Escolar Documentos
Profissional Documentos
Cultura Documentos
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/236069645
CITATIONS READS
26 553
2 authors:
Some of the authors of this publication are also working on these related projects:
PhD thesis: micro-level processes of identity development and career decisions View project
All content following this page was uploaded by Paul van geert on 15 April 2014.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Nonlinear Dynamics, Psychology, and Life Sciences, Vol. 17, No. 2, pp. 233-267.
2013 Society for Chaos Theory in Psychology & Life Sciences
1
Correspondence address: H.W. Steenbeek, Heijmans Institute, University of Groningen;
Grote Kruisstraat 2/1, 9712 TS Groningen, Netherlands. E-mail: h.w.steenbeek@rug.nl
233
234 NDPLS, 17(2), Steenbeek & van Geert
Kratochwill, 2000). However, the view that underlies these approaches is one
that characterizes individual students in terms of true scores on underlying latent
variables, the variance of which is to a specifiable extent explained by the
contribution of educational methods and practices that can be treated as
independent variables.
In this article we shall focus on learning-teaching trajectories
successful as well as unsuccessful ones as emergent and dynamic
phenomena resulting from the interactions in the entire educational context, in
particular the interaction between students and teachers. Although the terms
successful and unsuccessful are almost intrinsically vague, they feature
prominently in the thoughts of teachers, parents and educational policy makers.
A successful learning trajectory can be defined as a learning trajectory in
which the learner shows the progress in acquiring new knowledge and skills that
is expected on the basis of the learner's personal capabilities and the quality of
the education provided, based on proper engagement and activity in the learner
himself (see also Clements & Sarama, 2004; Crosnoe et al., 2010; Sarama &
Clements, 2009). In contrast, a learner with an unsuccessful learning trajec-
tory lags behind the progress in acquiring new knowledge and skills that the
learner is supposed to make given the learner's capabilities and education. The
classical approach to explaining the differences between successful and
unsuccessful learning trajectories, or the differences between more or less
successful ones, is by treating them as problems that relate to educational
effectiveness on the one hand and on the other hand by estimating the contribu-
tion of general determinants such as the students social economic status or the
influence of the curriculum (Rowe, 2003). However, in this article we shall use a
complex dynamic systems (CDS) approach to describe educational trajectories
as emergent properties of co-regulated real-time processes. The real-time
processes are those of activities taking place in students and teachers, in the con-
text of particular school classes, curricula and educational materials. We shall
review evidence showing that self-regulation (by the student) of these processes
is constantly interacting with other-regulation (regulation of the process by other
persons, in particular teacher), thus resulting in a process of co-regulation. We
shall argue that the short and long-term properties of these processes can be des-
cribed and at least qualitatively predicted by applying specific dynamic systems
model.
The outline of this article is as follows: First, current insights into the
co-and self-regulation of learning processes are briefly discussed. Section two
provides an overview of central features of a CDS approach to teaching-learning
processes. Section three concentrates on a conceptual model for the short-term
and long-term dynamics of learning, and discusses two examples of dynamic
models, one on the level of real time teaching learning activities, one on the
level of long-term changes in learning-teaching trajectories. Section four con-
cludes with a discussion about the added value of this approach for theory build-
ing and educational practice.
NDPLS, 17(2), Learning-Teaching Trajectories 235
LEARNING-TEACHING AS A CO-REGULATED PROCESS
A literature survey by Nolen (2009) showed that the five main themes
of current educational research were classroom achievement (36%), learning and
memory (33%), affects/motivation/beliefs (31%), cognition/reasoning (21%)
and teaching (21%). The strongest link between themes was the one between
classroom achievement and affects/motivation/beliefs, i.e. the question why stu-
dents do or don't do certain things that should lead to higher achievement.
Various lines of research and theory formation have attempted to
answer this question. One line emphasizes the importance of self-regulated
learning for children to become successful learners (Boekaerts & Corno,
2005; Miller & Brickman, 2004; Rothbart & Jones, 1998; Schunk &
Zimmerman, 2000, 2004; Zimmerman, 2001, 2008). Self-regulated learning
refers to a persons thoughts, feelings, and actions that affect the persons learn-
ing of knowledge and skills. A central aspect is that the learner is recognized as
an autonomous, intentional, interacting agent. This view features prominently in
Zimmerman and Schunks four-level model of development of self-regulation
skills (2004, 2008). However, despite this models explanation of the distinct
phases in self-regulation skills, it does not explain how the child moves from
one phase to the next one. Like their students, teachers are autonomous inten-
tional agents, that is, they are self-regulating beings (Butler, Lauscher, Jarvis-
Selinger, & Beckingham, 2004; Wehmeyer, Yeager, Bolding, Agran & Hughes,
2003), regulating their actions with those of their students, among others. They
are driven by their own motives, emotions and beliefs of what makes them effi-
cient teachers (Hoy, 2008).
This particular concept of self-regulation is strongly associated with
metacognition, deliberate choices, decisions and beliefs about achievement
(Dinsmore, Alexander, & Loughlin, 2008). The concept thus features a relative-
ly conscious and rational agent who is master of his own actions and feelings,
with the self as central executive (Loyens, Magda, & Rikers, 2008).
A related dominant line of research is that of the goal theory of
achievement and goal orientation theory (e.g. Papaioannou, Simou, Kosmidou,
Milosis, & Tsigilis, 2009; Roussel, Elliot, & Feltman, 2011). In an overview of
the literature on goal orientation, Kaplan and Maehr (2006) state that goal
orientations are overarching purposes of achievement behavior that are different
from the actual contents of what people try to achieve. In the educational con-
text, main goal orientations are those of mastery and performance. Whereas
mastery goal orientations refer to an individuals purpose of developing compe-
tence, performance goal orientations refer to the purpose of demonstrating
competence, i.e. demonstrating a high performance level. Other main goal orien-
tations are those to achieve extrinsic incentives, social goal orientations that
refer to social or interpersonal reasons for engaging in achievement behavior,
and finally avoidance orientations, which relate to work avoidance and academic
alienation. A number of authors associate the major goal orientations with a
basic behavioral contrast, namely that of approach versus avoidance (Elliot &
236 NDPLS, 17(2), Steenbeek & van Geert
Covington, 2001). According to Elliott and Thrash (2001) achievement goals
function as a network or integrated pattern of variables that together create an
orientation toward achievement tasks. By relating achievement goals to net-
works of variables many of which are strongly emotionally laden, it is also pos-
sible to explain apparently counterproductive goals and actions, known as
academic self-handicapping (the creation of impediments to successful perfor-
mance on tasks that the individual considers important; Urdan and Midgley,
2001). In short, major goal orientations are to a considerable extent driven by
motives that are emotional and partly or greatly outside the deliberate control of
the rational agent (Dai & Sternberg, 2004; Immordino-Yang & Damasio, 2008;
Meyer & Turner, 2006).
In addition, the literature pays increasing attention to the transactional
nature and the social embeddedness of learning (Cowie & van der Aalsvoort,
2000; DePaepe, de Corte, & Verschaffel, 2006; Fogel, 2009; Kumpulainen,
Hmelo-Silver, & Cesar, 2009; Murphy, 2007; Sorsana, 2008; Vosniadou, 2007).
The picture that begins to emerge from this literature is that of learning
as a complex intertwined system of components. These components are of very
different kinds: internal and external, emotional and cognitive, psychological
and physiological. They interact with one another and out of this interaction
macroscopic properties emerge, such as particular goal orientations, evaluations,
emotions, actions and achievements. The conscious and deliberate aspect of the
person, student or teacher, is one of the many components that contribute to this
dynamics and that on its turn is determined by this dynamics.
The CDS approach can serve as an overarching theoretical framework,
and can help us specify three important aspects of a comprehensive theory of
teaching and learning that are apparent in the current literature but that are still
underexposed.
First, to fully understand learning processes, insight is needed in the
temporal unfolding of learning processes in individual children, and in the
mechanisms that play a role in these processes of change (Howe & Lewis, 2005;
Lichtwarck-Asschof, van Geert, Bosma, & Kunnen, 2008; Steenbeek, Jansen &
van Geert, 2012; van Geert & Steenbeek, 2005, 2006). However, much research
reports only on averaged findings over groups of children, i.e., yielding group
data while leaving aside information about the way the individuals learning
processes unfold over time. Hamaker, Dolan, & Molenaar state that: (the)
dominant statistical approach in psychological research, comprising
interindividual techniques, is not necessarily appropriate if one wants to study
psychological mechanisms taking place within individuals (2005, p. 227; see
also Molenaar, 2004; Molenaar & Campbell, 2009 for a more thorough discus-
sion of the so called ergodicity problem). Therefore, models of the mechan-
isms of change as they apply to individual trajectories need to be used, i.e. at the
level where these mechanisms are supposed to operate1.
Second, a comprehensive theory of learning must be able to explain the
cyclical causal relationship between short-term learning processes of individual
children (what happens in the classroom on a particular day, during a particular
NDPLS, 17(2), Learning-Teaching Trajectories 237
lesson in real time) and emerging long term developmental learning trajectories
of those children (their academic progress, i.e., their developing insights over
years; see Grossen (2009) on the linking-of-time-scales). The theory must thus
be able to describe and explain how short-term events affect long-term changes,
and how long-term changes affect short term events.
Third, most research recognizes the importance of studying childrens
learning in context (van Geert, 2009). However, in existing explanatory models,
context is usually operationalized as an independent and distal variable, e.g. as
in studying the statistical association between the home environment and
academic performance of children (van Geert & Steenbeek, 2005). However,
context and person influence each other, there is a reciprocal determinism
between the two (Bandura, 1977). Explanatory models should incorporate con-
texts as proximal, dynamic factors that on the one hand result from the interac-
tion processes themselves, and on the other hand influence these interaction
processes (which is yet another illustration of cyclical causality). Recent insights
in dynamic systems suggest that the interaction between a person and a context
is actually a specific case of interactions across a network of multiple compo-
nents (for general accounts, see Newman, 2003, 2010; for applications in the
behavioral and developmental sciences, Borsboom, Cramer, Schmittmann,
Epskamp, & Waldorp, 2011; Cramer, Waldorp, van der Maas & Borsboom,
2010; van der Maas et al., 2006; van Geert, 1991, 1994, 2008). CDS provides
the appropriate tool for modeling, and by doing so understanding, processes that
emerge out of the reciprocal interactions between multiple components. An
example of such type of processes are the relations between goals, actions,
achievements, emotions, teaching practices of interacting agents, i.e., of teachers
and students.
In summary, the CDS approach should enable us to formulate
individual-based models of reciprocal causality among the multiple components
that constitute a learning and teaching process unfolding in real time, and to
understand how such models generate long-term process trajectories.
This point of view is also endorsed by a small but growing number of
authors which are currently arguing for a complexity movement in the
educational sciences (Drnyei, 2009; Goldspink, 2007; Goldspink & Kay, 2004;
Jrg, 2011; Jrg, Davis, & Goemans, 2007; Mowat & Davis, 2010; Namukasa,
2006; Nelson, 2004; Radford, 2006, 2007; Semetsky, 2005).
CDS THEORY APPLIED TO EDUCATIONAL PROCESSES
Components and Main Properties of Complex Dynamic Systems
A complex system consists of many interacting components or
elements, the aggregate behavior of which is typically more or different than the
sum of its parts (e.g. van Geert, 2008). A dynamic system is a way of describ-
ing how one state of a developmental system changes into another state
(Weisstein, 1999). That is, the properties of the components change over time,
as a consequence of the interactions between those components. The properties
238 NDPLS,
N 17(2),, Steenbeek & van Geert
changee in the state sp pace, which is the whole off possible statees distinguishabble
in the system;
s describbed by the set of dimensionss or variables nneeded to specify
the sysstem (or, in th his case, what specifies the ssystem as beinng an educationnal
system
m; van Geert, 20 008, p. 1873). The state spacce can be as siimple or as com m-
plex ass need be. Figu ure 1 shows an example of a ssimple state space, representiing
the inteerplay between n the amount ofo help given bby the teacher,, and the numbber
of arithhmetic problem ms solved by the student, oover subsequennt lessons. Ovver
time, itt stabilizes in an
a attractor statte.
Fig. 3. The agent model of student and teacher; a an agent has m multiple concern ns,
which govern the ap ppraisal of the current situattion (with as m
main compone ent,
ones own
o and the otther persons current
c actionss). Appraisals llead to particu
ular
actions
s; the effect of the
t actions are fed back into tthe next appraisal.
As Fig. 3 sh
hows, this iteraative, mutuallyy influencing pprocess that takkes
place over
o the course of an individu
ual instruction session, can bee conceived off as
the co
ore businesses of the chang ge process oveer the short-terrm time scale or
246 NDPLS, 17(2), Steenbeek & van Geert
micro-time. The mutual influences can give rise to smoothly occurring
interactions, which are associated with optimal functioning and optimal learning
(Nakamura & Csikszentmihalyi, 2002), i.e., the emergence of successful
learning trajectories. On the other hand, mutual interactions can also take the
form of non-smooth, erratic processes, associated with ineffective functioning
and learning, thus leading to attractor states of sub-optimal interaction sustaining
sub-optimal learning (and vice versa), and over the long term leading to
unsuccessful learning trajectories. Note that this iterative, mutually influenc-
ing process cannot be explained by using classical, individual-based self-regula-
tion models, such as the four-level model of Schunk and Zimmerman (2004). In
order to understand such processes they must be modeled in the form of itera-
tive, coupled dynamic systems. In this way, the short-term building blocks of
emerging successful versus unsuccessful learning trajectories can be made
explicit by modeling the (in-)balance of the student and teacher concerns for
Autonomy, for Relatedness, and for Competence, and what this means for their
behavior, appraisal and emotions during real time teaching-learning situations in
the classroom.
An Example of a Simulation of a Short-Term Learning-Teaching Coupling
In our previous work (Steenbeek & van Geert, 2005, 2007, 2008), we
have shown that this conceptual network can be used to build a general dynamic
systems model that generates temporal sequences of actions and interactions.
The model describes two agents, each of which is specified in the form of a
series of coupled equations. The equations refer to coupled changes in the
strength of a concern variable, a concern realization variable and its associated
emotional appraisal, an emotional expression variable, and a behavioral variable.
All variables are reduced to the simplest possible forms. The concern variable
for instance is described as a one-dimensional variable specifying a person's
dynamically varying between other-directed and self-directed interest (in a parti-
cular situation). The behavioral variable is reduced to either an other-directed or
a self-directed action. Agents perceive their own and the others actions and
emotional expressions and these perceptions are coupled to their internal states
and determine their next activity and emotional expression (for the technical
details of this dynamic model we refer to Steenbeek, 2006, Steenbeek & van
Geert, 2005, 2007, 2008). The model generates co-regulated sequences of
actions (as either other- or self-directed), of emotional expressions and of
internal appraisal states of the two agents.
We have applied this general model to dyadic interaction in children
and have shown that it predicts empirically observed properties of action pat-
terns in time. For instance, the model correctly predicted distributions of interac-
tion properties in dyadic play among children of different sociometric status
(Steenbeek, 2006; Steenbeek & van Geert, 2005, 2007, 2008). The model also
correctly predicted the occurrence of heavy-tailed distributions of the duration of
actual contact episodes during a play session and characteristic differences
between boys and girls (Steenbeek, van der Aalsvoort & van Geert, 2013).
NDPL
LS, 17(2), Lea
arning-Teachin
ng Trajectoriees 2247
Fig. 5. Example of an
a unsuccessfful learning tra
ajectory (or: an
n underachieviing
pattern) characterize
ed by a consttantly decreas ing performan nce level (outp
put
simulattion model).
DIISCUSSION
Potential Streng
gths of the CD S Approach
he CDS appro
In using th oach, includingg the use of conceptual aand
mic modelling of teaching-leearning processses, we aimed to
computtational dynam
NDPLS, 17(2), Learning-Teaching Trajectories 255
demonstrate that approaching learning as a complex dynamic process with
specifiable model properties can yield fruitful insights. For instance, we
discussed the importance of moment-to-moment iterations in building up
students learning experiences, e.g. in interaction with the teacher during
instruction sessions in arithmetic lessons. The dynamic agent model for instance,
suggested that students learning concerns (the balance in autonomy-relatedness,
and competence concerns) were not fixed, but that their strengths changed
during and as a consequence of the ongoing interaction. In addition, it high-
lighted the dynamic interplay of both students and teachers concerns and emo-
tional appraisals in forming short-term positive or negative learning experiences,
e.g. during a particular arithmetic lesson. It was shown how not only the
students, but also the teachers level of balance in autonomy-relatedness con-
cern, and a sufficient level of competence concern dynamically constituted
teaching-learning processes.
In addition, the long-term model showed that there exist dynamic
relationships between the students and teachers concerns, emotional appraisals,
and the students performance level. These dynamic relationships can be
supportive or competitive, competing, and symmetrical or asymmetrical. We
demonstrated that particular patterns of relationships act as control parameters
producing emergent learning-teaching trajectories, such as unsuccessful or
underachievement trajectories.
Because the conceptual models of temporal trajectories, for instance of
an arithmetic lesson or of the development of arithmetic performance during the
school year, involve many, often mutual and nonlinear interactions between
many variables in the child or children and the teacher, it is not possible to infer
the trajectories predicted by such models on the basis of simple conceptual and
linear extrapolation from a verbal model. Dynamic models, which provide
formal descriptions of these conceptual models, make it possible to infer such
trajectories in a computationally explicit and controllable way. They can also
help researchers in making their conceptual models more complete and explicit,
because the simulations will not generate the expected temporal patterns if some
essential component is missing from the dynamic model.
The surplus value for educational practice is that a better understanding
of the dynamics of teaching-learning processes provides possibilities for better
advising about and coaching of optimal learning trajectories. The help given to
teachers with the aim of changing their actions during instruction sessions where
necessary or wanted will greatly profit from a better understanding of the tem-
poral stream of those actions. This understanding will help teachers to focus on
their actions as a teacher, and how these actions influence the student in his
learning, e.g. of arithmetic, in the here-and-now of arithmetic lessons (Steenbeek
et al., 2012). In addition, we hope to have shown that in order to intervene in
classroom situations, one has to start from understanding the concerns of both
teacher and students and the active ways teacher and students try to realize these
concerns, that is of the dynamical way in which student and teacher activities are
co-regulated processes. For instance, when the teacher uses more or less contin-
256 NDPLS, 17(2), Steenbeek & van Geert
uous ratification to teach desirable behaviour patterns a practice we often
noticed in our observations of learning teaching interactions in special education
it is necessary to understand eventual nonlinear effects of ratification as a con-
sequence of the dynamics of basic concerns of teacher and student, instead of
treating ratification practice as an independent variable with a particular effect
size.
In summary, although various papers have suggested that our
understanding of teaching-learning and developmental processes can be
furthered by applying ideas from complex dynamic systems (see discussion on
the complexity movement in education), only very few have presented actual
dynamic models that go beyond verbal descriptions. In the present paper, we
have illustrated the possibilities of a short-term dynamic agent model on the one
hand and a long-term dynamic network model on the other hand. By doing so,
we hope to have demonstrated that it is at least feasible to construct such
models, that such models can be made to incorporate fundamental theory
regarding human concerns and teaching and learning, and that such models are
able to generate temporal patterns that show important qualitative similarities
with what we empirically know about the simulated phenomena.
Limitations and Future Steps
Further work needs to be done on completing the empirical picture of
possible educational trajectories that can emerge in individual students and their
teachers and educational contexts. If such models can be validated with empi-
rical datasets (e.g. Steenbeek et al., 2012), more complete pictures of possible
learning trajectories, and of how the social dynamics in the classroom contri-
butes to the long-term development of those trajectories, can be built.
Secondly, more work needs to be done in building dynamic simulation
models of teaching-learning processes, based on the general theory of action or
agent behavior on interacting time scales, and a general theory of mechanisms of
change, as described in this article. As mentioned before, the dynamic
simulation model is still work in progress (Section 2.3). In addition, it is
important to note that hypotheses derived from the model must still show their
validity in comparison with the empirical data. However, based on positive
results with validating our interaction model during play (Steenbeek, van der
Aalsvoort & van Geert, 2013; Steenbeek & van Geert, 2005, 2007, 2008), we
expect to show the validity of this model as it will be applied to concrete
instruction sessions. By iterating model building and empirical research based
on principles of complex dynamic systems, we might hope to find a way out of
limited linear explanations of the problems education is faced with in
contemporary society.
APPENDIX
Short-Term Model
The model has been written in Visual Basic for applications and runs
under Excel. The Excel file can be downloaded from
NDPLS, 17(2), Learning-Teaching Trajectories 257
http://www.paulvangeert.nl/articles_appendices.htm; see the section complex
dynamic systems approach of learning-teaching trajectories. A WORD file with
the annotated visual basic code can also be downloaded from the same webpage.
The Excel file contains a short explanation of how the parameter values can be
determined and the model can be run.
The model generates a time series of two behaviors in two agents,
namely self-directed behavior (e.g. the child is working alone on his assignment)
and other-directed behavior (e.g. the child is trying to attract the teacher's
attention). If both agents, namely the child and the teacher are simultaneously
showing other directed-behavior, they are interacting (e.g. they are working
together on the child's assignment). The model generates a parallel series of
emotional expressions, which corresponds with the codes neutral, negative, or
positive, in addition to a numerical value of intensity, from -4 to +4.
The parameters concern the (one) preference for involvement versus
autonomy (doing something together versus doing something alone), the contri-
bution of the behaviors to serialize agent of the concerns, the relation between
emotional appraisal (the continuous evaluation of the extent to which the con-
cerns are realized), the contribution of emotional expression to the value of the
basic concerns involvement and autonomy, the strength of the non-intentional
parameters behavioral symmetry versus behavioral continuity, the duration of
the activity in seconds, the level of randomization, and finally the memory span
(the temporal duration for which the concern realization is monitored by the
agent). The parameters can be determined qualitatively in a graphical mode, but
can also be numerically adjusted in the parameter sheet.
Steps in the Calculation of the Agent Model
1. specify all parameter values
2. initialize the variables for the calculation of the model
3. run main routine (420 times, corresponding with 7 min. of interaction):
a. Step 1: calculate the preferred behavioral state (preferred proportion
of Involvement over Autonomy, I.E. Acting Together Versus
Acting Alone) on the basis of the current values of the concerns
b. Step 2: determine the value of the behavioral weights, i.e. the
weights associated with the behaviors Acting Alone and Acting
Together respectively, on the basis of the current value of the
corresponding drives
c. Step 3: add the influence of the non-intentional parameters of
behavior (Symmetry and Continuity) to the behavioral weights
d. Step 4: calculate the value of the random factor for both types of
behavior, and add the random factor to the behavioral weights (the
random factor is based on a model of a constrained random walk)
e. Step 5: select the behavior Acting Alone or Acting Together, on the
basis of the highest value of the behavioral weights
f. Step 6: determine the level of the realized behavioral state (i.e.
concern realization) following the selected behavior, by means of
258 NDPLS, 17(2), Steenbeek & van Geert
the following substeps:
i. update the memory to cover the new level of concern
realization
ii. determine the influence of the selected behavior on the
realization of the concerns depending on the memory
iii. calculate the level of concern realization on the basis of the
memory
g. Step 7: calculate the strength of the drive on the basis of the
difference between the preferred behavioral state (see step 1) and
realized behavioral state
h. Step 8: calculate the emotion level on the basis of the drive
i. Step 9: calculate the emotional expression (positive, neutral,
negative) on the basis of the emotion level
j. Step 10: calculate the influence of the emotional expression on the
real-time value of the concerns (i.e. preferred behavioral states)
k. Step 11: return to Step 1