the attacker. Candidates to control parameters sought to explain whether
the balance of the attacker–defender dyad was disturbed or even broken at
specific critical values of the players’ interpersonal distance. The data
revealed that the attacker–defender dyad distance to the basket as a
coordinative variable was able to accurately describe two different states of
the dyadic system: i) when the defender counterbalance the attackers’
movements, the dyadic system stability remains unchanged, which is
considered an advantageous situation for the defender; or ii) when the
attacker was able to dribble past the defender and move closer to the basket.
The former system state was consistent with a breaking in the spatiotemporal
symmetry of the attacker-defender relations. Araújo et al. (2002) identified
this sudden change in the dyadic system balance as evidence of a ‘phase
transition’, which is a general feature of a non-linear dynamical system.
Also aiming to accurately describe the attacker–defender dyadic
interactions, the works of Passos et al. (2006, 2008, 2009a) proposed a
coordinative variable which was the angle formed between a vector linking
the defender to the attacker and an imaginary horizontal line parallel to the
try line (Figure 6.1).
The data from this coordinative variable displayed the intermittency
between stability and volatility periods that characterize attacker–defender
interactions. These volatility periods occur because of a marked decrease in
players’ interpersonal distances. Thus, interpersonal distances are candidates
for controlling parameters that tend to move attacker–defender systems
towards critical regions where opponent players become mutually dependent
and some control parameters gain influence over another. For instance, the
increase in the relative velocity (i.e. the difference between attacker and
defender velocity) increases its influence on the performance outcome (i.e. a
try or a tackle) but only within certain regions. An increase in the difference
of players velocities led the system for try outcomes whereas similar
velocities between players led it to tackle outcomes. Interestingly, outside
specific values of interpersonal distances (i.e. around 4 metres in the studied
sample), relative velocity had little or no influence on the performance
outcome. This is why Passos et al. (2008) suggested the notion of nested
control parameters to explain that the attacker–defender balance was broken
and a new state of coordination has emerged. The candidates for nested
control parameters that we identified (i.e. interpersonal distance and attacker–
defender relative velocity) are emergent system constraints because they
became spontaneously coupled without interference from an external agent.
Figure 6.1 The calculation of the coordinative variable (reprinted with permission of the Journal of
Motor Behaviour)

A different approach to analyzing potential control parameters is shown in

the studies from Correia et al. (2011) in rugby union and Cordovil et al.
(2009) in basketball. Both manipulated task constraints, aiming to test
potential system parameters. The work of Correia et al. (2011) examined
whether manipulating the initial starting locations of two defenders near the
try line influence the performance outcome (i.e. try or tackle). Data revealed
important changes in the displacement trajectories of defenders as a function
of initial starting distance. When defending participants started the task
positioned further apart from each other, they tended to move closer together
(possibly acquiring an ‘optimal’ interpersonal distance that help them to face
as a collective unit) and wait for the attacker close to the try line (i.e.
defenders tended to move laterally across the try line instead of going
forward, decreasing the distance to the attacker). On the other hand, as the
initial starting distance between defenders was decreased, they tended to run
forward in the direction of the attacker (i.e. decreasing the distance to contact
the attacker). The data revealed that when the defender's initial starting
distance was between 20 and 10 metres, a try results as the performance
outcome. Below four metres of the defender's initial distance, the tackle was
the preferred outcome. However, between eight and six metres of the
defender's starting interpersonal distance, both system outcomes coexist with
similar frequencies. We suggest that Correia et al. (2011) identified a critical
region with plenty of uncertainty concerning the outcome. Owing to the
manipulated task constraints, this study indicated that defenders’ starting
distances between eight and six metres exponentially augmented uncertainty
to the agents involved on this triadic system. Thus, the decisions and actions
must be sustained by information that is locally created owing to the players’
contextual dependency. These findings add relevant insights for practice
designs creating learning environments where players must deal with the
uncertainty and the dynamics of the information that emerges locally.
Other types of task-constraint manipulation are illustrated in the work of
Cordovil et al. (2009). Aiming to analyze potential control parameters, they
sought to understand whether the patterns of coordination in the attacker–
defender–basket system could be disturbed under the influence of common
task constraints, such as coaching instructions and players’ height. In a first
experiment, different performance instructions (neutral, playing
conservatively and risk taking) were given to the attacking players. The data
revealed that coaching instructions disturbed the attacker–defender
interactive behaviours (i.e. the conservative instructions led to an increase in
the variability of attackers running-line trajectories; also, the attackers took
more time to cross the mid-line court) but do not lead to differences in the
frequency of phase transitions (i.e. the number of times that the attackers
dribbled past the defender). Therefore, differences in coaching instructions
failed as a candidate for control parameter. In a second experiment, the
authors created dyads with different height relations, aiming to analyze
whether this individual structural constraint might work as a potential control
parameter. Contrary to the findings of the previous experiment, the data
revealed significant differences in the frequency of phase-transition
occurrences whenever the attackers were shorter in height than the defenders
(Cordovil et al. 2009). These findings implied that a different height relation
between players in the team sport of basketball may be interpreted as a
candidate for control parameter that influences the balance in attacker–
defender dyadic systems, providing a strong advantage for the attackers.
Another example of task-constraint manipulation was the work of
Headrick et al. (2012) in one-on-one dyadic behaviour in association football.
The aim of this investigation was to analyze whether the proximity to the
goal area constrained the spatiotemporal interactions within attacker–
defender dyads plus the ball. The results revealed that the defender-to-ball
distance stabilized at significantly higher values (approximately 1.7 metres
apart) when the interaction among attacker and defender occurred close to the
defender's goal (which was consistent with a lower risk-taking behaviour
from the defender) than when occurred far from the defender's goal area
(defender-to-ball distance was approximately 1.2 metres; Headrick et al.
2012). When stabilizing the defender-to-ball distances, the authors suggested
that attacker and defender became contextually dependent and the dyadic
system entered a critical region. These critical regions displayed different
characteristics depending on the attacker–defender distance to the goal area.
In summary, results from studies in dyadic and triadic behaviours in team
sports revealed that players' behaviours are highly sensitive both to changes
in initial task constraints (e.g. distance to the goal, initial distance between
defenders, differences in players' height) and also to small changes in
interpersonal interactions within critical regions demarcated by close
interpersonal distances (e.g. changes in relative velocity). This signifies that
opponent players interact differently, depending on performer and task
constraints that shape the perceptual– motor landscape (the system region)
for action. The information that emerges from player interactions within the
critical regions implies that players' behaviours are context dependent and
small changes in relevant candidates to control parameters (e.g. relative
velocity) often lead to transitions in the attacker–defender system's structural
organization (e.g. characterized by the relative positioning of each player to
the goal). These transitions can be captured by abrupt and nonlinear changes
in the coordinative variables. In the next section, we describe how these
concepts are useful and explain collective behaviours observed in team
sports.

Beyond dyadic behaviour in team ball sports

According to Oullier and Kelso (2009), one of the major problems that
persists when investigating interpersonal coordination is the difficulty in
manipulating or measuring the intensity/strength of coupling between agents.
Following what was described for dyadic behaviours, the step further is to
study collective behaviours in team sports. Based on studies in biological
systems, Passos and colleagues developed research aiming to analyze pattern-
forming dynamics and the strength of connection among players within the
team sport of rugby union (Passos et al. 2011a). The analysis of collective
behaviours in team sports requires two levels of analysis: i) intrateam
couplings, aiming to characterize how players co-adapt to the teammates
behaviour; and ii) interteam couplings, aiming to analyze attacker–defender
co-adaptive behaviours.
Concerning the intra-team analyses, Passos et al. (2011a) explored
interpersonal distances as a relevant variable that bounded teammates’
behaviours. Similar to other biological systems (e.g. schools of fish, flock of
birds), the results display that, in team sports, players’ behaviours are ruled
by functional interpersonal distances, which are sensitive to specific task
constraints such as the opponents’ proximity. The results revealed that
players within an attacker subunit tend to remain close to each other when
playing before the first defensive line and augment their interpersonal
distances when playing between the first and the second defensive line. An
issue that Passos et al. (2011a) aimed to solve was how to measure the
strength of connection between players within an attacker subunit. Running
correlation analyses were successfully used for that purpose on a four versus
two plus two situations in rugby union.
An interteam analysis requires testing for a coordinative variable that
accurately synthesizes and captures the behaviour of a set of players. Milho et
al. (2010) suggested the use of geometrical centres (or centroids) calculated
from the mean values for interpersonal distance between players within the
subunit over time. This measure was later used in association football by
Frencken et al. (2011) and Lames et al. (2010). The authors proposed that a
centroid position and surface area (i.e. the space covered by the outfield
players of a team) may provide meaningful system variables that capture the
collective behaviours and the interactions between attacking and defending
teams. The use of these sorts of coordinative variables might help to solve a
problem raised by Oullier and Kelso (2009) concerning the large number of
agents that were under analysis to describe social collective behaviours. Data
from Milho et al. (2010) confirmed that the centroids trajectory emerging due
to systemic couplings between players allowed a description of the teams’
collective behaviours as a whole rather than as standalone entities or a sum of
individual performances. This interteam coordinative variable was imported
from the study of attacker–defender dyads and again calculated from the
angle of vector-linked centroids (from the defence to the attack subunit) with
an imaginary horizontal line parallel to the try line. When plotted on time
data it was revealed that this interteam coordinative variable described
different states of attacker–defender coordination: i) when the defenders were
the players closer to the try line; ii) the moment that attackers overpass the
defenders; and iii) when the attackers become the players closest the try line.
Moreover, owing to a continuous decrease in attacker–defender interpersonal
distances, an increase in the variability of the interteam coordinative
occurred, highlighting when the attacker–defender balance is disturbed and
the system undergoes a phase transition. The intermittency between stability
and volatility periods on attacker–defender systems can be observed if we
calculate the first derivative of the interteam coordinative variable. The
volatility period characterized when attackers and defenders become mutually
dependent and defined the entry in critical regions (Passos et al. 2009b).
Within these critical regions, some parameters might gain influence over
the others, which moves the attacker–defender system to a performance
outcome. The work of Rodrigues and Passos (2011) described and explained
successful performance outcomes in attacker and defender subunits of rugby
union. This research was developed in situ during rugby union matches. The
first issue under investigation was which were the players under analysis? In
other words, what are the attackers that are functionally coordinated forming
a subunit evolving towards the score line? To solve this issue, the authors
used running correlations (calculated based on players' distance from the goal
line) as a tool that allowed measurement of the degree of coordination among
all the players involved in each attacking situation. Positive values meant that
both players were running in the same direction (presumably towards the goal
line); negative values meant that one player was running towards goal line
but the other was running apart; values close to zero meant weak correlation
between player movement displacement trajectories. The data revealed that
the behaviour of supporting players was strongly correlated with the ball
carrier's behaviour. But correlation values decreased with the proximity to
defenders, which sustains previous results from Passos et al. (2011a). The
coordination patterns among players within an attacker subunit seem to be
disturbed, owing to the influence of task constraints such as the approach of
defenders. Based on the running correlation values, Rodrigues and Passos
also identified three sorts of attacker–defender interactions that led to
different performance outcomes: i) a strong correlation among attackers and a
weak correlation among defenders usually led the attackers to succeed; ii) a
strong correlation among defenders and a weak correlation among attackers
often led the defenders to succeed; iii) when both attackers and defenders
displayed similar levels of correlation, it was not possible to identify an
outcome tendency (Rodrigues and Passos 2011).
The first conclusion from Rodrigues and Passos' research was that the
functional synergies (i.e. subunits of attacking and defending players) that are
formed within rugby union sub-phases of play could be captured through
running correlation analyses (Rodrigues and Passos 2011). Despite this
achievement, the use of running correlations requires some caution. In some
situations, the players could be running laterally in the same direction (for a
brief period of time), with a relatively stable distance to the try line, or one
player could be running laterally and the other running forward; for both
cases we might have weak correlations values. These data do not necessarily
mean that players are not coordinated but only that they are not achieving
symmetrical coordination tendencies (i.e. synchronously moving in the same
direction at the same time). However, because of the nature and laws of the
game, we reinforce that in almost all of the analyzed situations the running
correlations allowed the uncovering of the attacking and defending subunits
and explained their relative success in the match.
The second conclusion was that when a set of attacking players formed a
functional synergy, this could be disturbed by the approach of defenders (i.e.
a decrease in interpersonal distances between attackers and defenders), which
is captured by the oscillation in the running correlation values. This research
sustains the notion that forming functional synergies is an important
advantage in success during team sport performance and the co-adaptive
behaviour that is required by each player led to the emergence of functional
synergies within each subunit of play, which reciprocally bounded player co-
adaptive behaviour. We suggest that the work of Rodrigues and Passos
supports the notion of circular causality of co-adaptive behaviour and
functional synergies, as a general feature of performing in rugby union which
could be tested in other team ball sports (Rodrigues and Passos 2011).
How can these attacking and defending functional synergies (i.e. the game
subunits) break the initial stability featured by their relative positioning in
relation to the try line? Rodrigues and Passos (2011) found that what moves
the attacker–defender system to a phase transition was the increase in relative
velocity. The relative velocity was revealed as a relevant candidate for
control parameter that could explain the attacker subunit's success (Rodrigues
and Passos 2011). This result is consistent with the previous work of Passos
et al. (2008) presented earlier in this chapter. Similar to the previous work in
one-on-one dyads in rugby union, in this study, the relative velocity only
gained influence on the performance outcome within regions of short
interpersonal distance between attackers and defenders. These results
supported once again the argument that important changes in team sport
behaviours evolve within critical regions formed by the emergence of
contextual dependency in players' interactions. It was only ‘inside’ these
regions (i.e. regions of contextual dependency) that the attacker–defender
balance might be broken, presenting an advantage to the attackers. Outside
these regions, the game remains stable and without much change in attacker–
defender structural organization.
In summary, from the studies presented in this section, it is possible to
highlight the relevance of variables such as players' interpersonal distances to
describe collective functional behaviours in team sports. It is based on the
dynamics of interpersonal distances (i.e. the continuous adjustments towards
suitable interpersonal distances) that we sustain the existence of contextual
dependency among players within game sub-phases, which leads to the
emergence of functional synergies based on processes of co-adaptation. This
implies that a set of players often coordinates their behaviours to perform
collectively as a single unit. Also, interpersonal distances poised attacker–
defender systems within critical regions where contextual dependency
governs and is governed by players' interactive behaviours. The circular
causality between co-adaptive behaviour and functional synergies is based on
local informational rules (e.g. interpersonal distances and players relative
velocity) suggesting that attacker–defender interactions are self-organized.

Nonlinear analysis of team-game coordination

Research in intrapersonal coordination (e.g. on dual-limb or hand
movements) has demonstrated that patterns of coordination can be captured
in terms of cycling frequency and phase relations (Haken et al. 1985;
Schmidt et al. 1990; Schmidt 1997). Phase attractors, i.e. stable and strong
patterns of coordination, emerge by a coupling of limb or hand phase
relations and frequencies. To measure such coordination, Haken et al. (1985)
used the relative phase, which was demonstrated to be acting as a collective
variable capturing the spatiotemporal relations between two oscillating agents
by considering their amplitudes and frequencies (Oullier and Kelso 2009). An
oscillating agent can be considered to be a cyclical behaviour that occurs in
space and time such as a pendulum, the body sway or the heart rate measured
over time. The phase angle corresponds to the amount of time in which the
signal of the two oscillating agents is delayed or shifted (Rosenblum et al.
2001). For example, if two fingers or two players are moving to the left and
to the right at the same time, there is no delay registered in the coordination
of the agents and 0o of phase angle (i.e. an in-phase mode of coordination)
may be observed. On the contrary, if one finger or player is moving to the left
and the other is moving to the right at the same time, the phase angle values
display a delay of 180o (i.e. an anti-phase mode of coordination).
Intermediate values, such as –60o, indicates that the second oscillating agent
is leading the relation by onesixth of a cycle. The advantage of using relative
phase to capture nonlinear behaviours in comparison with other linear
measures (e.g. running correlations) is that relative phase measures the
amplitude and frequency of two signals to calculate the phase relation whilst
linear measures just consider the frequency of the signals. Disregarding the
amplitude of oscillation may mask the dynamical coupling of agents
(Rosenblum et al. 2001), which may make the complete analysis of the time-
evolving dynamics of a system difficult.
More recently, relative phase applications to intrapersonal coordination
have been extended to interpersonal coordination, namely in sports.
Considering that sport contests tend to exhibit forward–backward and left–
right behaviours (i.e. oscillating behaviours on the field) McGarry et al.
(2002) proposed that the emergent patterns of behaviour between players or
teams could also be captured using relative phase. This would allow
researchers to explain how the interaction between players constrains the
emergence of patterns of stability (i.e. high attraction to few coordination
modes between performers), variability (i.e. high attraction to larger number
of coordination modes between performers) and symmetry breaking in
systems organizational states (i.e. when new patterns of coordination emerge
during performance; Vilar et al. 2012a). This is a relevant issue that sport
scientists and coaches need to pursue and understand in analyses of team-
game performances (Araújo et al. 2006; Davids et al. 1994; Handford et al.
1997).
The first research in sport contests using relative phase analyses was
conducted in tennis and squash (Lames 2006; McGarry et al. 1999; Palut and
Zanone 2005). These studies showed that relative phase is a pertinent
collective variable that captured different modes of coordination between
players over time. In addition, the authors suggested self-organization as the
process through which the coordinated behaviours between players emerged.
Following the suggestion of McGarry et al. (2002) that game dynamics might
emerge from the coupling within players of the same team (i.e. intrateam
relations) and between players of opposite teams (i.e. interteam relations),
research was extended to team sports settings, such as basketball and futsal.
With the goal of understanding the intraand interteam couplings of players in
basketball, two studies were conducted by Bourbousson et al. (2010a,b).
Considering the interteam relations (i.e. coordination between all possible
attacking–defending dyads), data revealed strong in-phase attractions in the
longitudinal (basket-to-basket) displacements, especially between direct
opponents for attacker–defender relations (e.g. right wing of team A and left
wing of team B). In futsal (the five-a-side indoor association football)
Travassos et al. (2011) used relative phase to examine not only intra- and
interteam player relations but also the relation that each player developed
with ball trajectory. Results were expected to highlight how the location of
the ball is an important constraint on game behaviour, since players seek to
gain ball possession to score goals and prevent the opponents from doing the
same. The movement trajectories of the ball and players were recorded in
both lateral and longitudinal directions and investigated using relative phase
as a collective variable. Travassos et al. (2011) observed stronger in-phase
attractions between players and ball (specifically between defenders and ball
with a lag of 30o) than just between players. A lag of 30o means that the ball
led the spatiotemporal relation with players by one-twelfth of a cycle,
suggesting that players adjusted their positions according to and just after the
movement of the ball. Indeed, this finding demonstrates that ball dynamics
are an important constraint on behaviour that needs to be further considered
to analyze performance in team games. The analysis of intrateam dyadic
relations revealed stronger coordination patterns between defenders than
between attackers. Therefore, it was suggested that defenders tried to couple
their displacements in relation to the ball closing the paths to the goal, whilst
attackers explored high-variable spatiotemporal relations among each other to
disrupt the defensive structure. For interteam couplings, stronger inphase
modes of coordination were observed between the attacker and the closest
direct opponent. Similarly to a previous study of basketball (Bourbousson et
al. 2010a), these results demonstrated different modes of coordination for
different phases of the game (i.e. while attacking or defending). Finally, the
authors reported stronger in-phase modes of coordination for lateral
displacements than for longitudinal ones. These findings are opposite to the
results obtained by Bourbousson et al. (2010a) in basketball and may be
explained by the differences on the task constraints that bound the playing
conditions (five-versus-five in basketball and five-versus-four plus
goalkeeper in futsal). The numerical advantage of the attacking team in the
futsal task is suggested as having constrained the defending team to use a
zone-marking strategy, whilst players’ numerical equivalence in basketball
afforded the possibility of using a ‘man-to-man’ marking strategy. The
numerical relation between players was shown to constrain players’
interactions and the strength of coupling between them. Changing the
numerical relations is then proposed to be a major task constraint that
coaches may manipulate during practice tasks to promote specific adaptations
of interpersonal relations sustaining the use of zone and man-to-man
defensive playing strategies. In addition, in a second study also conducted in
futsal (Travassos et al. 2012) examined interpersonal coordination between
teams. The authors advanced the study of Bourbousson et al. (2010a) by
measuring the phase relations between the geometrical centre of each team
with regard to its location on the field of play referenced to the goal position.
Measuring variations in the angle between the geometrical centre of each
team and the goal position, Travassos et al. (2012) observed higher
tendencies towards specific modes of coordination than when using lateral
and longitudinal displacements (Figure 6.2). These results demonstrated
game behaviours anchored on the goal location. For example, the defending
team seeks to close the paths to the goal in relation to changes in ball
positioning, attacking player behaviours and goal location and not just based
on abstract lateral and longitudinal displacements. Measuring team relations
with regard to their location on the field of play referenced to the goal
position takes into account the key goal of the game and allows a functional
understanding of emergent patterns of coordination.
Also in futsal, Vilar et al. (2012b) investigated how the goal and ball
locations constrain interpersonal patterns of coordination between inter-team
(i.e. attackerdefender) dyads. The authors sought to capture dyadic relations
using relative phase analyses between the distances and the angles of each
player to the ball and to the centre of the goal. The distances captured the
dynamics of proximity between players to the ball or to the goal while the
angle measured the alignment between them. Vilar et al. (2012b) reported
predominant in-phase modes of coordination between attackers and defenders
for angles as well as for distances regarding the ball and the goal. However,
stronger in-phase modes of coordination were reported between attacker and
defenders when related to the goal than between players and the ball. These
data revealed that the location of the goal exerts a larger constraining
influence on dyadic coordination than the location of the ball. Finally, when a
goal was scored similar distances between attackers and defenders to the goal
were observed and instability in the players' angular relations to the goal was
registered, suggesting that the angular relations between attackers and
defenders may be candidates for control parameters that move this dyadic
system to a phase transition. At this point, the attacker and defender stability
was broken, emerging an advantageous situation for the attacker to kick at
goal without the defender being able to intercept its trajectory.
Figure 6.2 Angle between geometrical centre of each team and the goal position

Summarizing, relative phase was showed to capture attacker and defender

interpersonal interactions relative to locations of the goal and the ball.
Understanding the spatiotemporal relations among players and key
performance constraints that bound players' dyadic and collective behaviours
within competitive settings provides coaches with updated information for
manipulating task constraints, allowing them to create representative practice
designs to promote functional adaptations in players' performances.

Social networks as tools for describing player interactions

How teams perform collectively in competition is an issue that is highly
topical in sports sciences. Practitioners have been analyzing performance
with the goal of providing information that may improve team performance
and to create awareness among players and coaches of how individual players
can influence team patterns (McGarry 2009). One of such methods recently
proposed for analyzing performance in team ball sports is social networks
analysis (Duch et al. 2010; Passos et al. 2011b). These tools consider teams
as complex social systems, recognizing their ‘degeneracy’ (i.e. inherent
adaptive flexibility in the way in which teams achieve successful
performance outcomes; Edelman and Gally 2001). Next, we introduce the
instruments for network analysis supported on small-world networks and by
graph theory and discuss their application in the investigation of interpersonal
relations in team sports.
In team sports, the structure of the web of interpersonal interactions
between players constrains the organization and functioning of teams. Players
coordinate their actions to achieve patterns of collective behaviour that allow
them to satisfy game demands. Often, two players can become interconnected
for performing through a path of only a few passes between few players. This
allows different sets of players to become linked to form a subunit in a team
to perform successfully. The myriad of interactions that emerge among team
players during competitive performance might lead to the emergence of
distinct but functional equivalent patterns of play (Passos et al. 2011b).
Therefore, team sports may be conceptualized as a small-world social system,
in which system behaviour might evolve from the interpersonal interactions
among system agents (Barabasi and Oltvai 2004). The term small-world
effect was originally observed in studies of collective networks. A small-
world network is a connected simple graph in which each node is linked to a
relatively well-connected set of neighbouring nodes with short-cut
connections between some nodes (Watts and Strogatz 1998).
Recently, the small-world effect was shown to capture the rich interactions
among players in team sports (Duch et al. 2010; Passos et al. 2011b). One of
the main tools that have been used for the quantitative study of networks is
graph theory. Its concepts have been applied to discuss network issues in
complex social (e.g. traffic jams), biological (e.g. cells, schools of fish, flocks
of pigeons) or communication systems (e.g. the worldwide web). A graph
consists of a set of nodes (or vertices) together with a set of links (or edges)
that connect various pairs of nodes (Batten 2000). Nodes may have states and
links may have directions and weights. For example, recent research in water
polo (Passos et al. 2011b) examined the intrateam pattern-forming in
attacking sub-phases of water polo (i.e. from the moment a team gained ball
possession to the moment that ball possession was recovered by the
opponent). Each node corresponded to each one of the six offensive players
and two linkage levels were established: (i) when a player passed the ball to a
teammate; or (ii) when players changed position in the performance area due
to displacement of a teammate.
There are different ways to structure data relative to a graph in a computer
system. For example, an adjacency list consists on a list of links whose
element i →j shows a link going from node i to node j. Passos et al. (2011b)
build an adjacency matrix for each unit of attack to identify the proximity of
interacting players. The adjacency matrix is used to build a finite n × n
network where the entries represent the linkages between players (e.g. when
player A passes the ball to player B). Finally, graphs are frequently drawn as
node-link diagrams in which the vertices are represented as disks and the
edges are represented as line segments (Di Battista et al. 1994). Graphs may
be displayed as different shapes (complete graph, regular graphs, planar
graph, directed graph, weighted graph, etc.), depending on the nature of
problem that the graph is intended to illustrate. In the exemplar research in
water polo, the authors used a combination between a weighted graph, with
an associated number indicating the strength of the edge, and a directed
graph, displaying edges that were directional (Figure 6.3). The direction of
the arrows indicates the pass direction, i.e. the origin of the arrow represents
the player who passed the ball and the arrowhead represents the player who
received the ball. The width of the black arrows denoted the quantity of
passes from one player to another during performance (i.e. thicker arrows
illustrated more passes occurring between specific players and thinner arrows
represented fewer passes taking place among players; Passos et al. 2011b).
Figure 6.3 Exemplar data of a network in water polo (reprinted with permission of the Journal of
Science and Medicine in Sport)

One of the main potentials of network analysis tools is to provide

understanding of the strengths and weaknesses in specific performance
situations in a range of sports. By using different measurements of the
topological properties of networks in several ‘units of attack’, researchers
may describe the structural organization of the attacking performance of
players and teams (Passos et al. 2011b). For example, the order of the graph
informs about the number of nodes (i.e. players) acting on the network, whilst
the number of links (e.g. passes performed between players) is referred to as
the size of the graph. This analysis may inform about the number of different
players typically engaging in the offensive plays, as well how many passes do
teams perform until loosing or shooting the ball on goal. This is expected to
shed some light on the style of play adopted (e.g. ‘direct play’ or ‘possession
play’), that is, if teams seek to use shorter or longer passing sequences as a
means of scoring goals.
Using the same type of approach, Duch et al. (2010) analyzed the
performance of players in the association football European Cup 2008. The
representation of the ‘flow network’ of successful passing balls in different
matches, allowed comparison of the performance of the two teams and
identification of the most relevant players (i.e. the players who successfully
passed and received more balls to and from different teammates) of each
team when different team strategies were used. The most common analysis in
networks framework concerns the centrality measures of each node. For
example, the degree centrality of node n is the number of links connected to
n; that is, how many connections the player has. However, in a directed
graph, such as the ones typically considered in team sports, the in-degree of
each node corresponds to the number of nodes n’ of the system that are
directly linked to n by an inward-pointing link to n. Conversely, the number
of nodes n’ of the system that are directly linked to n by a link pointing
outward from n is called out-degree of n. These data may inform practitioners
about how many passes each player receives and how often he passes the ball
efficiently. Other different measures of centrality may also be considered,
such as: (i) the betweenness centrality, which refers to how many shortest
paths go through each player; and (ii) the closeness centrality, which informs
about how close the player is to other players. This measurement can be a
very useful way of accurately identifying the key ‘decision makers’ or
‘emergent leaders’ during important phases of competitive performance.
Similarly, the degree distribution gives a rough profile of how the
connectivity is distributed within the team, by displaying the total number of
players with degree k.
The connectivity of the network may be also a relevant topological
measure of a team, informing about the degree to which players interact with
each other (Batten 2000). A connectivity measure would generally take into
account two aspects of this interaction pattern: (i) which players are
interacting with one another; and (ii) what is the strength (frequency,
regularity, impact, etc.) of these interactions. For example, the shortest path
length (or distance) between node n and n’ refers to the minimum number of
links that must be traversed to travel from a node n to another node n’. This
may inform about how many passes typically occur for the ball to arrive from
a given player to a different one. Moreover, this technique may also provide a
collective measurement of team closeness. The characteristic path length of a
network is the median of the means of the shortest path lengths connecting
each node to all other nodes (Watts 1999).
Finally, the clustering coefficient is also suggested as a relevant
topological property that measures the degree to which players in a team
tended to be linked. To calculate the clustering coefficient of a node n, one
should first consider the number of nodes in the neighbourhood (k); that is,
how many nodes n’ are linked to n. Since, in team sports, the direction of the
pass is important (directed graph), for each neighbourhood there are k(k–1)
links that could exist among the vertices within the neighbourhood (k is the
total [in plus out] degree of the vertex). Thus, the clustering coefficient of a
given node is equal to the proportion of links between the nodes within its
neighbourhood divided by the number of links that could possibly exist
between them. The clustering coefficient of a specific node may inform
practitioners about subgroups of players who may coordinate their actions
very well among each other (e.g. left back, central midfielder and left forward
in association football). In addition, one may also calculate the clustering
coefficient of the network by averaging the clustering coefficient of all nodes
in a network. A large clustering coefficient implies that the team is well
connected locally to form a cluster.
In summary, we suggest that networks tools are able to provide
understanding about how players (i.e. nodes) may interact to form a team (i.e.
network). They can explain how the structure and specific topology of a team
might constrain the emergence of collective behaviours in team sports, which
can be used in tactical and strategic decision making for competition. The
possibility of plotting a pattern of play with an observable network structure
and topology allows us to identify the players engaged in more and less
frequent interactions within a team and to compare outcomes of successful
and less successful patterns of play during performance.
Concluding remarks and applications
This chapter has presented an overview of the conceptual and methodological
developments in the recent study of interpersonal coordination in team sports.
From the analysis of dyadic system interactions to the understanding of group
behaviours in subunits of play, interactions within sport teams seem to
assemble into functional synergies. The emergence of these functional
grouping tendencies are governed by locally generated information sources
from the relative positioning of other team players, motion directions and
changes in motion, which attest the significant meaning of inter- and
intrateam co-adaptation processes (Passos et al. 2011b).
The findings presented in this chapter might have some important
implications concerning performance analysis, as well as for learning and
training design. For example, social networks can help performance analysts
to enhance their analyses using simple notation data, such as passing actions
and switching positions. Moreover, other coordinative variables discussed in
this chapter can be used to evaluate the interactive performance behaviours of
team players in different subphases and levels of analysis. The recent
technological advances in player tracking systems, such as electronic portable
devices (Carling et al. 2008) and multi-player video-based systems (Barris
and Button 2008) offer a novel opportunity to develop evaluations of the
coordination tendencies emerging within sport teams during performance
(competition and training settings). For example: is the distance between the
three midfielders of a football team presenting similar values during
competition and practice sessions? The variables and measurement tools
presented here might support the innovative approaches to performance
analysis, which allow answering this type of relevant questions about
performance.
Regarding learning and training design, by changing the local available
information that guides player behaviour, coaches can manipulate the local
rules governing interactions between neighbour teammates and opponents,
inducing the emergence of new patterns of collective movement solutions.
For example, increasing the regular ratio of area per player could create
larger areas of influence which promote the need for players to fine tune their
positioning and their interpersonal distances to efficiently cover the spaces
during the defensive phases. A constraints-led approach has been suggested
as a practical and meaningful tool to promote the emergence of movement
patterns under constraints (Davids et al. 2008; Glazier 2010) which can be
used to enhance learning and performance.
References
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7: 653–76.
Araújo, D., K., D., Sainhas, J. and Fernandes, O. (2002) Emergent decision-
making in sport: a constraints-led approach, in L. Toussaint and P.
Boulinguez (eds) International Congress on ‘Movement, Attention and
Perception’. Poitiers, France: Universite de Poitiers, p. 77.
Barabasi, A. L. and Oltvai, Z. N. (2004) Network biology: understanding the
cell's functional organization. Nature Reviews Genetics, 5 (2): 101–13.
Barris, S. and Button, C. (2008) A review of vision-based motion analysis in
sport. Sports Medicine, 38 (12): 1025–43.
Batten, D. (2000) Sheep, explorers, and phase transitions, in D. Batten (ed.)
Discovering Artificial Economics: How Agents Learn and Economies Evolve.
Boulder, CO: Westview Press, pp. 81–115.
Bourbousson, J., Sève, C. and McGarry, T. (2010a) Space-time coordination
patterns in basketball: Part 1 Intra- and inter-couplings amongst player dyads.
Journal of Sport Sciences, 28 (3): 339–47.
Bourbousson, J., Sève, C. and McGarry, T. (2010b) Space-time coordination
patterns in basketball: Part 2 Investigating the interaction between the two
teams. Journal of Sport Sciences, 28 (3): 349–58.
Carling, C., Bloomfield, J., Nelsen, L. and Reilly, T. (2008) The role of
motion analysis in elite soccer: contemporary performance measurement
techniques and work rate data. Sports Medicine, 38 (10): 839–62.
Cordovil, R., Araújo, D., Davids, K., Gouveia, L., Barreiros, J., Fernandes,
O. and Serpa, S. (2009) The influence of instructions and body-scaling as
constraints on decisionmaking processes in team sports. European Journal of
Sport Science, 9 (3): 169–79.
Correia, V., Araújo, A., Duarte, R., Travassos, B., Passos, P. and Davids, K.
(2011) Changes in practice task constraints shape decision-making
behaviours of team games players. Journal of Science and Medicine in Sport,
15 (3): 244–9.
Davids, K., Button, C. and Bennett, S. (2008) Dynamics of Skill Acquisition.
A Constraints-led Approach. Champaign: Human Kinetics.
Davids, K., Handford, C. and Williams, M. (1994) The natural physical
alternative to cognitive theories of motor behaviour: an invitation for
interdisciplinary research in sports science? Journal of Sports Sciences, 12
(6): 495–528.
Di Battista, G., Eades, P., Tamassia, R. and Tollis, I. G. (1994) Algorithms
for drawing graphs: an annotated bibliography. Computational Geometry:
Theory and Applications, 4 (5): 235–82.
Duarte, R., Araújo, D., Correia, V. and Davids, K. (2012) Sport teams as
superorganisms: implications of sociobiological models of behaviour for
research and practice in team sports performance analysis. Sports Medicine,
42 (8): 1–10.
Duch, J., Waitzman, J. S. and Nunes Amaral, L. A. (2010) Quantifying the
performance of individual players in a team activity. PloS One, 5 (6):
e10937; doi:10.1371/journal.pone.0010937
Edelman, G. M. and Gally, J. A. (2001) Degeneracy and complexity in
biological systems. Procedings of the National Academy of Sciences of the U
S A, 98 (24): 13763–8.
Frencken, W., Lemmink, K., Delleman, N. and Visscher, C. (2011)
Oscillations of centroid position and surface area of soccer teams in small-
sided games. European Journal of Sport Science, 11: 215–23.
Glazier, P. S. (2010) Game, set and match? Substantive issues and future
directions in performance analysis. Sports Medicine, 40 (8): 625–34.
Haken, H., Kelso, J. A. S. and Bunz, H. (1985) A theoretical-model of phase-
transitions in human hand movements. Biological Cybernetics, 51 (5): 347–
56.
Handford, C., Davids, K., Bennett, S. and Button, C. (1997) Skill acquisition
in sport: some applications of an evolving practice ecology. Journal of Sports
Sciences, 15: 621–40.
Headrick, J., Davids, K., Renshaw, I., Araújo, D., Passos, P. and Fernandes,
O. (2012) Proximity-to-goal as a constraint on patterns of behaviour in
attacker-defender dyads in team games. Journal of Sport Sciences, 30 (3):
247–53.
Jensen, H. (1998) Self-Organized Criticality: Emergent Complex Behavior in
Physical and Biological Systems. Cambridge Lecture Notes in Physics, Vol.
10. Cambridge: Cambridge University Press.
Kauffman, S. (1993) The Origins of Order: Self-organization and Selection
in Evolution. New York: Oxford University Press.
Kelso, S. (2009) Coordination dynamics, in R. A. Meyers (ed.) Encyclopedia
of Complexity and System Science. Heidelberg: Springer, pp. 1537–64.
Lames, M. (2006) Modelling the interaction in game sports: relative phase
and moving correlations. Journal of Sports Science and Medicine, 5: 556–60.
Lames, M., Erdmann, J. and Walter, F. (2010) Oscillations in football: order
and disorder in spatial interactions between the two teams. International
Journal of Sport Psychology, 41 (4): 85.
McGarry, T. (2009) Applied and theoretical perspectives of performance
analysis in sport: scientific issues and challenges. International Journal of
Performance Analysis in Sport, 9: 128–40.
McGarry, T., Anderson, D. I., Wallace, S. A., Hughes, M. D. and Franks, I.
M. (2002) Sport competition as a dynamical self-organizing system. Journal
of Sports Sciences, 20 (10): 771–81.
McGarry, T., Khan, M. A. and Franks, I. M. (1999) On the presence and
absence of behavioural traits in sport: an example from championship squash
match-play. Journal of Sports Sciences, 17 (4): 297–311.
Milho, J., Passos, P., Leandro, H., Borges, J., Araújo, D. and Davids, K.
(2010) Collective decision making inter and intra-team analysis: new
challenges to match analysis. International Journal of Sport Psychology, 41:
95–6.
Oullier, O. and Kelso, J. A. S. (2009) Coordination from the perspective of
social coordination dynamics, in R. A. Meyers (ed.) The Encyclopedia of
Complexity and Systems Science. Heidelberg: Springer, pp. 8198–213.
Palut, Y. and Zanone, P. (2005) A dynamical analysis of tennis: concepts and
data. Journal of Sports Sciences, 23 (10): 1021–32.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Milho, J., and Serpa, S.
(2008) Information-governing dynamics of attacker–defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Passos, P., Araújo, D., Davids, K., Gouveia, L. and Serpa, S. (2006)
Interpersonal dynamics in sport: the role of artificial neural networks and 3-D
analysis. Behavior Research Methods, 38 (4): 683–91.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Serpa, S., Milho, J. and
Fonseca, S. (2009a) Interpersonal pattern dynamics and adaptive behavior in
multiagent neurobiological systems: conceptual model and data. Journal of
Motor Behavior, 41 (5): 445–59.
Passos, P., Araújo, D., Davids, K. W., Milho, J. and Gouveia, L. (2009b)
Power law distributions in pattern dynamics of attacker-defender dyads in
rugby union: phenomena in a region of self-organized criticality? Emergence:
Complexity and Organization, 11 (2): 37–45.
Passos, P., Milho, J., Fonseca, S., Borges, J., Araújo, D. and Davids, K.
(2011a) Interpersonal distance regulates functional grouping tendencies of
agents in team sports. Journal of Motor Behavior, 43 (2): 155–63.
Passos, P., Davids, K., Araújo, D., Paz, N., Mingue.ns, J. and Mendes, J.
(2011b) Networks as a novel tool for studying team ball sports as complex
social systems. Journal of Science and Medicine in Sport, 14 (2): 170–6.
Rodrigues, M. and Passos, P. (2011) Padroes de coordenacao interpessoal no
rugby: analise de comportamentos colectivos em jogo. Paper presented at the
XII Jornadas da Sociedade Portuguesa de Psicologia do Desporto, Portimao,
Portugal.
Rosenblum, M., Pikovsky, A., Kurths, J., Schafer, C. and Tass, P. A. (2001)
Phase synchronization: from theory to data analysis, in F. Moss and S. Gielen
(eds) Handbook of Biological Physics, Volume 4, Neuro-informatics.
Amsterdam: Elsevier Science, pp. 279–321.
Schmidt, R., Carello, C. and Turvey, M. (1990) Phase transitions and critical
fluctuations in the visual coordination of rhythmic movements between
people. Journal of Experimental Psychology: Human Perception and
Performance, 16 (2): 227–47.
Schmidt, R. C. (1997) Evaluating the dynamics of unintended interpersonal
coordination. Ecological Psychology, 9 (3): 189–206.
Sumpter, D. J. (2010) Collective Animal Behavior. Princeton, NJ: Princeton
University Press. Travassos, B., Araújo, D., Vilar, L., and McGarry, T.
(2011) Interpersonal coordination and ball dynamics in futsal (indoor
football). Human Movement Science, 30: 1245–59.
Travassos, B., Araújo, D., Duarte, R. and McGarry, T. (2012) Spatiotemporal
coordination patterns in futsal (indoor football) are guided by informational
game constraints. Human Movement Science, 31 (4):932–45.
Vilar, L., Araújo, D., Davids, K. and Button, C. (2012a) The role of
ecological dynamics in analysising performance in team sports. Sports
Medicine, 42(1): 1–10.
Vilar, L., Araújo, D., Davids, K. and Travassos, B. (2012b) Constraints on
competitive performance of attacker–defender dyads in team sports. Journal
of Sport Sciences, 30 (5): 459–69.
Watts, D. (1999) Networks, Dynamics and the Small-world Phenomenon.
American Journal of Sociology, 105, (2): 493–527.
Watts, D. and Strogatz, S. (1998) Collective dynamics of ‘small-world’
networks. Nature, 393 (6684): 440–2.
7 The measurement of space and
time in evolving sport
phenomena
Sofia Fonseca, Ana Diniz and Duarte
Araújo
This chapter provides an introduction to methods for analyzing both space
and time in behavioural data. An overview of methods for analyzing time
and, in particular, some methodologies in the so-called time and frequency
domains are presented to characterize dynamical behaviours across time.
Some important functions, such as the autocorrelation function and the
spectral density function, are presented and some illustrations are made. We
also present and describe methods for extracting, from a system of interacting
agents, spatial time-series data capable of describing underlying dynamic
system behaviours. Particular attention is given to applications in team sports.
Results from recent studies are considered to illustrate the models presented.
Efforts at player motion capture have traditionally involved a range of data
collection techniques, from live observation to post-event video analysis,
where players' movement patterns are manually recorded and categorized to
determine performance effectiveness. A variety of systems and methods have
been employed to analyze the motion of athletes during sports, where the
movements vary in duration, field position and surface, speed, acceleration,
direction technique and tactics (Barris and Button 2008).
Indeed, a large number of variables can be considered to describe and
classify individual and collective performance, from physical to
psychological and physiological markers. Of particular relevance to team ball
sports are spatial markers and, more precisely, information on the spatial
organization (i.e. space management) of the players, as members of a team
and as opponents. Moreover, sport science research suggests that
performance data in sports needs to be evaluated in a continuous manner,
given the variability inherent in such multivariate processes, which implies a
permanent update and adjustment of variables over time (Davids et al. 2003).
However, the classical methods of analysis are based on descriptive statistics,
such as the mean and the standard deviation, and tend to ignore the
dimension of time. In contrast, time-series methods, in the so-called time and
frequency domains, focus on the dynamical behaviours across time and allow
for modelling and inference.
Time series analysis methods
A time series is a sequence of observations, typically measured at successive
time instants spaced at uniform time intervals. Formally, a time series is often
denoted by {Y1,…,Yn}, where n is the length of the series. Figure 7.1 displays
a time series referent to the distance, in metres, from a given sailing boat to a
fixed optimal starting position, before the start of the regatta, recorded at 25
Hz, i.e. every 0.04 seconds (Araújo 2006). This time series has a length of n
= 579 points and duration of 23.12 seconds.
To assess the fundamental properties of a time series, it is assumed that
they are realizations of stochastic processes. A stochastic process is one that
evolves over time and whose evolution at each time step is governed, at least
in part, by probability. More precisely, a stochastic process is a family of
random variables {Yt, t ∈ T}, defined on a probability space. One can use the
term stochastic to refer to a behaviour that is influenced, to some degree, by
both deterministic and random processes. In the sailing example, the position
of the sailing boat relative to the optimal starting position may be seen as a
realization of a stochastic process, since the position of the boat at each time
is dependent on its previous positions with a given probability.
For some situations but not all, time series have a kind of stability that can
reasonably be modelled by stationary processes (see Chapter 5). A stationary
process is a stochastic process whose joint probability distribution does not
change when shifted in time or space. Consequently, parameters such as the
mean and the variance, if they exist, do not change over time or position. In
the sailing example, the distance from the sailing boat to the optimal starting
position seems to fluctuate slowly around a nearly constant or slightly
decreasing mean level and seems to have a cyclical component. This suggests
that the sailing boat approaches and departs from the fixed starting position in
a roughly cyclical manner.
When the mentioned characteristics (e.g. stability) are known, time series
can be studied in two domains of analysis, the time domain and the frequency
domain. These domains are presented in the next section and illustrated with
real and simulated time series.
Figure 7.1 Exemplar data of the distance from a sailing boat to a fixed starting position before the start
of the regatta; the length of this time series is 579 data points

Time domain analysis

In time domain analysis, the main goal is the evaluation of the behaviour of
the process under study over time. In this domain, two central concepts are
the mean and the autocorrelation function of the process. The latter function
gives the correlation value of the process at different times. Formally, and in
a wide sense, a stochastic process is said to be stationary if its mean is
constant across time, its variance is constant across time and its
autocorrelation function depends only on the time lag. Figure 7.2 illustrates
realizations of length n = 200 of a stationary process and of two non-
stationary processes. Note that the first time series (top panel) has a constant
mean and a constant variance, the second has an increasing trend (middle
panel) and the third has a growing variability (bottom panel).
Figure 7.2 Simulated time series of length n = 200 of a stationary process (top), a nonstationary process
with non-constant mean (middle) and a non-stationary process with non-constant variance (bottom)

When in the presence of non-stationary time series, the observed data

points can usually be transformed into realizations of stationary processes by
statistical techniques, such as differencing and others (e.g. Brockwell and
Davis 1991, but see Chapter 5).
If the process {Yt, t ∈ T} is stationary, then its autocorrelation function
Ρ(.) is defined as:
Ρ(k) = Corr(Yt,Yt+k), for all t,k.
The function Ρ(.) has some important properties, such as:
−1 ≤ Ρ(k) ≤ 1, for all k,
Ρ(−k) = Ρ(k), for all k.
The autocorrelation function Ρ(.) measures the strength of the statistical
dependence of the process at each time lag. A value of Ρ(.) close to 1 at a
given lag indicates strong positive correlation, while a value of Ρ(.) close to
−1 at a given lag indicates strong negative correlation (or anti-correlation).
Given a time series {Y1,…,Yn}, an usual estimator of the autocorrelation
function Ρ(.) is the sample autocorrelation function (.) defined by:

Notice that the sample autocorrelation function (.) can be computed for any
time series, even for series that are not realizations of stationary processes.
Some works suggest that these sample autocorrelations (.) should only be
used if the series length n satisfies n ≥ 50 and the time lag k satisfies k ≤ n/4.
The simplest kind of stationary process is the white noise process. A
stochastic process {Yt, t∈T} is said to be white noise if the random variables
have constant mean, usually equal to zero, constant variance and are
uncorrelated (see also Chapter 5). This implies that the autocorrelation
function Ρ(.) is zero everywhere, except at k = 0, and therefore is defined as:

Given a time series {Y1,…,Yn}, if the series is a realization of a white noise,

then the sample autocorrelations (.) should be relatively close to zero. More
precisely, at a significance level of 5%, the sample autocorrelations (.)
should lie between the critical bounds ± 1.960/√n. In contrast, one or more
high values of the sample autocorrelations outside the critical bounds
suggests that the time series is not a realization of a white noise.
Figure 7.3 illustrates a time series {Y1,…,Yn} of length n = 200 generated
from a white noise with mean zero and variance one. It also shows the sample
autocorrelation function (.) at lags k = 0, …, 50 with the bounds ± 1.960/
√200 = ±0.139.

Figure 7.3 Simulated time series of length n = 200 of a white noise (top) and respective sample
autocorrelation function at lags k = 0, …, 50 (bottom) with critical bounds (dashed lines)

Note that the time series fluctuates randomly around the mean and the
sample autocorrelations are very close to zero lying between the critical
bounds.
With respect to the sailing example shown earlier in this chapter, Figure
7.4 presents the time series {Y1,…,Yn} of length n = 579 relative to the
distance from a sailing boat to a starting position. It also provides the sample
autocorrelation function (.) at lags k = 0, …, 350 with the bounds ± 1.960/
√579 = ±0.081.
 Observe that the time series has a cyclical feature and the sample
autocorrelations also exhibit a cyclical pattern lying outside the critical
bounds. This implies that the sample autocorrelations are significant and take
positive values at certain lags and negative values at other lags. Although it is
unadvisable to compute the sample autocorrelations at lags larger than n/4,
some of these values are represented here merely to show their periodicity.

Frequency domain analysis

In frequency domain analysis, the main objective is the decomposition of the
process under study into a sum of sinusoidal components with fixed
frequencies. A key concept is the spectral density function of the process.
This function gives the amount of variance accounted for by each frequency
in the process and corresponds mathematically to the Fourier transform of the
autocorrelation function. The Fourier transform is a mathematical operation
that decomposes a function into its constituent frequencies. The spectral
density function allows for identifying dominant frequencies in the process
that may be associated to hidden periodicities. The frequency is the number
of occurrences of a repeating process per unit of time and the SI unit for
frequency is the Hertz (or cycles per second). Reciprocally, the period is the
duration of one cycle in a repeating process and the SI unit for period is the
second. The time domain analysis and the frequency domain analysis of time
series are complementary and contribute, under different perspectives, to a
better understanding of the observed phenomena.
Figure 7.4 Exemplar data of length n = 579 of the distance from a sailing boat to a fixed starting
position before the regatta starting (top) and respective sample autocorrelation function at lags k = 0,
…, 350 (bottom) with critical bounds (dashed lines)

If the process {Yt, t∈T} is stationary, then its spectral density function f(.)
has some important properties, such as:
f(λ) ≥ 0, for all λ,
f(−λ) = f(λ), for all λ,.
The spectral density function f(.) measures the intensity of each frequency in
the process over a range of frequencies. The larger the value of f(.) at a given
frequency, the larger the intensity (or power) of that frequency.
Given a time series {Y1,…,Yn}, the usual estimator of the spectral density
function f(.) is the normalized periodogram In(.) defined by:

The normalized periodogram In(.) is the modulus-squared of the discrete

Fourier transform of the time series with a simple normalization and the λj
are the frequencies.
If the process {Yt, t∈T} is a white noise with mean zero and variance σ2,
then its spectral density function f(.) is constant (flat) and is defined as:

Given a time series {Y1,…,Yn}, if the series is a realization of a white noise,

then the periodogram values In(.) should be reasonably similar at all
frequencies. In contrast, a high peak in the periodogram ordinates at a given
frequency suggests that the time series is not a realization of a white noise
and there is a preferred frequency. The significance of the largest peak in the
periodogram ordinates can be tested using the Fisher's test (e.g. Brockwell
and Davis 1991).
Figure 7.5 illustrates a time series {Y1,…,Yn} of length n = 200 generated
from a white noise with mean zero and variance one. It also shows the
normalized periodogram In(.) at frequencies λr = 2πr with r = 0, …, 0.5. Note
that the time series fluctuates randomly around the mean and the periodogram
values are reasonably similar at all frequencies.
Figure 7.5 Simulated time series of length n = 200 of a white noise (top) and respective normalized
periodogram at frequencies λr = 2πr with r = 0, …, 0.5 (bottom)

For the sailing example, Figure 7.6 presents the time series {Y1,…,Yn}of
length n = 579 relative to the distance from a sailing boat to a starting
position. It also provides the normalized periodogram In(.) at frequencies λr =
2πr with r = 0, …, 0.5. Observe that the time series has a cyclical feature and
the periodogram ordinates exhibit a high peak at a given frequency. The peak
is significant and occurs at the frequency 0.00518, which implies a period of
1/0.00518 = 193. This means that the series has indeed a cyclical pattern,
repeating itself roughly every 193 time units, i.e. every 7.72 seconds.
Figure 7.6 Exemplar data of length n = 579 of the distance from a sailing boat to a fixed starting
position before the regatta starting (top) and respective normalized periodogram at frequencies λr = 2πr
with r = 0, …, 0.5 (bottom)

In a stationary process, the autocorrelation function depends only on the

time lag. In this case, the range of the correlation (or stochastic memory) of
the process can be defined as the speed of the decay of the autocorrelation
function.
Formally, a stationary process {Yt, t∈T} is said to have short-range
correlation (or short memory) if its autocorrelation function Ρ(.) satisfies the
relation:
|Ρ(k)| ≤ c rk, k = 0, 1, …,
where c and r are two constants such that c > 0 and 0 < r < 1 and k is the lag.
This means that the function Ρ(.) decays to zero fast with a geometrically
bounded decay.
Figure 7.7 represents a realization {Y1, …,Yn} of length n = 200 of a
shortrange correlation process. It also shows the sample autocorrelation
function (.) at lags k = 0, …, 50 and the normalized periodogram In(.) at
frequencies λr = 2πr with r = 0, …, 0.5. It is clear that the time series
fluctuates rapidly around the mean, the sample autocorrelation function has a
large and negative value at lag one with non-significant values after that, and
the normalized periodogram has larger values at high frequencies (which
explains the series fluctuations).

Figure 7.7 Simulated time series of length n = 200 of a short-range correlation process (top), its sample
autocorrelation function at lags k = 0, …, 50 (middle) and normalized periodogram at frequencies λr =
2πr with r = 0, …, 0.5 (bottom)
 Formally, a stationary process {Yt, t∈T} is said to have long-range
correlation (or long memory) if its autocorrelation function Ρ(.) satisfies the
power law:
Ρ(k) ~ c k−(1.2d), k → ∞,
where c and d are two constants such that c ≠ 0, d ≠ 0, and d < 0.5, and k is
the lag. This means that the function Ρ(.) decays to zero very slowly with a
hyperbolic decay. Moreover, the process is said to have persistent long-range
correlation if 0 < d < 0.5, so that ∑k Ρ(k) = ∞, reflecting the fact that the
remote past has an influence into the present.
In the frequency domain, a long-range correlation process can be defined
as a process whose spectral density function f(.) satisfies the power law:
f(λ) ~ c ∞−2d, λ → 0,
where c and d are two constants such that c ≠ 0, d ≠ 0, and d < 0.5, and λ is
the frequency. This means that the function f(.) has a pole at zero if 0 < d <
0.5, that is f(0) = ∞, signifying that the low frequencies predominate and
therefore longterm oscillations are expected. These processes whose function
f(.) has the form f(λ) ~ λ −α, where α is a constant, are usually known as 1/f α
noise. This property of the spectral density function f(.) can be used to
distinguish different types of noise in terms of colour: α = 0 (white noise) and
α = 1 (pink noise), amongst others. It is worth mentioning that pink noise has
been found in a number of time series from human movement systems and
some stochastic models have been proposed for explaining these phenomena
(e.g. Diniz et al. 2010, 2011).
Figure 7.8 represents a realization {Y1,…,Yn} of length n = 400 of a long-
range correlation process. It also shows the sample autocorrelation function
(.) at lags k = 0, …, 50 and the normalized periodogram In(.) at frequencies λr
= 2πr with r = 0, …, 0.5. It is clear that the time series exhibits long non-
periodic oscillations around the mean, the sample autocorrelation function
has large and positive values with hyperbolic decay and the normalized
periodogram has larger values at low frequencies (which explains the series
fluctuations).
For the distinction between short- and long-range correlations, the length
of the time series is an important question. In fact, a realization of a stationary
process with long-range correlation can easily be mistaken for a realization of
a non-stationary process, if the series length is very small. The statistical
discrimination between short- and long-range correlations can be done using
several tests, such as the rescaled range method (R/S), the detrended
fluctuation analysis (DFA) and others (e.g. Palma 2007; but see Chapter 5).
With the sailing example, the method of analysis of the distance between a
sailing boat and an optimal point for the start of the regatta was illustrated in
terms of its time structure. Somehow, space was analyzed as a function of
time. It is also possible to analyze changes in the spatial structure across time.
After presenting some bases for performing time series analysis, we then
describe a method for performing spatial patterns analysis – Voronoi
diagrams – which is starting to be used in sport sciences.
Voronoi-based models for team sports analysis
Studying spatial patterns formed by a group of individuals gives us some
insight into the understanding of interactive behaviour, in both the individual
and collective dimensions. For example, spatial patterns characterized by
large distances between all individuals will indicate some source of
inhibition, whereas small distances will indicate some source of attraction. In
addition, larger interpersonal distances associated with a single individual
may reflect some source of avoidance. In the sports context, spatial pattern
analysis is thought to be a useful approach for characterizing and evaluating
players' space management, which is associated with patterns of interacting
behaviour.
Figure 7.8 Simulated times series of length n = 400 of a long-range correlation process (top), its sample
autocorrelation function at lags k = 0, …, 50 (middle) and normalized periodogram at frequencies λr =
2πr with r = 0, …, 0.5 (bottom)

The methods described below are based on a well-known two-dimensional

spatial tessellation (or decomposition), the Voronoi diagrams. Given a set of
n points distributed in a plane (Figure 7.9A), this spatial construction divides
the plane into n cells (Figure 7.9B), each associated with one and only one
point; in other words, each cell corresponds to a part of the plane that is
closer to one of the points (see Okabe et al. 2000). For instance, in a soccer
match, the points could be the position of the players, the plane the play area
and the cells the regions associated to each player.

Figure 7.9 Example of a set of points in a plane (A) and corresponding Voronoi diagram (B)

This spatial construction has already been suggested by other authors for
studying players' spatial distribution in team sports, having been applied in
variety of game settings, namely, electronic soccer games (Kim 2004),
robotic soccer (Law 2005), on-field hockey games (Fujimura and Sugihara
2005) and on-field soccer games (Taki et al. 1996). The Voronoi cells that
define the individual dominant region of each player (Taki et al. 1996;
Fujimura and Sugihara 2005) change continuously over time, owing to
continuous adjustments of the players' positions, which implies permanent
changes in the global spatial configuration. Thus, for an adequate application
of such markers in any team sports, these areas should be analyzed
throughout the duration of the game or trial. For instance, in the work by Kim
(2004), the spatial markers considered were area and number of vertices of
each Voronoi region; however, these were averaged, eliminating the temporal
component of the phenomenon, which in this particular study limits the
understanding and explanation of players' spatial relation.
The models described here allow the study of performance at different
levels of analysis: model 1 – player and team individual behaviour; and
model 2 – intra- and interteam interaction behaviour. Here, we present some
examples that, with some adaptations, can be applied to other sports. A
particular aspect to be considered in such an adaptation is the number of
players involved and the field dimensions. Codes are available upon request
from the lead author of this chapter.

Model 1: player and team individual behaviour

In this model, we consider the Voronoi diagram generated to the set of points
corresponding to the positions of players from two opponent teams during a
game or trial.
When formally analyzing these regions across time, one can better
understand player and team spatial interaction. For example, at a team level in
football (soccer), the attacker team is expected to free up space, increasing its
area of action, while the defender team tries to tie up space, decreasing its
area of action (McGarry et al. 2002). At a player level, in soccer for example,
players from the defender team are expected to maintain their space closed to
prevent attacker players to score, keeping their net tight while in danger
(Figure 7.10A); on the other hand, players from the attacker team are
expected to keep greater dominant regions but some will try to disturb
defenders' organization and create instability by changing their action (Passos
et al. 2008), as entering the defence zone (Figure 7.10B).
A detailed description of the method is presented next. Given a game or
trial of f frames (see Chapter 9), in a play area of dimension W × H metres,
consider that the trajectory of each of the P players involved was captured
and adequately converted to real coordinates, so that for each player a
collection of coordinates {(x,y)1,(x,y)f} is available.
The play area is mapped with a grid of W × H positions. At each frame (f),
the area of the Voronoi cell of player k (VA(k), k∈[1,P]) is the sum of all
grid positions (i,j) (where i = 1,..,W and j = 1,…,H) that are closer to that
player than they are to any other player. This can be mathematically defined
as:

where I(i,j) is a Boolean function that takes value 1 if player k is the closest
player to the grid position (i,j) and 0 otherwise
Figure 7.10 Spatial configurations: (A) attacker (grey areas) and defender (white areas) teams and; (B)
attacker player breaking defence organization; the arrow indicates the direction of the attack

Grid points that are equidistant between two or more players constitute the
boundaries of their respective regions and therefore are not added to the
corresponding areas.
The calculated areas can be used to investigate how the size of the Voronoi
cells changes over time for each team and/or for each player and related to
specific phases, events and/or characteristics of the game.
The model described was applied to data from futsal (Fonseca et al.
2012b). We considered 19 trials from five versus four plus goalkeeper plays,
all starting with similar conditions and each ending when the attack lost ball
possession. On average, plays lasted 848 (± 374) frames (corresponding to
approximately 0.57 (± 0.249) minutes), a minimum of 315 and maximum of
1558 frames (approximately 0.21 and 1.04 minutes, respectively). The main
results are presented below.
With model 1, it was possible to verify the following: (1) on average and
for all plays, the attacker team had Voronoi regions with larger areas in
comparison with those defined by the defender team (Figure 7.11); (2) the
area of these regions was more variable for the attacker team, which
presented, in each frame and across the duration of all trials, a larger standard
deviation of the Voronoi area (Figure 7.11). In addition, (3) the spatial
behaviour of the attacking players was more stable during each trial,
presenting Voronoi areas with smaller approximate entropy (see Chapter 5;
see also Fonseca et al. (2012a) for criteria for normalizing time series for
performing entropy analysis), as illustrated in Figure 7.12.
It is also interesting to study intra- and interteam behaviour, as we next
describe.

Figure 7.11 Example (one play) of the mean Voronoi area (VA) across time for each team; error bars
represent the standard deviation (adapted from Fonseca et al. 2011a)

Figure 7.12 Comparison of the mean entropy of Voronoi area (VA) between teams in the same trial;
error bars represent the standard deviation (*** P < 0.001); adapted from Fonseca et al. (2011a)

Model 2: intra- and interteam interaction behaviour

In this model, we consider Voronoi diagrams generated separately for the
competing groups, e.g. in a sports application, these could be the attacker and
defender teams. The superimposition of these two diagrams, illustrated in
Figure 7.13, allows an interesting spatial analysis of individual and collective
interaction behaviour. This graphical construction can be considered as
evaluating the similarity between the spatial distributions of two opponent
groups. If each pair of opponents is closely located, as illustrated in Figure
7.14A, the overlapped diagram will indicate an almost perfect fit between the
two Voronoi diagrams. On the other hand, for spatially unrelated pairs, the
described spatial composition will indicate the opposite, as in Figure 7.14B.

Figure 7.13 Construction of the superimposed Voronoi diagram (bottom) from considering, separately,
the Voronoi diagrams for team A (black dots) and team B (white dots)

To quantify this spatial agreement, two variables were defined, one to

measure the degree of coordination at an individual level and the other for
measuring coordination at a team level. More precisely, the maximum
percentage of overlapped area (max%OA) measures coordination at player
level and the percentage of free area (%FA) measures coordination at team
level.
Figure 7.14 Construction of the superimposed Voronoi diagram for (A) exclusively paired opponents
and (B) randomly located individuals

Figure 7.15 Measures from the superimposed Voronoi diagram: (A) maximum percentage of
overlapped area for each individual of the group marked with black dots; and (B) percentage of free
area (in black)

The max%OA is calculated for each player and represents the maximum
percentage of the Voronoi cell that is covered by the cell of an opponent; the
smaller this measure the greater the number of opponents in the
neighbourhood (Figure 7.15A). The %FA summarizes the degree of
similarity between the overlapped Voronoi diagrams, which is calculated by
extracting from the plane the sum of the max%OA calculated for one of the
groups (Figure 7.15B).
The %FA is inversely proportional to the degree of agreement between the
spatial configurations of both groups and therefore can be used to analyze
and characterize the interaction behaviour between them. This construction is
of particular interest for studying interaction behaviour in team sports, where
the two groups correspond to two competing teams.
Consequently, in team sports, when analyzing the spatial distribution of
two opponent teams playing in a well-defined area of known dimensions, it is
possible to consider reference values associated to two specific spatial
relations: i) when each player of the defender team assumes an exclusive
pairing with an opponent, which can be linked to a man-to-man defensive
method; and ii) when players from both teams are randomly distributed in the
playing area, which, while may not appear to be a reasonable assumption for
players' behaviour, provides a reference for the interteam spatial arrangement
when the location of each player is chosen, regardless of the location of the
others. To exemplify this, model 2 was applied to data from futsal (Fonseca
et al. 2011). We considered 19 trials from five versus four plus goalkeeper
plays, all starting with similar conditions and each ending when the attack
lost ball possession.
Reference values for this specific setting, ten players in a play area of 20
m2, were generated for randomly distributed players and exclusively paired
at different interpersonal (inter-pair) distances. For situations where players
are exclusively paired with an opponent, the percentage of free area increases
as the maximum distance allowed between the pairs increases. The lower and
upper reference values obtained for complete spatial randomness (CSR), i.e.
when all players are randomly allocated in space, were 0.22 and 0.50,
corresponding to the upper and lower limits of a 95% confidence envelope
for the %FA:

where %FA25:1000 and %FA95:1000 represent the order, out of 1000, of the
%FA obtained in simulated patterns of CSR.
Based on these reference values, it is possible to classify at each frame of a
trial how players are spatially distributed; in particular, observed values
below the lower limit of the 95% confidence envelopes for CSR suggest that
the defender team is likely to be applying a man-to-man defence method
(exclusive pairing). Figure 7.16 shows the %FA observed across the duration
of one trial in this study.
A spatial approach of futsal players' collective behaviour in a five versus
four plus one played in midfield suggests that players are not considering a
man-toman defence strategy, which is what was expected, given the setting of
this study. It also indicates that the zone defence method spatially relates to
an absence of interaction between players, as the observed value of %FA is,
during most of each trial, within the corresponding bands, [0.22, 0.50].

Figure 7.16 Example (one play) of the observed percentage of free area (%FA) across time (solid line)
and the 95% confidence interval for spatial random distribution (dashed lines)
Conclusion
In sport, space and time are key parameters to consider if individual or
collective behaviours are to be understood. However, space and time should
not be considered in abstract terms but embedded in the variables that capture
the interaction between the performer and the environment. In this chapter,
we have presented ways in which such analysis could be performed. First, we
described how time series could be analyzed in the time domain and in the
frequency domain. Moreover, we argued that the analysis of spatial patterns
formed by a group of individuals could give insights about individual and
collective behaviours. For this type of analysis, we presented a technique
called the Voronoi diagrams, a two-dimensional spatial decomposition of
geometrical space. The techniques described focus on the individual
behaviour of the player and team and on the intra- and interteam interaction
behaviour. These are possibilities that open new ways to understand the
complexity of sport behaviour.

Questions for students

1. What are the major differences between time domain and
frequency domain analysis and which are some relevant functions
in each of these methodologies?
2. What does the stochastic memory of a process indicate about the
dependence structure of that process over time?
3. Why may it be important for the statistical analysis of the
underlying process to have time series that are relatively long?
4. In the sports context, what do the areas defined by the Voronoi
diagrams represent?
5. What characteristics, besides spatial location and field area, could
be considered to weight the Voronoi cell of each player?
6. What are, in principle, the spatial characteristics of the interaction
behaviour between players in team sports that could be tested using
any of the models presented?
References
Araújo, D. (2006) Tomada de Decisão no Desporto. [Decision-making in
Sport]. Cruz Quebrada: Edições FMH.
Barris, S. and Button, C. (2008) A review of vision-based motion analysis in
sport. Sports Medicine, 38 (12): 1025–43.
Brockwell. P. J. and Davis, R. A. (1991) Time Series: Theory and Methods,
2nd edn. New York: Springer.
Davids, K., Glazier, P., Araújo, D. and Bartlett, R. (2003) Movement systems
as dynamical systems: the functional role of variability and its implications
for sports medicine. Sports Medicine, 33 (4): 245–60.
Diniz, A., Barreiros, J. and Crato, N. (2010) Parameterized estimation of
long-range correlation and variance components in human serial interval
production. Motor Control, 14: 26–43.
Diniz, A., Wijnants, M. L., Torre, K., Barreiros, J., Crato, N., Bosman, A. M.
T., Hasselman, F., Cox, R. F. A., Van Orden, G. C. and Delignières, D.
(2011) Contemporary theories of 1/f noise in motor control. Human
Movement Science, 30: 889–905.
Fonseca, S., Milho, J., Travassos, B. and Araujo, D. (2011) Modeling intra-
and inter-team spatial interaction in team sports, Symposium on A Complex
Systems Approach to Studying Behaviour in Sport. Organizer: Keith Davids.
13th FEPSAC European Congress of Sport Psychology, Madeira, Portugal.
Fonseca, S., Milho, M., Passos, P., Araújo, D. and Davids, K. (2012a)
Approximate entropy normalized measures for analyzing social
neurobiological systems. Journal of Motor Behavior, 44 (3): 179–83.
Fonseca, S., Milho, M., Travassos, B. and Araújo, D. (2012b) Spatial
dynamics of team sports using Voronoi diagrams. Human Movement Science,
31 (6): 1652–9.
Fujimura, A. and Sugihara, K. (2005) Geometric analysis and quantitative
evaluation of sport teamwork. Systems and Computers in Japan, 36 (6): 49–
58.
Kim, S. (2004) Voronoi analysis of a soccer game. Nonlinear Analysis:
Modelling and Control, 9(3): 233–40.
Law, J. (2005) Analysis of Multi-Robot Cooperation using Voronoi
Diagrams. Proceedings of the 3rd International Kemurdjian Workshop
‘Planetary rovers, space robotics and Earth-based robots-2005’, St
Petersburg, Russia, October 2005.
McGarry, T., Anderson, D. I., Wallace, S. A., Hughes, M. D. and Franks, I.
M. (2002) Sport competition as a dynamical self-organizing system. Journal
of Sports Sciences, 20 (10): 771–81.
Okabe, A., Boots, B., Sugihara, K., Chiu, S. N. and Kendall, D. G. (2008)
Spatial Tessellations: Concepts and Applications of Voronoi Diagrams, 2nd
edn. Hoboken, NJ: John Wiley.
Palma, W. (2007) Long-Memory Time Series: Theory and Methods. New
Jersey: Wiley.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Milho, J. and Serpa, S.
(2008) Informationgoverning dynamics of attacker-defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Taki, T., Hasegawa, J. and Fukumura, T. (1996) Development of motion
analysis system for quantitative evaluation of teamwork in soccer games, in
Proceedings of the International Conference on Image Processing, 16
September 1996, Lausanne. IEEE, Vol. 3, pp. 815–18.
8 Self-organizing maps and
cluster analysis in elite and sub-
elite athletic performance
Wolfgang Schöllhorn, Jia Yi Chow, Paul
Glazier and Chris Button
This chapter examines ways in which movement patterns can be analyzed as
performance contexts change or as a function of learning and development.
The methods described can be used to study the effects of important factors
such as fatigue, injury, learning, development or training in motor
performance.
Classifying objects can be considered as one of the major tasks in science
(Slife 1995). Classifying usually occurs at the beginning of the research
process and involves categorizing objects by criteria specified by the
investigator. Most often, these criteria are associated with implicit
assumptions and rely on a certain philosophical background. In the scientific
literature, two general types of classification study can be identified; those
adopting a confirmatory approach or an exploratory approach, with the
former being more common (Tukey 1980; Jaeger and Halliday 1998). In
confirmatory approaches, the classes are given in advance and are tested for
statistical significance. In exploratory approaches, by contrast, only the
criteria for the classification of the objects are provided. Once certain classes
are identified, the explorative approach can be followed by the confirmatory
approach so as to test the identified classes for significance.
For methodological and historical reasons, the quantitative investigation of
movement has typically been focused on the classification of simple
movements on the basis of time-discrete amplitudes of selected variables. In
biomechanical studies, these time-discrete variables (also known as
performance parameters) have typically been specified by deterministic or
hierarchical models and have some relationship with performance outcome.
Time-series data typically have to be reduced or collapsed to single data
points (maximum, minimum, average, value at specific event, etc.) and
pooled so that (confirmatory) statistical analyses can be performed. In
parallel, the analysis of movements has been mainly limited to the statistical
comparison of single variables. In cases where multiple variables have been
analyzed, potentially more useful practical knowledge can be derived.
When movement patterns are classified by means of multiple time-
continuous variables, a more holistic approach is pursued. The difference
between a timediscrete and time-continuous movement description can be
illustrated with an analogy. If we see a known person far away standing still,
it is often difficult to identify that person. Once he/she starts to walk, our
visual system receives additional information that increases the likelihood of
recognizing that person Similarly, if only the left elbow of the person walking
can be viewed, the probability of correctly identifying that person is rather
low. However, if the motion of other joints and segments are presented
simultaneously, the probability of recognizing the person improves. Indeed,
Johansson (1973) showed that perception of biological motion relies more
heavily on relative motion, rather than absolute motion characteristics of limb
and torso segments.
While time-discrete variables focus on instantaneous characteristics of
single joint or segment motion, time-continuous variables provide the
opportunity to categorize movement qualities such as types of movement
(jumping, running, etc.), modes of movement (springy, tentative, rushed, etc.)
and individual styles (individual expressions of types and modes) of
movements. According to pattern analysis, the quantification of coordination
is initiated by means of the determination of a reference pattern to which all
other patterns have to be compared with. Alternatively, the similarity of all
patterns to each other can also be determined.
Numerous questions can be addressed through the quantitative description
of movement patterns in sport on the basis of time-continuous variables. For
example, how similar are the average patterns of movement within a class of
movement? How are gait patterns similar within an individual across
different performance context or between different individuals? Even in the
study of plants or animals and their subordination into certain classes in
biology, a first step for time-continuous movement patterns is their
classification on the basis of certain criteria. Because the general
classification of movements already has been performed successfully by
numerous scientists (e.g. Hay 1993), mathematical algorithms are first
required that are relevant to everyday experience and observations. Thus,
plausible criteria for the classification of objects seem to be their relative
similarity or proximity in a most abstract understanding. The quantitative
sorting of objects by means of linear distances led to the development of
different forms of cluster analysis (Everitt et al. 2001), while the
mathematical simulation of neuronal assemblies was associated with artificial
neural nets that meanwhile provided plausible nonlinear groupings.
Without delving too deeply into the problem of interpreting similarity or
proximity (Everitt et al. 2001), the simplest procedure is to quantify a certain
quality of all objects and to determine the relative distance of these quantities
(e.g. three people walking should be clustered on the basis of their minimum
knee flexion during single stance phase in gait). The first step in such a
procedure would be to assign a quality or variable to ‘knee flexion’, such as
‘knee angle’. This quality is typically represented by a unit such as radians or
degrees. Hence, qualities can be compared by means of their relative size or a
vector distance. A commonly used measure for mathematical comparisons is
the Euclidean distance, which represents the distance between two objects on
a straight line. If the distance between two knee angles of two participants is
smaller than the distance relative to a third participant, the first two
participants would be assigned to the first cluster and the third participant to
another. Applying the same procedure to movement patterns that are
described by means of multiple time course variables, some vector and
matrix algebra is necessary that includes similarity and proximity measures
analogously. In contrast to cluster analysis, which exclusively sorts objects,
selforganizing maps (SOMS; as a specific form of artificial neural network
[ANN]) have the additional ability to reduce high-dimensional data to low-
dimensional data. Therefore, SOMs are often used as a preparatory tool for
cluster analysis.
In the following sections, we examine the application of cluster analysis
and SOMs to movement analysis in sport and human movement science.
Cluster analysis
As discussed earlier, the application of cluster analysis in sport can be
divided into exploratory and confirmatory approaches. In the exploratory
approach, hierarchical cluster analyses are typically applied to holistic
descriptions of movements. Starting with a single object, all other objects are
clustered successively by means of a chosen metric until all objects are
included. The exploration of the number and size of the clusters along the
relative metric provides either previously expected groups of movements or
leads to a new categorization that has to be interpreted creatively. In
summary, here the least information about the structure of the objects is
included in the classification process. In the second case of plausible
interpretations, the confirmatory approach, can be followed. Here, the classes
or at least the number of classes are given in advance and the cluster analysis
procedure classifies all objects accordingly on the basis of the chosen metric.
Subsequently, the identified clusters have to be validated and any cluster
differences has to be tested for significance. However, irrespective of the
approach, a dominant influence on the results is given by the chosen variables
as well as by their pre-processing. Thus, the selection of variables is directly
connected with the specificity of the hypothesis and the expectation of the
investigator while the pre-processing of the data has an indirect but strong
influence on the relative weighting of the variables with respect to the
resulting clusters (for further details, see Everitt et al. 2001; Rein et al.
2010a).
Owing to the increased processing capacity of computers by the end of the
last century, cluster analysis has since been applied increasingly to large sets
of time course oriented movement data without data reduction.

Exploratory approaches
One of the first exploratory applications of cluster analysis in the context of
movement pattern analysis in sports was provided by Müller (1986). Several
kinematic variable time courses and muscular activities were measured
during the demonstration of different skiing techniques for downhill skiers.
Discrete biomechanical parameters were extracted and classified by means of
a cluster analysis. It was found that irrespective of the snow and slope
conditions, the techniques could be classified as similar.
 An explorative cluster analysis on the basis of time-continuous variables in
discus-throwing movements during practice of a single high-performance
athlete provided evidence for a successful learning process that led to
enduring/lasting qualitative changes of the throwing technique (Schöllhorn
1993). Eight discusthrowing trials before and after a biomechanical feedback
intervention were described by means of the time courses of 40 joint angles
and angular velocities during the final throwing phase. Three trials were
performed within one competition before the intervention and five trials
during different competitions following the intervention. For data reduction,
these high-dimensional data were factor analyzed. The resulting factor-
loading matrices were compared by means of a structure comparison
algorithm, which led to a distance matrix. The subsequent cluster analysis
(Figure 8.1) clearly separated three trials (T791–T793) before a specific
training intervention from five trials (T84–T88 ) performed after the
biomechanical feedback. Figure 8.1 displays exemplarily the history of a
hierarchical cluster analysis applied to eight discus throws during a learning
process. The cluster analysis begins with determining the distances between
all trials and is followed by determining the two most similar (smallest
distance) trials (T791 and T793). The clustered trials are then considered as a
single new trial. Subsequently, the next two trials with the smallest distance
are clustered together and so on until all trials are clustered together. Similar
results revealed the cluster analysis of the same data when data reduction was
performed by means orthogonal reference functions (Schöllhorn 1995).
Figure 8.1 Clustering dendrogram resulting from an average linkage algorithm; horizontal lines
indicate the level of the rescaled distance at which the respective movements are grouped into one
cluster

Explorative cluster analysis of biomechanical data during several double

steps in running provided a separation of left and right contact phase as well
as of flight phase. Just by examining the ground contact phase of a single leg,
95% of the runners could be identified individually (Schöllhorn 1999). A
similar application of cluster analysis to kinematic and muscular variables in
long jumping allowed the identification of the athletes during take off (Jaitner
et al. 2001a,b). Muscular variables pertaining to either the envelope of the
rectified electromyography signals or their frequency spectra can be taken as
input variables for the classification process for the cluster analysis. Different
levels of expertise in handball players were also identified inter- and intra-
individually on the basis of timecontinuous, three-dimensional kinematic data
of the hip, shoulder, elbow and wrist joint by Schorer et al. (2007). Cluster
analysis yielded an assignment rate of 92% for the level of expertise.
A new direction in terms of cluster analysis applications in the field of elite
sports was taken by Jäger et al. (2003) by analyzing tactical behaviour of
teams in the form of player's coordinates during volleyball matches from the
2002 World Championships. Over 250 similar situations during a tournament
where six participating teams competed against each other, were video
recorded at 50 Hz. Positions (x and y coordinates) of the defending teams
during a defensive move in time-discrete and time-continuous form were
inputted to a cluster analysis that subsequently revealed team specific
behaviours. Contrary to the opinions of a coach, who believed that all teams
played with the same system, the specific behaviours of each team could be
identified by means of the movement of all six players during the defence
movement independent of the opponent. This individuality of team dynamics
has a number of potential implications for tactical training.

Confirmatory approaches
A confirmatory approach of cluster analysis with a single discrete duration
parameter was examined by Lames (1992). Using ideas from dynamical
systems theory and with a focus on investigating the hysteresis effects (see
Kelso 1995), the duration of the driving and pitching movement of several
golfers was measured for different distances. In order to initiate a hysteresis
effect, the golfers performed shots from 100 metres first, then decreasing
down to ten metres in five-metre steps and then increasing back to 100
metres. The subsequent cluster analysis of the movement durations led to
two, three, four or five clusters from which only the three-cluster solution
could be interpreted plausibly in accordance with the assumed hysteresis
phenomenon.
In a more recent study by Chow et al. (2008), which examined
coordination changes of novice participants in a football (soccer)-kicking
task, cluster analysis was also effectively used to determine intra-individual
differences in kicking patterns. Key kinematic data of various joint motions
were captured over a fourweek intervention period (40 trials per session over
12 sessions) and the kinematic data were used as input for a cluster analysis
procedure. Differences in coordination patterns within individual for each
session were effectively determined and the information provided valuable
insights about movement pattern variability, as well as the presence of
preferred kicking patterns. Such information is extremely critical in helping
researchers to understand the learning processes and to investigate
nonlinearity in learning evidenced by sudden transitions from one movement
pattern to another, as well as the emergence of pattern variability prior to a
transition.
 Another confirmatory application of cluster analysis was performed by
Ball and Best (2007) to determine the presence of weight transfer for two
styles of golf swing. Sixty-two golfers, from professionals to high-handicap
players, performed simulated drives, hitting a golfball into a net. While
standing on two force plates, the centre of pressure position relative to the
feet was quantified at eight swing events identified from a 200-Hz video.
Cluster analysis on the basis of these timediscrete parameters revealed two
major styles of golf swing: a front-foot style and a reverse style.
Nevertheless, validation procedures were required.

Validation of clusters
As cluster techniques will always identify groups of data depending on the
identification parameters, it is important to consider additional procedures to
validate them. For supporting and providing the statistical proof of the
resulting clusters, different approaches have been suggested. According to
Handl et al. (2005), cluster validation measures can be distinguished into
internal and external measures:
External validation measures comprise all those methods that evaluate a
clustering result based on the knowledge of the correct class labels . . .
Internal validation techniques do not use additional knowledge in the
form of class labels, but base their quality estimate on the information
intrinsic to the data alone.
(Handl et al. 2005: 3203)
With further extension to the work by Lames (1992), Rein et al. (2010b)
applied an internal validation approach to their cluster analysis of basketball
shooting. The phenomenon of phase transition in basketball hook shots with
decreasing and increasing distances from nine metres to two metres and back
to nine metres, with one-metres increments was investigated. The input
variables for the cluster analysis were 12 angle variables derived from a 13-
segment, rigid, three-dimensional body model. The clusters were interpreted
in the terminology of systems dynamics as attractors with certain criteria.
Only two of eight participants showed a clear expected phase transition
behaviour. Importantly, in this preliminary study, it was possible to identify
three distinctive shooting patterns with varying frequencies at different
shooting distances.
 As three different shooting patterns had been previously established, the
external validation procedures were adopted by Rein et al. (2010a). Two
studies in basketball served for testing the sensitivity of cluster analysis to
pre-processing and for testing the phenomenon of phase transitioning in
hook-shot technique. For the first experiment, four professional basketball
players had to throw from three different distances with three different
techniques. Owing to the impact of data normalization, the same analysis was
performed with z-transformed (average = 0, standard deviation = 1) and raw
data. Both sets were validated by means of bootstrapping and Hubert-Gamma
method. Overall, in the first experiment, the cluster analyses led to ‘entirely
feasible’ results and were able to reproduce a priori known differences
between diverse movement patterns. In the second validation study, two
basketball players were instructed to shoot baskets by means of a hook-shot
technique from distances between two to nine metres. In contrast to Rein et
al. (2010b), the task was limited by a lowered ceiling to force flight curves of
the ball that are dominated by the velocity of release than by the angle of
release, with the aim of causing a phase transition in the movement pattern
with increasing or decreasing distance. Only one participant showed strong
indications of the use of two distinct patterns whereas another participant
displayed ‘distinctively fewer differences’ as shooting distance was
manipulated. Bootstrapping and Hubert-Gamma values showed that a
validation procedure is necessary for the confirmatory approach of cluster
analysis.
The external validation approach pursues the probability of ending up with
certain clusters relative to arbitrary or accidental data. The internal approach
compares the variation within a cluster relative to the variation between
clusters (Bauer and Schöllhorn 1997). Both approaches demonstrate the
problem and importance of variable selection and preparation that is known
in the context of analyzing complex self-organizing systems (Haken et al.
1995). According to Haken et al. (1995), the selection of collective variables
or order parameters is highly dependent on the investigator's intuition. The
problem seems to become even bigger with increasing complexity of the
movement task and the possibilities of compensations. Yet there are no
general rules for the choice and selection of variables as well as for the
preparation of the data before cluster analysis.
In summary, cluster analysis provides a powerful dimension reduction tool
which can serve different purposes depending on the nature (i.e. explorative
or confirmatory) of the research undertaken. In contrast to cluster analysis,
ANNs have the ability to separate two classes nonlinearly (Figure 8.2) and
therefore lead to higher recognition rates and potentially more flexibility in
their use (Haykin 1994; Schöllhorn and Jäger 2006). As we discuss in the
next section, ANNs can be administered independently or in combination
with cluster analysis.
Self-organizing maps
SOMs, also known as Kohonen maps, are a specific type of ANN that can be
used to mathematically model specific characteristics of neuronal cell
assemblies. In contrast to supervised learning ANNs such as multi-layer-
perceptrons (MLPs), SOMs are trained using unsupervised learning to
typically produce a two-dimensional discrete representation of the input
space of the training samples, called a map. A big advantage of SOMs is the
way in which it captures low-dimensional views of high-dimensional data,
akin to multidimensional scaling. SOMs are different from other ANNs
because of their usage of a neighbourhood function that preserves the
topological properties of the input space. Similar to most ANNs, SOMs
operate in two phases. In the first phase, the map is trained using input
examples. The second phase, called mapping, automatically classifies a new
input vector. The competitive working process is also called vector
quantization. Components of a SOM are called nodes or neurons that are
associated with a weight vector of the same dimension as the input data
vectors and a position in the map space. The usual arrangement of nodes is a
hexagonal of rectangular grid. The procedure for placing a vector from input
data space on to the map is to first find the node with the closest weight
vector to the vector taken from input data space. Once the closest neuron is
located, it is assigned the values from the vectors taken from the input data
space. Mathematically, SOMs are sometimes associated with nonlinear forms
of principal component analysis.
Figure 8.2 Linear versus nonlinear separation

Owing to their capacity to map high-dimensional data to a low-

dimensional map whilst preserving the topological characteristics of the
original complex movement data, in principle, four different modes of
application in the analysis of complex movement patterns can be
distinguished (Figure 8.3A–D). In all approaches, an explorative strategy is
pursued.
(A) In relation to describing a single movement by means of multiple time-
discrete data, each movement is represented by a single vector that is
mapped to a single node in the map space. By mapping several
movements to the map space, SOMs function as classifiers that group
similar movements to each other, owing to the characteristics of
neighbourhood preservation.
Figure 8.3 Variations of data processing by means of self-organizing maps

When the complex sport movement is described by means of time course

data, three different approaches can be adopted. In all approaches, a single
movement is described by means of several variable time courses.
(B) In the first time course oriented approach, the input vectors are
constructed from all variable intensities at single instants. The number of
vectors of a single movement is given by the number of instants the
 movement was measured with. Here, each feature vector is mapped to a
single node in the map space and the whole movement is characterized
by a two-dimensional trajectory in the map space (Figure 8.3).
(C) The second time course oriented approach considers each variable
timecourse as a single vector. Consequently, the number of input vectors
per movement is given by the number of variables the movement was
described with. The SOM maps each variable time course characteristics
to a single node. Each resulting two-dimensional trajectory in the map
space links the different time course characteristics of a single
movement (Figure 8.3).
Once the movements are mapped to the node space (Figure 8.3B,C), the
coordinates of the trajectories form new vectors, which can be either fed into
another SOM or given into a cluster analysis to classify all movements by
their relative similarity. When the trajectories are put into another SOM, the
trajectories are mapped to different single nodes. The distribution of the
nodes over the whole grid then displays the relative similarity of the
movements in a two-dimensional space. However, these two-dimensional
coordinates can be mapped with a cluster analysis to a linear grouping. When
the coordinate vectors of the trajectories are directly put into a cluster
analysis, the movements are grouped according to their one-dimensional
similarity. In both cases (Figure 8.3B,C) when the SOMs are applied in a
first-order form to the map, the movements on a trajectory in the low-
dimensional node space the SOMs are taken for nonlinear data reduction
comparable to linear analogies such as principal component analysis or
Karhunen–Loève transformation. When SOMs are applied in the second-
order form for the mapping of the trajectories, we can consider them as
nonlinear classification procedures comparable to linear analogies such as
cluster analysis.
(D) Instead of modelling the movement qualities by means of two SOMs or
one SOM and a cluster analysis, a fourth approach is suggested. Here,
all variable time courses of a single movement are put in a single, rather
long, vector, which forms the input vector for the SOM. In this case, the
SOM directly provides the two-dimensional map where each node
corresponds to the model of a movement and the whole node grid
provides the relative similarities of several movements.
In the following, the SOM research that is related to the area of movement
execution in particular will be followed by the sport game related
applications of SOMs.
To date, most applications of SOMs in elite sports can be found in the form
of (B) or (C). An explorative combination of SOM and cluster analysis in
form of (B) was applied by Bauer and Schöllhorn (1997). Two high-
performance discus throwers were analyzed during different competition and
training events over one year. The athletes' movement during the final
throwing phase were described using a 14-segment, rigid body model by
means of 13 joint angles and 13 angular velocities plus the trunk orientation
angle and angular velocity, to achieve a physically complete description.
Feature vectors of the variable time courses were mapped on a 10 × 10-node
SOM. The resulting trajectories were subsequently placed into a cluster
analysis. The results revealed, firstly, a clear differentiation of the two
athletes. Secondly, they identified a successful intervention that resulted in a
distinct separation of three trials before and five trials after the intervention
period and, thirdly, sessions of training could be clearly separated from each
other and provided evidence for day-dependent throwing strategies.
However, the identification of signature movements for each discus thrower
as well as continuous variations called traditional model- and repetition-
oriented approaches into question.
Further evidence for the individuality of movement patterns in running
patterns was found by Schöllhorn and Bauer (1997). By analyzing three to
five double steps of 20 runners during a 2000-metre run by means of a SOM
and cluster analysis, on the basis of the same variables as in Bauer and
Schöllhorn (1997, it was possible to separate automatically contact phases of
the right/left foot and the flight phase, as well as to identify the athletes by
means of their kinematic data during a single ground contact phase.
Identification was based on a 91 per cent recognition rate for the left foot and
96 per cent for the right foot but only 75 per cent for the flight phase.
The same approach with SOM and cluster analysis was pursued by
Schöllhorn and Bauer (1998) in an investigation of world-class javelin
throwers from different nations over several years in the search for
individuality in elite sports. The same variables and data processing approach
as in Bauer and Schöllhorn (1997) were used for describing and mapping the
movements over a SOM towards a dendrogram. Most intriguingly, the results
revealed the recognition of individual movement patterns of two female
javelin throwers over several years. Obviously, the fingerprint-like throwing
patterns were identifiable over several years. Furthermore, evidence was
provided for a clear distinction of male and female throwing patterns as well
as for nation-typical throwing patterns.
Two consecutive SOMs were applied to establish whether the order of
movement executions affected variability over repeated trials (Janssen et al.
2010). Five participants performed ten first and ten second tennis serves in
blocked and alternating (serial) series. The first serves were mainly
accomplished with higher serving speed but less accuracy, whereas the
second serves were mainly characterized by higher precision, lower serving
speed. As the input dimension of the kinematic data (15 angles) was high, a
self-implemented network called 2-SOM (because of its structure, which
consisted of two series-connected SOMs) was chosen for analysis purposes.
Within the 2-SOM, the first SOM was used to reduce data dimension,
whereas the second SOM was used to classify the serve patterns. The 2-SOM
revealed a person recognition rate of 100 per cent using all serves of all
participants for only the blocked or serial condition. Within the ‘person
clusters', a network that was trained either with all individuals’ data from the
blocked or serial condition was able to distinguish between an individual's
first and second serve with an accuracy of 99 per cent (in the blocked
condition) and 90 per cent (in the serial condition), respectively. In general,
movement variability was greater for the serial condition compared with the
blocked condition over all participants.
Lees et al. (2003) reported the results of a study that applied SOMs with 12
× 8 neurons to analyze instep kicks by two soccer players for distance or
accuracy. The output patterns were repeatable for the same task for one
player. The authors claimed that the trajectories in the output map were
related to characteristics of the movement technique, although these
characteristics were not determined. In a subsequent study, Lees and Barton
(2005) used a similar approach to model several kicks by six soccer players.
Fourteen joint angles were obtained from three-dimensional coordinates for
80 equispaced time instants from take off for the last stride to the end of the
follow through of the kick. The output maps distinguished between right and
left-footed groups. Intra-player differences were small.
The analysis of 60 golf chip shots of four low-handicap male golfers over
distances of 4–24 metres in four-metre intervals, randomly assigned, was the
objective of the investigation of Lamb et al. (2011). From a 16-segment
model, a 24-dimensional input dataset consisting of joint angle time courses
was derived. Because of high heterogeneity in the training data between the
golfers, separate SOMs were trained on each player's respective kinematic
data. Subsequently, a second SOM was trained using the best-matching nodes
from the previous SOM for each respective player. The first SOMs varied
between 24 × 16 and 25 × 15 nodes, the second SOMs had in general 32 × 1
nodes. A U-matrix representation for each player allowed a phase specific
comparison within each player. The attractor layout diagrams were presented
as evidence of nonlinear phase transitions for three of the four shot distances.
A confirmatory approach was pursued as well when data were reanalyzed
from Bartlett et al. (2006) with respect to differences in treadmill running
with marker and no-marker conditions (Lamb et al. 2008). Kinematic time
course data were analyzed using a 29 × 23 hexagonal lattice. In general, the
findings of Bartlett et al. (2006) could be replicated. However, SOMs do
require more work to be applied properly. The ability to represent high-
dimensional coordination patterns on visualizable low-dimensional map hold
great potential (Bartlett et al. 2006).
In another study by this group, Lamb et al. (2010) showed how SOMs
could be used to objectively distinguish between three types of basketball
shots. Two SOMs were used in this investigation. Kinematic data from right
and left ankles, knees, hips and shoulder were processed and input to the first
SOM. Each variable was range normalized to maximum and minimum values
of +1 and –1. The map size of the first SOM was 42 × 13 nodes and was used
to analyze different phases of the movements, while the second SOM had 9 ×
6 nodes and was used for the classification of the complete trials. There
remains some debate over the best way to represent data (i.e. complete or
broken down into key phases). In particular, the trial SOM was considered
more relevant in providing a macro representation (see Lamb et al. 2010).
In contrast to this confirmatory approach, Beckmann et al. (2012)
employed a more exploratory approach for the possibility of identifying
individual-specific movement characteristics in different athletic events.
Specifically, five decathletes were recorded using two high-speed video
cameras and three-dimensionally reconstructed during the final throwing
phases of shot put, discus and javelin movements while competing in the
same decathlon. On the basis of a rigid 14-segment body model, time-
normalized angle and angular velocity time courses formed the input vectors
for the SOM and was compared with support vector machines. By inclusion
of all variables, the disciplines could be distinguished at a 100 per cent level.
When the throwing or shot put arm was excluded, no individual could be
identified by means of the SOM but an individual recognition rate over all
throwing disciplines by means of the support vector machines achieved at
98.5 per cent.
By coding the rallies as a series of zones, from where the players were
hitting the volleyball, an intriguing approach was suggested by Perl and
Lames (2000) that transferred the modelling by means of dynamic SOMs to
game sports. Approximately 5000 rallies of men and women's volleyball
matches of first and second German division formed the training set for the
20 × 20 node grid. However, the only plausible rallies that could be identified
were those in which single nodes corresponded to characteristic events in a
single rally; e.g. serve or offence smash. A similar approach has also been
applied to squash (Perl 2002) and table tennis (Perl and Baca 2003).
Summary
Overall, cluster analysis and SOMs enjoy increasing popularity among
movement scientists, owing to their capacity to explore and validate different
qualities in movement science and match analysis. Both methods of analysis
offer a fruitful basis for characterizing and interpreting high-dimensional
datasets. The need to balance exploratory and confirmatory approaches in
combination with timecontinuous and time-discrete approaches is one of the
biggest challenges for the coming years. The willingness to apply most recent
methodological developments from related disciplines is growing and offers a
promising wide new field of research.
References
Ball, K. A. and Best, R. J. (2007) Different centre of pressure patterns within
the golf stroke I: Cluster analysis. Journal of Sports Science, 25: 757–70.
Bartlett, R., Bussey, M. and Flyger, N. (2006) Movement variability cannot
be determined reliably from no-marker conditions. Journal of Biomechanics,
39: 3076–9.
Bauer, H. U. and Schöllhorn, W. (1997) Self-organizing maps for the
analysis of complex movement patterns. Neural Processing Letters, 5: 193–9.
Beckmann, H., Janssen, D. and Schöllhorn, W. I. (2012) Identifikation
individueller disziplinübergreifender Bewegungsstile [Identification of
individual and discipline independent movement styles], in Bundesinstitut für
Sportwissenschaft (ed.) BISp- Jahrbuch Forschungsförderung [Federal
Institute of Sport Science (ed.) Annual Report of Governmental Funded
Research Projects]. Bonn, Germany.
Chow, J. Y., Davids, K., Button, C. and Rein, R. (2008) Dynamics of
movement patterning in learning a discrete multiarticular action. Motor
Control, 12: 219–40.
Everitt, B. S., Landau, S. and Leese, M. (2001) Cluster Analysis, 4th edn.
New York: Arnold.
Haken, H., Wunderlin, A. and Yigitbasi. A (1995) An introduction to
synergetics. Open Systems and Information Dynamics, 3 (1): 97–130.
Handl, J., Knowles, J. and Kell, D. B. (2005) Computational cluster
validation in postgenomic data analysis. Bioinformatics, 21 (15): 3201–12.
Hay, J. G. (1993) The Biomechanics of Sports Technique, 4th edn.
Englewood Cliffs, NJ: Prentice Hall.
Haykin, S. (1994) Neural Networks: A Comprehensive Foundation. New
York: Macmillan.
Jaeger, R. G. and Halliday, T. R. (1998) On confirmatory versus exploratory
research. Herpetologica, 54: S64–6.
Jäger, J., Schöllhorn, W. I. and Schwerdfeger, B. (2003) A pattern
recognition approach for an opponent specific classification of tactical moves
in team sports, in E. Müller, H.
Schwameder, G. Zallinger and V. Fastenbauer (eds) Proceedings of the 8th
Annual congress of the European College of Sport Science. Salzburg:
Institute of Sport Science, pp. 370.
Jaitner, T., Mendoza, L. and Schöllhorn, W. I. (2001a) Analysis of the long
jump technique in the transition from approach to takeoff based on time-
continuous kinematic data. European Journal of Sport Science, 1: 1–12.
Jaitner, T., Ernst, H., Mendoza, L., Schöllhorn, W. I. (2001b) Changes of
EMG patterns during motor learning of ballistic movements, in: H. Gerber
and R. Müller, R. (eds) Proceedings of the XVIIIth Congress of the
International Society of Biomechanics (ISB). Zurich: ETH.
Janssen, D., Gebkenjans, F., Beckmann, H. and Schöllhorn, W. I. (2010)
Analyzing learning approaches by means of complex movement pattern
analysis. International Journal of Sport Psychology, 41 (Special Issue): 18–
21.
Johansson, G. (1973) Visual perception of biological motion and a model for
its analysis. Perception and Psychophysics, 14: 201–11.
Kelso, J. A. S. (1995) Dynamic Patterns: The Self-Organization of Brain and
Behavior. Cambridge, MA: MIT Press.
Lamb, P., Bartlett, R. M., Robins, A. and Kennedy, G. (2008) Self-organizing
maps as a tool to analyze movement variability. International Journal of
Computer Science in Sport, 7: 28–39.
Lamb, P., Bartlett, R. and Robins, A. (2010) Self-organising maps: an
objective method for clustering complex human movement. International
Journal of Computer Science in Sport, 9: 20–9.
Lamb, P. F., Bartlett, R. M. and Robins, A. (2011) Artificial neural networks
for analyzing inter-limb coordination: the golf chip shot. Human Movement
Science, 30: 1129–43.
Lames, M. (1992) Synergetik als Konzept in der Sportmotorik. [Synergetics
as a concept for sport movements]. Sportpsychologie, 6 (3): 12–18.
Lees, A., Barton, G. and Kershaw, L. (2003) The use of Kohonen neural
network analysis to qualitatively characterize technique in soccer kicking.
Journal of Sports Sciences, 21: 243–4.
Lees, A. and Barton, G. (2005) A characterisation of technique in the soccer
kick using Kohonen neural network analysis, in T. Reilly, J. Cabri and D.
Araujo (eds) Science and Football V: The Proceedings of the Fifth World
Congress on Science and Football. London: Routledge, pp. 83–8.
Müller, E. (1986) Biomechanische Analyse alpine Schilauftechniken.
[Biomechanical analysis of skiing techniques.] Innsbruck: Inn-Verlag.
Perl, J. (2002) Game analysis and control by means of continuously learning
networks. International Journal of Performance Analysis of Sport, 2: 21–35.
Perl, J. and Baca, A. (2003) Application of neural networks to analyze
performance in sports, in E. Müller, H. Schwameder, G. Zallinger, V.
Fastenbauer (eds) Book of Abstracts of the 8th Annual Congress of the
European College of Sport Science in Salzburg, Austria, 9–12 July 2003.
Available online at
www.informatik.unimainz.de/dycon/ABS_2003__Perl_Baca_Appl_NN.pdf
(accessed 7 June 2013).
Perl, J. and Lames, M. (2000) Identifikation von Ballwechselverlaufstypen
mit Neuronalen Netzen am Beispiel Volleyball. [Identification of rallies in
Volleyball by means of SOMs], in W. Schmidt and A. Knollenberg (eds)
Sport – Spiel – Forschung: Gestern. Heute. Morgen. Hamburg: Czwalina, pp.
211–16.
Rein, R., Button C., Davids, K. and Summer, J. (2010a) Cluster analysis of
movement patterns in multiarticular actions: a tutorial. Motor Control, 14:
211–39.
Rein, R., Davids, K. and Button, C. (2010b) Adaptive and phase transition
behavior in performance of discrete multi-articular actions by degenerate
neurobiological systems. Experimental Brain Research, 201: 307–22.
Schöllhorn, W. I. (1993) Biomechanische Einzelfallanalyse im Diskuswurf.
[Biomechanical single case study of discus throwing]. Frankfurt: Harri
Deutsch. Schöllhorn, W. I. (1995) Time course oriented analysis of
biomechanical movement patterns by means of orthogonal reference
functions. Paper presented at the XVth Congress of the International Society
of Biomechanics (ISB) 2–6 July, Jyvaskyla.
Schöllhorn, W. I. (1999) Complex individual movement styles identified by
means of a simple pattern recognition method, in P. Parisi, F. Pigozzi and G.
Prinzi (eds) Book of Abstracts of the 4th Annual Congress of the European
College of Sport Science, Rome, Italy, 14–17 July 1999. Cologne: European
College of Sport Science.
Schöllhorn, W. I. and Bauer, H. U. (1997) Linear–nonlinear classification of
complex time course patterns, in J. Bangsbo, B. Saltin, H. Bonde, Y.
Hellsten, B. Ibsen, M. Kjaer and G. Sjogaard (eds) Conference Proceedings
of the 2nd European College of Sport Science. Copenhagen: University of
Copenhagen, pp. 308–9.
Schöllhorn, W. I. and Bauer, H. U. (1998) Identifying individual movement
styles in high performance sports by means of self-organizing Kohonen
maps, in H. Riehle, and M. Vieten (eds) XVI International Symposium on
Biomechanics in Sports. Konstanz: Universitatsverlag, pp. 574–77.
Schöllhorn, W.I. and Jäger, J. M. (2006) A survey on various applications of
artificial neural networks in selected fields of healthcare, in R. Begg, J.
Kamruzzaman and R. A.
Sarker (eds) Neural Networks in Healthcare: Potentials and Challenges.
Hershey, PA: Idea Group Inc., pp. 20–58.
Schorer, J. Baker, J. Fath, F. and Jaitner, T. (2007) Identification of
interindividual and intraindividual movement patterns in handball players of
varying expertise levels. Journal of Motor Behavior, 39: 409–21.
Slife, B. D. and Williams, R. N. (1995) What's Behind the Research?
Discovering Hidden Assumptions in the Behavioral Sciences. Thousand
Oaks, CA: Sage Publications.
Tukey, J. W. (1980) We need both exploratory and confirmatory. American
Statistician, 34 (1): 23–5.
9 Single camera analyses in
studying pattern-forming
dynamics of player interactions
in team sports
Ricardo Duarte, Orlando Fernandes,
Hugo Folgado and Duarte Araújo
A network of patterned interactions between players characterizes
performance in team ball sports. Thus, interpersonal coordination patterns are
an important topic in the study of performance in such sports. A very useful
method has been the study of inter-individual interactions captured by a
single camera filming an extended performance area. The appropriate
collection of positional data allows investigating the pattern forming
dynamics emerging in different performance sub-phases of team ball sports.
This chapter outlines: (i) a simple and flexible motion analysis procedure to
capture the movement displacement trajectories of performers using a single
camera; and (ii) exemplar data illustrating the analysis methods employed in
the identification of pattern forming dynamics in a threeversus- three sub-
phase of association football near the scoring areas.
This chapter focuses on the methodological procedures for capturing the
relative positions of team players on the field using a single camera, as well
as appropriate analysis methods to analyze the pattern-forming dynamics of
player interactions. Team sport competitions can be characterized as complex
systems in which players continuously interact to contest ball possession and
positional advantage (Correia et al. 2012). This complex system regulates
and is regulated by players’ individual and collective interactions, which
destabilize or stabilize the system accordingly (Davids et al. 2005). The
inspirational background to the study of pattern forming dynamics in team
sports comes from coordination dynamics frameworks (Kelso 2009). The
ecological dynamics approach adopted some of those concepts and tools to
investigate emergent pattern-forming dynamics in decision making (Araújo et
al. 2006), adaptive behaviour (Davids et al. 2006) and social movement
coordination (Duarte et al. 2012a) at the individual environment scale in
representative sports settings. An important concept in these approaches is
the ‘order parameter’. Order parameters are collective variables synthesizing
the complementary relations of cooperation and competition among the
individual parts of a system, which may be used to capture and describe the
self-organizing patterns emerging in a complex, dynamical system (Kelso
2009). Given its dynamical nature and sensitivity to environmental changes,
when a system is displaced from equilibrium a phase transition in the ‘order
parameter’ occurs, switching from one coordinated pattern to another
(McGarry et al. 1999). Compound kinematic variables integrating relevant
ecological constraints are commonly used as order parameters, such as the
distance between the basket and attacking/defending players (Araújo et al.
2006) or angles formed between vectors linking players and an imaginary
line parallel to the try line in rugby union (Passos et al. 2009). To identify the
existence of an order parameter, changes in the state of a system must be
identified as consequences of variations in the ‘control parameter’ (Kelso
2009). Control parameters are variables that ‘move’ a complex system (i.e.
the order parameter) through different states, inducing nonlinear qualitative
behavioural changes (Kelso 2009). They can help explain why the
interactions of opposing players remain stable (despite the inherent
variability) or, in contrast, why attackers gain spatial and temporal advantage
over defenders, creating a phase transition in the system (Passos et al. 2008).
Interpersonal distance and relative velocity were often demonstrated to act as
control parameters in one-on-one performance sub-phases of invasion team
sports (Araújo et al. 2006; Passos et al. 2008; Duarte et al. 2010a).
To accurately determine and measure these order and control parameters
specific methodologies are needed for collecting the relative positions of
players in the pitch. One of the main methods for time–motion analysis of
player positions that has been widely used is the video-based system (Dobson
and Keogh 2007). These video-based systems are often used to analyze the
physical conditioning of players (e.g. Di Salvo et al. 2007) or capturing
player coordination dynamics (e.g. Travassos et al. 2012) and can be divided
into manual and automated video-tracking systems (Barris and Button 2008).
Despite providing a more time-efficient collection, automated video-based
systems are frequently expensive and rely on several pieces of fixed
apparatus (Barris and Button 2008), making it difficult to implement in some
research scenarios and on a large scale. Manual tracking video-based
systems, on the other hand, are often a more timedependent method, which
leads frequently to a focus on a smaller number of players, with reliability
(i.e. intra or inter-observer reliability) frequently described as a potential
limitation of these methods (Dobson and Keogh 2007).
This chapter presents a tutorial on how to implement a straightforward and
low-cost method to capture the relative players' positions on the pitch, with a
high degree of flexibility and reliability. The application of this method is
illustrated with the identification of pattern-forming dynamics in three-
versus-three subphases of association football near the scoring areas. The
emphasis is on the description of motion-analysis procedures and data-
analysis methods.
Motion analysis procedures applied to a three-versus-three
smallsided game
To illustrate the motion analysis procedures, a representative experimental
task was developed stimulating the creation of shooting opportunities near
the scoring area in a seven-a-side association football competition of youth
football players (age 11.8 ± 0.4 years). The task consisted of a goalkeeper
plus three-versus-three plus goalkeeper small-sided game, where, to create
shooting opportunities, the attacking team needed to make penetrating passes
into the space behind the defending team (Figure 9.1). The central
performance space was 20 × 20 metres and the scoring areas measured 14.5
metres in length, to simulate the goalkeeper's area according to International
Federation of Association Football (FIFA) rules. Thus, the off-side rule was
associated as a task constraint only inside the scoring areas. Six outfield
players were divided in two teams by their coaches. The small-sided game
lasted for six minutes.

Single-camera recordings
An important methodological issue of the present method is that camera
positioning does not need to be perpendicular with the field of performance;
i.e. the camera lens does not need to be perpendicular in relation to the pitch
plane of motion. Particularly in this experimental task, participants’ on-field
movement displacements were recorded using a regular digital video camera
statically positioned at approximately 30 metres from the pitch, to capture the
whole playing area. Owing to the facilities available in the stadium, the
camera was placed five metres above the ground, perpendicular to the
longitudinal component of the pitch and with an angle of depression of
approximately ten degrees. However, this method can be used with other
camera placements and specifications without losing accuracy (e.g. Duarte et
al. 2010b; Correia et al. 2012; Travassos et al. 2012). Before the beginning of
the experiment, several non-collinear control points corresponding to specific
landmarks visible in the video camera were measured for later calibrations
(see the camera calibration and object-plane reconstruction section below).
Previous studies showed that seven control points were sufficient to obtain
adequate accuracy levels in movement data (Fernandes et al. 2007).
Figure 9.1 Schematic representation of single-camera video motion capture

Image treatment
Video-recorded images of the small-sided game were transferred to digital
support, coded and saved as ‘.avi’ format and goal-scoring opportunities
identified throughout the game. From the 21 goal-scoring opportunities
occurred during the game, only ten plays were analyzed further, in which: (i)
the ball was not projected into an aerial trajectory; and (ii) there were no
changes in ball possession between teams. The video recordings of these
situations were split into unique video files, with minimum file size to
increase the computational performance during extraction of positional
variables and analysis.
The software package TACTO 8.0 (Figure 9.2; Fernandes et al. 2010) was
used to extract the virtual positional coordinates (measured in pixels units)
from participants' movement displacement trajectories. The procedure
consisted of following with a computer mouse cursor the middle point
located between the feet of each participant (chosen as working point). This
working point was used because it represents an estimate of the projection of
the player's centre of gravity on the pitch (Duarte et al. 2010a). Data were
obtained at 25 Hz as recommended by Fernandes and Caixinha (2004). The
TACTO package was also used to assess the virtual coordinates of the seven
control points previously selected that afterwards were used for calibration.
During these procedures, the computer resolution was set at 1280 × 800
pixels and the TACTO 8.0 window was fixed on a permanent position on
screen. It is necessary to keep this procedures unchanged during the
digitization of all trials to avoid improper data transformation owing to
changes in the image plane reference frame.

Figure 9.2 The TACTO 8.0 device window; manual tracking of a selected working point with a
computer mouse allows virtual coordinates of the tracked player/object to be obtained

Camera calibration and object-plane reconstruction

Camera calibration and object-plane reconstruction were made using
bidimensional (2-D) direct linear transformation (2D-DLT) method (Abdel-
Aziz and Karara 1971). This planar analysis used algorithms adapted by
Reinschmidt (1996) from the original DLT method formulated for tri-
dimensional analysis (Woltring and Huiskes 1990). The MATLAB®
(MathWorks Inc.) files used in these steps can be found at the website of the
International Society of Biomechanics (Reinschmidt 1996). The DLT method
is thought to deal reliably with some measurement errors such as optical
distortion and decentring distortion reported by other methods (Marzan and
Karara 1975; Kwon 2008). This method directly relates an object point
located in the object-space/plane and the corresponding image point on the
image-plane of the camera/digitizer unit. Two reference frames are defined:
the object-space/plane reference frame (the XYZ-system) and the image-plane
reference frame (the UV-system; Figure 9.3). The [x, y, z] is the object-space
coordinates of point O (i.e. the pitch coordinates attributed to the working
point of each digitized player in which z coordinates were always equal to
zero), while [u, v] is the image-plane coordinates of the image point I (i.e. the
virtual coordinates obtained with the TACTO package). The mathematical
description of the 2D-DLT method can be found online (Kwon 2003). In the
current study, owing to the use of 2-D analysis, eight DLT coefficients were
automatically determined to reflect the relationships between the object-plane
reference frame (i.e. the pitch coordinates) and the image-plane reference
frame (i.e. the digitized virtual coordinates). This procedure was used twice
for camera calibration and 2-D reconstruction of the players' pitch
coordinates in each time instant (Figure 9.4).

Figure 9.3 The direct linear transformation (2D-DLT) method for camera calibration and bi-
dimensional reconstruction
Figure 9.4 Converted pitch coordinates (metres) allow the reproduction of movement displacement
trajectories of players in the space of action

Quality control procedures

An important methodological issue when dealing with signal processing in
behavioural studies is to ensure that variations in the time-series data (e.g.
Cartesian coordinates of players’ movement displacement trajectories) are
only due to the natural and inherent variability in performance. As mentioned
by Goto and Mascie-Taylor (2007), variability in continuous quantitative data
may be due to the inherent variance but also human measurement error (e.g.
lack of consistence in digitizing) and/or instrumentation error (e.g. lack of
accuracy in the calibration process). Failure to treat these errors properly
results in amplified noisy velocity and acceleration data (Kwon 2008). To
guarantee the fidelity of the obtained positional data earlier described, three
steps were fulfilled: (1) using appropriate data filtering; (2) testing the
accuracy of the instrument; (3) training and testing the consistency/reliability
of the digitations.
Filtering
A widely accepted recommendation when working with kinematic data is the
use of appropriate data filtering to remove some data fluctuations due to
human or instrumentation error (Winter 2005). Signals with predominant
lower frequencies such as the ones used in our experimental task are more
suitable to possess higher frequencies of noise (Giakas 2004). Taking these
recommendations, we used a Butterworth low-pass filter to maintain the
original lower frequencies and remove the higher ones above a specific cut-
off frequency. The original (unfiltered) dataset was compared with different
cut-off frequencies. The similarity between unfiltered and the various filtered
datasets was taken as a criterion to select the cut-off frequency of 6 Hz for all
the analyzed trials based on residual analysis technique (Winter 2005).

Accuracy
To test the accuracy and validity of the instrument for the specific task under
analysis, we used the procedures suggested by Fernandes and Caixinha
(2004). We compared the distances obtained from digitization with a fine
path of a player from which we knew the real distances (i.e. a circuit on the
pitch with changing directions). The mean absolute percentage error showed
values less than five per cent for all the measurements, which were taken as
indicative of high accuracy (Fernandes and Caixinha 2004).

Consistence and reliability

To decrease the human measurement error and increase the internal
consistence of the tracking operator, a digitizing training programme of seven
consecutive days was completed. The protocol consisted of digitizing one
random trial/play per day (i.e. six outfield players involved in each play). On
the seventh day, the tracking operator completed the same protocol
performed in the pretest as posttest measures, digitizing the same players
twice interspersed by a break of five hours (see protocol details in Duarte et
al. 2010a). To assess the internal consistence and reliability between the
digitized trials, we used the coefficient of reliability (R) derived from the
intra-technical error of measurement (intra-TEM) as suggested by Goto and
Mascie-Taylor (2007). This accuracy index is based on the standard deviation
between repeated measures. In this illustrative case, results showed an
improvement from the pretest to post-test measurements. In the pretest,
values of R between the two measurements were higher than 94% for
longitudinal (goal-to-goal) component of motion and 84% for the lateral
(side-toside) component. In post-test measurements, R values were higher
than 95% for both components of motion in all digitized players, which
ensures internal consistence of the tracking operator (Ulijaszek and Kerr
1999). High values of reliability between different tracking operators (i.e.
inter-operator reliability) are also available in the literature (Serrano and
Fernandes 2011).

Variables computation
For the study of team sports as multi-agent dynamical systems, there are
some compound-motion variables that might capture the complex dynamics
of players' interpersonal interactions during competitive performance (see
Chapter 6). With the individual kinematic data of each player obtained by the
aforementioned procedures it is possible to obtain specific compound motion
variables. A previous study identified a high coupling tendency between the
movements of attacking and defending players in the three-versus-three sub-
phase of association football (Duarte et al. 2012b). Thus, based on players’
positional data, a single centroid value (i.e. the geometrical centre) for the six
outfield players was calculated. The centroid or group ‘centre of mass’ was
calculated as the mean position of the six outfield players over time
(Frencken et al. 2011). Next, using the longitudinal component of motion, we
calculated for each time instant the smallest distance of this centroid to the
defensive line (i.e. the boundary line between the central performance space
and the scoring area, see Figure 9.1) as a potential order parameter.
The inter-centroid distance (i.e. distance between the centroid of the
attacking and defending players; Folgado et al. 2012) and the relative stretch
index (i.e. the mean vectorial distances of players to their team centroid;
adapted from Bourbousson et al. 2010) were tested as control parameter
candidate variables. As suggested by Frencken et al. (2011) the inter-centroid
distances tend to decrease immediately before the creation of a shooting
opportunity, suggesting that the closeness of the two sub-groups of players
can influence the stability of the competing team relations. Our purpose was
also to assess whether differences in the relative stretch index of teams
influenced the stability of the relative positioning in reference to the goals.
Specifically conceived MATLAB files were used to compute these time-
series data.
Illustrative data on pattern-forming dynamics in association
football
The exploratory analyses of the centroid distances to the defensive line
suggested the existence of two behavioural states, before and after the instant
of an assisting pass (i.e. the last pass before a shot occurred; see top panel of
Figure 9.5).
Using normalization procedures that did not alter the structure of the
timeseries data, we subtracted each value of the order parameter from the
own mean (Rosenblum et al. 2001). This procedure allowed us to find a mean
point between the two states corresponding to values of zero (see bottom
panel of Figure 9.5). Thus, the first state displayed positive values of the
order parameter corresponding to the initial stable relations between teams
(i.e. where the defending sub-group successfully interacted together to
prevent shooting opportunities). The second state displayed negative values
of the order parameter that emerged from an arrangement of interpersonal
relationships that led to defensive system instabilities and goal-scoring
opportunities. The qualitative changes between the two identified states
observed in the bottom panel of Figure 9.5 by the shift from positive to
negative values suggested that the emergence of instabilities within these
small-groups (i.e. changes in their relative positioning leading to penetrations
in the scoring areas) may be characterized by some nonlinear properties such
as order–order transition (Kugler and Turvey 1987).
Analysis of first derivative values of the order parameter also revealed a
high rate of change at the moment prior to the appearance of system
instabilities (Figure 9.6). These results reinforced the idea that transitions
between the two identified states occurred suddenly from one state to
another, and not by a cumulative linear process.
Figure 9.5 The two behavioural states of the order parameter; top panel shows the distance of the single
centroid to the defensive boundary line of all the analyzed trials, synchronized by the assistance pass
instant; bottom panel shows the order parameter subtracted by the own mean to highlight the qualitative
changes associated with perturbations of the initial stability of teams (see text for details)

To better understand how the stability of the competing teams was

perturbed in the studied sub-phases of play, we tested whether the two control
parameters influenced the stability of the order parameter. Inter-centroid
distances revealed a trend for consistently present lower values (minimum of
1.89 ± 1.0 metres) at the instants immediately before the loss of stability in
teams (see top panel of Figure 9.7). Standard deviations also showed a
convergence towards low values at these instants. As these values were
observed only at these instants of play, it suggests that a decrease in the
distance between teams seemed to influence the emergence of instabilities in
the three-versus-three sub-phases near goal scoring areas.
Figure 9.6 Identification of the nonlinear qualitative changes between the two behavioural states; top
panel displays an exemplar play with increased slope in the transition between the two system states
(down-sampled to 1 Hz); bottom panel shows the average and standard deviation bands of the first
derivative of the order parameter in all plays, highlighting with an ellipse the high rate of change of the
order parameter; the slope or high rate of change indicates the order–order transition

Relative stretch index measures demonstrated a trend for increasing

positive values (2.56 ± 1.4 metres), with increased stability (low standard
deviation) before the perturbation of system stability (see bottom panel of
Figure 9.7). Positive values meant that attacking teams were more stretched
than defending teams. This finding implies that the instabilities characterizing
the movement of these small groups into the goal-scoring areas were
associated with a high dispersion of the attackers compared to the spatial
arrangement of the defending players.
Figure 9.7 Moving average and standard deviation values of the inter-centroid distances (left panel) and
relative stretch index (right panel); vertical dashed lines highlight the instant of the instabilities
corresponding to the zero crossing of the order parameter previously identified in Figure 9.5
Concluding remarks and future implications
This chapter outlined a low-cost and straightforward method to reconstruct
the players' movement displacement trajectories in the performance context
using a single camera. The high values of accuracy and reliability, as well as
its flexibility, suggest the possibility of using this method in a wide range of
training and competitive performance settings. Special care is recommended
with signal processing in which appropriate data filtering and short-term
digitization training programmes should be undertaken to remove high-
frequency content noise (i.e. human and instrumentation error) from the
inherent performance variability.
Obtaining positional data from players' on-field performance allowed us to
identify the pattern forming dynamics emerging from players' interactions
during a three-versus-three sub-phase of association football near the scoring
areas. A functional relevant order parameter was proposed based on previous
qualitative examinations of collective movement data. Sudden transitions
associated to qualitative behavioural changes were identified in this order
parameter (i.e. distance from a single centroid to the defensive boundary line)
by the influence of two intertwined control parameters (i.e. the inter-centroid
distance and relative stretch index).
More specifically, low values of inter-centroid distance and increasing
positive values of relative stretch index were shown to influence the
emergence of instabilities in the small-sided game that led to goal-scoring
opportunities. The complementary relations between these two control
parameters suggested that goal-scoring opportunities created at this level of
interpersonal interactions resulted from a combined influence of the attacking
and defending sub-groups. This influence was characterized by an
approaching of the two sub-groups to each other and an increase in the
difference between the dispersion values of the attacking and defending
players. Folgado et al. (2012) demonstrated that more skilled footballers
achieved higher inter-centroid distances during three-versusthree small-sided
games compared to less-skilled players. In the illustrative experiment
presented in this chapter, it was observed that inter-centroid distance values
tended to decrease to the lowest values only in the moments before the loss of
stability between the small groups of attacking and defending players. These
findings are in agreement with the results of Frencken et al. (2011) on
fourversus- four small-sided games, reporting a trend towards a decrease in
this variable when teams scored goals. Concerning the relative stretch index
data, Bourbousson et al. (2010) observed that basketball teams followed a
counterphase relation as a function of changes in ball possession. However,
in the plays presented here, no changes in ball possession between teams
occurred. The presented analyses revealed that the stability of the game was
perturbed only when the relative dispersion of the players showed a
consistent pattern, indicated by the low standard deviation towards increasing
positive values.
Analysis of group motion dynamics of attacking and defending sub-phases
of performance can allow to identify pattern forming dynamics of players'
interactions related to the emergence of instabilities in sub-groups relations,
such as goal-scoring opportunities. Coaches and practitioners can enhance
their training programmes by using these data to constrain small-sided games
for training interpersonal interactions during attacking and defending sub-
phases of team sport. On the other hand, researchers can use this approach
from nonlinear dynamical systems to investigate how key relational system
variables evolve over time and influence the interactional behaviour of team
players during performance.
References
Abdel-Aziz, Y. I. and Karara, H. M. (1971) Direct linear transformation from
comparator coordinates into object space coordinates in close-range
photogrammetry, in Proceedings of the Symposium on Close-Range
Photogrammetry. Falls Church, VA: American Society of Photogrammetry,
pp. 1–18.
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7 (6): 653–76.
Barris, S. and Button, C. (2008) A review of vision-based motion analysis in
sport. Sports Medicine, 38 (12): 1025–43.
Bourbousson, G., Seve, C. and McGarry, T. (2010) Space-time coordination
dynamics in basketball: Part 2. The interaction between the two teams.
Journal of Sports Sciences, 28: 349–58.
Correia, V., Araújo, D., Duarte, R., Travassos, B., Passos, P. and Davids, K.
(2012) Changes in practice task constraints shape decision-making
behaviours of team games players. Journal of Science and Medicine in Sport,
15 (3): 244–9.
Davids, K., Araújo, D. and Shuttleworth, R. (2005) Applications of
dynamical systems theory to football, in T. Reilly, J. Cabri and D. Araújo
(eds) Science and Football V. Abingdon: Routledge, pp. 547–50.
Davids, K., Button, C., Araújo, D., Renshaw, I. and Hristovski, R. (2006)
Movement models from sports provide representative task constraints for
studying adaptive behavior in human movement systems. Adaptive Behavior,
14: 73–94.
Di Salvo, V., Baron, R., Tschan, H., Calderon Montero, F., Bachl, N. and
Pigozzi, F. (2007) Performance characteristics according to playing position
in elite soccer. International Journal of Sports Medicine, 28: 222–7.
Dobson, B. P. and Keogh, J. W. L. (2007) Methodological issues for the
application of time-motion analysis research. Strength and Conditioning
Journal, 29 (2): 48–55.
Duarte, R., Araújo, D., Fernandes, O., Fonseca, C., Correia, V., Travassos,
B., Esteves, P., Vilar, L. and Lopes, J. (2010a) Capturing complex human
behaviors in representative sports contexts with a single camera. Medicina-
Lithuania, 46: 408–14.
Duarte, R., Araújo, D., Gazimba, V. and Fernandes, O. (2010b) A time-
motion analysis method to study people interactions in human movement
science, in Conference Proceedings of the 3rd Mathematical Methods in
Engineering International Symposium. Coimbra: Polytechnic Institute of
Coimbra, pp. 408–15.
Duarte, R., Araújo, D., Correia, V. and Davids, K. (2012a) Sport teams as
superorganisms: implications of sociobiological models of behaviour for
research and practice in team sports performance analysis. Sports Medicine,
42 (8): 633–42.
Duarte, R., Araújo, D., Freire, L., Folgado, H., Fernandes, O. and Davids, K.
(2012b) Intraand inter-group coordination patterns reveal collective
behaviours of football players near the scoring zone. Human Movement
Science, 31 (6): 1639–51.
Fernandes, O. and Caixinha, P. (2004) A new method of time–motion
analysis for soccer training and competition. Part II: Game activity and
analysis. Journal of Sports Sciences, 22 (6): 505.
Fernandes, O., Caixinha, P. and Malta, P. (2007) Techno-tactics and running
distance analysis by camera. Journal of Sports Science and Medicine, 6
(Suppl.10): 204–5.
Fernandes, O., Folgado, H., Duarte, R. and Malta, P. (2010) Validation of the
tool for applied and contextual time-series observation. International Journal
of Sport Psychology, 41: 63–4.
Folgado, H., Lemmink, K., Frencken, W. and Sampaio, J. (2012) Length,
width and centroid distance as measures of teams tactical performance in
youth football. European Journal of Sport Science, 1–6, iFirst article; doi:
10.1080/ 17461391.2012.730060
Frencken, W., Lemmink, K., Delleman, N., Visscher, C. (2011) Oscillations
of centroid position and surface area of soccer teams in small-sided games.
European Journal of Sport Science, 4: 215–23.
Giakas, G. (2004) Power spectrum analysis and filtering, in N. Stergiou (ed.)
Innovative Analyses of Human Movement. Champaign, IL: Human Kinetics,
pp. 223–58.
Goto, R. and Mascie-Taylor, C. (2007) Precision of measurement as a
component of human variation. Journal of Physiological Anthropology, 26:
253–6.
Kelso, J. A. S. (2009) Coordination dynamics, in R. A. Meyers (ed.)
Encyclopedia of Complexity and System Science. Berlin: Springer, pp.1537–
64.
Kwon, Y. H. (2003) DLT method. Available online at
www.kwon3d.com/theory/dlt/dlt.html (accessed 7 June 2013).
Kwon, Y. H. (2008) Measurement for deriving kinematic parameters:
numerical methods, in Y. Hong and R. Bartlett (eds) Handbook of
Biomechanics and Human Movement Science. Abingdon: Routledge, pp.
156–81.
McGarry, T., Khan, M. and Franks, I. (1999) On the presence and absence of
behavioural traits in sport: an example from championship squash match-
play. Journal of Sports Sciences, 17: 297–311.
Marzan, G. T. and Karara, H. M. (1975) A computer program for direct linear
transformation solution of the collinearity condition, and some applications
of it, in Proceedings of the Symposium on Close-Range Photogrammetric
Systems. Falls Church, VA: American Society of Photogrammetry, pp. 420–
76.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Milho, J. and Serpa, S.
(2008) Informationgoverning dynamics of attacker–defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Serpa, S., Milho, J. and
Fonseca, S. (2009) Interpersonal pattern dynamics and adaptive behavior in
multiagent neurobiological systems: conceptual model and data. Journal of
Motor Behavior, 41: 445–9.
Reinschmidt, C. (1996) Movement Analysis Software (2D-DLT) [Matlab
code]. International Society of Biomechanics. Available online at
http://isbweb.org/software/ movanal.html (accessed 7 June 2013).
Rosenblum, M., Pikovsky, A., Kurths, J., Schafer, C. and Tass, P.A. (2001)
Phase synchronization: from theory to data analysis, in R. Lipowsky, E.
Sackmann and A. J. Hoff (eds) Handbook of Biological Physics. Leiden:
Elsevier, pp. 279–321.
Serrano, J. and Fernandes, O. (2011) Reliability of a new method to analyse
and to quantify athletes’ displacement. Portuguese Journal of Sport Sciences,
11, (Suppl 2): 935–6.
Travassos, B., Araújo, D., Duarte, R. and McGarry, T. (2012) Spatiotemporal
coordination patterns in futsal (indoor football) are guided by informational
game constraints. Human Movement Science, 31 (4): 932–45.
Ulijaszek, S. A. and Kerr, D. A. (1999) Anthropometric measurement error
and the assessment of nutritional status. British Journal of Nutrition, 82: 165–
77.
Winter, D. A. (2005) Biomechanics and Motor Control of Human Movement,
3rd edn. New York: John Wiley and Sons.
Woltring, H. J. and Huiskes, R. (1990) Stereophotogrammetry, in N, Berme
and A, Capozzo (eds) Biomechanics of Human Movement. Worthington, OH:
Bertec Corporation, pp. 108–27.
10 Using virtual environments to
study interactions in sport
performance
Vanda Correia, Duarte Araújo, Gareth
Watson and Cathy Craig
This chapter highlights the potential of immersive, interactive virtual reality
technology to study dynamical interactions in sport performance. In view of
the continuous advances in this technology, the dilemma between
maintaining task representative design and guaranteeing experimental control
has ceased to exist. The advantages and disadvantages of this methodological
tool for the design of both practice and research-based experimental tasks that
intend to improve and investigate the dynamics of environment–agent
systems and information–action coupling in sports, are reviewed and
discussed. Virtual reality technology is highlighted, as it allows us to capture
the complexity of actions yet to control the presentation of information
(mainly visual and auditory but also haptic/tactile) that guides action.
Although the use of virtual reality environments has been extensively used
to investigate social interactions in many different fields of research (e.g.
neuroscience, journalism, therapy in treatment of psychological disorders)
only a few researchers have addressed its potential application to sport (e.g.
Bideau et al. 2010; Watson et al. 2011). In this chapter, we review and
discuss the advantages and disadvantages of this methodological tool for
improving and investigating the dynamics of environment–agent systems in
sports. That is to say, how this technology could be used to enhance sports
performance from a complex sciences perspective, by allowing for the
examination of individual athletes and sports teams’ behaviours as they
emerge as a result of the informational constraints imposed by and controlled
in the virtual environment. In other words, how the perceptually based
information guides decisions about how and when to act.
We will outline the importance of examining interpersonal goal-directed
interactions under simulated dynamic task constraints that warrant the active
exploration of information. This exploration is important to continuously
guide on-going action. We emphasize also how crucial it is to simultaneously
guarantee that relevant action information present in the ‘real’1 context of
performance is also available in virtual contexts to support functional
behaviour. In view of that, we provide illustrative empirical evidence of
recent applications of virtual environments for sports research, addressing the
implications for the design of both practice and research related tasks.
Dynamic interactions in sport performance
Players in teams are constantly interacting, co-adapting towards shared (i.e.
between teammates) and opposing (i.e. between opponents) performance
goals (e.g. Marsh, Richardson, Baron, and Schmidt 2006; Passos et al. 2011;
McGarry, Anderson, Wallace, Hughes, and Franks, 2002). Multiple
constraints are responsible for multiple effects in both interpersonal
interaction processes and outcomes (e.g. Passos 2010). That is to say, that
different dynamic interactions and consequent performance outcomes may
emerge as a function of many interacting individual, task and environment
constraints. Namely, in team ball games, players display functional
coordination tendencies, based on local interaction rules acting during
performance (Passos et al. 2011). These local interaction rules are context
dependent, expressing the performance task constraints influencing players’
behaviours (Passos et al. 2011).
The emerging interpersonal coordination patterns are functional
spatiotemporal adjustments or adaptations to changing circumstances (Araújo
et al. 2006; Kelso 1995; Kelso 2002; Warren 2006). In view of this fact, a
growing body of research on sport performance has been concerned with
interpersonal interactions phenomena (e.g. Araújo et al. 2004; Correia et al.
2012a; Esteves et al. 2011; Passos et al. 2008). Interpersonal coordination
emerges from interactions during performance, shaped by contextual,
individual and task constraints (Araújo et al. 2004; McGarry and Franks
2007; Passos et al. 2008). It is important to note that behaviour emergent
from the interaction of a subject with the context can be expressed by the
action and transitions in the course of the action (Araújo et al. 2006). It is
thus important to formalize this dynamic behaviour in ‘real contexts’ and to
identify potential informational constraints in these ‘less controlled’ in situ
situations, which can be then used as manipulated parameters or variables in
‘virtual environments’ to help not only to describe but also to explain how
emergent behaviour emerges under (manipulated) informational task
constraints.
Virtual environment technology
Virtual environment technology has been used for the simulation of
domainspecific environments of human behaviour in research, training and
education. A virtual environment consists of an artificial (i.e. synthesized)
environment generated using software and is mostly experienced via vision
and sound. This approach has been successfully applied to the study of
human behaviour in varying domains, such as neuroscience (e.g. Bohil et al.
2011; Tarr and Warren 2002); journalism (e.g. de la Pena et al. 2010);
therapy in the treatment of anxiety disorders (e.g. Gerardi et al. 2008; Villani
et al. 2007). The key element of a virtual environment system is the visual
display. This display can either be an enclosed head-mounted display (HMD;
Figure 10.1) or an open display such as a computer monitor or projection
screen. A higher level of immersion is possible with the HMD. Owing to the
diversity of HMDs, virtual environment systems vary with respect to their
field of view, resolution and weight, amongst other characteristics (see, e.g.
Stanney et al. 1998). Alongside, and sometimes integrated within the same
visual display device (e.g. as part of the HMD), auditory stimulation is often
used in the form of three-dimensional spatial surround sound (Bohil et al.
2011). The use of devices providing haptic or tactile feedback is also
increasing.
Figure 10.1 Illustrative immersive interactive virtual reality apparatus in the Movement Innovation
Laboratory at Queen's University, Belfast. It shows a participant wearing a pair of gloves with attached
hand trackers, a head-mounted display with attached head tracker and a back-pack housing the control
unit

Although participants' performance in the virtual environment simulated

task may be captured by means of a keyboard, mouse, joystick or other
device, the use of immersive and interactive virtual environments is growing.
Immersive interactive virtual environments must thus incorporate highly
sensitive tracking systems with head or body tracking devices (fixed to the
head or body segments, such as the hands, see Figure 10.1). These systems
are used to update in real time an egocentric viewpoint, besides allowing for
the capture of the action being performed. That is, these tracking systems
monitor in ‘real’ time (i.e. online) behavioural responses (e.g. biomechanical
data). Any movement in the real world is therefore updated in the virtual
context (i.e. data from the tracking system updates the participant's or user's
viewpoint of the unfolding action in the virtual context). Online updating
using a tracking system and HMD means that the participant is fully
immersed (i.e. they never leave the virtual world, even by averting their
gaze). Of note here is that the participants are allowed to choose where they
look in the virtual context. These features of recent virtual environment
technology help to enhance levels of immersion. A number and assortment of
sensory and motor channels linked to the virtual environment, which
simultaneously allows for the matching of motor inputs (such as unrestricted
body movements), amplifies feelings of immersion (Bohil et al. 2011).
The discussion of assumptions, concepts and underlying paradigms or
ontological positions is beyond the scope of this chapter but we briefly
mention some issues related to virtual environment use. A notion commonly
related with virtual environment simulations is that of presence, otherwise
known as subjective experience or experiential fidelity (Riccio 1995;
Stoffregen et al. 2003). It may be said that ‘the psychological product of
technological immersion’ (Bohil et al. 2011, p. 754) and its influence on
performance is regularly argued (e.g. Bideau et al. 2010; Craig and Watson
2011; Krijn et al. 2004; Villani et al. 2007). Presence is generally evaluated
by the use of subjective measurements (i.e. dependent on the expertise of the
researcher in using such instruments) such as self-reports, questionnaires and
interviews, and is based on attribute category rating scales (e.g. Sheridan
1992, 1996; Wagner et al. 2009) but also uses objective measurements such
as physiological indicators (Bohil et al. 2011; Wagner et al. 2009).
Another important issue is the notion of action (or functional) fidelity (e.g.
Moroney and Moroney 1998; Stoffregen et al. 2003). Action fidelity is the
relationship between behaviour in the simulator and behaviour in the
simulated system and it is essentially measured in terms of task performance
(Stoffregen et al. 2003). That is, it can be assessed by measures such as time
to perform the task, emergent action similar to that observed in the situation
being simulated. The simulation should primarily simulate the functions (i.e.
it should demand the same type of interaction and the same goals) of the
simulated context rather than prioritizing the ‘fidelity of the stimuli’ (Araújo
et al. 2005; Stoffregen et al. 2003). The subjective feeling of being there (i.e.
presence or experiential fidelity) can be seen as more concerned with
enhancing stimulus fidelity (of the virtual environment) (Stoffregen et al.
2003). Conversely, action fidelity puts emphasis on simulation adequacy in
terms of research, training and learning purposes (Stoffregen et al. 2003).
Therefore, it must be guaranteed that the functions or action demanded by the
simulation are similar to those of the simulated (natural) context of
performance.
The virtual environment has many advantages over video presentation or
natural practice or game performance situations for improving our
understanding of the complexity of action in sport (see, e.g. Vignais et al.
2009). The virtual environment is about simplifying the complexity of sport
by computer generating performance contexts and thus reducing the
multitude of variables that may be influencing the emergent behaviour. In
spite of this, it can be seen as shedding light on the complexity of action by
exerting control over some variables and observing their effect on the
dynamics of behaviour that emerges likewise in the complex natural context.
One of the advantages is based on the ability to precisely control the
presentation of sensory information (mainly visual and auditory but also
haptic/tactile) that is involved in the guidance of action. It also makes it
possible to couple and decouple perception and action (Craig et al. 2011;
Craig and Watson 2011). As Bohil et al. (2011) have stated in a recent
review, ‘Virtual reality provides a middle ground, supporting naturalistic and
contextually rich scenarios along with an exacting degree of control over key
variables’ (p. 752). This technology makes it possible to simulate a varying
number of performance situations using immersive, interactive virtual
environments. That is, visual information made available (manipulated) can
be displayed to both stationary and moving participants by means of
combined use of the HMD and motion-tracking systems.
Advantages of virtual environments over screen displays or video playback
have also been pointed out (Bideau et al. 2010). Some attempts have been
made to improve the recorded viewpoint of the match (e.g. in tennis: Farrow
and Abernethy 2003; Williams et al. 2002; in soccer: Savelsbergh et al. 2002;
Petit and Ripoll 2008). That is, this research attempts to provide participants
with the perspective of the game nearest to what they would experience if
they were performing in the game. However, the viewpoint is still not
controlled by the observer (i.e. an egocentric viewpoint). An egocentric
viewpoint refers to the naturalistic interaction between the head movements
of the observer and the updating in real time of the optical information
presented to the eyes of the observer. In other words the observer can ‘move
to perceive and perceive to move’ (Gibson 1979). In addition to the automatic
updating of viewpoint, the virtual environment allows for a binocular field of
vision or stereopsis (i.e. visual field calculated for each eye). In addition, it
allows for complete control of the information presented to performers and
the movement trackers or controllers allow for diverse types of behavioural
data to be captured (e.g. Craig and Watson 2011; Bideau et al. 2010; Brault et
al. 2012).
A potential limitation of these studies is the lack of tactile/force feedback
that may reduce performance on the virtual environment simulated task.
When a cable is used to connect the visual display device to the central
computer, a limitation that may be pointed out is the eventual restricted
movement (McMenemy and Ferguson 2007). Another potential limitation is
that the HMD may affect distance and depth perception (e.g. Willemsen et al.
2009) and the limited field of view restricts the level of peripheral vision that
is available to the observer (e.g. Knapp and Loomis 2004). Finally,
accessibility, costs and the required specialized technical assistance may also
limit the use of this technology (e.g. Rizzo and Kim 2005).
Research on expertise in sport, namely on visual perception in sport, has
focused on what discrete ‘cues’ may be used by individuals for effective
performance. The information sources under study are commonly hidden by
means of temporal and/or spatial occlusion paradigms and motor responses
are combined with eye movement registration techniques (e.g. Müller and
Abernethy 2006; Vaeyens et al. 2007; for a review see, e.g. Williams and
Ericsson 2005). To investigate information sources used for perceptual
judgments and actions, researchers have also made use of virtual
environments to manipulate information in a continuous rather than a discrete
fashion (e.g. Vignais et al. 2009; Bideau et al. 2003, 2004). Investigations in
sport carried out using a virtual environment have demonstrated the influence
of spatiotemporal optical invariants as information that shapes actions (e.g.
restricted movements, e.g. Dessing and Craig 2010) and judgments on actions
to be performed (e.g. by button pressing; e.g. Watson et al. 2011; Craig et al.
2006, 2009). However, more work is needed to understand how performers
interact and create goal-directed adaptive behaviour, coupled with the many
dynamic information sources in performance contexts (Correia et al. 2012).
This remits to the importance of guaranteeing representative design for
sport psychology, practice and experimental design (Araújo et al. 2005;
Pinder et al. 2011). Representative design (put forward by Brunswik 1956;
see also Hammond and Stewart 2001) means that experimental task
constraints, representative from the context for which the findings are
intended to be generalized are essential in investigating performance in sports
but also for the design of training and evaluation tasks meant to improve
players’ and teams’ performance (Araújo et al. 2006; Davids et al. 2007;
Pinder et al. 2011). An indicator that a task situation might not be
representative of a specific domain is that expert performance can drop to the
level of novices (Allard et al. 1980; Araújo et al. 2005). As virtual
environment technology progresses, it has progressively become more
financially accessible to research laboratories and, as a result, the potential
dilemma of task representative design versus experimental control grows
fainter and more outdated.
We next depict the potential contribution of immersive and interactive
virtual reality setups to investigate behaviour in various sport performance
tasks and contexts. To accomplish this, we consider illustrative empirical
studies that have applied this technology to sports.
Designing virtual contexts of representative interpersonal
interactions

Coupling perception and action

Gibson (1979) advocated that behaviour is grounded on perception–action
couplings. That is, information is picked up from the surrounding
environment and optic flow (i.e. the pattern of light generated by a particular
animal–environment interaction). Optic flow is generated by action so that
visual perception guides the action and, in turn, the action changes
perception. It has been suggested that virtual environments are good tools for
studying the perception– action loop in athletes (e.g. Bideau et al. 2010;
Craig and Watson 2011) by simulating the perceptual information and
measuring the resulting actions. Simulations of performance contexts and
situations must allow participants to explore the task constraints so that they
can pick up relevant information to guide goal-directed behaviour. In
immersive, interactive virtual environment setups, movement through the
optic array is simulated. This means that by moving, participants bring about
transformations in the optic array to create an optic flow. It is this detection
of the transformation in the optic array (i.e. picking up relevant information)
that guides movement. This is accomplished by means of the aforementioned
tracking systems that track the head and body so that the visual display
changes accordingly. This advanced virtual environment setup guarantees
‘forces resulting in flows and flows resulting in forces – of perception
entailing action and of action entailing perception’ (Richardson et al. 2008, p.
174). Hence, the specific perception–action couplings are maintained (Araújo
et al. 2005). Representative functional interactions arise when participants
continuously act to perceive information from the visual context of
performance and this information in turn shapes action. Experimental
designs, including using virtual contexts, should maintain specific and
functional perception–action couplings and allow and potentiate interpersonal
coordination.
A study by Watson et al. (2011) used an immersive interactive
environment with a button-pressing response task to investigate ‘pass-
through-ability’ in rugby union. Though decoupling action responses from
perception by relying on judgments alone (button-press responses), this was
an important study in highlighting how virtual environment technology can
be used to understand affordances that emerge in dynamic rugby situations
and how they may be defined in terms of the spatiotemporal dynamics of the
optics of the unfolding events.
Importantly, existing technological advances are keeping with the
ecological psychology hallmark of perception–action coupling, given that
participants are able to act instead of ‘perceive only’; i.e. they can move and,
for this forward movement, optic flow information is provided. For instance,
in the study by Dessing and Craig (2010) on how goalkeepers save curved
free kicks, hand movements in the direction of the approaching ball were
considered to be measures of task performance. In a recent study by Correia
et al. (2013), the influence of emerging gaps between defenders on a ball-
carrier's choice of action was investigated. In this experiment, participants
acted freely (owing to the task characteristics, they could essentially run or
pass the ball to one of two attacking teammates). Note, though, that it is
important to avoid a discrete look at actors performing discrete actions when
faced with virtual simulations of natural performance tasks; i.e. not only
considering the actions performed but their dynamics and how they are
related to the unfolding information dynamics. There is a lack of systematic
research about how performers are continuously using information
throughout the dynamic course of goal-directed action (Correia et al. 2013).
Although the action informs about perception, thus far, continuous action or,
better, the dynamics of that action has not been considered in sport research.

Dynamics of the athlete–environment system

It is important to consider that goal-directed action in interpersonal
interactions is dependent on initial conditions, emerging from perception–
action couplings under the influence of the local constraints and that
understanding the resultant variability is a key feature of behaviour (e.g.
Davids 2009; Newell 1986). This means that more individualized analyses
may be needed and the virtual environment may be a fine tool to accomplish
this. For instance, this could be investigated on different timescales of sport
performance, learning and development. Besides, the emphasis on constraints
in understanding individual differences must consider not only the
perception–action coupling but also the degeneracy property. Degeneracy
essentially concerns the capacity to make use of structurally different
components to achieve the same functional outcomes (Edelman and Gally
2001; Hong and Newell 2006; Davids et al. 2006). This property expresses
the flexibility and adaptability to fit task constraints for performance goal-
achievement (Edelman and Gally 2001; Davids et al. 2006) and may be
investigated by means of the manipulation of virtual environment task
constraints. That is, it is possible, by having full control over the
manipulation of task features, such as the kinematics of teammates and
opposing players’ behaviours, to assess the same individuals exploitation of
structurally different components to achieve the same functional outcomes.
The perception of what action is possible in a given task is also said to be
body and action scaled (e.g. Warren 1984; Warren and Whang, 1987;
Oudejans et al. 1996; Turvey et al. 1999; Marik 1987). Dimensionless body
ratios and individual action capabilities (i.e. intrinsic functional properties of
the individual) have been shown to relate directly to the environmental
properties. This individual scaling may also be considered in the build-up of
sports interaction research by means of virtual environment technology.
Online updated interaction between the participant's actions and the virtual
performance context are still scarce. Although there is a functional
relationship between the participants and the virtual context, the virtual
environment does not allow for true reciprocity as in the simulated athlete–
environment system. The influence of the environment on the participant is
present in most immersive interactive virtual environment designs. However,
the influence of the participant on the environment, such as on the behaviour
of other players involved in the virtual environment task (avatars), is not.
This should not be a problem, given that the great advantage of these systems
over the natural performance context is actually to have control over the
visual context and information made available (e.g. avatar movements).
One foremost characteristic is that this kind of setup overcomes some
difficulties that are found in situ, such as the impracticality of examining
sport-specific decision making of non-players (e.g. studying non-rugby
players in a rugby task where significant physical contact is allowed by the
game rules; Araújo et al. 2005).
Empirical evidence has also been provided supporting the role of visual
and nonvisual sensory information in the perception and realization of the
possibilities of action (e.g. Stoffregen and Riccio 1988). Emergent behaviour
expressed in interpersonal sport interactions depends also on multiple sensory
modalities, including information from both vestibular and kinaesthetic
modalities. Technological possibilities of a multimodal interface combining
threedimensional visual and sound displays, together with other sensory
modalities would be helpful for better investigating interpersonal interactions
in complex systems such as sport.

Figure 10.2 A schematic representation of what the ball-carrying participant could see in front of
him/her (i.e. the defensive line) in a virtual environment simulated three-versus-three rugby task
Implications of virtual environments for sport performance
research and practice organization
There is still a need to bridge the natural–virtual context gap. For instance,
descriptive or experimental investigations undertaken in the field of
performance may provide evidence of how the key constraints in natural
contexts influence behaviour. By way of virtual contexts (use of immersive
and interactive virtual reality technology), a coach, a psychologist or
player/athlete, may reproduce and experience those performance challenges
and thus potentiate not only the diagnostic of performance but also the
subsequent intervention (Craig et al. 2011; Watson et al. 2011). Virtual
environments could be helpful in off-field training in cases of injury, to
increase training load without significant increases in fatigue or even when
there are no facilities available for training because of weather conditions.
Key constraints on performers' behavioural events and outcome found in
natural contexts of performance, such as interpersonal distance and relative
velocity (Duarte et al. 2010; Passos et al. 2008), initial interpersonal distance
(Correia et al. 2012a) could be reproduced and manipulated in virtual
contexts. This potentially facilitates practising off field some aspects of
performance, which, owing to various aspects (e.g. excessive opportunities to
be tackled in the natural context may restrict the number of trials suitable to
avoid injury and excessive fatigue), cannot be performed with such a high
frequency on the field.
In tasks designed for both research and practice, the possibility of actually
intercepting or running through spaces, for example, must be provided to the
participants. Designed tasks must also contain objects in the virtual
environment context with properties that are relevant to an individual's
purpose within the task (i.e. that afford desired skills and adaptive and
functional behaviours).
As mentioned before, the types of cross-modal interactions that take place
during direct perception in the natural context should also be exploited.
Moreover, to our knowledge, immersive and interactive virtual environment
tasks applied to sports research has focused on the study of one participant
interacting with an entirely controlled a priori virtual simulated performance
context (including the behaviour of other players). Taking advantage of the
advance in technology, such as tracking systems and vision display devices,
further research could use more elaborated virtual worlds and could
investigate also the interpersonal coordination between players and teams, by
immersing more than one participant in the virtual environment simulated
sport task; i.e. to investigate interactions between teammates or opposing
players when facing pre-established movements of the other virtual players
involved. Although this might be not possible at the moment, it is certainly
being worked on (e.g. a simple Kinect4® or a Nintendo Wii®, has interaction
between players/participants) and this is of great interest for furthering this
area of research.
Considering the dynamics in the individual–environment relationship and
scaled to the individual (Marik 1987; Oudejans et al. 1996; Warren 1984;
Warren and Whang 1987; Turvey 1999), we may incorporate these
characteristics within virtual environments. For instance, when catching a
ball in a virtual context, the participant could see a longer arm or a faster arm
movement, and participants' height or jumping reach could be changed. This
would not be possible except through virtual means.
Conclusion
The potential for studying behaviour in sport through virtual environment
technology has been outlined. One of the main advantages it offers is to
experimentally control the information available and examine how this
information guides action in sport. Although such technologies are not yet
widely accessible, large displays and body-based interactive devices are
becoming increasingly advanced and accessible (e.g. Kinect4, Nintendo
Wii®), and immersive and interactive sport situations will be easier to
simulate while maintaining their functional fidelity.

Questions for students

1. How is perception–action reciprocity assumed to be sustained in a
virtual environment?
2. What role do tracking systems and displays play in providing a
realistic representation of a sporting context? How does the level of
immersion differ between an HMD-based display and a
desktop/screen-based projection?
3. What are the advantages and disadvantages of using immersive,
interactive virtual environment over more traditional methods (e.g.
video or eye tracking) when trying to understand performance?
4. How can you recreate a representative sports experimental setup in
a virtual environment?
5. How do you think virtual environment technology should be used
to improve performance in sport?
Note
1 We avoid using the term real context or real behaviour because interacting with virtual contexts
does not imply unreal behaviour or that the virtual context is unreal.
References
Allard, F., Graham, S. and Paarsalu, M. (1980) Perception in sport:
basketball. Journal of Sport Psychology, 2: 14–21.
Araújo, D., Davids, K., Bennett, S., Button, C. and Chapman, G. (2004)
Emergence of sport skills under constraints, in A. M. Williams and N. J.
Hodges (eds) Skill Acquisition in Sport: Research, Theory and Practice.
Abingdon: Routledge, pp. 409–34.
Araújo, D., Davids, K. and Serpa, S. (2005) An ecological approach to
expertise effects in decision-making in a simulated sailing regatta.
Psychology of Sport and Exercise, 6 (6): 671–92.
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7 (6): 653–76.
Bideau B., Kulpa, R., Menardais, S., Fradet, L., Multon, F., Delamarche, P.
and Arnaldi, B. (2003) Real handball goalkeeper vs. virtual handball thrower.
Presence: Teleoperators and Virtual Environments, 12 (4): 411–21.
Bideau, B., Multon, F., Kulpa, R., Fradet, L., Arnaldi, B. and Delamarche, P.
(2004) Using virtual reality to analyse links between handball thrower
kinematics and goalkeeper's reactions. Neuroscience Letters, 372: 119–22.
Bideau, B., Kulpa, R., Vignais, N., Brault, S., Multon, F. and Craig, C.
(2010) Using virtual reality to analyze sports performance. IEEE Computer
Graphics and Applications, 30 (2): 14–21.
Bohil, C. J., Alicea, B., and Biocca, F. A. (2011) Virtual reality in
neuroscience research and therapy. Nature Reviews, 12: 752–62.
Brault, S., Bideau, B., Kupla, R. and Craig, C. M. (2012) Detecting deception
in movement: the case of the side step in rugby. PLoS ONE, 7 (6): e37494;
doi: 10.1371/journal.pone.0037494 (accessed 7 June 2013).
Brunswik, E. (1956) Perception and the Representative Design of
Psychological Experiments, 2nd edn. Berkeley: University of California
Press.
Correia, V., Araújo, D., Cummins, A. and Craig, C. (2012) Perceiving and
acting upon spaces in a VR rugby task: expertise effects in affordance
detection and task achievement. Journal of Sport and Exercise Psychology,
34: 305–21.
Correia, V., Araújo, D., Vilar, L. and Davids, K. (2013) From recording
discrete actions to studying continuous goal-directed behaviours in team
sports. Journal of Sports Sciences, 31 (5): 546–53.
Craig, C. and Watson, G. (2011) An affordance based approach to decision
making in sport: discussing a novel methodological framework. Revista
Psicologia del Deporte, 20 (2): 689–708.
Craig, C., Berton, E., Rao, G., Fernandez, L. and Bootsma, R. (2006) Judging
where a ball will go: the case of curved free kicks in football.
Naturwissenschaften, 93 (2): 97–101.
Craig, C., Goulon, C., Berton, E., Rao, G., Fernandez, L. and Bootsma, R. J.
(2009) Optic variables used to judge future ball arrival position in expert and
novice soccer players. Attention, Perception, and Psychophysics, 71 (3): 515–
22.
Craig, C. M., Bastin, J. and Montagne, G. (2011) How information guides
movement: intercepting curved free kicks in soccer. Human Movement
Science, 30 (5): 931–41.
Davids, K. (2009) The organization of action in complex neurobiological
systems, in D. Araújo, H. Ripoll and M. Raab (eds) Perspectives on
Cognition and Action in Sport. New York: Nova Science Publishers, pp. 3–
13.
Davids, K., Bennett, S. J. and Newell, K. (2006) Movement System
Variability. Champaign, IL: Human Kinetics.
Davids, K., Araújo, D., Button, C. and Renshaw, I. (2007) Degenerate brains,
indeterminate behavior and representative tasks: implications for
experimental design in sport psychology research, in G. Tenenbaum and R.
Eklund (eds) Handbook of Sport Psychology, 3rd edn. New York: John
Wiley, pp. 224–44.
de la Pena, N., Weil, P., Llobera, J., Giannopoulos, E., Pomes, A., Spanlang,
B., Friedman, D., Sanchez-Vives, M. V. and Slater, M. (2010) Immersive
journalism: immersive virtual reality for the first-person experience of news.
Presence, 19 (4): 291–301.
Dessing, J. C. and Craig, C. M. (2010) Bending it like Beckham: how to
visually fool the goalkeeper. Plos ONE, 5 (10): e13161;
doi:10.1371/journal.pone.0013161 (accessed 7 June 2013).
Duarte, R., Araújo, D., Gazimba, V., Fernandes, O., Fologado, H.,
Marmeleira, J. and
Davids, K. (2010) The ecological dynamics of 1v1 sub-phases in association
football. Open Sport Science Journal, 3: 16–18.
Edelman, G. M. and Gally, J. (2001) Degeneracy and complexity in
biological systems. Proceedings of the National Academy of Sciences of the
USA, 98: 13763–8.
Esteves, P. T., de Oliveira, R. F. and Araújo, D. (2011) Posture-related
affordances guide attacks in basketball. Psychology of Sport and Exercise,
12: 639–44.
Farrow, D. and Abernethy, B., (2003) Do expertise and the degree of
perception–action coupling affect natural anticipatory performance?
Perception, 32 (9): 1127–39.
Gerardi, M., Rothbaum, B. O., Ressler, K., Heekin, M. and Rizzo, A. (2008)
Virtual reality exposure therapy using a virtual Iraq: case report. Journal of
Traumatic Stress, 21: 209–13.
Gibson, J. J. (1979) The Ecological Approach to Visual Perception. Hillsdale,
NJ: Lawrence Erlbaum Associates.
Hammond, K. R. and Stewart, T. R. (eds) (2001) The Essential Brunswik:
Beginnings, Explications, Applications. New York: Oxford University Press.
Hong, S. L. and Newell, K. M. (2006) Practice effects on local and global
dynamics of the ski-simulator task. Experimental Brain Research, 169: 350–
60.
Knapp, J. M. and Loomis, J. M. (2004) Limited field of view of head-
mounted displays is not the cause of distance underestimation in virtual
environments. Presence: Teleoperators and Virtual Environments, 13 (5):
572–7.
Kelso, J. A. S. (1995) Dynamic Patterns. The Self-Organization of Brain and
Behavior. Cambridge, MA: MIT Press. Kelso, J. A. S. (2002) The
complementary nature of coordination dynamics: selforganization and
agency. Nonlinear Phenomena in Complex Systems, 5 (4): 364–71.
Krijn, M., Emmelkamp, P. M. G., Biemond, R., de Wilde de Ligny, C.,
Schuemie, M. and van der Mast, C. A. P. G. (2004) Treatment of acrophobia
in virtual reality: the role of immersion and presence. Behavior Research and
Therapy, 42: 229–39.
Marik, L. S. (1987) Eye height-scaled information about affordances: a study
of sitting and stair climbing. Journal of Experimental Psychology: Human
Perception and Performance, 13 (3): 361–70.
Marsh, K. L., Richardson, M. J., Baron, R. M. and Schmidt, R. C. (2006)
Contrasting approaches to perceiving and acting with others. Ecological
Psychology, 18, 1–38.
McGarry, T. and Franks, I. (2007) System approach to games and
competitive playing: reply to Lebed (2006) European Journal of Sport
Science, 7: 47–53.
McGarry, T., Anderson, D. I., Wallace, S. A., Hughes, M. D. and Franks, I.
M. (2002) Sport competition as a dynamical self-organizing system. Journal
of Sports Sciences, 20, 771–81.
McMenemy, K. and Ferguson, S. (2007) A Hitchhiker's Guide to Virtual
Reality. Wellesley, MA: A. K. Peters.
Moroney, W. F. and Moroney, B. W. (1998) Simulation, in D. J. Garland, J.
A. Wise, V. D. Hopkins (eds) Human Factors in Aviation Systems. Hillsdale,
MI: Lawrence Erlbaum Associates, pp. 358–88.
Müller, S. and Abernethy, B. (2006) Batting with occluded vision: an in situ
examination of the information pick-up and interceptive skills of high- and
low-skilled cricket batsmen. Journal of Science and Medicine in Sport, 9:
446–58.
Newell, K. M. (1986) Constraints on the development of coordination, in M.
G. Wade and H. T. A. Whiting (eds) Motor Development in Children:
Aspects of Coordination and Control. Dordrecht, Netherlands: Martinus
Nijhoff, pp. 341–60.
Oudejans, R. R. D., Michaels, C. F., Bakker, F. C. and Dolne, M. A. (1996)
The relevance of action in perceiving affordances: perception of
catchableness of fly balls. Journal of Experimental Psychology: Human
Perception and Performance, 22 (4): 879–92.
Passos, P. (2010) Rugby. Cruz Quebrada: Faculdade de Motricidade Humana.
Passos, P., Araújo, D., Davids, K., Gouveia, L. F., Milho, J. and Serpa, S.
(2008) Information-governing dynamics of attacker-defender interactions in
youth rugby union. Journal of Sport Sciences, 26 (13): 1421–9.
Passos, P., Milho, J., Fonseca, S., Borges, J., Araújo, D. and Davids, K.
(2011) Interpersonal distance regulates functional grouping tendencies of
agents in team sports. Journal of Motor Behavior, 43 (2): 155–63.
Petit, J. P. and Ripoll, H. (2008) Scene perception and decision making in
sport simulation: a masked priming investigation. International Journal of
Sport Psychology, 39 (1): 1–19.
Pinder, R., Davids, K., Renshaw, I. and Araújo, D. (2011) Manipulating
informational constraints shapes movement re-organisation in interceptive
actions. Attention, Perception, and Psychophysics, 73: 1242–54.
Riccio, G. E. (1995) Coordination of postural control and vehicular control:
implications for multimodal perception and simulation of self-motion, in P.
Hancock, J. Flach, J. Caird and K. Vicente (eds) Local Applications of the
Ecological Approach to Human- Machine Systems, Hillsdale, NJ: Lawrence
Erlbaum Associates, pp. 122–81.
Richardson, M. J., Shockley, K., Fajen, B. R., Riley, M. A. and Turvey, M. T.
(2008) Ecological psychology: six principles for an embodied–embedded
approach to behaviour, in P. Calvo and A. Gomila (eds) Handbook of
Cognitive Science: An Embodied Approach. San Diego, CA: Elsevier, pp.
161–87.
Rizzo, A. and Kim, G. J. (2005) A SWOT analysis of the field of virtual
reality rehabilitation and therapy. Presence: Teleoperators and Virtual
Environments, 14 (2): 119–46.
Runeson, S. and Frykholm, G. (1981) Visual perception of lifted weight.
Journal of Experimental Psychology, 7 (4): 733–40.
Savelsbergh G. J. P., Williams A. M., van der Kamp J. and Ward P. (2002)
Visual search, anticipation and expertise in soccer goalkeepers. Journal of
Sports Sciences, 20: 279–87.
Sheridan, T. B. (1992) Musings on telepresence and virtual presence.
Presence: Teleoperators and Virtual Environments, 1 (1): 120–6.
Sheridan, T. B. (1996) Further musings on the psychophysics of presence.
Presence: Teleoperators and Virtual Environments, 5: 241–5.
Stanney, K. M., Mourant, R. R. and Kennedy, R. S. (1998) Human factors
issues in virtual environments: a review of the literature. Presence:
Teleoperators and Virtual Environments, 7 (4): 327–51.
Stoffregen, T. A. and Riccio, G. E. (1988) An ecological theory of orientation
and the vestibular system. Psychological Review, 95: 3–14.
Stoffregen, T. A., Bardy, B. G. Smart, L. J. and Pagulayan, R. J. (2003) On
the nature and evaluation of fidelity in virtual environments, in L. J. Hettinger
and M. W. Haas (eds) Virtual and Adaptative Environments: Applications,
Implications, and Human Performance Issues. Mahwah, NJ: Lawrence
Erlbaum, pp. 111–28.
Tarr, M. J. and Warren, W. H. (2002) Virtual reality in behavioural
neuroscience and beyond. Nature Neuroscience, 5: 1089–92.
Turvey, M. T., Shocklet, K. and Carello, C. (1999) Affordance, proper
function, and the physical basis of heaviness. Cognition, 73: B17–26.
Vaeyens, R., Lenoir, M., Williams, A. M., Mazyn, L. and Philippaerts, R. M.
(2007) The effects of task constraints on visual search behavior and decision-
making skill in youth soccer players. Journal of Sport Exercise Psychology,
29: 147–69.
Vignais, N., Bideau, B., Craig, C., Brault, S., Multon, F., Delamarche, P. and
Kulpa, R. (2009) Does the level of graphical detail of a virtual handball
thrower influence a goalkeeper's motor response? Journal of Sports Science
and Medicine, 8: 501–50. Villani, D., Riva, F. and Riva, G. (2007) New
technologies for relaxation: the role of presence. International Journal of
Stress Management, 14: 260–74.
Wagner, I., Broll, W., Jacucci, G., Kuutii, K., McCall, R., Morrison, A.,
Schmalstieg, D. and Terrin, J.-J. (2009) On the role of presence in mixed
reality. Presence: Teleoperators and Virtual Environments, 18 (4): 249–76.
Warren W. (2006) The dynamics of perception and action, Psychological
Review, 113: 358–89.
Warren, W. H., Jr. (1984) Perceiving affordances: visual guidance of stair
climbing. Journal of Experimental Psychology: Human Perception and
Performance, 10 (5): 683–703.
Warren, W. H., Jr. and Whang, S. (1987) Visual guidance of walking through
apertures: body-scaled information for affordances. Journal of Experimental
Psychology: Human Perception and Performance, 13 (3): 371–83.
Watson, G., Brault, S., Kulpa, R., Bideau, B., Butterfield, J. and Craig, C.
(2011) Judging the ‘passability’ of dynamic gaps in a virtual rugby
environment. Human Movement Science, 30 (5): 942–56.
Willemsen, P., Colton, M. B., Creem-Regehr S. H. and Thompson, W. B.
(2009) The effects of head-mounted display mechanical properties and field-
of-view on distance judgments in virtual environments. ACM Transactions
on Applied Perception, 6 (2): article no. 8; doi: 10.1145/1498700.1498702
Williams, A. M. and Ericsson, K. A. (2005) Perceptual–cognitive expertise in
sport: some considerations when applying the expert performance approach.
Human Movement Science, 24: 287–307.
Williams, A. M., Ward, P., Knowles, J. M. and Smeeton, N. J. (2002)
Anticipation skill in a real-world measurement, and transfer in tennis.
Journal of Experimental Psychology: Applied, 8 (4): 259–70.
11 Methods of measurement in
studying team sports as
dynamical systems
Daniel Memmert and Jürgen Perl
The theoretical roots of the approach of an artificial simulation by means of
neural networks lie back in the 1940s. Meanwhile, the claim of modelling
biological dynamics has been reduced to a more pragmatic application of the
great abilities of net-based concepts and tools. Today, two approaches are
mainly in use. They can complement each other, in particular in decision
making processes that occur in team sports. Unsupervised or self-organizing
maps or networks (SOM) can learn and recognize the patterns of match
situations on their own (Figure 11.1, left graphic), whereas supervised or feed
forward networks (FFN) can learn by supervision what are the best solutions
for a situation, once recognized (Figure 11.1, right graphic; see also Kohonen
1995; Hopfield 1982).
An FFN consists of a number of layers, each of which contains a number
of neurons. The layers are arranged in a sequential order. The neurons of each
layer can change the states of the neurons of the following layer by means of
specific functions (‘feed forward’), which, as a whole, model the network
behaviour. During the learning phase, every step from input to output layer is
completed by a comparison of calculated and expected results, the difference
of which is then used for feedback (e.g. back propagation): the feedback is
combined with a deviation-depending reinforcement and so adjusting the
parameters of the functions that calculate the changes of the neuron states.
After the learning phase; i.e. as soon as the calculated results are satisfyingly
accurate, the network can calculate output value-like actions to given input
value-like situations.
Figure 11.1 (left) Self-organizing map (SOM: neurons are grouped to clusters which form the output;
(right) feed-forward network (FFN): comparison of expected output Oexp and computed output Ocomp
is feedback and so changes the neuron connections to minimize the difference between expected and
computed output

A self-organizing network also consists of neurons which are not

organized in layers but are regularly arranged in a matrix. As opposed to
FFNs, the connections between neurons do not define possible signal
pathways but describe a topological neighbourhood relation. During the
learning phase, an input affects a particular neuron, to which it is closest or
most similar, by moving it towards the input. The neurons of a number of
neighbour shells are moved in the same direction, with decreasing intensity at
increasing distance. Owing to the similarity of controlled neuron movements,
the input values eventually cause a distribution of the neurons that is
topologically similar to the distribution of the input values. In particular,
classes of input values are mapped to clusters of neurons. After the learning
phase, input values can be classified as belonging to specific clusters. This
way, the high variety of datasets can be reduced to a small set of
characteristic types.
In this chapter, the focus is on pattern recognition and SOMs. Supporting
games by means of decision-optimizing FFNs has been done with simulated
games, e.g. in the field of robot soccer, but currently is still too complicated
for original games played by humans.
Developments and derivations of the basic SOM approach

Static and dynamic pattern analyses

There are two ways of net-based pattern analysis in sport. The first and
comparably simple one is that of analyzing time- or space-oriented
distributions, as for example player positions on the pitch. During a game,
every player distributes their positions over the pitch in a characteristic way,
depending on their tactical tasks and their technical and conditional skills.
The net can then find the typical clusters; i.e. specific areas with the most
frequented positions. In every moment, a group of players – defence or
offence – form patterns of positions on the pitch, depending on their tactical
task and the actions of the respective opponent group. The net can again find
typical clusters; i.e. the most frequent formations.
Different to this static pattern analysis – and significantly more difficult –
is a dynamic pattern analysis of complex game processes. For this purpose,
trajectories are very helpful. They connect situation-representing neurons to a
process-representing sequence of neurons. The advantage of trajectories is
that they reduce the sequence of high-dimensional process datasets to a
simple twodimensional sequence of neuron coordinates, which is much easier
to handle (see detailed explanations below). Moreover, they can be taken as
input to a second net in order to analyze trajectories as dynamic process
patterns. On the other hand, we are then confronted with two basic
disadvantages of SOMs: the number of available trajectories is much smaller
than that of the contained data sets and therefore it is normally not sufficient
for SOM training. They are also unable to learn continuously and follow
development of team behaviour over time. The approach of the dynamically
controlled network ‘DyCoN’, which has been developed by one of the
authors (Perl 2002a), is able to solve the problem.

DyCoN-approach
Owing to the fact that SOM training is controlled by an external algorithm
using parameters that run down to final values and so eventually cause the
end of the training process, a SOM that has been trained once cannot be
reactivated for further training. Therefore, continuing training would require
a complete rearrangement of the net and all controlling parameters, a practice
that cannot be done in a satisfying way.
The DyCoN concept, however, is different: each neuron contains an
internal memory and a self-controlling algorithm. The effect of the individual
neural self-control is that a DyCoN has no final state but it can always adjust
its internal memory to new input and can therefore learn continuously as well
as in separate phases (see Perl 2002a; Perl and Dauscher 2006).

Trajectories and two-level pattern analysis

Figure 11.2 shows how trajectory analysis works; the matrix represents the
trained network in which the small, grey shaded squares represent situations
of a game – as for instance are formations (i.e. sets of x–y coordinates) of
selected groups of players. Each grey shade stands for a specific type of such
situation (e.g. the rhombic formation of the back four in soccer), while areas
of the same grey shade represent variants of the same type (e.g. different
geometric forms of that rhombus). The situation datasets of the game activate
corresponding neurons of the network: starting with the light grey square that
is marked by an dark frame, the process runs through some light grey neurons
followed by some grey and dark, and so on. The resulting trajectory – i.e. the
embedded sequence of black arrows – is a two-dimensional mapping of the
process; i.e. the time series of recorded situations.
To analyze such trajectories, they are taken as training data for a second
network, which is then able to cluster the set of trajectories and find out the
characteristic types of processes. An example would be to assume that the
colour sequence ‘light grey, medium grey, dark grey, light grey’ from Figure
11.2 means a sequence of formations ‘rhombus, line, square, rhombus’ of the
back four, which might be corresponding to a particular defence strategy of a
soccer team. The second network learns those sequences, compresses them to
clusters or types and therefore represents the defence patterns of that team.
However, the combinatorial variety of such trajectories through some
hundreds of neurons is so large that it makes sense to reduce the original
neuron-oriented trajectories to type-oriented trajectories, as is shown in the
small embedded diagram of Figure 11.2: the trajectory has been replaced by
the sequence of corresponding colours/type numbers, resulting in the
sequence (5,5,5,5,5,5,3,3,3,3,2,2,2,2,2,3,3,3,3,3), which can be taken as an
input vector for the second network, as described above. One serious problem
remains: normally, a set of about 5400 input data (one per second) is
available from a soccer game to train a net. The procedure of dividing the
stream of input data into trajectories reduces the number of available data
objects dramatically to some hundreds, often missing the necessary amount
for a successful training by a long way. The advantage of DyCoN, however,
can compensate for the lack of data by two steps:

Figure 11.2 Network with a process trajectory and the corresponding type profile

• Monte Carlo simulation multiplies the original trajectories by varying

 the values of their components without changing the trajectory types;
• repeated training phases (as described above) help to acquire more
information from the available data.
Overview on the development of game analysis: concepts,
devices and tools
There are three main tracks of technical development that can be observed
since the early 1990s (see also Figure 11.3). Data have normally been
acquired manually by extraction from the video recording. Since around
2010, position recording has been supported by automatic devices, whereas
ball recording using automatic video analysis and action recording using net-
based pattern recognition have just started very recently (Perl et al. 2012).
Methods of analysis have mainly been reduced to event analysis, using
statistical approaches and mapping the complex game to simple numbers
and/or distribution. First approaches to process-oriented analyses started in
the second half of the 1990s in tennis, table tennis and handball (working
group of Mainz; Perl 2001), followed by net-based approaches in squash,
volleyball, soccer and general game analysis (working groups of Mainz and
Heidelberg; Memmert and Perl 2005) starting around the year 2000. Early
data processing tools mainly used video recorders and databases, while neural
networks increased their influence in game analysis around the late 2000s
(Perl 2002b). Again, the working groups of Mainz, Heidelberg and, more
recently, Cologne, have been the leading centres of that development (e.g.
Memmert and Perl 2009a,b). Currently, the research focus is on creativity
and game simulation in particular.
Meanwhile, neuronal networks have become a frequently studied and
commonly recognized possibility for data analysis and data simulation in
sport. By means of neuronal networks, different kinds of human movements
and several skills in sport can be analyzed. For example, handball (Memmert
and Perl 2009a), soccer (Memmert and Perl 2009b), squash (McGarry and
Perl 2004), field hockey (Memmert and Perl 2005), arm movements (Draye
et al. 1995; Cheron et al. 1996), arm and stand-up movements (Draye et al.
2002), as well as walking (Aminian et al. 1993; Köhle and Merkl 1998;
Schöllhorn et al. 2002; Tucker and White 1999; Schöllhorn 2004) were
modelled by means of neural nets. These studies demonstrate that additional
research questions linked with pattern learning, gate analysis, game analysis,
motor analysis and simulation processes can profit from using neuronal
networks. All in all, the role and development of neuronal networks is a topic
of current discussions in computer sport science and a new method of
analysis in studying team sports as dynamical systems.

Figure 11.3 (left): TeSSy input interface with video control (bottom), animation interface (left),
attribute selection and input panel (top); (right): examples of a stroke frequency matrix and a stroke
success matrix representing tactical concepts and technical skills, respectively
Game complexity: from simple to complex game dynamics
The first approaches to video-based game data recording and analysis date
back to the late 1970s. While many modern concepts and methods of game
analysis have already been developed, the weakness was – and still is – the
data extraction process, which normally had to be done manually. Thus, the
first case studies were restricted to structurally simple two-person games such
as tennis or squash, later followed by games like volleyball, in which the
teams are separated, acting like (abstract) super-players.
During the past five years or so, the combination of automatic position data
extraction and net-based process pattern analyses has made great progress,
with analyses of complex game processes such as those from handball,
basketball or even soccer (Perl and Memmert 2011).

Tennis
The first computer-based tennis analyses were restricted simply to strokes,
which were simply be described by the ‘from’ position and the ‘to’ position.
Figure 11.3 shows the interface of TESSY® (Razorcat Development GmbH,
Berlin), a videoand computer-based tennis analysis tool which was developed
during the 1990s, offering a collection of complex game-process analysis
features (Mussel et al. 2001).
TESSY enabled statistical evaluations of game situations and stroke
sequences, including stroke combinations within single rallies. Moreover,
tactical patterns, e.g. position clouds or stroke bundles, could be presented
graphically. Finally, simulations of games and their tactical concepts were
possible, as is shown in Figure 11.3. A player's technical skills (against a
fixed opponent or as mean values) can be characterized by a matrix of
success values of those ‘from–to’ strokes, while tactical concepts are
represented by a similar matrix containing the frequencies of strokes.
Obviously, as was in fact done in practice, such matrices can be used to
simulate the effects of tactical concepts and technical skills by just changing
the corresponding matrix values and analysing the changing game dynamics.

Squash
A player-specific process in squash can be defined as sequence of stroke
positions (cf. tennis) of a player (see McGarry et al. 1999). Figure 11.4 shows
an example of two strokes (left) and a net trained with the positions of four-
stroke sequences, taken from games of an international tournament in 1988.

Figure 11.4 Squash court with game process BR-FR-BL (left); trained net representing the most
frequent 4-position-sequences (right); BL = backhand left side, BR = backhand right side, FR =
forehand right side

The tactical pattern of a player in a game is the collection of all action

sequences in which he has been involved. In the case of stroke positions as
actions, however, it has to be considered that the stroke position of a player is
the target position of his opponent's stroke and therefore reflects the tactical
pattern of the opponent and not of the actual player.
In Figure 11.5, it can be seen that, in the semi-final, the tactical patterns of
both players mainly consisted of the same two sequences BR–BL–BL–BL
and BL–BL–BL–BL (see Figure 11.3); i.e. both players trying to keep their
opponent on the backhand side. In the final, the winner of semi-final 1, player
B, showed a quite different tactical behaviour (i.e. varying much more
between long line and cross playing). Again, both players showed nearly
identical patterns. It seems that even top players are not able to play tactical
patterns independently of their opponents but find a common rhythm, which
is not specific for a player but for the pair of players.
Figure 11.5 Tactical patterns of players depending on their opponents; A, B and D are the players. The
squares represent the corresponding networks, where the circles represent the players' tactical concepts.
Each circle corresponds to a stroke sequence, the frequency of which is encoded by the diameter of the
circle

Of course, analyses dealing with complex game dynamics based on tactical

patterns can also be done without neural networks. However, the main
advantage of self-organizing neural networks is that they are able to select the
most important 10–20 types of sequences themselves, instead of having to
manually select from 256 different possibilities – or even millions as is the
case in football. In the example of squash, the sequences are of length four,
where each component of the sequence can have the four entries BL, BR, FL,
FR, therefore summing up to 44 = 256. The same calculation with soccer
sequences of length 10 (meaning a duration of only 10 seconds) and only six
different formation types already sums up to 106 = 1.000.000 different
sequence types.

Volleyball
The first net-based approach to volleyball from 1999 (Perl and Lames 2000)
was similar to the squash approach above, resulting in a characterization of
the most important sequences and their frequencies as is presented in Figure
11.6.
Figure 11.6 The network was trained with game processes from volleyball, presenting the most
important sequences as circles, whose diameters represent the frequencies of the corresponding
sequences, three of them being explained in more detail

Different to squash, however, this network could not give information

about the individual technical or tactical behaviour and therefore its use was
restricted to statistical analyses of frequencies and distributions of typical
game processes only. The simple reason is that the net was unable to give
information about the actions of single players, owing to its conceptual
design.
In a second approach, the focus was on the formation of the players of a
team on the playground (Figure 11.7, right-hand graphic), resulting in a much
better insight into the tactical behaviour of the teams. After training, the
twodimensional neuron grid of the artificial neural network is organized into
clusters of similar configurations. Figure 11.7 shows a trained net with such
separated clusters (left). One cluster, which could, for example, correspond to
the presented formation type – is highlighted by a thick black frame.
Figure 11.7 (left): Trained net with a trajectory representing a sequence of player formations starting at
the ‘O’ and ending at the ‘X’; (right): scheme of a configuration prototype of a team formation, which
could, for example, correspond to the marked cluster

Testing original game data on the trained net again (as in squash) results in
activations of neurons, representing the types and frequencies of the activated
formations. Moreover, the order of appearance of the configurations can also
be transferred onto the net and results in a trajectory (grey line).
By way of example, Figure 11.8 shows the differences between two top
teams in the game of an international women's tournament: both nets show
the defence's preparation for the expected service of the opponent team. It can
easily be seen that Germany and Italy prefer quite different formations when
it comes to taking the service. Moreover, there is another remarkable
difference: while the Italian team obviously finds the optimal formation
comparably fast, only adjusting the formation itself (most of the moves are
inside the marked clusters), the German team prepares by changing the whole
formation (most of the moves are between the marked clusters), which means
much more movement and a certain restlessness that later affects their actions
negatively.
Figure 11.8 Trajectories showing the preparation of the defence against the opponent's service

Football (soccer)
Neural networks can already be used for the computer-supported analysis of
several more complex tactical performance factors in soccer. Based on
position data (Figure 11.9), soccer-specific situations can be objectivized by
using analysis software for the assessment of match situations (see Grunz et
al. 2009, 2012; Memmert and Perl 2006, 2009a,b; Memmert et al. 2011; Perl
et al. 2011).
The basis for the generation of the data is the collection of the x–y
coordinates of all 22 players and the ball for the entire match time of 90
minutes (‘tracking’). Using a sampling rate of 25 frames per second, an
amount of 135,000 x–y data per player is generated which equals a total
amount of 3,105,000 x–y data per game, including the ball. The basic idea is
that the developed neural networks make it possible to compare match scenes
from one or more games, to discover which sequences have led to which
results. As described for volleyball, the neurons activated by the single
datasets are being connected to trajectories that represent the two-
dimensional patterns of the match sequence.
Similar patterns of such match sequences are then assigned to a common
neuron or a cluster of adjacent neurons on a neural net of the second level and
form a characteristic type. The aim is to group sequences, e.g. a ‘quick build-
up’ on to the net in according clusters during the training, to automatically
detect the respective realizations during the game analysis. In this way, it is
possible to analyze extensive amounts of data online and to classify them
with regard to differences and similarities.

Figure 11.9 A two-dimensional replication of a match situation by means of position data

Another crucial aspect of the evaluation of a team's tactical behaviour is

the interaction of specific tactical groups, such as offence and defence. The
problem is that, despite the availability of the position data, an analysis of, for
example, tactical movements of a team formation is hardly operable with
conventional methods, owing to above-mentioned amount of data. Here, the
ability of neural networks to recognize patterns offers new possibilities (as
indicated in Figure 11.10). Specific match situations can be isolated from the
rest of the game and, thus, specific player positions can be learned by the net
as characteristic formations. In this way, it is possible to identify frequency
distributions of typical group formations as well as tactical interaction
processes. At the same time, the degree of the implementation of tactical
patterns can be identified, together with the ability of a team, to situationally
generate new patterns.
Figure 11.10 Net-based recognition of formation types and the recombination with position and time
information (Perl and Memmert 2011)

First validation studies show that approximately 90% of the match events
that are collected by means of traditional game analysis can also be detected
by neural networks (Grunz et al. 2012). These events include various group
tactics such as build-ups and set pieces (further differentiated into throw-ins,
free kicks and corner kicks) as well as goal scoring.
From quantitative to qualitative analysis – and back
Analyses of complex behavioural processes – such as team sports – are
intended to map the time-oriented flow of situations and actions to a time
series of data packages, which then have to be transferred and condensed into
a sequence of the characteristic and/or relevant pieces of information.
Obviously, statistical numbers like mean values or even distributions of types
of activities are neither sufficient nor adequate to characterize or analyze such
complex behaviour. In 1971, Günter Hagedorn formulated the problem of the
high time- and space-related meaning of qualitative game situations
(Hagedorn 1971), also called the context orientation of activities (e.g. see Perl
1999). It turns out that patterns are, for instance, useful for the necessary
quantitative–qualitative transfer (see Pattern recognition, below) and, as
described above, that self-organizing neural networks can be used to find,
characterize and analyze those patterns (e.g. Perl and Dauscher 2006).
Owing to the massive lack of game data, these first steps were normally
case studies, demonstrating ways and dealing with aspects as well as basic
phenomena. Of course, a well-based research requires information not only
about one game but about a collection of games, to recognize standards,
invariants, opponent- dependent behaviour and so on. By now, automatic
position recording is one very important step towards developing standard
routines for analyzing games and collecting information about them, which
opens the way to a high-level empirical analysis. Not only is this the
necessary step from data-based qualitative analysis to information-based
quantitative statistics; the approaches that have been developed in the
meantime allow much more: the combination of behavioural patterns and
statistical distributions of frequencies and success can be used for simulation
of training effects as well as tactical innovations (see Simulation, below).
Moreover, the combination of advanced network skills and statistical
analyses can help to recognize tactical creativity – and to improve it by
means of net-based simulation. Two examples are given in the next chapter.

Pattern recognition
Based on the approach depicted in Figure 11.10, Figure 11.11 exemplarily
shows some of the possibilities of the formation-based quantitative and
qualitative pattern analysis of game processes in soccer. On the left of Figure
11.11, the interaction analysis with the selection tool for the data of both
teams (here, the back-four of Team A (in light grey) and four offensive
players of Team B (in dark grey), the scrollbar for a review of the entire
match, the window to display the respective formation and the synoptic table
which lists the number of formations as well as the coincidence frequencies
for the entire match. Very frequently occurring formations are framed in a
dotted line, very rarely occurring ones in solid black. On the top right, the
window of a simplified formation, the group's field of attention and of the
combined formation of the offence and defence group. On the bottom right,
the window with the interaction process of the observed offence/defence
formation for a certain period of the match (Perl and Memmert 2011).

Figure 11.11 Example of the user interface of a tool for the combined quantitative and qualitative
analysis of formations in football (see explanation in the text)

Another example from team sport research illustrates the pattern-

recognition phenomenon by means of neural networks with regard to tactical
creativity. In team sports, tactical creativity at a behavioural level is defined
as unusualness, innovativeness, statistical rareness or even uniqueness of
solutions to a related sport situation (Memmert and Roth 2007; Memmert
2011). Using the example of tactical creativity, different types of creative
learning behaviour can be differentiated by means of neural networks. Such a
process-oriented analysis can help to find reasons for specific distinctive
features. In this specific case, the different learning patterns could be
explained by the fact that the training process was characterized by various
types of learning behaviour in the different training groups (Memmert and
Perl 2009b).
Figure 11.12 exemplarily shows an array of trajectories representing the
individual training process of the athletes; beginning with the dark grey and
ending with the light grey square, the individual sub-steps of the respective
process are displayed as red lines on the net. In the three steps of the process,
the trajectory runs through the grey shaded areas that depict the different
quality levels of the net (from light grey = very good to black = extremely
bad). After the entry on the net, the mean values of the qualitative analysis
between the groups were compared by means of c2 tests for the nominal-
scaled variable.

Figure 11.12 Trained neural network with grey shaded areas that illustrate different quality levels (top
left) and a representation of the trajectories of hockey training. The learning process begins in the dark
grey square and ends in the light grey square (Memmert and Perl 2009b); the colours of the neurons
correspond to those in the large net graphic (top left)

The development of hockey-specific tactical creativity of the 20 hockey

players represented in Table 11.1 shows very varied results over the 15
months of training: in 5 of 20 cases (25%), the performance increases at the
beginning but turns out to be worse in the end than in the middle of the
training process (up–down fluctuation process). The contrary behaviour was
observed for 30% of the subjects (down–up fluctuations). In 25% of the
cases, the performance increased monotonically, whereas it decreased
monotonically in 10% of the cases. In 10% of the cases, the performance
remained almost entirely the same.
Table 11.1 represents the collected results of the hockey, the soccer and the
control group regarding the development of tactical creativity. Interestingly,
the hockey group shows different result patterns than the soccer group. Only
a comparison of the c2 statistics of both fluctuation processes showed a
significant effect between the hockey and the soccer group. The subjects of
the hockey group showed stronger up–down fluctuations than those from the
soccer group and vice versa. Compared with the control group, there were no
significant differences regarding the fluctuations between the soccer and the
hockey group. Concerning the other three types of learning processes, no
crucial differences between the groups were found.

Table 11.1 Summary of the results of all trajectories of all three groups (hockey, soccer, control); the
five different types of learning behaviour are outlined in the second column

Source: Memmert and Perl 2009b

Such a process-oriented comparison of the results can help to detect
problems and to find reasons for specific distinctive features. In this specific
example, the different learning patterns could, for example, be explained by
the fact that the training process was characterized by various types of
learning behaviour in the different training groups.

Simulation
An advanced application of neural networks is the simulation of tactical
behaviour, creative actions and dynamic learning in games. The current step
of the game process is tested on the network, activating the corresponding
neuron, which then returns information in different semantic categories such
as type of activity, degree of creativity, probability of success or probability
of transition to other activities. The idea is to replace the current activity by a
simulated one, which could be more creative or successful, and to further
simulate the resulting process by means of transition and success matrices
(see Figure 11.3) and then to analyze the resulting simulated process with the
intention of improving the team's tactical behaviour. Mapped to a network,
this means that neurons should have the ability to represent not only frequent
but also – and in particular – rare actions. If such a net is calibrated with
respect to success or adequacy, the time series of a process is mapped to a
trajectory, where the neurons can be recognized to correspond to creative
actions (Grunz et al. 2009).

Questions for students

1. Characterize the differences between SOMs and FFNs.
2. Depict the differences between static and dynamic pattern analyses.
3. How does the DyCoN approach work?
4. Which are the three crucial technical developments for the analysis
of sport games by means of neural networks during the last
coupZle of years?
5. Describe the possibilities of the net-based game analysis in
different team sport games.
6. What are the differences between racket and team sports with
regard to the analysis by means of neural networks?
7. Which new results can be found by means of analysis of the
tactical creativity when using neural networks for pattern
recognition?
References
Aminian, K., Robert, P., Jéquier, E. and Schutz, Y. (1993) Level, downhill
and uphill walking identification using neural networks. Electronics Letters,
29 (17): 1563–5.
Cheron, G., Draye, J. P., Bourgeois, M. and Libert, G. (1996) A dynamic
neural network identification of electromyography and arm trajectory
relationship during complex movements. IEEE Transactions on Biomedical
Engineering, 43 (5): 552–8.
Draye, J. P., Cheron, G., Bourgeois, M., Pavisic, D. and Libert, G. (1995)
Identification of the Human Arm Kinetics using Dynamic Recurrent Neural
Networks. Neuro-COLT Technical Report Series NC-TR-95-017.
ESANN’1995 Proceedings, European Symposium on Artificial Neural
Networks, Brussels (Belgium), 19–21 April 1995. Egham: Royal Holloway
University of London, Department of Computer Science, pp. 33–8.
Draye, J. P., Winters, J. M. and Cheron, G. (2002) Self-selected modular
recurrent neural networks with postural and inertial subnetworks applied to
complex movements. Biological Cybernetics, 87: 27–39.
Grunz, A., Memmert, D. and Perl, J. (2009) Analysis and simulation of
actions in games by means of special self-organizing maps. International
Journal of Computer Science in Sport, 8: 22–36.
Grunz, A., Memmert, D. and Perl, J. (2012) Tactical pattern recognition in
soccer games by means of special self-organizing maps. Human Movement
Science, 31: 334–43.
Hagedorn, G. (1971) Beobachtung und Leistungsmessung im Sportspiel
[Observation and performance measures in team sports]. Leistungssport, 1:
17–22 [German].
Hopfield, J. J. (1982) Neural networks and physical systems with emergent
collective computational abilities. Proceedings of the National Academy of
Sciences of the USA, 79: 2554–8.
Köhle, M. and Merkl, D. (1998) Experiments in gait pattern classification
with neural networks of adaptive architecture, in Proceedings of the 8th
International Conference on Artificial Neural Networks, Skövde, Sweden.
Perspectives in Neural Computing. New York: Springer.
Kohonen, T. (1995) Self-Organizing Maps. New-York: Springer.
McGarry, T. and Perl, J. (2004) Models of sports contests: Markov processes,
dynamical systems and neural networks, in M. Hughes and I. M. Franks (eds)
Notational Analysis of Sport. London and New York: Routledge, pp. 227–42.
McGarry, T., Khan, M. A. and Franks, I. M. (1999) On the presence and
absence of behavioural traits in sport: an example from championship squash
match play. Journal of Sports Science, 17: 297–311.
Memmert D. (2011) Sports and Creativity, in M. A. Runco and S. R. Pritzker
(eds) Encyclopedia of Creativity, 2nd edn. San Diego: Academic Press, 2:
373–8.
Memmert, D. and Perl, J. (2005) Game intelligence analysis by means of a
combination of variance-analysis and neural networks. International Journal
of Computer Science in Sport, 4 (1): 29–38.
Memmert, D. and Perl, J. (2006) Analysis of game creativity development by
means of continuously learning neural networks, in E. F. Moritz and S.
Haake (eds) The Engineering of Sport 6. New York: Springer, 3: 261–6.
Memmert, D. and Perl, J. (2009a) Analysis and simulation of creativity
learning by means of artificial neural networks. Human Movement Science,
28: 263–82.
Memmert, D. and Perl, J. (2009b) game creativity analysis by means of
neural networks. Journal of Sport Science, 27: 139–49.
Memmert, D. and Roth, K. (2007) The effects of non-specific and specific
concepts on tactical creativity in team ball sports. Journal of Sport Science,
25: 1423–32.
Memmert, D., Bischof, J., Endler, S., Grunz, A., Schmid, M., Schmidt, A.
and Perl, J., (2011) World-level analysis in top level football. Analysis and
simulation of football specific group tactics by means of adaptive neural
networks, in C. L. P. Hui (ed.) Artificial Neural Networks: Application.
Rijeka, Croatia: InTech, pp. 3–12. Available from:
www.intechopen.com/articles/show/title/world-level-analysis-in-top-level-
football-analysis-and-simulation-of-football-specific-group-tactic (accessed
10 June 2013).
Mussel, D., Perl, J. and Schroder, H.-J. (2001) TeSSy 2000: Erfassungs- und
Analysesystem fur Tennis. [TeSSy 2000: system for collecting and analysing
tennis], in J. Perl (ed.) Sport and Informatik VIII. Koln: Straus, pp. 111–21.
Perl, J. (1999) Aspects and potentiality of unconventional modeling of
processes in sporting events, in B. Scholz-Reiter, H.-D. Stahlmann and A.
Nethe (eds) Process Modelling. Berlin-Heidelberg: Springer, pp. 74–85.
Perl, J. (2001) DyCoN: Ein dynamisch gesteuertes Neuronales Netz zur
Modellierung und Analyse von Prozessen im Sport, in J. Perl (ed.) Sport and
Informatik VIII. Koln: Straus, pp. 85–98.
Perl, J. (2002a) Adaptation, antagonism, and system dynamics, in G. Ghent,
D. Kluka and D. Jones (eds) Perspectives – The Multidisciplinary Series of
Physical Education and Sport Science, 4. Oxford: Meyer and Meyer Sport,
pp. 105–25.
Perl, J. (2002b) Game analysis and control by means of continuously learning
networks. International Journal of Performance Analysis of Sport, 2: 21–35.
Perl, J. and Dauscher, P. (2006) Dynamic pattern recognition in sport by
means of artificial neural networks, in R. Begg and M. Palaniswami (eds)
Computational Intelligence for Movement Science. Hershey, ID: Idea Group
Publishing: pp. 299–318.
Perl, J. and Lames, M. (2000) Identifikation von Ballwechselverlaufstypen
mit Neuronalen Netzen am Beispiel Volleyball. [Identification of rallies in
Volleyball by means of SOMs], in W. Schmidt and A. Knollenberg (eds)
Sport – Spiel – Forschung: Gestern. Heute. Morgen. Hamburg: Czwalina, pp.
211–15.
Perl, J. and Memmert, D. (2011) Net-based game analysis by means of the
software tool SOCCER. International Journal of Computer Science in Sport,
10: 77–84.
Perl, J., Memmert, D., Bischof, J. and Gerharz, Ch. (2006) On a first attempt
to modeling creativity learning by means of artificial neural networks.
International Journal of Computer Science in Sport, 5 (2): 33–8.
Perl, J., Memmert, D., Baca, A., Endler, S., Grunz, A., Rebel, M., and
Schmidt, A. (2011) Sensors, monitoring, and model-based data analysis in
sports, exercise and rehabilitation, in D. T. H. Lai, R. K. Begg and M.
Palaniswami (eds). Healthcare Sensor Networks: Challenges Toward
Practical Application. Boca Raton, FL: CRC Press, pp. 375–405.
Schöllhorn, W. (2004) Applications of artificial neural nets in clinical
biomechanics. Clinical Biomechanics, 19 (9): 876–98.
Schöllhorn, W. I., Schaper, H., Kimmeskamp, S. and Milani, T. L. (2002)
Inter- and intraindividual differentiation of dynamic foot pressure patterns by
means of artificial neural nets. Gait and Posture, 16: 159.
Tucker, C. A. and White, S. C. (1999) Neurocomputational approaches to
pattern recognition and time series analysis, in W. Herzog and A. Jinha (eds)
International Society for Biomechanics Congress XVIII, p. 2.
12 Team sports as dynamical
systems
Tim McGarry, Jürgen Perl and Martin
Lames
The challenge for understanding coordination was introduced by Meijer
(2001) as ‘Charles' problem’ (Charles V, 1500–1558) in reference of the
longstanding difficulty in comprehending how coordination might be
explained using mechanical (machine) metaphor. In short, Charles was
reportedly preoccupied with getting mechanical clocks (or pendulums) to
strike together in unison but was unsuccessful in doing so. Unfortunate for
Charles, an answer to the coordination problem was not discovered until
much later when Huygens, in 1664, reportedly sympathetic, behaviour
between pendulums when swung separately but suspended from a common
frame. Thus, two pendulums swinging from a common frame, given
sufficient time, self-coordinate into one of two possible rhythmic patterns,
that of in-phase or anti-phase (Meijer 2001). In-phase and anti-phase
coordination therefore constitute separate attractors for the coupled
pendulums with both pendulums drawn to one or the other attractor.
Importantly then, coordinated behaviours between coupled pendulums
emerges not from prescriptive control design by some outside agency (like
Charles) but, instead, from within by virtue of common information (energy)
flows. In short, coupled pendulums produce self-organized behaviours by
means of shared information exchange.
Relative phase
Relative phase is a metric that indicates the relative position of two points in
their given cycles at any instant. In-phase (zero or unity) represents the same
positions in the given cycles at any instant whereas half-phase (or anti-phase)
represents opposite positions, with other phase relations expressed within the
limits of zero through unity (or 360 degrees). For example, in-phase
represents two pendulums at the same points in their respective cycles with
both pendulums reaching the same zeniths at the same time whereas anti-
phase represents the anti-symmetric relation with the two pendulums
reaching opposing zeniths at the same time.
Human rhythmic coordination
Relative phase was used to describe the coordination features of simple
rhythmic coordinated actions expressed in the well-known ‘finger waggling’
experiments of Kelso and colleagues (Kelso et al. 1981; Kelso 1984; Haken
et al. 1985). The task was to flex and extend the index fingers of both hands
in the transverse plane paced at different oscillating frequencies. Increasing
coordination instabilities leading to spontaneous transition from anti-phase to
in-phase were reported as a result of increasing the cycling frequencies
beyond some critical value. Haken et al. (1985) modelled these dynamics
results using two coupled oscillators (pendulums) representing the two index
fingers, a theory of self-organizing behaviour subsequently applied to other
rhythmical movements, including coordination between different limbs
(Kelso and Jeka 1992), multi-limb movements (Kelso et al. 1991) and other
coordination patterns (deGuzman and Kelso 1991), including coordination
processes between the legs of different persons (Schmidt et al. 1990). The
same system description predicated on coupled oscillator dynamics therefore
describes coordination for many different complex rhythmic actions
including coordinated actions produced between persons. This latter finding,
in particular, prompted McGarry et al. (2002) to propose coupled oscillator
dynamics between players and teams as the theoretical underpinning for the
many and varied unique but nonetheless patterned game behaviours that
typify different sports.
Game sports: self-organizing dynamics and behavioural
perturbations
In consequence of results from investigating sports (squash) game behaviours
as a probability-based (Markov) process, McGarry and Franks (1996)
suggested that sports contest behaviours instead be considered as open
(complex) selforganizing systems attracting to certain (stable) patterns of
behaviour at the expense of other ones. In this context, the idea of
‘perturbations’ prompting temporary instabilities in otherwise coordinated
game behaviours was introduced as a key concept for developing appropriate
new descriptions for different sports. Perturbations in squash game
behaviours were subsequently affirmed by McGarry et al. (1996), who
reported good agreement among independent observers tasked with
identifying them. Similar evidence on perturbation behaviours in squash
games presented in the added context of dynamical self-organizing systems
are reported in McGarry et al. (1999). Further evidence of perturbations
identified from visual inspection has been reported for tennis (Jörg and
Lames 2009) and football (Hughes et al. 1998). Thus, behavioural
perturbations disrupt game equilibrium by producing system instability that is
recognized by simple observation. Relative phase offers the prospect of
identifying perturbations using a quantitative metric leading, possibly, to a
new paradigm for investigating game sports behaviours in future research.
The underlying premise for considering sports behaviours as self-
organizing dynamical systems is that, just as relative phase represents
coordination for rhythmic movement behaviour, including between persons
(Schmidt et al. 1990), so relative phase represents the spatiotemporal
coupling of players and teams underpinning the behavioural rhythms that
characterize the different game sports. In net games, anti-phase coordination
is predicted on the basis of alternating strokes, such that, as one player leaves
some neutral position to strike the shot, the other player waits for the pending
stroke in the same or equivalent neutral position (see later comment on a
common locus or separate loci of oscillations for the different racket sports).
In invasion games, however, since both teams want to score against each
other while preventing being scored against, in-phase behaviour between the
two teams is expected as the opposing players locomote in tandem with each
other. These are the basic hypotheses when considering relative phase in
game sports.
Preliminary analysis of player movement data using radial distance from
the T-position (approximate centre court location) provided good indication
of antiphase attraction in squash dyads, as hypothesized (McGarry et al.
1999). From these results, McGarry et al. (2002) proposed an accounting of
sports game behaviours predicated on coupled oscillator dynamics. Put
simply, the common descriptions of self-organizing behaviours resulting
from the coupling dynamics of pendulums (Huygens), waggling fingers
(Haken et al. 1985), different legs of different persons (Schmidt et al. 1990)
and squash players (McGarry et al. 1999) was extended to include team
sports as follows. Firstly, the reasoning of squash players oscillating about a
common locus (the T) in coupled fashion by virtue of shared information
exchanges was applied to the other racket sports, for example tennis and
badminton, while acknowledging the separate and different loci of oscillation
for the tennis players (e.g. the mid-point of the baseline) and badminton
players (e.g. centre half-court). Secondly, this same reasoning was extended
to include doubles play in these same sports, suggesting then that tennis
players couple with their double (teammate) as well as with their opponents,
thereby introducing the concept of multiple couplings as well as layered, or
nested, couplings – a coupling of couplings, if you will. Thirdly, this same
reasoning was further extended to speculate on other team sports, for
example basketball and football. The next section reports subsequent research
on these suggestions.
Attractors
Lames (1991) identified game sports as comprising ‘two parties (teams,
doubles or singles) that interact dynamically in order to score a goal/point
and simultaneously to prevent the opponent from scoring’ (p. 33), a definition
that acknowledges unique performance structure for game sports with both
parties sharing mutual competing objectives. Two main points should be
emphasized here. Firstly, strong couplings between two interacting parties are
typical of sports games. In most situations, the actions of one team cannot
properly be understood without knowing the actions of the opponent.
Secondly, these interactions are dynamic, acknowledging that intentions and
behaviours change in time. If a certain game action is successful for a given
player or team then the opponent has reason to change his (her) behaviour. If
the action is unsuccessful however, then the player or team will look for
something better to do. These interactions may further be understood in
regards to the enslavement principle that is typical for complex systems
(Haken 1993). The game state as order parameter enslaves the scope of
actions possible for the players and teams. For example, the actions of a
tennis player are constrained in part by the position and actions of the
opponent and ball. Similarly, the actions of a football player are likewise
constrained by game context, such as position of ball, teammates and
opponents. Thus, lower-level behaviours of players and teams are influenced
by higher-level game behaviours and, on the other hand, these higherlevel
game behaviours emerge from the lower-level behaviours produced by the
interactions between players and teams.
At this point, we trust that striking analogies are apparent between features
of game sports and properties owned by dynamical systems. As such, it may
be useful to consider the ultimate objectives of both parties as attractor states
in which a game exists in phase space containing all possible game states.
Figure 12.1 offers an abstract model of football game behaviour with goal
scoring considered as possible attractor states for both teams. Here, the game
progresses from top to bottom with game behaviour observed as meandering
in phase space between the two attractors. The state of attraction for one team
necessarily constitutes a state of repulsion for the opposing team. In this
accounting, the phase meandering between the two attractors is a product of
the constant interactions of opposing players and teams as they pursue their
competing objectives of attraction (repulsion) and repulsion (attraction).
Beyond general abstraction however, the challenge is to identify appropriate
variables that might adequately describe the spatial, temporal and/or
situational state of a football match at any given point in time. Possible
considerations for addressing this challenge using artificial neural networks
are presented later in this chapter.

Figure 12.1 Illustration of football as a complex dynamical system (Lames and McGarry 2007)
Net games

Tennis
Palut and Zanone (2005) first presented relative phase analysis in tennis. Four
tennis players of national level were instructed to play a rally from the
baseline while not trying to win the point in the first seven strokes. Two-
dimensional coordinate positions of the players on the tennis court were
recorded at 25 Hz for 40 rallies, from which lateral distances from the centre
line of the tennis court (i.e. the longitudinal line that divides the tennis court
into two equal parts) were obtained. Relative phase of the two players in the
lateral direction was then computed using the Hilbert transform procedure.
The results reported a bimodal distribution demonstrating approximate anti-
phase and in-phase behaviours corresponding with baseline exchanges
between line and cross-court shots, respectively. For example, a player
producing a line shot thereafter moves in the direction of the midline while
the opponent retrieving the shot moves from the midline in the opposite
direction to the player, thus yielding anti-phase. Alternatively, a player
producing a cross-court shot once more travels towards the midline following
the shot, whereas the opponent this time leaves the midline in the same
direction as the player to retrieve the shot, thereby producing in-phase. As
such, the rallies exhibited intermittent phase transitions between the generally
stable properties of in-phase and anti-phase, indicating phase attractions
within the tennis dyads as hypothesized by virtue of shared information
exchange between the players.
Lames and Walter (2006) analyzed a single rally in a top women's tennis
game with the aim of investigating relative phase in reference to game
behaviour. Firstly, it was demonstrated that a rally in tennis produced cyclical
behaviours as demonstrated in the circles observed in the phase plane results
(Figure 12.2). Secondly, transitions between in-phase and anti-phase as
reported by Palut and Zanone (2005) were once more observed (not shown).
Since these phase transitions occurred by regular switching between cross
play and line play, however, the more important challenge for game
understanding is that of identifying the ‘critical fluctuations’ in relative phase
(cf. perturbations), indicating the destabilizing of a phase relation before a
possible phase transition. In addition, relative phase measures in both lateral
and longitudinal directions are required for a more complete accounting of
tennis behaviour, as information pertaining to important tactical aspects of
game behaviour such as approaching the net cannot be obtained from lateral
data.

Figure 12.2 Phase space for two players in a tennis rally (Lames and Walter 2006); Serena Williams
(left) Justine Henin (right); △, □ = strokes of Williams; ▪, ♦ = strokes of Henin; going for the ball to
strike and returning to a neutral position results in cyclical structures in a speed/position phase space

Squash
McGarry et al. (1999) used radial distance from the T to investigate
interaction among squash dyads and consequently reported single anti-phase
coordination for all four squash rallies investigated. Radial distance was
selected on the reasoning that it offers a single metric to express the two-
dimensional movement kinematics of both players at any instant. As noted,
however, when investigating baseline tennis behaviour as a dynamical self-
organizing system, Palut and Zanone (2005) restricted analysis to the lateral
direction only and, moreover, selected velocity instead of displacement as the
kinematic metric of choice. To address these issues, McGarry and Walter
(2012) applied Hilbert analysis to squash game behaviour for purposes of
investigating the movement kinematics of squash dyads separately in lateral,
longitudinal and radial directions using displacement and velocity metrics,
with the aim of reporting on the similarities and dissimilarities that exist
between these various selected measures. Speaking generally, the findings
demonstrated strong phase attractions within squash dyads with varying
combinations of direction and kinematic metrics producing varying results.
More specifically, bimodal phase attractions were reported for both lateral
and longitudinal directions for both displacement and velocity metrics,
although the phase attraction values differed depending on the kinematic
metric, whereas the radial direction produced only single anti-phase attraction
for both metrics. As with the results from Palut and Zanone (2005), the
bimodal phase relations of the squash dyad in the lateral directions are
attributed to transitions between line and cross-court shots and, similarly, to
transitions between short and long shots in the longitudinal direction. These
results furthermore indicate that additional information for understanding
game behaviour is derived from analyzing movement kinematics in both
directions instead of a single direction which results necessarily in some loss
of information.
The different results reported by McGarry and Walter (2012) were
interpreted to indicate that selection of direction and kinematic metrics are
important considerations when investigating coordination dynamics of game
sports. Of more importance, however, was the observation that the varying
outcomes from the varying initial conditions for analysis nonetheless
conformed to common dynamical system descriptions, as expected given the
universal underpinnings of self-organizing complex systems.
Invasive games

Basketball
Team sport behaviour in basketball resulting from dynamical interactions of
dyads comprising players (Bourbousson et al. 2010a) and teams
(Bourbousson et al. 2010b) was investigated. These authors recorded the
kinematic trajectories of individual players and then undertook relative phase
analysis of all possible player combinations, yielding dyads comprising
players from the same team and from opposing teams. The results indicated
in-phase coordination between dyads, with stronger attractions observed in
the longitudinal direction (basket-to-basket) than the lateral (side-to-side)
direction. Moreover, the phase attractions were influenced by the particular
make up of the playing dyad, with dyads comprising direct opponents
identified from playing position reporting stronger phase attractions than
other permutations. This result is explained by the basketball teams using
individual marking defensive strategy rather than zone defence. Other phase
attractions reported were anti-phase in the lateral direction observed for the
playing dyads comprising the wing players from the same teams, a result
attributed to these players working in concert to increase width when
attacking and decrease width when defending.
The kinematic data for each team were obtaining from the geometric
means of the individual players data. These data were then subjected to
relative phase analysis as before, thereby investigating game dynamical
behaviour at the level of team instead of the level of players. As expected,
similar results regarding inphase attractions between teams was reported,
with stronger phase locking in the longitudinal direction than the lateral
direction. In-phase attraction between teams was furthermore stronger than
between players as anticipated from statistical considerations.

Football
Frencken et al. (2011) also used team centroids (geometric means) as well as
surface areas to analyze playing behaviours in small-sided (five versus five)
football games. The distance between team centroids was taken as an
indication of game pressure with lesser distances between teams indicating
higher pressure exerted by one or both teams on the other. The surface area
contained within the perimeter of a team configuration was interpreted as an
index of player (or team) distribution with higher values indicating higher
dispersion of players (Frencken and Lemmink 2008). Visual inspection
indicated strong in-phase couplings between teams on both variables. A
crossing of team centroids was also reported before some of the goals were
scored possibly representing behavioural perturbation in these instances.
Lames et al. (2010) also reported dynamical analysis of a football game
using relative phase (see also Cordes et al. 2011; Siegle and Lames 2013).
Position data of all players from the 2006 FIFA (International Federation of
Association Football) World Championship final were recorded at 25 Hz
using automated image processing techniques (Beetz et al. 2005) and relative
phase between teams was obtained from centroid data aggregated to 1 Hz
using Hilbert transform. Figure 12.3 presents data for both team centroids
from the first half and demonstrates strong in-phase coupling in the
longitudinal (forward–backward) direction (Xrp = 0.002° ± 5.254°). Three
main perturbations from in-phase are noted from data inspection, each of
which is associated with significant breaks from open play. The first
perturbation (minute 6) was associated with a penalty kick resulting in a goal
to France, the second perturbation (minute 19) corresponded to the equalizing
goal by Italy and may well be a result of the time taken to restart the game,
and the third perturbation (minute 33) was attributed to game injury lasting
more than a minute. These results highlight the strong in-phase couplings
attributed to behavioural interactions between teams produced in open play,
as contrasted against the weaker coupling behaviours observed during periods
of inactive game behaviour, as expected. As before, strong in-phase
couplings between teams is predicted on the basis of shared information
exchanges between players and teams as they contest the game using
common, if competing, objectives.
The coupling of teams in the lateral (side-to-side) direction was marginally
stronger than the longitudinal direction as indicated in reduced phase
variability (Xrp = 0.010° ± 3.844° versus Xrp = 0.002° ± 5.254°). In contrast
to the results from the team centroids just noted, the coupling of the team
ranges was weaker in both lateral (Xrp = 0.130° ± 18.250°) and longitudinal
(Xrp = 0.128° ± 18.319°) directions as represented by increased phase
variability. Since the lateral and longitudinal range values for a team are
determined from the maximum and minimum player coordinates within a
team configuration, this finding is expected, as the range is more sensitive
than the centroid (mean) to changes in player movements. Regarding game
behaviour, the result indicates that the teams are less coupled on measures of
dispersion (cf. surface area) than on measures of central tendency.

Figure 12.3 Longitudinal team centres, differences and relative phase for Italy and France during the
first half of the 2006 World Cup final game

Relative phase in the longitudinal direction for the midfield lines

comprising four players per team (Xrp = 0.000° ± 5.398°) returned similar
values as the entire teams, although higher sensitivity to tactical behaviour
was noted. For example, perturbation in relative phase in the corresponding
midfield lines was observed in the ninth minute. This perturbation was due to
a free kick awarded to France producing attacking play and leading to a goal-
scoring opportunity. Similar observation of a scoring opportunity created by
one of the French midfield players was noted in the 36th minute. Further
analysis of a single playing dyad comprising a French attacker and Italian
defender showed strong attraction to inphase (Xrp = 0.000° ± 11.223°), as
expected, with three main perturbations from this phase relation. The first
perturbation was associated with the scoring opportunity for France in the
ninth minute noted above for the midfield line associations. The second
perturbation in the playing dyad occurred in the 24th minute and was
attributed to dribbling action by the French attacker ultimately leading to a
scoring opportunity. The last perturbation was produced in the 43rd minute
when the French attacker entered the penalty box without attention from the
defender, who was preoccupied with assisting a teammate in tackling another
opposing player. The coupling of a second playing dyad comprising an
Italian attacker and French defender demonstrated stronger in-phase
attraction than the first playing dyad (Xrp = 0.000° ± 6.551° versus Xrp =
0.000° ± 11.223°) with two main perturbations identified from the data. The
first perturbation was associated with the ninth-minute scoring opportunity
for France mentioned already, with the French defender leaving defending
duties temporarily to join the attack. The second perturbation was associated
with a failed attack by France during which the French defender moved
towards the halfway line while the Italian attacker remained in vicinity of the
penalty box.
In short, these summary findings together with results from other
investigations of football games revealed strong in-phase attractions in both
longitudinal and lateral directions, from which it is suggested that coupling
attraction may serve as an indicator of game quality. The more a team acts as
a single unit with its behavioural organization coupled to the opposing team,
then perhaps the better is the tactical performance. The coupling attractions
should furthermore be considered in context of ball possession, however,
with the defending players and team looking to establish in-phase
associations with their opponents and the attacking players and team seeking
simultaneously to break from it and free space, perhaps by way of
perturbation. These assertions are consistent with the interpretation of results
from futsal (indoor football) game behaviour (Travassos et al. 2011, 2012),
although additional evidence is required to elucidate further on these possible
important associations.
Artificial neural networks
In this section, we extend consideration of game sports behaviours as
selforganizing dynamical systems by using artificial neural networks for
identifying patterned behaviours in large data sets. Indeed, artificial neural
networks themselves contain self-organizing features derived from rule-based
approximations of nervous system function that, importantly, allow for
automated learned recognition of structured patterns. For purposes of
continuity, the following section relates only to uses of neural networks for
analysis of football.
Football behaviour results from complex dynamic processes that make
understanding difficult beyond the limits provided by simple comparisons
such as action frequencies. For example, the vast amounts of data obtained by
automated position recording methods are often reduced to providing general
information on position distributions of players and their corresponding
movement kinematics. In contrast, self-organizing neural networks such as
dynamically controlled networks, DyCoN (Perl 2001), are useful for game
analysis in their ability for recognizing dynamic behavioural patterns
associated with playing strategies of tactical groups, such as offence and/or
defence. One example for sports practice is identifying patterned behaviours
in the changing time-dependent constellations of the playing configurations.
For obvious reasons, automated position records are most useful for obtaining
the required data, although in many instances the amount of data available is
too large for detecting important information, even for artificial neural
networks, which require large data sets. The data must therefore be reduced
before analysis by neural networks, as detailed in Perl and Memmert (2011).
For additional detail on research in football behaviour using neural networks,
the reader is referred to Perl (2008), Grunz et al. (2009), Memmert and Perl
(2009a, 2009b) and Grunz et al. (2011).
One way to reduce the amount of data is to separate the playing
constellations from their positions on the playing surface denoted by
geometrical centre, as demonstrated in Figure 12.4. This technique reduces
the large number of constellations to smaller numbers of formations, which
can be learned by the network and reduced further to characteristic types by
virtue of patterned features. Applying this protocol for each time point
produces information regarding the playing formations, the formation types
and their positions. These data, combined with other corresponding
information added manually regarding activities and outcomes, enables wide
ranging analyses, from the analysis of dynamic processes regarding tactical
behaviours to statistical analysis.

Figure 12.4 Separating a constellation of players on the playground into its formation and position

Following training of the network, the formations and formation types

contained in each dataset of the position data can be recognized and mapped
as a trajectory (Figure 12.5). Figure 12.5 shows a net of neurons depicted in
white and/or grey-shaded squares, where each shade represents a formation
type, as exemplified in Figure 12.4. Neurons of equal shade represent variant
formations of the same characteristic type; representing these variants by
single type reduces the number of different formation types to approximately
ten. This reduction of formation types has two important advantages. Firstly,
it enables statistical analyses on reasonable distribution numbers and,
secondly, the formation trajectories are smoothed, thus allowing easier
comparisons.
Figure 12.5 A trajectory of formations on the net and its reduction to a formation type trajectory

Figure 12.5 demonstrates operation of the network. The position datasets

of the game activate corresponding neurons within the network, starting with
the one identified with the black mark (bordered square). The process then
passes through some light-grey neurons, followed by some middle-grey and
some dark-grey ones, and so on. Reduced to the most important formation
types, which represent the specific behaviours of the corresponding tactical
groupings (see embedded graphic at top left, Figure 12.5), the trajectories
become much simpler to follow and comprehend. For example, the formation
types could represent various defensive formations, with different formations
indicated in the changing colours from light grey to medium grey to dark
grey and back to medium grey. Thus, an advantage of net-based analysis is
that the net is able to recognize different formation types and their dynamic
transitions within a team automatically without need of additional
information.
Net-based approaches for analyzing football behaviour:
SOCCER
In this section, some net-based approaches of behavioural (tactical) and
statistical analyses are introduced. This research uses the software application
SOCCER developed by Perl in 2010. For more information on SOCCER and
the spectrum of analyses it offers, see Perl and Memmert (2011). Figure 12.6
addresses the distribution of formations of tactical groupings and their
interactions with opponent groupings. In Figure 12.6, the defence group of
team A and the offence group of team B during the first half is represented,
where the defence types 3 (890) and 5 (1,689) and the offense types 2 (1,428)
and 4 (1,039) are most frequent. Note that frequency values denote time (in
seconds) – for example, team B played formation 4 against different
formations of team A for an aggregate of 1,039 seconds, that is for
approximately 17 minutes of the first half. The formation of offence type 4
for team B is shown in the left-hand box, while the right-hand box
demonstrates the formation of defence type 5 for team A. As can be seen
from the matrix, the combination of both types appeared to be an aggregate
of 624 seconds in the first half, comprising multiple separate instances of
varying durations. These data denote playing configurations representing the
situational context of the game at the respective points in time. They should
not necessarily be taken as indicating ball-related interactions between
offensive and defending units, however, as the ball may or may not be
possessed by players within these particular formations. This said, the most
recent version of SOCCER distinguishes between team formations with ball-
related and non-ball-related associations.
Figure 12.6 Frequencies of formations and their correlations

SOCCER offers many types of distribution analyses but little information

by way of underlying game dynamics that produced these frequency counts.
Regarding game dynamics, the phase trajectories of formation types of
offence group B and defence group A are presented in Figure 12.7 (from the
21st to the 30th minute). Here, the meaning of phase is somewhat different
from earlier usage reported in this article. Unlike before, phase does not
represent position relations between the players or teams but, instead, it
represents the formation relations between the tactical groupings being
investigated. In the example presented, the time-depending distribution of the
correspondences between type B4 and type A5 is highlighted for
approximately 200 of the 640 seconds observed in the first half. These B4–
A5 playing formation frequencies indicate this particular combination of
playing configurations between the two teams as being a common phase of
game behaviour.
Specific phase patterns may be of interest to sports practitioners for
addressing questions regarding particular tactical concepts in the context of
specific formation conditions, as well as for evaluating them by the
success/failure outcomes of certain team activities (e.g. ball possession, goal
scoring opportunities). In fact, such tactical phase patterns could be detected
using neural networks on a second level of analysis. In the following, we
offer an example of a feature analysis which demonstrates the usefulness of
combining quantitative and qualitative aspects for advancing understanding
of game behaviour.
Figure 12.8 presents the time section from Figure 12.7. First inspection of
the formation phases shows that team B has frequent changes between 2 and
4 while team A has similar changes between 5 and 3, an expected finding,
given that these are the most frequent formations of the respective teams.
Second inspection shows correspondences between these changes, two of
which are marked in Figure 12.8. The process starts in T21 with a phase of
[B,A] = [4,5] and, ignoring spikes, is followed by change in B to [B,A] =
[2,5], and then change in A to [B,A] = [2,3] before change in B to [B,A] =
[4,3]. This sequential change in formation combinations between the two
teams, which appears in similar fashion between T29 and T30, is represented
in the preceding sentence by the use of boldface type.
Figure 12.7 Distribution of a typical pair of formations between minutes 21 and 30

Figure 12.8 Example of a typical tactical pattern produced between the two teams

Initial impressions may suggest randomness in these patterns, the more so

since the phase changes are not observed as idealized (error free) patterns but
instead appear disturbed by secondary formation types (error). Restated, the
patterned formations are not obvious and unambiguous but, instead,
demonstrate variability typical of real-world data. Taking the respective
success of the team actions into consideration offers additional information
and perspective. In addition to the automated position data, there are
corresponding success valuations of time-dependent team activities, such as
flanking, tackling and shooting. Here, ‘success’ is not used in a game
theoretical way where the aggregate of win and loss equals some fixed value
(e.g. zero or unity). For example, a successful ball win by team A may be
accompanied with successful tackling by team B without necessarily
reducing the success awarded either team. In addition, success in the context
of a formation does not necessarily mean success of the corresponding
tactical group, simply that while the tactical group was in that formation the
team was awarded a successful action.
 Figure 12.9 demonstrates the results of combining quantitative and
qualitative analysis. By way of example of combining SOCCER-based
analysis formations with corresponding success values, we introduce the
following nomenclature: SB([B,A] = [4,5]) = 0.89 which means that the
tactical formation (type 4) of team B when competing against the tactical
formation of team A (type 5) yielded an 89% success rate for Team B. Note
that references to tactical formations of both teams represent specific playing
configurations within these teams and not necessarily either or both teams in
their entirety. Given the nomenclature explained above, the sequences that
yielded most success were SB([B,A] = [4,5]) = 0.89, SB([B,A] = [2,5]) =
0.82, SB([B,A] = [2,3]) = 0.83, SB([B,A] = [4,3]) = 0.84, SA([A,B] = [5,4])
= 0.87, SA([A,B] = [5,2]) = 0.84, SA([A,B] = [3,2]) = 0.91 and SA([A,B] =
[3,4]) = 0.94.

Figure 12.9 Prototype of the tactical pattern from Figure 12.5, together with success values

Figure 12.9 shows the simplified phase diagrams of teams A and B

combined with the corresponding success values. The figure is reminiscent of
the strategic trees of game theory, where the players respond with tactical
decisions to drive out the opponent(s) from successful areas and/or escape
from unsuccessful ones. One interpretation then from Figures 12.8 and 12.9
might be that team B tries unsuccessfully to leave an unwanted situation. In
phase 1, the attacking team (B) has a success rate of 89%, whereas the
defending team (A) has 87% success, indicating that this particular
constellation is not significantly advantageous for team B. As such, team B
might look to change attacking behaviour by changing formation from, say,
type 4 to type 2 (phase 2) although this formation yielded less success than
the previous one. This change in attacking formation for team B, moreover,
results in a changed defensive formation for team A from type 5 to type 3
(phase 3). These changes in the last two phases advantages team A at the
expense of team B. Of course, additional interpretation of the specific team
formations in respect of game context is required for possibly identifying
future optimized game behaviours. For now, therefore, this type of
interpretation must remain speculative but it nonetheless serves to
demonstrate how repeated patterns identified by neural network analysis as
tactical invariants could usefully be applied in future sports practice.
In brief, neural network analysis of game behaviour offers the prospect for
knowledge advancement when data are reduced to kernel information
comprising formation types (or playing constellations). This analytical
technique permits detection of tactical behavioural patterns not available to
more traditional statistical methods, all the while retaining key information
for purposes of statistical analyses if desired. Moreover, patterns such as
those from Figure 12.8 can be detected by a second-level network, if
required. For complex pattern recognition the inherent ability of
‘fuzzification’ helps the net to recognize patterns from other tactical
formations, even if they should be disturbed by randomness and/or variation.
A combined approach using networks and tools like SOCCER offers a
productive combination of qualitative pattern recognition and quantitative
statistical analysis.
Summary
In this chapter, we have presented consideration of sports games as
selforganizing systems on the basis of coupled oscillator dynamics. Relative
phase was introduced as a metric for investigating the behavioural dynamics
of both net games (e.g. tennis and squash) and invasive games different (e.g.
basketball and football), with in-phase and anti-phase in some instances
representing attractor states for game behaviours at the expense of other
possible phase relations. The notion of behavioural perturbations serving to
destabilize the system from these phase attractions was introduced and
considered important for game description. Moreover, the presence of
behavioural perturbations in game sports has been validated from visual
inspection although their predicted corresponding associations with relative
phase variability await future demonstration for the most part. Beyond
considerations of perturbations, relative phase and transitions between phase
attractors, artificial neural networks were introduced as a means of
identifying game structure from large datasets using football data as example.
As with earlier demonstrations using relative phase, various unique phase
structures (playing configurations) were identified by the neural networks,
together with time-dependent phase changes indicating the underlying game
dynamics. In closing, the approaches outlined in this article go some way to
describing and explaining the behavioural structure of sports games and their
underpinning dynamics.
References
Beetz, M., Kirchlechner, B. and Lames, M. (2005) Computerized real-time
analysis of football games. IEEE Pervasive Computing, 4 (3): 33–9.
Bourbousson, J., Seve, C. and McGarry, T. (2010a) Space-time coordination
dynamics in basketball: Part 1. Intra- and inter-couplings among player
dyads. Journal of Sports Science, 28: 339–47.
Bourbousson, J., Seve, C. and McGarry, T. (2010b) Space-time coordination
dynamics in basketball: Part 2. The interaction between the two teams.
Journal of Sports Science, 28: 349–58.
Cordes, O., Siegle, M., Stockl, M., Durus, M., Beetz, M. and Lames, M.
(2011) Kopplung von Mannschaften, Mannschaftsteilen und Spielern im
Fusball – Berechnung mit Hilfe der relativen Phase, in D. Link and J.
Wiemeyer (eds) Sportinformatik trifft Sporttechnologie. Hamburg: Czwalina,
pp. 186–90.
deGuzman, G. C. and Kelso, J. A. S. (1991) Multi-frequency behavioural
patterns and the phase attractive circle map. Biological Cybernetics, 64: 485–
95.
Frencken, W., Lemmink, K. Dellemann, N. and Visscher, C. (2011)
Oscillations of centroid position and surface area of soccer teams in small-
sided games. European Journal of Sport Science, 11: 215–23.
Frencken, W. G. P. and Lemmink, K. A. P. M. (2008) Team kinematics of
small-sided soccer games: a systematic approach, in T. Reilly and F.
Korkusuz (eds) Science and Football VI. London: Routledge, pp. 161–6.
Grunz A., Memmert D. and Perl J. (2009) Analysis and simulation of actions
in games by means of special self-organizing maps. International Journal of
Computer Science in Sport, 8: 22–37.
Grunz, A., Endler, S. Memmert, D. and Perl, J. (2011) Netz-gestutzte
Konstellations- Analyse im Fusball, [Net-based constellation analysis in
soccer] in Link and Wiemeyer (eds) Schriften der Deutschen Vereinigung für
Sportwissenschaft, vol. 217. Hamburg: Czwalina, pp. 111–15.
Haken, H. (1993) Advanced Synergetics: Instability Hierarchies of Self-
organizing Systems and Devices. New York: Springer.
Haken, H., Kelso, J. A. S. and Bunz, H. A. (1985) Theoretical model of phase
transitions in human hand movements. Biological Cybernetics, 51: 347–56.
Hughes, M., Dawkins, N., David, R. and Mills, J. (1998) The perturbation
effect and goal opportunities in soccer. Journal of Sports Sciences, 16: 20.
Jörg, D. and Lames, M. (2009) Perturbationen im Tennis – Beobachtbarkeit
und Stabilitat, in M. Lames, C. Augste, O. Cordes, Ch. Dreckmann, K.
Gorsdorf and M. Siegle (eds) Gegenstand und Anwendungsfelder der
Sportinformatik. Hamburg: Czwalina, pp. 86–96.
Kelso, J. A. S. (1984) Phase transitions and critical behaviour in human
bimanual coordination. American Journal of Physiology: Regulatory,
Integrative and Comparative Physiology, 15: R1000–4.
Kelso, J. A. S. and Jeka, J. J. (1992) Symmetry breaking dynamics of human
multilimb coordination. Journal of Experimental Psychology: Human
Perception and Performance, 18: 645–68.
Kelso, J. A. S., Buchanan, J. J. and Wallace, S. A. (1991) Order parameters
for the neural organization of single, multijoint limb movement patterns.
Experimental Brain Research, 85: 432–44.
Kelso, J. A. S., Holt, K. G., Rubin, P. and Kugler, P. N. (1981) Patterns of
human interlimb coordination emerge from the properties of non-linear, limit
cycle oscillatory processes: theory and data. Journal of Motor Behaviour, 13:
226–61.
Lames, M. (1991) Leistungsdiagnostik durch Computersimulation: Ein
Beitrag zur Theorie der Sportspiele am Beispiel Tennis. Frankfurt: Harry
Deutsch.
Lames, M. and Walter, F. (2006) Druck machen und ausspielen: Die relative
Phase und die Interaktion in den Ruckschlagsportspielen am Beispiel Tennis.
Spectrum der Sportwissenschaften, 18 (2): 7–24.
Lames, M., Ertmer, J. and Walter, F. (2010) Oscillations in football – order
and disorder in spatial interactions between the two teams. International
Journal of Sports Psychology, 41 (4 Supplement): 85–6.
McGarry, T. and Franks, I. M. (1996) In search of invariant athletic
behaviour in competitive sport systems: An example from championship
squash match-play. Journal of Sports Sciences, 14: 445–56.
McGarry, T. and Walter, F. (2012) Sport competition as a dynamical self-
organizing system: example from the movement coordination kinematics of
squash players. International Journal of Motor Learning and Sport
Performance, 2: 59–67.
McGarry, T., Anderson, D. I., Wallace, S. A., Hughes, M. and Franks, I. M.
(2002) Sport competition as a dynamical self-organizing system. Journal of
Sports Sciences, 20: 771–81.
McGarry, T., Khan, M. A. and Franks, I. M. (1996) Analyzing championship
squash match-play: in search of a system description, in S. Haake (ed.) The
Engineering of Sport. Rotterdam: Balkema, pp. 263–9.
McGarry, T., Khan, M. A. and Franks, I. M. (1999) On the presence and
absence of behavioural traits in sport: an example from championship squash
match-play. Journal of Sports Science, 17: 298–311.
Meijer, O. G. (2001) An introduction to the history of movement science, in
M. L. Latash and V. Zatsiorsky (eds) Classics in Movement Science.
Champaign, IL: Human Kinetics, pp. 1–57.
Memmert, D. and Perl, J. (2009a) Game creativity analysis by means of
neural networks. Journal of Sports Sciences, 27: 139–49.
Memmert, D. and Perl, J. (2009b) Analysis and simulation of creativity
learning by means of artificial neural networks. Human Movement Science,
28: 263–82.
Palut, Y. and Zanone, P. S. (2005) A dynamical analysis of tennis players’
motion: concepts and data. Journal of Sports Science, 23: 1021–32. Perl, J.
(2001) DyCoN: Ein dynamisch gesteuertes Neuronales Netz zur
Modellierung und Analyse von Prozessen im Sport, [A dynamically
controlled neural network for modelling and analysis of processes in sport] in
J. Perl (ed.) Sport and Informatik VIII. Koln: Straus, pp. 85–98.
Perl, J. (2008) Modelling, in P. Dabnichki and A. Baca (eds) Computers in
Sport. Southampton: Wit Press, pp. 121–60.
Perl, J. and Memmert, D. (2011) Net-based game analysis by means of the
software tool SOCCER. International Journal of Computer Science in Sport,
10 (2): 77–84.
Schmidt, R. C., Carello, C. and Turvey, M. T. (1990) Phase transitions and
critical fluctuations in the visual coordination of rhythmic movements
between people. Journal of Experimental Psychology: Human Performance
and Perception, 16: 227–47.
Siegle, M. and Lames, M. (2013) Modeling soccer by means of relative
phase. Journal of Systems Science and Complexity, 26: 14–20.
Travassos, B., Araújo, D., Vilar, L. and McGarry, T. (2011) Interpersonal
coordination and ball dynamics in futsal (indoor football). Human Movement
Science, 30: 1245–59.
Travassos, B., Araújo, D., Duarte, R. and McGarry, T. (2012) Spatiotemporal
coordination patterns in futsal (indoor football) are guided by informational
game constraints. Human Movement Science, 31 (4): 932–45.
Part 3
Complexity sciences and sport
performance
13 Ecological dynamics as an
alternative framework to
notational performance
analysis
Luís Vilar, Carlota Torrents, Duarte
Araújo and Keith Davids
Practitioners and sports scientists have been always seeking to identify key
factors or characteristics that can distinguish between successful and less
successful players and teams (Nevill et al. 2009; Reilly et al. 2000).
Quantitative analysis of performance provides coaches with additional
information that describes performance in detail beyond that which they can
access through recall of personal observations (Borrie et al. 2002). Such
critical information allows coaches to improve performance during matches
and practice, through the improvement of feedback in an appropriate form
(Franks 1997).
Traditional performance analysis have sought to understand performance
by identifying the behaviours that are important for a given sport (Lames and
McGarry 2007). Notational analysis techniques have described the
performance tendencies of players and teams, and strengths and weaknesses
in specific performance situations in a range of sports (e.g. playing long or
spreading wide during transitions phases in association football). This
methodology has sought for performance indicators that helped to
characterize successful and unsuccessful performance and creating awareness
among players and coaches of how individual players can influence team
patterns (Hughes and Franks 2004; McGarry 2009).
Performance indicators are variables that can be categorized as scoring
indicators (e.g. goals, baskets, winning shots, errors, the ratios of winners to
errors and goals to shots) or quality indicators (e.g. turnovers, tackles,
passes/ball possession; Hughes and Bartlett 2002). Performance indicators
are often ranked through the use of statistical procedures, such as factor
analysis and multiple regressions, according to their influence on sport
outcome. In addition, performance indicators may also be considered by their
combinatory temporal relationships, typically recorded in a discrete
sequential fashion to operationalize patterns of play of each team. For
example, analysts have been also considering the sequential actions of
players within the same team. By recording discrete action frequencies in a
number of games in a sequential ‘who[did]–what–where–when’ fashion,
researchers have identified teams' temporal patterns of play (Borrie et al.
2002; McGarry 2009). This sequential analysis also allows the analysis of
interrelationships between performance variables. Temporal patterns are able
to reveal those aspects of social interaction that are not immediately
observable, detecting the hidden structures underlying an interactive
situation, such as a game (Anguera 2005; Anguera and Jonsson 2003;
Fernandez et al. 2009).
To provide meaningful insights to researchers about successful
performance, indicators must be highly correlated with any associated
outcomes (i.e. winner, error or neutral outcome; Hughes and Bartlett 2002),
which has not been always the case (McGarry 2009). For example, one of the
most intriguing indicators of team performance in association football has
been the style of play (e.g. ‘direct play’ or‘ possession play’), measured by
the number of passes that a team takes to score a goal (Franks et al. 1990;
Grehaigne 1999; Hughes and Franks 2005; Hughes et al. 1988). Although
high-level practitioners have emphasized the use of longer passing sequences
as a means of scoring goals, notational analysts have reported that the strike
ratio of goals from shots is better for ‘direct play’ than for ‘possession play’
(Hughes and Franks 2005).
The presumed correspondence of this finding to competitive outcomes
remains unclear, owing to a lack of a theoretical understanding of how to
interpret the data. This discrepancy suggests that neglecting the constraining
influence of information from the performance environment, such as the
active role of opponents in shaping players' actions and decision making, may
fail to provide understanding on successful performance. By omitting
reference to the why and how of performance that underlie the structure of
recorded behaviours, which would define their functional utility, notational
analysis has been shown to be somewhat reductionist (Glazier 2010;
McGarry 2009). These facts point to the need for descriptive analysis, such as
that provided by notational techniques, to be complemented with a sound
theoretical rationale that explains sport performance at the player–
environment scale of analysis. In this chapter, we present ecological
dynamics as a reliable framework for the analysis of performance in team
sport, since it recognizes the ‘degeneracy’ (inherent adaptive flexibility in
achieving successful performance outcomes) of players, including sports
teams. This framework provides an understanding of how players perform
successfully, by using information from their environment, and explains how
the same successful performance outcomes can emerge from different
movement or tactical patterns (a phenomenon also known as motor
equivalence).
Ecological dynamics approach to performance analysis
From an ecological dynamics perspective, performers and teams are
neurobiological and social neurobiological systems, respectively, in which
patterns emerge from the interaction of their many degrees of freedom or
constraints, through a process of self-organization (Kauffman 1993; Warren
2006). Self-organization in complex neurobiological systems is not a random
or completely ‘blind’ process in which any pattern can result; rather, it is
influenced by surrounding informational and physical constraints and
intentions of a performer (Seifert and Davids 2012). Constraints are features
that surround a complex system and reduce the number of configurations that
are available to it as it interacts with the performance environment (Davids et
al. 2013). Most interest for researchers is to examine how these macroscopic
patterns of coordination are constructed, sustained and dissembled, through
the continuous exchange of information between the athlete and their
performance environment (Araújo et al. 2006; Kugler and Turvey 1987).
Ecological dynamics captures the intertwined relationships of the
performer and their performance environment through an information-based
perspective (Warren 1984; Warren and Whang 1987). This rationale was
grounded on James Gibson's (1979) arguments that organisms, including
humans, do not necessarily need representations of the world (e.g. knowledge
about its properties) to be able to perceive its structure. Rather, the energy
flows or arrays that surround individuals as they displace though the
performance environment can act as specifying information, allowing
individuals to directly perceive the properties of the environment and guide
action. For example, the rate of dilation of an approaching ball on the retina
of a goalkeeper informs him/her about the time-to-contact remaining before
the ball arrives at the catching point. In principle, it is not necessary for the
goalkeeper to separately compute either distance or speed of the ball to
perceive time-to-contact information, under a constant approach velocity
(Lee et al. 1983). The perception of specifying informational variables allows
the perception of opportunities for action or affordances (e.g. a gap between
two defenders promotes a pass to an attacking teammate) offered by the
environment with the respect to the individual action capabilities, so that
individuals directly perceive what they can or cannot do (Turvey 1992). In
addition, individuals were shown to be also able to perceive affordances for
others; that is, to identify relations between other performers (e.g. teammates
and opponents) and key environmental objects (e.g. the locations of the ball
and goal) that others may use to guide their behaviour (Richardson et al.
2007; Stoffregen et al. 1999). In this sense, ecological dynamics suggests that
interpersonal coordination in team sports is grounded in the players' ability to
identify the multidimensional spectrum of competitive affordances for
themselves and others (both teammates and opponents) and acting adaptively
towards acquiring (collective) goals.
When performers act upon affordances, a synergy emerges at the
ecological scale, attracting the system towards a stable state of coordination
(i.e. an attractor or preferable mode of coordination; Warren 2006).
Hypothetically, if neither the individual constraints nor the environmental
properties change, behaviour would remain attached to the same specific
attractor. However, owing to the complex spatiotemporal relations among
performers that characterize team sports, ecological constraints change on a
moment-to-moment basis. On one hand, constraints of the individual may
change, even though the performance environment may remain static. For
example, in a team game like futsal (a type of five-a-side association football
played on an indoor court) when a fatigued defender late in performance
cannot accelerate quickly enough to intercept a shot that it would have been
possible to intercept earlier in competitive performance (Fajen et al. 2009).
On the other hand, changes may occur in the performance environment while
the performer's capabilities remain constant. For example, at any moment in
futsal, a goal path may open and a shooting chance may be offered.
Milliseconds later, a defender may move into the line of the ball's trajectory
with the goal and a successful shot is no longer possible (Fajen et al. 2009).
The instability that characterizes the interacting constraints in team sports
makes opportunities for action continuously come and go instantaneously,
leading to fluctuations in the organizational states of games (e.g. increased
variability in the way that attackers and defenders coordinate their actions;
Araújo and Davids 2009). When these fluctuations are powerful enough to
break the existing balance between performers (i.e. if the equilibrium
between attacking and defending players is successfully destabilized), a
symmetry-breaking process occurs. That is, a previously stable state of the
game transits to a new dynamic state of organization (e.g. an attacker dribbles
past a first defender, inducing a second defender to cover and leading to a
change in the structure of the defending team; Davids et al. 2003).
 In summary, ecological dynamics provides explanations about the way in
which the interaction between players and information from the performance
environment constrains the emergence of patterns of stability (i.e.
coordination between performers), variability (loss of coordination between
performers) and symmetry breaking in organizational states (i.e. how new
patterns of coordination emerge in performance) in such systems (Vilar et al.
2012a). This is precisely what sport scientists and coaches need to understand
in analyses of team game performance (Araújo et al. 2006; Davids et al.
1994; Handford et al. 1997).
Interpersonal coordination in team sports
In its recent years, ecological dynamics has examined coordination among
players in team sports by considering the pattern forming dynamics of
attacker–defender sub-systems as a basic unit of investigation (Araújo et al.
2004; Davids et al. 2006). For example, Araújo and colleagues (2004; 2006)
provided empirical data about one-versus-one sub-phases in basketball,
showing that the attacker's and defender's displacements were highly coupled.
The defender was shown to counteract the movements of the attacker, in
order to maintain system symmetry (to prevent the immediate attacker from
gaining a positional advantage that allows him/her to score). However, as the
attacker approached the location of the basket, critical fluctuations were
observed in the dyadic stability and a sudden change in the organization of
the system occurred towards one of the two following possible states: (i) an
advantage for an attacker (i.e. the defender may not be able to balance the
attacker's actions and the attacker may move past the defender); or (ii) an
advantage for a defender (i.e. the attacker was not able to break system
symmetry and a defender may intercept the ball). This research provided
some understanding about how players continuously coupled their actions to
information from their opponent and the location of the target, to acquire a
desirable state of coordination that allow them to attain their goals in one-
versus-one sub-phases in team sports.
In order to conduct appropriate performance analysis, research should
examine the spatiotemporal relations that constrain pattern-forming dynamics
in team games players during actual competitive performance, such as an
international championship (Vilar et al. 2012b). In competitive scenarios, the
existence of more than one attacker–defender dyadic system performing
simultaneously is a distinct constraint is likely to be a major influence on the
inter-player coordination dynamics. In the next section, we discuss some
extensive work on performance analysis in futsal, seeking to provide
understanding on how players coordinate their actions and attain success by
using information from other players (teammates and opponents) and the
locations of the goal and the ball.

Exemplar analysis of performance in futsal

Previous research on one-versus-one sub-phases in basketball showed that a
phase transition in the attacker's and defender's distances to the basket
precipitated a scoring event. However, in the competitive environment of
futsal, it could also be suggested that a goal could be scored merely as a
consequence of instabilities in a defender's alignment between the goal and
an attacker's position (Vilar et al. 2012b). In this sense, we recorded ten futsal
games between five national teams in the 2009 Lusophony Games and
analyzed the displacement trajectories of the four outfield players in 13
sequence of plays that ended in a goal being scored (N = 52). We considered
the coordination between the distances and angles of the attacker and nearest
defender to the centre of the goal (Vilar et al. 2012b). To capture stabilities
and instabilities in attacker–defender phase relations, as well as phase
transitions in a given data sequence, we used relative phase (see Chapter 7).
Analysis of the interpersonal coordination between attackers and defenders
in five-versus-five competitive environments in futsal showed that stable
patterns of coordination emerged from changes to both players' distances and
angles to the goal (Figure 13.1C and 13.1F, respectively). Individual analysis
of the coordination between the attacker who scored the goal and the nearest
defender showed that the defender seemed to be always closer to the goal
than the attacker 5 and aligned with the goal and his direct opponent (Figure
13.1A and 13.1D). In-phase patterns of coordination suggest that while the
attacker was in possession of the ball and the defender was between the goal
and attacker, the stability of the dyadic system was maintained (Figure 13.1B
and 13.1E). However, as the sequence of play evolved towards the goal, the
defender's efforts to maintain system stability were often insufficient and
critical fluctuations precipitated more than one phase transition. States of
system stability showed less duration, providing attacker 5 with opportunity
to score a goal. Attacker 5 seemed to have used lateral displacement to
increase the angle to the goal relative to the defender's position, while
decreasing the distance to the goal, to break system symmetry. This analysis
suggested that only when symmetry-breaking processes emerged near the
goal and the defenders did not have the collective ability to re-establish
dyadic system stability, a goal opportunity presented itself. Leading the
systems towards critical regions of instable coordination is suggested to
constrain the attackers' ability to create opportunities to score.
Figure 13.1 The constraint of goal location on coordination processes in dyadic systems presented in
decomposed format: (left column) distances of each player to the centre of the goal; (right column)
angles of each player to the centre of the goal; (a) and (b) exemplar data from attacker five [A5] and
nearest defender [Def]; (A) and (B) exemplar data from attacker five [A5] and nearest defender [D];
(C) and (D) dynamics of the relative phase of the exemplar data from A5 and nearest D; (e) and (f)
frequency histograms of the relative phases of all A–D dyadic systems (n = 52) (data from Vilar et al.
2012a)

However, in futsal, breaking symmetry with the defender does not

guarantee per se goal scoring. Attackers must shoot the ball and override
simultaneously the opposition of the immediate defender and the goalkeeper
in order for success to be obtained. To examine the influence of spatial and
temporal constraints on shooting performances, we used the model of the
required velocity to intercept moving objects (Peper et al. 1994). This model
suggests that catching a ball is related to the individuals' ability to gear the
velocity of the hand to a specified value that ensures that the hand is located
at the right place and the right time, regardless of where this might be. Using
the same data shown previously, we examined the locations of the ball, the
nearest defender and the goalkeeper, from the moment when the ball was shot
until it was intercepted or entered the goal, in plays that ended in a defender's
interception, in a goalkeeper's save and in a goal. We computed the
interception points of the defender and the goalkeeper by recording their
shortest distance to the ball's trajectory during the act of shooting. Since we
considered the working point from each player, the players' distances to the
interception points considered half of the players' shoulder-to-shoulder width
(0.40 metres, estimation) and the radius of the futsal ball used in this
tournament (0.10 metres). We also calculated the time for the ball to arrive at
each player's interception point. Finally, the required velocities of the
defender and the goalkeeper to intercept the ball were computed by dividing
each player's distance to the interception point by the time for the ball to
arrive to each player's interception point (Figure 13.2).

Figure 13.2 Mean values and standard error of the required velocity for intercepting the ball of (A)
defender and (B) goalkeeper in shots that ended in a defender's interception, in a goalkeeper's save and
in a goal. The represented levels of statistical significance are P < 0.05 (*), P < 0.01 (**) and P < 0.001
(***). Note that the required velocity of the goalkeeper was not measured when the defender
intercepted the ball, since it is impossible to compute the goalkeeper's interception point
 The mean values of the required velocity of the defender to intercept the
ball were significantly lower in plays ending in a defender's interception
(mean [M] = 3.29, standard error [SE] = 0.39) than in plays ending in a
goalkeeper's save (M = 32.16, SE = 10.11) and in plays ending in a goal. This
finding suggests that the time taken for the ball to arrive at the interception
point was higher than the defender's ability to move to the interception point.
That is, the time allowed for Ecological dynamics as an alternative
framework 235 Figure a defender to close the gap between him and the
interception point and to intercept the ball was within the defender's action
capabilities. In this sense, to score goals, attackers need to move in order to
‘pull’ the opponents away from an imaginary line between him and the centre
of the goal.
Similarly, the mean values of the required velocity of the goalkeeper to
intercept the ball were significantly lower in plays ending in a goalkeeper's
save (M = 3.29, SE = 0.39) than in plays ending in a goal (M = 12.97, SE =
4.41). This result suggests that the time for the ball to arrive at the
interception point was greater than the time needed for the goalkeeper to
arrive at the same point. These data suggest that attackers need to be able to
identify an opportunity in the performance environment to shoot the ball
without allowing the immediate defender and the goalkeeper to move fast
enough to intercept the ball. Such opportunities emerge not only from the
information from the performance environment but also from attackers'
action capabilities. That is, attackers should scale information they perceive
according to their own capabilities. Decisions where, when and how to shoot
should always be placed in a performance context and should be guided by
the information of both the time for the ball and for the opponents to arrive at
potential interception points (Watson et al. 2011).

Implications and applications of ecological dynamics perspective

on sport analysis and performance
Analyzing performance from an ecological dynamics perspective provides a
theoretical rational that explains how reductionist notational analysis may be.
To enhance the validity of theoretical interpretations, analysts need to move
beyond merely documenting performance statistics to study the emergent
interactions between players, in key areas of the field, which underpin
success in team sports (Vilar et al. 2012a). For example, previous research on
shooting performance has investigated independently the placement of the
ball in the goal (Alcock 2010; Hughes and Wells 2002; López-Botella and
Palao 2007; Morya et al. 2004), the velocity of ball flight (Alcock 2010;
Kerwin and Bray 2006) and the pitch location of a shot (Alcock 2010; Ensum
et al. 2000). However, without considering how these variables relate to a
defenders' positioning on field and a goalkeeper's positioning in goal and
their ability to displace towards possible interception points, researchers will
not be able to explain how successful and unsuccessful shots occur during
futsal competitive performance. As we have stated in the introduction,
sequential analysis has also studied the interaction between variables in sport
teams (Johnson 2006), but ecological perspective provides the theoretical
framework for identifying the key informational variables of sport success.
Its application requires a collective variable to be found that describes the
behavioural dynamics of the system. At present, investigation is focused on
analyzing the space–time movement trajectories of the ball kinematics, as
well as those of the players (Travassos et al. 2011).
Ecological dynamics also has major implications for the design of
representative training tasks in team sports. The key informational variables
that players use to control their performances should be simulated (i.e.
represented) in training tasks, allowing them to become better attuned to
functionally coupling information and movement during practice (Davids et
al. 2005). Major implications of these ideas also exist for development
programmes in team sports. Instead of deconstructing tasks as repetition drills
for learners, coaches should simplify tasks (e.g. by reducing the numbers in
teams during small-sided practice games: three-versus-three, five-versus-four
or six-versus-four) to facilitate players' performances (Vilar et al. 2012a).
Conclusions
In this chapter, we have shown how ecological dynamics provides
understanding about how players use information from their performance
environment to attain successful performances in team sports. Our
programme of research has shown that attackers seek to break symmetry with
their nearest opponents, as defenders try to maintain system symmetry by
remaining between their own goal and the immediate attacker. Slight changes
in player behaviours may induce a symmetrybreaking process in the state of
the dyadic systems. In this case, when the defenders do not have the
collective ability to re-establish dyadic system stability, a shot at goal may
occur. Attackers must perceive the locations of the immediate defender and
the goalkeeper and shoot the ball in a specific direction and with a specific
velocity that requires the opponents to displace faster than they are able to
intercept the ball.
Our results are encouraging and provide evidence that notational analysis
based on performance statistics can be too reductionist. Moreover, this
approach has potential in areas such as training and programming in team
sports, proposing the simplification of tasks instead of deconstructing them.
By unveiling the influence of interacting task constraints on the emergent
selforganized behaviours of players during performance, ecological dynamics
reveals itself as a powerful tool for both researchers and practitioners in sport
performance analysis.

Questions for students

1. What are performance indicators? Give some examples.
2. What are affordances? Give one example.
3. What information should attackers use to create opportunities to
score?
4. How may ecological dynamics overcome notational analysis
limitations? Give one example.
5. What implications does ecological dynamics have for practice?
References
Alcock, A. (2010) Analysis of direct free kicks in the women's football
World Cup 2007. European Journal of Sport Science, 10 (4): 279–84.
Anguera, M. T. (2005) Microanalysis of T-patterns: analysis of
symmetry/asymmetry in social interaction, in L. Anolli, S. Duncan, M.
Magnusson and G. Riva (eds) The Hidden Structure of Social Interaction:
From Genomics to Culture Patterns. Amsterdam: IOS Press, pp. 51–70.
Anguera, M. T. and Jonsson, G. K. (2003) Detection of real-time patterns in
sport: interactions in football. International Journal of Computer Science in
Sport, 2: 118–21.
Araújo, D. and Davids, K. (2009) Ecological approaches to cognition and
action in sport and exercise: ask not only what you do, but where you do it.
International Journal of Sport Psychology, 40 (1): 5–37.
Araújo, D., Davids, K., Bennett, S., Button, C. and Chapman, G. (2004)
Emergence of sport skills under constraints, in A. M. Williams and N. J.
Hodges (eds) Skill Acquisition in Sport: Research, Theory and Practice.
London: Routledge, pp. 409–33.
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7: 653–76.
Borrie, A., Jonsson, G. and Magnusson, M. (2002) Temporal pattern analysis
and its applicability in sport: an explanation and exemplar data. Journal of
Sports Sciences, 20: 845–52.
Davids, K., Handford, C. and Williams, M. (1994) The natural physical
alternative to cognitive theories of motor behaviour: an invitation for
interdisciplinary research in sports science? Journal of Sports Sciences, 12
(6): 495–528.
Davids, K., Glazier, P., Araújo, D. and Bartlett, R. (2003) Movement systems
as dynamical systems. the functional role of variability and its implications
for sports medicine. Sports Medicine, 33 (4): 245–60.
Davids, K., Renshaw, I. and Glazier, P. (2005) Movement models from
sports reveal fundamental insights into coordination processes. Exercise and
Sport Sciences Reviews, 33 (1): 36–42.
Davids, K., Button, C., Araújo, D., Renshaw, I., and Hristovski, R. (2006)
Movement models from sports provide representative task constraints for
studying adaptive behavior in human movement systems. Adaptive Behavior,
14 (1): 73–95.
Davids, K., Araújo, D., Vilar, L., Renshaw, I. and Pinder, R. (2013) An
ecological dynamics approach to skill acquisition: implications for
development of talent in sport. Talent Development and Excellence, 5 (1):
21–34.
Ensum, J., Williams, M. and Grant, A. (2000) Analysis of the attacking set
plays in Euro 2000. Insight: The FA Coaches Association Journal, 4: 36–9.
Fajen, B., Riley, M. and Turvey, M. (2009) Information, affordances, and the
control of action in sport. International Journal of Sport Psychology, 40 (1):
79–107.
Fernandez, J., Camerino, O., Anguera, M. T. and Jonnson, F. K. (2009)
Identifying and analyzing the construction and effectiveness of offensive
plays in basketball by using systematic observation. Behavior Research
Methods, 41 (3): 719–30.
Franks, I. (1997) Use of feedback by coaches and players, in T. Reilly, J.
Bangsbo and M. Hughes (eds) Science and Football III. London: E. and F. N.
Spon, pp. 267–8.
Franks, I., Partridge, D. and Nagelkerke, P. (1990) World Cup 90: A
Computer Assisted Technical Analysis of Team Performance Technical
Report for the Canadian Soccer Association. Vancouver: University of
British Columbia.
Gibson, J. (1979) The Ecological Approach to Visual Perception. Boston,
MA: Houghton Mifflin.
Glazier, P. (2010) Game, set and match? Substantive issues and future
direction in performance analysis. Sports Medicine, 40 (8): 625–34.
Grehaigne, J. (1999) Systemic approach and soccer, in M. Hughes (ed.)
Notation of Sport III. Cardiff: Centre for Performance Analysis, UWIC, pp.
1–8.
Handford, C., Davids, K., Bennett, S. and Button, C. (1997) Skill acquisition
in sport: some applications of an evolving practice ecology. Journal of Sports
Sciences, 15: 621–40.
Hughes, M. and Bartlett, R. (2002) The use of performance indicators in
performance analysis. Journal of Sports Sciences, 20: 739–54.
Hughes, M. and Franks, I. (2004) Notational analysis: a review of the
literature, in M. Hughes and I. Franks (eds) Notational Analysis of Sport, 2nd
edn. London: Routledge, pp. 59–106.
Hughes, M. and Franks, I. (2005) Analysis of passing sequences, shots and
goals in soccer. Journal of Sports Sciences, 23 (5): 509–14.
Hughes, M. and Wells, J. (2002) Analysis of penalties taken in shoot-outs.
International Journal of Performance Analysis in Sport, 2: 55–72.
Hughes, M., Robertson, K. and Nicholson, A. (1988) An analysis of the 1984
World Cup of Association Football, in T. Reilly, A. Lees, K. Davids and W.
Murphy (eds) Science and Football. London: E. and F. N. Spon, pp. 363–7.
Johnson, J. G. (2006) Cognitive modeling of decision making in sports.
Psychology of Sport and Exercise, 7 (6): 631–52.
Kauffman, S. (1993) The Origins of Order: Self-organization and Selection
in Evolution. New York: Oxford University Press.
Kerwin, D. J. and Bray, K. (2006) Measuring and modelling the goalkeeper's
diving envelope in a penalty kick, in E. F. Moritz and S. Haake (eds)
Engineering of Sport: Developments for Sports. New York: Springer Science
Business Media, 1: 321–6.
Kugler, P. and Turvey, M. (1987) Information, Natural Law, and the Self-
assembly of Rhythmic Movement. Hillsdale: Lawrence Erlbaum Associates.
Lames, M. and McGarry, T. (2007) On the search for reliable performance
indicators in game sports. International Journal of Performance Analysis in
Sport, 7 (1): 62–79.
Lee, D. N., Young, D. S., Reddish, P. E., Lough, S. and Clayton, T. M.
(1983) Visual timing in hitting an accelerating ball. Quarterly Journal of
Experimental Psychology, 35 (Pt 2): 333–46.
López-Botella, M. and Palao, J. M. (2007) Relationship between laterality of
foot strike and shot zone on penalty efficacy in specialist penalty takers.
International Journal of Performance Analysis in Sport, 7: 26–36.
McGarry, T. (2009) Applied and theoretical perspectives of performance
analysis in sport: scientific issues and challenges. International Journal of
Performance Analysis in Sport, 9: 128–40.
Morya, E., Bigatão, H., Lees, A. and Ranvaud, R. (2004) Evolving penalty
kick strategies: World Cup and club matches, 2000–2002. Journal of Sports
Sciences, 22: 512–13.
Nevill, A., Holder, R., and Watts, A. (2009) The changing shape of
“successful” professional footballers. Journal of Sports Sciences, 27 (5),
419–26.
Peper, C., Bootsma, R., Mestre, D. and Bakker, F. (1994) Catching balls:
how to get the hand to the right place at the right time. Journal of
Experimental Psychology: Human Perception and Performance, 20: 591–
612.
Reilly, T., Williams, A. M., Nevill, A. and Franks, A. (2000) A
multidisciplinary approach to talent identification in soccer. Journal of Sports
Sciences, 18 (9): 695–702.
Richardson, M., Marsh, K. and Baron, R. (2007) Judging and actualizing
intrapersonal and interpersonal affordances. Journal of Experimental
Psychology: Human Perception and Performance, 33 (4): 845–59.
Seifert, L. and Davids, K. (2012) Intentions, perceptions and actions
constrain functional inter- and intra-individual variability in the acquisition of
expertise in individual sports. Open Sports Sciences Journal, 5 (Suppl 1-M8):
68–75.
Stoffregen, T., Gorday, K., Sheng, Y. and Flynn, S. (1999) Perceiving
affordances for another person's actions. Journal of Experimental
Psychology: Human Perception and Performance, 25 (1): 120–36.
Travassos, B., Araújo, D., McGarry, T. and Vilar, L. (2011) Interpersonal
coordination and ball dynamics in futsal (indoor football). Human Movement
Science, 30: 1245–59.
Turvey, M. (1992) Affordances and prospective control: an outline of the
ontology. Ecological Psychology, 4 (3): 173–88.
Vilar, L., Araújo, D., Davids, K. and Button, C. (2012a) The role of
ecological dynamics in analysing performance in team sports. Sports
Medicine, 42 (1): 1–10.
Vilar, L., Araújo, D., Davids, K. and Travassos, B. (2012) Constraints on
competitive performance of attacker–defender dyads in team sports. Journal
of Sports Sciences, 30 (5): 459–69.
Warren, W. (1984) Perceiving affordances: visual guidance of stair climbing.
Journal of Experimental Psychology: Human Perception and Performance,
10 (5): 683–703.
Warren, W. (2006) The dynamics of perception and action. Psychological
Review, 113 (2): 358–89.
Warren, W. and Whang, S. (1987) Visual guidance of walking through
apertures: bodyscaled information for affordances. Journal of Experimental
Psychology: Human Perception and Performance, 13 (3): 371–83.
Watson, G., Brault, S., Kulpa, R., Bideau, B., Butterfield, J. and Craig, C.
(2011) Judging the ‘passability’ of dynamic gaps in a virtual rugby
environment. Human Movement Science, 30 (5): 942–56.
14 Talent development and
expertise in sport
Elissa Phillips, Keith Davids, Duarte
Araújo and Ian Renshaw
Complexity sciences have been used to study and explain the rich patterns
formed in complex systems such as animal collectives, weather systems, the
human brain and movements in team sports, where patterns emerge from
seemingly random component trajectories (Bak and Chialvo 2001; Kauffman
1993; Sumpter 2006). From this description, it is clear that an emerging
expert performance can be viewed as a complex system, composed of many
degrees of freedom on many system levels. The potential for interaction
between system components provides the platform for rich patterns of
behaviour to emerge as individuals interact with dynamically changing
environments. This new perspective reveals that compensatory adaptation in
performance achievement occurs as the result of system trade-offs between
specificity and diversity of behaviours (Edelman and Gally 2001). These
ideas are harmonious with a dynamical systems theoretical perspective on the
influence of interacting constraints. This overarching theoretical framework
proposes that expert levels of performance can be achieved in diverse ways
as individual performers attempt to satisfy the unique constraints on them
(Davids et al. 2003).
The role of neurobiological degeneracy in expertise acquisition
A brief overview of the strengths and weaknesses of theoretical ideas and
empirical methods used in environmental and genetic research on expertise in
sport suggests that neither specific approach provides enough explanatory
power to account for all the data on expertise in sport. The implicit basis of
the deliberate practice perspective is the adage ‘all healthy individuals are
created equal’. Analysis of the literature on genetic constraints on variability
of performance does not support this conclusion but this interpretation of the
literature should not be taken to imply that expert performance in sport is
biologically predetermined. There is clear evidence rejecting the idea that
single gene variants can predispose an athlete to superior performance
manifested in a specific domain, without clear and detailed consideration of
the performance context (e.g. a gene that is widespread in cricket fast
bowlers). Rather, the effects of interacting constraints on acquisition of
expertise in sport have been noted since, despite variations in genetic
structure, maximal heritability of particular traits includes strong
environmental components. Complex neurobiological systems are composed
of many interacting parts and levels, which self-organize under constraints
(Davids and Baker 2007; Frank et al. 2006; Schöllhorn 2003).
One of the main reasons for the failure to identify single gene variants
responsible for sport performance has been the inherent degeneracy of
athletes considered as neurobiological systems. Degeneracy in this context
refers to the ability of structurally different components to be coordinated
together to achieve the same behavioural goal (Edelman and Gally 2001).
Complex, neurobiological systems have been conceptualized as pleiotropic
and degenerate, with huge numbers of degrees of freedom and the distinct
ability to adapt to different task and environmental demands (Chow et al.
2008; Davids et al. 2007b). Pleiotropy concerns the multiple effects of
expression of phenotypes or behaviours from one constraint, such as
environment and genes, which may have phenotypic effects (Baker and
Davids 2006). Pleiotropy provides neurobiological systems with a variety of
alternate performance solutions (Davids et al. 2007a), while degeneracy
refers to the ability of structurally different components to be coordinated
together to achieve the same behavioural goal (Edelman and Gally 2001). At
the level of gene networks, degeneracy promotes evolutionary fitness by
ensuring that genetic diversity supports functional adaptation to variable
environments. Genes function in networks and single genes can produce
multiple effects, leading to multiple phenotypic expressions. This is why
gene expression is an inherently stochastic process (Kaerns et al. 2005).
The degenerate relationship between system components and system
output in developing experts is important because it implies that there are
many different pathways to achieving expert performance. Dynamically
varying performance environments interact with the inherent degenerate
nature of human movement systems, signalling a new view on the acquisition
of expertise in sport. This new perspective reveals that compensatory
adaptation in performance achievement occurs as the result of system trades-
off between specificity and diversity of behaviours (Edelman and Gally
2001). These ideas are harmonious with a dynamical systems theoretical
perspective on the influence of interacting constraints. This overarching
theoretical framework proposes that expert levels of performance can be
achieved in diverse ways as individual performers attempt to satisfy the
unique constraints on them (Davids et al. 2003). From this viewpoint,
expertise can be defined as the optimal satisfaction of unique, interacting
constraints on each individual in specific performance domains. Genetic
diversity may be responsible for a small part of training or performance
response differences between individuals and only when there is a favourable
interaction with important environmental constraints are performance
benefits observed. This description of key influences on athletic performance
has implications for considering the effects of time spent in practice in sport.
Given differences in genetic contributions, performance variations are more
likely to assert themselves under intensive practice regimes.
The characteristics of pleiotropy and degeneracy in athletes as complex,
neurobiological systems highlight the need for a multidimensional framework
(Davids and Baker 2007; Simonton 1999). Expertise attainment in a given
skill depends on many additional constraints outside the cognitive domain,
including but not limited to genetics, social and physical environment,
opportunity, encouragement and the effect of these variables on physical and
psychological traits (Wolstencroft 2002). Monodisciplinary approaches to the
acquisition of expertise, focusing on effects of genetic or environmental
constraints alone, fail to emulate the complementary nature of the
relationship between individual, task and environmental constraints (Abbott
et al. 2005; Beek et al. 2003; Davids and Baker 2007). Davids and Baker
(2007) lamented the absence of an explanatory framework to examine the
interactionist perspective of expertise development in sport and suggested the
use of a complex systems framework incorporating key ideas from dynamical
systems theory. As we note in the following sections of this chapter, this
theoretical approach provides a viable platform for explaining the dynamic
relationship between an individual's genetic disposition and the environment
and the acquisition of expertise in sport through variable pathways and
processes.
Constraints on acquiring expertise
In sports performance contexts, the expression of expertise is limited or
shaped by interacting constraints at many system levels. Although numerous
constraints might act on any given system, they have been classified into
organismic, task and environmental constraints (Newell 1986). The concept
of constraints was proposed by Newell (1986) as boundaries or qualities that
constrain the interactions of system components. Constraints are the
numerous variables that form each individual expert's developmental
trajectory. These constraints include individual characteristics such as
experience, learning, development, morphology and genes which interact to
shape performance and the acquisition of expertise in sport (Davids et al.
2008).
It is important to identify the range of constraints on the acquisition of
expertise, requiring a multidisciplinary framework of analysis. Given that an
individual is born with distinguishing physical characteristics (with a degree
of genetic influence), expertise research is concerned with how
environmental constraints affect the development of skill and the expression
of genotypes.
Performance emerges from the intrinsic dynamics of experts; these
preferred coordination tendencies come from the interaction of
environmental, task and organismic constraints (including development,
experience, genes and learning of each performer; Kelso 1991).
Understanding the nature of each individual's intrinsic dynamics is central to
understanding how expert performance develops in sport. As individuals
progress towards a state of expertise and explore different performance
solutions, their intrinsic dynamics will alter and diversify. The learner's
ability to adapt to constraints, with dynamic performance solutions, will
affect their rate of learning. Learning to perform a new task with dynamics
that are similar to a previously learned task can harness an existing landscape
of intrinsic dynamics in learners and may provide a more rapid transition to a
new required movement pattern. This is the basis of talent transfer. In
contrast, acquiring task dynamics which are dissimilar to those of a
previously learned task (e.g. tennis and squash movement pattern dynamics)
may lead to a longer process of learning because the specific learner's
intrinsic dynamics may need to be significantly re-shaped (Renshaw et al.
2009).
These ideas in dynamical systems theory have important implications for
understanding the development and maintenance of expert performance.
Importantly, expertise research on developmental histories, has found both
similarities and differences in expert developmental pathways, suggesting
that experts can adopt different pathways and strategies as they acquire
expertise in athletic performance (Durand-Bush and Salmela 2002; Gould et
al. 2002; Holt and Dunn 2004). How the intrinsic dynamics of developing
experts are continually shaped by genetic and environmental constraints
needs to be understood. The effects of environmental constraints on
phenotypic gene expression suggests that athletes with what may be
perceived as less favourable genetic dispositions may still achieve expert
levels of performance given an appropriate skill acquisition environment
(Baker and Horton 2004). Alternatively, genetically gifted athletes may fail to
achieve expert status without a rich environment for acquiring and practising
skills. Rich learning environments do not necessarily imply a need for
purpose-built state of the art training facilities. In fact, there is evidence of
sporting champions emerging from the most basic learning and performance
environments.
To summarize so far, it seems unlikely that a singular common optimal
pathway to performance expertise exists because of the degenerate
neurobiological system characterizing each individual performer and the
effect of interactions between environmental and personal constraints on the
intrinsic dynamics of each learner. Expert performers are able to generate
different types of functional performance solutions, depending on differences
in their intrinsic dynamics (e.g. varying ball speed or type and line and/or
length in cricket fast bowling). For example, in cricket fast bowling Pyne et
al. (2006) examined the relationship between junior and senior high-
performance athletes' anthropometric and strength characteristics and
bowling speed. They found differences in the variables and strength of
correlation of predictors of peak ball speed between age groups and
suggested growth and biological maturation largely accounted for greater
peak ball speed in seniors. A more comprehensive examination including
technical/coordination variants under a variety of tasks' demands such as
different delivery types would be needed to gain greater insight into the
individual dynamics and movement solutions of such a group (e.g. varying
ball speed or type and line and/or length in cricket fast bowling).
The role of metastability in acquiring expertise
Conceptualized as dynamical neurobiological systems, emerging expert
athletes exhibit complexity and metastability, owing to the potential for
interaction between their subsystems. The inherent degeneracy of
neurobiological systems and the nonlinear interactions between system
components enable the existence of more than one stable pathway to
expertise. This characteristic of dynamical systems is termed multistability
(Araújo et al. 2006; Edelman and Gally 2001; Kelso 1995). Multistable
systems can also exhibit metastability when modes of behaviour or
performance are weakly stable or weakly unstable (i.e. close to an instability
point) and it manifests itself in the switching between modes of organization
or structure (Fringelkurts and Fringelkurts 2004; Kelso 2008). The
characteristic of metastability has important implications for our
understanding of emerging expertise. The rich interactions that can occur
between personal, task and environmental constraints can create a fertile
‘metastable’ region of the performance landscape in which small changes in
one constraint can lead to large transitions in system organization. For
example, small alterations in experience, practice and/or development,
combined with small variations in genetic structure, might induce continuous
and abrupt changes (i.e. bifurcations)1 in the set of possible behaviours
available to a developing athlete as these constraints are satisfied. In this way,
the whole dynamical landscape of expertise can suddenly change with small
variations in responses to constraints impinging on the developing athlete.
For example, the emergence of novel actions or responses to performance
problems can influence future system behaviours by changing the probability
of occurrence of other potentially functional behaviours. These ideas also
need to be considered in light of suggestions that critical periods exist for
acquisition of skills in developing children. Critical periods have been
identified as brief windows of time and space during which a complex
system's organisation is most open to modification from external and internal
constraints. It has been argued that motor learning can be enhanced when
developing athletes are located within these critical periods (Anderson and
Ward 2002). These ideas should be interpreted in relation to data tracking the
performance trajectories of skilled and less-skilled developing footballers in
the UK (Ward et al. 2007). These findings suggested that a greater proportion
of performance variability in these groups could be explained by the
acquisition of skill, in the absence of physical and maturational advantages.
Harnessing the evolutionary strategy of co-adaptation to incur
performance paradigm shifts in sport
The challenge for coaches and sport scientists is to identify when individual
athletes enter metastable regions or critical periods while developing their
expertise levels so that performance paradigm shifts in expertise and skill
acquisition might be triggered, as we exemplify in this section. By ensuring
exposure to metastable regions of the performance landscape during
development, experts can discover new modes of behaviour to satisfy
interacting perceptual, affective and task constraints. These new modes of
performance are likely to emerge as novel solutions to performance problems
as developing athletes co-adapt their responses to challenging constraints
imposed by opponents, coaches or performance environments. For example,
after the famous ‘turn’ away from defenders, performed by Johan Cruyff at
the 1974 Football World Cup, millions of young players began to imitate and
practise the move in training, as coaches directed their search for novel ways
to avoid defensive marking.
When self-organizing systems, such as the developing expert, are poised in
a state near this region, different types of behaviour (including development
pathways) can emerge depending on changes in relevant system control
parameters. For this reason, a system poised in the metastable region may
enter a critical state which is ready for change. Such states of criticality in
complex systems are recognised as ‘a global state that is acutely context
sensitive’ (Van Orden et al. 2003, p. 332). Van Orden et al. (2003) showed
how ‘criticality’ emerges from a balance between constraints on complex
systems. The implication is that a small difference in circumstances can lead
to transitions along the expertise pathway for the system. This raises some
important issues about attempting to identify future champions. It suggests
that sometimes apparently minor changes to individual, task or environmental
constraints can result in giant leaps or paradigm shifts in performance. For
example, exposing athletes to a minor change in technique, a change of
coach, an opportunity to play at a higher level or interaction with a previous
champion could be the turning point in a developmental pathway. From a
dynamical systems theoretical perspective, these turning points and critical
states are likely to be unique to each individual athlete and it may be difficult
to identify patterns in the developmental histories of expert performers. This
is an area that future talent development research needs to explore in more
detail.
In evolutionary theory, the influential role of metastability in organismic
change has been observed. Kauffman's (1993) modelling of evolutionary
processes from the perspective of spontaneous self-organizing system
dynamics has provided valuable insights for understanding how developing
experts, viewed as complex, open systems, may undergo transitions along the
developmental pathway. According to Kauffmann (1993), evolution occurs
as different systems pressurise each other, co-adapting function, structure and
organization in the process. In this process, rich and varied patterns of
behaviour can emerge as organisms subtly co-adapt their actions in response
to a range of specific interacting constraints to achieve specific system
outcomes or goals.
Models of evolutionary processes arising from spontaneous self-organizing
system dynamics provide valuable insights for understanding the acquisition
of expertise (Kauffman 1995). In these models, phase transitions in system
behaviour are most prevalent in metastable regions where co-evolving system
components compete to modify a system's landscape of behaviours. The
coevolving adaption of system components within critical regions is typically
an emergent process because of the evolved coupling between system
components. In developing experts, new behaviours and performance levels
can emerge out of fluctuations created by interactions between
interdependent constituents of the whole performance system. Random
interactions between system components can alter into more organized forms
of interactions as one key system parameter (a control parameter) changes in
value. Fluctuations in a developing system's organization can create
instabilities which provide a useful platform to assist transitions in expertise
with practice and experience. Near this critical state, interactions between
system components can become correlated, in a domino effect, capturing
global system interactions and leading to a sudden reduction from multiple
options to one (a sudden collapse in the critical state). Criticality provides a
fertile region for a functional mix of creativity and constraint to emerge in
dynamic learning environments. It affords new opportunities for behaviours
to emerge which can fit newly arising states of organization during
development and skill acquisition. As each individual component in a co-
adapting system changes over time, the whole system can move inexorably in
the cocreated landscape towards a region of self-organizing criticality poised
for a transition (Bak and Chialvo 2001). In this respect, variability in a
developing athlete can be a useful process to exploit in order to facilitate the
transition to a new state of expertise in sport.
To summarize so far, co-adaptation is an evolutionary strategy that has
implications for the way that constraints can influence the process of talent
development by forcing the developing expert to find new functional
performance solutions. It may represent a useful expertise development
strategy for practitioners, as different performers and/or technologies attempt
to pressurise individuals to seek unique performance solutions. Furthermore,
the process occurs naturally as a sport develops (i.e. through technique
changes, rule changes or equipment change or as athletes pose each other
new performance problems). New solutions to performance problems emerge
as talented individuals learn how to assemble creative movement solutions
during practice.
The developmental strategy of seeking to move sport performers to a
metastable region of their performance landscape might provide a platform to
help them create new, more adaptive, modes of action with respect to
emerging task and goal constraints of the performance environment. For
example, in high jump, early athletes used a scissor technique to jump with
the inside leg leading, followed closely by the other leg in a scissoring
motion. The scissors action then progressed to the western roll and the
Fosbury flop, which represents a good example of the metastable region in
the sport of high jumping. The Fosbury flop was a performance solution that
emerged coinciding with the use of raised, softer landing areas in modern
times. Directing the body over the bar head and shoulders first and sliding
over on one's back would have likely led to injury without modern landing
mats.
Changes to expert movement skill performance in other sports have
resulted from advances in equipment design, with these changes interacting
with environmental constraints of a physical nature. For example, an
interesting question concerns why certain types of expertise can be traced to
specific cultural groups. In a well-known performance paradigm shift in the
sport of football, the first players to ‘bend’ the trajectory of balls when
shooting (adding swerve and dip to flight) were South Americans in the
1960s, rather than European players. This was because, prior to the 1970s,
footballs had a leather skin that picked up moisture as a game progressed, so
during winter games played in the northern hemisphere, the ball increased in
mass to almost double its original value. In the generally drier climates of
South America the ball did not pick up as much moisture and mass, so
aerodynamic forces had more influence on the ‘lighter’ ball, enabling skilled
players to acquire expertise in bending and curving ball trajectories. With the
advent of waterproof balls in the 1970s, Northern European experts were able
to recreate the same effects as their South American counterparts (James
2008). James (2008) also noted recent research observing that the
aerodynamic drag of non-spinning footballs has fallen by as much as 30%
over the last 36 years, leading to the ball being able to travel greater
distances. These changes in task constraints (e.g. physical characteristics of
equipment) have influenced the tactical strategies available to teams and
consequently re-emphasised the acquisition of specific shooting and long
passing skills in players. In other sports, performance paradigm shifts have
been created by:
• equipment changes (e.g. the change from bamboo to fibreglass pole
vaults leading to the world record increasing from around 4.5 metres to
6.14 metres in a vault by Sergey Bubka in 1993);
• changes to playing surfaces (such as international hockey matches on
artificial turf instead of grass), which may explain the ebb and flow of
current performance standings and international rankings between
countries;
• rule changes, such as the turn-over law in rugby union or the distance of
the three-point line in the Olympics versus the US National Basketball
Association (NBA). It was argued that the change in this distance (in
NBA it is 7.23 metres while at the Olympics it was 6.25 metres) resulted
in the allstar American team struggling to score their usual number of
three-pointers in the early rounds of the 2008 Olympic Games
competition (Thamel 2008).
In elite sport, the drive for success means that performers are being
challenged constantly to co-adapt to succeed. This is particularly true of
professional televized sports where opportunities to analyze strengths and
weaknesses of opponents are enhanced by available television and videotape
footage. Through co-adaptation, players need to add new skills or strategies
in the off-season in order to continue to challenge opponents with new
problems (Cowdrey 1974). Essentially, players have to constantly reinvent
themselves or demonstrate an ability to adapt to the strategies developed by
opponents to counteract previously observed strengths. A good example of
this adaptation was observed in 2008 in cricket when the use of ‘switch
hitting’ by Kevin Pietersen (an English batsman) was first formalized. He
developed a strategy of changing his stance (from his typically right-handed
position) as a bowler in the process of delivering a ball, by ‘jumping’ into a
left-handed stance in order to overcome the restrictive field placements of
opponents in one-day cricket.
Finally, co-adaptation is a process which occurs intra-individually as well
as inter-individually, as different subsystems within an individual athlete self-
organize in relation to each other. For example, an athlete's cognitive
subsystem might constrain their action system and vice versa in a co-adaptive
manner. For example, after World War II, in the sport of athletics the sub-
four-minute mile was viewed by many as an impossible dream. However,
once Roger Bannister broke the record in 1954, it was almost immediately
broken again by John Landy, who improved his personal best from 4.02.1
minutes to 3.57.9 minutes. Interestingly, Landy had been the favourite to
become the first athlete to break the four-minute barrier, having run 4.02 six
times in the 15-month period prior to Bannister breaking the record. Landy
suggested that the first time he ran 4.02 was unexpected and was performed
without any pressure. He remarked: ‘Once I was aware of what I could do, I
was always under pressure. I was under pressure externally and internally and
I think that made a lot of difference’ (North 2004).
The idea of subsystems co-adapting to constraints imposed by other
subsystems of the body is influential in a dynamic systems analysis of motor
development across the lifespan (Thelen 1995). For example, during
children's motor development, dynamic systems analysts emphasize the idea
that specific behaviours may not have yet appeared in developing children,
because specific subsystems act as system ‘rate limiters’ and are ‘lying in
wait’ for another critical subsystem to reach a critical level (e.g. changes in
the muscle-to-fat ratio in infants to enable upright postural control). In the
performance of specific sports, various subsystems could be critical to the
performance development in athletes, such as strength, speed, mobility or
game understanding as a result of numerous experiences. John Landy's
experience suggests that, in some athletes, a major constraint on performance
might be imposed by the cognitive subsystem acting as a rate limiter in the
form of psychological effects of prior expectations or motivation on
performance.
The role of movement variability in successful co-adaptation
From a dynamical systems perspective, expertise acquisition can be
construed as a noisy, nonlinear process, in which, counterintuitively, system
variability can be functional. In this process, it is important to note that
subsystems can change over differing timescales. The concept of rate limiters
which shape the trajectory of expertise acquisition in individuals places into
sharp focus the need for theoretical frameworks to capture differences in rates
of growth and development in young athletes. An important point to note in
the development of talent is the different timescales associated with various
subsystems of each performer. A developing athlete viewed as a complex
system can transit to a new state of organization (e.g. expertise) when the
behaviour of important subsystems suddenly become mutually entrained.
Through this process of entrainment, like co-adapting biological organisms
seeking to optimize their relative ‘fitness’ on an evolutionary landscape,
important subsystems can become dependent on what is occurring in other
key subsystems.
A phase transition in the expertise levels of athletes might therefore be
shaped by a change in the relationship between an athlete's subsystems. This
change may emerge as a result of development, experience and
practice/training and might push the whole system to a state of non-
equilibrium (i.e. the region of metastability). In nonlinear dynamics, if a
developing system is driven to the edge of its current basin of attraction, the
probability of a new state of organization emerging (e.g. a new level of
expertise) increases, owing a breaking of symmetry in the initial system
structure. This occurrence exemplifies the process of ‘selfconstruction’ that
Kauffmann (1993) defined in systems that evolve over time. The interaction
between critical subsystems creates a flow of information that drives the
athletic system to a desegregation of its components. Expert skill acquisition
can be promoted by exploiting metastability tendencies in athletes, by
creating diverse learning environments, late specialization (e.g. from
approximately 13–15 years of age; Côté et al. 2007) into sport and discovery
learning processes. The process of ‘self-construction’ may be harnessed by
coaches and athletes to propel learners along different expertise pathways by
incurring performance paradigm shifts.
The functional role of variability in expert performance may have an
important role to play in expert skill acquisition in sports. For example,
Morriss et al. (1997) examined the variability of men's javelin throwing in the
1995 world athletics championships. They found that the finalist from the
event exhibited a range of movement solutions which was reliant on different
upper-body contributions to javelin release speed. The silver medallist
demonstrated a predominance of linear shoulder movements (shoulder
horizontal flexion and extension), whereas the winner of the event used
shoulder rotation movements combined with elbow flexion (which had a
velocity of 18% greater than any competitor). The other finalists used
variations of these movement patterns, supporting the notion of degenerate
movement solutions reliant on the self-organization process (Bartlett and
Robins 2008) An interesting aside here is to examine how changes in the
specification of the javelin in 1986 led to changes in the optimal throwing
technique for javelin throwers. As a result of a desire to develop javelins that
landed ‘point first’ on every throw, the projectile's centre of gravity was
moved forward by four centimetres. According to Bartlett and Best (1988),
the change resulted in the optimal technique being a ‘higher release angle’.
They also suggested that the changes would benefit more powerful athletes
who could generate high release speeds. Sudden inter-athlete changes in
performance levels are probably explained by changes in interacting personal
and task constraints.
Methodological implications in expertise research: research
and practice
What are the methodological implications for adopting a multi-dimensional
theoretical approach to the acquisition of expertise and talent development?
One major implication is that a range of methods is needed to examine these
processes. While qualitative expertise research has identified unique
pathways of development as athletes progress through transitions associated
with different developmental stages (Abbott and Collins 2002; Bloom 1985),
this concept has not been examined comprehensively for its implications in
relation to talent development. Unique pathways to expertise development
imply the existence of a range of functional performance solutions as a result
of unique interactions of individual, environmental and task constraints
during expert skill acquisition. The specific intrinsic dynamics of learners is
shaped by previous experiences, individual rate limiters and environmental
factors, suggesting that the notion of a general theory of expertise is
unrealistic.
The methodological implication of these ideas in ecological psychology is
that practice environments should be high in representative design. Egon
Brunswik, the ecological psychologist, originally coined the term
representative design (Brunswik 1955). Brunswik defended the view that
cues or perceptual variables should be sampled from an organism's
environment so as to be representative of the environmental stimulation to
which it had adapted and which formed the focus of an experimenter's
generalization. It has been noted that the term could refer equally to the
composition of practice task constraints so that they represent the
performance environment which learners are faced with (Araújo et al. 2007).
The acquisition of expertise is predicated on representative design in sports
training programmes and refers to the accurate sampling of the information
present in performance environments to be reflected in practice task
constraints. For example, the use of bowlers rather than ball machines in
cricket batting training means that the batter is able to attune to specifying
information from the bowler's actions rather than the less-specifying
information available from the bowling machine during practice (Renshaw et
al. 2007). Obviously, only the former specifying sources of information for
regulating action are present in the performance context, which has clear
implications for designing representative practice tasks. Representative
practice task design ensures that developing individuals can gain expertise in
picking up affordances for action in performance environments. However, it
should be noted that the training activities of athletes in performance and
learning environments, designed by coaches, can lead to substantial changes
in an athlete's abilities to perceive and act. This pedagogical activity results in
a constant interaction of abilities and affordances over time, leading to
dynamic relationships between the individual and practice and performance
environments, with new self-organizing behaviours emerging (Chemero
2008). Developing experts will become sensitive selectively to the
performance environment (their niche) relative to the things that they can do.
Therefore, the design of this environment is critical in influencing the
development of the performer's ability to perceive and act. Affordance and
abilities causally interact, as changes in abilities will lead to changes in the
layout of available affordances, which will change the way abilities are
exercised in real-time action (Chemero 2008).
The critical decision for coaches when working with talented young
performers is when to expose them to practice and competitive environments
that are representative of a new performance level. For example, it is clearly
worthwhile for talented young performers to spend a percentage of training
time working with senior players to gain exposure to the constraints on
performance imposed by highly skilled competitors. However, it would still
be prudent for these developing athletes to continue to play against peers in
their own age group to maintain levels of confidence and motivation (Côté et
al. 2007). Clearly, deciding when to expose the developing performer to this
higher performance level is a complex decision for the practice of coaching,
since it creates instability to allow the performer opportunities to adapt their
skills but may threaten his or her selfconfidence (Abbott et al. 2005).
Nonlinear dynamics, expertise in sport and talent development
pathways
Since expertise has a multi-dimensional basis with a range of constraints
impinging on its acquisition, the implication is that talent development
programmes need to have a multi-dimensional framework. There are a
number of multi-dimensional models of talent development (Abbott et al.
2005; Simonton 1999), which have conceptualized talent as a nonlinear
process, predicting that there is a range of developmental trajectories and
timescales that eventually result in the achievement of sporting expertise.
These models have criticized traditional talent identification programmes for
overemphasising the role of early identification and the role of physical and
structural variables (such as physical dimensions, weight and height) and not
accounting for variations in maturation rates of adolescents owing to the
presence of performance rate limiters (Abbott et al. 2005).
Conceptualizing expertise acquisition as a complex system has numerous
implications for talent development programmes. From this viewpoint, the
aim of talent development is to aid individuals in gaining the expertise
needed to satisfy the unique constraints impinging on them in specific
performance environments. To achieve this aim, talent development
programmes should aspire to promote the transition of individuals into a
metastable region of the perceptual–motor landscape. This can be achieved
by identifying key constraints on each individual and facilitating
development by manipulating these constraints to encourage exploration of
coordination. This process in expertise acquisition requires a comprehensive
understanding of the intrinsic dynamics of each individual. Talent
development programmes can harness these nonlinearities by developing
strategies to induce phase transitions in individual performers. This aim
might be best achieved by understanding how to force individuals into the
metastable region of the perceptual–motor landscape of practice where a
strategy of co-adaptation can underpin the emergence of creative behaviours
(typical of the profile of expert performers in line with the findings of
Durand-Bush and Salmela (2002). The implication of this approach for
coaches is to incorporate tasks that promote adaptability and creativity in
performers. In cricket, this aim could be achieved by carefully designing
games that require batters to explore the boundaries of their skill set. For
example, an approach to promote co-adaptation by players could involve
setting up a practice task where the batter and bowler are on the same team
and the field is split into eight zones but only six fielders are provided. The
aim for the bowler is to bowl the ball so that the batter can hit into these
‘free’ zones. The fielding captain can vary these empty regions to challenge
the batter to hit to areas where they may have a weaknesses. Despite the
counterintuitive decision to have the batter and the bowler on the same team,
this task requires bowlers to learn to bowl to force the batter to hit where they
want them to, an essential skill at higher levels of performance. The batter is
forced to be adaptable, since, even if the delivery is not ‘perfect’ for the
required shot, they have to improvise and assemble a functional motor
solution. The advantages for the fielding captain are the need to work out the
strengths and weaknesses of opponents. This example highlights how
nonlinear pedagogy can be used to great effect to force co-adaptive
behaviours in learners as long as the design of game tasks are based on
grounded principles of the game (e.g. Renshaw et al. 2007). It has been
argued that traditional models need to be adjusted to consider the different
rates of development of potentially talented athletes (Abbott et al. 2005).
Subsystem behaviours shape an individual athlete's intrinsic dynamics.
Because of variations in athletes' intrinsic dynamics, individual rates of skill
acquisition are likely to progress at different time scales (Liu et al. 2006).
Talent development models need take into account the different rates of
learning and growth and maturation processes experienced by individuals on
their pathway to expertise (not least it must take into account the relative age
effect to maximize the available talent; Cobley et al. 2009). These different
rates of learning can be influenced by key constraints which act as ‘rate
limiters’ causing systems to find new coordination solutions (Handford et al.
1997). Rate limiters can be defined as system controllers; they are
components or systems which limit the development of an individual in sport
for example (Thelen and Smith 1994). In this instance, children's strength in
key muscle groups may act as a rate limiter that inhibits them from
demonstrating the skills that they have acquired already through practice and
experience in sport. The complexity of neurobiological systems and
associated rate limiters requires future research to adopt multidisciplinary
methods to capture the nature of expert skill acquisition in sport.
Some current models of talent development are harmonious with these
theoretical ideas, although their tenets are not necessarily predicated
conceptually on these insights. For example, Simonton (1999) proposed that
talent emerges from multidisciplinary, multiplicative and dynamic processes
and is likely to operate as an intricate system beyond the scope of the polar
nature–nurture debate. He pioneered mathematical equations to model the
potential components that contribute to talent development. These
components were weighted by relevance and included reference to genetic
dispositions (e.g. height or endurance capacity), environmental (e.g. social
and familial support) and developmental constraints. Subsequently the model
was described as emergenic and epigenetic, comprising components that
interact and change with time (Simonton 1999). The emergenic aspect
proposes that potential talent consists of multiple components, including all
physical, physiological, cognitive and dispositional traits that facilitate the
manifestation of superior expertise in a specific domain. Beyond individual
differences, Simonton's (1999) model captures the dynamics of epigenetics.
Epigenetics is seen in the diverse components that make up talent that slowly
appear and differentiate over time in an individual and ultimately depend on
underlying neurological, muscular, cultural, skeletal, social, psychological,
physiological and environmental variables (Obler and Fein 1988). However,
they emerge gradually during the course of long-term interactions between
the internally developing organism and appropriate environmental constraints
(Simonton 1999). This system is complicated, as it includes the evolutionary
interaction of components, therefore any examination of talent development
that uses this model needs to be holistic, impartial and sophisticated
(Simonton 1999). Although this mathematical model is a useful starting point
for sport scientists since it attempts to operationalize key concepts such as
multidisciplinary talent development practices and multiple developmental
trajectories towards potential expertise (Abbott et al. 2005), it lacks
theoretical power as it is not conceptualized within a theoretical framework.
This weakness could be mediated by including dynamical systems theory as a
viable rationale for talent development as an emergenic and epigenetic
process. Furthermore, on a practical level, adopting Simonton's (1999) model
may be extremely difficult as its efficacy is predicated on identifying all
components that contribute to expertise in any one specific sporting domain.
Identification of every component is essential because as the model is said to
be multiplicative, a score of zero for any one factor means that expertise
cannot be achieved.
Vaeyens et al. (2008) provided an insightful model of talent identification
capturing the dynamic nature of talent and its development, focusing on
potential for development and inclusion rather than early identification. Such
a model can be strengthened with a dynamical system theoretical approach
using concepts discussed previously, including metastability, nonlinearity,
co-adaptation and degeneracy. The interaction of constraints within
neurobiological systems means the intrinsic dynamics of each learner will
result in unique pathways to expert performance in sport. Degeneracy of
individuals will result in individualized performance solutions as interacting
constraints are optimally satisfied when striving for acquisition of expertise
in sport. In support of this view, qualitative sports expertise research has
found differing developmental factors to be important to elite athletes. For
example, Baker et al. (2005) examined the early sport involvement of elite
ultra-triathletes and found early sport specialization was not a requirement for
acquiring expertise. This observation has been supported by research into
elite swimmers, who found early specialization was related to less time on the
national team and an earlier end to their careers compared with swimmers
who specialized later (Barynina and Vaitsekhovskii 1992). Although, the idea
of multilateral development has many strengths and early specialization many
potential pitfalls, early specialization may be an appropriate strategy for some
potential experts. For example, it could be argued that early specialization is
necessary to reach ‘legend’ status in a sport, given that this was the pathway
of a number of individuals such as Tiger Woods, Sachin Tendulkar and
Venus and Serena Williams. These different pathways would depend on the
individual and also the chosen sport. Sports which emphasize physiological
constraints rather than technical constraints require very different expertise
sets, for example (e.g. ultra-marathon running compared with clay-pigeon
shooting).
The notion of optimal movement solutions for a specific performance task,
generalizable to all sport performers, is inappropriate, given the complex and
degenerate nature of neurobiological systems (Glazier and Davids 2009).
Characteristics such as complexity, pleiotropy and degeneracy make it
challenging to identify a common optimal movement solution for all
performers, requiring a remodelling of movement behaviour in dynamic
performance contexts. Dynamical systems researchers have suggested that
common optimal movement patterns probably do not exist and that
variability is an intrinsic feature of skilled motor performance, providing
flexibility to adapt performance in dynamic environments (Glazier et al.
2003). As Sutton (2008) identified, any specific skilled action is based on
‘flexible intelligent action in real time that requires attention to the
continuous coupling of perception and action, and the mutually modulatory
dynamics operating between brain, body and world’ (p.770).
These findings provide some challenges for talent identification
programmes that emphasize physical and test performance measures as a
major criterion for referencing expertise in sport (see the earlier section on
‘Results of research with quantitative methodologies'). Abbott et al. (2005)
noted several major problems associated with identifying potential athletes
solely on current physical and performance measures. Underpinning these
concerns was the nonlinear process of development and the difficulty of
predicting changes in genetically and environmentally driven variables,
which were often unstable. Typically, identification of talent has focused on
discrete performance measures with limited task representativeness. As such,
these measures have revealed limited information about an athlete's potential
and adaptability in the competitive environment (Morris 2000). Abbott et al.
(2005) suggested that talent identification processes should be focused on the
future performance capacity of athletes, rather than current performance
measures and highlighted the importance of future research on the factors that
underpin successful development towards one's potential (Simonton 1999).
However, assessing future performance capacity must take into account the
current intrinsic dynamics of each individual in order to identify the current
relative position on the developmental pathway. Predicting ultimate
performance capacity becomes an even more complex issue when one tries to
predict what a sport will look like in five to ten years' time, as technological
advances are continually changing the nature of sports. Elite-level sport is
moving fast and is evolving continually, owing to rule changes, new training
methods and performance strategies. What is clear from this discussion is the
need for talent identification and development programmes to provide
flexible learning environments that produce performers who have the
capacity to adapt to constant change, within a context of expertise.
Conclusion
Complex systems theory is an appropriate functional framework for expert
performance research because it can address questions that other frameworks
do not have the language and tools to pose. This chapter has discussed
several concepts with important implications for expertise and talent
development researchers, including the concepts of self-organization under
constraints; emergence; metastability, creativity; and degeneracy, stabilities
and instabilities over different timescales. These ideas on the dynamic
interactions between genes and environment and the functional role of
degeneracy in the human nervous system require a multidisciplinary
theoretical perspective to explain the acquisition of expertise in sport. Within
this overarching theoretical framework, it has been argued that the same
performance outcomes can be achieved in diverse ways as individual
performers attempt to satisfy the unique constraints on them (Davids et al.
2003). Genetic diversity may be responsible for a small part of training or
performance response differences between individuals and only when there is
a favourable interaction with important environmental constraints are
performance benefits observed. Phenotypic expression of behaviour might be
best understood at the level of individual interactions with key environmental
and task constraints. This is particularly relevant for researchers considering
effects of time spent in deliberate practice in sport, since there is often little
mention of genetic constraints on the acquisition of expertise. Given
differences in genetic contributions, performance variations are more likely to
assert themselves under intensive practice regimens.
The acquisition of expertise is domain specific and is about adaptation to
the environment through satisfying unique constraints which impinge on each
developing expert. Expertise acquisition emphasizes the changing nature of
the performer–environment relationship through development and gaining
experience through training, practice, coaching and competing. A
comprehensive examination of expertise involves identifying the intrinsic
dynamics of each individual and the specific rate limiters and constraints
which shape their behaviour. Each individual athlete comes to a performance
context with a particular set of intrinsic dynamics that has already been
shaped by genes, development and early experiences. Individualized
pathways to expert performance are expected because of the uniqueness of
these dynamics constraints.
Note
1 Abrupt changes in the organization of a system are called bifurcations or
branchings, since typically they signify changes in the number of possible
action modes available.
References
Abbott, A., Button, C., Pepping, G.-J. and Collins, D. (2005) Unnatural
selection: talent identification and development in sport. Nonlinear
Dynamics, Psychology, and Life Sciences, 9 (1): 61–88.
Abbott, A. and Collins, D. (2002) A theoretical and empirical analysis of a
‘state of the art’ talent identification model. High Ability Studies, 13 (2): 157–
78.
Anderson, G. S. and Ward, R. (2002) Classifying children for sports
participation based upon anthropometric measurement. European Journal of
Sport Science, 2 (3) 11: 1–13.
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7 (6): 653–76.
Araujo, D., Davids, K. and Passos, P. (2007) Ecological validity,
representative design, and correspondence between experimental task
constraints and behavioral setting: comment on Rogers, Kadar, and Costall
(2005) Ecological Psychology, 19 (1): 69–78.
Bak, P. and Chialvo, D. R. (2001) Adaptive learning by extremal dynamics
and negative feedback. Physical Review E, 63 (3): 031912; doi:
10.1080/17461390200072301
Baker, J. and Davids, K. (2006) Genetic and environmental constraints on
variability in sport performance, in K. Davids, S. J. Bennett and K. Newell
(eds) Movement System Variability. Champaign, IL: Human Kinetics, pp.
109–29.
Baker, J. and Horton, S. (2004) A review of primary and secondary
influences on sport expertise. High Ability Studies, 15 (2): 211–28.
Baker, J., Côté, J. and Deakin, J. (2005) Expertise in ultra-endurance
triathletes early sport involvement, training structure, and the theory of
deliberate practice. Journal of Applied Sport Psychology, 17 (1): 64–78.
Bartlett, R. and Robins, M. (2008) Biomechanics of throwing, in Y. Hong
and R. Bartlett (eds) Handbook of Biomechanics and Human Movement
Science. New York: Routledge, pp. 285–95.
Bartlett, R. M. and Best, R. J. (1988) The biomechanics of javelin throwing; a
review. Journal of Sports Sciences, 6: 1–38.
Barynina, I. I. and Vaitsekhovskii, S. M. (1992) The aftermath of early sports
specialization for highly qualified swimmers. Fitness and Sports Review
International, 27 (4): 132–3.
Beek, P. J., Jacobs, D. M., Daffertshofer, A. and Huys, R. (2003) Expert
performance in sport: views from the joint perspectives of ecological
psychology and dynamical systems theory, in J. L. Starkes and K. A.
Ericsson (eds) Expert Performance in Sport. Champaign, IL: Human
Kinetics, pp. 321–44.
Bloom, B. S. (1985) Developing Talent in Young People. New York:
Ballantine.
Brunswik, E. (1955) Representative design and probabilistic theory in a
functional psychology. Psychological Review, 62 (3): 193–217.
Chemero, A. (2008) Self-organization, writ large. Ecological Psychology, 20:
257–69.
Chow, J. Y., Davids, K., Button, C. and Koh, M. (2008) Coordination
changes in a discrete multi-articular action as a function of practice. Acta
Psychologica, 127 (1): 163–76.
Cobley, S., Baker, J., Wattie, N. and McKenna, J. (2009) Annual age-
grouping and athlete development: a meta-analytical review of relative age
effects in sport. Sports Medicine, 39 (3): 235–56.
Côté, J., Baker, J. and Abernethy, B. (2007) Practice and play and the
development of sport expertise, in G. Tenenabum and R. Eklund (eds)
Handbook of Sport Psychology, 3rd edn. 10. Degenerate Brains,
Indeterminate Behavior, and Representative Tasks: Implications for
Experimental Design in Sport Psychology Research. New Jersey: Wiley, pp.
184–202.
Cowdrey, M. C. (1974) Tackle Cricket. London: Stanley Paul.
Davids, K. and Baker, J. (2007) Genes, environment and sport performance:
why the nature–nurture dualism is no longer relevant. Sports Medicine, 37
(11): 961–80.
Davids, K., Araújo, D., Button, C. and Renshaw, I. (2007a) Degenerate
brains, indeterminate behavior and representative tasks: implications for
experimental design in sport psychology research, in G. Tenenbaum and R.
Eklund (eds) Handbook of Sport Psychology, 3rd edn. 10. Degenerate
Brains, Indeterminate Behavior, and Representative Tasks: Implications for
Experimental Design in Sport Psychology Research. New York: Wiley, pp.
224–44.
Davids, K., Button, C. and Bennett, S. J. (2007b) Coordination and control of
movement in sport: An ecological approach. Champaign, IL: Human
Kinetics.
Davids, K., Button, C. and Bennett, S. (2008) Dynamics of Skill Acquisition:
A Constraints-Led Approach. Champaign, IL: Human Kinetics.
Davids, K., Glazier, P., Araujo, D., and Bartlett, R. (2003) Movement
systems as dynamical systems: the functional role of variability and its
implications for sports medicine. Sports Medicine, 33 (4): 245–60.
Duchon, A. P., Kaelbling, L. P. and Warren, W. H. (1998) Ecological
robotics. Adaptive Behavior, 6 (3–4): 473–507.
Durand-Bush, N. and Salmela, J. (2002) The development and maintenance
of expert athletic performance: perceptions of World and Olympic
champions. Journal of Applied Sport Psychology, 14 (3): 154–71.
Edelman, G. M. and Gally, J. A. (2001) Degeneracy and complexity in
biological systems. PNAS: Proceedings of the National Academy of Sciences
of the USA, 98 (24): 13763–8; doi: 10.1073/pnas.231499798 (accessed 10
June 2013).
Frank, T. D., Peper, C. E., Daffertshofer, A. and Beek, P. J. (2006)
Variability of brain activity during rhythmic unimanual finger movements, in
K. Davids, S. J. Bennett and K. Newell (eds) Movement System Variability.
Champaign, IL: Human Kinetics, pp. 271–6.
Fringelkurts, A. A. and Fringelkurts, A. A. (2004) Making complexity
simpler: multivariability and metastability in the brain. International Journal
of Neuroscience, 114: 843–62.
Gibson, J. J. (1979) The Ecological Approach to Visual Perception. Boston:
Houghton Mifflin.
Glazier, P. S. and Davids, K. (2009) Constraints on the complete
optimization of human motion. Sports Medicine, 39 (1): 15–28.
Glazier, P. S., Davids, K. and Bartlett, R. M. (2003) Dynamical systems
theory: a relevant framework for performance-orientated sports biomechanics
research. Sportscience, 7. Available online at
www.sportsci.org/2003/index.html (accessed 10 June 2013).
Gould, D., Dieffenbach, K. and Moffett, A. (2002) Psychological
characteristics and their development in Olympics champions. Journal of
Applied Sport Psychology, 14 (3): 172–204.
Handford, C., Davids, K., Bennett, S. and Button, C. (1997) Skill acquisition
in sport: some applications of an evolving practice ecology. Journal of Sports
Sciences, 15 (6): 621–40.
Holt, N. L. and Dunn, J. G. H. (2004) Toward a grounded theory of the
psychosocial competencies and environmental conditions associated with
soccer success. Journal of Applied Sport Psychology, 16 (3): 199–219.
James, D. (2008) The physics of winning: engineering the world of sport.
Physics Education, 43 (5); doi: 10.1088/0031-9120/43/5/006. Available
online at http://iopscience.iop.org/0031-9120/43/5/006 (accessed 10 June
2013).
Kaerns, M., Elston, T. C., Blake, W. and Collins, J. J. (2005) Stochasticity in
gene expression: from theories to phenotypes. Nature Reviews Genetics, 6:
451–64.
Kauffman, S. (1995) At Home in the Universe: The Search for Laws of
Complexity. London: Viking.
Kauffman, S. A. (1993) The Origins of Order: Self-organization and
Selection in Evolution. New York: Oxford University Press.
Kelso, J. A. S. (1991) Anticipatory dynamical systems, intrinsic pattern
dynamics and skill learning. Human Movement Science, 10: 93–111.
Kelso, J. A. S. (1995) Dynamic Patterns: The Self-organization of Brain and
Behavior. Cambridge, MA: MIT Press.
Kelso, J. A. S. (2008) An essay on understanding the mind. Ecological
Psychology, 20: 180–208.
Liu, Y.-T., Mayer-Kress, G. and Newell, K. M. (2006) Qualitative and
quantitative change in the dynamics of motor learning. Journal of
Experimental Psychology: Human Perception and Performance, 32 (2): 380–
93.
Morris, T. (2000) Psychological characteristics and talent identification in
soccer. Journal of Sports Sciences, 18 (9): 715–26.
Morriss, C. J., Bartlett, R. M. and Fowler, N. (1997) Biomechanical analysis
of the men's javelin throw at the 1995 world championships in athletics. New
Studies in Athletics, 12: 32–41.
Newell, K. M. (1986) Constraints on the development of coordination, in M.
G. Wade and H. T. A. Whiting (eds) Motor Skill Acquisition in Children:
Aspects of Coordination and Control. Amsterdam: Martinus Nijhoff, pp.
341–60.
North, S. (2004) Landy took it all in his stride and is still making the running.
Sydney Morning Herald, 1 May. Available online at
www.smh.com.au/articles/2004/04/30/1083224590040.html (accessed 10
June 2013).
Obler, L. K. and Fein, D. (1988) The Exceptional Brain: Neuropsychology of
Talent and Special Abilities. New York: Guilford Press.
Pyne, D. B., Duthie, G., Saunders, P. U., Petersen, C. A. and Portus, M. R.
(2006) Anthropometric and strength correlates of fast bowling speed in junior
and senior cricketers. Journal of Strength and Conditioning Research, 20 (3):
620–6.
Renshaw, I., Oldham, A. R., Golds, T. and Davids, K. (2007) Changing
ecological constraints of practice alters coordination of dynamic interceptive
actions. Journal of Sport Sciences 7 (3): 157–67.
Renshaw, I., Davids, K., Chow, J. and Shuttleworth, R. (2009) Insights from
ecological psychology and dynamical systems theory can underpin a
philosophy of coaching. International Journal of Sport Psychology, 40 (4):
540–602.
Schöllhorn, W. I. (2003) Coordination dynamics and its consequences on
sport. International Journal of Computer Science in Sport, 2: 40–6.
Simonton, D. K. (1999) Talent and its development: an emergenic and
epigenetic model. Psychological Review, 106 (3): 435–57.
Sumpter, D. J. T. (2006) The principles of collective animal behaviour.
Philosophical Transactions of the Royal Society B: Biological Sciences, 361:
5–22.
Sutton, J. (2008) Batting, habit and memory: the embodied mind and the
nature of skill. Sport in Society, 10 (5): 763–86.
Thamel, P. (2008) Poor 3-Point Shooting Can't Stop U.S. Basketball Team.
New York Times, 12 August: D2. Available online at
www.nytimes.com/2008/08/13/sports/olympics/13games.html (accessed 10
June 2013).
Thelen, E. (1995) Motor development: a new synthesis. American
Psychologist, 50 (2): 79–95.
Thelen, E. and Smith, L. B. (1994) A Dynamic Systems Approach to the
Development of Cognition and Action. Cambridge, MA: MIT Press.
Vaeyens, R., Lenoir, M., Williams, A. M. and Philippaerts, R. M. (2008)
Talent identification and development programmes in sport: current models
and future directions. Sports Medicine, 38 (9): 703–14.
Van Orden, G. C., Holden, J. G. and Turvey, M. T. (2003) Self-organization
of cognitive performance. Journal of Experimental Psychology: General, 132
(3): 331–50.
Ward, P., Hodges, N. J., Starkes, J. L. and Williams, M. A. (2007) The road
to excellence: deliberate practice and the development of expertise. High
Ability Studies, 18 (2): 119–53.
Wolstencroft, E. (2002) Talent Identification and Development: An Academic
Review. Edinburgh: Sportscotland.
15 Creativity in sport and dance
Ecological dynamics on a hierarchically
soft-assembled perception–action
landscape
Robert Hristovski, Keith Davids, Duarte Araújo, Pedro Passos, Carlota
Torrents, Alexandar Aceski, and Alexandar Tufekcievski
Innovative and creative goal-directed behaviours in complex neurobiological
systems, such as apes and birds, has been studied extensively (see, e.g.
Reader and Laland 2001; Taylor et al. 2010) and involves the discovery of
novel patterns of behaviour by an organism. There has not been the same
focus on creative behaviour in sport, although much research in sport
sciences and pedagogy has been aimed at improving athletic performance.
Traditionally, performance optimization methods have been implemented to
identify the set of movement parameters that might maximize competitive
outcomes for specific elite athletes (e.g. Bartlett 2007). However, as in the
case of Dick Fosbury, the elite high jumper, sometimes exploration of novel
movement patterns can not just improve performance but actually push it to a
new, higher level.
Such advances occur, not by optimizing model parameters of an existing
wellestablished technique but by construction of new coordination patterns.
This creative advance can resolve problems with a previous technique, shape
performance outcomes and enrich movement culture, making specific sports
more diverse and more aesthetically attractive. On the other hand the
sociocultural influence on the performer or the team put strong constraints of
nonspecific and specific kind (for more on this issue see Hristovski et al.
2011). The changes in socioculturally induced nonspecific task constraints,
like the type of tool used or environment on which is acted upon, may be
instrumental in eliciting innovative performer–environment relations (e.g.
Bril et al. 2005). On the other hand, specific social influences, i.e. imitation
of extant performer–environment relations, are constraining athletes to
comply with these movement forms within the sociocultural milieu.
However, complying with extant movement forms is just the opposite from
creativity defined as process of exploration and production of novel and
functional behaviour. Drazin et al. (1999) and Sternberg and Lubart (1996)
provide detailed explanations of these two basic types of creative behaviour,
respectively.
Inventions of novel performance solutions occur regularly in sport and
have been well documented in activities like track and field, exemplified by
the ‘O'Brien’ (or rotational) technique in shot put or the ‘straddle’ and
‘Fosbury flop’ techniques which replaced more traditional and less successful
‘scissors' and ‘eastern cut-off’ high-jump techniques. In the last few decades
alone in gymnastics, 14 novel technical elements have appeared. Yet despite
this process, creativity in sport performance has seldom been the subject of
systematic research, revealed by the absence in sport science of established
theoretical rationales for studying and explaining creative behaviour. In the
following text, we present a model of creative behaviour in sports and dance
and some new empirical results based on the ecological dynamics approach
(Araújo et al. 2006).
Multistability as a prerequisite for creativity in neurobiological
systems
The multitude of solutions or states available in complex performer–
environment systems is a consequence of the wealth of potential nonlinear
couplings available between system components (or between system agents
in sports teams as social neurobiological systems (e.g. Challet et al. 2000).
Degeneracy, pleiotropy and multistability exemplify performer–environment
properties that support such nonlinearity and complexity. In short,
degeneracy means that different structural components may be assembled to
satisfy the same task goal constraints and pleiotropy means that the same
structural component can have a role in satisfying different task goals.
Multistability (see Chapter 1 and 2 for definition) is a prerequisite for an
important property of complex systems metastability; i.e. the capacity to
possess numerous functional, coexistent pattern-forming propensities (e.g.
Bressler and Kelso 2001; Kello et al. 2008). It is a prerequisite, since
metastable dynamics can exist only in systems which have more than one
attracting states under the same longer-term constraints. In such systems, the
behaviour can dwell for some time in the vicinity of the attractor or attractive
point and then switch to another one. The monostability of linear systems
does not allow such metastability. This property of performer–environment
systems, allows system states (qua performance solutions) to be soft-
assembled, emerging under specific constellations of boundary conditions as
constraints. These system states are not preformed but emerge under
interacting constraints, basic principles that underpin the constraints-led
perspective on skill acquisition in sport (Davids et al. 2008; Davids et al.
2013; Davids et al. 2003; Araújo et al. 2004, 2006; Davids et al. 2005; Chow
et al. 2006; Newell 1986).
The distinctive configurations of constraints between learners, based on a
platform of system degeneracy and pleiotropy, are manifested in how each
individual attempts to satisfy specific task constraints during practice.
Inventing a new technique or a movement form (i.e. coordination) is always
defined as being in the coordination stage of learning, which can be
afterwards, through practice, made more functional (for more detailed
discussion, see Hristovski et al. 2012). The invented coordination, of course,
may be supported and contextualized by already-stabilized skills. Individual
creativity is a product of the interactions of three nonlinear properties: cause–
effect nonproportionality, parametric (constraint) control and multistability in
complex systems (for definitions, see Chapter 1 and 2; Hristovski et al.
2009). Within these interactions, nonlinear pedagogy (e.g. Chow et al. 2006)
frames the individuality of learning pathways and individual creation of
performance solutions for a given movement task. Based on these arguments,
we next outline a nonlinear, complex systems model of creativity in sports
and physical activity.
Ecological dynamics on a hierarchically soft-assembled
potential landscape: a model outline
The empirically-based, hierarchical structure of human movement variability
(see e.g. Chow et al. 2008) provides an explanation of the behavioural
characteristic of creativity within the framework of statistical mechanics of
disordered systems; i.e. systems with many different but ordered states,
particularly the replica symmetry-breaking framework first developed by
Parisi (e.g. Parisi 1979). The hierarchical structure of movement variability
implicates a rugged landscape with many nested metastable minima
(Hristovski and Davids 2008), within the global basin of attraction. For
clarification, one such landscape is depicted for fixed task constraints in
Figure 15.1.

Figure 15.1 Schematic (i.e. one-dimensional) presentation of the corrugated hierarchically soft-
assembled potential landscape with two confining barriers on both sides (reproduced with kind
permission from Nonlinear Dynamics, Psychology and Life Sciences, Springer)

The main idea behind this model is the absence of any simple symmetries
of potential order parameters (see Chapter 1 and 2) governing discrete and
whole body movements. In other words, in principle, it is hard to relate two
or more actions through simple symmetry transformations. For example, in
the HKB (Haken, Kelso and Bunz) model (Haken et al. 1985), the relative
phase order parameter exhibits mirror (change of the sign) symmetry, which
allows one to relate a set of patterns; i.e. values of the order parameter, with
those symmetries. However, a set of whole-body movements/postures or
multiarticular discrete movements in sport or dancing, generally, cannot be
related by simple symmetries and their rearrangements. Hence, the
relatedness of any set of discrete and whole-body movements may be made
by the so-called overlap order parameter q, which measures the mutual
correlation of this kind of actions. It can be defined as a cosine of the angle
between two random vectors, i.e. replicas, in real or formal space of any
finite dimension (e.g. Domany et al. 2001). Under absolutely relaxed
constraints, one may expect that these replicas are totally uncorrelated and
their average correlation is <q> = 0. We can say that the system is replica
symmetrical. Any replica, i.e. performer–environment configuration, can
emerge. However, under constraining influences of the task, properties of the
individual performer and the environment, this symmetry and ergodicity is
broken and clusters of correlated actions arise (for detailed examples, see
Hristovski et al. 2011, 2012). Some similar replicas, i.e. configurations, are
more likely to occur and some are unlikely. Constraints break system
symmetry, produce a phase transition and also create ergodicity breaking
high barriers on the both sides of the potential landscape, confining all
available actions within the internal space (Figure 15.1). It can be observed
that the action landscape soft-assembles under specific constraints
configurations.
These correlated clusters form a hierarchical landscape. Actions lying in
one attractor basin (see Chapter 1 for definition), separated by smaller
barriers are more correlated than those separated by higher barriers. Thus, we
can define order parameters at each level, with those defining lower levels
depending on increasingly subtle constraints. These order parameters form a
tree-like structure. Note that, while slowly changing (i.e. quasi-stationary)
interacting constraints, are forming the deterministic structure of the
landscape (see Figure 15.1 and the previous text), there is also a need of
stochastic agitation force within the performer–environment system, mainly
contributed by the interaction between performer's intention to move and the
unpredictable and quickly changing physical and information constraints,
which are the driving force of reconfigurations. Hence, both the deterministic
structure and the stochastic drive interact in reconfiguring the systems
dynamics. The exploratory dynamics within this landscape, may be seen as a
hopping of the system from one basin to another or equivalently as a random
walk on a tree. The hopping or the random walk has a meaning of
reconfigurations within the system that are taking place. Hopping over larger
barriers means larger reconfigurations and vice versa.
This modelling shows how exploration is a requisite of creative behaviour;
i.e. inventing novel or innovating extant movement forms and actions with
respect to the extant sociocultural milieu (for details see Hristovski et al.
2011). Without it, a complex neurobiological system cannot find novel
functional behavioural solutions. In Hristovski et al. (2011), exploratory
breadth Q was defined as being equal to the average escape probability over
all possible state basins of attraction (see Saxton 1996), Q = We. Escape
probabilities for each movement mode are defined as We = 1 – Wc, where Wc
is the conditional probability of staying inside the same attractor (Hristovski
et al. 2009). In other words, Wc measures the trapping strength of the
attractor; i.e. the probability of being able to achieve the same performance
outcomes sequentially. The larger the average escape probability We, the
larger the exploratory breadth Q of the system and vice versa. In general, it
can be said that: for any performer–environment system containing a large
amount of degrees of freedom, there always exists a set of constraints which
maximizes the functional action versatility, i.e. exploratory breadth, defined
as maximum action entropy (see e.g. Hristovski et al. 2006; Pinder et al.
2012). A more thorough exposition of this model and an experimental
example of novel action emergence in martial arts can be found in Hristovski
et al. (2011) and Hristovski et al. (2012).
In the following section, we further illustrate how this theoretical model
provides an analysis of movement exploration structure and dynamics applied
in context of dance improvisation. The general predictions of a hierarchical
structure and softly assembled dynamics were preliminarily tested. A
conceptual model of creativity in team sports within the general framework
of dynamical systems is also presented.
How creative behaviours emerge under ecological constraints
in dance and team sports

Creative behaviour in contact improvisation dance: a soft-

assembled hierarchy model analysis
Contact improvisation is a form of dance where contact points between and
movements of partners constrain each other's subsequent movements. What
are the creative aspects of contact improvisation? Contact improvisation may
be thought of as a bank of emergent human movement forms and an
expression of continuous exploration and discovery of idiosyncratic postures
and actions supported by immediate affordances; i.e. opportunities for action.
It is used systematically as a dance research method for identifying
innovative set choreography. The invented novel forms of movement, being
themselves creative products, could afterwards be stylized and put to a further
use. These properties of contact improvisation make it a highly creative
activity satisfying both, the process and the product definitions of creativity
as a phenomenon (for these definitions, see (Drazin et al. 1999; Sternberg
and Lubart 1996). This characteristic of contact improvisation fits nicely
within the general approach in our experiments for discovering the
dynamical, perception–action landscape of sport/dance performers which
involves probing system activity by letting it evolve autonomously under
specific task constraints manipulations over a period of time.
Here, we describe some results of an analysis of a typical contact
improvisation session lasting 450 seconds under no special instructional
constraints, except those provided by the contact with the partner, i.e. visual
and haptic information, and the force of gravity. Sequences of
actions/postures were analyzed to determine their complex dynamical
characteristics. Actions/postures were defined on a coarse-grained scale
containing 52 movement/posture components, such as support/contact
characteristics and directions and planes of motion of body segments
according to established observational methodology (see Torrents et al. 2010;
Castañer et al. 2009). To the active components a value of 1 was ascribed and
to the inactive components a value of 0. Hence, a binary matrix was formed
with a time resolution of 1 second. Each 1-second window was defined as a
52- component binary vector representing the action configuration during the
same time interval. Reconfigurations, i.e. mutations, of action patterns were
calculated as Hamming distances between any two binary vectors. For
example, the change of one component of the vector from 1 to 0 or vice versa
has a Hamming distance equal to 1. Hence, the Hamming distance actually
measures the height of the potential barrier between two configurations.
Overlap order parameter q was used to determine the structure of the
potential landscape of the dancer and its dynamic properties. The overlap was
defined in two intrinsically related association measures: as a cosine
similarity and as a Pearson correlation between two binary configuration
vectors. A hierarchical principal component analysis (HPCA) was performed
on the data using the second measure (for the plausibility of using principal
component analysis on binary variables, see Joliffe 2002, p. 339), with the
aim of detecting the possible nested attractor basin structure of the dancer's
action landscape. In order to determine not only the structure of dancer's
complex movement patterns but also their exploratory dynamics, the dynamic
overlap qd(t) was calculated as an average cosine autosimilarity of the
overlap between configurations with increasing time lag.
The HPCA initially revealed 25 principal components. Seven primary
principal components accounting for 81.3% of the total variance were taken
into account. The first, the fifth, the sixth and the seventh principal
component were weakly but significantly correlated <q> = 0.401 ± 0.02. The
other 18 principal components contained statistically rare and short-lived
reconfigurations having a role of fluctuations. This structure is interesting in
itself. While the dominant dynamics were confined within the seven principal
components, the reconfiguration space was with significantly higher
dimension. This gives a picture of lower-dimensional global dynamics
connected by high-dimensional quick reconfigurations. Seventeen
movement–posture components had very high scores on these seven principal
components. Consequently, the dynamics were saturated predominantly by
these emergent movements-posture patterns.
The secondary level consisted of two principal components containing the
seven primary principal components as substructure. Hence, the HPCA
procedure reduced the dancer's dynamics dimensionality from 449
configuration vectors to the two slowest collective variables (order
parameters) containing seven faster variables, each of which was formed by
even faster movement/posture configuration vectors. The median dwell time
of these configuration vectors was 5.4 seconds. This structure revealed a
metastable dynamics in a soft-assembled hierarchy of collective variables
with different characteristic time scales (for intuitive understanding, see
Figure 15.1). The global and the slowest collective variable consisted of
degrees of freedom such as foot–floor surface support, support of the partner
with lower and upper limbs and the trunk-torso-back-pelvis surfaces. They
were present, with only quick interruptions, over the whole observation
timescale. This persistent collective variable formed the global basin of
attraction and continually constrained the faster degrees of freedom such as a
change in the direction of movement, lateral or forward flexion of the trunk,
arm movements, and so on. These degrees of freedom under constraining
influence of the slow collective variable formed shorter-lived metastable
states (local basins of attraction) on the lower hierarchical level. In other
words, the slowest collective variable formed rather persistent pinning
surfaces between partners, nucleuses, around which more quick movement
motifs were created.
In the upper panel of Figure 15.2, one can see a zoomed example of
metastable dynamics of the first 35 seconds, where the action system of a
dancer transits from one local metastable configuration attractor basin to
another, meanwhile dwelling inside one of them for several seconds. These
four principal components extracted from the first 35 time-ordered
configurations were correlated <q> = 0.64 ± 0.05 and belonged to a
secondary global confining attracting basin (secondary principal component).
At each moment, several opportunities for action coexist, owing to
informational constraints from the partners and the floor, the gravity and the
abundance of nonlinearly coupled degrees of freedom within the system of
dancers. Which one of these coexistent propensities will become a temporary
solution, i.e. a creative product, depends on subtle and quick information, i.e.
fast perceptual-motor degrees of freedom and decisions dressed with
randomness, e.g. who will lead and who will follow in the immediate future
interval. Aside from their common motive to move, randomness is a crucial
part of the dynamics because partners do not deterministically guide each
other, i.e. with perfect probability (P = 1). Each temporary posture or
movement solution is a creative product embedded within the creative
exploration dynamics. Hence, ‘exploratory’ or ‘process' and ‘product’
properties of creative behaviour are becoming obvious (Drazin et al. 1999;
Sternberg and Lubart 1996).
 The darkest areas of the centre panel of Figure 15.2 correspond to
maximum overlap q values of certain configurations with the certain principal
component and represent the locally attracting configurations. The increasing
lightness in the conturogram signifies a lower overlap and represents the local
attracting basin around the attractor configurations. One can notice that from
the darkest areas (local potential minima) going outwards the lightness
increases, reaching in some areas the lightest grey colour, which represents
saddles or barriers of the potential landscape. Looking at the lower panel of
Figure 15.2, where Hamming distances count the number of reconfigurations
or mutations that have occurred, we see that those lighter shaded areas
correspond to the intervals of reconfigurations of the action system of the
dancer. Hence, reconfigurations correspond to crossing over the potential
barriers (saddles) between two metastable minima.
The evolution of the cosine autosimilarity, i.e. the average dynamic
overlap <qd(t)>, is given in Figure 15.3. If the dancers stayed in one posture
during whole observation time lasting seven minutes, the average dynamic
overlap would be a constant equal to one; i.e. <qd(t)> = 1 for all time lags
(see the dashed line on Figure 15.3). On the other hand, if during the
performance, the dancers were exploring distant configurations of the whole
state space (a case of unbroken ergodicity), the average dynamic overlap
<qd(t)> would drop to zero exponentially fast for the observation time of
seven minutes (see the dotted curve on Figure 15.3). Neither of these cases
was obtained in the observed performance. The fact that the relaxation of
<qd(t)> was not exponential, means that there was no single characteristic
time of the dynamics. In other words, the dynamics contains a distribution of
relaxation times and thus a distribution of attractor basins of different depth
(barriers of different height), which is consistent with the predictions of the
model of soft-assembled hierarchy explained briefly previously. According to
the model, Figure 15.3 can be interpreted in the following way. The dancer
explores subsequent movement configurations, which are correlated; i.e.
lying within a local basin of attraction defined by the correlated principal
component substructure and makes, on average, small gradual
reconfigurations. On average, after longer times, s/he hops into attractor
basins which are less correlated with previous ones; i.e. moves within another
set of principal components. These reconfigurations with time accumulate
and bring the dancer far enough from a certain initial configuration. That is
why the dynamical overlap <qd(t)> decreases for the first 35 time lags,
forming the initial relaxation dynamics. The linear fit in log–log plot of this
initial relaxation was β = −0.203; with explained variance R2 = 0.977. This
variable roughly shows the rate of exploration of the system. One can see
that even movement configurations separated by 20 seconds possess overlaps
larger than 0.5. However, on average, movement configurations separated in
time by more than 35 seconds show a constant average overlap of
approximately <qd(t)> = 0.45. This is the value of the average overlap of all
the dancer's movement configurations that emerged within the seven-minute
contact improvisation. In other words, this is the self-overlap value of the
superbasin of attraction confining all of the metastable valleys of movement
configurations captured by the hierarchical principal component structure.
Since the dynamic overlap forms a plateau at values far from zero, the system
does not explore arbitrarily distant movement configurations (ergodicity is
broken on the observational time scale of seven minutes), because the slowest
collective variable (see the previous discussion), confine the emergent
metastable opportunities for action and, consequently, the movement
dynamics, in a relatively large but limited region of performer–environment
state space. Nevertheless, the plateau value <qd(t)> = 0.45, shows a relatively
high exploratory; i.e. creative fluency of the system. The fact that the
dynamical overlap, during the observation time of seven minutes, did not
decrease to zero means that the global relaxation (equilibration) time of the
slowest, order parameters (secondary principal components) diverged (did
not go to zero) and the system was in a nonequilibrium state. This means that
while there was an equilibration within the confined region (primary principal
components), the space of all possible configurations remained far from
being fully explored. Such slow dynamics is consistent with predictions of
the previously discussed theoretical model of soft-assembled hierarchical
dynamics. The exploratory breadth of dancer's movement system Q (see
previous section) defined as an average escape probability from a certain
configuration was Q = 0.37, showing a significant exploratory capacity under
relaxed constraints.
Figure 15.2 (Top) snapshots of the four metastable states of dancers for the first 35 seconds of
improvisation; (middle): overlaps q of the four metastable body configurations with principal
components (axis Y) and pathway of their dynamics. Overlap values are given in the legend on right;
(bottom): reconfigurations are given as nonzero Hamming distances of the dancer's action system; after
some reconfigurations take place, the action system relaxes and dwells in a state where no further
reconfigurations occur for some time; i.e. zero reconfiguration; this process represents the metastable
attractor configurations (movement/posture pattern)

Figure 15.3 The profile of the average dynamic overlap qd(t) for different time lags; its dynamics
proceed on three timescales (from seconds to several minutes) and does not converge to zero during the
observation time scale

In this section we have shown how exploratory, i.e. process and product
aspects of creative behaviour can be analyzed within the model of soft-
assembled hierarchy. The creative process unfolds on more time scales,
owing to the hierarchical structure of the perception–action landscape. The
soft-assembled hierarchical landscape model predicts also other interesting
phenomena, such as aging: the more time the learner spends in a confined
and thus correlated region of the landscape, the less responsive s/he becomes
to a change. Based on the obtained consistency of the experimental results
with respect to the predictions of the theoretical model, we further
hypothesize that, together with Q, the slope of initial relaxation â and the
value of the <qd(t)> plateau, are good candidates for assessing the creative
capacity of performer–environment systems under different types and
strengths of constraints (for details see Hristovski et al. 2011). Future work is
needed to test other predictions of the model.
Conceptual modelling of creative behaviour in team sports: the
need to play within critical regions of interpersonal distance
Attackers' interactions aim to actively explore space–time windows that
emerge because of defender displacements. On the other hand, defender
displacements aim to cover the possible paths to goal, which demands high
levels of interpersonal coordination among the players in defence. However,
space–time windows will only emerge if the attackers' movements are
powerful enough to disturb the defenders' interpersonal coordination and, to
do that, attackers' actions must be performed within short distances of
attacker–defender interpersonal distance (Passos et al. 2008).
Thus, sudden changes in the attacker–defender structural organization can
only happen when the attacker–defender systems moved towards regions of
very short interpersonal distances, where the contextual dependency among
players emerge, characterizing the performance region as critical. Within
these critical regions, the players' contextual dependency moves the system
from equally poised options to a single option, that emerges under the
influence of task and environmental constraints. In other words, within these
critical regions, creativity occurs as ongoing performance solutions emerge
and are annihilated, until a sudden change occurs where a single, i.e. creative
solution, emerges (Passos et al. 2009). In this sense, within critical regions,
exploratory metastable behaviour emerges as a precursor to the creative
product, i.e. the single solution. Hence, both the exploratory and product
phase of creative behaviour exist (Drazin et al. 1999 and Sternberg and
Lubart 1996). The players' contextual dependency creates local information
that originates at a specific moment in time and space where a gap in the
defensive system emerges and the attackers exploit it to move closer to the
goal or even score. This process underlines the notion that creativity in team
sports is based upon a self-organization mechanism that only occurs within
critical regions.
Creativity in team sports is sustained by the nonlinear interactions among
players, which enable nonproportional, abrupt and unpredictable environment
for the opponents. As in any other social system, the way that each player
interacts with others in the neighbourhood of play influences the behaviours
of players within the same team and this is a requisite to disturb the actions of
opponents (Fajen et al. 2009). From an attacker's perspective, the decisions of
the ball carrier and support players are based on the perceptions that they
have created of the defenders' relative positioning, running-line trajectories
and proximity to each other (Passos et al. 2008). On the other hand, the
decision making of defenders depends on the perception that they have of the
ball carrier's actions as well as the behaviours of the support players (Passos
et al. 2008). These variables include interpersonal distances, the speed and
running-line trajectories that contain important information concerning the
attackers' ability to perform different actions. These variables contain
information that are perceived by the players and specify the action
possibilities of each opponent or teammate (Gunns et al. 2002; Weast et al.
2011). This is where creativity emerges, with the need for attackers to
perform deceptive actions that creates the impression of multiple different
possibilities for action. These deceptive actions can also be characterized by
intrateam coordination, where attackers perform a set of previous established
movements that are intended to open a space–time window against a stable
opposing team. This is when creativity is needed again and players need to
reorganize, avoiding defenders. This reorganization process is grounded on
situational information concerning defenders' relative positions, number,
speed and distance to goal (Travassos et al. 2011; Cordovil et al. 2009;
Passos et al. 2011). These sources function as task constraints that attackers
use to avoid defenders. The reorganization of attackers is grounded on
situational information that emerges because of opponent players' nonlinear
interactions and is selforganized.
Typically, attacker–defender interactions are characterized with many
subtle fluctuations in the attacker–defender balance but also with few abrupt
changes in the attacker–defender structural organization, meaning that
suddenly the attackers gain an advantage and are in a crucial position to
score.
To summarize, in this chapter, we have outlined a model of creative
exploration and solution (product) emergence as a soft-assembly of actions
under ecological constraints. The obtained structural and dynamical hierarchy
of the creative behaviour, consistent with the model predictions, was
demonstrated in a contact improvisation dance example. The creative process
unfolds on different timescales, owing to the hierarchical structure of the
perception–action landscape. Hence, creativity is a nested process, both
structurally and temporally. The experimental examples provided here are
extensions of those in Hristovski et al. (2011), where an emergence of a
novel punch in martial arts was treated within the framework of the same
theoretical model. The role of slowly changing or constant environmental and
personal constraints, (i.e. gravity and morpho-anatomical structure in creative
exploratory activity) is provided by their slow contextualizing function.
Quickly changing and stochastic physical and informational constraints form
the base of the unpredictability function in creativity. Through this process
they mould the goal-directed activity of complex systems potentially leading
to inventions of new movement forms or team actions. By manipulation of
these key constraints, athletes structure different types of contexts which
eventually lead to soft-assembly of novel forms of actions. In this way,
highly motivated athletes, by self-generated experimentation in the full space
of constraints, may facilitate the emergence of new and functional action
forms.
References
Araújo, D., Davids, K., Bennett, S., Button, C. and Chapman, G. (2004)
Emergence of sport skills under constraints, in A. M. Williams and N. J.
Hodges (ed.) Skill Acquisition in Sport: Research, Theory and Practice.
London: Taylor and Francis, pp. 409–33.
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7: 653–76.
Bartlett, R. M. (2007) Introduction to Sport Biomechanics: Analysing Human
Movement Patterns, 2nd edn. London: Routledge.
Bressler, S. L. and Kelso, J. A. S. (2001) Cortical coordination dynamics and
cognition. Trends in Cognitive Sciences, 5 (1): 26–36.
Bril, B., Roux, V. and Dietrich, G. (2005) Stone knapping: Khambhat (India),
a unique opportunity? in V. Roux and B. Bril (eds) Stone Knapping, the
Necessary Conditions for an Uniquely Hominid Behaviour. McDonald
Institute Monograph Series, Cambridge: McDonald Institute for
Archaeological Research, pp. 53–72.
Castañer, M., Torrents, C., Anguera, M. T., Dinušova, M. and Jonsson G. M.
(2009) Identifying and analyzing motor skill responses in body movement
and dance. Behavior Research Methods, 41 (3): 857–67.
Challet, D., Marsili, M. and Zecchina, R. (2000) Statistical mechanics of
systems with heterogeneous agents: minority games. Physical Review Letters,
84 (8): 1824–7.
Chow, J. Y., Davids, K., Button, C., Shuttleworth, R., Renshaw, I. and
Araújo, D. (2006) Nonlinear pedagogy: a constraints-led framework to
understand emergence of game play and skills. Nonlinear Dynamics,
Psychology and Life Sciences, 10 (1): 74–104.
Chow, J. Y., Davids, K., Button, C. and Rein, R. (2008) Dynamics of
movement patterning in learning a discrete multiarticular action. Motor
Control, 12 (3): 219–40.
Cordovil, R., Araujo, D., Davids, K., Gouveia, L., Barreiros, J. and
Fernandes, O. and Serpa, S. (2009) The influence of instructions and body-
scaling as constraints on decisionmaking processes in team sports. European
Journal of Sport Sciences, 9 (3): 169–79.
Davids, K., Araújo, D., Shuttleworth, R. and Button, C. (2003) Acquiring
skill in sport: A constraints-led perspective. International Journal of
Computer Sciences in Sport, 2: 31–9.
Davids, K., Renshaw, I. and Glazier, P. (2005) Movement models from
sports reveal fundamental insights into coordination processes. Exercise and
Sport Science Reviews, 33: 36–42.
Davids, K., Button, C., and Bennett, S. (2008) Dynamics of Skill Acquisition.
A Constraints-led Approach. Champaign: Human Kinetics
Davids, K., Araújo, D., Vilar, L., Renshaw, I. and Pinder, R. (2013) An
ecological dynamics approach to skill acquisition: implications for
development of talent in sport. Talent Development and Excellence, 5 (1):
21–34.
Domany, E., Hed, G., Palassini, M. and Young, A. P. (2001) State hierarchy
induced by correlated spin domains in short-range spin glasses. Physical
Review B, 64: 224–406.
Drazin, R., Glynn, M. A. and Kazanjian, R. K. (1999) Multilevel theorizing
about creativity in organizations: a sensemaking perspective. Academy of
Management Review, 24: 286–307.
Fajen, B., Riley, M. and Turvey, M. (2009) Information, affordances and the
control of action in sport. International Journal of Sport Psychology, 40: 79–
107.
Gunns, R. E., Johnston, L. and Hudson, S. M. (2002) Victim selection and
kinematics: a point-light investigation of vulnerability to attack. Journal of
Nonverbal Behavior, 26 (3): 129–58.
Haken, H., Kelso, J. A. S. and Bunz, H. A. (1985) Theoretical model of phase
transitions in human hand movements. Biological Cybernetics, 51: 347–56.
Hristovski, R., Davids, K. and Araújo, D. (2006) Affordance – controlled
bifurcations of action patterns in martial arts. Nonlinear Dynamics,
Psychology and Life Sciences, 4: 409–44.
Hristovski, R. and Davids, K. (2008) Metastability and situated creativity in
sport. Paper presented at the 2nd International Congress of Complex Systems
in Sport, 4–8 November, Madeira, Portugal.
Hristovski, R., Davids, K. and Araújo, D. (2009) Information for regulating
action in sport: metastability and emergence of tactical solutions under
ecological constraints, in D. Araújo, H. Ripoll and M. Raab (eds)
Perspectives on Cognition and Action in Sport. New York: Nova Science
Publishers, pp. 43–57.
Hristovski, R., Davids, K., Araújo, D. and Passos, P. (2011) Constraints-
induced emergence of functional novelty in complex neurobiological
systems: a basis for creativity in sport. Nonlinear Dynamics, Psychology and
Life Sciences, 15 (2): 175–206.
Hristovski, R., Davids, K., Passos, P. and Araújo, D. (2012) Sport
performance as a domain of creative problem solving for self-organizing
performer–environment systems. Open Sports Science Journal, 5 (Suppl 1-
M4): 26–35.
Joliffe, I. T. (2002) Principal Component Analysis, 2nd edn. New York:
Springer.
Kello, C. T., Anderson, G. G., Holden, J. G. and Van Orden, G. C. (2008)
The pervasiveness of 1/f scaling in speech reflects the metastable basis of
cognition. Cognitive Science, 32 (7): 1217–31.
Newell, K. M. (1986) Constraints on the development of coordination, in M.
G. Wade and H. T. A. Whiting (eds) Motor Development in Children.
Aspects of Coordination and Control. Dordrecht, Netherlands: Martinus
Nijhoff, pp. 341–60.
Parisi, G. (1979) Infinite number of order parameters for spin-glasses.
Physical Review Letters, 43: 1754–6.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Milho, J. and Serpa, S.
(2008) Informationgoverning dynamics of attacker–defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Passos, P., Araújo, D., Davids, K., Gouveia, L., Serpa, S., Milho, J. and
Fonseca, S. (2009) Interpersonal pattern dynamics and adaptive behavior in
multi-agent neurobiological systems: a conceptual model and data. Journal of
Motor Behavior, 41 (5): 445–59.
Passos, P., Milho, J., Fonseca, S., Borges, J., Araújo, D. and Davids, K.
(2011) Interpersonal distance regulates functional grouping tendencies of
agents in team sports. Journal of Motor Behavior, 43 (2): 155–63.
Pinder, R., Davids, K. and Renshaw I. (2012) Metastability and emergent
performance of dynamic interceptive actions, Journal of Science and
Medicine in Sport, 15 (1): 1–7.
Reader, S. M. and Laland, K. N. (2001) Primate innovation: sex, age, and
social rank differences. International Journal of Primatology, 22: 787–805.
Saxton, M. (1996) Anomalous diffusion due to binding: a Monte Carlo study.
Biophysical Journal, 70: 1250–62.
Sternberg, R. J. and Lubart, T. I. (1996) Investing in creativity. American
Psychologist, 51: 677–88.
Taylor, A. H., Elliffe, D., Hunt, G. R. and Gray, R. D. (2010) Complex
cognition and behavioural innovation in New Caledonian crows. Proceedings
of the Royal Society B Biological Sciences, 277 (1694): 2637–43.
Torrents, C., Castañer,M., Dinušova, M. and Anguera, M. T. (2010)
Discovering new ways of moving: observational analysis of motor creativity
while dancing contact improvisation and the influence of the partner. Journal
of Creative Behavior, 44 (1): 45–61.
Travassos, B., Araújo, D., Vilar, L. and McGarry, T. (2011) Interpersonal
coordination and ball dynamics in futsal (indoor football) Human Movement
Science, 30 (6): 1245–59.
Weast, J. A., Shockley, K. and Riley, M. A. (2011) The influence of athletic
experience and kinematic information on skill-relevant affordance perception.
Quarterly Journal of Experimental Psychology, 64 (4): 689–706.
Part 4
Complexity sciences and
training for sport
16 Variability in neurobiological
systems and training
Chris Button, Ludovic Seifert, David
O'Donovan and Keith Davids
Variability in achieving consistent performance outcomes has become
increasingly recognized as important in preparing for dynamic performance
contexts like sport. Here, we examine the theoretical basis for viewing
variability as functional and examine the implications for understanding
training for sport in individual and team sports. The issue of a putative
optimal movement pattern, common to all learners, is challenged. In this
chapter, we describe motor expertise as the capacity to functionally adapt
behaviours to satisfy key constraints in order to achieve intended
performance outcomes. Darwin, amongst many others, recognized the
fundamental value of individual variation and adaptation for the functional
behaviour and long-term survival of biological organisms. Similarly, at the
timescale of perception and action, success in sport is underpinned by the
capacity of athletes to capitalize on their individual strengths and to respond
appropriately to different challenges. Indeed, the principles of overload in
physical training (i.e. frequency, intensity, duration, type) are built upon
biological adaptation through which the human body becomes increasingly
prepared to function more efficiently and to produce sufficient energy to
attain higher performance goals. The importance of adaptability in motor
control was originally raised by Bernstein (1996) when he conceptualized the
notion of resourcefulness (i.e. stability and initiative) as an important
property of an organism's dexterity (p. 221). Whilst performing a complex
coordination pattern may be difficult for a learner in sport, more challenging
is its functional adaptation to a specific performance context; i.e. in
responding to interacting constraints (task, environmental, or organismic) that
continually change over time (Newell 1986). Biryukova and Bril (2002)
commented that ‘the dexterity is not movements in themselves, but their
ability to adapt to external constraints’ (p. 65).
A corollary of these theoretical ideas is the acceptance that there is not one
optimal pattern of coordination towards which all developing learners should
aspire but, instead, that expertise concerns ‘individual-constraint coupling’
(Seifert et al. 2013). Davids and Glazier (2010) postulated that this requisite
adaptability of complex movement systems was founded on pertinent
neurobiological system properties including degeneracy, defined as ‘the
ability of elements that are structurally different to perform the same function
or yield the same output’ (Edelman and Gally 2001: 13763). They proposed
that evidence of intra-individual movement variability often observed in
experts could play a functional role; for instance, it could correspond to
several types of movement and/or to the ability to use co-existing modes of
coordination (i.e. exploit multistability and metastability in complex
neurobiological systems). This relationship between multistability and
metastability in expert performance illustrates the requisite subtle balance
needed between stability and variability in sport performance (Jantzen et al.
2008), which arises from the complex interactions between intentions, actions
and perceptions of individual performers.
Traditionally, there has been an overriding tendency to define expertise as
the capacity to both reproduce a specific movement pattern consistently and
to increase the automaticity of movement (Seifert et al. 2013). In
representational accounts of expertise, movement variability is seen as noise
(i.e. an artefact limiting the system's processing of data input and output),
which should be minimized (Summers and Anson 2009). Interestingly, in
recent times, several computational models of control have surfaced,
attempting to cast the problem of variability for motor control in a more
positive light (i.e. computational modelling, Wolpert et al. 2003; optimal
control theory, Todorov and Jordon 2002). However, for some time, research
from a dynamical systems theoretical orientation has shown that movement
system variability is not noise detrimental to performance, error or deviation
from a putative expert model, which should be corrected in the beginner.
Movement system variability instead indicates the functional flexibility
needed to respond to dynamic performance constraints (Davids et al. 2003).
In this context, Schöllhorn et al. (2009) have more radically argued for the
value of adding noise to the initial conditions of performance to stimulate
learning by forcing the individual to adapt to varying constraints of the
context.
The functional role of variability has been historically overlooked by
researchers because of the types of tasks studied (e.g. pursuit tracking) and
limitations of data collection and analysis tools. For example, Newell and
Slifkin (1998) indicated that the magnitude of performance variability has
traditionally been assessed by the standard deviation or variance over trials;
these statistical indicators attempt to characterize the data distribution and the
amount of noise in a single measurement. The standard deviation measure
indicates the magnitude of variability (i.e. the amplitude, the spatial aspect of
the distribution of performance over trials) but not the structure of system
variability (Newell and Slifkin 1998). Instead, studying the temporal structure
of variability by spectral analysis of noise provides information on its
deterministic or stochastic nature. In this chapter, we overview empirical
research from sports science employing a range of emerging data analysis
tools more suited to examining the structure of movement variability (for a
review of variability analysis in medicine, see Bravi et al. 2011). As we will
demonstrate in a later section, the functional role of movement variability has
typically been explored in performance of ball skills. Our aim for this chapter
is to extend our breadth of understanding by exemplifying how intra-
individual and inter-individual movement variability could play a functional
role in a range of physical activities common to sport, such as different forms
of locomotion, object manipulation and team invasion games.
Functional movement variability in sport

Gait
Human gait is seemingly characterized by smooth, regular, repeating patterns.
The ‘control problem’ in terms of mechanical regulation of gait is that there
are many more muscle actuators involved than independent equations that
define the system (Vaughan 1996). In fact, when analyzed in detail, there are
notable strideto- stride fluctuations even under constant environmental
conditions (Hausdorff 2007) and these fluctuations seem to be a prominent
feature when key parameters such as gait speed and balance are considered
(i.e. Jordan et al. 2006, 2007; van Emmerik and Wagennar 1996). Hausdorff
(2007) suggests that fluctuations in the stride interval exhibit long range,
fractal-like correlations similar to those found in heartrate beat fluctuations.
Put simply, in the short term, the stride interval is dependent on other nearby
cycles but this dependency weakens in a power-law fashion. Interestingly, the
fractal scaling underpinning gait dynamics differs significantly for subgroups
who exhibit much less stable and effective gait patterns (e.g. infants and the
elderly; Hausdorff 2007).
These ideas were demonstrated in a study by Jordan et al. (2007) who
required participants to walk for 12-minute trials at 80%, 90%, 100%, 110%
and 120% of their preferred walking speed. A range of gait-cycle variables
was investigated, including the intervals and lengths of steps and strides and
the impulse from the vertical ground reaction force profile. Detrended
fluctuation analysis revealed the presence of U-shaped long range
correlations in gait-cycle variables. The reduced strength of long-range
correlations at certain locomotion speeds (100–110%) was interpreted as
reflective of enhanced stability and adaptability at preferred speeds (see also
Li et al. 2005).
Button et al. (2010) pointed out that stable attractor states in healthy and
pathological gaits are an important, functional feature of stable coordination
and intra-individual and inter-individual variability amongst these patterns
are discernible with the appropriate analysis tools. It seems that such
differences are more likely to be identified when using time-continuous
analysis tools (i.e. selforganizing Kohonen maps) rather than summative,
discrete statistics (Schöllhorn et al. 2002). For example, a number of
discernible patterns can be detected from the phase relations of pelvis and
thorax segments as gait speed changes (van Emmerik and Wagenaar 1996).
Stability analysis of relative phase and range of motion data have also been
used to identify hysteresis effects in transitions between stable patterns (i.e.
direction of speed dependent).
A particularly distinctive form of gait (i.e. competitive race walking) has
also been considered in relation to functional variability. Donà et al. (2009)
explored the use of functional principal component analysis for assessing and
classifying the kinematics of the knee joint in competitive race walkers.
Functional principal component analysis was applied bilaterally to the sagittal
knee angle data, because knee joint motion is fundamental to race walking
technique. Scatterplots of principal component scores provided evidence of
athletes' technical differences and asymmetries, even when traditional
analysis (mean plus or minus standard deviation curves) was not effective.
Whilst there were certain features, such as the absence of a flight phase, that
were common for all seven participants, principal components provided
indications for the classification of race walkers and identified potentially
important technical differences between higher- and lower-skilled athletes
(Donà et al. 2009).
Bradshaw et al. (2007) were also interested in whether skilled athletes
showed evidence of functional variability, in this case related to the sprint
start. Indeed, high biological movement variability (in comparison with
systematic error) was observed for the joint velocities of ten track sprinters.
Of particular interest, regression analysis indicated that a decrease in ten-
metre sprint time was associated with an increase in the variability of the lead
ankle step velocity.
From this brief overview, it is apparent that functional variability
manifested as stride-to-stride fluctuations is a consistent feature of several
types of gait patterns (e.g. walking, running, sprinting). Movement variability
is most noticeable around transitions in speed, which is an underlying feature
of all types of gait that contribute to balance and stability. It is relevant to
note that only certain kinds of data analysis tools were suited to detecting the
functional fluctuations that subserve gait dynamics.

Breaststroke swimming
Whether locomoting on ground, over an object or through aquatic
environments, it seems that biological organisms exhibit strong preferences
and many global similarities in terms of the cyclical patterns they use to
move. In skilled breaststrokers, one cycle is composed of an alternation of
propulsions (i.e. arm propulsion during the leg glide; leg propulsion during
the arm glide), a brief glide time with the body fully extended and a
synchronization of arm and leg recoveries (Chollet et al. 2004). In
comparison with beginners, during performance, skilled swimmers display a
high level of intra-individual coordination pattern variability, exemplified by
a high intracyclic knee and elbow angular variability and several modes of
arm–leg coordination, depending on swim speed (Seifert et al. 2010). Expert
swimmers need to organize different coordination patterns for each
performance phase. They display an out-of-phase pattern of coordination of
their arms and legs during propulsion (i.e. flexion or extension of a pair of
limbs while the other pair of limbs is fixed in extension), an in-phase
coordination mode during glide (i.e. extension of the arms and legs) and an
anti-phase coordination mode during recoveries (i.e. extension of the arms
during leg flexion) (Seifert et al. 2010; Figure 16.1). In a cycle of 1.5–2.0
seconds, expert swimmers are able to alternate between these three modes of
coordination.
In contrast, owing to their different mix of intentions, perceptions and
actions, learner swimmers typically demonstrate low intra-individual
coordination variability but very high inter-individual coordination variability
(Figure 16.1), notably a bi-stability of the arm–leg coordination modes that
could lead to several intermediate profiles. The first coordination mode
corresponds to an isocontraction of the nonhomologous limbs: that is, the in-
phase muscle contraction of the arms and legs (Baldissera et al. 1991). One
way to enhance system stability is to synchronize the flexion movements of
both arms and legs, as well as the extension movements like an ‘accordion’,
supporting emergence of low intracyclic arm–leg coordination variability
(Figure 16.1). The accordion mode of coordination corresponds to a
superposition of two contradictory actions (Seifert et al. 2010): leg
propulsion during arm recovery and arm propulsion during leg recovery. It is
not mechanically effective and does not provide high swim speed because
each propulsive action is thwarted by a recovery action. As observed in other
studies of interlimb coordination, this coordination mode appears to be the
most stable and the easiest to perform for learners (for an overview, see Kelso
1995).

Figure 16.1 Continuous relative phase (CRP) between elbow and knee through a complete cycle for 24
beginners (left panel) and for 24 expert swimmers (right panel), showing lower inter-individual
variability for experts

The second mode of coordination often observed in beginners is the

flexion movement of one set of limbs that occurs during the extension of the
other set, following the principle of isodirection, which consists of making
movements in the same direction (e.g. arms and legs go forward or backward
on the longitudinal axis; Baldissera et al. 1991). This arm–leg coordination
corresponds to a superposition of the propulsive phases of the arms and legs.
When the superposition of these two propulsions is complete, the arm–leg
coordination pattern resembles the movements of ‘windscreen wipers’ and
shows low intracyclic coordination variability. Finally, although there is bi-
stability in the modes of arm–leg coordination of beginning breaststroke
swimmers, several combinations of these two modes of coordination may
arise for a variety of reasons. From a constraintled approach, the emergent
behaviours of learners are understandable, because they relate to the
overarching need to maintain buoyancy, stability and breathing capacity in
aquatic environments. These suggestions were supported by a cluster analysis
of the movement patterns of 24 beginning swimmers (Seifert et al. 2011b).
The emergent movement patterns were: (i) stable coordination modes, based
on principles of isocontraction and/or isodirection; (ii) indicative of beginners
in an exploratory phase with regard to environmental constraints; and (iii) the
task constraint or goal of the action ‘propel the body forward and fast’
imposed by the teacher or coach was perceived differently by beginners. This
was because their priority was not just to move forward in the water but also
to balance (e.g. to stay in a ventral position), float (e.g. to stay at the water
surface) and breathe (e.g. to avoid bringing hands to the chest in order to
keep the head above water) and see in the aquatic environment.
In summary, these comparisons between the movement patterns of
beginners and expert breaststroke swimmers reveal the influence of different
intentional constraints. The findings stimulate a number of interesting
practical questions, including: (i) whether beginners should be encouraged to
imitate expert swimmers; and (ii), whether coordination patterns observed in
beginning swimmers should be considered as ‘full of errors’ simply because
they differ from a putative ideal movement model based on an analysis of
expert swimmers. A novice's movement pattern may actually reflect different
performance priorities, compared with an expert (i.e. in swimming to
maintain buoyancy, rather than the intention of advancing rapidly through the
water).

Ice-climbing
The way in which intentions, actions and perception of information guide
movement pattern formation has also been shown in a study of ice climbers
(Seifert et al. 2011a). Ice climbing involves climbing with ice tools in each
hand and crampons on each foot on frozen water falls, the properties of
which vary stochastically in shape, steepness, temperature, thickness and ice
density. Since these environmental constraints are neither predictable or
controllable, this task requires successful climbers to use numerous types of
movements (e.g. swinging, kicking and hooking actions) and patterns of
interlimb coordination (e.g. horizontally, diagonally and vertically located
angular positions) during performance by exploiting complex neurobiological
system properties of degeneracy and multistability. For instance, climbers
could either swing their ice tools to create their own holes or hook an existing
hole (owing to the actions of previous climbers or by exploiting the presence
of natural holes), supporting the functional role of intra-individual variability.
Seifert et al. (2011a) examined the performance of beginners and skilled
climbers as they climbed a frozen waterfall. They assessed interlimb
coordination patterns by using the angle between the horizontal line and the
displacement of the heads of the left- and right-hand ice tools for the upper-
limb coordination. Lower-limb coordination patterns corresponded to the
angle between the horizontal line and the displacement of the left and right
crampons (Figure 16.2).
 When both groups of climbers climbed an ice fall with a quasi-vertical
slope (range between 80–90 degrees), beginners showed low levels of intra-
individual movement and interlimb coordination variability, as they varied
their upper-limb and lower-limb coordination patterns much less frequently
and extensively than the experts. As in the study of the novice breaststroke
swimmers, Seifert et al. (2011a) observed patterns of movement that were
indicative of the intentions idiosyncratic to this group. Beginners mostly used
horizontally and diagonally located angular positions (since limb anchorages
are at the same level for the horizontal angle, the arms or legs appear in an in-
phase coordination mode). This highly stable behaviour resembled climbing
up a ladder and led them to maintain a static ‘X’ body position with arms and
legs extended or with arms flexed and legs extended, corresponding to a
freezing of the degrees of freedom of the motor system. Owing to their lack
of attunement to information from properties of the ice fall, novices tended to
swing their ice tools and kick with their crampons more frequently than
experts, in patterns synonymous with achieving deep anchorages, instead of
exploiting existing holes in the ice fall. Like the novice breaststrokers, the
novice ice climbers tended to prioritize stability and security of posture in
interacting with environmental constraints, rather than speed and efficiency
of movement.

Figure 16.2 Angle between horizontal, left limb and right limb (left panel); modes of limb coordination
as regards the angle value between horizontal, left limb and right limb (right panel). The angle between
the horizontal line and the left and right limbs was positive when the right limb was above the left limb
and negative when the right limb was below the left limb

Conversely, expert ice climbers showed high intra-individual movement

and coordination pattern variability as they exploited affordances in the
specific icefall properties. The efficient individual–environment coupling of
the experts was probably predicated on neurobiological system degeneracy
(Edelman and Gally 2001), since they varied the types of movement and the
interlimb coordination modes they exploited to achieve their task goals.
Indeed, the multistability of their complex movement systems allowed them:
(i) to swing the ice tools and to kick the crampons in many different ways –
in horizontal, diagonal, vertical and crossed angular positions (for instance,
crossing the arms is not a natural action but it enabled experts to exploit
information and hook existing holes in the ice fall); and (ii), to hook the ice
tools into already existing holes (i.e. exploiting information on the shape of
the ice fall) and to place the crampons in the holes previously made by their
own ice tools, instead of using repetitive ice-tool swinging and crampon
kicking as observed in beginners.
As in swimming, an individual's behaviours relate to differentiated
intentions and perception of the task and environmental constraints. Notably,
the beginners were functioning independently of the icefall properties,
because they were mainly focusing on keeping their equilibrium, with respect
to gravity, under control. These observations suggest that manipulation of
task constraints by a coach or teacher would enable the beginners to further
interact with the icefall properties in a secure learning environment, to
balance their independency/ dependency to the environmental constraints and
to gradually support the emergence of a wider repertoire of movement and
multi-stability of coordination patterns (Seifert et al. 2011a).

Throwing: Boccia
Goal-directed throwing actions typically need to balance the requirements of
speed and accuracy and the manipulation of task goals is a powerful
constraint on emergent throwing patterns. For example, in throwing tasks, the
alteration of target location simultaneously imposes constraints on both
movement speed (and by extension movement force) and accuracy.
Variability in movement kinematics has been demonstrated in throwing tasks
including javelin (Bartlett et al. 1996), basketball (Button et al. 2003),
underarm (Kudo et al. 2000) and overarm throwing (Wagner et al. 2012) and
a Frisbee®-throwing task (Yang and Scholz 2005).
This body of work confirms that kinematic variability is influenced by task
constraints. In precision throwing tasks, the relative invariance in inter-trial
spatial release points (Hirashima et al. 2002) of thrown objects and decreases
in kinematic variability closer to object release (Wagner et al. 2012) point to
the observed variability in joint parameters serving a functional purpose. For
example, increased variability in distal joints like the elbow and wrist
probably compensate for variations in proximal joints like the shoulder, in
that individuals can use a number of joint configurations to maintain desired
sets of release parameters (e.g. height, speed, and angle) to satisfy various
task constraints.
Compensatory variability has also been observed in a three-dimensional
kinematic analysis of athletes with motor system disorders, such as cerebral
palsy. Cerebral palsy is a clinical term summarizing a group of non-
progressive conditions arising from damage to the brain during early
development. Individuals with cerebral palsy display obvious movement
difficulties in performing tasks of everyday living, typically showing longer,
slower movements, with disturbed speeds and trajectories, particularly at the
elbow, wrist, and fingers (Jaspers et al. 2009).
Boccia is a precision throwing sport in the Paralympics, similar in nature to
lawn bowls or the French game of pétanque. It is played indoors on a firm,
flat surface, where the aim of the game is to land six balls closer to the target
ball than the balls thrown by an opponent. The sport requires a high degree of
muscle control, concentration and tactical awareness. O'Donovan et al.
(2011) recorded the kinematics of four cerebral palsy athletes who had
competed at the Paralympics. In this study, the athletes were required to
throw to four different distances (three metres, five metres, seven metres and
nine metres). Three-dimensional movement analysis showed that the athletes
displayed kinematic variability comparable to typically developed persons
but in the context of distinctly individual movement patterns arising from the
organismic constraints imposed by their type of cerebral palsy. Adopting an
underarm, typically pendulum- like throwing style, all four athletes produced
relatively invariant ball release locations, despite systematically adjusting key
release parameters (primarily adapting release angle and speed to changes in
target distance) to modify the distance thrown. Distal joints (e.g. elbow and
wrist) typically showed increased relative variability compared with proximal
joints (e.g. shoulder), indicating that kinematic variability is necessary to
adapt to task constraints and preserve release parameters in throwing (Figure
16.3).

Figure 16.3 Trajectory plots from O'Donovan et al. (2011)

Southard (2002) showed that unskilled participants demonstrated highly

coupled coordination between the wrist and elbow during overarm throwing
for speed, whereas experts exploited a less rigidly coupled joint synergy to
transfer energy from biomechanical sequencing. Skilled throwers were able
to flexibly and robustly adapt to changes in target size and position, as
reflected by less frequent coordination changes compared with lower-skilled
throwers and scaling of preferred movement patterns (Bartlett et al. 2007).
That being said, even skilled performances show intra- and inter-individual
differences, indicating that, even in homogenous groups such as elite athletes
and people with disabilities, there is no optimal coordination solution for any
given task (Glazier and Davids 2009).

Team invasion games: Association football (soccer)

In comparison with precision-throwing actions, movement variability is much
more obvious, even to the untrained observer, in team invasion games such as
soccer. Rapid technological developments in player tracking systems now
provide movement analysts with an intriguing opportunity of measuring the
coordination dynamics between teams and players in competition. For
example, Frencken et al. (2011) measured player positions at 45 Hz in small-
sided games used in training. Two measures were proposed as candidate
collective variables: a position representing the average position of the
players in the team and some measure of the dispersion or spread of the team.
The most robust average position is the team centroid, calculated from the
mean of the players' positions. Frencken et al. (2011) report strong positive
correlations for the team centroids of the outfield players across each entire
game. Hence, teams as a whole moved together (in-phase coordination)
across the course of a game. Indeed, the analysis by Lames et al. (2010) of
different sports team centroids indicates that similar principles may underpin
the collective organization of teams in invasion games.
Duarte et al. (2012) also found a strong symmetrical relation between the
centroids of the two teams in three-a-side soccer games. Importantly, as the
play proceeded to critical moments (i.e. when a final pass was made to a
shooter) the distance between the team's centroids decreased. The authors
interpreted this feature as a necessary loss in stability between the teams
before a decisive moment, such as a shot or a turnover in possession. The
surface area of the teams was highly variable, although the attacking team
typically covered a greater area of the pitch. Therefore, it seems that either
the surface area of the teams in smallsided games does not capture the
coordination dynamics well (see also Frencken et al. 2011) or that stable
states of play are not a common feature of small-sided games.
Various measures can be used to express the dispersion of sports teams.
For example, Bourbousson et al. (2010) used a ‘stretch index’ to define the
average distance of the players in a basketball team from the team centroid.
Moura et al. (2012) analyzed player trajectory information from eight
Brazilian First Division Championship matches involving 16 different teams.
From visual inspection of the graphs for pairs of teams in all games and
throughout each game, they considered that the two teams' Frobenius norms
(average interpersonal distance for all outfield players), showed anti-phase
relationships, whereas the team surface areas did not. Both dispersion
measures indicated that, when a team lost possession, the team dispersion
decreased; when a team gained possession, the dispersion increased.
Additionally, teams had greater Frobenius norms and surface areas defending
when a shot was made on their goal than when they tackled (P < 0.01). When
attacking, teams had greater Frobenius norms and surface areas when they
were tackled than when they shot at goal. Hence, compression of both the
defending and attacking team's Frobenius norm and surface area was
beneficial to performance in those plays.
Clearly, the movement patterns of individual players in a soccer match are
coupled with the movements of teammates and opponents. These functional
groupings allow a defending team to protect their goal whilst also permitting
the attacking side to probe for space and time to shoot. Nested beneath the
level of grouped actions in competitive football sits a layer of individual
player variability, the unpredictable nature of which contributes to the
exciting, dynamic nature of the world's most spectated sport (Bartlett et al.
2012).

Team invasion games: Field hockey

Although field hockey has received much less consideration than association
football in the sports science literature, interesting data have recently been
reported by Brétigny et al. (2011) concerning inter-individual variability. The
aim of this study was to determine coordination profiles for the field hockey
drive. Nine elite female players accustomed to different playing positions (i.e.
defenders, midfielders and attackers) were asked to play drive shots typical of
a game situation. The high standard deviation values in joint kinematics were
indicative of inter-individual variability; that is, several drive solutions.
Cluster analysis of the upper-body kinematics identified two distinct profiles
among the nine expert players. For the two profiles, the global coordination
pattern of movement was inphase for the right arm and out-of-phase for the
left lead arm, suggesting a segmental sequencing. This interlimb organization
represents the deep layer of the movement and stayed unchanged across the
different field roles. Conversely, the task constraints and, more particularly,
the temporal pressure typically encountered at each field role, affected the
superficial layer of the movement (i.e. kinematic parameters). Notably, the
initial phase of the drive movement (represented by the backswing) was not
universal but, conversely, a ground for technical profiles. Players achieved a
compromise between: (i) long total and relative backswing durations, which
permit adaptations to task constraints and increase drive amplitude to provide
great velocity to the ball but which also increase the risks of opponents'
interceptions; (ii) short total and relative backswing durations, which are a
real advantage in contexts of temporal pressure as the preparation time is
shortened but which can be detrimental in terms of adaptation and ball
velocity.
It was suggested that the short-grip drive was very useful in contexts of
temporal pressure (Bretigny et al. 2011). As the hands are placed lower on
the stick shaft, the lever arm is shorter, resulting in smaller overall amplitudes
and shorter durations in comparison with the classic field-hockey drive.
Preparation time (backswing and downswing) is thus reduced, with no loss in
ball velocity. This stick-holding technique seems best adapted by midfielders
and forwards. Several participants played these roles and formed the
‘temporal-effectiveness group’. They had shorter drives and smaller overall
amplitudes. The velocity variables, especially ball velocity, were not as high
but drive rapidity gave them an important advantage in the context of
temporal pressure. The performance level did not differ among these two
groups but the task constraints from one field role to another varied greatly,
leading to different expert profiles. Similar to the golf swing, for which two
weight transfer styles were identified (Ball and Best 2007), expert field-
hockey drive performance has no universal solution. The participants were
able to exploit the stability and variability of their expert movement patterns
to functionally adapt to dynamic task constraints (exemplified by the
exploitation of the short grip drive in some performance circumstances).
Summary and practical implications
In this brief review of functional movement variability in sport, it has been
demonstrated that both inter- and intra-individual variability is an inherent
feature of performance across the broad continuum of expertise. The
enhanced focus on variability in movement at an individual level has resulted
from changes in theoretical influences in the literature on motor control and
from advances in data collection and analysis techniques. The existence of
inter-individual variability implies that there is no optimal movement pattern
for a given activity and that movement expertise is a reflection of the
performer's ability to adapt to dynamic constraints. Intra-individual
variability shows us that an individual must reciprocally link their actions
with relevant environmental information to consistently achieve performance
outcomes. Characterizing learners as complex, biological systems promotes
awareness by practitioners that a learner's coordination solutions are the
products of self-organization and that periods of movement variability (or
instability) should be valued as part of the learning process (Chow et al.
2007).
To encourage acquisition of functionally relevant coordination solutions
performer–environment interactions should be manipulated through altering
relevant task, environmental and performer constraints. Thereby, constraints
operate on different timescales, which has important implications for the
practitioner's judgment of the learner's rate of progress. When learning a new
coordination pattern, more permanent behavioural changes take longer to
appear than immediate adaptations to task constraints during practice.
Practitioners should understand that some behaviours might represent
transient adaptations to immediate task constraints imposed during practice,
which interact with organismic constraints related to developmental status
(Seifert et al. 2013).
A struggling learner can be viewed as a system that is temporarily trapped
in a stable attractor state that does not correspond well with a behavioural
solution satisfying task demands. As Davids et al. (2008) suggest to
practitioners, a strategy of perturbing the movement system may be necessary
to help the learner to let go of previous movement experiences. Techniques
such as altering task constraints like rules, space, equipment and number of
opponents are useful ways to induce variability in movement and encourage
exploration for alternative information sources and movement solutions. Note
that the learner may need additional encouragement and reassurance at this
important stage, as performance could fluctuate as a consequence of the
perceptual-motor reorganization.
In this chapter, we have demonstrated the value of employing
individualized analyses in research on movement to gain a clearer picture of
how performers exploit variability. As more studies employ methods such as
coordination profiling and cluster analysis, researchers are becoming
increasingly convinced that varied movement trajectories emerge from the
interplay among the specific task, environmental and organismic constraints
unique to each situation. This is particularly apparent within sport, where
such factors change frequently and unexpectedly. As we have demonstrated,
expert performers are increasingly recognized as having an ability to
continually adapt their techniques as perceptual demands change. The
mechanisms by which humans progress to this level of control as a function
of learning or relearning provide a fruitful focus for future research.
References
Baldissera, F., Cavalleri, P., Marini, G. and Tassone, G. (1991) Differential
control of inphase and anti-phase coupling of rhythmic movements of
ipsilateral hand and foot. Experimental Brain Research, 83: 375–80.
Ball, K. A. and Best, R. J. (2007) Different centre of pressure patterns within
the golf stroke I: cluster analysis. Journal of Sports Sciences, 25: 757–70.
Bartlett, R., Muller, E., Lindinger, S., Brunner, F. and Morriss, C. (1996)
Three-dimensional evaluation of the kinematic release parameters for javelin
throwers of different skill levels. Journal of Applied Biomechanics, 12 (1):
58–71.
Bartlett, R., Wheat, J. and Robins, M. (2007) Is movement variability
important for sports biomechanists? Sports Biomechanics, 6 (2): 224-43.
Bartlett, R., Button, C., Robins, M., Dutt Mazumder, A. and Kennedy, G.
(2012) Analysing team coordination patterns from player movement
trajectories in soccer: methodological considerations. International Journal
of Performance Analysis in Sport, 12 (2): 398–424.
Bernstein, N. A. (1996) On dexterity and its development, in M. Latash and
M. T. Turvey (eds) Dexterity and its Development. Mahwah, NJ: Lawrence
Erlbaum Associates, pp. 1–244.
Biryukova E, and Bril, B. (2002) Bernstein et le geste technique. Revue
d'anthropologie des Connaissances, 2: 49–68 [French].
Bourbousson, J., Seve, C. and McGarry, T. (2010) Space-time coordination
dynamics in basketball: Part 2. The interaction between the two teams.
Journal of Sports Sciences, 28: 349–58.
Bradshaw, E. J., Maulder, P. S. and Keogh, J. W. (2007) Biological
movement variability during the sprint start: performance enhancement or
hindrance? Sports Biomechanics, 6: 246–60.
Bravi, A., Longtin, A. and Seely, A. J. E. (2011) Review and classification of
variability analysis techniques with clinical applications. Biomedical
Engineering Online, 10: 90; doi: 10.1186/1475-925X-10-90. Available online
at www.biomedical-engineeringonline.com/content/10/1/90 (accessed 12
June 2013).
Bretigny, P., Leroy, D., Button, C., Chollet, D. and Seifert, L. (2011)
Coordination profiles of the expert field hockey drive according to field roles.
Sports Biomechanics, 10: 339–50.
Button, C., MacLeod, M., Sanders, R. and Coleman, S. (2003) Examining
movement variability in the basketball free-throw action at different skill
levels. Research Quarterly for Exercise and Sport, 74 (3): 257–69.
Button, C., Moyle, S. and Davids, K. (2010) Comparison of below-knee
amputee gait performed overground and on a motorized treadmill. Adapted
Physical Activity Quarterly, 27: 96–112.
Chollet, D., Seifert, L., Leblanc, H., Boulesteix, L. and Carter M. (2004)
Evaluation of the arm-leg coordination in flat breaststroke. International
Journal of Sport Medicine, 25: 486–95.
Chow, J. Y., Davids, K., Button, C., Shuttleworth, R., Renshaw, I. and
Araújo, D. (2007) The role of nonlinear pedagogy in physical education.
Review of Educational Research, 77: 251–78.
Davids, K. and Glazier, P. (2010) Deconstructing neurobiological
coordination: the role of the biomechanics-motor control nexus. Exercise and
Sport Science Reviews, 38: 86–90.
Davids, K., Glazier, P., Araújo, D. and Bartlett, R. M. (2003) Movement
systems as dynamical systems: the role of functional variability and its
implications for sports medicine. Sports Medicine, 33: 245–60.
Davids, K., Button, C. and Bennett, S. J. (2008) Dynamics of Skill
Acquisition: A Constraints-Led Approach. Champaign, IL: Human Kinetics.
Donà, G., Preatoni, E., Cobelli, C., Rodano, R. and Harrison, A. J. (2009)
Application of functional principal component analysis in race walking: an
emerging methodology. Sports Biomechanics, 8 (4): 284–301.
Duarte, R., Araújo, D., Freire, L., Folgado, H., Fernandes, O. and Davids, K.
(2012) Intraand inter-group coordination patterns reveal collective behaviors
of football players near the scoring zone. Human Movement Science, 31 (6):
1639–51.
Edelman, G. M. and Gally, J. A. (2001) Degeneracy and complexity in
biological systems. Proceedings of the National Academy of Sciences of the
USA, 98: 13763–8.
Frencken, W., Lemmink, K., Delleman, N. and Visscher, C. (2011)
Oscillations of centroid position and surface area of soccer teams in small-
sided games. European Journal of Sport Science, 11: 215–23.
Glazier, P. S. and Davids, K. (2009) The problem of measurement
indeterminacy in complex neurobiological movement systems. Journal of
Biomechanics, 42 (16): 2694–6.
Hausdorff, J. M. (2007) Gait dynamics, fractals and falls: finding meaning in
the stride-tostride fluctuations of human walking. Human Movement Science,
26 (4): 555–89.
Hirashima, M., Kadota, H., Sakurai, S., Kudo, K. and Ohtsuki, T. (2002)
Sequential muscle activity and its functional role in the upper extremity and
trunk during overarm throwing. Journal of Sports Sciences, 20 (4): 301–10.
Jantzen, K. J., Oullier, O. and Scott Kelso, J. A. (2008) Neuroimaging
coordination dynamics in the sport sciences. Methods, 45: 325–35.
Jaspers, E., Desloovere, K., Bruyninckx, H., Molenaers, G., Klingels, K. and
Feys, H. (2009) Review of quantitative measurements of upper limb
movements in hemiplegic cerebral palsy. Gait and Posture, 30 (4): 395–404.
Jordan, K., Challis, J. H. and Newell, K. M. (2006) Long range correlations
in the stride interval of running. Gait and Posture, 24: 120–5.
Jordan, K., Challis, J. H. and Newell, K. M. (2007) Walking speed influences
on gait cycle variability. Gait and Posture, 26: 128–34.
Kelso, J. A. S. (1995) Dynamic Patterns: The Self-Organization of Brain and
Behavior. Cambridge, MA: MIT Press.
Kudo, K., Tsutsui, S., Ishikura, T., Ito, T. and Yamamoto, Y. (2000)
Compensatory coordination of release parameters in a throwing task. Journal
of Motor Behavior, 32 (4): 337–45.
Lames, M., Erdmann, J. and Walter, F. (2010) Oscillations in football: order
and disorder in spatial interactions between the two teams. International
Journal of Sport Psychology, 41: 85–6.
Li, L., Haddad, J. and Hamill, J. (2005) Stability and variability may respond
differently to changes in walking speed. Human Movement Science, 24 (2):
257–67.
Moura, F. A., Martins, L. E. B., Anido, R. D., De Barros, R. M. L. and
Cunha, S. A. (2012) Quantitative analysis of Brazilian football players'
organisation on the pitch. Sports Biomechanics, 11: 85–96.
Newell, K. M. (1986) Constraints on the development of coordination, in: M.
G. Wade and H. T. A. Whiting (eds) Motor Development in Children:
Aspects of Coordination and Control. Dordrecht, Netherlands: Martinus
Nijhoff, 341–61.
Newell, K. M. and Slifkin, A. B. (1998) The nature of movement variability,
in: J. P. Piek (ed.) Motor Behavior and Human Skill: A Multidisciplinary
Perspective. Champaign, IL: Human Kinetics, pp. 14–60.
O'Donovan, D. P. (2011) Kinematics of throwing in paralympic boccia
athletes with cerebral palsy. Unpublished Masters thesis, University of
Otago. Available online at: http://otago.ourarchive.ac.nz/handle/10523/2009
Schöllhorn, W. I., Nigg, B. M., Stefanyshyn, D. J. and Liu, W. (2002)
Identification of individual walking patterns using time discrete and time
continuous data sets. Gait and Posture, 15: 180–6.
Schöllhorn, W. I., Mayer-Kress, G., Newell, K. M. and Michelbrink, M.
(2009) Time scales of adaptive behavior and motor learning in the presence
of stochastic perturbations. Human Movement Science, 28: 319–33.
Seifert, L., Leblanc, H., Chollet, D. and Delignières, D. (2010) Inter-limb
coordination in swimming: effect of speed and skill level. Human Movement
Science, 29: 103–13.
Seifert, L., Wattebled, L., L'Hermette, M. and Hérault, R. (2011a) Inter-limb
coordination variability in ice climbers of different skill level. Education and
Physical Training in Sport, 1: 63–8.
Seifert, L., Leblanc, H., Herault, R., Komar, J., Button, C. and Chollet, D.
(2011b) Interindividual variability in the upper-lower limb breaststroke
coordination. Human Movement Science, 30: 550–65.
Seifert, L., Button, C. and Davids, K. (2013) Key properties of expert
movement systems in sport: an ecological dynamics perspective. Sports
Medicine, 43 (3): 167–78.
Southard, D. (2002) Change in throwing pattern: critical values for control
parameter of velocity. Research Quarterly for Exercise and Sport, 73 (4):
396–407.
Summers, J. J. and Anson, J. G. (2009) Current status of the motor program:
revisited. Human Movement Science, 28: 566–77.
Todorov, E. and Jordan, M. I. (2002) Optimal feedback control as a theory of
motor coordination. Nature Neuroscience, 5 (11), 1226–35.
van Emmerik, R. E. A. and Wagenaar, R. C. (1996) Effects of walking
velocity on relative phase dynamics in the trunk in human walking. Journal
of Biomechanics, 29 (9): 1175–84.
Vaughan, C. L. (1996) Are joint torques the Holy Grail of human gait
analysis? Human Movement Science, 15 (3): 423–43.
Wagner, H., Pfusterschmied, J., Klous, M., von Duvillard, S. P. and Müller,
E. (2012) Movement variability and skill level of various throwing
techniques. Human Movement Science, 31 (1): 78–90.
Wolpert, D. M., Doya, K. and Kawato, M. (2003) A unifying computational
framework for motor control and social interaction. Philosophical
Transactions of the Royal Society of London, 358: 593–602.
Yang, J.-F. and Scholz, J. P. (2005) Learning a throwing task is associated
with differential changes in the use of motor abundance. Experimental Brain
Research, 163 (2): 137–58.
17 Individual pathways of change
in motor learning and
development
Yeou-Teh Liu and Karl M. Newell
Individual differences are observed in motor learning and development;
however, studies typically analyze the averaged data over groups of
participants. Based on the ergodic theorem of mathematics, it is clear that the
processes of human motor learning and development are non-ergodic as
reflected in non-stationarity and heterogeneity. Given this, it is necessary to
analyze the intra-individual data to unravel the characteristics of the change
processes. We present a landscape model of multiple timescales as a
framework to describe the individual pathways of change in motor learning
and development from a dynamical systems perspective. Examples of
individual differences, including those in the context of sport skills, are
provided from the evolving attractor landscape of multiple timescales.
Motor learning and development are individual but related processes that
are reflected in changes of motor ability. Studies of motor learning and
development typically examine the motor performance of groups of people
over a period of time in order to observe changes due to the
learning/development processes. Although individual differences have been
acknowledged in the motor skills acquisition literature (e.g. Ackerman 1987;
Fleishman 1978), these differences are often masked, if not eliminated
entirely, by way of averaging data over the group (Liu et al. 2006).
One general assumption of averaging data is that the individual differences
that appear in the motor performance are from the incidental random-like
movement variability that does not reflect the stable performance of the
participants. In this view, averaging over the participants will minimize or
eliminate this individual noise-like variability and allow the general trend of
learning and development to be revealed more robustly. The results from
analyzing the averaged data, however, may only reflect a constructed learning
trend of a group of participants, because the individual learning
characteristics tend to be lost in the averaging process. Moreover, averaging
learning data can produce a general performance trend that is not present in
the data of any participant. Thus, the challenge is to identify the individual
characteristics of interest to be analyzed before averaging the performance
data.
In this chapter, we focus on the individual pathways of change in motor
learning and development from a dynamical systems perspective. We use an
epigenetic landscape model as the theoretical framework for examining the
performance change at the individual level. We outline parallels in sport to
the individual differences identified in a dynamical systems approach to
laboratory experiments. Our approach limits itself to individual difference
phenomena that are related to the dynamics of the behaviour – we do not
draw on individual differences in motor learning and development that have
been interpreted by information processing frameworks (Ackerman 1987;
Fleishman 1978).
Motor learning and development processes are non-ergodic
Human motor learning and development systems are considered to be
reflections of deterministic and stochastic processes. From the classic ergodic
theorems of mathematics, any measurable stochastic processes have to meet
the stationary and homogeneous conditions to guarantee that the inter-
individual structure and the intra-individual structure of the processes are
equivalent, so that the result of an analysis from one structure can be
generalized to the other (Molenaar 2004). A non-ergodic process does not
have an a priori relationship between the interindividual structure and the
intra-individual structure, making the inference from individual to group data
challenging.
The stationary condition for a Gaussian stochastic process implies a
constant mean and higher-order (moment) measures without a trend or
periodicity. In motor learning and development, however, the mean function
and variance are typically changing over the learning/development sequence (
e.g. Adams 1952), hence violating the stationary condition of ergodicity. The
homogeneous condition dictates that each individual in a sample population
possesses the same dynamics of change. When the condition of homogeneity
is violated, the result obtained from the group analysis may not be applicable
at the individual level. Motor learning and development reflect the product of
the evolving individual interacting with the many contexts of the
environment to realize the task demands. Studies have shown different
individual patterns of learning that are due to individual differences (e.g. Liu
et al. 2006).
Thus, motor learning and development processes tend to be non-stationary
and heterogeneous within the observable conditions or groups, leaving them
nonergodic in nature in the technical sense (Molenaar 2004). This is because
motor learning and development by definition are characterized by the
change of behaviour over time and, therefore, by definition are non-ergodic.
It follows that it is necessary to analyze the intra-individual data (for
example, the time series of individual performers) to unravel the
characteristics of these processes.
The likelihood of individual pathways of change in motor learning and
development will be magnified in the performance of certain motor tasks,
including sport tasks. This is because there is degeneracy of the perceptual-
motor system in realizing task goals. That is, the same task goal can be met
with different solutions of movement coordination and control. We postulate
that the nature of the particular sports skill influences the number of potential
solutions as reflected in within and between subject variability. Intuitively, it
would seem that the maximal performance tasks of sport (e.g. weight lifting,
track and field) will tend to have fewer viable solutions than other tasks, such
as soccer and the mid-range passing of the ball in the context of the game.
This task-related difference in the potential range of movement solutions will
influence the presence of individual differences in the pathways of change in
motor learning and development.
Multiple processes of performance change
The studies of motor learning and development, although they may focus on
different populations (adults vs. children and elderly) and different tasks
(phylogenetic vs. ontogenetic movements), share the common goal of
characterizing behavioural change over time. In addition to the comparative
paradigm where pretest and post-test are performed and analyzed to examine
the change of behaviour, the learning (performance) curve has also been a
focus of research attention. For nearly two decades, the predominant
description was the ‘power law’ function (Newell and Rosenbloom 1981),
which covered a range of task outcomes and context domains but considered
to be the single function that characterized learning behaviour (see Heathcote
et al. 2000; Newell et al. 2001). In fact, many forms of mathematical
functions have been proposed to account for the different shapes of
performance curves over time (see Mazur and Hastie 1978; Thurstone 1919;
Welford 1987). These different shapes of performance curves may be due to
different task constraints that were imposed on the performers and reflect
different levels of change, such as continuous scaling improvement of the
task or discontinuous jumps between coordination patterns.
Additionally, the adaptive phenomena such as warm-up, fatigue, inhibition
and noise which were frequently observed in the learning curve data but
usually discarded or ignored for analysis also contribute to the overall change
of performance in learning and development. These adaptive features in
performance are also most obvious to the teachers and coaches but are often
misinterpreted in terms of their role within learning. The different levels of
learning and development, as well as the adaptive phenomena observed
during the performance sequences, reflect the multiple contributions of the
evolving set of dynamical subsystems to learning and development, each with
its own changing time scale. The power function that is considered to be
scale free (infinite timescales) could be the result of the averaging processes
over different learning curves (individual learners) and/or over different
segments of learning curves (Newell et al. 2001).
Dynamical systems and landscape metaphor of change
The use of an evolving landscape as a metaphor to describe the behavioural
change in learning and development can be traced back to Waddington's
(1957) epigenetic landscape. Waddington was interested in the genetic
expression of the embryo and the landscape model was a metaphor for
considering the dynamics of developmental growth and change. In
Waddington's landscape model, the elevations of the landscape depict the
stability of each developing behaviour (Figure 17.1). The two axes
perpendicular to the vertical axis of elevation represent the emergence of
particular activities along the developmental time. For example, the only
stable pattern of an infant at birth might be the ‘lying down with arms and
legs curled’ pattern. Additional stable patterns emerge later on in the infant's
motor development sequence, such as lying in the prone position with chin up
and then sitting and crawling. Although no specifics of the landscape were
provided in Waddington's model, a dynamical systems interpretation of the
epigenetic landscape has been outlined for the case of prone progression in
infants (Newell et al. 2003).

Figure 17.1 Waddington's (1957) schematic of the epigenetic landscape

The concept of the landscape model is individually based. The landscape

of motor behaviour describes the ability to perform different motor activities
and the height of the landscape at each activity location specifies the stability
of the performance. For example, in the HKB (Haken-Kelso-Bunz) model of
the finger oscillation tasks, while the 0-degree in-phase and 180-degree anti-
phase relations are both stable at low frequencies, only the in-phase relation
will remain stable and be observed when frequency is increased to cross the
threshold for transition (Figure 17.2). However, the specific value for the
threshold is individually dependent; that is, the value of b/a in Figure 17.2,
where the concave shapes at ± 180 degrees dissolve into a convex shape,
depends on the specific performer.
In the landscape example of Figure 17.2, the elevations of the landscape
determine the stability regions in the landscape and the motor performance
follows the descending slope toward the nearby fixed-point attractor. If, now,
the task is to perform a 90-degree relative phase, under the current landscape
structure of Figure 17.2, there is no attractor at the 90-degree region;
therefore, a restructuring of the landscape to form a ‘well’ at the 90-degree
area is required. Learning a new coordination pattern, such as oscillating two
fingers in 90-degree relative phase, would be modelled as changing the motor
landscape (see Zanone and Kelso 1997). From the dynamical systems
approach, new coordination learning involves a phase transition (Liu et al.
2006) and, although sufficient practice usually leads to a transition to the new
coordination pattern, the number of practice trials (practice time) required to
induce a transition is not constant among individual learners.
Figure 17.2 The landscape dynamics of the basic equation of the HKB (Haken-Kelso- Bunz) model
(Haken et al. 1985), with the same parameters as in Kelso (1995, Figure 2.7)

Figure 17.3 shows the hypothetical landscape change for the 90-degree
relative- phase learning (Newell et al. 2009). The top panel of Figure 17.3
depicts the landscape at the beginning stage of the 90-degree relative-phase
learning where, although the trajectory from initial position C shows a
temporarily stable target pattern of 90-degree relative phase, the trajectories
from all initial positions still tend toward the stable in-phase pattern.
Continuing practice results in further change in the layout of the landscape.
The middle panel of Figure 17.3 shows the landscape at transition where
selected initial positions will tend to the stable 90- degree relative phase
target pattern but the majority of the initial positions still converge to the in-
phase pattern. The landscape continues to deform with further practice and
the probability of converging to the stable 90-degree relative phase from any
initial position is greatly increased. Practice contributes to the change of
landscape and the change of landscape creates a new stable coordination
pattern.
Figure 17.3 (a) Landscape of learning the 90-degree phase task of the HKB model; at the beginning of
practice (C = 0.4) only temporary stabilization of the target phase x0 = 0.25 can be achieved when
starting from special initial conditions close to C; (b) right at the transition (C = 0.425) the target phase
x0 = 0.25 shows one-sided stability: initial conditions close to C will be attracted to the new attractor.
Note that, in this situation, the system is very sensitive to noise perturbations; (c) after sufficient
practice (C = 0.525), all initial conditions close to the target attractor x0 = 0.25 will converge to the
fixed point (reproduced from Newell et al. 2001)
Averaging masks the individual learning processes
The existing structure of the motor landscape determines the ability to
perform particular coordination patterns (whether there is an attractor at the
particular location); performing a particular coordination pattern will also
modify the structure of the motor landscape. When the current landscape
conforms to the required coordination pattern of the target task, the learning
curve usually shows improvement on the precision and/or stability of the
performance and the learning dynamics may be modelled as approaching to a
fixed-point attractor (Newell et al. 2009). Theoretically, the dynamics close
to the fixed point show an exponential function that provides a single
timescale as the basis of improvement rate. However, if the learning curves
are averaged among different individual learners, it is likely that the resulting
improvement rate does not capture any individual's learning rate. The
exponential nature of the learning curve is also lost in the averaging process.
Within the class of fixed-point-attractor learning, although the dynamics of
learning can be approximated by an exponential function, there may be other
dynamics involved in the process. Based on the frameworks of multiple
timescales of learning and development, a two-timescales landscape model of
a sensory–motor learning task has been established to examine the co-
existing adaptation and learning dynamics (Newell et al. 2009). The two-
timescale model of adaptation and learning was derived from a
decomposition of the performance dynamics into separate adaptation and
learning processes.
Figure 17.4 shows the graphical illustration of the two-timescale model.
The fast timescale, indicating a large change of performance level, describes
the adaptation phenomenon that is observed at the beginning of each practice
session. The slow timescale, on the other hand, representing the small
changes over time, reflects the persistent change of learning. Other processes
of different timescales, such as fatigue/inhibition, may also be identified and
included to make the model more comprehensive (e.g. Newell et al. 2010).
These different processes dominate different parts of the practice sequence
and will be masked if averaging over the sequence of trials is implemented in
data processing. The contemporary work in neuroscience has also shown
multiple processes to memory consolidation that each have their own
timescale of change on the performance dynamics (Kandel 2006; Shadmehr
and Holcomb 1997; Tse et al. 2007). The averaging technique has been
widely practised in the processing of learning data. However, the analyses
based on averaged data do not reflect well the characteristics of the individual
learning performance.
Self-organization in learning and development
Practice is one of the most important factors in motor skill learning. Through
practice, learners try to produce the target movement under the organismic,
environmental and task constraints (Newell 1986). Practice sessions provide
the opportunity for the learners to organize the movement systems and
perform the target movement repeatedly, the movement subsystems interact
with one another en route to the equilibrium state of the system.
Consequently, the emergent movement performance reflects the results of
self-organization process of the complex dynamical systems.

Figure 17.4 A two-timescale landscape model associated with Snoddy's (1926) score data (black dots)
as elevation levels. The four clusters correspond to the four practice sessions. The x behavioural
variable corresponds to the slow timescale (shallow dimension), whereas the y variable corresponds to
the fast timescale (steep dimension)

Self-organization is a characteristic of the complex dynamical systems

where the subsystems within and between levels interact with one another
without central/external commands. Individuals learning to perform a new
coordination pattern tend to undergo the self-organization processes. For
example, in the study of learning to ride a pedalo (Chen et al. 2005), subjects
practised 350 trials of pedalo task over seven days. The participants were
only informed about the goal of the pedalo task, which was to ride a
designated distance as fast and as smooth as possible, no additional
instruction as to how to ride the pedalo was given. The participants had to
organize their movement systems during practice trials to learn the task.
The results of data analyses showed that, although the movement time and
movement smoothness measures indicated a practice-day effect
(improvement over seven days), the principal components analysis on the
individual participants showed that different participants had a different
number of principal components before and after seven days of practice and
the variance accounted for by these components was also different. These
data indicate that participants all improved their pedalo performance with
practice but the movement coordination patterns and the way the movements
were changed was not the same for each individual participant. These results
suggest that individual learners underwent selforganization process during
practice in order to meet the requirement of the task dynamics. The
interaction between the movement systems and the instrument (pedalo) and
within the movement subsystems are the processes of selforganization.
Individual differences and the landscape model of multiple
timescales
There are multiple levels of analysis to human movement that evolve over
time. The learning of tracing a particular shape (start tracing), learning to
produce a specific relative phase (90 degree) or the lifetime evolution of the
postural and locomotion patterns can all be captured with the landscape
model. The general idea is that the shape of the landscape (that is, the
elevation and slope of each relevant location of the landscape that represents
a particular quality of special interest) describes the stability of the movement
performances. The different levels of analyses are reflected in the different
timescales of the landscape models and these landscapes are continuously
evolving.
As noted previously, learning and development are individual processes
and the individual differences have been observed and reported in various
relevant literatures. Based on the landscape metaphor of movement learning
and development, the individual differences may come from two sources: the
initial state and the rate of deformation. If the individual differences reside in
the initial state, there is a ‘well (attractor)’ that exists in the hypothetical
movement landscape of the performer, it is the incidental choice of the initial
attempt that differentiate the outcome performances (see Newell and
McDonald 1992). For example, in a motor learning study where the
combined spatial and temporal feedback were given to the learners to
improve the mirror tracing task, the results of the cluster analysis identified
three groups of different initial strategies – the movement time group, the
spatial error group and the mixed group (King et al. 2012). Although the final
performances among the three groups did not show any significant
differences, these different initial strategies were sustained throughout the
subsequent exploration of the perceptual work space (Figure 17.5).
For the type of differences where the source comes from the rate of
landscape deformation, it is assumed that, for the individual learner, the
elevation is high at the location of the landscape where the target movement
is specified; the change of the landscape is required to invoke the transition of
the movement dynamics. In addition to the structure of the landscape at the
initial stage of learning, the ‘hardness’ of the landscape also contributes to the
rate of change of the landscape. In a rollerball learning study, where
participants were asked to learn to accelerate a top encapsulated in a seven-
centimetre diameter spherical shell, seven days of practice were given to all
the participants to learn the task (Liu et al. 2006). Among the eight successful
participants who learned to accelerate the rollerball by the end of the
experiment, the number of trials used to reach the criterion of success ranged
from 10 to 300. There were three participants who did not reach the goal of
learning to accelerate the rollerball within seven days of practice.
Figure 17.5 Initial and final performance of the mirror tracing task; (A) breakdown of one-dimensional
performance score measure into movement time and spatial error (movement time) components. Filled
and open symbols indicate outcome score on trial 1 and trial 50, respectively. Triangle, circle and
square symbols reflect movement time (MT), mixed (MIXED) and spatial error (SE) group assignment
(see text for detailed explanation); (B) performance score on trial 1 and trial 50 as a function of group;
error bars indicate standard error

There are additional examples of individual differences in the learning

literature, as well as in practical situations such as athletic training, skill
learning and infant motor development (e.g. Chow et al. 2008). The
landscape model of multiple timescales provides a theoretical framework for
systematically examining the source and the influence of individual
differences in motor learning and development.
Individual differences are ubiquitous in every aspect of human
performance. This feature is especially significant in the processes of learning
and development. Not only in the data analysis procedure in research, where
learning and development data are often treated as a group, in practical
settings such as activity classes and sports training, skills and performances
are also taught and evaluated under a group criterion. The multiple processes
of change provide a base for understanding the source of individual
differences. The challenge for teachers and coaches is to identify and
recognize the consequences of these processes to better assist the
development of motor skills.
Concluding comments
We have shown in this chapter that the motor learning and development
processes are non-ergodic (Molenaar 2004). Therefore, it is necessary to
analyze the intraindividual data of motor learning and development in order
to reveal the characteristics of these processes. We have also used the
landscape model of multiple timescales from dynamical systems theory
(Newell et al. 2001) as the framework to emphasize the importance of
examining the individual pathways of learning and development in motor
behaviour. Over 100 years of research on human motor behaviour have
mostly analyzed learning data over groups of participants. In the light of the
individual nature of the learning and development of movement, further
research should examine the individual pathways of learning and
development to obtain a better understanding of these processes and
contribute to the theories and practices of related fields.
Acknowledgements
This paper was supported in part by NSF grant 0848339.
References
Ackerman, P. L. (1987) Individual differences in skill learning: an integration
of psychometric and information processing perspectives. Psychological.
Bulletin, 102: 3–27.
Adams, J. A. (1952) Warm-up decrement in performance on the pursuit rotor.
American Journal of Psychology, 65: 404–14.
Chen, H.-H., Liu, Y.-T., Mayer-Kress, G. and Newell, K. M. (2005) Learning
the pedalo locomotion task. Journal of Motor Behavior, 37: 247–56.
Chow, J. Y., Davids, K., Button, C. and Rein, R. (2008) Dynamics of
movement patterning in learning a discrete multiarticular action. Motor
Control, 12: 219–40.
Fleishman, E. A. (1978) Relating individual differences to the dimensions of
human tasks. Ergonomics, 21: 1007–19.
Haken, H., Kelso, J. A. S. and Bunz, H. A. (1985) Theoretical model of phase
transitions in human hand movements. Biological Cybernetics, 51: 347–56.
Heathcote, A., Brown, S. and Mewhort, D. J. K. (2000) The power law
repealed: the case for an exponential law of practice. Psychonomic Bulletin
and Review, 7: 185–207.
Kandel, E. R. (2006) In Search of Memory: The Emergence of a New Science
of the Mind. New York: Norton.
King, A. C., Ranganathan, R. and Newell, K. M. (2012) Individual
differences in the exploration of a redundant space-time motor task.
Neuroscience Letters, 529: 144–9.
Liu, Y.-T., Mayer-Kress, G. and Newell, K. M. (2006) Qualitative and
quantitative change in the dynamics of motor learning. Journal of
Experimental Psychology: Human Perception and Performance, 32: 380–93.
Mazur, J. E. and Hastie, R. (1978) Learning as accumulation: a
reexamination of the learning curve. Psychological Bulletin, 85: 1256–74.
Molenaar, P. C. M. (2004) A manifesto on psychology as idiographic
science: bringing the person back into scientific psychology, this time
forever. Measurement: Interdisciplinary Research and Perspective, 2: 201–
18.
Newell, A. and Rosenbloom, P. S. (1981) Mechanisms of skill acquisition
and the law of practice, in J. R. Anderson (ed.) Cognitive Skills and Their
Acquisition. Hillsdale, NJ: Erlbaum, pp. 1–55.
Newell, K. M. (1986) Constraints on the development of coordination, in M.
G. Wade and H. T. A. Whiting (eds) Motor Skill Acquisition in Children:
Aspects of Coordination and Control. Amsterdam: Martinus Nijhoff, pp.
341–60.
Newell, K. M. and McDonald, P. V. (1992) Searching for solutions to the
coordination function: learning as exploratory behavior, in G. E. Stelmach, J.
Requin (eds) Tutorials in Motor Behavior II. Advances in Psychology, Vol.
87. Amsterdam: Elsevier, pp. 517–32.
Newell, K. M., Liu, Y-T. and Mayer-Kress, G. (2001) Time scales in motor
learning and development. Psychological Review, 108: 57–82.
Newell, K. M., Liu, Y-T. and Mayer-Kress, G. (2003) A dynamical systems
interpretation of epigenetic landscapes for infant motor development. Infant
Development and Behavior, 26: 449–72.
Newell, K. M., Mayer-Kress, G., Hong, S. L. and Liu, Y. T. (2009)
Adaptation and learning: characteristic time scales of performance dynamics.
Human Movement Science, 28: 655–87.
Newell, K. M., Mayer-Kress, G., Hong, S. L. and Liu, Y-T. (2010)
Decomposing the performance dynamics of learning through time scales, in
P. C. M. Molenaar and K. M. Newell (eds) Individual Pathways of Change in
Learning and Development. Washington, DC: American Psychological
Association, pp. 71–86.
Shadmehr, R. and Holcomb, H. H. (1997) Neural correlates of motor memory
consolidation. Science, 277: 821–5.
Thurstone, L. L. (1919) The learning curve equation. Psychological
Monographs, 26: 1–51.
Tse, D., Langston, R. F., Kakeyama, M., Bethus, I., Spooner, P. A., Wood, E.
R., Witter, M. P., Witter, M. P. and Morris, R. G. M. (2007) Schemas and
memory consolidation. Science, 316: 76–82.
Waddington, C. H. (1957) The Strategy of the Genes. London: George Allen
and Unwin.
Welford, A. T. (1987) On rates of improvement with practice. Journal of
Motor Behavior, 19, 401–15.
Zanone, P. G. and Kelso, J. A. S. (1997) The coordination dynamics of
learning and transfer: collective and component levels. Journal of
Experimental Psychology: Human Perception and Performance, 23: 1454–
80.
18 A constraints-based approach
to the acquisition of expertise in
outdoor adventure sports
Keith Davids, Eric Brymer, Ludovic
Seifert and Dominic Orth
Previous work has shown that changes in behaviour, as a function of
learning, emerge as a consequence of continuous interactions between
learners and a performance environment (Chow et al. 2011; Davids et al.
2008). These interactions, with other individuals and key objects, surfaces
and events during learning, need not be the result of programmed, formalized
instructions but can occur through unstructured, exploratory activity (Davids
et al. 2012). As a result of their continuous interactions with the performance
environment, learners become adept at exploiting sources of information
available as properties of the environment for regulating and changing their
behaviours.
Research has also demonstrated several important properties to emerge in
nonlinear complex systems under the effect of constraints, including: system
multi- and metastability, adaptive variability, degeneracy and attunement to
affordances for action (Araújo et al. 2004; Davids et al. 2008; Hristovski et
al. 2006a; Kelso 1995; Seifert et al. 2011; Seifert et al. 2013). This chapter
outlines a constraints-based framework for explaining how these key
neurobiological system properties may influence processes of learning and
performance in the context of outdoor adventure sports. With reference to
current empirical work, we discuss how functional patterns of behaviour
might emerge from engagement in outdoor adventure sports.
The constraints-based model focuses on change in individuals over
different timescales and is generically suited to the study of learning and
performance in a variety of outdoor adventure sports contexts (Brymer and
Renshaw 2010, Brymer and Davids 2013). This is because it has a strong
focus on the relationship between the individual performer, the task and the
environment (social and physical) as the appropriate scale of analysis for
understanding behavioural change on different timescales. There is an
important and functional role of movement pattern variability in supporting
the necessary short-term performance adaptations required between and
within skilled individuals in outdoor adventure sports and physical activities
(Davids et al. 2003, 2006). Over the longer timescale of learning,
understanding of these complex system properties has signalled that the
acquisition of functional performance behaviours does not emerge from
repetitive and imitative practice to gradually reduce a perceived ‘void’
differentiating the behaviours of learners and a putative expert model. Rather,
learners' behaviours need to be understood in terms of their functionality in
achieving their intentions. For example, in kayaking or canoeing, learners
need to discover functional patterns of behaviour that provide stability of the
system formed by each individual and the interaction with the boat, the
paddle and the water during exploratory practice in safe settings, such as
small lakes and slow-moving rivers. More advanced exploratory practice can
be undertaken as the ecological constraints of performance are manipulated
by a coach exposing learners to more dynamic environmental settings, as
might be found in open-water settings, oceans or fastmoving rivers. A
constraints-based framework theoretically verifies why there exists no ideal
motor coordination solution for any adventure sport, in an absolute sense,
towards which all learners should aspire (Brymer and Renshaw 2010).
Different categories of constraints are resources that limit or set the
boundaries for the emergence of coordination patterns in human movement
systems. For example, personal constraints, relevant to each individual
athlete, are structural or functional and refer to characteristics of an
individual, such as genes, anthropometric properties, strength, endurance,
body shape, fitness level, technical abilities, age, and so on. In adventure
sports, psychological factors like beliefs, fear, anxiety, emotional readiness
and motivation obviously play a significant role in shaping the way that
participants approach a task such as climbing a vertical surface two hundred
feet off the ground or paddling through extreme white-water rapids (Brymer
and Schweitzer 2013; Brymer and Renshaw 2010). These personal factors
play a significant role in determining the responses to outdoor adventure
settings adopted by individuals.
 Environmental constraints are external to each individual and can be social
and physical, reflecting the environmental conditions of a task in outdoor
sports (e.g. height, light, temperature, altitude, gravity, climbing tethered by a
rope in a group). Outdoor adventure sports are reliant on the natural
environment (water, air and land) and environmental and weather conditions
represent a major influence on participation and emergent behaviours. For
example white-water kayaking depends on water flow in a river: no water, no
kayaking (Brymer, Downey and Gray, 2009). Related to this category are
task constraints, which include the goal of the task, instructional information,
the equipment and the nature of the surface (e.g. texture of rock in climbing,
airflow in BASE jumping). Task constraints shape the movement pattern
variability exhibited by adventure sport athletes. For instance, functional
performance behaviours in rock and ice climbers emerge from their
interactions with specific informational properties of a particular rock cliff
(i.e. shape, steepness, type of rock, overhang) or a frozen waterfall (i.e.
shape, steepness, ambient temperature at the surface, overhang, ice thickness
and density of an icefall; Seifert et al. 2013). This is because the conditions
of the rock and ice (as affordances) are mostly unpredictable when viewed
from the ground before the climb.
Constraints and affordances in rock climbing
Complex biological systems, even simple ones without nervous systems,
have the capacity to use information to regulate their functional behaviours in
complex environments. For example, Reid et al. (2012) demonstrated that
slime mould, a brainless unicellular organism, functions as a complex system
to explore its environment, enhance its spatial awareness and use externalized
memory mechanisms to find food or avoid nasty substances. It uses an
information regulation system to maintain its functionality. Smaller units of
the slime mould oscillate at different frequencies when food is sensed by
molecular binding at its outer molecular surface area. Oscillating frequency
increases with attraction to an environmental source of information or
decreases in the presence of a repellent source. The membrane structure of
the organism softens or hardens as a result of increasing or decreasing
oscillation frequency to absorb food or repel a substance like salt,
respectively. Our work has shown that complex neurobiological systems,
with highly structured nervous systems, use similar information detection
systems to explore and act in their performance environments. Attunement to
affordances is one such exploratory system exploited by complex
neurobiological systems.
For example, during the sport of rock climbing, performance is
characterized by individuals interacting with various task and environmental
constraints. Gravity could be viewed as an environmental constraint because
quadrupedal vertical locomotion involving the minimal support of at least
one limb is required to prevent falling under the force of gravity. One major
climbing task constraint on the fluency of a climber's movements is the
interface of limb extremities with the rock cliff surface, since the performer
has to maintain body equilibrium on a more or less vertical climbing surface
(Bourdin et al. 1999; Quaine and Martin 1999), while combining upper- and
lower-limb movements to ascend (Sibella et al. 2007). Assessment of
climbing fluency through analysis of the geometric entropy index from the
three-dimensional body centre of mass displacement (Cordier et al. 1994;
Sibella et al. 2007) is a relevant indicator to understand how a climber
alternates climbing with time spent maintaining body equilibrium under
control in a tripodal position (Bourdin et al. 1999). Expert climbers exhibit a
low geometric entropy index value since they travel a great distance up a
surface slope between each grip hold to maintain energy efficiency (Cordier
et al. 1994, 1996). A high value of climbing movement fluency suggests a
more functional level of individual–environment coupling. For instance,
Sibella et al. 2007 emphasized the relationship between a lower geometric
entropy index and the capacity of rock climbers to move when using fewer
than three holds. This observation is known as the ‘three-holds-rule’: if a rock
climber uses a smaller number of holds he/she has to be quick enough to
maintain equilibrium on the surface. Conversely, if the number of holds is
equal to or greater than three, it is more likely that the rock climber will climb
slowly, because his/her equilibrium is always under control (Sibella et al.
2007).
Climbers can achieve these different intentional aims by increasing their
attunement to affordances to regulate their actions in different climbing
contexts. Ecological psychology (Gibson 1966, 1979) proposed that
affordances are opportunities in the environment that invite actions and
experts are more able to functionally exploit them in their behaviour. The
implication is that affordances do not exist independently of an individual's
perceptions and intentions. In the context of rock climbing, affordances
(called ‘climbing opportunities’ by Boschker et al. 2002) imply that the
coordination dynamics of action emerge from a mutual coupling of a
climber's perceptions and intentions with the specific properties of a climbing
surface, such as a rock cliff (i.e. shape, steepness, type of rock). In line with
ideas on the important role of affordances, Boschker et al. (2002) previously
reported how expert rock climbers recalled more information and focused on
the functional properties of a climbing wall, neglecting to perceive its
structural features. Conversely, in their study, beginners were not able to
recall such functional properties of the wall for action and they tended to
report almost exclusively the structural features of the holds (Boschker et al.
2002).
As environmental properties of the rock cliff are mostly not predictable
from the ground, an important challenge is to examine whether and how
expert climbers detect information when they proceed to a pre-ascent
climbing route visual inspection (i.e. route preview; Sanchez et al. 2012) and
how they recall climbing surface properties once they are in the ascent
(Pezzulo et al. 2010) compared the capability of expert and novice climbers
to preview and recall the sequences of holds composing easy, difficult and
impossible routes. When the climbers were voluntary distracted between the
route preview and the recall, they showed that a greater level of movement
expertise enabled a better recall of sequences of holds on difficult routes.
These findings suggest that route previewing on a climbing wall activates a
motor, embodied simulation, which relies on the motor competence of the
climbers. In this way, Sanchez et al. (2012) highlighted that route previewing
did not influence movement output performance but influenced movement
form. Notably, climbing fluency was better after a preview of the route, since
the climbers made fewer and shorter stops during their ascent. Finally, these
findings showed that, with increasing levels of expertise, climbers previewed
and recalled perceptual variables that are more functional (i.e. those which
can specify actions) under a variety of different performance circumstances
compared to novices who tended to focus and recall structural properties of
the climbing environment.
Affordances for ice climbing in skilled and unskilled climbers
The coordination of a climber's actions with the properties of a frozen
waterfall are mediated by ice picks for use with the hands and crampons on
the feet. The ice fall can vary because key properties are stochastically
distributed through the surface. For example, ambient temperature can
modify the ice density in certain regions of the ice fall, causing changes to the
structure of the surface and the placement of holes for actions, such as
hooking with an ice pick and kicking with the crampons. Icefall properties
such as these provide affordances for climbers who perceive, use and shape
movement opportunities from their own unique perspective. For example,
two climbers on the same crag would be working with the same
environmental properties but individual differences, such as limb length,
body length and emotional regulation, would result in different perceptions
and actions emerging during performance. Objectively, a crag might have
various climbing affordances but, because of different personal constraints,
not all climbers will be able to take advantage of a specific affordance.
Environmental constraints are not under the control of the climber, since
this task requires experience at using numerous types of actions (e.g.
swinging, kicking and hooking actions) during exploratory behaviour.
Patterns of interlimb coordination (e.g. horizontally, diagonally and vertically
located angular positions) can also be used to explore functionality of the
neurobiological system properties of degeneracy and multistability. For
instance, climbers could either swing their ice tools to create their own holes
in the surface of the ice fall or hook an existing hole (formed by the actions
of the lead climber or exploiting the presence of natural holes). The latter
strategy is more energy efficient and requires some expertise in perceiving
the affordances of holes for supporting body weight. This conceptualization
of skill in ice climbing supports the functional role of intra-individual
variability in exploring affordances of icefall properties for action. Research
undertaken by Seifert et al. (2011) on ice-climbing performance has assessed
interlimb coordination patterns by using the angle between the horizontal line
and the displacement of the heads of the left- and right-hand ice tools for
upper-limb coordination. Lower-limb coordination patterns corresponded to
the angle between the horizontal line and the displacement of the left and
right crampons (Figure 18.1; Seifert et al. 2011).
Figure 18.1 (left) Angles identified for the horizontal planes of the left and right limbs in the upper and
lower body of ice climbers; (right) modes of limb coordination as regards the angle value between
horizontal, left limb and right limb (from Seifert et al. 2011)

Unskilled ice climbers showed low levels of intra-individual movement

and coordination pattern variability, as they varied their upper- and lower-
limb coordination patterns much less frequently and extensively than the
experts. Beginners mostly used horizontally and diagonally located angular
positions (since limb anchorages are at the same level for the horizontal
angle, the arms or legs appear in an in-phase coordination mode). This
behaviour resembled climbing up a ladder and led them to maintain a static
‘X’ body position with arms and legs extended or with arms flexed and legs
extended, corresponding to a freezing of the motor system degrees of
freedom (Bernstein 1967). Moreover, beginners tended not to hook the ice
tool into existing holes in the ice fall but tended to swing the ice tool into or
out of the holes. Beginners mainly focused on attaining a deep anchorage on
the ice fall for both ice tools and crampons, by numerous episodes of ice-tool
swinging and crampon kicking, suggesting that the icefall properties were
either not perceived or were not used by them as relevant affordances. While
this behaviour might enhance their stability on the ice fall, it also led them to
greater levels of fatigue.
These data exemplify how unskilled ice climbers tended to perceive the
affordance of an icefall surface as requiring a significant amount of stability
with respect to the force of gravity. From a Gibsonian perspective, it was
likely that they perceived the icefall surface as ‘fall-offable’, which led them
to typically prioritize stability and security of posture in interacting with
environmental constraints, rather than speed and efficiency of displacement
up the surface. For novice climbers extending their range of skills in more
difficult terrains, psychological factors such as fear and anxiety might
interfere with their ability to use affordances that might have been perceived
and used when nearer the ground or on simpler surfaces. Use of affordances
is often mediated by psychological constraints such as fear in the transition
from novice to high-level adventure sport performer (Brymer and Schweitzer
2013; Brymer and Renshaw 2010).
Conversely, in the skilled ice climbers, the frozen waterfall afforded
opportunities to hook their ice tools and kick in their crampons. They also
tended to show high levels of intra-individual movement and coordination
pattern variability, supporting an efficient balance between
dependency/independence of their actions from the environment as they
exploited affordances in the specific icefall properties. The efficient
individual–environment coupling of the experts was probably predicated on
the neurobiological property of degeneracy (Davids and Glazier 2010;
Edelman and Gally 2001) since they varied the types of movement and the
interlimb coordination modes they exploited to achieve their task goals.
Indeed, the multistability of their complex movement systems allowed them:
(i) to swing the ice tools and to kick the crampons in many different ways
(horizontally, diagonally, vertically and in crossed angular positions),
exploiting the functionality of intra-individual movement pattern variability.
For instance, crossing the arms is not a natural action but it enabled skilled
climbers to exploit information and hook existing holes in the ice fall; and
(ii), to hook the ice tools into already existing holes (i.e. exploiting
information on the shape of the ice fall) and to place the crampons in the
holes previously made by their own ice tools, instead of using repetitive ice
tool swinging and crampon kicking, as observed in beginners.
From a constraint-based perspective, these behaviours are completely
understandable and related to differentiated perceptions of personal (e.g.
fear), task and environmental constraints and significantly varying intentions.
Beginners tended to function independently of the ice fall's properties for
climbing upwards at speed, because their main goal was to keep their
equilibrium, with respect to gravitational forces, under control. Manipulation
of task constraints under the guidance of a climbing instructor would enable
the beginners to further interact with the icefall properties in a secure learning
environment, to balance their independence of/dependence on environmental
constraints when performing. This pedagogical approach would provide
gradual support for the emergence of a wider repertoire of movement by
exploiting system multistability in coordinating their actions while allowing
novice climbers to come to terms with and move through individual
constraints such as fear.
Knowledge of performance environments in outdoor adventure
sports
A constraints-based approach also highlights the importance of the role of
knowledge of the performance environment, which underpins the detection of
these affordances to regulate actions (Gibson 1966; Araújo and Davids 2011;
Davids and Araújo 2010b). Gibson (1966) proposed that knowledge of the
environment is embedded in knowing how to realize an action because it
involves perception of affordances used to control action directly (Araújo and
Davids 2011; Davids and Araújo 2010b). Expert performers are able to
transit functionally between various functional coordination solutions in ice
climbing by exploiting system multistability, notably by picking up
affordances for action. In contrast, novices tend to pick up and use sources of
information that may be only partially functional in a particular performance
situation because they do not specify actions. A similar experience occurs in
white-water kayaking, where a novice would invariably cross a fast-moving
current by using a standard but stable ferry glide, which involves continuous
arm work with the paddle and trunk work keeping the boat on edge and at an
appropriate angle to the current. The same move by an expert might take one
stroke, as the expert is attuned to the nuances of the water. By perceiving and
using and even shaping aspects of the current possibly considered too
dangerous by the beginner, the expert glides effortlessly across the current
using the upstream face of a standing wave. In outdoor adventure sports,
knowledge of the environment can also mean the difference between life and
death, as the adventurer makes a decision based on an assessment of the
environment and about the likely success of an action in partnership with a
particular environmental condition (Brymer and Gray 2010).
Related to the notion of performance dependency/independence is the
concept of metastability, another important property of complex, dynamical
movement systems. System metastability emerges when a subtle blend arises
between behavioural stability and instability, which research indicates can be
exploited to achieve adaptive functional performance goals in sport
(Hristovski et al. 2006a,b; Pinder et al. 2012). The same process is also
apparent in outdoor adventure sports. Metastability has been defined as a
transient or semi-transient behaviour or a ‘dynamically stable’ state of system
organization (Kelso 1995, 2012, 2012). In a metastable performance region,
component tendencies for action independence coexist with tendencies to
couple actions with affordances, explaining how rich and varied sequences of
goal-directed behaviour can spontaneously emerge in highly dynamic
adventure environments. Metastability helps an individual to adapt their
motor behaviours to achieve particular performance goals (Chow et al. 2011;
Hristovski et al. 2006a,b; Pinder et al. 2012). In a metastable performance
region, one or several movement patterns are weakly stable (when there are
multiple attractors) or weakly unstable (when there are only attractor
remnants) and switching between two or more movement patterns occurs
according to interacting constraints.
Movement variability as adaptive skilled behaviour
In sport, traditional approaches to the study of performance and expertise
mostly focus on performance outputs and their consistent achievement. An
important challenge is to pay closer attention to movement organization in
studying expertise in outdoor adventure sports, since the existence of several
expert performance profiles may imply that there actually is no putative
expert model of performance towards which all learners should aspire.
Ecological dynamics and its emphasis on emergent behaviours under
interacting constraints distinguishes variability in movement organization, a
healthy sign of adaptive behaviour in indeterminate biological movement
systems, from variability in movement output, which is synonymous with
performance inconsistency and, therefore, less functional (Davids et al.
2006).
Research in ecological dynamics has shown that movement system
variability is not necessarily noise that is detrimental to performance, error
(Newell and Corcos 1993; Newell and Slifkin 1998; Newell 2006) or a
deviation from a putative expert performance model that should be corrected
in beginners. Considering the functional role of movement variability leads to
an exploration of what adaptive behaviour means, so that it could be more
appropriate to consider the term adaptability rather than variability.
Adaptability relates to an appropriate ratio between stability (i.e. persistent
behaviours) and flexibility (i.e. variable behaviours; Davids et al. 2003; Li et
al. 2005; van Emmerik and van Wegen 2000; Warren 2006) and is essential
to skilled performance in outdoor adventurous sports. Expert behaviour is
characterized by stable and reproducible movement patterns that are
consistent over time, resistant to perturbations and reproducible, in that a
similar movement pattern may recur under different task and environmental
constraints. It is not stereotyped and rigid but flexible and adaptive. Even if
movement patterns could show regularities and similarities within their
structural components, an individual is not fixed into a rigidly stable solution
but can adapt a movement pattern in a functional way, as neurobiological
complex systems reveal degeneracy (Edelman and Gally 2001). In white-
water kayaking, for example, no two rapids are alike; however, there are only
a limited number of ways of crossing a fast moving current. The expert is
able to adapt (and perhaps merge) the basic processes to fit emergent
environmental affordances. In this process, there is a fine line between
stability and instability as the expert coordinates extreme boat edge, boat
angle, body movement, paddle stroke and mental capacity to decide and act.
The expert is able to expend less effort by exploring this fine line between
stability and instability, as it take less energy to cross the river when the
kayak is in a position of impending instability but there is also a potential
cost if the fine line is crossed.
An ecological dynamics model of expertise articulates both stability and
flexibility: experts and non-experts each have their stable states and
sometimes share the same coordination modes; however, a particularity of
expert performance is the capacity for adaptability, i.e. to produce behaviour
which is stable when needed and variable when needed. In fact, although
human movement systems naturally tend to move toward stable states, as
more economical organization modes (Hoyt and Taylor 1981; Sparrow 2000;
Sparrow and Newell 1998), stability and flexibility should not be construed
as opposites. Flexibility should not be interpreted as a loss of stability but,
conversely, as a sign of adaptability (van Emmerik and van Wegen 2000;
Warren 2006). From there, Bartlett et al. (2007) indicated three functional
roles of movement pattern variability: (i) to adapt to interacting constraints;
(ii) to facilitate (structural or not) changes in coordination modes and, at the
same time, maintaining functional performance through degeneracy or
redundancy; and (iii) to reduce the risk of injury.
Mason (2010) has highlighted four avenues for degeneracy in biological
systems that could help us to understand how expert individuals functionally
adapt their motor behaviours to exhibit high levels of performance outcomes
in dynamic outdoor adventure sport contexts. Firstly, ‘redundancy can create
the opportunity for degeneracy to arise as the function of the original
structure is maintained by one copy, while any other copy is free to diverge
functionally’ (Mason 2010, p. 282). Secondly, degeneracy can occur through
parcellation, when an initial structure is subdivided into smaller units that
can still perform the initial function and can also be functionally redeployed
(Mason 2010, p. 282). Thirdly, degeneracy may emerge through a
coordinative structure that realises a function in combination. This means
that, whether or not a structure is able to perform an initial function
independently, another one is available for modification. Lastly, degeneracy
may exist when two or more independent structures converge upon the same
function. These four avenues for degeneracy emphasize the potential
adaptation in human movement systems that coaches and teachers could
encourage to emerge in various individual motor responses to satisfy a task.
In white-water kayaking, for example, there are numerous ways of exiting
a fast water section of a river to enter a slow moving eddy. If the exit/entry
move requires a fast turn then one of the most efficient ways of undertaking
this in a kayak is to coordinate boat, body and paddle in such a way that the
whole system is potentially unbalanced. In its basic form, many students of
kayaking would recognise this move as a bow rudder. The paddler drives for
the eddy employs a turning stroke on the opposite side then edges the boat to
near imbalance and plants a blade into the eddy at an angle that, if effectively
undertaken, ensures balance. If ineffectively undertaken, the move could
result in a capsize, which is not recommended on fast-moving rocky rapids.
A ‘novice’ in this situation would most likely err on the side of ensuring that
the boat was not edged too far and that the stroke was planted parallel with
their feet and a short distance from the boat. The top arm would be securely
in front of their head to ensure strength and minimize risk of injury.
However, the turn speed and precision would not be optimal for a smaller
eddy or for turns that require sharp, instant precision. For this reason, an
expert would invariably edge the boat to a position that would be unstable on
its own but would balance this by varying the position of their body and the
planted paddle in the turn. Depending on the environmental context, the
expert might even undertake a move where the top hand is behind the head,
thus ensuring further system imbalance. The potential downside of this move
is that risk of injury is heightened because the system is so far out of balance.
If the coordination of boat, body and paddle is not finely tuned, injury is
possible and capsize is assured.
At an inter-individual level, movement pattern variability has been
observed both at novice and expert level, suggesting that neurobiological
degeneracy is a common property in human motor behaviour. However,
degeneracy occurs in different ways as regards to expertise level. Owing to
extensive experience in various performance contexts, experts exploit to the
fullest their individual properties according to the task demands and the
environmental constraints. As stated previously, when the gap existing
between the pre-existing movement pattern repertoire of an individual and the
task demands is low and/or when the tasks demands are weak, multistability
of movement patterns could emerge, giving support for neurobiological
degeneracy. For instance, expert climbers regularly use several hand grasping
patterns and body positions for a given hold (e.g. crimp, gaston, jug, mono,
pinch, pocket, sloper and undercling grasping pattern; bridge, campus,
crossover, deadpoint, flag, heel hook, knee bar and mantle body positions;
Phillips et al. 2012) exhibiting several individual climbing profiles. In
contrast, novices tend to demonstrate a basic quadrupedic climbing pattern
that resembles climbing a ladder.
Conclusions and implications
A constraints-based framework enables a new understanding of expertise in
outdoor adventure sports by considering performer–environment couplings
through emergent and self-organizing behaviours in relation to interacting
constraints. Expert adventure athletes, conceptualized as complex, dynamical
movement systems, pick up affordances for action to regulate adaptive
transitions between functional movement behaviours. For example, icefall
properties contain affordances that can induce variable motor coordination
patterns in expert climbers, whereas beginners use a basic and stable motor
organization to achieve the main goal of maintaining body equilibrium with
respect to gravity. Movement pattern variability could play a functional role
as individuals adapt their behaviours to ecological constraints of performance
by exhibiting multistability and metastability. The properties are exploitable
by coaches and educators who can use system instability to stimulate
creativity and skill acquisition. In this way, expertise relates to the
neurobiological property of adaptability, a subtle blend between stability and
flexibility, as experts are able to be stable when needed and variable when
needed. We highlighted a new emphasis on how novices and experts
individually manage motor system degrees of freedom in coordinative
structures through redundancy or degeneracy as they structurally adapt
system and sub-system organization to achieve functional goals. The main
implications for adventure athletes are to identify and manipulate key
constraints to perturb and create emergence of appropriate behaviours rather
than to encourage the imitation of a single response in reference to a putative
ideal expert model. Indeed, imitating so-called ‘expert behaviours’ could lead
to frustration and a prolonged skill-acquisition process, as novices may
encounter difficulties in matching the required behaviours. Using a
constraint-led approach could lead to the emergence of individualized
movement responses directly related to the preexisting intrinsic dynamics of
a performer in outdoor adventure sports.
References
Araújo A. and Davids K. (2011) What exactly is acquired during skill
acquisition? Journal of Consciousness Studies, 18 (3–4): 7–23.
Araújo, D., Davids, K., Bennett, S. J., Button, C. and Chapman, G. (2004)
Emergence of sport skills under constraint, A. M. Williams and N. Hodges
(eds) Skill Acquisition in Sport: Research, Theory and Practice, 2nd edn.
London: Routledge, pp. 409–33.
Bartlett, R., Wheat, J., and Robins, M. (2007) Is movement variability
important for sports biomechanists? Sports Biomechanics, 6, 224–43.
Bernstein, N. A. (1967) The Co-ordination and Regulation of Movement.
New York: Pergamon Press Elmsford.
Boschker, M. S. J., Bakker, F. C. and Michaels, C. F. (2002) Memory for the
functional characteristics of climbing walls: perceiving affordances. Journal
of Motor Behavior, 34: 25–36.
Bourdin, C., Teasdale, N., Nougier, V., Bard, C. and Fleury, M. (1999)
Postural constraints modify the organization of grasping movements. Human
Movement Science, 18: 87–102.
Brymer, E. and Davids, K. (2012) Ecological dynamics as a theoretical
framework for development of sustainable behaviours towards the
environment. Environmental Education Research, 19 (1): 45–63.
Brymer, E. and Davids, K. (2013) Experiential learning as an constraint-led
process: an ecological dynamics perspective. Journal of Adventure education
and Outdoor Learning, doi: 10.1080/14729679.2013.789353.
Brymer, E. and Gray, T. (2009) Dancing with nature: rhythm and harmony in
extreme sport participation. Journal of Adventure Education and Outdoor
Learning, 9 (2): 135–49.
Brymer, E. and Gray, T. (2010) Developing an intimate ‘relationship’ with
nature through extreme sports participation. Leisure/Loisir, 34 (4): 361–74.
Brymer, E. and Renshaw, I. (2010) An introduction to the constraints-led
approach to learning in outdoor education. Australian Journal of Outdoor
Education, 14 (2): 33–41.
Brymer, E. and Schweitzer R. (2013) Fear is good for your health: a
phenomenological understanding of fear and anxiety in extreme sport.
Journal of Health Psychology, 18 (4): 477–87.
Brymer E., Downey, G. and Gray, T. (2009) Extreme sports as a precursor to
environmental sustainability. Journal of Sport and Tourism, 14 (2–3): 193–
204.
Chow, J. Y., Davids, K., Hristovski, R., Araújo, D. and Passos, P. (2011)
Nonlinear pedagogy: Learning design for self-organizing neurobiological
systems, New Ideas in Psychology, 29: 189–200.
Cordier, P., Dietrich, G. and Pailhous, J. (1996) Harmonic analysis of a
complex motor behaviour. Human Movement Science, 15 (6): 789–807.
Cordier, P., France, M. M., Pailhous, J. and Bolon, P. (1994) Entropy as a
global variable of the learning process. Human Movement Science, 13: 745–
63.
Davids, K. and Araújo, A. (2010a) The concept of ‘organismic asymmetry’ in
sport science. Journal of Science and Medicine in Sport, 13: 633–40.
Davids, K. and Araújo, A. (2010b) Perception of affordances in multi-scale
dynamics as an alternative explanation for equivalence of analogical and
inferential reasoning in animals and humans. Theory and Psychology, 20 (1):
125–34.
Davids K. and Glazier, P. (2010) Deconstructing neurobiological
coordination: the role of the biomechanics-motor control nexus. Exercise
Sport Science Reviews, 38 (2): 86–90.
Davids, K., Araújo, D., Shuttleworth, R. and Button, C. (2003) Acquiring
skill in sport: a constraints-led perspective. International Journal of Computer
Science in Sport, 2, 31–9.
Davids K., Bennett S., and Newell, K. (2006) Movement System Variability.
Champaign, IL: Human Kinetics.
Davids, K., Button C. and Bennett S. (2008) Dynamics of Skill Acquisition: A
Constraintsled Approach. Champaign, IL: Human Kinetics.
Davids, K., Araújo D., Hristovski, R., Passos, P. and Chow, J.-Y. (2012)
Ecological dynamics and motor learning design in sport, in A. M. Williams
and N. Hodges (eds) Skill Acquisition in Sport: Research, Theory and
Practice, 2nd edn. London: Routledge, pp. 112–30.
Edelman, G. M. and Gally, J. A. (2001) Degeneracy and complexity in
biological systems. Proceedings of the National Academy of Sciences of the
U S A, 98 (24): 13763–8.
Gibson, J. J. (1979) An Ecological Approach to Visual Perception. Boston,
MA: Houghton-Mifflin.
Gibson, J. (1966) The Senses Considered as Perceptual Systems. Boston,
MA: Houghton Mifflin.
Hoyt, D. F. and Taylor, C. R. (1981) Gait and the energetics of locomotion in
horses. Nature, 292: 239–40.
Hristovski, R., Davids, K. and Araújo, D. (2006a) Affordance-controlled
bifurcations of action patterns in martial arts. Nonlinear Dynamics,
Psychology, and Life Sciences, 10 (4): 409–44.
Hristovski, R., Davids K., Araújo, D. and Button, C. (2006b) How boxers
decide to punch a target: emergent behaviour in non linear dynamic
movement systems. Journal of Sports Science and Medicine, 5: 60–73.
Kelso, J. A. S. (1995) Dynamic patterns: the self-organization of brain and
behaviour. Cambridge, MA: MIT Press.
Kelso, J. A. S. (2012) Multi-stability and meta-stability: understanding
dynamic coordination in the brain. Philosophical Transactions of the Royal
Society B Biological Sciences, 367: 906–18.
Li, L., Haddad, J. M. and Hamill, J. (2005) Stability and variability may
respond differently to changes in walking speed. Human Movement Science,
24: 257–67.
Mason, P. H. (2010) Degeneracy at multiple levels of complexity. Biological
Theory, 5 (3): 277–88.
Newell, K. M. (1986) Constraints on the development of coordination, in M.
G. Wade and H. T. A. Whiting (eds) Motor Development in Children: Aspect
of Coordination and Control. Dordrecht: Martinus Nijhoff, pp. 341–60.
Newell, K. M. and Corcos, D. M. (1993) Issues in variability and motor
control, in K.M. Newell and D. M. Corcos (eds) Variability and Motor
Control. Champaign, IL: Human Kinetics Publishers, pp. 1–12.
Newell K. M. and Slifkin A. B. (1998) The nature of movement variability,
in J. P. Piek (ed.) Motor Behaviour and Human Skill: A Multidisciplinarity
Perspective. Champaign, IL: Human Kinetics Publishers, pp. 143–60.
Pezzulo, G., Barca, L., Lamberti Bocconi, A. and Borghi, A. M. (2010) When
affordances climb into your mind: advantages of motor simulation in a
memory task performed by novice and expert rock climbers, Brain and
Cognition, 73: 68–73.
Phillips, K. C., Sassaman, J. M. and Smoliga, J. M. (2012) Optimizing rock
climbing performance through sport-specific strength and conditioning.
Strength and Conditioning Journal, 24 (3): 1–18.
Pinder, R. A., Davids, K. and Renshaw, I. (2012) Metastability and emergent
performance of dynamic interceptive actions. Journal of Science and
Medicine in Sport, 15 (5): 437–43.
Quaine, F. and Martin, L. (1999) A biomechanical study of equilibrium in
sport rock climbing, Gait and Posture, 10: 233–9.
Reid, C. R., Latty, T., Dussutour, A. and Beekman, M. (2012) Slime mold
uses an externalised spatial ‘memory’ to navigate in complex environments.
PNAS Proceedings of the National Academy of Sciences of the USA, 109
(43): 17490–4.
Sanchez, X., Lambert, P., Jones, G. and Llewellyn D. J. (2012) Efficacy of
pre-ascent climbing route visual inspection in indoor sport climbing,
Scandinavian Journal of Medicine and Science in Sports, 22: 67–72.
Seifert, L., Wattebled, L., L'Hermette, M. and Hérault, R. (2011) Inter-limb
coordination variability in ice climbers of different skill level. Education and
Physical Training in Sport, 1: 63–8.
Seifert, L., Button, C. and Davids, K. (2013) Key properties of expert
movement systems in sport: An Ecological Dynamics perspective. Sports
Medicine, 43 (3): 167–78.
Sibella, F., Frosio, I., Schena, F., and Borghese, N. A. (2007) 3D analysis of
the body center of mass in rock climbing. Human Movement Science, 26:
841–52.
Sparrow, W. A. (2000) Energetics of Human Activity. Champaign, IL:
Human Kinetics.
Sparrow, W. A. and Newell, K. M. (1998) Metabolic energy expenditure and
the regulation of movement economy. Psychonomic Bulletin and Review, 5:
173–96.
van Emmerik, R. E. A. and van Wegen, E. E. H. (2000) On variability and
stability in human movement. Journal of Applied Biomechanics, 16: 394–
406.
Warren, W. H. (2006) The dynamics of perception and action. Psychological
Review, 113 (2): 358–89.
19 Skill acquisition and
representative task design
Ross A. Pinder, Ian Renshaw, Jonathon
Headrick and Keith Davids
Egon Brunswik's (1956) insights have revealed that representative task design
is a key concept in understanding the organization of task constraints in
experiments, evaluation tests and learning programmes in sport.
Representative design implies that these environments need to be predicated
on key information sources found in specific performance contexts. Although
Brunswik's ideas have, until now, failed to be fully appreciated in a wide
range of experimental and behavioural sciences, many of the concepts have
begun to be accepted in the study of complex systems in sport (Araújo and
Davids 2009; Beek et al. 2003; Davids 2008; Dicks et al. 2008; Fajen et al.
2009; Pinder et al. 2011b,c). Here, we discuss the ideas of representative task
design and examine its implications for constructing experimental and
learning environments in sport. We provide principles for sport scientists,
experimental psychologists and pedagogues to recognize the potential
application and adaptation of Brunswik's original concepts, with examples
from various sports to demonstrate how the model can be applied in practice.
We also discuss the role of representative design in supporting the
psychology of learning and creating holistic learning environments for
learners as complex systems, in addition to considering the integrated
emotional engagement of participants for the enhancement of learning and
practice task design.
Egon Brunswik's representative design
Traditional research designs in sports science and motor learning have been
historically reductionist and systematic in nature (Dhami et al. 2004). The
traditional distinction between experimental control and ‘field’ research has
been recognized as a false dichotomy, where reductionist approaches are
being complemented by functional paradigms of movement coordination
considered in humans modelled as complex dynamical systems. This
paradigm has been largely supported by theoretical and empirical work from
ecological psychology (Brunswik 1956; Gibson 1979). The insights of James
Gibson (1979) revealed why specific environments need to be carefully
structured and organized so that they maintain functional relations between
processes of perception and action during performance. In ecological
psychology, there have been recent attempts to mediate the links between
Gibsonian and Brunswikian theoretical and methodological approaches
(Araújo et al. 2007; Kirlik 2009; Vicente 2003), to broaden ecological
research and develop cumulative knowledge. Current neo- Gibsonian
research into event perception coincides with ideas on organism–environment
relationships demonstrated in the lens model (Figure 19.1; Brunswik 1952,
1956), where research is now confronting the problem of uncertainty
highlighted by Brunswik's theory of probabilistic functionalism (Kirlik 2009).
The integration of Egon Brunswik's and James Gibson's theoretical work
provides an advantageous methodological approach for studying complex
systems in sport, maintaining the scale of analysis as the performer–
environment relationship (Kirlik 2009; Reed 1996; Vicente 2003).
Egon Brunswik proposed the term representative design as an alternative
to systematic design in the 1950s, advocating that the study of human
performance and psychological processes should be at the level of organism–
environment interaction. That is, a focus should be maintained on the
complex system formed by a sports performer and their performance
environment. These ideas proposed that experimental stimuli must be
sampled from an organism's (e.g. performer's) natural environment, so as to
be representative of the stimuli to which it is adapted and to which
experimental data are intended to be generalized (Brunswik 1956).
Generalization of findings beyond the constraints of the experiment is highly
problematic when studying the adaptability of human movement in dynamic
environments (see Chapter 15). There is a need to adequately sample
environmental constraints which facilitate the emergence of functional human
behaviours based on the informational variables of the specific performance
context of interest. In sport, these variables emerge from the interpersonal
interactions between performers and between performers and key events,
objects and surrounding energy structures present in the performance
environment (exemplified by how a soccer player interacts with the ball, the
opposition and performers within their own team): the basis of sport as a
complex system.

Figure 19.1 The role of Brunswik's lens model in understanding informational variables for complex
systems in sport – analysis of a tennis serve (model adapted from Araujo and Kirlik 2008)

Brunswik (1956) concluded that the adequate sampling of situations and

problems in psychological experimentation should take precedence over the
concerns of population sampling. His ecological approach to cognition
(considered here to encompass intentionality, see Searle 1983; Shaw 2001),
perception and action was implemented through the theoretical framework of
probabilistic functionalism (Hammond and Stewart 2001; Kirlik 2009).
Simply stated, this idea signifies that, in sport, performers try to use multiple
sources of imperfect information to infer about future events or aspects of the
dynamic environment (e.g. the way a rugby defender might use kinematic
information from an opponent's movements to predict their attacking
intentions). In describing organism–environment interactions, Brunswik
(1956) referred to distal variables (remote properties of the environment, such
as an opponent's intentions) and proximal variables or cues (information
sources directly available, such as vision of an opponent's movements).
Importantly, this process is inherently probabilistic, with variables available
from the environment providing different levels of functionality. Perceptual
judgements are based on the inferences of proximal cues (perceptual
variables), with different sets of cues available in different environmental
conditions. The ability to identify and selectively use informational variables
(or in Brunswikian terms, cues) has been considered to be one of the major
factors influencing the ability to predict future behaviour of other performers
in complex sport tasks (van der Kamp et al. 2008; Williams et al. 1999). As
informational sources differ in their degrees of functionality, they may also
vary in the degree to which they are intercorrelated with each other (for
example the relationship between multiple limb segments of a tennis server's
action).
Importantly, cues must be combined in a context-specific manner, allowing
for the development and refinement of robust performer–environment
couplings, known as vicarious functioning. Owing to the interdependence of
the processes of intentionality, perception and action, and the ways in which
human movement systems have evolved, the separation of these processes
during experimental or learning designs, often in the name of enhancing
internal validity, is of major concern (Michaels and Carello 1981; van der
Kamp et al. 2008). Simulations (e.g. experimental or practice designs
attempting to simulate aspects of a performance environment) that decouple
the processes of perception and action do not permit perceptual and action
subsystems to function, as they have been developed or trained to do in a
particular sport performance context, nor do they allow for the same
emotional or psychological engagement of an individual with the task. The
mutual interdependence between intentionality, perception and action
suggests that these processes should be allowed to function in an
interdependent manner in an experiment or learning programme (Araújo et
al. 2007; Pinder et al. 2011b; van der Kamp et al. 2008). As achievement of
an action cannot be defined without reference to the environment, functional
systems of an individual (such as ventral and dorsal visual streams, or
engagement of the limbic system) are viewed as contributing factors to task
goal achievement (Kirlik 2001). In representative design, there is a strong
emphasis on the relations between the participant and the environment, which
is often neglected in traditional approaches to the study of human behaviour,
such as in cognitive psychology (e.g. Dunwoody 2006).
 Perceptual motor skill research has all too frequently attempted to focus on
the precise (presumed singular) source of information that perceivers
putatively use to achieve a specific task outcome (Withagen and van
Wermeskerken 2009), assuming that some source of optimal information will
be available to the perceiver (Reed 1996). Studies focusing on perceptual
learning have now begun to cast doubt over this assumption of information
singularity, with a greater emphasis on understanding information plurality
when individuals use various informational variables to regulate their actions,
changing their use of variables over time (Jacobs et al. 2001; Runeson and
Andersson 2007; Withagen and Michaels 2005). The use of a variety of
perceptual variables to regulate behaviours is supported by a better
understanding of neurobiological complexity, especially of system
degeneracy, which performers can exploit to contextually adapt their actions
under changing task constraints (Davids and Araújo 2010). The perceiver is
likely to use a variety of relatively reliable perceptual cues to provide
themselves with the required information to support action (Kirlik 2009;
Runeson and Andersson 2007). For example, a performer may use a
combination of kinematic cues from a baseball pitcher's action, such as wrist
position, seam position and shoulder alignment, to predict the speed and
angle of the upcoming pitch. The establishment of robust organism–
environment relationships (i.e. information–movement couplings), therefore,
is analogous to Bernstein's (1967) mastering degrees of freedom problem.
Essentially, several different couplings may be available for exploitation in
goal-directed behaviour, with learning characterized by the refinement of
established information–movement couplings. The lens model (see Figure
19.1) explains how performers cope with complex and dynamically changing
environments through exploiting perceptual and motor system degeneracy
(Davids 2008); essentially, a pictorial representation of vicarious functioning
(see Goldstein 2006).
In considering sport performance as a complex system, the major
implication is that practitioners need to ensure that all key informational
variables are available in particular experimental and learning environments
to allow performers to detect information and perceive affordances for action.
Note that ecological validity was originally defined as the statistical
correlation between proximal cues available in the environment (perceptual
variables) and the extent to which they depict the distal criterion state of the
environment (see also, Araújo et al. 2007; Brunswik 1956; Pinder et al.
2011b). This idea is exemplified by the statistical correlation between the
presence of an observable kinematic detail of a tennis server's action and the
resultant action (e.g. a cross-court serve). The definition of representative
design emphasizes the need to ensure that task constraints of experiments
represent (i.e. simulate) the particular task constraints of the performance
environment that forms the focus of study. In this way, the distinction
between ecological validity and representative design is not a trivial issue, as
the clarification allows for a more comprehensive and principled approach to
studying complex system interactions in sport.
Principles of representative learning design
Brunswik's (1956) concept of representative task design provides a critical
framework for the study of intention, perception and action processes in sport
(Dicks et al. 2008). A key concept of experimental design in the analysis of
complex human behaviour is the importance of manipulating task constraints
(Araújo et al. 2004), which are central to the process of harnessing inherent
self-organization tendencies in complex neurobiological systems. In sport,
representative design supports the generalization of task constraints in
learning designs to the ecological constraints of performance (Araújo et al.
2006; Davids 2008), implying that a performer needs to be able to search the
environment for reliable information to support and direct action. Learning
designs need to be predicated on the ability of performers to use information
from the performance environment, which is vital for the acquisition of a
functional and efficient relationship between perceptual and motor processes
in the control of action (Le Runigo et al. 2005). Skill acquisition programmes
in sport, therefore, need to be designed to enhance the existing coupling
between an individual's intentions and perception and action subsystems.
Based on these ideas, it is important for sport practitioners to identify the
major ecological constraints of a particular performance environment,
especially the informational constraints.
Despite technological and methodological advances, many experimental
and practice tasks in the analysis of dynamic human behaviours in sport fail
to provide an adequate platform to facilitate the intertwined processes of
cognition, perception and action to function as a complex system. For
example, Dicks et al. (2010) demonstrated significant changes in the
detection of information when performers responded under experimental task
constraints varying in levels of functional coupling between perception and
action processes (see also Farrow and Abernethy 2003). Soccer (association
football) goalkeepers were observed to use different information sources
under the constraints of video simulations and ‘in situ’ tasks representative of
intercepting a penalty kick in competition. These data demonstrated the need
for the analysis of behaviour at an ecological scale of performance (e.g.
performers directly interacting with the performance environment), ensuring
that task designs allow for the emergence of functional behaviours
representative of the performance environment of interest. The presence of
this key characteristic of system complexity can be assessed and achieved
through the careful manipulation of key task constraints (e.g. informational
constraints) during practice and experimentation. The design of learning
environments that effectively capture and enhance functional movement
responses representative of performance contexts should be constructed using
the theoretical principles of representative learning design (Pinder et al.
2011b).
Representative learning design (Figure 19.2) provides a principled
framework which theoretically captures how insights from ecological
dynamics and nonlinear pedagogy approaches can be used to ensure that
experimental and practice task constraints are representative of sport
performance environments as integrated complex systems (see Renshaw et
al. 2010). Assessment of representative learning design in specific practice
tasks allows sport scientists to understand the functionality and limitations of
particular training environments. Understanding how the complexity sciences
can inform representative learning design may provide opportunities to
optimize learning programmes in sport and inform use of performance
enhancement tools, such as projection technology, during practice.
Practitioners should consider both the functionality and action fidelity of a
simulation design. This will ensure that the complexity of the performance
context is considered and the coupling between key cognitive, perceptual and
action processes is maintained. Functionality provides a measure of action
achievement and the information on which this level of attainment is based.
Considering the functionality of a task ensures that: a) the degree of success
between two or more contexts is controlled and compared; and b),
participants are able to use comparable information sources for goal
attainment and control of movement behaviour between
learning/experimental and performance tasks (e.g. by comparing visual
search strategies or detection of information between contexts).
Figure 19.2 Principles for the assessment of representative learning design

Action fidelity (see Stoffregen et al. 2003) is a measure of the degree of

association between behaviour in two or more contexts, allowing scientists or
practitioners to examine whether the actions and decisions of a complex
neurobiological system remain the same under experimental or learning
designs and their respective performance contexts. Essentially, action fidelity
provides information about the fidelity of a performer's actions between the
simulator (e.g. a practice situation that is attempting to simulate one or all
aspects of a performance environment) and the simulated context (the
performance environment of interest). For example, large video simulations
are a common research and learning tool in perceptual-motor analysis in
sport; however, the concept of simulating aspects of performance should be
extended to all practice and learning tasks. Experimental, practice and
learning designs can be viewed as simulations of the performance
environment of interest (e.g. a ball projection machine simulating aspects of
facing a ‘live’ bowler in performance – also see below). Key measures of
behaviour, such as time taken to complete a task and observed movement
organisation during action, are suitable variables for assessing action fidelity
of simulation environments (Araújo et al. 2007). In the next section, we
demonstrate how the principles outlined here have and may be used to assess
the representative design of practice and learning designs in sport using short
examples from empirical findings.
Representative learning design for complex systems in sport
Considering sport performance as an integrated complex system, as outlined
throughout this book, can encourage the development of a more
representative design for learning, practice and experimental settings. To
date, the primary focus of representative (learning) design has been ensuring
that links between perception and action processes and the informational
variables provided in complex interactions in sports performance are
representative. Essentially, up to now, this focus has been on ensuring the
presence of specifying information in the external environmental. However,
information from within the performer can also act as a constraint, shaping
cognitions, feelings and behaviours during learning. In the following section,
we build on these ideas and provide examples from external environmental
and internal intrinsic dynamics.
The significance of ensuring representative designs for experimental and
learning tasks in the analysis and acquisition of skilled human behaviour is
exemplified by the use of ball projection machines. Our work has shown that
using projection machines with developmental athletes can create significant
differences in timing and control of their actions compared to when facing a
‘live’ opponent (Pinder et al. 2011a; Renshaw et al. 2007). Junior cricket
batters initiated key movements significantly later when playing against a
projection machine set to the same speed (approximately equal to 28 metres
per second) and with similar ball trajectory characteristics as a ball delivered
by a ‘live’ performer. These delays in movement initiation resulted in
significant adaptations in movement organization and reduction in quality of
bat–ball contact during interception; similar to findings in tennis (Shim et al.
2005). Technologies, such as projection machines, have the potential to
change the key ecological (informational) constraints between practice and
performance and demonstrate how caution should be taken in designing
practice simulations. Practitioners should use complex systems modelling to
assess the representative design of such tasks to inform how and when to use
technology in learning design, as the removal of key information sources may
limit a performer's ability to create robust and functional information–
movement couplings required in the performance environment (Pinder et al.
2011d).
Analyses of complex human behaviour in invasion sports (e.g. soccer,
rugby, basketball) have also demonstrated that sometimes apparently small
modifications in task constraints within aspects of the experimental design
can lead to substantial and critical changes in the functional and emergent
performances of individuals. Analyses of team games sub-phases (e.g. one-
versus-one, twoversus- one) have shown that performers are highly attuned to
the presence and movements of opponents, regulating movement
organization and decisionmaking behaviours within these complex systems.
For example, successful attacking performance in soccer and rugby union has
been highly constrained by the relative interpersonal coordination and
velocities of opponents (Duarte et al. 2010; Passos et al. 2008) or simply the
presence of defenders (Orth et al. 2012). Furthermore, Headrick et al. (2012)
demonstrated that the emergent attacking strategies of soccer players were
influenced by the proximity to goal and the subsequent intentions of both the
player in possession and the defender. These findings suggest that for
simulations (e.g. practice tasks) to demonstrate fidelity, practitioners should
consider the state and context of specific game situations, allow performers to
practice against opponents in the same areas of the field, as they would in
competition, and use standard pitch markings to orient players within those
tasks. This is an important issue that is often overlooked in the design of
practice tasks, particularly with junior or developmental performers. Subtly,
in invasion games such as football, pitch size needs to be carefully considered
from the physiological and skill acquisition perspective. For example, while
the use of small-sided games is strongly recommended from the viewpoint of
nonlinear pedagogy (see Renshaw et al. 2010), practitioners need to also
carefully consider the player numbers and pitch dimensions in terms of the
physiological workloads of players (Hill-Haas et al. 2011). Principled
frameworks from a complex systems perspective are ideal for understanding
the interacting and complex nature of performer–environment relationships
and how the careful manipulation of task constraints can lead to substantial
changes to other areas or subsystems. Altering the spatiotemporal constraints
available to players on the ball is likely to lead to significant qualitative
changes in ball reception strategies and the time taken from ball reception to
pass initiation. These ideas are discussed in Renshaw et al. (2012) where a
‘game intensity index’ is proposed as a way of controlling and manipulating
the time–space available to players. These ideas are important for those
studying representative learning design in complex sports tasks and provide
some pointers for future research.
 Representative design also has significant implications for the
identification and development of talent in skill acquisition in sport. The
developed principles, based on identifying and manipulating the major
constraints during learning, provides a principled basis for the design of
performance evaluation tests in talent programmes (Vilar et al. 2012),
allowing performance and physiological capacities of potential performers to
be assessed. Just as experimental and learning designs should be based on the
performance context of interest, testing protocols should ensure that the tests
exhibit representative design (Araújo et al. 2007; Pinder et al. 2011b). For
example, it has been highlighted how static drills used in the identification of
potentially talented performers lack the key informational variables and
functionality between perception and action subsystems as they would
contribute in competitive performance (Vilar et al. 2012), nor do they
consider emotional or social aspects of sports performance; in such instances,
fundamental relationships between performers and key informational
variables of a complex sporting system are jeopardised.
Representative learning design and the psychology of learning
As previously discussed, the concept of representative (learning) design has
been outlined primarily in relation to the informational aspects of an
environment that provide the context for a performer's action. Another factor
to be considered for the design of practice tasks is their interaction with
psychological constraints such as emotions, attention, goal orientations and
self-efficacy. When sampling complex tasks with the aim of simulating
experiences from competition in practice, consideration of the psychological
aspects of performance is important in enabling practitioners and sport
scientists to design representative learning environments that faithfully
replicate interactions between task constraints and individual constraints such
as emotions, behaviours and cognitions. From this perspective, the performer
is seen as an interconnected system, which suggests that there is a
requirement for supporting the emotional and psychological engagement of
the individual in practice to simulate the intensity experienced in competitive
performance. Learning to cope with emotions created by complex
performance environments such as anger, fear, excitement, frustration or
anxiety is just as important as learning a technical skill. The development of
functional perception–action couplings is intertwined with emotions via
vertical integration between the lower emotional (e.g. limbic system) and
higher, cognitive systems (cortex). The emotional and psychological
engagement of individual performers is an important consideration to further
enhance the representative design of practice, learning and experimental
tasks, allowing for a more integrated perspective of the learner.
Emotions underpin numerous motivational aspects of learning, ‘playing an
integral part in shaping mental abilities, processes and skills’ (Lewis 2004, p.
221). Additionally, at a neurobiological level, learning is strengthened
through emotional engagement in learning activities (Lewis 2004; Tucker et
al. 2000). Similarly, when individual learners are required to fashion new
strategies out of older established stable patterns that have been functional in
the past, they have to learn to deal with and understand the accompanying
emotional turmoil associated with the high levels of variability that are
typically experienced in the cognition–action relationship around these phase
transitions (Lewis 2004; Thelen and Smith 1994). Unfortunately, coaches
tend to have limited skills and know - ledge of how to incorporate
psychological skills into learning design for complex systems (Harwood
2008). One way of enhancing a practitioner's ability to implement
psychological goals in practice could be to implicitly embed them into the
design of practice tasks. For example, Renshaw et al. (2012) demonstrated
how nonlinear pedagogy could be used to develop autonomy, competence
and relatedness; the key components underpinning self-determining
behaviour (Deci and Ryan 2002). Deci (1980) defined emotion as:
a reaction to a stimulus event (either actual or imagined). It involves
change in the viscera and musculature of the person, is experienced
subjectively in characteristic ways, is expressed through such means as
facial changes and action tendencies, and may mediate and energise
subsequent behaviours.
(Deci 1980: 85)
In essence, emotional activation constrains cognitive processes such as
attentional processes by appraising situations as beneficial or harmful (Lewis
2004). Those situations that are identified as significant trigger an emotion
and demand cognitive or motor responses. Consequently, exposure to
emotionally engaging tasks leads to changes in physiological responses (e.g.
heart rate, skin responses), subjective experiences (e.g. what a performer
consciously experiences prior to, during and after an event) and specific and
individual action tendencies (Vallerand and Blanchard 2000). In effect, as
individuals interact in complex and dynamic environments, changes in
emotions influence intentions, perceptions and actions. For example,
momentary failure in a practice bout in judo may lead to an immediate
intention to avoid experiencing failure (avoidance goal orientation) and
adoption of a more defensive strategy (Gernigon et al. 2004). In essence, an
increase in negative emotions such as anxiety owing to short term failure
could result in the reduction of performance degrees of freedom (e.g. biasing
or narrowing perception; Pessoa 2011) or reducing movement options (Parfitt
et al. 1990). Conversely, success is more likely to produce positive emotions
such as pride and satisfaction (Lewis 2004) and lead to goal orientations that
aim to demonstrate competence and lead to more attacking, assertive and
aggressive actions (Gernigon et al. 2004).
Psychological aspects of performance can also be linked to, and
exemplified by Brunswik's lens model. Referring to the example given
(Figure 19.1), the previous success and subsequent perceived competence of
a performer will influence how they approach a task. Likewise, the context of
the task (e.g. first or second serve) can influence whether the actions of the
performer tend to approach success (e.g. hit an attacking return on a second
serve) or avoid failure (e.g. only try to return the ball back into play on a first
serve). In this way, psychological aspects considered above can be viewed as
contributing towards the functionality of the practice or learning task design,
and an enhancement of the representative design of complex sports
interactions.
Emotion and learning in nonlinear pedagogy
Learning something new is an emotionally demanding experience,
irrespective of the learner being a complete beginner or an established
champion. For the beginner, who is required to construct a new coordination
pattern or a champion who has to move away from a successful strategy to
adapt to new rules, equipment or clever opponents that are always emerging
in complex environments, new learning requires risk-taking that can
potentially threaten self-efficacy and create somatic and cognitive anxiety as
they struggle to deal with the relative ‘failure’ that comes with trying to
develop something new. As Australian Test cricketer, Ed Cowan (2011, p.
11) eloquently put it when describing how he tried to improve his ability to
‘hit’ in T20 games (a short version cricket that requires batters to score by
regularly clearing the boundary), ‘learning “to hit” from scratch meant
dealing with the feeling of initial inadequacy’. Clearly, exploration
commensurate with becoming less reliant on tried and trusted stable
behaviours in search of new, potentially more advantageous, ways could be
threatening on a number of levels and requires appropriate psychosocial
support from coaches and teachers who understand the multilayered
consequences of ‘risk’ in learning.
Practitioners need to carefully consider the impact of manipulating
constraints that lead to performance failure and the consequent search for
new solutions. As discussed earlier from a complex systems perspective,
moving performers into metastable regions that result in self-organizing
phase transitions is accompanied by the creation of emotions that impact on
intentions and actions of each individual. High levels of emotion can be used
to index transition phases (Lewis 2004) and reflect the competition between
new emergent solutions and previously established patterns. In their initial
learning design, coaches need to understand that learners will always be
attracted to patterns that ‘work’ and that this would be indexed directly by
‘what feels good’. This in turn is shaped by the intentionality of each learner
to pursue his/her goals (e.g. to improve, please the coach, demonstrate
competence or protect self-worth) as effectively as they can (Lewis 2004). As
such, coaches should not only sample the informational constraints on actions
of learners but should also predict the likely psychological consequences
caused by potentially stressful practice tasks, based on knowledge of each
individual learner. On an individual level, because responses are likely to
reflect stable cognitive–emotional attractor patterns laid down in earlier
phases of development (Lewis 2004), identifying previous learning histories
of individuals is an important requirement when programming learning
activities. Consequently, coaches might not always want to create high levels
of representative learning design in respect of emotional responses for all
practice tasks. This process can be carefully controlled if the outlined
framework is followed and coaches are able to make principled decisions
based on sampling of competition and practice environments.
Conclusions
Egon Brunswik's original insights and concepts have proven to be highly
applicable and adaptable (Dhami et al. 2004; Wigton 2008). The
development of a functional and principled framework in sport demonstrates
how sport scientists, coaches and pedagogues can assess and provide
representative experimental, testing and learning tasks which maintain,
enhance or emphasize the functional and intertwined relationships between
performers' perceptions, actions, intentions and emotions. Considering sport
performance as an integrated complex system encourages the development of
more representative learning, practice and experimental designs, particularly
in supporting the role of information in the environment and the emotional
and psychological engagement of the individual in practice. While, the role of
emotion in learning has been considered by psychologists, we suggest that by
considering how emotions couple with intentions, perceptions and actions
within a representative learning design may reveal new insights on learning
via a nonlinear pedagogy.

Questions for students

1. Choose a sports task (e.g. springboard diving, baseball batting,
tennis serving):
• What are the key perception and action processes which should be
maintained in practice to ensure a representative degree of
functionality?
• What measures would allow you to assess the fidelity of action
between learning and performance contexts?
2. What are the implications of representative learning design for task
decomposition in the learning of complex sports skills?
3. How do representative learning environments support and foster
positive psychological experiences in performers?
4. Why is it important to consider emotion in the design of learning
tasks?
References
Araújo, D. and Kirlik, A. (2008) Towards an ecological approach to visual
anticipation for expert performance in sport. International Journal of Sport
Psychology, 39: 157–65.
Araújo, D. and Davids, K. (2009) Ecological approaches to cognition and
action in sport and exercise: ask not only what you do, but where you do it.
International Journal of Sport Psychology, 40 (1): 5–37.
Araújo, D., Davids, K., Bennett, S., Button, C. and Chapman, G. (2004)
Emergence of sport skills under constraints, in A. M. Williams and N. J.
Hodges (eds) Skill Acquisition in Sport: Research, Theory and Practice.
London: Taylor and Francis, pp. 409–33.
Araújo, D., Davids, K. and Hristovski, R. (2006) The ecological dynamics of
decision making in sport. Psychology of Sport and Exercise, 7: 653–76.
Araújo, D., Davids, K. and Passos, P. (2007) Ecological validity,
representative design, and correspondence between experimental task
constraints and behavioral setting: comment on Rogers, Kadar, and Costall
(2005) Ecological Psychology, 19 (1): 69–78.
Beek, P. J., Jacobs, D. M., Daffertshofer, A. and Huys, R. (2003) Expert
performance in sport: views from joint perspectives of ecological psychology
and dynamical systems theory, in J. L. Starkes and K. A. Ericsson (eds)
Expert Performance in Sport: Advances in Research on Sport Expertise.
Champaign, IL: Human Kinetics, pp. 321–44.
Bernstein, N. A. (1967) The Control and Regulation of Movements. London:
Pergamon Press.
Brunswik, E. (1952) The Conceptual Framework of Psychology. Chicago, IL:
University of Chicago Press.
Brunswik, E. (1956) Perception and the Representative Design of
Psychological Experiments, 2nd edn. Berkeley, CA: University of California
Press.
Cowan, E. (2011) In the Firing Line: Diary of a Season. Sydney: NewSouth
Publishing.
Davids, K. (2008) Designing representative task constraints for studying
visual anticipation in fast ball sports: what we can learn from past and
contemporary insights in neurobiology and psychology. International
Journal of Sport Psychology, 39 (2): 166–77.
Davids, K. and Araújo, D. (2010) The concept of ‘organismic asymmetry’ in
sport science. Journal of Science and Medicine in Sport, 13 (6): 633–40.
Deci, E. L. (1980) The Psychology of Self-Determination. Lexington, MA: D.
C. Heath and Company.
Deci, E. L. and Ryan, R. M. (eds) (2002) Handbook of Self-Determination
Research. Rochester, NY: University of Rochester Press.
Dhami, M. K., Hertwig, R. and Hoffrage, U. (2004) The role of
representative design in ecological approach to cognition. Psychological
Bulletin, 130 (6): 959–88.
Dicks, M., Davids, K. and Araújo, D. (2008) Ecological psychology and task
representativeness: implications for the design of perceptual-motor training
programmes in sport, in Y. Hong and R. Bartlett (eds) The Routledge
Handbook of Biomechanics and Human Movement Science. London:
Routledge, pp. 129–39.
Dicks, M., Button, C. and Davids, K. (2010) Examination of gaze behaviors
under in situ and video simulation task constraints reveals differences in
information pickup for perception and action. Attention, Perception and
Psychophysics, 72 (3): 706–20.
Duarte, R., Araújo, D., Gazimba, V., Fernandes, O., Folgado, H.,
Marmeleira, J. and Davids, K. (2010) The ecological dynamics of 1v1 sub-
phases in association football. Open Sports Sciences Journal, 3: 16–18.
Dunwoody, P. T. (2006) The neglect of the environment by cognitive
psychology. Journal of Theoretical and Philosophical Psychology, 26: 139–
53.
Fajen, B. R., Riley, M. A. and Turvey, M. T. (2009) Information, affordances
and the control of action in sport. International Journal of Sport Psychology,
40: 79–107.
Farrow, D. and Abernethy, B. (2003) Do expertise and the degree of
perception-action coupling affect natural anticipatory performance?
Perception, 32 (9): 1127–39.
Gernigon, C., d'Arripe-Longueville, F., Delignieres, D. and Ninot, G. (2004)
A dynamical systems perspective on goal involvement states in sport. Journal
of Sport and Exercise Psychology, 26: 572–96.
Gibson, J. J. (1979) The Ecological Approach to Visual Perception. Boston,
MA: Houghton Mifflin.
Goldstein, W. M. (2006) Introduction to Brunswikian theory and method, in
A. Kirlik (ed.) Adaptive Perspectives on Human–Technology Interaction.
Oxford: Oxford University Press, pp. 10–24.
Hammond, K. R. and Stewart, T. R. (2001) Introduction, in K. R. Hammond
and T. R. Stewart (eds) The Essential Brunswik: Beginnings, Explications,
Applications. New York: Oxford University Press, pp. 3–11.
Harwood, C. (2008) Development consulting in a professional football
academy: the 5Cs coaching efficiency program. Sport Psychologist, 22: 109–
33.
Headrick, J., Davids, K., Renshaw, I., Araújo, D., Passo, P. and Fernandes,
O. (2012) Proximity-to-goal as a constraint on patterns of behaviour in
attacker-defender dyads in team games. Journal of Sports Sciences, 30 (3):
247–53.
Hill-Haas, S. V., Dawson, B., Impellizzeri, F. M. and Coutts, A. J. (2011)
Physiology of small-sided games training in football: a systematic review.
Sports Medicine, 41 (3): 199–220.
Jacobs, D. M., Runeson, S. and Michaels, C. F. (2001) Learning to visually
perceive the relative mass of coliding balls in globally and locally constrained
task ecologies. Journal of Experimental Psychology: Human Perception and
Performance, 27 (5): 1019–38.
Kirlik, A. (2001) On Gibson's review of Brunswik, in K. R. Hammond and T.
R. Stewart (eds) The Essential Brunswik: Beginnings, Explications,
Applications. Oxford: Oxford University Press, pp. 238–42.
Kirlik, A. (2009) Brunswikian resources for event perception research.
Perception, 38 (3): 376–98.
Le Runigo, C., Benguigui, N. and Bardy, B. G. (2005) Perception-action
coupling and expertise in interceptive actions. Human Movement Science, 24:
429–45.
Lewis, M. D. (2004) The emergence of mind in the emotional brain, in A.
Demetriou and A. Raftopoulos (eds) Cognitive Developmental Change. New
York: Cambridge University Press, pp. 217–40.
Michaels, C. F. and Carello, C. (1981) Direct Perception. Englewood Cliffs,
NJ: Prentice Hall.
Orth, D., Davids, K., Araújo, D., Renshaw, I. and Passos, P. (2012) Effects of
a defender on run-up velocity and ball speed when crossing a football.
European Journal of Sport Science, iFirst article: 1–8; doi:
10.1080/17461391.2012.696712
Parfitt, G., Hardy, L. and Pates, J. (1990) Somatic anxiety and psychological
arousal: their effects upon a high anaerobic, low memory demand task.
International Journal of Sport Psychology, 26: 196–213.
Passos, P., Araujo, D., Davids, K., Gouveia, L., Milho, J. and Serpa, S.
(2008) Informationgoverning dynamics of attacker-defender interactions in
youth rugby union. Journal of Sports Sciences, 26 (13): 1421–9.
Pessoa, L. (2011) Reprint of: Emotion and cognition and the amygdala: from
‘what is it?’ to ‘what's to be done?’, Neuropsychologia, 49 (4): 681–94.
Pinder, R. A., Davids, K., Renshaw, I. and Araújo, D. (2011a) Manipulating
informational constraints shapes movement reorganization in interceptive
actions. Attention, Perception and Psychophysics, 73 (4): 1242–54.
Pinder, R. A., Davids, K., Renshaw, I. and Araújo, D. (2011b) Representative
learning design and functionality of research and practice in sport. Journal of
Sport and Exercise Psychology, 33: 146–55.
Pinder, R. A., Renshaw, I., Davids, K. and Kerhervé, H. (2011c) Principles
for use of ball projection machines in elite and developmental sport
programmes. Sports Medicine, 41 (10): 793–800.
Pinder, R. A., Davids, K. and Renshaw, I. (2011d) Learning design for use of
ball projection technologies with different skill levels in ball skill acquisition.
Paper presented at the Technologies in Sport Symposium.
Reed, E. S. (1996) Encountering the World: Toward an Ecological
Psychology. New York: Oxford University Press.
Renshaw, I., Oldham, A. R. H., Davids, K. and Golds, T. (2007) Changing
ecological constraints of practice alters coordination of dynamic interceptive
actions. European Journal of Sport Science, 7 (3): 157–67.
Renshaw, I., Chow, J. Y., Davids, K. and Hammond, J. (2010) A constraints-
led perspective to understanding skill acquisition and game play: a basis for
integration of motor learning theory and physical education praxis? Physical
Education and Sport Pedagogy, 15: 117–37.
Renshaw, I., Oldham, A. R. H. and Bawden, M. (2012) Nonlinear pedagogy
underpins intrinsic motivation in sports coaching. Open Science Journal, 5
(Suppl 1-M10): 88–99.
Runeson, S. and Andersson, I. E. K. (2007) Achievement of specificational
information usage with true and false feedback in learning a visual relative-
mass discrimination task. Journal of Experimental Psychology: Human
Perception and Performance, 33 (1): 163–82.
Searle, J. R. (1983) Intentionality: An Essay in the Philosophy of Mind.
Cambridge: Cambridge University Press.
Shaw, R. (2001) Processes, acts, and experiences: three stances on the
problem of intentionality. Ecological Psychology, 13 (4): 275–314.
Shim, J., Carlton, L. G., Chow, J. W. and Chae, W. K. (2005) The use of
anticipatory visual cues by highly skilled tennis players. Journal of Motor
Behavior, 37 (2): 164–75.
Stoffregen, T. A., Bardy, B. G., Smart, L. J. and Pagulayan, R. (2003) On the
nature and evaluation of fidelity in virtual environments, in L. J. Hettinger
and M. W. Haas (eds) Virtual and Adaptive Environments: Applications,
Implications and Human Performance Issues. Mahwah: Lawrence Erlbaum
Associates, pp. 111–28.
Thelen, E. and Smith, L. B. (1994) A Dynamic Systems Approach to the
Development of Cognition and Action. Cambridge, MA: The MIT Press.
Tucker, D. M., Derryberry, D. and Luu, P. (2000) Anatomy and physiology
of human emotion: Vertical Integration of brainstem, limbic, and cortical
systems, in J. Borod (ed.) Handbook of the Neuropsychology of Emotion.
New York: Oxford University Press, pp. 56–79.
Vallerand, R. J. and Blanchard, C. M. (2000) The study of emotion in sport
and exercise: Historical, definitional, and conceptual perspectives, in Y. L.
Hanin (ed.) Emotions in Sport. Champaign, IL: Human Kinetics, pp. 3–38.
van der Kamp, J., Rivas, F., van Doorn, H. and Savelsbergh, G. (2008)
Ventral and dorsal contributions in visual anticipation in fast ball sports.
International Journal of Sport Psychology, 39 (2): 100–30.
Vicente, K. J. (2003) Beyond the lens model and direct perception: toward a
broader ecological psychology. Ecological Psychology, 15 (3): 241–67.
Vilar, L., Araújo, D., Davids, K. and Renshaw, I. (2012) The need for
‘representative task designs’ in efficacy of skills tests in sport: a comment on
Russell, Benton and Kingsley (2010) Journal of Sports Sciences, 30 (16):
1727–30.
Wigton, R. S. (2008) What do the theories of Egon Brunswik have to say to
medical education. Advances in Health Sciences Education, 13: 109–21.
Williams, A. M., Davids, K. and Williams, J. G. (1999) Visual Perception
and Action in Sport. London: Routledge.
Withagen, R. and Michaels, C. F. (2005) The role of feedback information
for calibration and attunement in perceiving length by dynamic touch.
Journal of Experimental Psychology: Human Perception and Performance,
31 (6): 1379–90.
Withagen, R. and van Wermeskerken, M. (2009) Individual differences in
learning to perceive length by dynamic touch: evidence for variation in
perceptual learning capacities. Attention, Perception and Psychophysics, 71
(1): 64–75.
Index
accumulated effort 66–7, 70–2, 75–7
action fidelity 178, 324–5
adaptation 44, 47, 65, 77–8, 105, 113, 115, 117, 120, 136, 176, 241, 242, 245, 247–9, 251, 253–4,
256–7, 287–8, 299, 306, 314, 319, 325; biological 277; co-adaptation 105, 113, 120, 245, 247–9,
253–4; compensatory 241–2; functional 117, 242, 277; performance 306; psychobiological 77
adaptive flexibility 117, 230
affordances 181, 231, 237, 251–2, 256, 265, 283, 306–15, 322; attunement to 306, 308, 313
aging 270
anti-persistent 74–5, 77
anti-phase 48, 50–3, 113, 208–10, 212–14, 223, 280, 287, 289, 296; attraction 213; behaviour 212;
coordination 208, 210, 213, 280
association football see also soccer 56, 109, 111, 114, 119, 120, 160, 161, 162, 167, 171, 214, 229–31,
286–7, 323
attacker-defender dyad 107, 109, 111, 114, 233
attention 20–1, 25, 27, 31–3, 327
attention focus 66–7, 75–6
attractor 5–9, 13–16, 47, 67–8, 73, 88–9, 113, 150, 156, 208, 210–11, 223, 231, 262, 264, 266–9, 279,
289, 293, 296, 298–9, 301, 313, 239; fixed-point attractor 296, 299
attractor landscape 293, 295–9, 303
autocorrelation function 99, 125, 127–30, 132–5
avatars 182

basin of attraction 5–7, 14–15, 250, 263, 267, 269–70

badminton 210
baseball 322
basketball 55, 107–9, 114–15, 150–1, 156, 171, 195, 210, 212, 214, 223, 233, 248, 284, 286, 325
behavioural pattern 27, 31, 201, 217, 222
bifurcation 7–8, 11–13, 28, 31, 34, 47, 65, 67, 245, 257; definition 257; pitchfork 13; saddle-node 13
boccia 284–5
bootstrapping 151
breaststroke swimming 280
Brunswik's lens model 320, 322, 328
Brunswik's probabilistic functionalism 320–1
Butterworth low-pass filter 166

cognition 18–20, 26, 35, 47, 321, 323, 325, 327

classification 53–4, 98, 145–7, 149, 154
cluster analysis 146–51, 154–5, 157, 187, 301
collective variable see also order parameter 7–11, 46–8, 54, 65–6, 73, 77, 79, 106–8, 110–11, 121, 151,
160, 236, 266, 267, 270, 286
complexity 3–5, 7, 46, 69, 93–4, 98–9, 142, 151, 175, 179, 195, 241, 244, 254–5, 262, 322–4;
neurobiological 322; of behavior 5; system 99, 323
compound kinematic variable 161
configurations 137, 141, 217, 220–1, 223, 269–70; movement 269–70; playing 217, 220–1, 223; spatial
137, 141
contextual dependency 106, 109, 113, 271
control parameter 8–9, 11–16, 22–3, 32, 34, 44, 46–7, 55, 66, 78–9, 106–10, 112, 116, 167–8, 171, 246
constraints 11, 56, 69, 71, 108–10, 112, 115, 117, 161–2, 175–6, 180–2, 231, 335?>237, 242–8, 250–1,
254, 256, 261–5, 271–2, 282–4, 285, 287–9, 295, 299, 306–8, 310–13, 315–16, 319, 320, 322–3,
325–6; based-model 306; ecological 161, 231, 265, 307, 316, 323; environmental 11, 71, 242–7, 254,
256, 271, 282–4, 307–8, 310–13, 315, 320; informational 264; interacting 47, 232, 241–3, 246, 255,
262, 264, 277, 313; structural 109; task 56, 69, 108–10, 112, 115, 117, 162, 175–6, 180–2, 237, 245,
248, 250–1, 256, 261–3, 265, 272, 282, 285, 287–9, 295, 299, 307–8, 312, 319, 322–3, 325–6
cooperative effects 20, 65
coordination 19–35, 44–9, 50–2, 54–7, 65–6, 69, 71, 105–15, 120–1, 150, 156, 160–1, 176, 181, 184,
213, 223–33, 270, 326, 280, 282, 284, 286, 295, 299–300, 307, 309–11; attackerdefender 111;
dynamics 19–35, 44–9, 51–2, 54–7, 65–6, 69, 71, 160–1, 213, 223–33, 286, 309; interpersonal 23,
105–10, 114–15, 120, 160, 176, 181, 184, 231–33, 270, 326; interteam 105; intrateam 105, 271;
intrapersonal 113–14; patterns of 24, 28, 44–5, 50, 52, 54–5, 109, 112–16, 150, 156, 160, 176, 209,
231–3, 245, 261, 280, 282, 284, 295, 299–300, 307, 310–11; state of 107–8, 231–2; variable 21–3,
28, 106–8, 110–11, 121
coordinative structure 31, 46, 56, 105, 314, 316
coupling angle 49–51
coupling tendency 167
creativity 194, 201–5, 247, 253, 256, 261– 3, 265, 270–2, 316; tactical 201–3
cricket 241, 244, 248, 251, 253, 324–5, 329
critical fluctuation 3, 14, 213, 232–3
critical regions 107, 110–11, 113, 233, 246, 271
critical slowing down 8, 14
critical states 246–7
criticality 106, 246–7
cross-correlations 48–9
cross-modal interactions 184

dance 265
decentring distortion 164
decision making 31, 34–5, 47, 120, 160, 182, 190, 230, 271
degeneracy 117, 182, 230, 241–2, 244, 255–6, 262, 277, 282–3, 294, 306, 310–11, 314–16, 322
degrees of freedom: biomechanical 27, 47, 56; motor system 311, 316; performance 328; spectral 68,
70–1, 77
deterministic models 145
detrended fluctuation analysis (dfa) 86, 92–3, 101, 134, 279
dimension reduction tool 151
direct linear transformation 164
direct perception 184
displacement trajectories 233
distal cues 321
dorsal visual system 321
dyadic system 107, 109–10, 116, 120, 233–4, 237
dynamics: cognition 19; intrinsic 55, 75, 243–4, 253, 255–7, 316, 325; overlap 266–70; pattern analysis
191, 205; pattern forming 160, 161, 232; systems 3, 13, 87, 91, 100, 249

early specialization 255

ecological dynamics model 160, 229–32, 236–7, 262, 313–14, 323
ecological psychology 181, 251, 308, 311, 319, 320
ecological validity 322
effort tolerance 62
egocentric viewpoint 178–9
embedding theorem 89
emergence 3, 9, 18, 23, 25, 31, 75–7, 105, 112–14, 117, 120–1, 150, 167, 169, 171, 232, 245, 251, 253,
256, 265, 272, 281, 284, 296, 307, 313, 315, 316, 320, 326
emotions 327–30
entropy 85, 91–4, 97, 100, 138–9, 143, 265, 308; approximate entropy (ApEn) 92–4, 138; definition
91–2; geometric entropy index 308; Kolmogorov-Sinai entropy (KS) 92–3; sample entropy
(SampEn) 85, 94–5, 97–101
epigenetics 254
ergodicity 15, 264, 269–70, 294; breaking 264, 270
equilibrium 160, 209, 232, 284, 300, 308, 312, 316
experimental design 45, 180–1, 323, 325, 329
environment-agent systems 175
expertise 54, 149, 178, 180, 241–78, 306, 310, 313–16; acquisition of 241–3, 246, 250–1, 255–6; in
climbing 308–12, 336?>315; in kayaking 312, 314–15
expert performance 180, 241–4, 250, 255, 257, 278, 313–14, 330
exploratory 57, 307, 310; behaviour 57, 310; practice 307
exploration 14–15, 147, 175, 253, 261, 264–5, 267, 269, 272, 289, 301, 313, 329
exploratory breadth 264–5, 270
exponential function 299

fatigue-induced spontaneous termination point (FISTP) 66–8, 71, 73

fidelity 165, 178, 184, 324–6
field hockey 51, 136, 195, 287–8
finger waggling 208, 296–8
fluency 270, 308–9
fluctuations 3, 14–15, 26, 31, 46, 67–8, 70, 75, 77, 85–6, 92–3, 133–4, 165, 203–4, 213, 232–3, 246,
266, 272, 279–80; fluctuation process 203
functional behaviour 175, 184, 245, 261
functional synergies 105, 112–13, 120–1
fractal 75, 91–2, 99, 279
futsal 56, 114–15, 138, 141, 216, 231, 233–6

gait 46, 48, 50–1, 53–4, 99, 146, 279–80

genetic 241–5, 254–6, 295
goal-directed adaptive behaviour 180
goal orientation 327–8
goal-scoring opportunities 163, 167, 171–2

head-mounted display 176–7

hierarchically soft-assembled 261, 263, 270
Hilbert transform 52, 212, 215
HKB model 47, 296–8
homogeneity assumptions 293, 294
Hubert-Gamma method 151
Hysteresis 13, 31, 79, 149, 279
ice-climbing 282, 307–12
identification parameters 150
immersion 176, 178
individual-constraint coupling 277
individual differences in learning 293–4, 297, 299–303, 306–7, 309–10
individual environment coupling 283, 308, 311
in-phase 48, 50–2, 113–16, 208–9, 212–16, 223, 280, 283, 286, 311; attraction 114–16, 214, 216;
behaviour 210, 212; coordination 208, 214, 280, 283, 286, 311
information 18–21, 25–8, 33–5; action coupling 175; detection 308, 323–4; movement coupling 55,
322, 325
in situ 111, 176, 182, 323
instability see also loss of stability 19, 22–3, 26–7, 137, 169–71, 209, 232, 245, 252, 288, 213–16
intentions 21, 25, 27, 32, 35, 67, 73, 76, 78, 106, 210, 230, 264, 278, 280, 282–4, 307, 309, 321, 323,
328
inter-centroid distance 167, 169, 171
internal validity 321
interpersonal distances 105–7, 110–13, 121, 161, 183–4
intra-individual analysis 294
intra-individual variability 310–11
intra-technical error of measurement 166

Judo 328

Karhunen-Loève transformation 154

Kayaking 307, 314–15
Kohonen maps 151

learning 19–20, 25–34, 92; design 56, 321, 323–7, 329, 330; motor 14, 26, 254, 293–304; process 204,
250, 288, 299; task 28–30, 299, 324–5, 328–9; timescales of 293, 295–6, 299–301, 303, 306; rate
299; supervised 151
level of description 19, 20–4, 29
limbic system 321, 327
limitations in averaging 293, 294, 295, 299, 303
locomotion 53, 278–9, 301, 308
long-range correlations 91, 99, 133–5, 279
loss of stability see also instability 8, 31, 47, 67, 72–3, 76–8, 169, 171, 314
memory recall 309
mesoscopic protectorate 24
metastability 14, 26–7, 46–7, 244, 245, 250, 254, 256, 262, 306, 312, 316; performance region 47, 313;
region
movement form 262, 264–5, 272, 309, 313
movement output 309, 313
multifunctionality 6, 31
multi-layer-perceptrons 151
multistability 15, 27, 31, 65, 282, 284, 306, 310–11, 312, 315–16
337?>
naturalistic 22, 45, 179
neighbourhood preservation 152
networks 5, 11, 117, 121,152, 190, 194–7, 199–202, 205, 211, 217, 223; artificial neural 152, 190,
194–7, 199–202, 205, 211, 217, 223; social 117, 121
neurobiological system 11, 54, 56, 230, 242, 244, 254–5, 262, 264, 277, 278, 282–3, 285, 306, 308,
310, 323, 324
nodes 152, 154, 156–7
noise 92, 134, 128–9, 131, 134; brown 92; pink 92, 134; white 92, 128–9, 131, 134
nonlinearity 10, 15, 18, 78, 85, 87–90, 98, 101, 110, 113–14, 250, 254, 262
nonlinear dynamical systems 4–6, 87, 172
nonlinear pedagogy 253, 263, 323, 327, 329–30
non-stationarity 92, 98–9, 127–8, 134, 229–30, 236–7

optic array 181

optic flow 180–1
optical distortion 164
optimization 44, 261
order parameter see also collective variable 7, 11, 65–7, 69, 73, 79, 151, 160–1, 167–71, 211, 263–6,
270
order-order transition 167, 169
overlap order parameter 264, 266

parametric stabilization 21, 27

pattern formation 22, 282
pattern recognition 191, 194, 201–2, 205, 222–3
pedalo riding 300–1
perception-action coupling 180–2, 327
perceptual work-space 301
performance curves 295, 299
performance indicators 229–30, 237
performance variability 171, 245, 278
perception-action landscape 265, 270, 272
persistence 74–5, 77, 133
phase space 88–9, 211–12
phase relation 11, 29, 48, 51, 113–15, 171, 208, 213, 216, 223, 233, 279, 287, 296–7, 301, 311
phase transition see also transition 3, 8, 11, 13–14, 23, 30–2, 46–7, 52, 70, 107, 109, 111–12, 116, 150,
160–1, 212–13, 233, 297, 327, 329
pleiotropy 242, 255, 262
position data 195, 199–200, 214, 218, 221
postural stability 93, 301, 308, 310, 312, 315
power function 295
power law 295
practice 109, 117, 183, 241, 251, 256; deliberate 241, 256; design 109, 117, 321; organization 183;
representative 117, 251
principal components analysis 52–3, 266, 300
process analysis 194–5
proximal cues 321–2
psychobiological integration 62, 66, 76–7, 79
psychobiological model 64, 72–3

qualitative analysis 201–3, 221

quantitative analysis 201–2, 221, 223, 229

race walking 279

racket sport 205, 210
reciprocity 46, 182
recurrence quantification analysis (RQA) 89–91, 98
redundancy 97, 314, 316
rehabilitation 34, 44, 56–7
relative phase 26, 28, 30, 46, 48, 51, 55, 113–14, 116–17, 208–10, 212–16, 223, 233–4, 264, 279, 281,
296–7, 301
relative stretch index 167, 170–1
relative velocity 107–8, 110, 112–13, 161, 183
relaxation time 8–9, 15, 28, 68, 269–70
repeller 5–7, 14, 19, 25
replica symmetry breaking 263
representative design 180, 251, 319–23; experimental task 161; functional interaction 181; learning
design 323, 326–27, 329–30
residual analysis technique 166
rock climbing 307–9, 315
rollerball 301–3
rugby union 110–13, 161, 181–3, 321, 325–6
rugged potential landscape see also rugged energy landscape 14–15
rugged energy landscape see also rugged potential landscape 68
running correlations 110–12, 114

sampling rate 89–90, 94, 99

scale invariance 75
1/f scaling 86
selection via instability 26
338?>self-determination theory 327
self-efficacy 327–8
self-implemented network 155
self-interaction 5, 10
self-organization 3, 10, 24–5, 35, 46, 71, 114, 122, 160, 230, 250, 256, 271, 288, 299–301, 323
self-organizing maps 54–5, 152–4, 190–2, 205
sequential analysis 229, 236
sequential dependence 85–7, 89, 92
signal processing 165, 171
signature movements 155
simulation 190, 193–5, 201, 204
simulation of constraints 323–5, 327
skill acquisition 28–30, 45, 244–54, 262, 316, 323, 326
soccer see association football 55–7, 87, 135–7, 149, 155–6, 179, 191–9, 202–4, 219, 286–7, 295, 320,
323, 325–6
soft-assembled 77–8, 262–3, 265–70
spatial awareness 308
spatial feedback 301
spatial randomness 141
spectral density function 125, 129–31, 134
squash 194–8, 209–10, 213, 223
stability 7, 11, 19, 24–5, 27–8, 31–4, 55–6, 63, 72, 77–8, 107, 111–12, 114, 126, 296–7, 306, 311,
313–14, 316, 329
stability analysis 279
stationary 90, 92, 94, 98–9
stationarity 90
stochastic processes 126–8
surface area 214, 216
symmetry 12–13
symmetry breaking see also decision making 12–13
synergies 46
tactical pattern 195–7, 201, 221–2, 230
talent development 250, 252–4, 256
talent identification 250, 252–4, 256
task-specific device 56
team dynamics 149, 155
team centroid 55, 167, 214, 215, 286
team sports 105–7, 109–10, 112–14, 117, 120, 125, 134, 136, 141, 190, 195, 201–2, 205, 208, 210, 214,
230–2, 236–7
temporal feedback 301
tennis 52, 114, 155, 157, 179, 194–6, 209–13, 223, 322
time-discrete variables 145–6
time series 125–35, 138, 142, 294, 301
time series analysis 89–90, 134, 294, 301
topological control parameter 32
track and field 294
tracking operator 166

variability 56, 150, 155, 241, 263, 278–8, 293, 306–7, 310–11, 313–16; adaptive 306, 310–11;
compensatory 284; functional 279–80; movement 56, 155, 241, 263, 278–88, 293; movement pattern
56, 150, 306–7, 311, 314–16; movement system 56, 278, 313
vectors 94–6, 99 coding 49–51
ventral visual system 321
vicarious functioning 321
video-tracking 161
virtual positional coordinates 163
volition states 66, 71–2
volleyball 61, 149, 157, 194–5, 197, 199
two-landscape model 299, 301, 303

Waddington's landscape model 295

weight lifting 294