Complejidad Futbol
Complejidad Futbol
Complejidad Futbol
comportamiento táctico
http://hdl.handle.net/10803/404150
Directores
Robert Hristovski
Carlota Torrents Martín
2017
A Emili. Por tu generosidad, humildad y bondad eternas.
Simplemente gracias
Agradecimientos
A Fátima, por estar siempre ahí, en los momentos de mayor dificultad para animarme a
continuar, y en los de mayor euforia y alegría tras la publicación de cada uno de los
artículos. Gracias por tu infinita paciencia durante estos “tres años de sufrimiento”, como tú
dices; pero que también han servido para sentir tu admiración por mi trabajo y mi persona.
Empezamos juntos el que hasta ahora ha sido el proyecto más importante de mi vida para
en unos meses tener entre nuestros brazos al que pasará a ser el más importante de las
nuestras.
A Carlota Torrents, por hacerme el mejor regalo después de aquella clase de expresión
corporal que nunca olvidaré, decir que sí a dirigir mí trabajo final de máster, en fútbol…
que se ha acabado convirtiendo en una tesis doctoral. Gracias por la oportunidad que me
diste y por no dejar de constreñirme, siempre con única intención de aumentar mis grados
de libertad.
To Robert Hristovski, for gifting myself with your infinite knowledge. For always being
available to solve my doubts and ready to find a solution to any problem. Thank you for
supervising this thesis and for hosting me for the three months which I was in Skopje, one
of the best academic and personal experiences of my life.
A Bruno Gonçalves, porque tú has sido mucho más que un compañero, un verdadero
amigo. Te considero un director más de esta tesis, no sólo has ayudado a mejorar el trabajo
que aquí se presenta, sino que también me has ayudado a mejorar como investigador y
2
como persona. Gracias por todo lo que me has enseñado y por lo que todavía te queda por
enseñarme.
A Lorena Torres, porque aunque solo hayamos compartido cafés y conversaciones durante
la primera parte del partido, sin duda tu apoyo y compañía me sirvió para superar quizás los
momentos más complicados de este proceso. Gracias por hacer de directora cuando la jefa
no pudo estar presente y por compartir conmigo algunos de tus múltiples proyectos.
A Jaime Sampaio, por tu profesionalidad y por dar luz verde a todos los proyectos
propuestos. Gracias por tratar de buscar siempre la innovación científica e intentar
encontrar las aplicaciones prácticas a cada una de las investigaciones.
Al Reus Deportiu, Lleida Esportiu y C.D. Ebro de Zaragoza, a todos sus jugadores y
cuerpos técnicos por colaborar en el diseño del protocolo y recogida de datos para la
investigación. En especial a Emili Vicente, Josep Maria Martínez “Joma” y Jose Luis Arjol
por facilitar e intermediar con cada club para hacer posible la realización de los estudios
con jugadores de fútbol masculino profesionales.
A Àlex Gordillo, Pere Lavega y Cristòfol Salas por ofrecerme la oportunidad de trabajar
como profesor en las asignaturas de aprendizaje motor, teoría y práctica del juego motor, y
deportes colectivos I. Además a los dos últimos por ayudarme desde la subdirección y
dirección del INEFC de Lleida.
A Montse Sanmartí, por su complicidad, sus consejos y su ayuda durante los tres años de
beca con toda la parte burocrática y en el desarrollo de mis funciones como secretario de la
Comisión Académica de Doctorado.
A todos los miembros del grupo de investigación Sistemas Complejos y Deporte, por hacer
crecer y mejorar la investigación desde esta perspectiva. En especial a mis compañeros
Daniel Aragonés, Pablo Vázquez, Sergi Garcia y Agne Slapsinskaite quienes me han
ayudado desde el inicio a profundizar en el conocimiento de los sistemas complejos y
3
enseñarme la posibilidades que nos da la utilización de esta teoría en la investigación
científica. Gracias también, a Natalia Balagué, la coordinadora del grupo, por toda su
energía, su forma de transmitir conocimiento y su predisposición para iniciar siempre
nuevos proyectos que nos permiten seguir creciendo personal y académicamente.
To Alexander Aceski, for welcoming me into your home, introducing me to your friends
and giving me your friendship during my time in Skopje. I will never forget those three
months. Thank you Tejonazo! You will be a friend for life.
4
Resumen
5
Resum
6
Abstract
Complexity sciences and specifically non-linear dynamic system theory offer a set of
principles, concepts and methodological tools that have provided grateful advances for the
study of football. Recently, empirical evidence has shown the emergence of coordinative
patterns at different levels of analysis and the influence of different game or training task
constraints. Nevertheless, it is essential to go further in the understanding of the inherent
dynamics or temporal structure of these processes. For this reason, the aim of this thesis
was to identify the dynamics of tactical behavior emerging under the influence of
constraints in different game scenarios of male football. Two general (the principal
components and the lagged average cosine similarity) and four specific (the tactical actions
and field zones as well as the distance to the target and to the nearest opponent) variables
were treated as order parameters in order to characterize the structure and dynamics of
tactical behavior. Under the influence of the set of constraints derived from the
competition, the numerical imbalances in small-sided games or the restriction of the
interaction space during a training game situation, a hierarchical structure of behavior was
identified in which the player or team explores many different metastable tactical solutions.
From the results of the different studies, it can be concluded that the adequate manipulation
of the constraints allows shaping a structure and dynamics of behavior more regular and/or
stable or more variate and/or flexible depending on the goals.
7
Índice de contenidos
Agradecimientos ..................................................................................................................... 2
Resumen ................................................................................................................................. 5
Resum ..................................................................................................................................... 6
Abstract................................................................................................................................... 7
1. Introducción .................................................................................................................. 10
1.1. Bases teóricas para estudiar la complejidad en el fútbol ....................................... 10
1.2. Dinámicas de comportamiento táctico en deportes colectivos .............................. 15
1.2.1. Dinámicas interpersonales (jugador-jugador) ................................................ 16
1.2.2. Dinámicas intergrupales (equipo-equipo) ...................................................... 17
1.2.3. Dinámicas multinivel (jugador-equipo y/o equipo-jugador) .......................... 18
1.3. Rol de los constreñimientos en el entrenamiento .................................................. 19
1.4. Objetivo y estructura de la tesis ............................................................................. 21
2. Métodos ......................................................................................................................... 23
2.1. Participantes ........................................................................................................... 23
2.2. Adquisición y preparación de los datos ................................................................. 23
2.3. Análisis de los datos .............................................................................................. 26
2.3.1. Análisis Jerárquico de Componente Principales (hierarchical Principal
Component Analysis) .................................................................................................... 26
2.3.2. Análisis de superposición dinámica qd(t) (Dynamic overlap) ........................ 27
2.3.3. Análisis complementarios .............................................................................. 28
3. Estudio I - Soft-assembled Multilevel Dynamics of Tactical Behaviors in Soccer ..... 31
4. Estudio II - Timescales for exploratory tactical behaviour in football small-sided
games ................................................................................................................................... 44
5. Estudio III - Emergence of Exploratory, Technical and Tactical Behavior in Small-
Sided Soccer Games when Manipulating the Number of Teammates and Opponents ....... 53
6. Estudio IV - Dynamics of tactical behavior in football when manipulating players'
space of interaction ............................................................................................................... 69
7. Discusión ....................................................................................................................... 86
7.1. Síntesis de los principales hallazgos ...................................................................... 86
7.2. Aportaciones teóricas y metodológicas ............................................................... ..93
7.3. Aplicaciones prácticas ......................................................................................... ..95
8
7.4. Limitaciones y perspectivas de investigación futuras.......................................... 100
8. Conclusiones ............................................................................................................... 104
9. Referencias .................................................................................................................. 107
9
1.Introducción
Las recientes aportaciones de la teoría de sistemas dinámicos no lineales (TSDN) al
estudio de fenómenos relacionados con el deporte (Vázquez, 2017) han inspirado también
la investigación reciente en el fútbol. Concretamente, la propuesta teórica denominada
Coordination Dynamics (Kelso, 1995) define, explica y predice cómo se forman, adaptan,
persisten y cambian los patrones coordinativos en los seres vivos. Desde su aparición en la
década de los 80’ ha proporcionado conceptos y herramientas para verificar empíricamente
el comportamiento emergente en diferentes contextos relacionados con el aprendizaje
motor (Kelso, 1984) y el deporte (Araujo, Davids y Hristovski, 2006), y, más
recientemente, para el estudio del fútbol (Duarte, 2012). Este enfoque, junto con algunas
ideas de la psicología ecológica, permite abordar la comprensión de este deporte colectivo
examinando cómo múltiples variables pueden constreñir el comportamiento de los sistemas
neurobiológicos (deportistas) y de éstos actuando en conjunto (equipo/s), modificando los
diferentes estados estables de organización (patrones de comportamiento). En este primer
capítulo se introducen detalladamente los principios y conceptos clave que nos ayudan a
comprender el fútbol como un fenómeno complejo, dinámico y no lineal. En segundo lugar,
se revisa la literatura centrada en cómo estos principios y conceptos han servido para
estudiar la emergencia de comportamiento táctico y/o patrones coordinativos en deportes
colectivos a diferentes niveles de análisis. En tercer lugar, se recogen algunos trabajos que
proponen modificar estos comportamientos a partir de la manipulación de constreñimientos
relacionados con las tareas de entrenamiento. Y finalmente, se detalla la estructura y el
principal objetivo de la tesis.
Cualquier variable que actúe como fuente de información durante el juego podrá
influir en las dinámicas de comportamiento del sistema (Fajen & Warren, 2003). Desde la
perspectiva que se aborda en esta tesis, estas variables reciben el nombre de parámetros de
control (Haken, 1983). Los parámetros de control son representaciones matemáticas de los
constreñimientos que actúan sobre cualquier sistema, entre ellos los biológicos,
11
presionándolo hacia diferentes estados de organización o patrones coordinativos concretos.
La manipulación de los constreñimientos, especialmente aquellos relacionados con las
tareas de entrenamiento, condicionará la emergencia de patrones de comportamiento
determinados (Davids, Button, Araújo, Renshaw y Hristovski, 2006). La modificación o
variaciones de este tipo de variables o parámetros de control en la dirección adecuada
pueden producir cambios funcionales en el comportamiento en contextos competitivos o de
rendimiento.
12
probables o los que muestren mayor correlación entre sí, reciben el nombre de atractores,
por la atracción que generan dichos estados sobre las trayectorias que sigue el sistema
próximas al atractor. Por contra, los estados más inestables indican la baja probabilidad de
que emerja ese comportamiento o estado coordinativo. El paisaje potencial que se muestra
en la figura 1 representa el espacio de estados donde se puede apreciar el total de
configuraciones que realiza un jugador. En este ejemplo, la combinación de la distancia del
balón respecto a su portería (variable y) y la trayectoria de movimiento del jugador respecto
al balón (variable x) se considera como parámetro de orden. En rojo se pueden apreciar los
estados que atraen al sistema mientras que en azul se representan aquellos comportamientos
que son menos probables que realice el jugador. A modo de ejemplo se puede decir que
cuando el balón está cerca de la portería propia, el jugador tiende a moverse hacia el balón,
pero nunca en una trayectoria frontal (0º). Además se puede ver que cuando el balón está a
la altura del área, nunca se alejará del balón (180º), sin embargo este movimiento es
probable que se dé cuando el balón está en campo contrario, aunque la tendencia (atractor)
será a ofrecerse al portador del balón.
13
Figura 1. Espacio de estados (configuraciones) de un jugador y la dinámica asociada. El potencial
(Vi) indica un paisaje con muchos mínimos metastables (atractores) que atrapan temporalmente el
comportamiento del sistema.
14
concreta del espacio de estados disponibles, mientras que en escalas de tiempo mayores,
transita entre todos los estados metaestables (Bovier, 2006).
Por lo tanto, es de particular interés para las ciencias del movimiento y el deporte
descubrir cómo influyen ciertos constreñimientos (parámetros de control) en la dinámica
exploratoria y el comportamiento táctico emergente (parámetros de orden). Especialmente
aquellos de alta validez ecológica, es decir, aquellos ampliamente utilizados durante el
entrenamiento o que puedan aparecer durante la competición. Estos constreñimientos
desempeñarán un papel esencial para provocar cambios en el comportamiento individual y
colectivo, así como, para determinar los contextos favorables en los cuales deportistas y
equipos sean capaces de coordinarse funcionalmente con el fin de alcanzar un objetivo.
15
En el caso de los deportes colectivos de invasión masculinos se pueden identificar y
estudiar dinámicas de comportamiento táctico a tres niveles distintos1: A nivel
intrapersonal, donde el jugador trata de desarrollar patrones de comportamiento o
configuraciones de movimiento que le permitan dar solución a la heterogeneidad de las
tareas-problema que debe solucionar durante el desarrollo del juego. A nivel interpersonal,
en el que los jugadores del equipo tratan de coordinar sus acciones o movimientos para
conseguir objetivos compartidos. Finalmente, a nivel intergrupal, donde se analizan y
estudian los patrones de comportamiento o coordinación entre dos equipos enfrentados con
el fin de puntuar y evitar que lo haga el rival, para finalmente ganar el partido o no
perderlo. A continuación se describen los niveles interpersonales e intergrupales, ya que no
se han encontrado estudios específicos de fútbol a nivel intrapersonal.
Utilizando como referencia los avances en otras disciplinas como la vela, el criquet
o el boxeo (Araujo et al., 2006), el rugby fue uno de los primeros deportes colectivos en los
que se estudió la dinámica de comportamiento identificando parámetros de orden y control
relevantes que permitieran describir con precisión el comportamiento del sistema en sub-
fases específicas del juego (ej. 1 contra 1) (Passos et al., 2008). Bourbousson y
colaboradores (2010) estudiaron las dinámicas de comportamiento entre diadas de
jugadores de baloncesto utilizando la fase relativa (ver para revisión completa Kelso, 1995;
McGarry, 2006; Palut & Zanone, 2005) como parámetro de orden para identificar fases
(estados) estables e inestables así como transiciones en las relaciones de fase. Ellos
1 No se han encontrado estudios que sigan esta línea de investigación en deportes colectivos de
equipos femeninos. Por este motivo, todos los ejemplos que se describen en este apartado y los
siguientes hacen referencia al comportamiento de jugadores o equipos masculinos.
16
evidenciaron que los jugadores (jugador-compañero, jugador-oponente) tendieron a
moverse coordinados en fase, es decir, moviéndose en la misma dirección a la vez en el eje
longitudinal. Por otro lado, identificaron estrategias de marcaje al hombre así como los
momentos en los que alguna perturbación (finta, dribling, etc) provocó la pérdida de
estabilidad en el estado de coordinación estable obligando al sistema jugador-oponente a
coordinarse de otro modo en caso de no poder recuperar la estabilidad en el estado anterior.
El análisis de fase relativa también se utilizó para analizar la coordinación de movimiento
entre jugadores en situaciones específicas de 5 contra 4 más el portero permitidas por el
reglamento de fútbol sala, donde tras ser notificado, un jugador de campo puede sustituir al
portero para generar superioridad ofensiva (Travassos, Araújo, Vilar y McGarry, 2011).
Estos autores incluyeron el balón en el análisis y demostraron que los defensores están
mucho más acoplados al movimiento del balón mostrando una menor distribución de
frecuencias en torno a los 0 grados (coordinación en fase) y por lo tanto mayor estabilidad o
rigidez en su comportamiento en relación al balón. También se ha demostrado como, tras
un periodo aproximado de tres meses (trece semanas), jugadores de fútbol sala
desarrollaron comportamientos mucho más regulares (estables) que en el test inicial
(Sampaio & Maçãs, 2012). Tras el mismo periodo de tiempo, Silva y colaboradores (2016)
demostraron que el número de acoplamientos entre los jugadores aumentó
significativamente disminuyendo el tiempo necesario para actuar de forma sincronizada.
Otros factores que parecen resultar clave para competir de forma sincronizada son el nivel
del oponente y la densidad de partidos (Folgado, Duarte, Fernandes y Sampaio, 2014;
Folgado, Duarte, Marques y Sampaio, 2015). Estos autores demostraron que cuanto mayor
es el nivel del oponente más sincronizado actúa el equipo, así como que el hecho de jugar
partidos entre semana disminuye los niveles de sincronización de los equipos. También se
observó que la sincronización es mayor entre los jugadores que ocupan posiciones más
retrasadas y que disminuye a medida que la posición de los jugadores es más adelantada.
17
información relevante para conocer el comportamiento colectivo (Duarte et al., 2012). Las
dos primeras variables fueron procesadas también mediante la fase relativa en baloncesto
(Bourbousson, Seve y McGarry, 2010) demostrando que los equipos opuestos están
fuertemente atraídos hacia modos de coordinación en fase. Sin embargo, perturbaciones en
el eje latitudinal, es decir, periodos en los que se perdía este modo de coordinación,
anticipaban lanzamientos a canasta. Por otro lado, los estados estables en modos de
coordinación en fase relativos al índice de dispersión mostraban una clara relación con los
intercambios de posesión del balón, es decir, cuando un equipo tiene la posesión del balón
tiende a expandirse mientras el equipo oponente se comprime. En situaciones de juego
reducido en el fútbol diferentes análisis de correlación mostraron resultados similares
(Duarte, Araújo, Freire, et al., 2012; Frencken, Lemmink, Delleman y Visscher, 2011). Sin
embargo, un estudio reciente demostró que esto no pasaba en situaciones de 11 contra 11,
ya que cuando un equipo tendía a expandirse el otro también, sin que este índice de
dispersión pudiera asociarse a los momentos de posesión en partidos de fútbol (Moura et
al., 2016). Además, estos autores identificaron cuatro modos de coordinación. Las
frecuencias de modos de coordinación anti-fase (expansión del equipo atacante y
contracción del defensor) en el primer tercio de las secuencias ofensivas fueron
significativamente superiores en aquellas secuencias que finalizaron con lanzamiento a
portería. Teniendo en cuenta la velocidad promedio de subgrupos de jugadores se demostró
que durante la confrontación (defensores-atacantes, medios-medios, atacantes-defensores)
estos subgrupos tendían a competir acoplados (Memmert, Lemmink y Sampaio, 2017), es
decir que si los atacantes de un equipo aumentaban la velocidad de desplazamiento también
lo hacían los defensores oponentes. Destacar que esta sincronización disminuyó
significativamente en las tres confrontaciones durante la segunda parte. Estos mismos
autores identificaron una perturbación significativa en la diferencia de velocidad promedio
entre los diferentes subgrupos enfrentados en instantes previos al gol.
18
asociación (correlación) entre comportamientos identificados a diferentes niveles de
organización jerárquica del sistema y que se caracterizan por su evolución en escalas
espacio-temporales distintas (Hristovski et al., 2011). Así, en un nivel superior ciertos
comportamientos evolucionan más lentamente cambiando a escalas de tiempo más largas,
mientras que en un nivel inferior los cambios se producen más rápido por lo que esos
comportamientos permanecen estables en escalas temporales más cortas. Bourbousson y
col (2014) encontraron esta asociación entre las dinámicas colectivas e individuales
identificando que una pérdida de estabilidad de un estado de coordinación en fase
(sincronía) en el eje latitudinal entre los centros geométricos y el índice de dispersión de los
equipos inducía a la penetración del jugador con balón. Otro estudio a destacar es el de
Duarte y colaboradores (2013) que demostró como la baja irregularidad (estabilidad) a
nivel colectivo puede ser compensada y asociada a los valores más altos a nivel individual
como un tipo de compensación recíproca (Riley, Richardson, Shockley y Ramenzoni, 2011)
o sinergia multinivel. A pesar de estos ejemplos aislados todavía no ha sido demostrado
empíricamente esta relación en los diferentes niveles de organización del sistema.
20
1.4. Objetivo y estructura de la tesis
En este primer capítulo se han presentado las bases teóricas que permiten estudiar
las dinámicas de comportamiento apoyándonos en la TSDN, con un resumen de los
resultados derivados de investigaciones que se han centrado en el estudio de la
coordinación a diferentes niveles de análisis en el deporte colectivo, para finalizar con una
breve exposición del rol de los constreñimientos en la emergencia de comportamiento.
21
que pertenece al artículo titulado: “Timescales for exploratory tactical behaviour
in football small-sided games”, publicado en la revista Journal of Sport Science.
- Estudio III: Determinar cómo el número de compañeros y oponentes afecta al
comportamiento exploratorio de equipos profesionales y amateur en juegos
reducidos, correspondiente al artículo titulado: “Emergence of Exploratory,
Technical and Tactical Behavior in Small-Sided Soccer Games when
Manipulating the Number of Teammates and Opponents”, publicado en la
revista Plos One.
- Estudio IV: Idenficar cómo las restricciones espaciales afectan tanto al
comportamiento táctico exploratorio como a la interacción del jugador con
balón con su entorno y compañeros, dentro del artículo titulado: “Dynamics of
tactical behavior in football when manipulating players' space of interaction”,
enviado a la revista Plos One, actualmente en revisión.
22
2.Métodos
2.1. Participantes
23
detalles consultar Folgado, Duarte, Fernandes y Sampaio, 2014). Los datos recogidos a
frecuencias de 5 y 15 hercios (Hz) fueron filtrados y suavizados a 1 Hz (1 dato por
segundo) con el objetivo de facilitar la sincronización con los datos obtenidos a partir de la
observación.
Tabla 1. Ejemplo de matriz binaria que recoge las configuraciones de movimiento en cada
segundo de la unidad observada.
Tiempo (s.)
1 2 3 … m
Categorías
1 1 1 0 … 1
2 0 0 1 … 0
3 0 1 0 … 0
… … … … … 0
n 0 1 0 1 1
24
media y grande respectivamente, posicionados sobre el carril central del sector medio-
ofensivo y desplazándose rápido hacia delante y basculando ligeramente hacia la izquierda;
activándose en este caso las categorías pertinentes. Esta categorización se determinó a
partir de variables discretas (ej. zonas del campo o acciones tácticas) o de la segmentación
de variables continuas (ej. distancia al oponente más cercano o la velocidad de dispersión y
contracción del equipo).
Figura 2. Ilustración de las categorías (valores de variables) cuya combinación definen las
configuraciones de movimiento individuales (a: acción táctica, b: contexto de interacción, c: zona,
d: velocidad, e: distancia al oponente más cercano, f: distancia a la línea de fondo) y colectivas (g:
índice de dispersión, h: velocidad de dispersión/contracción, i: profundidad, j: amplitud, k: sector, l:
carril, m: velocidad longitudinal centro del equipo, n: velocidad latitudinal del centro del equipo).
25
la relación entre los componentes, es decir, los valores del parámetro de orden, modificando
así las propiedades de estabilidad del sistema (ver tabla 2).
26
PCA en variables binarias ver Joliffe, 2002). Los resultados de los componentes principales
(CPs) extraídos se utilizaron para determinar las categorías más destacadas que definieron
cada CP (es decir, el patrón de comportamiento). Las puntuaciones de los CPs se estimaron
como: 𝐶 = 𝑅−1 𝑆 donde R es la matriz de correlación de las variables iniciales
(configuraciones) y S es la matriz de estructura (Fulgosi, 1988). Esta matriz de estructura
permite visualizar la dinámica de evolución de patrones de comportamiento a través de las
correlaciones entre las configuraciones ordenadas en el tiempo y los CPs extraídos. La
principal característica del hPCA es la ausencia del supuesto de ortogonalidad de los CPs.
Por ello se utilizó el método de rotación Oblim Directo con delta = 0 (Westerhuis, Kourti y
MacGregor, 1998) para considerar la posibilidad de una estructura de orden superior en los
datos debido a la correlación entre los componentes extraídos en cada orden. Por lo tanto,
los resultados del primer orden de CP extraídos son tratados como una nueva entrada para
una mayor reducción de la dimensionalidad, estableciendo un orden superior de CP. El
procedimiento continúa hasta el momento en que no se detectan correlaciones significativas
entre los CPs extraídos. El número de CP significativos se determinó por el criterio de
Kaiser-Gutmann, pero únicamente se seleccionaron aquellos que acumularon el 80% de la
varianza explicada (Fabrigar, Wegener, MacCallum y Strahan, 1999).
Esta medida se puede utilizar como parámetro de orden para determinar las
propiedades dinámicas de comportamiento exploratorio (Hristovski et al., 2013; Torrents et
al., 2015). Captura la similitud promedio de las configuraciones a distancias de tiempo cada
vez mayores entre sí. Por lo tanto es capaz de detectar el ritmo o velocidad, y la amplitud
del comportamiento exploratorio en diferentes escalas temporales. El promedio de
superposición dinámica <qd(t)> se calculó como el coseno de similitud promedio de las
superposiciones o solapamientos entre las configuraciones a diferentes lapsos de tiempo, y
se ajustó a la siguiente ecuación, derivada de sistemas con una estructura jerárquica
intrincada (Sibani & Dall, 2003):
27
Donde qstat es el valor asintótico (es decir, estacionario) de la superposición dinámica, t son
las distancias o lapsos de tiempo y α es el exponente dinámico. Mientras que qstat detecta la
amplitud exploratoria de los jugadores, el valor de α captura el ritmo, o velocidad, de
exploración. Si los jugadores realizaran el mismo comportamiento durante todo el tiempo
de observación, la media de la dinámica exploratoria sería constante e igual a uno (<qd(t)>
= 1) para el total de lapsos de tiempo. Sin embargo, si los jugadores desarrollaran
comportamientos completamente diferentes en cada unidad de tiempo durante toda la
práctica, la dinámica exploratoria tendería a cero (<qd(t)> = 0), es decir, todos los
comportamientos serían completamente aleatorios entre sí. El hecho de que la relajación de
la curva teórica no sea exponencial permite detectar la escala temporal en la que el
comportamiento exploratorio se satura, y, por lo tanto, diferenciar dos escalas temporales
distintas dentro de la dinámica de juego.
𝑆 = − ∑ 𝑝𝑖 ln 𝑝𝑖
𝑖=1
28
donde pi es la probabilidad de cada configuración calculada dividiendo la frecuencia de
cada configuración por la totalidad de las frecuencias (𝑝𝑖 = 𝑛𝑖 /𝑁). Estas probabilidades
nos han posibilitado representar la red de probabilidad de pases entre jugadores, así como
calcular el potencial (cuencas de atracción) del espacio de estados representado a partir de
la combinación de dos variables de comportamiento. Los valores del potencial se
calcularon a partir de la siguiente ecuación (Balescu, 1975):
𝑉𝑖 = 𝑄 ln(𝑝𝑖 ⁄𝑁)
donde Q es la varianza estandarizada del sistema asumiendo que el ruido del sistema es
constante (Q = 1), y N es el número total de configuraciones.
29
Tabla 2. Contextualización y resumen de los diferentes parámetros y análisis utilizados en los diferentes artículos.
Contexto Parámetro de Nivel de Parámetro de Análisis Relación de variables que definen las
Artículo
de juego control descripción orden complementarios configuraciones de movimiento
Fuerza de atracción
(Trapping strength)
Juego
campo Número de Coordinación Dynamic Entropía (Shannon Acción táctica, contexto de interacción, zona
II
reducido oponentes intrapersonal overlap qd(t) entropy) del campo, velocidad
(SSG)
Paisaje potencial de
acción
Componentes
Juego Acciones del jugador con balón,
Número de principales y
campo Coordinación características de la acción con balón,
III oponentes y -
reducido interpersonal acciones de los compañeros sin balón,
compañeros Dynamic
(SSG) acciones de los oponentes
overlap qd(t)
Análisis de redes
Juego
Limitación del Coordinación Acción táctica, distancia al oponente más
campo Dynamic Dwell time
IV espacio de intrapersonal e cercano, distancia a la zona de marca,
amplio overlap qd(t)
interacción interpersonal velocidad
(LSG) Paisaje potencial de
acción
30
3. Estudio I - Soft-assembled Multilevel
Dynamics of Tactical Behaviors in
Soccer 2
2
Ric, A., Torrents, C., Gonçalves, B., Sampaio, J. y Hristovski, R. (2016). Soft-assembled
10.3389/fpsyg.2016.01513
31
ORIGINAL RESEARCH
published: 05 October 2016
doi: 10.3389/fpsyg.2016.01513
This study aimed to identify the tactical patterns and the timescales of variables during
a soccer match, allowing understanding the multilevel organization of tactical behaviors,
and to determine the similarity of patterns performed by different groups of teammates
during the first and second halves. Positional data from 20 professional male soccer
players from the same team were collected using high frequency global positioning
systems (5 Hz). Twenty-nine categories of tactical behaviors were determined from eight
positioning-derived variables creating multivariate binary (Boolean) time-series matrices.
Hierarchical principal component analysis (PCA) was used to identify the multilevel
structure of tactical behaviors. The sequential reduction of each set level of principal
Edited by:
components revealed a sole principal component as the slowest collective variable,
Duarte Araújo,
University of Lisbon, Portugal forming the global basin of attraction of tactical patterns during each half of the match. In
Reviewed by: addition, the mean dwell time of each positioning-derived variable helped to understand
Raoul Huys, the multilevel organization of collective tactical behavior during a soccer match. This
Centre national de la recherche
scientifique, France
approach warrants further investigations to analyze the influence of task constraints
John Komar, on the emergence of tactical behavior. Furthermore, PCA can help coaches to design
University of Rouen, France
representative training tasks according to those tactical patterns captured during match
*Correspondence:
competitions and to compare them depending on situational variables.
Angel Ric
[email protected] Keywords: exploratory behavior, interpersonal coordination, complex systems, collective variables, principal
component analysis
Specialty section:
This article was submitted to
Movement Science and Sport INTRODUCTION
Psychology,
a section of the journal
In team sports settings, the interaction between team players and environment gives rise to
Frontiers in Psychology
interpersonal coordination movements that dynamically arise during the game. Interpersonal
Received: 04 June 2016 interactions are thought to be non-linear and some studies have shown this explicitly (Schmidt
Accepted: 20 September 2016
et al., 1990; Richardson et al., 2007). Under conditions of non-linear interpersonal interactions,
Published: 05 October 2016
behavioral patterns are hypothesized to be spontaneously organized at the macroscopic level
Citation: as a result of functional grouping of components which are temporarily assembled, through a
Ric A, Torrents C, Gonçalves B,
self-organization process (Gréhaigne et al., 1997; McGarry et al., 2002; Araujo et al., 2006). Many
Sampaio J and Hristovski R (2016)
Soft-assembled Multilevel Dynamics
investigations have studied the emergence of coordination patterns on individual (Travassos et al.,
of Tactical Behaviors in Soccer. 2012), dyadic (Sampaio and Maçãs, 2012) and collective levels of analysis (Silva et al., 2014).
Front. Psychol. 7:1513. Notwithstanding, there is little literature that attempts to identify the theoretically existing relation
doi: 10.3389/fpsyg.2016.01513 among these levels of game constraints (Bourbousson et al., 2014).
During soccer matches, there is no particular and fixed property related to the existence of multiple separated timescales.
predetermined movement or social coordination to be executed At quick timescales, the system appears to be in equilibrium,
for the achievement of a performance goal (score a goal) or i.e., in a temporarily stable state, but explores a limited part
sub-goal (avoid conceding one). Hence, there is a lack of of its available state space. At longer timescales, however, it
convergence of the exploratory behavior toward a pre-defined undergoes transitions between such metastable states (Bovier
intra- or inter-personal configuration. This situation allows very and Den Hollander, 2016). These previous studies allow us to
subtle and consequently unpredictable interactions between the hypothesize that the characteristic timescales of team behaviors
environmental information flow and the performer’s organismic could define the interpersonal coordination at different levels of
constraints to decide which particular affordance, from the set system organization and help to understand the nested structure
of affordances, will be realized at each moment. Gibson (1979) of the dynamics of tactical behaviors.
originally used this term referring to opportunities for action. The The dynamical systems approach has established the bases for
interrelatedness of affordances (Bruineberg and Rietveld, 2014) understanding the emergence of interpersonal coordination in
allows us to understand the emergence of behavior at different team sports requiring the identification of relevant coordination
levels of game constraints. Affordances offered by situational variables [described as collective variables or modes and
variables, such as quality or expertise of the opponent, can their amplitudes, i.e., order parameters (Haken, 2006)]. These
constrain ball possession strategies (Lago, 2009). In the same collective variables describe the spatiotemporal pattern of
way, having ball possession far from or close to the opponent’s interpersonal coordination and the changes of coordination
goal typically constrains the emergence of specific tactical task (Riley et al., 2011) that occur in response to game constraints.
solutions, for example, deep mobility of players while keeping In a team sport setting, the relative phase between two oscillating
the ball far from the opponent’s goal, or, choosing a pressing components has been frequently used as a collective variable to
strategy when the ball is lost in the opponent’s field. All these capture the macroscopic order or coordination of the dyadic
interconnected situations are characterized by their evolution system. Bourbousson et al. (2010), through the relative phase
on different timescales (Mendes et al., 2007; Ric et al., 2016). between the oscillations of mean player dispersion around the
This study of Ric et al. (2016) showed a clear separation of team geometrical center (i.e., stretch index) of two confronted
dynamical timescales in small-sided games. Slower evolution basketball teams, found that team players tended to expand and
was characteristic of ball possesion or dispossesion and quicker contract together when both teams were moving from basket to
evolution was characteristic for specific actions belonging to the basket showing clear inter-team coordination patterns. Travassos
former states. This state of affairs clearly reflects the model- et al. (2011) showed that interpersonal coordination between
independent definition of dynamic property of metastability1 as a defenders was stronger than between attackers demonstrating
1 Over time several models were proposed that can account for metastability:
that attacking players show greater variability in the movement
Stochastic hopping within the enduring potential landscape (a simple bistable
patterns performed during futsal matches. Gonçalves et al. (2014)
potential scenario Eyring, 1935; Freidlin and Wentzell, 2012) or a much more applied relative phase analysis between the mean position of
involved quenched disorder scenario (Cugliandolo and Kurchan, 1993), hopping specific group players (i.e., team geometrical center) over time
due to the changing stability of the fixed point attractors (minima become maxima during a soccer match. They found that stronger coordination
and vice versa, i.e., an annealed disorder scenario Badii and Politi, 1999), transient movements appeared between the consecutive line forces,
dynamics in a homoclinic channel (Kelso, 1995), and transient dynamics in a stable
heteroclinic channel (Rabinovich et al., 2008) as well as the self-organized criticality
that is, defender-midfielders and midfielders-attackers. Other
scenario (Bak et al., 1987; Kello et al., 2008). techniques, commonly used to capture coordination patterns,
All these models claim descriptions of metastability. However, they differ. like cross-correlations and vector coding, have recently been
Quenched and annealed disorder models differ in the timescales ascribed to the studied (Moura et al., 2016).
interactions (constraints) among the dynamic degrees of freedom of the system.
Whereas these studies focused on the coordination between
Both systems are multistable in the sense of having a vast number of well-defined
fixed point attractors i.e., minima. While in the quenched systems interactions are two oscillating variables, principal component analysis (PCA) has
assumed to be to a good approximation constant with respect to the dynamical
degrees of freedom, in annealed disorder systems they change on the same dynamics comes from the enduring or changing landscape or any other form of
timescale. Hence the difference in the behavior of these systems: enduring vs. quick dynamics irrelevant. The enduring landscape scenario, in this case, becomes
changing potential landscapes. These two models are stochastic in a sense that the a useful tool to represent the dynamics of the configurations (see Hristovski et al.,
interactions between the dynamic degrees of freedom are random. Real complex 2011; Ric et al., 2016). Of course, alternative models may exist as is the case in
systems, such as teams of players, may even be a combination of both these polar many areas of science. Here we find the scenario of hopping dynamics on an
types, showing both types: constant and quickly changing interactions. This would enduring landscape as being the most parsimonious at present, although a mixed
produce landscapes with some enduring and some changing basins of attraction. quenched-annealed disordered scenario would be rather more realistic.
Contrary to these models, homoclinic and heteroclinic channel metastability are Bovier and Den Hollander (2016) define metastability in a model-independent
purely dynamic and no stochastic influence is needed although, as was the case way as a property related to the existence of multiple separated timescales. At
in the heteroclinic channel scenario, noise, i.e., stochasticity, can be present. The short time-scales, the system appears to be in equilibrium, but in fact, explores
self-organized criticality type of metastability arises as a domino effect, upsetting only a limited part of its available state space. At longer timescales, it undergoes
minimally stable states. transitions between such metastable states. This definition satisfies the models
On the other hand the data collected on the observational timescale enables with enduring and changing potential landscapes as well as the homoclinic and
the calculation of the probabilities of configurations. This further enables the heteroclinic channel models of metastability. However, it does not satisfy the
presentation of the configurational probabilities as coming from the well-known metastability of the self-organized criticality type in which timescales are not
statistical mechanics relation (via the maximum entropy method) between separated and independent. This example shows that the definition of metastability
configuration probabilities and minima depths. Of course, such a depiction of is somewhat flexible. Nevertheless, in all of these models there exist at least two
probabilities, taken on the observational timescale, makes whether this quicker timescales (for convenience called quick and long-term).
been successfully proposed as a multivariate statistical method Data Collection and Preparation
for analyzing high dimensional (i.e., >2) movement coordination Positional data from the outfield players were collected using
patterns (Daffertshofer et al., 2004; Forner-Cordero et al., 2005). 5 Hz GPS devices (SPI Pro, GPSports, Canberra, Australia).
In essence, PCA reduces the dimensionality of large data sets, Each one was placed on the upper back of the players. Latitude
obtaining a smaller number of subjacent components that explain and longitude coordinates were exported from the units and
most of the variance and summarizes the information of the computed using dedicated routines in Matlab R2014b software
original variables. PCA has been used in the past three decades (MathWorks, Inc., Massachusetts, USA; for complete guidelines,
as an efficient method of defining the essential, i.e., collective, see Folgado et al., 2014).
variables in a range of complex dynamical processes such Configurations of tactical behaviors expressed in team-related
as protein folding (Matsunaga et al., 2007; Hayward and De positioning-derived variables were computed to determine the
Groot, 2008; Maisuradze et al., 2009) and brain dynamics (Jirsa structural and dynamic characteristics of the team. Eight
et al., 2002). In sport science, PCA has allowed identifying the collective measures were processed from the outfield players:
performance of an interpersonal precision task (Ramenzoni et al., stretch index, team length, team width, longitudinal position of
2012) to determine the impact of different training programs team geometrical center (x axis) and lateral position of team
on cardio-respiratory coordination (Balagué et al., 2016), and to geometrical center (y axis) (Duarte et al., 2013). Whereas team
identify different dancing coordination patterns depending on geometrical center (also named centroid) and stretch index have
concrete task constraints (Bronner and Shippen, 2015; Torrents been previously defined, team length, and width were calculated
et al., 2015). In soccer, PCA has been applied to capture player as the difference between the maximum and minimum positions
positional patterns and their variability (Barros et al., 2006). of players in the field’s longitudinal and latitudinal dimensions,
Moura et al. (2015) used this method by applying it to each respectively, in each time unit. Also included was the speed of
mean player position in the field in order to obtain the collective displacement (meters per second) at which team geometrical
organization during the matches of subsequent rounds in the center was moving (differentiated lateral and longitudinal axes)
European Championship and the positional variability of all and the speed of contraction and expansion in order to capture
players. Despite these investigations, the use of PCA to detect the how the team was behaving (Bourbousson et al., 2010). This last
pattern-forming dynamics on a collective level over the course of variable was calculated by differencing the each data point of the
a competitive match and the relation between the adjacent levels stretch index with the previous one.
of game constraints remains unexplored. In this sense, several Data were down sampled into 1 Hz in order to define the team
authors have emphasized the need to analyze the timescales configuration at each second. A two-step cluster analysis was
of tactical behavior that unequivocally define the multi-level performed to determine automatically the boundary values of
game dynamics because nested levels in dynamical systems each positioning-derived variable. To determine the zone around
are inevitably connected with characteristic timescales of their which the geometrical center was located, the field was divided
evolution (Haken, 2006). Therefore, the aim of this study was to into four sectors and three corridors (Costa et al., 2011; Sarmento
identify the tactical patterns and the timescales of positioning- et al., 2013). Finally, 29 categories were determined to create
derived variables that define the patterns during a soccer match, multivariate binary (Boolean) time series matrices (Table 1). A
allowing understanding the multilevel organization of tactical value of 1 was ascribed to the active categories and value of 0
behaviors. to the inactive ones, representing the full configuration (tactical
pattern) during the same time interval of 1 s. Each 1-s window
MATERIALS AND METHODS was defined as a 29-component binary vector (column). A total
of 2700 configurations (45 min × 60 s) for each half were finally
Participants and Procedure obtained.
Twenty male professional soccer players from the same team (age
= 22.8 ± 4.4 years; professional playing experience = 6.9 ± 4.1 Data Analysis
years) participated in an official pre-season match across the Principal Component Analysis
two halves against another team. Goalkeepers participated in Hierarchical PCA (hPCA) was performed to define the collective
the match but were excluded from the analysis. Due to the (state) variables of the coordinated behaviors of teams on
characteristics of pre-season matches, the ten starter outfield separated levels of organization. The higher levels were
soccer players participated in the first half. Eight of them hypothesized to be dominantly defined by categories that possess
were substituted at halftime and the other two in the middle larger dwell times. The initial data were multivariate binary
of the second half in order to ensure a more constant team matrices: 29 categories × 2700 time-ordered configurations (for
formation and avoid the effects of cumulative fatigue (Mohr the suitability of using principal components analysis with binary
et al., 2005). The match was played on a natural turf pitch variables see Joliffe, 2002). The initial system of PCs was rotated
(100 × 68 m) following the official soccer rules. The players under the Direct Oblimin method with δ = 0 (Westerhuis
analyzed belonged to the visiting team. The final score of the et al., 1998) to consider the possibility of a higher-order
match was 1-0. All players provided written informed consent structure in data because of the correlation between the extracted
to participate in the experiment. The local institutional Research components in each order. The number of significant principal
Ethics Committee approved the study, which also conformed to components (PCs) was determined by the Kaiser-Gutmann
the recommendations of the Declaration of Helsinki. criterion but only those that cumulatively accounted for ≥80%
TABLE 1 | Categories of each collective positioning-derived variable with solutions. Principal component scores (see Joliffe, 2002) were
the boundary values, identifying a total of 29 categories. used to determine the most salient categories which defined each
1–2: Stretch index <16.75 m Small SIS
principal component (i.e., tactical pattern). Principal component
>16.75 m Large SIL
scores were estimated as: C = R−1 S, where R is the correlation
matrix of time-ordered game configurations and S is the PC
3–6: Speed of spread < −0.6 m/s Quick contraction QC structure matrix, i.e., the correlations between 2700 time-ordered
−0.6 to 0 m/s Slow contraction SC game configurations and PCs (Fulgosi, 1988).
0–0.6 m/s Slow expansion SE The structure matrix was used to visualize the dynamics of
>0.6 m/s Quick expansion QE team configurations in the space spanned by the extracted PCs.
Finally, to compare the structure of first-level PCs between both
7–10: Length <27.72 m Small LS halves, Tucker’s congruence coefficient was used to determine
27.72–35.4 m Medium LM the degree of similarity between principal components (Lorenzo-
35.4–42.71 m Large LL Seva and ten Berge, 2006).
>42.71 m Very large LVL
Analysis of Timescales of Positioning-Derived
11–14: Width <34.49 m Small WS
Variables
34.49–43.68 m Medium WM
The aim of the analysis was to identify the dynamic properties of
43.68–53.59 m Large WL the game assessed by the associated dwell (waiting or residence)
>53.59 m Very large WVL times of positioning-derived variables. Dwell times assess how
long a certain variable remains in a well-defined state before
15–18: Sector <25 m Ultra-defensive UDS
leaving it and switching to another. Hence, they are useful in
25–50 m Mid-defensive MDS
this respect since their averages show the speed of evolution of
50–75 m Mid-offensive MOS
the variable in question. The shorter the average dwell time the
>75 m Ultra-offensive UDS
quicker the evolution (changing the states) and vice versa. The
19–21: Corridor <21.33 m Right RC pooled averages of the active categories (i.e., with 1 ascribed) were
21.33–42.66 Central CC calculated in order to find out the average time, in seconds, that
>42.66 Left LC
the team was dwelling on each positioning-derived variable.
In addition, the video-recorded match was analyzed to
22–25: Longitudinal speed < −1.06 m/s Quick drop back QDB calculate the timescale (i.e., average dwell time) on which ball
of team center possession switched from one team to another. The beginning
−1.06 to 0 m/s Slow drop back SDB of ball possession started when: the goalkeeper took the ball in
0–1.06 m/s Slow forward move SFM his hands, the second touch of the player winning back the ball,
>1.06 m/s Quick forward move QFM or the first touch of any teammate after a pass, deflection, or
clearance by the teammate who first touched the ball (Castellano,
26–29: Lateral speed of < −0.6 m/s Quick to the right QR
2008). It is important to note that when play was stopped due to
team center
an interruption (e.g. corner kick, fouls, goals, etc.) ball possession
−0.6 to 0 m/s Slow to the right SR
was assigned to the team responsible for restarting the game.
0–0.6 m/s Slow to the left SL
Due to non-Gaussian distributions of the dwell times, the
>0.6 m/s Quick to the left QL
non-parametric Kruskal-Wallis test was performed in order to
compare dwell times of variables to identify possible slow-
and fast-evolving processes. It was conducted to compare the
of the explained variance were further selected (Fabrigar et al., timescales of all positioning-derived variables between both
1999). halves. The partial eta square (pη2 ) value is reported as a
The hPCA differs from the ordinary PCA in that it is measure of effect size and is interpreted according to the
free from the assumption of orthogonality of the principal following criteria: significant but weak (ES ≤ 0.04), moderate
components. In other words, it is based on more general (0.04 < ES ≤ 0.36) and strong (ES > 0.36) (Tabachnick and
assumptions of non-orthogonality of correlation/covariance Fidell, 2007). Comparisons between halves were assessed via
matrix eigenvectors (PCs) and treats orthogonality as a special standardized mean differences, computed with pooled variance
case. Just as the ordinary PCA is based on reducing the original and respective 90% confidence intervals. Dwell time data were
high dimensional data to a lower dimensional space based on log-transformed to reduce bias arising from a non-uniformity
correlations/covariance of observed data, the hPCA treats the error. Uncertainty in the differences was expressed as 90%
correlations of the extracted first order PCs as a new input to of confidence limits (CL) and as probabilities that the true
further reduce the dimensionality to fewer and higher-order effect was substantially greater or smaller than the smaller
PCs. The procedure continues until the moment when no practical difference at the threshold of 25% (declared possible).
further meaningful correlations/covariance between the PCs is These probabilities were used to make a qualitative probabilistic
detected. If the first order PCs share little meaningful variance, mechanistic inference about the true effect. The scale was as
the results of hPCA are comparable to the orthogonal PCA follows: 25−75%, possible; 75−95%, likely; 95−99%, very likely;
>99%, most likely. A difference was assessed as being unclear small stretch index which slowly shrank, with the team centroid
if the CI overlapped both substantially positive and negative located in middle offensive sector and central corridor while they
thresholds by ≥5%. The Cohen d effect size with 90% CL was were dropping back (see Figure 2). The percentage of the total
calculated using pooled standard deviation for comparisons and variance that explained this pattern in the first part was twice that
the magnitude ranges for mean differences were: 0–0.2 trivial; > of the second. These patterns could describe the main positioning
0.2–0.6 small; > 0.6–1.2 moderate; > 1.2–2 large; > 2 very large structure when they were defending in the first half. On the
(Hopkins et al., 2009). other hand, the most frequent tactical pattern in the second
half (PC1) had a significant degree of similarity (rc = 0.78)
with PC9 of the first half. They are defined as defensive patterns
RESULTS and were characterized by a small stretch index, related to the
The Primary Level of Tactical Patterns medium length and small width, with the team centroid located
PCA initially revealed 22 principal components. Twelve of in the middle defensive sector and right corridor. The players
these represented 80.8 and 80.91% of the total variance for were slowly reducing their effective playing space and dropping
the first and second half, respectively. The high positive and back in PC1 but this was not clearly defined in PC9. Defending
negative component scores on these twelve principal components patterns were the most stable patterns in both halves, but,
represented those tactical variables (categories) that occurred whereas in the first half the team was located in the opponent’s
and decayed jointly (Figure 1). The salient structure of PCs was field for defending, in the second they were placed mostly in
defined by those categories with a high absolute (positive or their own field. The congruence coefficient between PC2 in both
negative) component score. Scores close to 0 indicate that the halves showed a significant similarity between them (rc = 0.73).
corresponding category does not contribute or hardly contributes These offensive patterns were defined by a large stretch index,
to the PC. High positive scores (>1) informs that those categories with the team keeping the distances between the players and their
are simultaneously active while high negative scores (< − 1) geometrical center mostly stable. The team was moving forward
report that the corresponding categories are active together but and quickly to the left, with the team centroid located in central
are active when the positive ones are inactive and inversely. offensive sector and central corridor. The third PC of the first half
The structures of PCs showed significant congruence between was defined by a small but slowly increasing dispersion of players,
PC1 of the first half and PC3 of the second (rc = 0.70). They can with a long length and medium width. Besides, the location of
be defined as defensive patterns because of the medium width and the team center was around the ultra-offensive sector and it was
FIGURE 1 | Illustration of ten outfield players (numbered gray circles) and the eight positioning-derived variables: (a) stretch index, (b) speed of
dispersion, (c) length, (d) width, (e) sectors, (f) corridors, (g) longitudinal speed of team center, and (h) latitudinal speed of team center.
FIGURE 2 | Larger values than 1 (blue dashed line) and/or lower than −1 (red dashed line) refer to the tactical behaviors that defined the primary-level
of principal components.
slowly moving forward to the right showing the dominance of persistent patterns over time. The congruence coefficients of
the game in the opponent’s field. Its highest congruence value these highest-level PCs reached a value of 0.69.
(rc = 0.61) is related with PC8 of the second half, explaining The Kruskal-Wallis test showed a significant effect between
two similar offensive patterns for each half. In addition, PCs also the timescales of all positional variables including ball possession:
showed significant congruence coefficients: PC12 of the first half [H (8, N = 3156) = 845.8; X 2(8) = 513.51; P < 0.0001; pη2 =
with PC11 of the second (rc = 0.70) PC11 with PC10 (rc = 0.76) 0.314] for the first half and [H (8, N = 3257) = 911.66; X 2(8) =
and PC10 with PC5 (rc = 0.79). The rest of combinations did not 527.63; P < 0.0001; pη2 = 0.324] for the second. Table 2 shows
reach coefficients of congruence higher than 0.70, however their means and standard deviations of timescales for each variable
structure can be consulted in Figure 1. differentiating both halves. The probability of finding true
differences between halves was possible for timescales of stretch
Metastable Dynamics of Tactical Patterns index and ball possession, with a positive and negative effect,
Figure 3 represents the loadings (correlations) of the time respectively. An unclear tendency was reported for the dwell
configuration vector on/with the PCs. Note that the configuration time of the centroids allocated in the corridors. Finally, trivial-to-
vector dwells for some time projecting dominantly on a few PCs unlikely differences between the first and second half were found
and then, on a longer timescale, quickly transits to form another in the timescales for the remaining collective positional variables.
temporarily stable set of dominant projections. These sets of Multiple comparisons between the means of all eight variables
dominant projections of configurations on the PCs represent the (see Table 3) allowed identifying three groups of variables that
attractors of the team dynamics. The dominance of projections possessed different characteristic timescales. Those variables that
refers to the dominant PC content of the configurations, while the evolved over the timescales of tens of seconds, i.e., stretch index,
dwell time of the dominant projection to its stability or attraction sectors, and corridors, showed statistical differences with all other
strength (the longer the dwell time the greater the stability or variables, whereas no significant differences were detected when
attraction strength and vice versa). compared between them (stretch index vs. sectors p = 1.00;
As can be seen in Figure 3, during the first 5 min the system stretch index vs. corridors p = 1.00; sectors vs. corridors p = 1.00
was projecting mostly on PC1, PC5, and PC9 defined by a small for the first half and p = 1.00; p = 1.00; p = 1.00 for the second).
stretch index that slowly contracted and team positioning was Similarly, significant differences were absent in comparisons of
slowly dropping back. Although the locations of the team center those variables that quickly evolved, i.e., over the timescale of a
were in different sectors, all these patterns were defensive. Then, few seconds. For the first half: speed dispersion vs. longitudinal
the switches between these PCs at the beginning of the match speed p = 0.05, speed dispersion vs. lateral speed p = 0.272
were brought about by the small differences in some of the and longitudinal vs. lateral speed p = 1.00; and for the second
categories that defined each one of these PCs. It is interesting half: speed dispersion vs. longitudinal speed p = 0.11; speed
to note that the dominant projection of the game configuration dispersion vs. lateral speed p = 1.00 and longitudinal vs. lateral
when the goal was scored (second 1148) corresponds to PC10. speed 8 p = 1.00. The significant differences of length and width
The small dispersion of players was not altered much more, variables with the rest but not between them (for the first half
but the team center quickly moved back to the right side from p = 0.272 and for the second half p = 0.375), lead us to think
the central corridor of the middle defensive sector. That pattern that another intermediate timescale, which evolved over several
clearly defines the defense of the opponent’s counterattack. seconds, could exist.
After the goal, the game configuration transited to a dominant
projection on PC11 defined by the large dispersion, slowly
growing up from the middle of the field quickly moving to DISCUSSION
the opponent’s midfield. Finally, the team was mostly stable
performing PC2, previously defined, for 7 min before transiting The current study explores the soft-assembly of tactical patterns
to another less stable tactical pattern. and the timescales of positioning-derived variables that define
them during a soccer match, allowing understanding the
Multilevel Organization of Tactical Patterns multilevel organization of tactical behaviors as defined by the
First-level PCs correlations were then subjected to a further timescales of evolution of collective patterns. For this goal,
higher-order analysis revealing second-level PCs. Four PCs in a hPCA and the dwell time measure were used to identify,
the first half of the match and five in the second half were respectively, nested correlated movement configurations as well
identified. Furthermore, the second-level correlated structures as the characteristic timescales of their change (Hristovski
systematically produced a sole PC on the third-level for the first et al., 2013). The main results afford that most stable/persistent
half (see black bars in Figure 4), whereas two PCs were extracted movement patterns were well-related to defensive behaviors. In
from the second half. Therefore, by a further iteration of the fact, team contraction, expressed through lower stretch index
procedure a fourth-level PC was extracted only in the second values and the drop-back of the team, are suggested as being key
half (see gray bars in Figure 4). The lower-order PCs are more indicators of team behavior in the defensive phase (Bourbousson
sensitive to detailed changes in impinging game constraints, et al., 2010; Frencken et al., 2011). These results do not indicate
while the highest-level PC (third-level PC in the first half and that the team performed more defensive patterns, rather the
fourth-level PC in the second) captures the most robust and observed patterns are likely to be most stable (Travassos et al.,
stable structure of associations within the data, defining the most 2011; Ric et al., 2016). The degree of similarity between the PC2
FIGURE 3 | Correlations (loadings) of time-ordered game configurations with the first-level PCs.
extracted for both halves elucidates that the offensive playing style would happen must be considered in the training task designed
of the team is clearly defined. In this sense, it can be said that (Barreira et al., 2014).
independently of the players, the playing style proposed by the In this study the hypothesized soft-assembled multilevel
coach strongly constrained the emergence of concrete tactical dynamics (Hristovski, 2012; Hristovski et al., 2013) in a soccer
behaviors. These patterns, which explain the larger percentage match have been corroborated. The salient correlated patterns
of variance, describe collective behavior when the team or the (PCs) in the first level of analysis revealed tactical behaviors
opponent perform long ball possessions, by stabilizing their on higher levels. This finding helps to understand the nested
behavior and during which small reconfigurations take place. organization of tactical behavior. At the highest level, a sole
Notwithstanding, some PCs also exhibit some patterns that collective variable (PC) for each half exhibits the essence of
corresponded to game transitions (losing or winning back the tactics. The levels evolve on different timescales and only the
ball; Sarmento et al., 2013). For example, PC12 in the first half and tactical variables that evolved over longer timescales significantly
PC5 in the second defined counterattacks by the team analyzed. contributed to the structure of highest-order PCs. That is,
The small stretch index was quickly expanded and the team was the speed of team geometrical center evolves on a scale of
moving rapidly forward starting from the mid-defensive sector. seconds, the stretch index (player dispersion) remains below
However, it is possible to detect some tactical patterns defined by 16.75 m during more than 20 s on average (timescales of tens of
a pressing strategy to win back the ball by using player distances seconds), before switching expansion, and vice versa. Therefore,
from their team geometrical center and its depth in respect of the hierarchy is a consequence of the correlated lower-order
goal (Frias and Duarte, 2014). In this hypothetical case, the team PCs. The most time-persistent categories (those with long dwell
geometrical center would move forward while team dispersion times) create correlations between the lower-order PCs. This was
would reduce in size. This structure of PCs could be related with revealed in the highest order PCs where two out of three most
PC7 and PC8 in the first half and with PC6 and PC9 in the second. time-persistent categories were those that had by far the highest
Previous findings revealed that depending on the situational scores. In summary, whereas the highest level that captures the
context of the match the teams switch between different essentials of team tactical behavior corresponds to slower changes
functional states (Frencken et al., 2012). The results have (a few tens of seconds), the lowest order of PCs quickly evolve,
shown those of specific tactical patterns brought about by being more susceptible to the sensitive changes of constraint-
the influence of key events, like goal-scoring (Lago-Peñas and induced actions. It is important to note that this emergence
Dellal, 2010). Figure 2 allows identifying the types of patterns of tactical behaviors in soccer as in other team sports results
that appeared at different game moments. In this sense, the from reciprocal influences (bottom-up, top-down) of performer-
goal was preceded by a counterattack by the opposing team. environment interactions on different timescales (Hristovski
Consequently, coaches should consider training specifically those et al., 2011).
patterns which might lead to conceding/scoring a goal. Even so, Although previous studies have demonstrated that the average
such defensive patterns are obviously preceded by an attacking values of these positioning-derived variables were significantly
phase, then ball recovery strategies and the context in which this different between both halves and also dependent on ball
FIGURE 4 | Component scores of the highest level PC for the first and second half.
TABLE 2 | Mean values ± SD for different collective positional variables and ball possession and corresponding difference comparisons between first
and second half.
Mean ± SD (n) Differences observed for the first half compared with second
Stretch index 23.94 ± 22.70 (111) 26.44 ± 25.96 (102) 15.36; ± 25.97 34/65/1 Possibly ↑ 0.14; ± 0.23
Speed of dispersion 3.37 ± 3.32 (800) 3.24 ± 2.93 (833) −2.48; ± 6.08 0/100/0 Most likely trivial −0.03; ± 0.08
Length 7.27 ± 7.56 (370) 7.55 ± 8.46 (357) 2.12; ± 11.21 1/99/0 Very likely trivial 0.02; ± 0.12
Width 9.20 ± 9.10 (289) 9.27 ± 9.52 (291) 0.18; ± 12.57 1/98/1 Very likely trivial 0.00; ± 0.14
Sectors 22.05 ± 18.21 (119) 24.37 ± 25.31 (110) −3.27; ± 22.53 4/86/10 Likely trivial −0.03; ± 0.22
Corridors 37.10 ± 47.07 (73) 29.96 ± 32.69 (90) −0.68; ± 33.40 10/79/11 Unclear −0.01; ± 0.26
Longitudinal speed of team center 4.50 ± 5.03 (598) 3.98 ± 3.96 (678) −5.34; ± 7.71 0/99/1 Very likely trivial −0.06; ± 0.09
Lateral speed of team center 3.82 ± 3.84 (707) 3.89 ± 3.96 (693) 1.38; ± 7.67 0/100/0 Most likely trivial 0.02; ± 0.09
Ball possession 30.30 ± 32.30 (89) 26.19 ± 25.91 (103) −11.02; ± 22.82 2/72/27 Possibly ↓ −0.11; ± 0.24
TABLE 3 | Multiple Comparisons p-values (2-tailed) and Cohen’s d results for (a) the first half and (b) the second half.
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
Stretch index 1 — 1 —
Speed of dispersion 2 *vl — 2 *vl —
Length 3 *l *m — 3 *l *m —
Width 4 *l *m — 4 *l *m —
Sectors 5 *vl *l *m — 5 *vl *l *m —
Corridors 6 *vl *l *l — 6 *vl *l *l —
Longitudinal speed of team center 7 *l *s *s *m *vl *vl — 7 *vl *s *m *vl *vl —
Lateral speed of team center 8 *vl *m *m *vl *vl — 8 *vl *m *m *vl *vl —
–, Diagonal cell. *p < 0.05. Letters denote the magnitude: s, small; m, moderate; l, large; vl, very large.
possession (Clemente et al., 2013), the dwell times of most of rank of opponent, score of the previous match, playing home or
these variables were not significantly different comparing the first away, which remain invariant during a longer timescale, i.e., days.
and second halves. This invariance in the timescales has given rise Moreover, this analysis can help coaches to verify if the essence
to the hierarchical structure of tactical patterns in both halves of tactics or playing style performed during training sessions
being mostly the same. However, the additional level identified is definitively developed during competition and if it remains
in the second half suggests that tactical patterns carried out by invariant from match to match. Moreover, it would allow coaches
the second group of teammates performed more varied behaviors to optimize training drills developing their style of play and/or
because of the less correlated behaviors in the second-level of identify the tactical behaviors performed during the match to
principal components. In spite of this, the principal component design a representative training task. Finally, the characteristic
extracted in the highest-level of both halves reached a high value timescales of collective behaviors allow understanding the
of similarity. This congruence in score distribution suggests that formation of the hierarchically nested structure of tactical
independently of the group of teammates, the essence of team patterns in an ecological context and presents a rationale to define
tactics was similar during the match. soft-assemble multilevel dynamics in soccer matches.
These findings open the way to analyzing how teams behave
during a competition and studying the influence of different AUTHOR CONTRIBUTIONS
constraints (e.g., score, substitutions, instructional constraints,
opponent level, etc.) over the course of the match. Furthermore, AR worked on the design of the study, collection, analysis
due to the limitation of the unknown opponent and ball position, and interpretation of data, and drafting the manuscript. CT
more research is needed to ascertain the soft-assemble action participated in the conceptualization and design of the study,
landscape of different teams and to compare the collective data collection and reviewed the content of the manuscript. BG
tactical patterns of two confronted teams. The positioning of worked on the design of the study, data collection, statistical
the opponent would allow analyzing the coupling between the analysis, and reviewed the content of the manuscript. JS
teams’ behavior, and to detect if some of them lead pattern- participated in the conceptualization and design of the study
forming. The results should be balanced considering that only and reviewed the content of the manuscript. RH conceived the
one match was used to capture the tactical behaviors. Despite this approach to data analysis, data interpretation, and drafting of the
limitation, this study leaves an open path to exploring match-to- manuscript. The authors approved the final version and agree to
match invariant behaviors. In addition, this approach warrants be accountable for all aspects of the work.
further investigations on using ecological task constraints
during training to develop the potential landscape of tactical FUNDING
patterns allowing the spontaneous emergence of novel modes
of coordination and/or specific tactical performance solutions. This study is supported by the National Institute of Physical
This study also provides the possibility to determine relevant Education of Catalonia, Generalitat de Catalunya.
timescales for scoring goals and shooting at goal using a large
goal-scoring sample. ACKNOWLEDGMENTS
CONCLUSION We would like to thank the players who volunteered to
participate in this study. We gratefully acknowledge the support
This approach can help to identify tactical patterns during of the Generalitat de Catalunya government project Grup de
different matches, comparing the degree of similarity between recerca en Sistemes Complexos i Esport (2014 SGR 975). We
them. Further analysis would allow determining the influence would also thank the reviewers for their thorough review that
on team tactical behaviors of different situational variables, e.g., helped us a lot to improve the present manuscript.
REFERENCES Frencken, W., Lemmink, K., Delleman, N., and Visscher, C. (2011). Oscillations of
centroid position and surface area of soccer teams in small-sided games. Eur. J.
Araujo, D., Davids, K., and Hristovski, R. (2006). The ecological dynamics Sport Sci. 11, 215–223. doi: 10.1080/17461391.2010.499967
of decision making in sport. Psychol. Sport Exerc. 7, 653–676. doi: Frencken, W., Poel, H., De Visscher, C., and Lemmink, K. (2012). Variability of
10.1016/j.psychsport.2006.07.002 inter-team distances associated with match events in elite-standard soccer. J.
Badii, R., and Politi, A. (1999). Complexity: Hierarchical Structures and Scaling in Sports Sci. 30, 1207–1213. doi: 10.1080/02640414.2012.703783
Physics. Cambridge: Cambridge University Press. Frias, T., and Duarte, R. (2014). Man-to-man or zone defense? Measuring team
Bak, P., Tang, C., and Wiesenfeld, K. (1987). Self-organized criticality: dispersion behaviors in small-sided soccer games. Trends Sport Sci. 3, 135–144.
an explanation of the 1/f noise. Phys. Rev. Lett. 59, 381–384. doi: Fulgosi, A. (1988). Faktorska Analiza. Zagreb: Školska Knjiga.
10.1103/PhysRevLett.59.381 Gibson, J. J. (1979). The Ecological Approach to Visual Perception. Boston, MA:
Balagué, N., González, J., Javierre, C., Hristovski, R., Aragonés, D., Álamo, J., Houghton, Mifflin and Company.
et al. (2016). Cardiorespiratory coordination after training and detraining. Gonçalves, B. V., Figueira, B. E., Maçãs, V., and Sampaio, J. (2014). Effect
A principal component analysis approach. Front. Physiol. 7:35. doi: of player position on movement behaviour, physical and physiological
10.3389/fphys.2016.00035 performances during an 11-a-side football game. J. Sports Sci. 32, 191–199. doi:
Barreira, D., Garganta, J., Guimarães, P., Machado, J. C., and Anguera, M. T. 10.1080/02640414.2013.816761
(2014). Ball recovery patterns as a performance indicator in elite soccer. Gréhaigne, J. F., Bouthier, D., and David, B. (1997). Dynamic-system analysis of
Proc. Inst. Mech. Eng. Part P J. Sports Eng. Technol. 228, 61–72. doi: opponent relationships in collective actions in soccer. J. Sports Sci. 15, 137–149.
10.1177/1754337113493083 doi: 10.1080/026404197367416
Barros, R. M. L., Cunha, S. A., Magalhaes, W. J., and Guimarães, M. F. Haken, H. (2006). Information and Self-Organization: A Macroscopic Approach to
(2006). Representation and analysis of soccer players’ actions using principal Complex Systems, 3rd Edn. Heidelberg: Springer Verlag.
components. J. Hum. Mov. Stud. 51, 103–116. Available online at: http://hdl. Hayward, S., and De Groot, B. L. (2008). Normal modes and essential dynamics.
handle.net/11449/31748 Methods Mol. Biol. 443, 89–106. doi: 10.1007/978-1-59745-177-2-5
Bourbousson, J., Deschamps, T., and Travassos, B. (2014). From players to teams: Hopkins, W. G., Marshall, S. W., Batterham, A. M., and Hanin, J. (2009).
towards a multi-level approach of game constraints in team sports. Int. J. Sports Progressive statistics for studies in sports medicine and exercise science. Med.
Sci. Coach. 9, 1393–1406. doi: 10.1260/1747-9541.9.6.1393 Sci. Sports Exerc. 41, 3–12. doi: 10.1249/MSS.0b013e31818cb278
Bourbousson, J., Seve, C., and McGarry, T. (2010). Space–time coordination Hristovski, R. (2012). Sport performance as a domain of creative problem solving
dynamics in basketball: part 2. The interaction between the two teams. J. Sports for self-organizing performer-environment systems. Open Sports Sci. J. 5,
Sci. 28, 349–358. doi: 10.1080/02640410903503640 26–35. doi: 10.2174/1875399X01205010026
Bovier, A., and Den Hollander, F. (2016). Metastability: A Potential-Theoretic Hristovski, R., Davids, K., Araújo, D., and Passos, P. (2011). Constraints-induced
Approach. Berlin: Springer International Publishing. emergence of functional novelty in complex neurobiological systems: a basis for
Bronner, S., and Shippen, J. (2015). Biomechanical metrics of aesthetic perception creativity in sport. Nonlinear Dynamics. Psychol. Life Sci. 15, 175–206.
in dance. Exp. Brain Res. 233, 3565–3581. doi: 10.1007/s00221-015-4424-4 Hristovski, R., Davids, K., Araújo, D., Passos, P., Torrents, C., Aceski, A., et al.
Bruineberg, J., and Rietveld, E. (2014). Self-organization, free energy minimization, (2013). “Creativity in sport and dance: ecological dynamics on a hierarchically
and optimal grip on a field of affordances. Front. Hum. Neurosci. 8:599. doi: soft-assembled perception-action landscape,” in Complex Systems in Sport, eds
10.3389/fnhum.2014.00599 K. Davids, R. Hristovski, D. Araújo, N. Balagué, C. Button, and P. Passos
Castellano, J. (2008). Análisis de las posesiones de balón en fútbol : frecuencia, (London: Routledge), 261–274.
duración y transición. Mot. Eur. J. Hum. Movemen. 21, 179–196. Jirsa, V. K., Jantzen, K. J., Fuchs, A., and Kelso, J. A. S. (2002). Spatiotemporal
Clemente, F. M., Couceiro, M. S., Martins, F. M. L., Mendes, R., and Figueiredo, forward solution of the EEG and MEG using network modeling. IEEE Trans.
A. J. (2013). Measuring collective behaviour in football teams: inspecting the Med. Imaging. 21, 493–504. doi: 10.1109/TMI.2002.1009385
impact of each half of the match on ball possession. Int. J. Perform. Anal. Sport Joliffe, I. T. (2002). Principal Component Analysis, 2nd edn. New York, NY:
13, 678–689. Springer.
Costa, I. T., Garganta, J., Greco, P. J., Mesquita, I., and Maia, J. (2011). System Kello, C. T., Anderson, G. G., Holden, J. G., and Van Orden, G. C. (2008). The
of tactical assessment in Soccer (FUT-SAT): development and preliminary pervasiveness of 1/f scaling in speech reflects the metastable basis of cognition.
validation. System 7, 69–83. Cogn. Sci. 32, 1217–1231. doi: 10.1080/03640210801944898
Cugliandolo, L. F., and Kurchan, J. (1993). Analytical solution of the off- Kelso, J. A. S. (1995). Dynamic patterns: The Self-Organization of Brain and
equilibrium dynamics of a long-range spin-glass model. Phys. Rev. Lett. 71, Behavior. Complex adaptive systems series. Cambridge, MA: MIT Press,
173–176. doi: 10.1103/PhysRevLett.71.173 Bradford Book.
Daffertshofer, A., Lamoth, C. J. C., Meijer, O. G., and Beek, P. J. (2004). PCA in Lago, C. (2009). The influence of match location, quality of opposition, and match
studying coordination and variability: a tutorial. Clin. Biomech. 19, 415–428. status on possession strategies in professional association football. J. Sports Sci.
doi: 10.1016/j.clinbiomech.2004.01.005 27, 1463–1469. doi: 10.1080/02640410903131681
Duarte, R., Araújo, D., Folgado, H., Esteves, P., Marques, P., and Davids, K. (2013). Lago-Peñas, C., and Dellal, A. (2010). Ball possession strategies in elite soccer
Capturing complex, non-linear team behaviours during competitive football according to the evolution of the match-score: the influence of situational
performance. J. Syst. Sci. Complex. 26, 62–72. doi: 10.1007/s11424-013-2290-3 variables. J. Hum. Kinet. 25, 93–100. doi: 10.2478/v10078-010-0036-z
Eyring, H. (1935). The activated complex in chemical reactions. J. Chem. Phys. 3, Lorenzo-Seva, U., and ten Berge, J. M. F. (2006). Tucker’s congruence coefficient
107–115. doi: 10.1063/1.1749604 as a meaningful index of factor similarity. Methodology 2, 57–64. doi:
Fabrigar, L. R., Wegener, D. T., MacCallum, R. C., and Strahan, E. J. (1999). 10.1027/1614-2241.2.2.57
Evaluating the use of exploratory factor analysis in psychological research. Maisuradze, G. G., Liwo, A., and Scheraga, H. A. (2009). Principal component
Psychol. Methods 4, 272–299. doi: 10.1037/1082-989X.4.3.272 analysis for protein folding dynamics. J. Mol. Biol. 385, 312–329. doi:
Folgado, H., Duarte, R., Fernandes, O., and Sampaio, J. (2014). Competing with 10.1016/j.jmb.2008.10.018
lower level opponents decreases intra-team movement synchronization and Matsunaga, Y., Li, C.-B., and Komatsuzaki, T. (2007). Anomalous diffusion in
time-motion demands during pre-season soccer matches. PLoS ONE 9:e97145. folding dynamics of minimalist protein landscape. Phys. Rev. Lett. 99:238103.
doi: 10.1371/journal.pone.0097145 doi: 10.1103/PhysRevLett.99.238103
Forner-Cordero, A., Levin, O., Li, Y., and Swinnen, S. P. (2005). Principal McGarry, T., Anderson, D. I., Wallace, S. A., Hughes, M. D., and Franks, I. M.
component analysis of complex multijoint coordinative movements. Biol. (2002). Sport competition as a dynamical self-organizing system. J. Sports Sci.
Cybern. 93, 63–78. doi: 10.1007/s00422-005-0582-y 20, 771–781. doi: 10.1080/026404102320675620
Freidlin, M. I., and Wentzell, A. D. (2012). Random Perturbations of Dynamical Mendes, R. S., Malacarne, L. C., and Anteneodo, C. (2007). Statistics of football
Systems, 3rd Edn. Berlin; Heidelberg: Springer. dynamics. Eur. Phys. J. B 57, 357–363. doi: 10.1140/epjb/e2007-00177-4
Mohr, M., Krustrup, P., and Bangsbo, J. (2005). Fatigue in soccer: a brief review. J. people. J. Exp. Psychol. Hum. Percept. Perform. 16, 227–247. doi: 10.1037/0096-
Sports Sci. 23, 593–599. doi: 10.1080/02640410400021286 1523.16.2.227
Moura, F. A., Santana, J. E., Vieira, N. A., Santiago, P. R. P., and Cunha, S. Silva, P., Travassos, B., Vilar, L., Aguiar, P., Davids, K., Araújo, D., et al.
A. (2015). Analysis of soccer players’ positional variability during the 2012 (2014). Numerical relations and skill level constrain co-adaptive behaviors of
UEFA European Championship: a case study. J. Hum. Kinet. 47, 225–236. doi: agents in sports teams. PLoS ONE 9:e107112. doi: 10.1371/journal.pone.01
10.1515/hukin-2015-0078 07112
Moura, F. A., van Emmerik, R. E. A., Santana, J. E., Martins, L. E. B., de Barros, R. Tabachnick, B. G., and Fidell, L. S. (2007). Multivariate analysis of variance and
M. L., and Cunha, S. A. (2016). Coordination analysis of players’ distribution in covariance. Using Multivar. Stat. 3, 402–407.
football using cross-correlation and vector coding techniques. J. Sports Sci. doi: Torrents, C., Ric, A., and Hristovski, R. (2015). Creativity and emergence of specific
10.1080/02640414.2016.1173222. [Epub ahead of print]. dance movements using instructional constraints. Psychol. Aesthetics Creat.
Rabinovich, M. I., Huerta, R., Varona, P., and Afraimovich, V. S. (2008). Transient Arts 9, 65–74. doi: 10.1037/a0038706
cognitive dynamics, metastability, and decision making. PLoS Comput. Biol. Travassos, B., Araújo, D., Davids, K., Vilar, L., Esteves, P., and Vanda, C.
4:e1000072. doi: 10.1371/journal.pcbi.1000072 (2012). Informational constraints shape emergent functional behaviours during
Ramenzoni, V. C., Riley, M. A., Shockley, K., and Baker, A. A. (2012). Interpersonal performance of interceptive actions in team sports. Psychol. Sport Exerc. 13,
and intrapersonal coordinative modes for joint and single task performance. 216–223. doi: 10.1016/j.psychsport.2011.11.009
Hum. Mov. Sci. 31, 1253–1267. doi: 10.1016/j.humov.2011.12.004 Travassos, B., Araújo, D., Vilar, L., and McGarry, T. (2011). Interpersonal
Ric, A., Hristovski, R., Gonçalves, B., Torres, L., Sampaio, J., and Torrents, C. coordination and ball dynamics in futsal (indoor football). Hum. Mov. Sci. 30,
(2016). Timescales for exploratory tactical behaviour in football small-sided 1245–1259. doi: 10.1016/j.humov.2011.04.003
games. J. Sports Sci. 34, 1723–1730. doi: 10.1080/02640414.2015.1136068 Westerhuis, J. A., Kourti, T., and MacGregor, J. F. (1998). Analysis of
Richardson, M. J., Marsh, K. L., Isenhower, R. W., Goodman, J. R. L., and multiblock and hierarchical PCA and PLS models. J. Chemom. 12,
Schmidt, R. C. (2007). Rocking together: dynamics of intentional and 301–321.
unintentional interpersonal coordination. Hum. Mov. Sci. 26, 867–891. doi:
10.1016/j.humov.2007.07.002 Conflict of Interest Statement: The authors declare that the research was
Riley, M. A., Richardson, M. J., Shockley, K., and Ramenzoni, V. C. (2011). conducted in the absence of any commercial or financial relationships that could
Interpersonal synergies. Front. Psychol. 2:38. doi: 10.3389/fpsyg.2011.00038 be construed as a potential conflict of interest.
Sampaio, J. E., and Maçãs, V. M. (2012). Measuring tactical behaviour in football.
Int. J. Sports Med. 33, 395–401. doi: 10.1055/s-0031-1301320 Copyright © 2016 Ric, Torrents, Gonçalves, Sampaio and Hristovski. This is an
Sarmento, H., Barbosa, A., Anguera, M. T., Campaniço, J., and Leitão, J. (2013). open-access article distributed under the terms of the Creative Commons Attribution
“Regular patterns of play in the counterattack of the FC Barcelona and License (CC BY). The use, distribution or reproduction in other forums is permitted,
Manchester United football teams,” in Performance Analysis of Sport IX, eds provided the original author(s) or licensor are credited and that the original
D. Peters and P. O’Donoghue (London: Routledge), 59–106. publication in this journal is cited, in accordance with accepted academic practice.
Schmidt, R. C., Carello, C., and Turvey, M. T. (1990). Phase transitions and No use, distribution or reproduction is permitted which does not comply with these
critical fluctuations in the visual coordination of rhythmic movements between terms.
3
Ric, A., Hristovski, R., Gonçalves, B., Torres, L., Sampaio, J. y Torrents, C. (2016).
44
5. Estudio III - Emergence of Exploratory,
Technical and Tactical Behavior in
Small-Sided Soccer Games when
Manipulating the Number of Teammates
and Opponents 4
4
Torrents, C., Ric, A., Hristovski, R., Torres-Ronda, L., Vicente, E. y Sampaio, J. (2016).
when Manipulating the Number of Teammates and Opponents. PloS ONE, 11(12),
53
RESEARCH ARTICLE
1 National Institute of Physical Education of Catalonia (INEFC), University of Lleida, Lleida, Spain, 2 Sts.
Cyril and Methodius University, Skopje, Macedonia, 3 Department of Health and Kinesiology, Exercise and
a1111111111 Sport Nutrition Lab, Texas A&M University, College Station, Texas, United States of America, 4 CreativeLab,
a1111111111 Research Center in Sports Sciences, Health Sciences and Human Development (CIDESD), Vila Real,
a1111111111 Portugal, 5 Sport Sciences Department, Universidade de Trás-Os-Montes e Alto Douro, Vila Real, Portugal
a1111111111
* [email protected]
a1111111111
Abstract
OPEN ACCESS The effects that different constraints have on the exploratory behavior, measured by the
Citation: Torrents C, Ric A, Hristovski R, Torres-
variety and quantity of different responses within a game situation, is of the utmost impor-
Ronda L, Vicente E, Sampaio J (2016) Emergence tance for successful performance in team sports. The aim of this study was to determine
of Exploratory, Technical and Tactical Behavior in how the number of teammates and opponents affects the exploratory behavior of both pro-
Small-Sided Soccer Games when Manipulating the
fessional and amateur players in small-sided soccer games. Twenty-two professional (age
Number of Teammates and Opponents. PLoS ONE
11(12): e0168866. doi:10.1371/journal. 25.6 ± 4.9 years) and 22 amateur (age 23.1 ± 0.7 years) male soccer players played three
pone.0168866 small-sided game formats (4 vs. 3, 4 vs. 5, and 4 vs. 7). These trials were video-recorded
Editor: Sergio Pellis, University of Lethbridge, and a systematic observation instrument was used to notate the actions, which were subse-
CANADA quently analyzed by means of a principal component analysis and the dynamic overlap
Received: June 8, 2016 order parameter (measure to identify the rate and breadth of exploratory behavior on differ-
ent time scales). Results revealed that a higher the number of opponents required for more
Accepted: December 7, 2016
frequent ball controls. Moreover, with a higher number of teammates, there were more
Published: December 22, 2016
defensive actions focused on protecting the goal, with more players balancing. In relation to
Copyright: © 2016 Torrents et al. This is an open attack, an increase in the number of opponents produced a decrease in passing, driving and
access article distributed under the terms of the
controlling actions, while an increase in the number of teammates led to more time being
Creative Commons Attribution License, which
permits unrestricted use, distribution, and spent in attacking situations. A numerical advantage led to less exploratory behavior, an
reproduction in any medium, provided the original effect that was especially clear when playing within a team of seven players against four
author and source are credited. opponents. All teams showed strong effects of the number of teammates on the exploratory
Data Availability Statement: All data files are behavior when comparing 5 vs 7 or 3 vs 7 teammates. These results seem to be indepen-
available from the Open Science Framework dent of the players’ level.
database osf.io/8weh9.
Introduction
Tactical creativity refers to varying, atypical and flexible decisions and actions, and it plays a
key role in team ball sports due to the complex and chaotic determinants of performance [1,2].
Coaches are especially interested in developing a more flexible game to increase unpredictable
behavior through emergent properties [3] and gain advantage over the opponents [1]. Tactical
creativity can be described by the classical characteristics of creativity defined by Guilford [4]:
the unusualness of tactical actions defines originality; the variety of tactical actions defines flex-
ibility; and the number of tactical actions generated for certain situations defines fluency [5].
Therefore, it is important to design practice tasks that can foster players’ creative behavior and,
consequently, improve performance outcomes. Despite this, few studies have addressed the
topic of creativity in sport performance, and there remains a need to establish solid modeling
approaches for studying and explaining creative behavior [6,7].
The most widely used definitions of creativity refer to obtaining a useful and original prod-
uct or behavior [8–13], however, several authors with closer ties to education have particularly
focused the concept on individual and exploratory behavior. This conceptualization of creativ-
ity classifies as creative any exploratory behavior without regard to whether it leads to a novel
final outcome or not [14]. Exploratory behavior may be defined as a subsequent realization of
a large number of movement configurations under specific constraints of each performer [15].
Therefore, exploratory activity is strongly dependent on the set of constraints imposed on the
system. This observation implies that during training programs no explicitly prescriptive list of
possible actions needs to be provided beforehand to the performer so that s/he might realize
them [6]. This approach is relevant to team sports, which are based on improvisation and on
the interaction between the performers and the environment. The exploratory behavior that
emerges in team sports will depend on the affordances of the individuals involved, and on the
influence of the environment. Affordances have been defined as opportunities [16], invitations
[17] or solicitations [18] for action, but always as a scaled relation between certain properties
of the environment and a performer’s effectiveness, i.e. abilities [19,20]. It is the affordances
that channel the exploratory activity of the performer. The interrelatedness of affordances [18]
allows us to understand the emergence of behavior at different levels of game constraints.
Therefore, it may be necessary to provide task constraints allowing players to adapt constantly
their behavior to the unpredictable opponents’ behavior. Practice tasks should seek to promote
varied and flexible behavior so that players learn to be more adaptive, and the inclusion of
exploratory tasks can help them to be more creative.
A coordination dynamics approach, based on complex systems theories and ecological psy-
chology, offers a useful theoretical explanation of behavior based on performer-environment
relationships [21,22]. Under this approach, individuals and teams can be modeled as complex
social and nondeterministic systems [23]. From this perspective, functional patterns of coordi-
nated behavior emerge from ongoing interactions between system components through a pro-
cess of self-organization between players, the environment, and specific task constraints [24].
Hristovski, Davids, and Araújo [25] showed how the manipulation of task constraints
might enable the emergence of innovative and functional tactical behaviors. In boxing, it was
observed that changing the distance of the participants to a punching bag led to spontaneous
and novel behaviors, in this case, specific striking techniques that had not been practiced or
taught before. It was also shown how the spontaneous and innovative behaviors emerge only
for some specific task constraints. Extremes of very high and very low risk-of-being-hit con-
straints were suppressing the innovative behavior and only medium cost constraints enabled
emergence of novel actions. Torrents, Ric, and Hristovski [26] showed how the manipulation
of task constraints in dancers had a significant effect on the type of configurations performed,
as well as on their explorative behavior (i.e., the degree of exploration of dancers’ movement/
postures). Unusual rules resulted in a greater exploration of movement possibilities. Moreover,
results showed that novel dancers explored a significantly smaller region of the potential land-
scape than experts did, suggesting that this measure may help measuring the dancers’ perfor-
mances [27]. These observations of the emergence of human innovative behaviors are
important from the perspective of Boden’s [28] definition of transformational creativity, since
they explain how manipulating constraints, enables a transformation within a neurobiological
action system without reference to abstract rule-governed behaviors. Traditionally, coaches
design partial and separate practice exercises for technical, physiological, and tactical aspects,
and in some cases there is little correspondence with actual game situations [29]. However,
coaches need to develop practice scenarios that are representative of the actual competition in
order to achieve a better coupling between perception and action during the games [30].
In soccer, a common way to modify task constraints is implementing small-sided games, a
training task widely described in the literature [31,32]. These training scenarios exploit inher-
ent self-organizing tendencies and can simultaneously develop technical skills (related with the
movement of players’ body with the ball) and tactical behaviors (spatial-temporal movements
related with the shared task-goals in offense or defense) in competitive performance environ-
ments. The game is unpredictable and helps players maintain higher levels of motivation in
comparison with non-specific training situations. During small-sided games without external
instructions, the information flows, allowing continuous spontaneous interactions between
teammates and opponents [23]. The most common responses analyzed when playing small-
sided games have been those related to physiological or physical nature [31,33–35], in addition
to technical [36–38] and tactical aspects related to players’ positioning [36, 39]. In particular,
considering the number of players involved, Jones and Drust [40], and Katis and Kellis [41]
reported that the number of technical actions increased with a decrease in the number of play-
ers. However, Abrantes, Nunes, Maçãs, Leite, and Sampaio [42] observed no differences in
player effectiveness in relation to the density of players. Similarly, Duarte, Batalha, Folgado,
and Sampaio [43] founded that, in semi-professional futsal players, the lower the number of
players during the small-sided games, the more frequent the technical actions (e.g., successive
contacts with the ball, the number of dribbles, and the number of tackles). Casamichana and
Castellano [44] studied the frequency of appearance of technical actions in soccer and found
they became more common as the density of players, i.e. individual playing area was reduced.
Specifically, there were differences between small (32 x 23 m) and medium-sized (50 x 35 m)
pitches for the variables control and shoot, clearance, and putting the ball in play. Between
small and large (62 x 44 m) pitches there were also differences for interception, control and
dribble, clearance, and putting the ball in play. Owen et al. [45] found differences in the techni-
cal-related performance with the ball between small- (4 vs. 4), medium- (5 vs. 5 to 8 vs. 8), and
large-sided games (i.e., 9 vs. 9 to 11 vs. 11), where passing, receiving, dribbling, and shots, were
more frequent as the number of players decreased. Nevertheless, very little research has studied
technical responses in games involving unequal numbers of players, even though this kind of
numerical (dis-) advantage is common, and likely to be associated with more game perturba-
tions and, in consequence, more scoring opportunities [32]. Silva et al. [39] evaluated the effect
of different numerical relations (5vs5, 5vs4 and 5vs3) on positioning considering inter-individ-
ual, intra- and inter-team coordination. They found that these asymmetries constrained play-
ers’ individual dominant regions, modified the teams’ dispersion (decreased in inferiority), the
teams’ relative positioning on field (approached the opponent’s goal in superiority) and the
space between line-forces on wings and sectors.
Although small-sided games are well described in the literature, research related to their
effect on the creative behavior of the players is very scarce. In general, very limited research
has been conducted so far focused in the development of training methods that may foster cre-
ative behavior in sport. Some of the proposals to date include the use of diversification, deliber-
ate play, deliberate practice, or deliberate coaching (see [1] for a review). However, measuring
creativity in sport is challenging, and previous research has shown how difficult it can be to
reach an agreement about what constitutes an original response [46]. Due to the subjective
nature of this aspect of tactical creativity, the present study takes into account criteria related
to the variety and quantity of responses (i.e., flexibility and fluency) [4], but does not consider
the rarity or originality of responses. For this reason, we will address exploratory behavior
instead of tactical creativity to describe the players’ creative behavior. Moreover, we propose
unusual training situations with the aim of stimulating greater exploration of the tactical possi-
bilities, considering the suggestion by Stokes [47,48] where in order to produce a creative solu-
tion, constraints not only may preclude existing responses, but also promote novel, and
frequently, opposites. We hypothesize that playing in difficult scenarios such as playing in
numerical inferiority will facilitate the emergence of more varied behavior. We further hypoth-
esize that the modification of the number of opponents and teammates will produce the use of
different number and type of technical and tactical actions. Therefore, the aim of this study
was to examine how the constraints such as number of opponents and teammates affect the
technical, tactical, and exploratory behavior in small-sided games, in both professional and
amateur players.
Procedure
The study was conducted over two separated sessions, one for PRO and one for AMA, both
following the same procedures. The players were distributed into two teams (a, b) with a fixed
number of players (4), and two teams (c, d) with a variable number of players (3, 5, or 7). The
head coach of the professional team and the teacher of the students was the same person and
used subjective evaluation to distribute the players in the different teams. Their physical, tech-
nical, and tactical performances were the criteria used to constitute balanced teams [31]. Both
sessions involved three different small-sided games: GK + 4 vs. 7 + GK; GK + 4 vs. 5 + GK;
and GK+ 4 vs. 3 + GK, lasting 3 min each, with 4 min of passive recovery (Fig 1). These three
small-sided games formats were repeated twice, yielding a total of six situations for each team.
All these games were played with a numerical mismatch between the teams involved, and in a
randomized order. Each team always competed against the same team (i.e., AMAa vs. AMAc;
AMAb vs. AMAd; PROa vs. PROc; PROb vs. PROd). Neither professional nor amateur players
were familiar with the 4 vs. 7 situations. All the small-sided games were played on a 40 × 30 m
artificial turf pitch, and in accordance with the official soccer rules. To encourage high work-
rate maintenance, coaches or other players were allowed to give verbal encouragement to all
players, but not specific feedback related with players’ performance or their technical/tactical
behavior. In addition, several balls were placed around the pitch perimeter with the aim of
increasing the effective playing time. In order to avoid the effect of the scoring, the scoreboard
turned to 0 when any team got two goals. Each session ended with a 10 min cool-down, con-
sisted of static stretching exercises.
All small-sided games were video-recorded using three cameras, and analyzed by two
independent experienced soccer coaches (Graduated in Sports Sciences, certified with UEFA
PRO license and with experience as coaches). For all small-sided games, observers took into
account the actions and characteristics of the actions performed by both teams. These actions
or characteristics were defined on a coarse-grained scale of 51 categories (see Table 1) adapt-
ing the proposals of Owen, Twist and Ford [49] and Costa, Garganta, Greco, Mesquita and
Maia [50]. These actions were evaluated each second, following the suggestions of Mendes
and Malacarne [51]. The two repetitions of the same small-sided games were sequenced and
summed, resulting in three 6 min situations for each AMA team and three 6 min situations
for each PRO team. The time during which the players analyzed were not playing (i.e., the
ball left the pitch or the goalkeeper was controlling the ball) was removed, and all trials were
equalized, resulting in an overall analysis of 204 seconds for each type of small-sided games.
The interval and ratio scales, i.e. positional data, are maximally objective and due to this
property they intrinsically have larger potential for generalization. However, space-time data
are not able to provide direct information on which tasks individual players were exploring
and in what context, within the game. Because we were interested in the technical-tactical
and exploratory behavior in the discrete task-space [52] the nominal scale was the scale of
choice. Each 1 s window was defined as a 51-component binary vector representing the full
action configuration during the same time interval. A value of 1 was ascribed to the active
category and a value of 0 to the inactive one. Both observers analyzed one game (6 minutes)
of each team to calculate reliability. The inter-observer agreement yielded a minimum Kappa
coefficient of 0.82, providing a satisfactory guarantee of data reliability. One of the observers
analyzed the other trials of all teams.
Table 1. Observation instrument defined on a coarse-grained scale and based on the technical/tacti-
cal actions of players (Adapted from Owen et al. [45] and Costa et al. [53])*.
Role Technical/tactical Definition
action
Attacker with Run to the ball Player who runs towards the ball with the intention of reaching it
ball Wait Player remains still until the ball is close to where he is, at a point
where another action can be performed
Control Player receives an intended pass or makes contact in attempting
to maintain possession of the ball
Pass Player in possession sends the ball to a teammate
Shoot Player in possession intentionally sends the ball towards the goal
in an attempt to score
Protect Putting one’s body between the ball and one or more opponents
with the aim of keeping possession
Drive Movement of ball carrier towards the goal or changing direction in
order to play in other areas of the pitch line
Feint Gesture or movement to deceive the opponent
Dribble Player in possession, with ball at feet, runs with ball, beats or
attempts to beat an opponent
Intercept Player makes contact with or stops the ball, enabling him to regain
possession and preventing an opponent’s pass from reaching its
intended destination
Deflect Unintentionally changing the trajectory of the ball after it was
kicked by the opponent
Clear Intentionally moving the ball away from a zone or situation close to
one’s own goal
Anticipate Action intending to dispossess an opponent who is in possession
of the ball.
Attackers Wait Remaining in a fixed position or walking, waiting the action of the
without ball ball carrier, teammates and opponent
Support Player’ movements towards the ball carrier offering a passing
option aimed on keeping the ball possession
Unmark Movement of players between the last defender and towards the
goal line amplifying the effective playing space and offering a long
pass option
Defenders Press Actions to regain the ball or attempt to make the opponent lose the
ball
Delay Actions to slow down the opponent’s attempt to move forward with
the ball
Dissuade Take up a position with respect to an opponent without the ball to
make it difficult for him to receive or to prevent him from receiving
the ball
Balance Positioning of off-ball defenders in response to movements of
attackers in an attempt to achieve numerical stability or superiority
with respect to the opposition
Withdraw Move back to a position between the line of the ball and one’s own
goal
*The number of players who performed each action (1, 2, 3 or more) was defined each second.
doi:10.1371/journal.pone.0168866.t001
Data analysis
In order to analyze the influence of the task constraints on the use of the different type and
number of actions, a hierarchical principal components analysis [54] under Direct Oblimin
rotation, with Delta = 0, aiming data reduction was performed (for the suitability of using
principal components analysis with binary variables, see [55]). The number of significant first
level principal components was determined by identifying those that accounted for 80% of
the explained variance. The component correlation matrix of the first-order principal compo-
nents was then subjected to a further higher-order analysis. The component score matrix was
used in order to detect the most salient collective and individual action categories that emerged
during play.
To identify a team’s collective exploratory behavior during games, the average dynamic
overlap, <qd(t)>, was calculated as an average cosine auto-similarity of the overlaps between
configurations with increasing time lag (for more details, see Hristovski et al. [7]). This mea-
sure captures the average similarity of configurations or game patterns at ever increasing time
distances (i.e., time lags) from each other. Hence, it is capable of detecting the rate and breadth
of exploratory behavior over different time scales. For example, if a team uses different types of
attacking actions (i.e. running to the ball, then control it and then pass it, but in the following
attacks the players use other combinations of actions) and defending actions (sometimes press,
sometimes delay and others balance, for instance) will have a larger exploratory breadth than a
team of players who perform the same actions over time. Had the players remained in the
same position or had they repeated the same action throughout the observation time, then the
average dynamic overlap would be a constant equal to 1 (i.e., <qstat> = 1) for all time lags. On
the other hand, if, during the game, the players explored distant configurations of all the possi-
ble combinations of actions, then the average overlap (<qstat)>) would be zero. This will be
detected in a short time scale of some seconds, but also for larger time scales of minutes.
The average dynamic overlap was fitted by the following equation, which is derived for sys-
tems with an intricate hierarchical structure [56]:
< qd ðtÞ >¼ ð1 qstat Þt a þ qstat ð1Þ
where qstat is the asymptotic (i.e., stationary) value of the dynamic overlap, t is the time lag,
and α is the dynamic exponent. qstat detects the long-term exploratory breadth of the team,
and α the rate of exploration. The fit between the theoretical prediction and the experimental
data was assessed by the quasi-Newton method of least squares estimation, with the goodness
of fit being based on the percentage of explained empirical variance.
In the current study, the sample average <qstat> values under different task constraints
using a one-way ANOVA were compared across the different scenarios. A statistical signifi-
cance level of 95% was selected and effect sizes (Cohen’s d) were computed to identify the mag-
nitude of the effects.
Results
Principal component analysis
Under the Kaiser-Guttman criterion the hierarchical principal components analysis initially
revealed between 9 and 14 primary principal components (PCs) in each file. The second-level
PC structure was formed by salient correlated clusters of first-order PCs. A significant dimen-
sional reduction was obtained, resulting in between three and five PCs to analyze. The second-
ary-level correlated structures systematically produced a sole third-level PCs under the
majority of the trials, although a fourth level was present in some cases. By way of an example,
the results from this highest level PC for AMAb and PROa (fixed teams) are presented in Fig
2. The highest level PC of the remaining teams was very similar.
These results showed that the essential technical/tactical actions of soccer were the most fre-
quent in all situations and in both AMA and PRO players. When attacking, the most common
actions used by the player in possession of the ball were controlling, passing, and driving the
ball, whereas waiting or supporting the player with the ball were the actions most used by
Fig 2. Highest level PCs for AMAb and PROa (fixed teams).
doi:10.1371/journal.pone.0168866.g002
attacking players without the ball. On the other side, when defending, results showed less vari-
ability in the different tactical actions used, being the most common combination: one player
pressing, one delaying, one dissuading, and one balancing and/or one withdrawing. Results
for the variable team were very similar, with the exception that the number of players without
the ball, who performed more technical/tactical actions as the number of players in the team
increased.
Due to these similar results for all situations in the highest level PC, which seems to capture
the essential characteristics of soccer, it was important to analyze the first set of primary PCs of
each trial. For the fixed team, an increase in the number of opponents led to an increase in the
time spent defending, this being the case for both PRO and AMA players. A further finding for
both levels when defending was a decrease in the number of players dissuading and an increase
in the number of players balancing. In 4 vs. 7 games, the most frequent scenario was all the
players balancing. In relation to attack, an increase in the number of opponents produced a
decrease in passing and driving actions among PRO players, whereas AMA players showed a
decrease in passing and controlling actions. For both levels there were fewer players waiting
during the game in inferiority. For the variable team, an increase in the number of teammates
led to more time being spent in attacking situations. No clear differences appeared with respect
to the player with the ball, but an increase in the number of teammates did produce an increase
in the number of players waiting and supporting. When defending, there was a clear increase
in the number of players involved in balancing. In all the situations, the most frequent combi-
nation was one player pressing, one delaying, one dissuading, and/or one or more balancing.
Exploratory dynamics
The goodness of fit between the theoretical curve and the experimental data (n = 102) for the
exploratory dynamics (see Eq 1) reached values of R2 = .91 ± 0.04. This result confirms that Eq
(1) is able to capture the essential hierarchical dynamics during small-sided games. Results
from the exploratory dynamics analysis are summarized in Table 2 and Fig 3, where it is pre-
sented the averages of the stationary overlap order parameter, <qstat>, for each trial, compar-
ing the values of each team in all conditions. Results show different values of exploration but
with clear differences between the different small-sided games. For the fixed teams, AMAa
showed a small effect of the number of opponents, just comparing 5 OPP vs 7 OPP. AMAb
Table 2. Mean values (±SD) for the stationary part of the overlap order parameter (<qstat)>) obtained for the three conditions in all teams.
Fixed teams 3 OPP 5 OPP 7 OPP p (Cohen’s d) Variable teams 3 TM 5 TM 7 TM p (Cohen’s d)
AMAa 0.19±0.04 0.19±0.03 0.19±0.02 a) - AMAc 0.2±0.04 0.19±0.02 0.28±0.02 d) p < .01 (0.21)
b) p < .05 (0.36) e) p < .01 (4.27)
c) - f) p < .01 (2.65)
AMAb 0.22±0.03 0.17±0.03 0.16±0.03 a) p < .01 (1.73) AMAd 0.18±0.03 0.19±0.04 0.27±0.03 d) -
b) - e) p < .01 (2.00)
c) p < .01 (2.01) f) p < .01 (2.74)
PROa 0.20±0.04 0.17±0.02 0.17±0.03 a) p < .01 (0.87) PROc 0.2±0.05 0.2±0.03 0.29±0.03 d) -
b) - e) p < .01 (3.27)
c) p < .01 (0.76) f) p < .01 (2.18)
PROb 0.20±0.03 0.19±0.03 0.18±0.03 a) p < .05 (0.43) PROd 0.15±0.04 0.19±0.04 0.28±0.03 d) p < .01 (1.01)
b) - e) p < .01 (2.8)
c) p < .05 (0.66) f) p < .01 (4.02)
Abbreviations: PRO = professional players; AMA = amateurs players; OPP = opponents; TM = teammates. Significant differences are identified as: a) 3
OPP vs. 5 OPP; b) 5 OPP vs. 7 OPP; c) 3 OPP vs. 7 OPP; d) 3 TM vs. 5 TM; e) 5 TM vs. 7 TM; f) 3 TM vs. 7 TM.
doi:10.1371/journal.pone.0168866.t002
Fig 3. Averages of the stationary overlap order parameter, <qstat>, for each trial.
doi:10.1371/journal.pone.0168866.g003
showed strong effects of the number of opponents when comparing 3 vs 5 OPP and 3OPP vs 7
OPP. PROa and PROb showed moderate effects when comparing 3 OPP vs 5 OPP and 3 OPP
vs 5 OPP. In the case of the variable teams, playing with seven teammates clearly produced a
lower exploratory breadth compared with the other conditions. All teams showed strong
effects of the number of teammates when comparing 5 TM vs 7 TM and 3 TM vs 7 TM.
Discussion
This study explored the use of constraints as a way of enhancing players’ exploratory behavior
in soccer, and also analyzed the emergence of individual and collective action configurations
when varying the number of opponents or teammates in the context of numerically unbal-
anced teams. The analysis was only focused on observable motor behavior, therefore, disre-
garding important behavioral factors such as participants’ emotions, thoughts, the influence of
the partners’ behavior, or the subjective meaning of their behavior. In addition, the effective-
ness of play was not analyzed, mainly due to the controversies found in using any objective
measure of overall performance.
The results showed a topological and temporal structure of relatedness between the config-
urations emerging when playing soccer [7]. In general, certain technical/tactical actions
appeared with a high frequency, while others were statistically rare and short-lived reconfigu-
rations that constitute fluctuations following dynamical systems principles [21]. Highly fre-
quent combination of actions loaded predominantly on the first PC, capturing most of the
variance. The variation in the number of opponents and teammates had a significant effect on
the type of action configurations performed by the players in lower order PCs, although this
effect was not appreciated in the highest level PC, which captures the essential characteristics
of the game but not the peculiarities of each situation. This result gives a clear picture of the
emergent hierarchical organization of soccer, in which the general task constraints of the game
give rise to a stable and general structure that is captured by the high similarity of the highest
order PCs [57]. However, more immediate and peculiar task constraints, specific to the relative
number of players on the pitch and, perhaps, to personal constraints of the players involved in
the game, induce a more differentiated structure of decisions-actions at the individual and col-
lective level, and this is captured in lower order PCs.
An increase in the number of opponents produced an increase of the frequency of defensive
patterns and, especially, an increase in the number of players controlling and a decrease in the
number waiting. An increase in the number of teammates produced an increase in the fre-
quency of attacking patterns. By contrast, an increase in the number of opponents produced a
decrease in the use of basic actions such as passing or driving by players in possession of the
ball, while an increase in the number of teammates produced more defensive actions focused
on protecting the goal, with more players balancing, although it did not produce an increase in
the number of players pressing or dissuading. This finding should be taken into account by
coaches when designing their training sessions, since it suggests that training in situations of
numerical inferiority could improve a team’s defensive skills, not only because of the increased
time spent in that situation but also because the players become more active when defending
in an attempt to prevent shooting opportunities by the opponents. In light of these results, it is
suggested that coaches alternate situations of numerical inferiority and superiority in training
sessions by using attacking jokers who play for the team in possession of the ball [58]. This
strategy could help coaches to take advantage of the observed effects of inferiority in defense
and superiority in attack. In general, compared with full-sided games, small-sided games offer
more opportunities for players to practice attacking actions such as passing, shooting, drib-
bling, heading, or supporting teammates with the ball, as well as other defense-related skills
such as intercepting the ball [44]. Moreover, small-sided games provide situations closer to
instability [23,41,45], since in full-sided team games the global system is harder to destabilize
because some subgroups may be inactive [23].
In terms of the emergence of flexible and fluent behavior, players seem to show more
exploratory behavior when playing with a numerical disadvantage. This situation seems to
force players to vary their game more, whereas a numerical advantage appears to produce
game that is less exploratory and less varied. This was especially evident when playing with
seven teammates, since the exploratory breadth was clearly lower in all cases. This result is in
agreement with the proposal of Stokes [48,59], who considers constraints not only as barriers
or structures but also as a possibility to promote a search in different parts of the problem
space. This seems to be a general principle, since it has also been found in individual sports
training sessions. For example, in order to enhance exploratory behavior one can induce con-
straints that suppress more habitual actions lying within the comfort zone [26]. The newly
induced constraints effectively release the level of constraints that play a crucial role in generat-
ing the habitual behavior. As a consequence, new functional behaviors may emerge [6]. Never-
theless, extreme difficult scenarios probably will produce the contrary effect, and players
would not be able to cope with the situation. These results may also be related to that obtained
when studying the positioning dynamics of small-sided games [60], as an increase in the num-
ber of teammates seemed to promote regularity in spatial organization in AMA teams,
although in PRO teams the variation in players’ irregularity ranged from trivial to small. PRO
players probably anticipate better the need to optimize collective decision-making. However,
by studying individual variables, it has been showed that playing in inferiority produces less
varied game patterns, especially in defense [57]. It may be the case that the smaller degree of
diversity at the individual tactical behavior level is compensated with an increased diversity at
the team level, and vice versa [61]. These possible synergic mutually compensating processes
may be a key to understanding the multi-level interdependencies in team sports. That is, each
individual performer may reciprocally influence and be influenced by team’s behavior in a
multilevel synergic relation [62].
In general, these results suggest that easier game situations can promote more regular and
less varied play, while more difficult scenarios force the players to explore the varieties of tacti-
cal/technical actions that they can perform. This conclusion is consistent with the results
obtained in other contexts, for example, the recent analysis of dance improvisation by Torrents
et al. [26]. In that study unusual and more difficult instructional constraints produced a more
varied dance and led to greater exploration of movement possibilities. Previous research on
tactical creativity has shown that it can be improved by practice, with deliberate play or delib-
erate practice [63], but to our knowledge, there is no literature related to the type of specific
task constraints that can enhance creative behavior. The present study suggests that discomfort
may be a good way of enhancing creativity, as has been suggested in other fields such as man-
agement, art or creative services [64], but more research is needed to verify that this theory is
applicable to sport specific settings. More research is also needed in order to evaluate if there is
a curvilinear effect of discomfort on the exploratory behavior in team sports as Hristovski,
Davids and Araújo [25] observed in boxing by modifying the risk-of-being-hit constraints.
The effect of the three small-sided games formats seems to be similar for both AMA and
PRO players. The effect of varying the number of opponents and teammates on the configura-
tions of actions or exploratory behavior seems to be independent of the players’ level. On the
contrary, Silva et al. [39] have found that skill level is determinant in the perception of different
possibilities for action when evaluating positioning of the team. This model could be useful for
studying the emergence of creative behavior in any sport activity, especially team sports or
improvisational situations, and also for exploring the influence of task constraints on the
teaching-learning process.
Acknowledgments
We would like to thank the players who volunteered to participate in this study. We gratefully
acknowledge the support of the Generalitat de Catalunya government project Grup de recerca
en Sistemes Complexos i Esport (2014 SGR 975) and Project Symbiotic technology for societal
efficiency gains: Deus ex Machina (DEM), NORTE-01-0145-FEDER-000026, co-financed by
Fundo Europeu de Desenvolvimento Regional (FEDER) by NORTE 2020.
Author Contributions
Conceptualization: CT AR RH EV LT JS.
Data curation: AR RH.
Formal analysis: AR RH.
Funding acquisition: CT JS.
Investigation: CT EV LT JS.
Methodology: CT AR RH EV LT JS.
Project administration: CT.
Resources: CT EV LT JS.
Software: RH.
Supervision: CT RH LT JS.
Validation: CT AR RH EV LT JS.
Visualization: CT AR.
Writing – original draft: CT AR RH EV LT JS.
Writing – review & editing: CT RH LT JS.
References
1. Memmert D. Teaching tactical creativity in sport: Research and practice. Teaching Tactical Creativity in
Sport: Research and Practice. Taylor and Francis Inc.; 2015. http://www.scopus.com/inward/record.
url?eid=2-s2.0-84942093386&partnerID=tZOtx3y1
2. Memmert D, Roth K. The effects of non-specific and specific concepts on tactical creativity in team ball
sports. J Sports Sci. 2007; 25: 1423–1432. doi: 10.1080/02640410601129755 PMID: 17786695
3. Yamamoto Y, Yokoyama K. Common and Unique Network Dynamics in Football Games. PLoS One.
2011; 6: e29638. doi: 10.1371/journal.pone.0029638 PMID: 22216336
4. Guilford JP. The Nature of Human Intelligence. McGraw-Hill, Inc.; 1967.
5. Memmert D. Tactical creativity in team sports. Res Phys Educ Sport Heal. 2014; 3: 13–18.
6. Hristovski R, Davids K, Araújo D, Passos P. Constraints-induced emergence of functional novelty in
complex neurobiological systems: a basis for creativity in sport. Nonlinear Dynamics Psychol Life Sci.
2011; 15: 175–206. PMID: 21382260
7. Hristovski R, Davids K, Araújo D, Passos P, Torrents C, Aceski A, et al. Creativity in sport and dance:
Ecological dynamics on a hierarchically soft-assembled perception-action landscape. In: Davids K,
Hristovski R, Araújo D, Balagué N, Button C, Passos P, editors. Complex systems in sport. London:
Routledge; 2013. pp. 261–274.
8. Amabile T. Growing up creative. Nurturing a lifetime of creativity. New York: CEF Press; 1989.
9. Kasof J. Explaining Creativity: The Attributional Perspective. Creat Res J. 1995; 8: 397–403.
10. Mumford MD, Simonton K. Creativity in the Workplace: People, Problems, and Structures. J Creat
Behav. 1997; 31: 1–6.
11. Runco MA. Creativity. Annu Rev Psychol. 2004; 55: 657–687. doi: 10.1146/annurev.psych.55.090902.
141502 PMID: 14744230
12. Runco MA, Charles RE. Judgments of originality and appropriateness as predictors of creativity. Pers
Individ Dif. 1993; 15: 537–546.
13. Sternberg RJ, Lubart TI. Investing in Creativity. Psychol Inq. 1993; 4: 229–232.
14. Drazin R, Glynn MA, Kazanjian RK. Multilevel theorizing about creativity in organizations: A sensemak-
ing perspective. Acad Manag Rev. 1999; 24: 286–307.
15. Hristovski R, Davids K, Passos P, Araújo D. Sport Performance as a Domain of Creative Problem Solv-
ing for Self-Organizing Performer-Environment Systems. Open Sports Sci J. 2012; 5: 26–35.
16. Gibson JJ. The ecological approach to visual perception. Boston: Houghton Mifflin; 1979.
17. Withagen R, de Poel H, Araújo D, Pepping GJ. Affordances can invite behavior: Reconsidering the rela-
tionship between affordances and agency. New Ideas Psychol. 2012; 30: 250–258.
18. Bruineberg J, Rietveld E. Self-organization, free energy minimization, and optimal grip on a field of affor-
dances. Front Hum Neurosci. 2014; 8: 1–14.
19. Turvey M. Affordances and prospective control: An outline of the ontology. Ecol Psychol. 1992; 4: 173–
187.
20. Chemero A. An outline of a theory of affordances. Ecol Psychol. 2003; 15: 181–195.
21. Kelso JAS. Dynamic patterns: the self-organization of brain and behavior. Complex Adaptive Systems
series. MIT Press, Bradford Book, Cambridge, MA; 1995.
22. Kelso JAS. Coordination Dynamics. In: Meyers RA, editor. Encyclopedia of complexity and system sci-
ence. Springer; 2009. pp. 1537–1564.
23. Davids K, Araújo D, Correia V, Vilar L. How small-sided and conditioned games enhance acquisition of
movement and decision-making skills. Exerc Sport Sci Rev. LWW; 2013; 41: 154–161.
24. Newell KM. Constraints on the development of coordination. In: Wade MG, Whiting TA, editors. Motor
development in children Aspects of coordination and control. Dordrecht: Martinus Nijhoff; 1986. pp.
341–360.
25. Hristovski R, Davids K, Araújo D. Information for Regulating Action in Sport: Metastability and Emer-
gence of Tactical Solutions. In: Araujo D, Ripoll H, Raab M, editors. Perspectives on Cognition and
Action in Sport. New York; 2009. pp. 43–57.
26. Torrents C, Ric A, Hristovski R. Creativity and emergence of specific dance movements using instruc-
tional constraints. Psychol Aesthetics, Creat Arts. 2015; 9: 65–74.
27. Torrents C, Hristovski R, Coterón J, Ric A. Interpersonal coordination in dance: Inspiration or constraint.
In: Passos P, Chow JY, Davids K, editors. Interpersonal Coordination and Performance in Social Sys-
tems. Oxon: Routledge; 2016. pp. 85–93.
28. Boden MA. What is Creativity. In: Boden MA, editor. Dimensions of Creativity. MIT Press; 1996. Pp.
75–118.
29. Mitchell S, Oslin J. Ecological task analysis in games teaching: tatical games model. In: Davis WE,
Broadhead GD, editors. Ecological task analysis and movement. Human Kinetics; 2007. pp. 161–178.
30. Araujo D, Davids K, Passos P. Ecological validity, representative design, and correspondence between
experimental task constraints and behavioral setting: comment on Rogers, Kadar, and Costall (2005).
Ecol Psychol. 2007; 19: 69–78. Available: http://www.tandf.co.uk/journals/titles/10407413.asp
31. Aguiar MVD, Botelho GMA, Gonçalves BS V, Sampaio JE. Physiological responses and activity profiles
of football small-sided games. J Strength Cond Res. LWW; 2013; 27: 1287–1294.
32. Hill-Haas S V., Dawson B, Impellizzeri FM, Coutts AJ. Physiology of small-sided games training in foot-
ball: A systematic review. Sport Med. 2011; 41: 199–220.
33. Los Arcos A, Vázquez JS, Martı́n J, Lerga J, Sánchez F, Villagra F, et al. Effects of small-sided games
vs. interval training in aerobic fitness and physical enjoyment in young elite soccer players. PLoS One.
Public Library of Science; 2015; 10.
34. Castellano J, Puente A, Echeazarra I, Usabiaga O, Casamichana D. Number of Players and Relative
Pitch Area per Player: Comparing Their Influence on Heart Rate and Physical Demands in Under-12
and Under-13 Football Players. PLoS One. 2016; 11.
35. Halouani J, Chtourou H, Dellal A, Chaouachi A, Chamari K. Physiological responses according to rules
changes during 3 vs. 3 small-sided games in youth soccer players: stop-ball vs. small-goals rules. J
58. Ric A, Hristovski R, Torrents C. Can joker players favor the exploratory behaviour in football small-sided
games? Res Phys Educ Sport Heal. 2015; 4: 35–39.
59. Stokes PD. Using Constraints to Create Novelty: A Case Study. Psychol Aesthetics, Creat Arts. 2009;
3: 174–180. Available: http://205.186.157.44/imats/Serra2009.pdf
60. Gonçalves B, Marcelino R, Torres-Ronda L, Torrents C, Sampaio J. Effects of emphasising opposition
and cooperation on collective movement behaviour during football small-sided games. J Sport Sci.
2016; 34: 1346–1354.
61. Riley MA, Richardson MJ, Shockley K, Ramenzoni VC. Interpersonal synergies. Front Psychol. 2011;
2: 1–7.
62. Schöner G. Recent Developments and Problems in Human Movement Science and Their Conceptual
Implications. Ecol Psychol. 1995; 7: 291–314.
63. Memmert D. Sports and Creativity. In: Runco E-CMA, Pritzker SR, editors. Encyclopedia of Creativity (
Second Edition). Academic Press; 2011. pp. 373–378. http://dx.doi.org/10.1016/B978-0-12-375038-9.
00207-7
64. Hearn G, Rodrigues JHP, Bridgstock R. Learning processes in Creative Service teams: towards a
dynamic systems theory. In: Hearn G, Bridgstock R, Goldsmith B, Rodgers J, editors. Creative Work
Beyond the Creative Industries: Innovation, Employment and Education. Edward Elgar Publishing
Company; 2014. pp. 175–192.
5
Ric, A., Torrents, C., Gonçalves, B., Torres-Ronda, L., Sampaio, J. y Hristovski, R.
69
RESEARCH ARTICLE
relational) variables would allow coaches to identify the effects of training drills on teams
and players’ behaviour. More research is needed considering these type variables to under-
stand how the manipulation of constraints induce a more stable or flexible dynamical struc-
ture of tactical behaviour.
Introduction
The analysis of performance in sport through the collection and subsequent processing of data
has been widely used to provide useful information for coaches [1]. This performance analysis
has sought to obtain indicators of execution, such as offensive technical actions or successful
defensive events, and has ranked them using statistical procedures to characterize football per-
formance [2–5]. Although these previous studies have led to advances in football performance,
the notation of discrete actions and/or events has not provided information about the certain
performance contexts [6]. For instance, ball location relative to the scoring targets constrains
the emergence of spatiotemporal coordinated team behaviours [7]. Various studies have
employed other measures such as players’ trajectories, interpersonal distances, relative angles
between players or velocities as state variables [8–11]. They have served to define the coordina-
tion system states (patterns) at different levels of analysis (i.e., player, team or game). Hence,
the challenge in performance analysis is to capture key contextual information that helps to
describe and model the varied game scenarios explored by the team and players during both
training and competition [12,13].
A combination of ball events and positional data is needed to understand the players’ and
team’s performance. Thus, several indicators such as player-player and player-ball dyadic
coordination, intra- and inter-team synchronization, pattern-forming dynamics, time
required to regain ball possession, ball possession percentage, number of passes and their
length have been used to characterize individual and collective performance [14–17]. The
notation of passes has also helped to quantify the main interactions between the player in pos-
session of the ball and her or his teammates. These data can be used to measure closeness and
betweenness centrality scores [18–20], flow motifs [21] or a combination of zone and player
passing measures [22] to identify team and player performance and styles of play. Accordingly,
Yamamoto and Yokoyama [23] found the existence of a power law structure in the network of
passing behaviour. This degree distribution suggests that self-organisation phenomena occurs
during ball flow football dynamics and characterize a scale-free network where few players acts
as a game hub. Due to the small number of nodes (11 players), Narizuka and colleagues [24]
included the player location when performing a pass. They found that the degree distributions
fitted with a truncated gamma distribution, showing that each player moves around his own
home position when interacting with the ball. Cotta and colleagues [25] also identified this
positional feature when characterizing the Spanish national team style of play, by analysing
passing network during the 2010 FIFA world cup. Such ball-passing dynamics not only influ-
ence the teams’ dynamics, but also the local players’ dynamics (e.g., movement reconfigura-
tions) [26,27]. That is, after a pass, the player in possession of the ball switches, change the
topology of the passing network (i.e. ball-passing probability distribution) and the perceptual-
motor workspace of the player in possession of the ball.
Football players’ perceptual-motor workspace is the dynamic interface between informa-
tion flows coming from perception, and kinetic flows from action [28,29]. The channelling of
players’ exploratory activity is provided by affordances, i.e., possibilities for action mainly
offered by the environment [30,31]. In fact, they map environmental information onto
movement behaviour within the perceptual-motor workspace in accordance with the demands
of the task goal. Due to the multiple ways to score a point in the football game, there is no
predetermined interpersonal coordination to be executed in order to attain the goal. Hence,
there is a lack of convergence of the exploratory behaviour towards a pre-defined intra- or
interpersonal configuration. This situation allows very subtle and consequently unpredictable
interactions between the environmental information flow and the performer’s organismic
constraints, to decide which particular action, from the perceptual-motor workspace, will be
performed at each moment.
These informational variables constrain the system, pushing it towards different states of
organization or concrete coordination patterns [32,33]. The manipulation of constraints, espe-
cially those related to the tasks, will induce the stability of patterns and the functional explora-
tion through the transitions between organizational states inherent to sports teams [34]. The
adequate modification or variations of this type of constraints can produce functional changes
in behaviour in competitive and performance contexts. These qualitative changes in the coor-
dinative system states are characterized by the loss of stability from a previous state to the cur-
rent one. In this sense, behavioural dynamics correspond to the change of trajectories, i.e.
bifurcations, that occur in the performance state space, making reference to the whole set of
possible coordinative states of a system [35].
Key environmental information, and the manipulation of task constraints, will allow play-
ers to explore a variety of functional movements. That is to say, in order to foster the emer-
gence of functional tactical behaviours, coaches should manipulate training drills enhancing
the number of possibilities for action that can satisfy goals, or suppressing habitual behaviours
[36,37]. In individual sports, it has been experimentally demonstrated that constraining such
habitual behavioural modes can promote the emergence of novel behaviours. In fact, this
manipulation promotes an indirect release of constraints and increases the likelihood of spe-
cific emergent behaviour and/or enhances the potential for exploratory behaviour [36,38].
This exploration enhancing mechanism was called ‘the connected door effect’, and it enables
greater fluency and flexibility as well as innovative actions [39].
In association football, several studies have reported the effects of manipulating task con-
straints. For example, increasing the number of goals amplifies possibilities for scoring and,
consequently, a stabilization of protecting goal patterns will likely emerge [40]. The higher the
number of players in small-sided games requires a higher level of collective organization and
optimized space occupation [41]. The number of players involved in the match clearly affects
the stability of the team structure, constrains players to using a specific pitch location and
defines roles during the match. Curiously, when increasing the relative space per player by
reducing the number of players, the players’ positioning tends to be more irregular; however,
when the field dimension increases that tendency changes and players’ positioning becomes
more regular [42].
Furthermore, Vilar and colleagues [43] found that, for most of the match time, two team
members were located in center-back and middle sub-areas of the effective playing space while
one player was located in the back- and front-wing and center-front sub-areas. However, their
results showed that players’ numerical relations were differently distributed depending on the
sub-areas, postulating that local numerical dominance plays a key role in offensive and defen-
sive success. Several studies have explored the influence of numerical imbalances on perfor-
mance [44–46]. For example, Gonçalves and colleagues [47] analyzed the effects on tactical
behaviour when playing with different numbers of opponents or teammates and found higher
values of irregularity and variability in players’ positioning relative to their centroid (i.e., the
geometrical center of the team) when playing with more than one teammate or opponent.
Moreover, under numerical superiority, the players’ distance from their nearest opponent
increased, expanding the space and time for interaction. Conversely, Ric and colleagues [48]
analyzed the similarity of individual tactical patterns and found that numerical superiority
allowed players to explore more varied configurations. However, it was reported that this
capacity to explore different patterns expired in time lags up to 20-to-30 seconds, recommend-
ing the use of wild-card players that enter and leave the game in that timescale. In basketball,
Bourbousson and colleagues [49] found that disruption in inter-team coordination fosters the
drive action, proposing a task including an extra player, who systematically plays in offense to
perturb the coordination modes of defenders. These situations require an increase in players’
breadth of attention to solve the numerical disadvantage, allowing the player in possession of
the ball and the supporting players to act within an advantageous context of play. These find-
ings may encourage coaches to use task constraints to preserve the players’ specific locations
and roles, as well as the local and temporary nature of numerical imbalances in order to respect
the representative nature of training drills.
According to all the presented literature, a task was designed where players were con-
strained to create local imbalances while preserving their specific playing location and roles.
The effects of these constraints on the exploratory dynamics and behavioural tendencies
remain unknown. Therefore, the aim of this study was to identify how players’ spatial restric-
tions affected their exploration of tactical behaviour and constrained the perceptual-motor
workspace of players in possession of the ball. We hypothesized that spatial restrictions would
increase the rate and breath of player’s exploratory behaviour. Moreover, coach staff expected
that these task constraints would allow players to possess the ball far from the opponent in
closer zones to the target, enhancing the passing flow.
Methods
Participants and procedures
Twenty-one male professional football players (age: 25.1±4.1 years; playing experience: 18.8
±5.3 years) participated in the study. All the players were informed about the procedures and
signed a consent form to participate voluntarily in the study. Players were divided into two
teams, taking into account their playing positions and their physical, technical and tactical lev-
els according to a coach’s subjective criteria. This was done to ensure that the performance
level of each team to be comparable [50]. The analyzed team comprised ten outfield players
plus a goalkeeper. The opposing team comprised nine outfield players and a goalkeeper. The
opponent goalkeeper was placed in the opponent target zone situated 37.5 m away from the
penalty area, supporting the offensive action of his teammates as an extra player. Based on the
expert knowledge of experienced football coaches and on earlier studies [43,51], the pitch was
divided into nine different sub-areas between the penalty area and the target zone, and the
players were placed according to their position roles (see Fig 1). The head coach then con-
strained the players’ movements between the sub-areas according to the following instructions:
a) players were not allowed to move out of the sub-areas during the game, except for the player
in possession of the ball; b) all players were allowed to move to any adjacent sub-area, and; c)
non-specific spatial constraints. Each of these situations was played for a 5-minute period with
a 3-minute period of passive rest. The three situations were repeated twice in a randomized
order. The sample size was determined by a power analysis, computed using G Power 3.1 [42]
for an effect size of d = 1, α < 0.05, power (1 – β) = 0.95. In similar studies of exploratory
dynamics [38], large effect sizes were observed (the smallest value was d = 0.84). The study was
approved by the Clinical Research Ethics Committee of the Catalan Institute of Health, Uni-
versity Hospital Arnau de Vilanova, Generalitat de Catalunya (Project-CEIC 1325), which fol-
lows the recommendations of the Declaration of Helsinki.
Data collection
All players wore, on their upper backs, a GPS device (SPI-Pro, GPSports, Canberra, ACT, Aus-
tralia) that captured the latitudinal and longitudinal coordinates with a sampling frequency of
5 Hz. The time series were exported and transformed into a Cartesian coordinate system using
dedicated routines in Matlab R2014b (MathWorks, Inc., Massachusetts, USA) (see Folgado
et al. [15] for complete guidelines). The players’ positioning data were resampled using two-
frames moving average filter to reduce tracking error noise. In addition, the player who
received the ball was notated, as was the time that elapsed between his reception of the ball and
that of another teammate before a pass. The dynamic players’ positional data was used to cal-
culate movement speed [15], distance from the nearest opponent [52], distance from the target
[53] and tactical actions [48,54]. Tactical actions were estimated from players’ trajectories rela-
tive to the position of the player in possession of the ball. The data collected for each player
yielded configuration states derived from the 37 variables pertaining to the four tactical mea-
sures named above (see Table 1). Players changed their states during the 5-minute game (i.e.,
300 seconds each game). Thus, every window of one second was defined as a 37-component
binary vector (column) representing the full configuration state, ascribing a value of 1 for
active categories and 0 for the inactive ones. This enabled the formation of a temporal 37 x 300
multivariate binary (Boolean) matrix.
Table 1. Data collected to assess the tactical patterns of each player, formed by 37 categories. Each
data vector represented a player’s configuration in a 4D-category space.
https://doi.org/10.1371/journal.pone.0180773.t001
Data analysis
Dynamic overlap qd(t) was used to determine the region of the performer-environment state
space explored by the players and the rate of exploration on different timescales [38,48,55].
The overlap was defined as a cosine similarity between two binary configuration vectors at
ever-increasing time distances (i.e., time lags), capturing the mean similarity of configuration
states. The mean dynamic overlap was then fitted by the following equation, which is derived
for systems with an intricate hierarchical structure (see Sibani & Dall [56]):
< qd ðtÞ >¼ ð1 qstat Þt a þ qstat
From the known mean dynamic overlap (<qd(t)>) at every time lag (t), three different
parameters were calculated. qstat is the asymptotic value of the dynamic overlap (i.e. the hori-
zontal line towards which the curve, adjusted by the equation, tends to infinity). The α param-
eter is the slope of the curve. Finally, T is the time point where, for a fixed value of 0.05, the
asymptotic value intersects with the curve of the non-linear model (see upper panels of Fig 2
for a better interpretation).
In order to understand the ball dynamics, the pooled means of time that a player was in
possession of the ball before passing it to a teammate, losing it or scoring a goal were calcu-
lated. The total passes performed with successful reception by a teammate, as well as turnovers
(losing possession of the ball) and goals scored were also taken into account, obtaining the
relative frequencies of players’ passing interactions. Transition probabilities were calculated
dividing the number of each player’s passes to his teammates, turnovers and goals by the total
number of player interactions. Thus, the most probable sequences of passes between players
allowed the inter-player interaction tendencies and the characteristic passing channels to
reach the target/scoring zone to be identified.
Fig 2. The upper panels show an example of evolution in the mean dynamic overlap of the same player for three different task constraints of a
player: restricted (left), semi (center), free (right). The blue lines represent the adjusted curve to the non-linear function, the grey lines represent the
stationary <qstat> value when the curve tends to infinity, the red lines represent the time lag in which, for a fixed value of 0.05, the asymptotic value intersects
with the curve. The lower panels show the mean values for <qstat> (left), α exponent (center) and time lag (right). Differences in mean are expressed as
percentages (±90% CL). The asterisks indicate the likelihood for the magnitude of the true difference in means as follows: *possible; **likely; ***very likely;
****most likely. The letters denote the effect sizes: T = trivial; S = small; M = moderate; L = large.
https://doi.org/10.1371/journal.pone.0180773.g002
In addition, distance from the nearest opponent and distance from the target were the state
variables, and the 2D space that they spanned was the configuration (state) space of the player
in possession of the ball. The configuration space contains all possible configurations under
the task constraints. The relative frequencies of the player in possession of the ball’s configura-
tion space were taken into account for each task constraint. The probabilities of each configu-
ration were calculated as limit (large N) relative frequencies for stationary distributions:
pi ¼ ni =N , where ni is the frequency of the configuration and N is the total frequencies. This
data allowed a potential landscape to be depicted, which represented the configuration space
of the player in possession of the ball for each task constraint. The potential values Vi were cal-
culated following [57] as: Vi ¼ Q lnðPi =N Þ; where Q = 1 is the standardized variance term; and
N is the total number of configurations in the space spanned by the two state variables.
Statistical analysis
Practical differences among task constraints were analysed using magnitude-based inferences
[58]. All data were first log-transformed to reduce bias arising from non-uniformity error. A
descriptive analysis was performed using mean and standard deviations for each variable (the
mean shown is the back-transformed mean of the log transformation). Uncertainty in the dif-
ferences was expressed as 90% of confidence limits (CL). Quantitative chances were assessed
qualitatively and reported using the following scale: 25−75%, possible; 75−95%, likely; 95
−99%, very likely; >99%, most likely. A difference was assessed as being unclear if the confi-
dence interval (CI) overlapped both substantially positive and negative thresholds. Cohen’s d
effect size at 90% CL was calculated using pooled standard deviation for comparisons, and the
magnitude ranges for mean differences were: 0–0.2 trivial; >0.2–0.6 small; >0.6–1.2 moderate;
>1.2–2 large; >2 very large [48]. In addition, the Chi-square test was used to compare the fre-
quencies of players’ interactions and frequencies of the configuration space between the three
task constraints.
Results
Dynamics of players’ exploratory behaviour
Dynamic overlap analysis allowed the slow dynamics on a long timescale (where players’
exploration became sufficiently saturated), and the quick dynamics on a shorter timescale
(related to the initial relaxation part of the overlap) to be determined (see upper panels of Fig 2
as an example of one player for each task constraint). The lower panel of Fig 2 shows the
descriptive and statistical analysis for all three parameters extracted from the non-linear
model. Long-term exploratory breadth most likely decreased (higher <qstat>) when playing
under restricted space compared to semi (large effect) and free space (large effect) conditions.
Also, a possible decrease was identified by comparing semi and free space scenarios (small
effect). When playing under restricted space conditions, players’ exploratory dynamics quickly
attained the stationary value (higher α) and likely decreased during the other two training situ-
ations. So, the larger the slope (α exponent), the quicker the players’ exploration. There were
unclear differences between semi and free space conditions. Finally, the time lag in which the
exploratory behaviour became saturated most likely (moderate/large effect) increased in the
tasks under semi and free space conditions, respectively, compared to the restricted condition.
Table 2. Descriptive analysis (mean±SD) of mean time of players’ ball possession, frequencies of passing interaction and frequencies of the
player in possession of the ball’s configurations. Difference in means, uncertainty in the true differences, based on probability chances, and Standardized
Cohen’s d differences among training game situations.
Restricted Semi Free Difference in means, Chances for smaller/ Uncertainty in the Standardized Cohen’s
%; ±90% CL similar/ greater value true differences d; ± 90% CL
a) -4.6 ±17.3 11/52/36 unclear -0.11±0.41
Δt ball 3.19±1.47 3.03 2.71 b) -7.1 ±21.0 12/41/47 unclear -0.17±0.51
possession ±1.08 ±0.73
n 20 20 20 c) -8.7 ±13.8 3/44/53 possibly # -0.21±0.35
a) -15.6 ±24.2 5/30/65 unclear -0.32±0.52
Passing 0.72±1.27 0.57 0.72 b) -1.3 ±20.4 17/61/22 unclear -0.02±0.38
interactions ±0.94 ±1.01
n 110 110 110 c) 25.1 ±26.4 83/17/1 likely " 0.42±0.39
a) -26.8 ±13.4 0/9/91 likely # -0.37±0.22
Configuration 3.22±6.18 2.63 2.40 b) -34.9 ±9.9 0/0/100 most likely # -0.51±0.18
space ±4.26 ±3.08
n 108 108 108 c) -4.6 ±15.9 2/86/12 likely trivial -0.06±0.20
Note: Δt = mean time; CL = confidence limits; " = increase; # = decrease. Comparisons between the three different training situations are identified as: a)
restricted vs semi, b) restricted vs free and c) semi vs free
https://doi.org/10.1371/journal.pone.0180773.t002
space condition (~-9%) compared to the semi space condition, while an unclear effect was
shown between the restricted and the other two task constraints. Fig 3 shows that the mean
time of players in ball possession was larger under restricted condition, specially central
defenders.
The Chi-square test did not show any significant differences (p<0.05) for each pair compar-
ison. The passing interactions likely increased under the free space condition compared to the
semi space condition (difference in means, %; ±90% CL: -25.1; ±26.4). Unclear differences
were shown when comparing the restricted space condition with the other two task constraints,
although a small effect (-0.32±0.52) was found between restricted and semi space conditions
(see Table 2). As shown in the network for the restricted space situation (see upper panels of
Fig 3), the most probable passing channel was found among players situated in the central cor-
ridor, connecting with the wings to score a goal or to pass to the forward for scoring. Not
allowing the players to move out of their zone decreased the possibilities of passing interaction
between full-backs and their teammates. This tendency changed when players were allowed to
move to the adjacent sub-area, as did the probability of interaction between midfielders and
forwards. In contrast, the probability of interaction with forwards decreased in the situation
where players’ movements were not restricted. In addition, the probability of a player passing
to the central midfielder increased.
Fig 3. Upper panels: Network diagrams obtained from each task constraint. Size of nodes represents the mean
time of players in possession of the ball. The width represents the frequency number of passes. Probability of passing
interactions was depicted as the following soften scale: 0 –blue, 0.5 –yellow, 1 –red. Lower panels: Potential landscapes
formed by two state (coordinative) variables of the player in possession of the ball (distance from the target and nearest
opponent) under the three different task constraints. The 3D deeper wells correspond to 2D-projected more stable (i.e.,
more probable) red areas. The blue areas correspond to unstable coordinative states. Less stable coordinative states
are more likely to decay into more stable states.
https://doi.org/10.1371/journal.pone.0180773.g003
free space conditions). When comparing semi and free space conditions, there were likely triv-
ial (-4.6±15.9%) differences with a trivial effect (Table 2).
The high-value surfaces of the potential landscape define unstable coordinative states while
the minimum values define highly stable coordinative states (see lower panels of Fig 3). The
first thing to notice is that the changing task constraints brought about not only quantitative
changes but also qualitative changes, i.e. bifurcations, in the configuration space. The stability
profile and also the number of stable minima changed as a consequence of varying the task
constraints. However, under all three task constraints, there were also invariant features of the
potential landscape. On the one hand, a “free player” coordinative state was constantly unsta-
ble. This meant that the player in possession of the ball was five meters away from his nearest
opponent. Under the restricted space condition, deeper valleys were depicted, showing that the
player in possession of the ball worked within highly stable contexts. Semi space conditions,
such as those where the ball was taken into distant zones, afforded less stable local states. How-
ever, when the ball was fewer than 30 meters from the scoring zone, the opponent delayed the
ball carrier’s action (i.e., distance from the nearest opponent less than five meters), depicting
other stable configurations for all three conditions. Moreover, all potential landscapes showed
how the pressure increased when the players took the ball closer to the target zone.
Discussion
This study aimed to present the influence of spatial restrictions on players’ movements on the
exploratory dynamics of tactical behaviour, ball flow dynamics and performance contexts
when players were in possession of the ball. The main finding suggests that restricting players’
movements out of their home sub-area enhanced their exploration in a short timescale, but
not their long-term exploratory behaviour. Furthermore, results did not reveal that spatial
restrictions help players to possess the ball in advantageous conditions, i.e. far from the oppo-
nent in advanced zones. However, allowing players to move out of their home sub-area fos-
tered the ball flow dynamics and the inter-player passing relations.
In terms of exploratory dynamics, players produced a higher long-term exploratory breadth
(i.e., lower <qstat>) when allowed to move out of their specific locations. While the results
might seem obvious, previous studies have shown that increasing the space of interaction leads
to decreases in the variability of players’ spatial distribution [42,59]. These differences may be
due to the multivariate analysis used on here, which involved tactical actions, as well as to tra-
jectories, distances, and movement-speed state variables. The players’ exploration quickly
became saturated under restricted spatial condition. This means that players were constantly
exploring new task solutions but quickly expired the whole set of possibilities for action. On
the contrary, relaxing spatial restrictions increased long-term exploratory behaviour. Some
previous results showed that a high number of opponents generally impaired long-term tacti-
cal exploration [48]. However, these constraints increased the probability of exploring certain
specific coordinative states, e.g., depth mobility back to the defence.
The manipulation of task constraints changes the probabilities of emergent tactical be-
haviours and the exploration in different regions of the task solution space [55,60] Players’
behaviour can be defined by several variables that shows the relations between the system com-
ponents. In this sense, performance coordinative variables, such as distance to opponents and
the target can serve as parsimonious macroscopic descriptors of what happens at the micro-
scopic level [53,61].
The depiction of these state variables on a performance configuration space gives a clear
picture, thus allowing it to be determined whether or not the task constraints analysed foster
an advantageous performance game scenario. For example, the restricted space condition
allowed players to play the ball far from their nearest opponent when decreasing the distance
from the target zone. This task constraint forced players to keep a large team surface area,
thereby fostering penetration actions or passing the ball to a free teammate [7,49]. During the
semi condition, this coordinative state had a greater degree of instability. Conversely, allowing
players to move to the adjacent sub-area increased the stability of coordinative states, with the
player in possession of the ball delayed in zones far from the target. The depth of the potential
of the player in possession of the ball clearly showed how forbidding players to move out of
their home sub-area can limit the exploration of task solutions yet foster the stability of con-
crete task solution states [48,62]. Moreover, it was shown that the change of task constraints
changes not only the quantitative aspects of the game, e.g. modifying the stability degree of
extant stable coordinative states, but also causes a formation of new stable states. This means
that changes of task constraints may create qualitatively new contexts of exploration for players
that cannot be interchangeable with respect to expected effects.
Network analysis allowed quantifying players performance [20] and identifying different
player solicitations. In that sense, the right fullbacks were no able to succeed during their inter-
ventions with the ball in all three scenarios. However, playing under the semi space condition
reduced the barriers to interacting with full-backs compared to the other two conditions. Pre-
vious network analysis in football has shown preferred passing channels and how they changed
depending on the opposing team [25,63]. Positional constrains might be considered to simu-
late opponents’ defensive strategies. Allowing players to move on the adjacent sub-areas, wing
opponents were able to press players located in the central corridor, leaving the fullbacks and
wing forward players free of a marker, fostering the association with them. However, that task
constraints lead to a lower centralized passing interaction patterns and higher intra-team well-
connected passing relations, these behaviours are related with a better team performance
[18,19]. Grund [64] found that high levels of interaction (i.e., passes per minute weighted by
possession) led to increased team performance (i.e., goals). The more the players were in pos-
session of the ball, the fewer passes per minute could be performed. According to this state-
ment, relaxing players’ positional constraints would enable an increase in team performance
regarding the quickness of ball flow dynamics and decreasing the players’ dwell time in posses-
sion of the ball. Larger passes samples would probably allow to identify the common features
and emergent properties of the dynamic network topology under the effects of different task
constraints [23,24].
In light of these results, when the coach used sub-spaces to constrain players’ behaviours,
the exploratory behaviours of players quickly attained the stationary value, showing larger val-
ues of overlapped behaviours on a shorter timescale. Nevertheless, these positional restrictions
helped to find advantageous scenarios for the player in possession of the ball. It has been dem-
onstrated that stability in the degree of free movement of players can be constrained by chang-
ing the relative space per player by increasing the number of players [52]. Five-a-side formats
increase the percentage of time spent from four to six meters away from the nearest opponent
but decrease the probability of taking action from six meters free of opponents, as more advan-
tageous contexts of play. Interestingly, Silva and colleagues [59] established that, with a larger
(152 m2) relative space per player, the percentage of time spent in numerical advantage very
likely increased by changing the field dimension instead of the number of players. Under
imbalanced small-sided game conditions, the radius of free movement most likely increased
for those players who competed in numerical superiority [47]. From these and previous results,
a clear role of the constraints with respect to coaches’ goals arises. If the goal of the coach is to
generally increase the individual player’s exploratory behaviour, he/she would have to allow
players to move to other sub-areas of the pitch for the training task. If the goal is to elicit and/
or stabilize specific coordinative states, then restricting positions to a concrete zone may be an
optimal training context. Therefore, coaches should carefully design their training tasks. For
this purpose, reading later studies that evidence the tactical effects of modifying tasks con-
straints is advisable. In addition, more research is needed in order to better understand the
effects of task constraints involving large numbers of players, in which the spatiotemporal
dynamics of game constraints is emphasized.
Previous research using a football match highlighted the importance to create local imbal-
ance on specific zones of the area of play [43]. Further research should include other relational
variables, such as angles, in order to analyse the relation between at least three system compo-
nents (e.g. ball-player-target). Moreover, multivariate analysis (e.g. Principal Component
Analysis) would allow knowing the (co) relation between the already relational variables cap-
tured characterizing the dynamical structure of tactical behaviour under the influence of task
and environmental constraints. The dynamic nature of constraints and their evolution on dif-
ferent time scales should be also considered for investigation.
Conclusion
Performance analysis can be improved by providing information about the dynamics of foot-
ball games. The combination of ball events and positional data can help coaches to understand
the effect of a task constraint on individual and collective behaviour. Moreover, the combina-
tion of relational variables can give a clear picture about the tactical performance of players
and its dynamics. Network analysis is widely used during football matches, but its use in train-
ing settings is still limited. Here, it has been reported new knowledge on how ball flow dynam-
ics could be constrained through pitch positioning restrictions. Spatial restrictions did not
stimulate the long-term players’ exploratory breadth, but increased the rate of exploration to
perform different tactical solutions on a shorter timescale. Dynamic overlap might be consid-
ered as a potential order parameter for performance analysis in sport. The depiction of rela-
tional variables also provide relevant information for tactical performance in a macroscopic
level. Overall, coaches can foster players’ exploration and/or to stabilize concrete coordinative
states by constraining the players’ space of interaction behaviour. In this sense, depending on
the training goals of the coaches i.e. enhancing the exploration of task solutions or focusing on
smaller set of task solutions, they can manipulate the constraints to attain each of them.
Author Contributions
Conceptualization: Angel Ric, Carlota Torrents, Bruno Gonçalves, Jaime Sampaio, Robert
Hristovski.
Data curation: Angel Ric, Bruno Gonçalves, Robert Hristovski.
Formal analysis: Angel Ric, Bruno Gonçalves, Jaime Sampaio, Robert Hristovski.
Funding acquisition: Carlota Torrents, Lorena Torres-Ronda.
Investigation: Angel Ric, Carlota Torrents, Bruno Gonçalves, Jaime Sampaio, Robert
Hristovski.
Methodology: Angel Ric, Carlota Torrents, Bruno Gonçalves, Robert Hristovski.
Project administration: Carlota Torrents.
Resources: Angel Ric, Carlota Torrents, Jaime Sampaio.
Software: Angel Ric, Bruno Gonçalves, Jaime Sampaio.
Supervision: Carlota Torrents, Bruno Gonçalves, Lorena Torres-Ronda, Jaime Sampaio, Rob-
ert Hristovski.
Validation: Carlota Torrents, Bruno Gonçalves, Jaime Sampaio, Robert Hristovski.
Visualization: Angel Ric, Bruno Gonçalves, Robert Hristovski.
Writing – original draft: Angel Ric, Carlota Torrents, Bruno Gonçalves, Lorena Torres-
Ronda, Jaime Sampaio, Robert Hristovski.
Writing – review & editing: Angel Ric, Carlota Torrents, Bruno Gonçalves, Lorena Torres-
Ronda, Jaime Sampaio, Robert Hristovski.
References
1. Hughes M, Franks IM. Notational analysis of sport: Systems for better coaching and performance in
sport. Psychology Press; 2004.
2. Castellano J, Casamichana D, Lago C. The use of match statistics that discriminate between successful
and unsuccessful soccer teams. J Hum Kinet. 2012; 31: 137–147.
3. Rampinini E, Impellizzeri FM, Castagna C, Coutts AJ, Wisløff U. Technical performance during soccer
matches of the Italian Serie A league: Effect of fatigue and competitive level. J Sci Med Sport. 2009; 12:
227–233. https://doi.org/10.1016/j.jsams.2007.10.002 PMID: 18083631
4. Liu H, Gómez M-A, Gonçalves B, Sampaio J. Technical performance and match-to-match variation in
elite football teams. J Sports Sci. 2016; 34: 509–18. https://doi.org/10.1080/02640414.2015.1117121
PMID: 26613399
5. Sampaio J, McGarry T, Calleja-González J, Jiménez Sáiz S, Schelling I Del Alcázar X, Balciunas M.
Exploring game performance in the National Basketball Association using player tracking data. PLoS
One. 2015; 10. https://doi.org/10.1371/journal.pone.0132894 PMID: 26171606
6. James N. Notational analysis in soccer: past, present and future. Int J Perform Anal Sport. 2006; 6: 67–
81.
7. Travassos B, Araújo D, Duarte R, McGarry T. Spatiotemporal coordination behaviors in futsal (indoor
football) are guided by informational game constraints. Hum Mov Sci. 2012; 31: 932–945. https://doi.
org/10.1016/j.humov.2011.10.004 PMID: 22672740
8. Vilar L, Araújo D, Davids K, Travassos B, Duarte R, Parreira J. Interpersonal coordination tendencies
supporting the creation/prevention of goal scoring opportunities in futsal. Eur J Sport Sci. Taylor & Fran-
cis; 2014; 14: 28–35. https://doi.org/10.1080/17461391.2012.725103 PMID: 24533492
9. Travassos B, Araújo D, Davids K, Esteves PT, Fernandes O. Improving passing actions in team sports
by developing interpersonal interactions between players. Int J Sport Sci Coach. Multi Science Publish-
ing; 2012; 7: 677–688.
10. Corrêa UC, Pinho ST De, Letı́cia S, Clavijo AR, Souza TDO, Tani G, et al. Revealing the decision-mak-
ing of dribbling in the sport of futsal Revealing the decision-making of dribbling in the sport of futsal. J
Sports Sci. Routledge; 2016; 34: 2321–2328. https://doi.org/10.1080/02640414.2016.1232488 PMID:
27868480
11. Passos P, Araújo D, Davids K, Gouveia L, Milho J, Serpa S. Information-governing dynamics of
attacker-defender interactions in youth Rugby Union. J Sport Sci. 2008; 26: 1421–1429. https://doi.org/
10.1080/02640410802208986 PMID: 18923958
12. McGarry T. Applied and theoretical perspectives of performance analysis in sport: Scientific issues and
challenges. Int J Perform Anal Sport. 2009; 9: 128–140.
13. Balagué N, Torrents C, Pol R, Seirul-lo F. Entrenamiento integrado. Principios dinámicos y aplica-
ciones. Apunt Educ fı́sica y deportes 2014; 116:60–68. https://doi.org/10.5672/apunts.2014-0983.es.
(2014/2).116.06
14. Travassos B, Araújo D, Vilar L, McGarry T. Interpersonal coordination and ball dynamics in futsal
(indoor football). Hum Mov Sci. 2011; 30: 1245–1259. https://doi.org/10.1016/j.humov.2011.04.003
PMID: 21683464
15. Folgado H, Duarte R, Fernandes O, Sampaio J. Competing with lower level opponents decreases intra-
team movement synchronization and time-motion demands during pre-season soccer matches. PLoS
One. 2014; 9: e97145. https://doi.org/10.1371/journal.pone.0097145 PMID: 24817186
16. Tenga A, Larsen Ø. Testing the Validity of Match Analysis to describe Playing Styles in Football. Inter-
national Journal of Performance Analysis in Sport. 2003; 3:90–102.
17. Kempe M, Vogelbein M, Memmert D, Nopp S. Possession vs. Direct Play: Evaluating Tactical Behavior
in Elite Soccer. Int J Sport Sci. 2014; 4: 35–41. https://doi.org/10.5923/s.sports.201401.05
18. Gonçalves B, Coutinho D, Santos S, Lago-Penas C, Jiménez S, Sampaio J. Exploring Team Passing
Networks and Player Movement Dynamics in Youth Association Football. PLoS One. Public Library of
Science; 2017; 12: e0171156. https://doi.org/10.1371/journal.pone.0171156 PMID: 28141823
19. López Peña J, Touchette H. A network theory analysis of football strategies. In: (ed.) CC, editor. Sports
Physics: Proc 2012 Euromech Physics of Sports Conference; Palaiseau: Editions de l’Ecole Polytechni-
que; 2012. p. 517–28.
20. Duch J, Waitzman JS, Amaral LA. Quantifying the performance of individual players in a team activity.
PLoS One. 2010; 5(6):e10937. https://doi.org/10.1371/journal.pone.0010937 PMID: 20585387
21. Gyarmati L., Kwak H., Rodriguez P. Searching for a unique style in soccer, in Proceedings of the 2014
KDD Workshop on Large-Scale Sports Analytics. arXiv: 1409.0308.
22. Cintia P, Giannotti F, Pappalardo L, Pedreschi D, Malvaldi M. The harsh rule of the goals: data-driven
performance indicators for football teams. Data Science and Advanced Analytics (DSAA), 2015 36678
2015 IEEE International Conference on. IEEE; 2015. pp. 1–10.
23. Yamamoto Y, Yokoyama K. Common and Unique Network Dynamics in Football Games. PLoS One.
2011; 6: e29638. https://doi.org/10.1371/journal.pone.0029638 PMID: 22216336
24. Narizuka T, Yamamoto K, Yamazaki Y. Statistical properties of position-dependent ball-passing net-
works in football games. Phys A Stat Mech its Appl. 2014; 412: 157–168. https://doi.org/10.1016/j.
physa.2014.06.037
25. Cotta C, Mora AM, Merelo JJ, Merelo-Molina C. A network analysis of the 2010 FIFA world cup cham-
pion team play. J Syst Sci Complex. 2013; 26: 21–42. https://doi.org/10.1007/s11424-013-2291-2
26. Ric A, Torrents C, Gonçalves B, Sampaio J, Hristovski R. Soft-Assembled Multilevel Dynamics of Tacti-
cal Behaviors in Soccer. Front Psychol. 2016; 7: 1513. https://doi.org/10.3389/fpsyg.2016.01513 PMID:
27761120
27. Wang Q, Zhu H, Hu W, Shen Z, Yao Y. Discerning Tactical Patterns for Professional Soccer Teams: An
Enhanced Topic Model with Applications. Proceedings of the 21th ACM SIGKDD International Confer-
ence on Knowledge Discovery and Data Mining. New York, NY, USA: ACM; 2015. pp. 2197–2206.
10.1145/2783258.2788577
28. Newell KM, McDonald P V, Kugler PN. The perceptual-motor workspace and the acquisition of skill. In:
Requin J, Stelmach GE, editors. Tutorials in motor neuroscience. Springer; 1991. pp. 95–108.
29. Kugler PN, Turvey MT. Resources for ecological psychology. Information, natural law, and the self-
assembly of rhythmic movement. Hillsdale, NJ, England: Lawrence Erlbaum Associates; 1987.
30. Gibson JJ. The ecological approach to visual perception. Boston, MA, US: Houghton, Mifflin and Com-
pany; 1979.
31. Bruineberg J, Rietveld E. Self-organization, free energy minimization, and optimal grip on a field of affor-
dances. Front Hum Neurosci. 2014; 8: 599. https://doi.org/10.3389/fnhum.2014.00599 PMID:
25161615
32. Haken H. An approach to self-organization. Self-organizing systems: The emergence of order. New
York; 1987. pp. 417–437.
33. Kelso JAS. Dynamic patterns: the self-organization of brain and behavior. Complex Adaptive Systems
series. MIT Press, Bradford Book, Cambridge, MA; 1995.
34. Hristovski R, Davids K, Araújo D, Passos P. Constraints-induced emergence of functional novelty in
complex neurobiological systems: a basis for creativity in sport. Nonlinear Dynamics Psychol Life Sci.
2011; 15: 175–206. PMID: 21382260
35. Araujo D, Davids K, Hristovski R. The ecological dynamics of decision making in sport. Psychol Sport
Exerc. Elsevier; 2006; 7: 653–676. https://doi.org/10.1016/j.psychsport.2006.07.002
36. Hristovski R, Davids K, Araújo D. Information for Regulating Action in Sport: Metastability and Emer-
gence of Tactical Solutions. In: Araujo D, Ripoll H, Raab M, editors. Perspectives on Cognition and
Action in Sport. New York; 2009. pp. 43–57.
37. Santos SDL, Memmert D, Sampaio J, Leite N. The Spawns of Creative Behavior in Team Sports: A Cre-
ativity Developmental Framework. Front Psychol. 2016; 7: 1282. https://doi.org/10.3389/fpsyg.2016.
01282 PMID: 27617000
38. Torrents C, Ric A, Hristovski R. Creativity and emergence of specific dance movements using instruc-
tional constraints. Psychol Aesthetics, Creat Arts. 2015; 9: 65–74. https://doi.org/10.1037/a0038706
39. Hristovski R, Davids K, Passos P, Araújo D. Sport Performance as a Domain of Creative Problem Solv-
ing for Self-Organizing Performer-Environment Systems. Open Sports Sci J. 2012; 5: 26–35. https://
doi.org/10.2174/1875399X01205010026
40. Travassos B, Gonçalves B, Marcelino R, Monteiro R, Sampaio J. How perceiving additional targets
modifies teams’ tactical behavior during football small-sided games. Hum Mov Sci. Elsevier B.V.; 2014;
38: 241–50. https://doi.org/10.1016/j.humov.2014.10.005 PMID: 25457422
41. Aguiar M, Gonçalves B, Botelho G, Lemmink K, Sampaio J. Footballers’ movement behaviour during 2-
, 3-, 4- and 5-a-side small-sided games. J Sports Sci. Routledge; 2015; 33: 1259–1266. https://doi.org/
10.1080/02640414.2015.1022571 PMID: 25782702
42. Silva P, Esteves P, Correia V, Davids K, Araújo D, Garganta J. Effects of manipulations of player num-
bers vs. field dimensions on inter-individual coordination during small-sided games in youth football.
International Journal of Performance Analysis in Sport. 2015. pp. 641–659.
43. Vilar L, Araújo D, Davids K, Bar-Yam Y. Science of winning soccer: Emergent pattern-forming dynamics
in association football. J Syst Sci Complex. 2013; 26: 73–84. https://doi.org/10.1007/s11424-013-2286-
z
44. Sampaio JE, Lago C, Gonçalves B, Maçãs VM, Leite N. Effects of pacing, status and unbalance in time
motion variables, heart rate and tactical behaviour when playing 5-a-side football small-sided games. J
Sci Med Sport. 2014; 17: 229–233. https://doi.org/10.1016/j.jsams.2013.04.005 PMID: 23683687
45. Silva P, Travassos B, Vilar L, Aguiar P, Davids K, Araújo D, et al. Numerical relations and skill level con-
strain co-adaptive behaviors of agents in sports teams. PLoS One. 2014; 9: e107112. https://doi.org/10.
1371/journal.pone.0107112 PMID: 25191870
46. Vilar L, Esteves P, Travassos B, Passos P, Lago-Peñas C, Davids K. Varying numbers of players in
small-sided soccer games modifies action opportunities during training. Int J Sport Sci Coach. Multi Sci-
ence Publishing; 2014; 9: 1007–1018.
86
evolucionan un conjunto de variables tácticas derivadas de los datos de posición de los
jugadores (Duarte, Araújo, Folgado, et al., 2012) definen diferentes niveles de organización
dentro de la dinámica de comportamiento táctico. Concretamente, el índice de dispersión
(Stretch index) y la localización del centro geométrico del equipo evolucionaron en una
escala temporal superior a veinte segundos. El centro geométrico del equipo permaneció
localizado en un sector y carril del campo durante un promedio de 22 y 37 segundos,
respectivamente, durante la primera parte, y de 24 y 30 durante la segunda. Precisamente la
estabilidad temporal de estas variables (prolongación de una categoría activa conforme
transcurre el tiempo) fueron las que arrojaron correlaciones significativas en el último
orden de componentes principales indicando las bases sobre las que se estructura el
comportamiento táctico del equipo. Curiosamente la alternancia en la posesión del balón
transcurrió en una escala temporal similar por lo que las dinámicas de cambio de estas
variables pueden estar asociadas a ella (Bourbousson, Seve, et al., 2010; Frencken et al.,
2011). En cambio, las variables que mostraban la velocidad de expansión y contracción y
de movimiento del centro geométrico del equipo (en metros por segundo) cambiaron en una
escala de pocos segundos y únicamente arrojaron valores significativos en los componentes
principales extraídos en el primer nivel de análisis. A pesar de que los once jugadores que
finalizaron la primera parte fueron distintos a los once que terminaron la segunda,
existieron tendencias colectivas similares. Por ejemplo, en ambas partes se pudo identificar
que el equipo se replegaba hacia un posicionamiento intermedio en torno al centro del
campo basculando ligeramente hacia la izquierda, sin embargo en la primera parte lo
hicieron en campo contrario desde el carril central, mientras que en la segunda en campo
propio desde el carril derecho. Esto, entre otros aspectos, puede indicar que el equipo
contrario atacaba preferentemente por su banda derecha. El hecho de que este patrón sea
defensivo no implica que el equipo estuviera la mayor parte del tiempo defendiendo, sino
que los comportamientos defensivos de los jugadores son más estables (mayor correlación)
que los ofensivos (Travassos, Araújo, Duarte y McGarry, 2012; Travassos et al., 2011). A
pesar de ello este tipo de análisis también permitió identificar patrones ofensivos e
inmediatos a la recuperación o pérdida del balón así como de transición entre ellos. Por otro
lado, la correlación entre estos patrones de primer orden captura la estructura jerárquica de
variabilidad de comportamiento que culmina con la obtención de una base global de
87
comportamiento táctico que atrae al sistema. Por lo tanto, la emergencia de patrones
tácticos se realiza de forma totalmente correlacionada, es decir que las bases de
comportamiento que atraen al sistema están integradas en otras, formando subatractores
dentro de atractores. En resumen, la correlación horizontal en la dinámica de
comportamiento táctico forma y estabiliza patrones de comportamiento en un orden
superior (vertical), que a su vez gobierna aquellos de orden inferior, los cuales residen en
escalas espacio-temporales más pequeñas. Por todo ello, en la adquisición y desarrollo de
comportamiento táctico no existe una ruta o dirección preferida (de arriba hacia abajo o de
abajo hacia arriba), sino que debe entenderse como una relación circular resultante de la
influencia recíproca entre los diferentes nivel o escalas espacio-temporales (Haken, 1983).
Estos resultados confirmarían pues una estructura jerárquica de la dinámica de
comportamiento en deportes colectivos, previamente ya identificada en deportes o prácticas
motrices de oposición (Hristovski et al., 2011) y de cooperación (Torrents et al., 2015).
88
lapsos de tiempo de más de sesenta segundos presentaron altos valores de similitud. Por lo
tanto, la dinámica exploratoria muestra tendencias distintas dependiendo de la escala
temporal, lo que sugiere nuevamente una dinámica metaestable, o lo que es lo mismo,
temporalmente estable a corto plazo pero inestable a largo. El cálculo del tiempo promedio
(en segundos) que los jugadores tardaron en cambiar de acción táctica, contexto de
interacción, zona del campo o velocidad de movimiento (τ1 = 3,09 ± 3,01 seg.); y el
correspondiente al conjunto de acciones ofensivas o defensivas hasta que el equipo perdía o
recuperaba el balón (τ2 = 12,71 ± 9,32 seg.) corroboró la existencia de escalas temporales
significativamente separadas, definiendo inequívocamente dos niveles dentro de la
dinámica de comportamiento táctico. La representación de las probabilidades condicionales
de las configuraciones de comportamiento en un paisaje potencial permitió comparar la
dinámica metaestable de las tres situaciones en desequilibrio numérico. La rugosidad de los
diferentes paisajes mostró claramente como la situación en superioridad numérica provocó
una dinámica de comportamiento más flexible, mientras que las situaciones en inferioridad
numérica mostraron un paisaje mucho más abrupto. Esto supuso una dinámica de
comportamiento más rígida al jugar en inferioridad, donde los jugadores tendieron a
realizar principalmente coberturas ofensivas y balances defensivos en campo propio. Los
resultados del promedio de los valores estacionarios de la dinámica exploratoria, entropía y
fuerza de atrapamiento de las configuraciones (la probabilidad de ser capaz de repetir las
mismas configuraciones secuencialmente) también demostraron que a medida que
aumentaba el número de oponentes el comportamiento táctico individual fue menos
variado, irregular (impredecible) y flexible.
89
obtenidos en la sincronización de cada jugador con el equipo. Estos resultados sugieren que
la estabilidad del comportamiento a nivel colectivo puede compensar los valores superiores
de irregularidad (entropía) del movimiento a nivel individual, estableciéndose algún tipo de
compensación recíproca entre los comportamiento individuales y el colectivo (Riley et al.,
2011). Más adelante entraremos en detalle sobre este tipo de sinergia multinivel
considerándola de gran importancia, ya no solo a nivel teórico, sino también a nivel
práctico.
90
estuvo formada por la combinación de acciones desarrolladas por los jugadores en cada
segundo de juego. La correlación entre las configuraciones del equipo en un primer nivel de
análisis y la aplicación secuencial del mismo permitió capturar la esencia de
comportamiento colectivo constituyendo la base global de atracción en la división o
coordinación de tareas entre los jugadores (Duarte, Arajo, Correia y Davids, 2012; Eccles,
2010). Precisamente la heterogeneidad de las tareas fue una de las características esenciales
que mostró el comportamiento de los jugadores independientemente del aumento de
oponentes o compañeros. Sin embargo, lo que se pudo apreciar con el aumento del número
de oponentes es que el número de acciones que mostraron correlaciones significativas fue
superior, lo que supone un aumento de las posibilidades de combinar acciones por parte del
equipo. A nivel defensivo, disminuyeron las acciones de disuasión en pro de dar balance
defensivo. La gran superioridad numérica obligó a los jugadores a proteger el espacio
cercano a la portería disminuyendo la distancia al oponente más cercano seguramente
porque más jugadores trataron de buscar líneas de pase interiores, lo que generó amplitud
por fuera (Gonçalves et al., 2016). La alta densidad de oponentes mermó las posibilidades
de pase, probablemente por ello el recurso ofensivo del pase dejo de ser tan significativo
con el aumento del número de oponentes. Por el contrario, el incremento de compañeros
disminuyó el número de correlaciones significativas entre acciones, lo que muestra una
mayor rigidez en la base general de comportamiento colectivo, confirmando los resultados
obtenidos de amplitud exploratoria del equipo. Concretamente, el incremento del número
de compañeros hizo que el otro equipo se replegara, posiblemente por ello las acciones de
desmarque dejaron de aparecer en la estructura de comportamiento. Defensivamente, el
aumento de compañeros supuso que más jugadores dejaran de tener un par, es decir, que de
uno a tres jugadores centraron sus tareas defensivas en dar balance defensivo sosteniendo
las acciones individuales de los compañeros que buscaban la recuperación del balón.
Aunque con baja resolución, la base global de coordinación colectiva de acciones
individuales crea un núcleo de comportamiento colectivo sobre el que se vertebran
combinaciones de acciones más concretas. Éstas se desarrollan en una escala temporal más
corta, debido a los sutiles y rápidos cambios de constreñimientos físicos e informacionales
que guían las reconfiguraciones y adaptación del equipo.
91
Los juegos reducidos en desequilibrio numérico son tareas habituales de
entrenamiento, sin embargo, se ha demostrado que, durante un partido, este tipo de
desequilibrios a los que se enfrentan los jugadores varía a lo largo del tiempo y es diferente
dependiendo de la posición que ocupe cada jugador (Vilar, Araújo, Davids y Bar-Yam,
2013). Por ello, en el estudio IV, manipulando la restricción del espacio en el que los
jugadores podían actuar, se pretendía provocar situaciones de desequilibrio numérico con el
objetivo de conocer si este constreñimiento provocaba contextos favorables para actuar con
el balón, y conocer sus efectos sobre la dinámica de pases y sobre la dinámica de
comportamiento exploratorio. Los paisajes potenciales de acción del jugador con balón,
definido por la distancia respecto al oponente más cercano y a la zona de marca, arrojó de
forma probable y muy probable diferencias significativas entre restringir el espacio de
interacción a los jugadores y hacerlo de forma parcial o no restringirlo. El hecho de
restringir el espacio dio como resultado un paisaje mucho más abrupto, lo que supone un
comportamiento más rígido que en las otras dos situaciones. A pesar de las diferencias
generales, la probabilidad de que los jugadores del equipo tuvieran el balón en zonas
avanzadas libres de marca (> 5 metros) fue prácticamente nula, por lo que se puede decir
que fue un estado inestable independientemente de los constreñimientos impuestos por el
entrenador. La limitación del espacio de interacción hizo que aumentara el tiempo
promedio en el que lo jugadores poseían el balón, aunque los resultados únicamente
mostraron posibles diferencias significativas entre la limitación parcial del espacio de
interacción y el juego libre. Las restricciones espaciales de los jugadores incrementaron la
relación espacio-temporal del atacante con el balón y el oponente más cercano aunque esto
únicamente se tradujo en una mayor distribución de probabilidades de pase entre
compañeros en la situación con limitación parcial. En cuanto a la dinámica exploratoria, la
restricción total del espacio de interacción disminuyó la amplitud de exploración de los
jugadores. Sin embargo, aumentó el ratio o rapidez de exploración en escalas temporales
más cortas, es decir, las reconfiguraciones de los jugadores fueron mayores en intervalos de
tiempo menores.
92
7.2. Aportaciones teóricas y metodológicas
En el fútbol masculino el parámetro de orden fase relativa (la diferencia en las fases
de oscilación de dos variables continuas) ha sido el más utilizado para capturar el
comportamiento coordinado o estados coordinativos que surgen de la interacción jugador-
jugador o equipo-equipo (Duarte, Araújo, Davids, et al., 2012; Folgado et al., 2014, 2015;
Gonçalves, Figueira, Maçãs y Sampaio, 2014; Silva, Chung, et al., 2016). A partir de
oscilaciones en el posicionamiento de dos jugadores o variables derivadas del equipo en su
conjunto, como el centro geométrico del equipo o el índice de dispersión, la fase relativa
describe los patrones espacio-temporales de la coordinación rítmica y los cambios en la
sincronización que se producen en respuesta a los constreñimientos del juego. Las
correlaciones cruzadas (cross-correlations) y vector coding (Moura et al., 2016), las
running correlations (Duarte, Araújo, Freire, et al., 2012) o el análisis cluster phase
(Duarte et al., 2013) también han permitido detectar la existencia de estos estados
metaestables siendo identificados como valores del parámetro de orden. Estos tipos de
93
análisis permiten detectar la simetría entre dos variables continuas. Sin embargo, la
heterogeneidad de las tareas ejecutadas por los jugadores (acciones técnico-tácticas) y su
desarrollo en diferentes situaciones, es decir, zona del campo, posicionamiento relativo
entre compañeros y oponentes o la posesión o no del balón, entre otras, son variables
discretas que, generalmente, no pueden ser analizadas como simetrías simples o espejo
(ɛ·ɜ), o reajustes en una variable continua. Por eso el parámetro de orden overlap q permite
medir la simetría o similitud (correlación) entre este tipo de variables discretas (Torrents et
al., 2015), así como entre variables continuas, e incluso la combinación de ambas. En este
sentido este parámetro de orden puede ser más general y versátil, y puede utilizarse en
casos multidimensionales, es decir, que involucren a más de dos variables, y que no sean
exclusivamente continuas. Además es muy importante destacar que captura la dinámica en
múltiples escalas temporales, ya que recoge los valores de similitud en lapsos de tiempo
incrementales (Hristovski et al., 2013).
Los diferentes análisis de datos utilizados en esta tesis abren un amplio abanico de
posibilidades ofreciendo nuevas herramientas y procedimientos estadísticos para estudiar el
comportamiento táctico. Por un lado, el análisis jerárquico de componentes principales
permite identificar patrones combinando múltiples variables que pueden definir el
comportamiento táctico colectivo o individual. En el estudio I se presenta una selección de
las variables más estudiadas en la literatura reciente, sin embargo, este tipo de análisis
permite la utilización de aquellas que puedan tener mayor interés para un investigador o
investigadora, o incluso aquellas que los miembros de un staff técnico puedan considerar
como variables de rendimiento. En el estudio III se pudo identificar la correlación existente
entre las acciones que realizan los diferentes jugadores de un equipo y las de los oponentes,
de manera que se puede obtener información relevante para verificar que los jugadores de
un equipo hacen una presión adecuada respetando los diferentes roles derivados de ella, o si
ofensivamente los compañeros sin balón ofrecen diferentes soluciones al portador. Aunque
los resultados no fueron publicados, este tipo de análisis también se pudo aplicar sobre las
matrices de datos utilizados en los estudios II y IV, de los que se podría haber identificado
la correlación entre las acciones de cada jugador con otras variables contextuales como la
zona del campo donde se realizó, la posición relativa respecto a compañeros y oponentes, la
distancia a la portería y al oponente más cercano, y la velocidad de movimiento al realizar
94
cada acción, permitiendo obtener una información más contextualizada de la acción
individual de cada jugador (McGarry, 2009). Sin embargo, a partir de las probabilidades
condicionales de las configuraciones de comportamiento derivadas de la combinación de
dos de las variables anteriores, se definió el paisaje potencial de comportamiento de cada
jugador. Estos paisajes permiten dar información sobre aquellos estados más estables o
comportamientos que atraen al sistema (jugador/equipo), es decir que poseen una alta
probabilidad de que se den, así como los más inestables. Además, la rugosidad de estos
paisajes permite identificar qué tareas o modificaciones de ellas promueven
comportamientos más variados y/o flexibles, así como cuales favorecen la estabilización de
ciertos comportamientos, en función de la localización y la profundidad de las bases de
atracción del comportamiento.
Se puede decir que el fútbol, como otros deportes o prácticas motrices que se
desarrollan en entornos con gran incertidumbre y donde, por lo tanto, prima la adaptación,
debe ser pensado como un banco emergente de comportamiento donde deportista y/o
equipo continuamente exploran diferentes soluciones sobre las múltiples tareas o
95
situaciones impuestas por el comportamiento del equipo oponente. Tradicionalmente, los
programas de formación y entrenamiento se han basado en la prescripción de movimientos
predeterminados proporcionando una serie de técnicas o movimientos preestablecidos para
que el jugador o jugadora sea capaz de reproducirlos durante la competición. Sin embargo,
tal y como se ha mostrado, el comportamiento no se puede aislar de su contexto, es decir,
una misma acción será diferente en función de si se realiza en campo propio o rival, si un
jugador es el último o si está situado entre compañeros y oponentes, si el oponente más
cercano está a 2 o a 10 metros, o que a pesar de su correcta ejecución, ésta provoque un
contexto desfavorable para la consecución del objetivo a corto o largo plazo. En el fútbol,
por lo tanto, se pueden identificar una amplia variedad de comportamientos que vienen
definidos por su contexto. Dentro de esa variedad se pueden identificar configuraciones o
patrones concretos sin necesidad de ser impuestos por un agente externo (staff técnico o
docente). Así, mediante la manipulación de constreñimientos se pueden diseñar contextos
favorables que provoquen cambios significativos en la distribución de probabilidades de
comportamientos concretos modulando el paisaje potencial de acción de jugadores,
jugadoras o equipos. La modificación de uno o varios parámetros de control en la dirección
y con la intensidad adecuada puede reducir la probabilidad de que se den comportamientos
no deseados o incrementar la de los deseados para que sean más estables. Sin embargo, se
debe considerar que si desarrollamos comportamientos excesivamente estables aumentará
la rigidez del sistema perdiendo la capacidad de adaptación ante situaciones imprevistas.
Junto con la estabilidad otra característica fundamental para tener una buena capacidad de
adaptación es la flexibilidad. La estimulación de ambas durante el entrenamiento permitirá
la exploración de múltiples soluciones funcionales de mayor o menor estabilidad. El
sistema deportista-equipo deberá ser lo suficientemente flexible como para que, ante
comportamientos del oponente que disminuyan la funcionalidad de su comportamiento, sea
capaz de transitar hacia otro comportamiento o solución funcional.
96
podría afectar la perturbación de la dinámica de juego a partir de manipulación del número
de jugadores en estas escalas temporales donde la exploración tiende a saturarse (Ric,
Hristovski y Torrents, 2015). Los resultados mostraron que provocar desequilibrios
numéricos de diferente magnitud en periodos de 15, 30, 45 y 60 segundos permite a los
jugadores ampliar su capacidad exploratoria, es decir, posibilita una mayor variedad de
comportamientos. Esto implica que si el objetivo del entrenador o entrenadora es
desarrollar más flexibilidad de comportamiento en sus equipos, la modificación constante
en breves periodos de tiempo de las condiciones iniciales de juego durante el desarrollo de
la tarea puede ser una estrategia interesante. Esta estrategia favorecerá la exploración de
soluciones en un entorno donde los desequilibrios cambian temporalmente, provocando
mayor o menor incertidumbre en función de la posición y variación del espacio efectivo de
juego (Vilar et al., 2013),.
Los resultados de los estudios II y III muestran como los mismos constreñimientos
pueden provocar diferentes tendencias a nivel individual y colectivo. Si aumentamos el
número de oponentes el comportamiento exploratorio de los jugadores a nivel individual
aumenta, sin embargo a nivel colectivo disminuye. Estos resultados pueden parecer
contradictorios pero muestran una compensación recíproca o sinergia multinivel. El
aumento de oponentes provoca configuraciones o patrones de comportamiento colectivos
mucho más estables, pero menos a nivel individual. El comportamiento jugando en
inferioridad numérica a nivel colectivo puede ser muy rígido, es decir, puede ser que haya
un jugador presionando, otro en cobertura, otro haciendo balance defensivo y un cuarto
replegando. Sin embargo, a nivel individual puede ser que cada jugador esté cambiando
constantemente de tarea, es decir, puede ser que el mismo jugador pase de estar haciendo
un balance defensivo a hacerle la cobertura a un compañero que ha pasado a presionar. Por
lo tanto, la superioridad del oponente puede implicar que el jugador a nivel individual deba
realizar múltiples tareas para favorecer la estabilidad a nivel colectivo. Estos hallazgos
hacen que se deba tener en cuenta el nivel del sistema (deportista o equipo) sobre el que se
quiere incidir, ya que la manipulación de la tarea puede provocar unos efectos a nivel
individual y no tenerlos o que sean contrarios a nivel colectivo, y viceversa.
97
Tal y como se ha comentado anteriormente, la dinámica exploratoria se define por
el parámetro de orden <qd(t)>, y concretamente por los parámetros alfa y qstat, cuya
combinación podría determinar el perfil de comportamiento de los jugadores a diferentes
escalas temporales. A modo de ejemplo, en la figura 2 se diferencian cuatro combinaciones
posibles entre estos dos parámetros: a) valor qstat alto y valor α bajo (línea azul), lo que
significa que el jugador explora una región reducida dentro del abanico de posibilidades de
comportamientos a realizar y desarrolla configuraciones similares (acciones, zonas,
contexto, distancias, velocidades, etc.) de forma más prolongada en el tiempo; b) valor α y
qstat altos (línea verde), esto significa que el jugador cambia rápidamente de configuración
de movimiento, pero no explora un gran conjunto de configuraciones diferentes; c) valor α
y qstat bajos (línea amarilla), esto se asocia con un cambio lento de comportamientos, pero,
a largo plazo, una exploración de un amplio conjunto de configuraciones de movimiento; y
d) valor qstat bajo y valor α alto (línea roja), este perfil se asocia con cambios rápidos en las
configuraciones de movimiento y con la exploración de un amplio abanico de
comportamientos a largo plazo.
98
Figura 2. Cuatro perfiles de la superposición dinámica <qd(t)> para distancias de tiempo
incrementales definidos por los parámetros alfa y qstat. Su dinámica evoluciona en escalas
temporales distintas (de segundos a varios minutos) y no converge a 0 (comportamiento aleatorio)
durante el periodo de observación.
99
7.4. Limitaciones y perspectivas de investigación futuras
A pesar de que este tipo de análisis podría tener transferencia inmediata a la realidad
de un club profesional, no resulta fácil obtener datos de posicionamiento para avanzar en el
análisis de rendimiento táctico. Actualmente, la última tecnología en sistemas de
seguimiento a través de video cámaras instaladas en la mayoría de estadios de los equipos
masculinos de las grandes ligas europeas permite conocer el posicionamiento de los
jugadores y del balón con una gran frecuencia de recogida, precisión y de forma
instantánea. Sin embargo, la tecnología GPS utilizada en esta tesis para recoger los datos de
posicionamiento impidió capturar el posicionamiento del balón durante las tareas
experimentales, y el posicionamiento del equipo rival durante el partido. El
posicionamiento del rival y del balón hubiera permitido comprender mejor la dinámica de
juego.
La localización del balón, así como el contexto que envuelve al que lo porta, tendrá
grandes implicaciones en el análisis del juego y obviamente será uno de los principales
condicionantes de la dinámica de comportamiento táctico. El hecho de tener que registrar
las acciones con balón y determinar el poseedor del balón de forma manual ralentizó la
preparación de los datos. Este hecho supone una limitación porque a pesar de que los datos
de posicionamiento de los jugadores se registraron con una frecuencia de 5 y 15 hercios
(Hz), tuvieron que ajustarse a 1 Hz para sincronizarlos con los datos obtenidos a través de
la observación sistemática y mantener unos valores de fiabilidad elevados. Si bien es cierto
que comprobamos que la similitud de configuraciones en una escala temporal inferior a 1
segundo mostró valores muy próximos al máximo (≈1), es decir, se mantuvo estacionaria
en una escala de décimas de segundo previa a la relajación de la curva (escala de
segundos). Esto implica que haberlo hecho con una frecuencia superior podría haber
alterado los resultados, ya que como se aprecia en la figura 3 la función de ajuste a la
distribución de los datos hubiera tenido que ser redefinida utilizado una función alternativa
(ej. stretched exponential function). Por ello, para futuros trabajos donde la frecuencia de
recogida sea mayor y pueda suponer una distribución diferente de los datos será necesario
utilizar funciones o modelos matemáticos distintos que se ajusten a dicha distribución.
100
Figura 3. Ejemplo de ajuste de la función de la curva teórica y datos experimentales.
Otro de los motivos por los que se utilizó una frecuencia de 1Hz es porque los
softwares estadísticos actuales no son capaces de procesar grandes matrices de datos. Por
ello, y para próximos trabajos que supongan el tratamiento de un mayor volumen de datos
será imprescindible desarrollar algoritmos que permitan el procesamiento de matrices
mayores y permitan disponer de los resultados con una mayor brevedad.
101
impuestos por el entrenador o la entrenadora todavía no ha estado sujeta a una investigación
sistemática, postulándose como una propuesta de estudio innovadora para el futuro.
Por otro lado, existe otro tipo de modelos matemáticos que podrían ser utilizados
para explorar los cambios discretos de patrones tácticos. Los modelos de cambio de
régimen (regime switching models), basándose en el enfoque de cadenas ocultas de Markov
(hidden Markov chain), permitirían estimar la probabilidad condicional de que se
produzcan cambios discretos entre diferentes regímenes (patrones) (Smith et al., 2017). Por
ejemplo, lateral y extremo intercambiando su posición en profundidad durante el juego, o la
alternancia en la presión de punta y media-punta en la salida de balón con defensas
centrales. Esto permitiría investigar la manera en que los comportamientos a nivel
individual responden a cambios cualitativos en un nivel superior del sistema. Es decir,
evaluar cómo los parámetros de control que evolucionan más lentamente (goles,
expulsiones, sustituciones, etc.) constriñen el comportamiento a nivel colectivo (valores de
parámetros de orden) que cambian más rápido, y cómo éstos constriñen las acciones
individuales de los jugadores o jugadoras que evolucionan más rápido, es decir, en una
escala temporal más corta.
El antes mencionado posicionamiento del balón podría ser considerado para estudiar
si el intercambio de pases entre deportistas sigue algún patrón en concreto. La anotación de
pases, aunque no durante el entrenamiento, ha permitido explorar las interacciones entre los
jugadores durante la competición mediante análisis de redes (Duch, Waitzman y Amaral,
2010; Gama et al., 2014; Gonçalves et al., 2017). Sin embargo, el conocimiento de la
longitud de esos pases permitiría explorar la posibilidad de que el desplazamiento del balón
pudiera seguir algún tipo de patrón que combine la alternancia de pases cortos con otros de
trayectorias más largas. Este tipo de patrones son parte de una estrategia general de
movimiento que adoptan muchas organismos o sistemas en una amplia variedad de
contextos, como por ejemplo, patrones identificados en moléculas de gas, patrones de caza
de tiburones o patrones de desplazamiento de personas en un parque de atracciones, y
reciben el nombre de vuelos de Levy. La distribución de distancias o longitud en los
desplazamientos (ej. pases) en forma de ley de potencia encierra en sí mismo una propiedad
102
fractal o de escala invariante, es decir, que estaríamos hablando de un patrón repetido a
diferentes escalas.
103
8. Conclusiones
Las conclusiones generales de esta tesis son:
104
La manipulación de la relación numérica en juegos reducidos tiene efectos
diferentes en la dinámica de comportamiento exploratorio del equipo e individual. A
nivel colectivo, los equipos que jugaron en superioridad mostraron una amplitud
exploratoria inferior, observándose mayores diferencias cuando aumentó el número
de compañeros. Sin embargo, a nivel individual, el aumento del número de
oponentes disminuyó la flexibilidad, variedad e incertidumbre de soluciones tácticas
individuales. En este sentido, se debe tener en cuenta que los mismos
constreñimientos pueden tener efectos diferentes sobre el comportamiento
exploratorio individual y colectivo. Esto también implica que existe algún tipo de
sinergia multinivel, donde el bajo grado de variedad táctica a nivel colectivo puede
ser compensado a nivel individual y viceversa.
105
Dependiendo de los objetivos del entrenador o de la entrenadora, las tareas de
entrenamiento mejorarán la exploración de soluciones tácticas o se centrarán en
estabilizar un conjunto más reducido de comportamientos. La manipulación
adecuada de los constreñimientos permite moldear una estructura y dinámica de
comportamiento táctico más o menos estable y/o flexible y más o menos regular y/o
variada. El análisis jerárquico de componentes principales y la superposición
dinámica o dynamic overlap <qd(t)> se postulan como parámetros de orden capaces
de caracterizar esta estructura dinámica.
106
9. Referencias
Abrantes, C. I., Nunes, M. I., Maçãs, V. M., Leite, N. M. y Sampaio, J. E. (2012). Effects
of the number of players and game type constraints on heart rate, rating of perceived
exertion, and technical actions of small-sided soccer games. Journal of Strength and
Conditioning Research, 26(4), 976–981. doi: 10.1519/JSC.0b013e31822dd398
Aguiar, M., Gonçalves, B., Botelho, G., Lemmink, K. y Sampaio, J. (2015). Footballers’
movement behaviour during 2-, 3-, 4- and 5-a-side small-sided games. Journal of
Sports Sciences, 33(12), 1259–1266. doi:10.1080/02640414.2015.1022571
Almeida, C. H., Duarte, R., Volossovitch, A. y Ferreira, A. P. (2016). Scoring mode and
age-related effects on youth soccer teams’ defensive performance during small-sided
games. Journal of Sports Sciences, 34(14), 1355-1362.
doi:10.1080/02640414.2016.1150602
Araújo, D., Passos, P., Esteves, P., Duarte, R., Lopes, J., Hristovski, R. y Davids, K.
(2015). The micro-macro link is understanding sport tactical behaviours: Integrating
information and action at different levels of system analysis in sport. Movement &
Sport Sciences, 2015(89), 1–32. doi:10.1051/sm/2015028
107
Bourbousson, J., Seve, C. y McGarry, T. (2010). Space–time coordination dynamics in
basketball: Part 2. The interaction between the two teams. Journal of Sports Sciences,
28(3), 349–358.
Davids, K., Button, C., Araújo, D., Renshaw, I. y Hristovski, R. (2006). Movement models
from sports provide representative task constraints for studying adaptive behaviour in
human movement studies. Adapt Behav, 14. doi:10.1177/105971230601400103
Domany, E., Hed, G., Palassini, M. y Young, A. P. (2001). State hierarchy induced by
correlated spin domains in short-range spin glasses. Phys. Rev. B, 64(22), 224406.
doi:10.1103/PhysRevB.64.224406
108
Duarte, R. (2012). Capturing Interpersonal coordination processes in association football:
from dyads to collectives. (Tesis doctoral, Universidade Técnica de Lisboa, Lisboa,
Portugal). Recuperado de: https://www.repository.utl.pt/handle/10400.5/4522
Duarte, R., Arajo, D., Correia, V. y Davids, K. (2012). Sports teams as superorganisms:
Implications of sociobiological models of behaviour for research and practice in team
sports performance analysis. Sports Medicine, 42(8), 633–642. doi:10.2165/11632450-
000000000-00000
Duarte, R., Araújo, D., Correia, V., Davids, K., Marques, P. y Richardson, M. J. (2013).
Competing together: Assessing the dynamics of team–team and player–team
synchrony in professional association football. Human Movement Science, 32(4), 555–
566. doi:10.1016/j.humov.2013.01.011
Duarte, R., Araújo, D., Davids, K., Travassos, B., Gazimba, V. y Sampaio, J. E. (2012).
Interpersonal coordination tendencies shape 1-vs-1 sub-phase performance outcomes
in youth soccer. Journal of Sports Sciences, 30(9), 871–877.
doi:10.1080/02640414.2012.675081
Duarte, R., Araújo, D., Folgado, H., Esteves, P., Marques, P. y Davids, K. (2012).
Capturing complex, non-linear team behaviours during competitive football
performance. Journal of Systems Science and Complexity, 26(1), 62–72.
doi:10.1007/s11424-013-2290-3
Duarte, R., Araújo, D., Freire, L., Folgado, H., Fernandes, O. y Davids, K. (2012). Intra-
and inter-group coordination patterns reveal collective behaviors of football players
near the scoring zone. Human Movement Science, 31(6), 1639–1651.
doi:10.1016/j.humov.2012.03.001
109
Fabrigar, L. R., Wegener, D. T., MacCallum, R. C. y Strahan, E. J. (1999). Evaluating the
use of exploratory factor analysis in psychological research. Psychological Methods,
4(3), 272–299. doi:10.1037/1082-989X.4.3.272
Folgado, H., Duarte, R., Fernandes, O. y Sampaio, J. (2014). Competing with lower level
opponents decreases intra-team movement synchronization and time-motion demands
during pre-season soccer matches. PloS One, 9(5), e97145.
doi:10.1371/journal.pone.0097145
Folgado, H., Duarte, R., Marques, P. y Sampaio, J. (2015). The effects of congested
fixtures period on tactical and physical performance in elite football. Journal of Sports
Sciences, 33(12), 1238–1247. doi:10.1080/02640414.2015.1022576
Gabin, B., Camerino, O., Anguera, M. T. y Castañer, M. (2012). Lince: Multiplatform sport
analysis software. Procedia -Social and Behavioral Sciences, 46, 4692–4694.
doi:10.1016/j.sbspro.2012.06.320
Gama, J., Passos, P., Davids, K., Relvas, H., Ribeiro, J., Vaz, V. y Dias, G. (2014).
Network analysis and intra-team activity in attacking phases of professional football.
International Journal of Performance Analysis in Sport, 14(3), 692–708.
Gibson, J. J. (1979). The ecological approach to visual perception. Boston, MA, US:
Houghton, Mifflin and Company.
110
Gonçalves, B., Coutinho, D., Santos, S., Lago-Penas, C., Jiménez, S. y Sampaio, J. (2017).
Exploring Team Passing Networks and Player Movement Dynamics in Youth
Association Football. PloS One, 12(1), e0171156.
Gonçalves, B., Marcelino, R., Torres-Ronda, L., Torrents, C. y Sampaio, J. (2016). Effects
of emphasising opposition and cooperation on collective movement behaviour during
football small-sided games. Journal of Sports Science, 34(14), 1346–1354.
doi:10.1080/02640414.2016.1143111
Gonçalves, B. V., Figueira, B. E., Maçãs, V. y Sampaio, J. (2014). Effect of player position
on movement behaviour, physical and physiological performances during an 11-a-side
football game. Journal of Sports Sciences, 32(2), 191–199.
doi:10.1080/02640414.2013.816761
Hristovski, R., Davids, K., Araújo, D., Passos, P., Torrents, C., Aceski, A. y Tufekcievski,
A. (2013). Creativity in sport and dance: Ecological dynamics on a hierarchically soft-
111
assembled perception-action landscape. En K. Davids, R. Hristovski, D. Araújo, N.
Balagué, C. Button y P. Passos (Eds.), Complex systems in sport (pp. 261–274).
London: Routledge.
Hristovski, R., Davids, K., Passos, P. y Araújo, D. (2012). Sport Performance as a Domain
of Creative Problem Solving for Self-Organizing Performer-Environment Systems.
The Open Sports Sciences Journal, 5(1), 26–35. doi:10.2174/1875399X01205010026
Joliffe, I. T. (2002). Principal component analysis (2nd editio). New York: Springer.
Kelly, D. M. y Drust, B. (2009). The effect of pitch dimensions on heart rate responses and
technical demands of small-sided soccer games in elite players. Journal of Science and
Medicine in Sport, 12(4), 475–479. doi:10.1016/j.jsams.2008.01.010
Liu, H., Gómez, M.-A., Gonçalves, B. y Sampaio, J. (2016). Technical performance and
match-to-match variation in elite football teams. Journal of Sports Sciences, 34(6),
509–18. doi:10.1080/02640414.2015.1117121
McGarry, T. (2006). Identifying patterns in squash contests using dynamical analysis and
human perception. International Journal of Performance Analysis in Sport, 6(2), 134–
147.
McGarry, T., Anderson, D. I., Wallace, S. A., Hughes, M. D. y Franks, I. M. (2002). Sport
competition as a dynamical self-organizing system. Journal of Sports Sciences,
112
20(10), 771–781. doi:10.1080/026404102320675620
Moura, F. A., van Emmerik, R. E. A., Santana, J. E., Martins, L. E. B., Barros, R. M. L. de
y Cunha, S. A. (2016). Coordination analysis of players’ distribution in football using
cross-correlation and vector coding techniques. Journal of Sports Sciences, 34(24),
2224-2232. doi:10.1080/02640414.2016.1173222
Palut, Y. y Zanone, P.-G. (2005). A dynamical analysis of tennis: Concepts and data.
Journal of Sports Sciences, 23(10), 1021–1032. doi:10.1080/02640410400021682
Parisi, G. (1983). Order Parameter for Spin-Glasses. Phys. Rev. Lett., 50(24), 1946–1948.
doi:10.1103/PhysRevLett.50.1946
Passos, P., Araújo, D., Davids, K., Gouveia, L., Milho, J. y Serpa, S. (2008). Information-
governing dynamics of attacker-defender interactions in youth Rugby Union. Journal
of Sports Science, 26(13), 1421-1429. doi:10.1080/02640410802208986
Ric, A., Hristovski, R. y Torrents, C. (2015). Can joker players favor the exploratoy
behavior in football small-sided games? Research in Physical Education , Sport and
Health, 4(2), 35–39.
113
synergies. Frontiers in Psychology, 2, 1–7. doi:10.3389/fpsyg.2011.00038
Rosso, B. D. (2014). Creativity and Constraints: Exploring the Role of Constraints in the
Creative Processes of Research and Development Teams. Organization Studies, 35(4),
551–585. doi:10.1177/0170840613517600
Sibani, P. y Dall, J. (2003). Log-Poisson statistics and full aging in glassy systems. EPL
(Europhysics Letters), 64(1), 8. doi:10.1209/epl/ i2003-00109-0
Silva, P., Chung, D., Carvalho, T., Cardoso, T., Davids, K., Araújo, D. y Garganta, J.
(2016). Practice effects on intra-team synergies in football teams. Human Movement
Science, 46, 39–51. doi:10.1016/j.humov.2015.11.017
Silva, P., Duarte, R., Sampaio, J., Aguiar, P., Davids, K., Araújo, D. y Garganta, J. (2014).
Field dimension and skill level constrain team tactical behaviours in small-sided and
conditioned games in football. Journal of Sports Sciences, 32(20), 1888-1896.
doi:10.1080/02640414.2014.961950
Silva, P., Esteves, P., Correia, V., Davids, K., Araújo, D. y Garganta, J. (2015). Effects of
manipulations of player numbers vs. field dimensions on inter-individual coordination
during small-sided games in youth football. International Journal of Performance
Analysis in Sport, 15(2), 641-659.
Silva, P., Garganta, J., Araújo, D., Davids, K. y Aguiar, P. (2013). Shared knowledge or
shared affordances? insights from an ecological dynamics approach to team
coordination in sports. Sports Medicine, 43(9), 765–772. doi:10.1007/s40279-013-
0070-9
114
Silva, P., Travassos, B., Vilar, L., Aguiar, P., Davids, K., Araújo, D. y Garganta, J. (2014).
Numerical relations and skill level constrain co-adaptive behaviors of agents in sports
teams. PloS One, 9(9), e107112. doi:10.1371/journal.pone.0107112
Silva, P., Vilar, L., Davids, K., Araújo, D. y Garganta, J. (2016). Sports teams as complex
adaptive systems: manipulating player numbers shapes behaviours during football
small-sided games. SpringerPlus, 5(1), 191. doi:10.1186/s40064-016-1813-5
Smith, D., Ric, A., Goncalves, B., Torrents, C., Sampaio, J. y Hristovski, R. (2017).
Exploring coordination in soccer players’ movement using reginme switching state
space models. En Vth World Congress on Science & Football, Rennes (France), 31st
May – 2nd June, 2017.
Stergiou, N., Buzzi, U., Kurz, M. y Heidel, J. (2004). Nonlinear Tools in Human
Movement. In N. Stergiou (Ed.), Innovative analyses of human movement (pp. 63–90).
Champaign, IL, Human Kinetics.
Torrents, C., Ric, A. y Hristovski, R. (2015). Creativity and emergence of specific dance
movements using instructional constraints. Psychology of Aesthetics, Creativity, and
the Arts, 9(1), 65–74. doi:10.1037/a0038706
Travassos, B., Araújo, D., Vilar, L. y McGarry, T. (2011). Interpersonal coordination and
ball dynamics in futsal (indoor football). Human Movement Science, 30, 1245–1259.
doi:10.1016/j.humov.2011.04.003
Travassos, B., Gonçalves, B., Marcelino, R., Monteiro, R. y Sampaio, J. (2014). How
115
perceiving additional targets modifies teams’ tactical behavior during football small-
sided games. Human Movement Science, 38, 241–50.
doi:10.1016/j.humov.2014.10.005
Travassos, B., Vilar, L., Araújo, D. y McGarry, T. (2014). Tactical performance changes
with equal vs unequal numbers of players in small-sided football games. International
Journal of Performance Analysis in Sport, 14(2), 594–605.
Vázquez, P. (2017). Unified framework for the study of sport-related behavior. Universidad
de Barcelona, Barcelona.
Vilar, L., Araújo, D., Davids, K. y Bar-Yam, Y. (2013). Science of winning soccer:
Emergent pattern-forming dynamics in association football. Journal of Systems
Science and Complexity, 26(1), 73–84. doi:10.1007/s11424-013-2286-z
Vilar, L., Araújo, D., Davids, K., Correia, V. y Esteves, P. T. (2013). Spatial-temporal
constraints on decision-making during shooting performance in the team sport of
futsal. Journal of Sports Sciences, 31(8), 840–846.
doi:10.1080/02640414.2012.753155
Vilar, L., Araújo, D., Davids, K., Travassos, B., Duarte, R. y Parreira, J. (2014).
Interpersonal coordination tendencies supporting the creation/prevention of goal
scoring opportunities in futsal. European Journal of Sport Science, 14(1), 28–35.
doi:10.1080/17461391.2012.725103
Vilar, L., Duarte, R., Silva, P., Chow, J. Y. y Davids, K. (2014). The influence of pitch
dimensions on performance during small-sided and conditioned soccer games. Journal
of Sports Sciences, 32(19), 1751–1759. doi:10.1080/02640414.2014.918640
116
Motor Organizations and Development in Infancy and Early Childhood Chateu de
Rosey, France (pp. 23–37). Dordrecht: Springer Netherlands. doi:10.1007/978-94-
009-2071-2_2
Withagen, R., Poel, H. J. d., Araújo, D. y Pepping, G.-. J. (2012). Affordances can invite
behaviour: reconsidering the relationship between affordances and agency. New Ideas
Psychol, 30(2), 250–258. doi:10.1016/j.newideapsych.2011.12.003
117