Table of Contents
Structure as Behavior:
Exploring Elements of the System Dynamics Modeling Process*
Ignacio J. Martinez-Moyano'
University at Albany, State University of New Y ork
First Draft: February 5, 2003
Final Draft: May 15, 2003
Abstract: This paper explores the implications of considering the structure of the system as changing
over time. Following the tradition of system dynamics and deeply believing that “the structure conditions
the behavior,” this paper makes the case for the analysis of the dynamics of the structure and its
implications. In addition, the questions of what structure is and what behavior is, in system dynamics, are
explored.
Keywords: Structure, Behavior, Supra-Structure, and System Dynamics Methodology
Base Motivation...
Structural Evolution...
Influencing the Structure.
Conclusion and Future Research
Appendix 1— Equations for Simple Model
*] thank Rogelio Oliva, Mohammad Mojtahedzadeh, Aldo Zagonel, Luis Luna, Vedat Dikert, and my colleagues at
the University at Albany for comments and assistance. All opinions and, especially, all errors are mine.
1 Comments welcome at int! 17797 @albany.edu or (518) 442-5257.
Page 1
Base Motivation
The motivation to write this paper was twofold. First, I felt that my understanding of how the
structure conditions the behavior of a system was weak and that I needed to strengthen it By doing so, I
started to realize that many of the mechanisms that we use in system dynamics to understand how the
structure conditions the behavior are not suited to think about the dynamic nature of the structure itself.
Second, I realized that a deeper understanding of the structure of a model did not, necessarily, informed.
me about the structure and nature of the problem in the system under study. Thus, I began to question
how! could leam about problematic situations in complex systems in such a way that I could gain better
understandings of them. Not only about the deepness of the structure of a model, but also about
altemative ‘more realistic’ models to represent the reality that I wanted to understand in order to be able
to intervene and modify a certain problematic occurrence.
Following the tradition of system dynamics, and deeply believing that “the behavior of a system.
arises from its structure,” (Sterman, 2000, p. 107) this paper makes a case for the analysis of the dynamics
of the structure. In addition, the paper is geared towards understanding how to influence conceptual jumps
froma ‘model 1,’ representing reality, to a ‘model 2’ being a better representation of that same reality.
In this paper I develop two major themes. First I explore the rationale of model’s change and system! s
change. Next, I briefly describe what system dynamics is and some of its assumptions to clarify the
specific area under study and to be able to put forward some questions. Then, I explore the concepts of
structure and behavior in system dynamics as well as the implications that arise when considering the
dynamics of the structure. As a final point, the paper ends with a concluding section and ideas about
possible future research in this area.
Page 2
Basic Rationale”
Let us suppose that we can represent the structure of the real world and the structure of the model of
the real world with two circles “M” and “R”. Let us assume that these representations are accurate and
complete. If these two representations adequately represent the structures of both the model and the real
Tf we create a model that replicates completely the structure of reality and superimpose the two circles
‘we would see only one because M =R. Therefore, any type of understanding that we could draw from the
model we could say applies to the real world. However, a model that has the exact same detail as reality is
as complex as reality is, therefore is very difficult to use efficiently (Sterman, 2000, p. 89). Furthermore,
we know that there is no model that can replicate the real world exactly, because by definition, all models
are simplifications of reality, and all models are wrong (Stennan, 2002, p. 525).
When using the system dynamics method, part of the process of generating understanding about the
model, the problem, and the system, is to meticulously test and explore the model to gain as much
understanding as possible (Richardson and Pugh, 1981; Sterman, 2000, Ch 3). The mode] is key to the
process and must be subject to many tests, trials, and experiments in order to gain understanding and
confidence on the understanding. By focusing on a model, we generate a tacit implication that the
structure under study is fixed. Sterman (2002), speaks to the need to focus on the modeling process
instead of on the model by saying that “focusing on the process of modeling rather than on the results of
any particular model speeds leaming and leads to better models.” (p. 521) Additionally, it has been
proposed that the boundary of the system—in tems of time horizon—helps identify the relevant structure
under study, specifically, the structure that does not change over the time period of enquiry. The
underlying assumption used is that the modeler can identify the structure that does not change during the
time frame of the modeling process. Unfortunately, perception of the modeler and access to information
to evaluate if the structure is not changing is sometimes limited to the modeler. Furthennore, defining the
? T would like to acknowledge the invaluable help that I received from Mohammad Mojtahedzadeh in shaping and
Clarifying the rationale presented here.
Page 3
system boundary and the degree of aggregation of the model has been identified as two of the most
difficult’ steps in successful modeling practice (Starman, 2000, p. 100).
Tt seams that, even highly skilled modelers are subject to mistakes in the crucial definition of model
boundary. Due to the fact that the models we build are static and have a fixed structure, we might think
that the real world we are studying behaves in the same way. However, it might not be the case in every
Situation.
As said before, it has been argued that it is crucial to focus on the process of modeling instead of on
the modd itself (Forester, 1985). In this paper, following and expanding on Forresters’s (1985)
argument, I argue that modelers, if they can focus on the processes instead of on the model AND if they
can acknowledge the dynamic nature of the structure of the real system, will be able to expand the
insights generated by system dynamics interventions.
Static Real World Structure Dynamic Real World Structure
(A) (B)
Figure 1— Static versus Dynamic Real World Structure
We have come to know that the iterative nature of the modeling process (Sterman, 2000, pp. 87-88)
creates the possibility for different models to arise as products of the process. Figure 1 (A) depicts the
possibility of having two models (M1 and M2) that are related to the real world structure (R). Model 2 in
this figure; represents a better and expanded way of looking at the real world structure (R) under study.
So far, the process through which a modeler creates the evolutionary change from M1 to M2 is not well
understood and/or documented. This process tends to rely mainly in the creativity of the modeler. To date,
many of the mechanisms devised to understand the structure of the model better, like eigenvalue analysis
3 Italics added.
Page 4
or automated ways like DIGEST (Mojtahedzadeh, 2001), would not take the modder from M1 to M2.
Instead, these mechanisms help the modeler go one layer deeper into the original model or theory about
the world. The more sophisticated we become at analyzing a determined structure, the deeper we will be
able to go into the same view of the world.
If the real world is changing in its structure over time as depicted in Figure 1(B)—represented by R1
and R2, then gaining deeper understanding of the same theory or worldview might be counter productive
and even misleading. As shown in Figure 1(B), a first model (M1) was related to the real world observed.
at time 1. Then the model evolved to (M2) in an attempt to gain a better and further understanding of the
teal world in order to intervene and try to change the problematic situation but at the same time the real
wodd structure changed from R1 to R2. Nevertheless, thanks to the evolution of the model from M1 to
M2, the model—M2 in this case—still has some relation to the real world. If the modeler had decided to
gain a deeper understanding of M1 only, instead of allowing the model to evolve to M2, the model’s
relation to the real world at R2 would have been lost.
System Dynamics: Some Assumptions, The Modeling Process, and Terminology
System Dynamics was developed in the late 1950's and early 1960's at the Massachusetts Institute of
Technology’s Sloan School of Management by Jay W. Forrester as he consciously applied control
piinciples to management and economics (Lane and Oliva, 1994, p. 219). System dynamics is a
computer-aided approach to policy analysis and design (Richardson, 1996, p. 656). This approach can be
applied to dynamic problems arising in complex social, managerial, economic, or ecological systems. In
general, any dynamic system characterized by interdependence, mutual interaction, information feedback,
and circular causality can be examined using system dynamics (Richardson, 1996). System dynamics
concentrates on the circular causality and feedback nature of systems in order to understand their behavior
throughout time. The main purpose of system dynamics is to try to discover the ‘structure’ that conditions
the observed behavior of the system over time. System dynamicists try to pose ‘dynamic’ hypotheses that
endogenously describe the observed behavior of systems. The ‘endogenous’ view is ciitical to system
Page 5
dynamics modeling allowing the existence of feedback rich explanations for certain types of
Phenomenon. System dynamics is fundamentally interdisciplinary and is grounded in the theory of
nonlinear dynamics and feedback control developed in mathematics, physics, and engineering (Stenman,
2000, pp. 45).
Since its creation, many people in system dynamics including Jay Forrester (1979, p. 14) have said
that “the most important and difficult step in system dynamics is perception of a model structure
appropriate to the chosen purpose.” One of the questions constantly asked is: how is a model structure
detenmined? Many answers revolve around words such as, art, craft, intuition, expertise, process, style,
and science. However, the word ‘reality’ is hardly heard. According to Forrester (1979, p. 15),
“perceiving model structure is akin to the process of invention,” and “[perception of model structure] is
not easy.” Forrester goes on by saying that (p. 15), to him, “it is a trial and error process” where models
are formulated, tested, evaluated, discarded, and replaced over time in “a gradual shaping of a unity
between the structure of the real system’, the behavior of the real system, the [structure of the] model, the
behavior of the model, and the model builder’ s purpose.”
If the modeling process that Forrester describes takes a considerable length of time, then many
changes—to the model and to the real world—can occur during the process. The structure of the real
system can change, the behavior of the real system .can change, the model builder’ s purpose can change,
and the model can be changing. These series of changes can be highly interrelated in the context of a
system dynamics intervention. Probably, the changes in the model development will influence
‘understanding that will trigger action that, in tum, will change the structure of the real world before the
whole exacise is over (a representation of this idea can be seen in Figure5). The structure of the real
system then becomes a moving target that the modeler is trying to capture. Nevertheless, in the system
dynamics literature, mentions of the dynamics of the structure are virtually nonexistent. The focus of the
modeling process is on understanding the dynamics of the behavior assuming that the structure is fixed
(Richardson and Pugh, 1981; Sterman, 2000).
“ Ttalics added
Page 6
According to Forrester (1979), “a system dynamics model contains policies that generate certain
behavioral responses; these policies are constant for the duration of the model simulation” (p. 16). This
static view of the structure makes things easier for the modeler, but what about the behavior of the
structure of the real would, is that static?
Perceiving the structure of the model may be the most important and difficult step in system
dynamics modeling. Thus, it makes sense to try to purposefully leam from the structure and emphasize
the differences between leaming from the structure and leaming from the behavior as separate sources of
system understanding and insight Also, if the structure is so important—among other things is the one
that conditions the behavior—it makes sense to try to understand the dynamic nature of the structure once
itis perceived. It can be hypothesized that most system dynamics interventions will take place more
rapidly than changes in the structure of the system. However, even when that is the case, acknowledging
the dynamic nature of the structure can provide additional insights.
Mathematically, the basic structure of a system dynamics model (Richardson, 1996, p. 657) is a
system of coupled, nonlinear, first-order differential (or integral) equations that can be waitten in the
form:
dx _* .
(1) Hh =x(t) =f[x(0,u]; Given x(t)
Wrere:
x(t) =n" Order vector of system states (orlevels)_ “(t) = Vector of exogenous inputs
X(t) ) =Initial value for state vector at t =, J () =Nonlinear vector function
dx
a =x() =Time derivative of the state vector
The set of equations representing a system dynamics mode! can always be manipulated into the
form shown in equation (1). This can be identified as the structure of the model. However, is this what
system dynamics is all about? Is system dynamics modeling more than just mathematics?
Page 7
Understanding the Structure
Structure and Behavior
1 had always thought that when we said ‘structure’ within the system dynamics field, we all thought
about the same thing. However, that might not always be the case. To clarify what structure is, what
behavior is, and the relationship between the two might be a challenge. Forrester (1968b, p. 406), in his
discussion about the nature of system dynamics as a theory of structure, argues that “it may be helpful to
distinguish two aspects of a system investigation—that relating to structure, and that relating to dynamic
behavior.” Then he goes on to say, “The two [structure and behavior] are intimately interwoven because it
is the structure which produces the behavior”. Following Forrester (1968a; 1968b; 1968c; 1969), I think
that if we can understand how to leam from behavior and how to leam from structure we could move
ahead in our understanding of how to influence change in complex systems. Possibly a theory of
structural evolution will allow us to leam from structures the way that we historically seem to be leaming
from the evolution of behavior. In a way, when we build a system dynamics model, we create a structure
that supports a theory of behavioral evolution. After we create that structure—which is a mental
construct—we try to leam the way in which that specific, static and fixed, structure creates the evolution
of behavior of the systems that concems us.
When system dynamicists are trying to claim something about the real world by means of mental
constructs, which we adjust to fit our views of the world using our own mental models, are generating
behavior conditioned by a predefined structure. This type of behavior is in itself the product of a structure
that conditions what we do when we create a simulation model. This behavior is the result of a set of
tules—wnitten and unwritten—that shape our actions and understandings about system dynamics
modeling and what it is that it can do for us and for the world.
Definitions of Structure
Forrester (1979) argues that an important focus of system dynamics is in the policy domain. He says
(p. 2) that “system dynamics focuses on policy and how policy determines behavior.” Policy, to Forrester,
Page 8
is the criteria for decision-making; the niles. He says that policy “is the rationale that determines how a
stream of decisions will be modulated in response to changing inputs of information.” He argues that
system dynamics deal with policy and structure and the resulting behavior. Forrester is very clear about
the focus on policy because, essentially, the interconnectedness of policy is the structure that produces the
observed behavior. According to Richardson (2002a), in the system dynamics world, ‘structure is an
interconnected set of stocks and flows and feedback loops.’ System dynamicists like to focus on, and
think in tems of, accumulations, causal connections, flows of change in accumulations, and parameters.
According to Senge (1990, p. 44), structure, or better said ‘systemic structure’ is “concemed with the key
interrelationships that influence behavior over time, [...], interrelationships not between people but
among key variables” of the system under study. According to Sterman (2000, p. 107), “the behavior of a
system arises from its structure” and “the structure consists of the feedback loops, stocks and flows, and
nonlinearities created by the interaction of the physical and institutional structure of the system with the
decision-making processes of the agents acting within it.” However, we need to be very conscious that
other disciplines have specific definitions of ‘structure’ too. Organizational theorists, and sociologists,
among others, also talk about structure.
Tt is complicated to talk about structure and behavior because these two words can mean different
things to different individuals even within the same community. As Richards (1986, p. 27) points out,
“words are arbitrary symbols and they have no inherent meaning,” they take on the meaning of the
context in which a person encounters them. Additionally, the specifics of the context itself are the product
of a process of personal or individual interpretation. Context is the key to meaning because words, as they
pass from context to context, change meanings. The meaning is not in the words themselves but in
people’s heads.
When we are using system dynamics to try to understand the world and do something about it, are we
changing the structure of the real world? Are we changing the behavior of the real world? Can we change
the behavior of the real world without changing the structure of it? If we consider stocks, flows, and
Page 9
parameters as part of the structure, then any change to one would have to be considered a structural
Undiscovered Rework
Generation of
Undiscovered
Rework \
Cumulative Real v
Real Progress Progress Cummulative Real
Rate Progress Perceived
Figure 2—Part of Project Model (adapted from Richardson and Pugh, 1981, p. 199 Figure 4.28)
For example, in the case of a standard project model (Richardson and Pugh, 1981, p. 199), a
parameter like fraction satisfactory FSAT has an influence both on the level of cumulative real progress
and undiscovered rework (see Figure 2 above). Having a value of 0.5 means something about the
accumulation of work and rework. However, fraction satisfactory having a value of one would mean that
the whole structure of the model related to undiscovered rework is inactive and tumed off. Therefore,
changing fraction satisfactory from 0.5 to almost anything but one will not change the structure of the
mode, while changing it to one will wipe out a whole portion of it Again, some parametric changes do
change the structure yet most do not change it.
Structure and Understanding
Quite commonly, a picture of an iceberg is used to depict the power of systems thinking and system
dynamics interventions—see the upper part of Figure 3 that includes events, tendencies, and structure—as
described in prose by Senge (1990, p. 52). The lower part of the fig™ure (the additional fuzzy iceberg like
figure below the upper iceberg) is an addition offered by the author. The picture in Figure 3 conveys the
idea that event-based explanations to phenomena are very superficial—above the waterline and therefore
eesily visible—and doom their holders to a reactive stand in the world. Tendencies-based explanations are
Page 10
better in the sense that these are responsive and focused on long-term implications of the phenomena.
Structure-based explanations are the most powerful and least common because of the difficulties inherent
to this type of explanations.
Events
)- Tens
Structure
Evolution
@ Supra - Structure
Figure 3—Supre-structure to Events
The strucuure is hidden and difficult to see making it virtually impossible for laypersons to ‘uncover’
it The structure, in order to create generative explanations of the phenomena, needs to be uncovered
(Senge, 1990). As Senge (1990) argues, “[though rare] , structural explanations, when they are clear and
widely understood, have a considerable impact.” (p. 53) The overall idea is that the leverage points for
change lie in the structural layer of understanding. Now, let us go deeper into this way of looking at
explanatory power. What if the structure is in itself just the tip of another deeper, more burdensome
iceberg that has elements that conditions its behavior. Is this behavior what we perceive as structure? This
would be like uncovering a deeper layer of understanding to clarify what it is that conditions the structure,
which is conditioning the behavior that we are interested in. Possibly, the structure that we are able to
pexveive is just a ‘snap shot’ of the dynamic structure. Just like an event is nothing but a ‘snap shot’ of the
dynamic behavior. The point is, that when we use systems thinking or system dynamics in interventions
in the real world, looking at a snap shot of the behavior is not enough. Having a static view of the
behavior of the real world is what has kept the solutions for problematic situations in low-leverage mode
Page 11
in the real system. System thinkers and system dynamicists argue that we should try to see the overall
picuure of the behavior. We should try to have a dynamic view of the behavior in order to have a better
opportunity to leam key insights that will help change the world. Even further, we say that we should go
beyond dynamic behavior focusing on structural explanations to uncover the causal relationships that are
conditioning the observed dynamic behavior. We say that the tue source of deep leaming and change lies
in the structure. Only if we are capable to see the structure and leam from it, will we be able to find high
leverage points for change and be able to change the world in an efficient, effective, and long-lasting way.
Subsequently, what about if we took the same approach towards the structure itself? What if we try to
grasp the deep conditioner of the structure—herein called the supra-structure—and try to leam both the
dynamics of the structure and the supra-structure that is conditioning the way the structure appears today?
The change in paradigm would be to focus on the structure as the behavior of a deeper structure that is
Operating in the real world. I know that some fraction of the readers will be thinking that this recursive
approach has an infinite number of possible layers to it. That if the structure has a supra-structure that
conditions its behavior, then the supra-structure by extension should have a supra-supra-structure
conditioning its behavior as well. This story could go on and on, down or up, the chain forever. This
would be similar to the paradox of Wigner's Friend of quantum physics related to cansciousness and
parallel universes (for an explanation of the paradox go to Wolf, 1989, pp. 215-225).
Tt seems to be that, in the system dynamics literature, there are no clear indications that lead us into
the clarification of the dynamics of the structure, only the dynamics of the behavior. In discussing
impediments to leaming and the iterative nature of system dynamics modeling, Stenman (2000, pp. 20,
87-80), depicts the role of a higher order structure in the process of modeling and leaming that can be
very well related to the ideas of a dynamic structure.
Tt might be that not only we have problems understanding complex systems because they are rich in
feedback (Sterman, 1994), but we also have problems understanding complex systems because these
complex systems are changing their structure while we are trying to understand them using a fixed
structure approach.
Page 12
Misperception of Structure
One interesting consideration with respect to identifying the structure of a system is whether or not
‘we could ger it wrong. Could we be leaming the wrong structure? Could this be what Forrester calls
dangerous models? According to Forrester (2001) there are a few very useful models, a lot of
unimportant—those that neither help nor cause problems—models, and some dangerous models—those
that cause problems when they are used. Forrester says that the world would be better off without these
‘wrong’ models. Furthermore, if we get it wrong, what is the effect on behavior? What are the problems
that are created by our inability to perceive the correct structure? Is this a problem of perception or
capacity? If this is a problem of perception, then we could address it by working on better structure-
perceiving mechanisms. However, if this is a problem of fundamental capacity, we would be facing a
much more complicated issue. Part of the problem lies in the way you can build capacity in an
organization or a field of study. One way to build capacity in the field would be by clarifying the rules of
engagement in intervention projects using system dynamics. Clearly, this could create unintended
consequences that might be worth exploring in advance. Oliva’ s (Forthcoming) work on calibration
addresses the problem of misperception of structure by treating the ‘wrong’ structures as altemative
hypotheses that we should be able to discard using scientific procedures. Oliva (Forthcoming, p. 2-3)
argues that “in testing a DH [dynamic hypothesis] is not enough for the model to match the observed
behavior, the behavior generated by the model has to be right for the right reasons.”
Understanding Changes in Structure
Structural Evolution
Structural evolution has to do with how structure changes over time. It is related to the different
characteristics of the structure of a system at different points in time. The question now is how do you
measure changes in structure? What it is that we would be graphing over time? The number of equations?
The number of variables (stocks and flows)? The [number of, or characteristics of] dominant loops?
Certain indicators of the dominant loops could be used. For example, overall contributions to the behavior
Page 13
using eigenvalue elasticity indexes or the pathway participation metrics developed by Mojtahedzadeh
(1996) might be used.
A theory of structural evolution might be geared towards generating insightful system stories
(Mojtahedzadeh, 2001) about the way the structure changes. The research interest moves from
understanding the behavior derived from a particular structure to changes in this structure over time.
If itis true that structure is changing we should be really careful about clarifying which structure is
changing. Is the model structure the one that is changing or is the real world structure the one that is.
changing? Fey (2002), in a recent work related to organizational change, presents the idea of ‘pattem
feedback control’ (PFC). PFC relates desired changes in behavior with changes in structure. He argues
that (p. 1) “system dynamics (SD) is intended to solve dynamic problems in existing living systems by
achieving improved future time pattems for problematic system variables” and “since the general time
pattems of a system's variables are created by the system’ s feedback control structure operating through
time, improved pattems can only be obtained by changing the system’ s feedback structure in ways that
will produce improved future pattems.”
Here the “variable” of interest is not a single time value for a clearly defined entity, but
the analyst’s perception of the nonplanar map of the systents dominant loop geometry
that existed during the period when the pattems in the time histories were created. [...]
These unusual feedback entities are necessary because the patterns cause the problems
and the loop geometry causes the patterns, so the dynamic problem has to be solved by
understanding and then changing the loop geometry.
The question is: How can you ‘optimize’ the structure in order to ‘optimize’ the desired behavior?
Traditional optimization algorithms will try to ‘optimize’ the behavior by finding values of parameters
that, within a fixed structure and according to a certain objective function, will generate the closest fit in
behavioral pattems (Oliva, Forthooming). However, these changes are parametric changes and not
necessarily structural changes. What could one feed back as trend of dominant structure? A numexical
qofile of the dominant structure perhaps?
The behavior of interest switches from behavior over time to structure over time. Instead of
concentrating on calibrating the parameters of the structure to ‘optimize’ the generated behavior, now the
Page 14
focus of the effort would be on calibrating the new structural behavior to optimize the fina! structure.
Possibly, how this works is that you click <run> and generate a certain behavior over time for the full
time horizon of inquiry, lets say 120 months. Then an automated piece of software identifies the
prominent structure for all the variables in the model and creates a pattem for you. However, because we
know that the existence of a dominant structure does not imply that other loops are not active, the
indicator to be chosen must be treated with caution. Additionally, we have to be aware that changes in
parameter values will change the pattem in dominant structure. In order to understand the stability of the
pattem we could apply parametric sensitivity analysis to the dominant structure pattem generation. If the
dominant structure pattem is sensitive to parameter changes then we could conclude that the dominance is
weak or unstable. If the pattem tums out to be not sensitive to parametric changes then the dominance is
strong or more stable. Consider the example of the simple system presented in Figure 4. In this simple
model there are 8 variables in total—see the equation list in Appendix 1. Out of the eight variables used,
four are variables and four are constants (sounds contradictory, right?). The four constants do not vary
over time and therefore are not considered for the dominant structure pattem definition.
IBS Business
BC Business _ |_Structurey
we cid \
BCN Business
Construction Normal LFO Land Fraction
BLM Business Occupied
Land Multiplieta-—~ }
Area Are
BLM F LBS Land per
Multiplier Business Structure
The prominent-structure analysis of this simple model would include four variables—BC
Business Construction, BS Business Structures, LFO Land Fraction Occupied, and BLM Business Land
Multiplier. The prominent structure for each variable would be identified over time and recorded—see
possible arrangement in Table 2—first with the original set of parameters (U), then with different sets of
parameters (U’) for the sensitivity analysis.
Page 15
Original Set (U)
Causal Route / Prominent Structures
BLM
ees LFO Land BS
Be ee HE Fraction —t—) Business
wee business 2 Mutipler Oooxpied Snuaties
usiness
Structures: Contract” Neg. BS
si “Business
: Structures:
& (~ Land F >
a
2 (R) BC Busines [BS Busines...] BLM Busine
s
a
BC Busines
From t=0.00 to t=10.63 From t=10.75 to t=24.00
LFO Land BS
3 Fraction —*—j Business
‘6. Occupied Structures
: LFO Land F. i vo LM Busine.
E [ES Busines..] BC Busines SR (B) ec Busines
4H
2
x
From t-0.00 to t-10.63 From t=10.75 to t-24.00
a Elements corresponding to F/ow BC Business Construction
NN
Elements corresponding to 4ux BLM Business Land Multiplier
Altemate Set (U’)
Table 2—Prominent Structure Pattem
Page 16
Then this pattem of dominance is somehow ‘feed back’ to change the structure in the way that makes
more sense to what I want. I do it, or the software does it, and then I have finished iteration 1 of the
process. Then one does it again for as many iterations as one wants.
Forrester (1979, p. 16) says that “a system dynamics model contains policies that are constant for the
duration of a model simulation” and that “the policies are laws of human behavior, for the circumstances
‘within the system’ excluding any possibility of change during the simulation process. Subsequently,
being conscious that when you hit the rwn button you have a fixed structure generating the behavior—a
snap-shop of the structure. And being conscious that the main concem of the process many times is with
hitting nm and being able to get a behaviorally deductive artifact to analyze. Maybe, we could try to
change the concem from the run button to something else. Possibly, the focus of attention of the overall
process should be on the mental and social processes that are triggered before and after we hit the run
button. Actually, hitting the-wn button is just an excuse to activate a leaming process about the system,
and the model, that allow us to create an intervention that can lead to changes in the real world (see
Figure 5).
Influencing the Structure
Figure 5 expands Saeed’s (1992) view of the system dynamics modeling process (p. 252) and
includes the effects of understanding the structure and the behavior—system conceptualization—on
actions that actors make to change the real world. Saeed’ s (1992) main concem while discussing his view
of the system dynamics modeling process is with model creation and the representation of reality, not
with how enhanced understanding changes reality. Saeed (1992) stresses two types of processes present in
asystem dynamics modeling intervention: the structure-validating process and the behavior-validating
process. These two processes, according to Saeed, are the mechanisms through which actors fine-tune the
models that represent real systems in order to understand better the nature of the problematic behavior
identified. The basic mechanism used is the comparison and reconciliation of the model with the
perception of the real world. I argue that through those very same processes, actions are triggered that
Page 17
change the very same elements that are trying to be understood and identified. Three more elements are
suggested: two actor- centered processes, and an autonomous process.
The behavior-influencing-processes loop identifies the way in which actors, as a result of perceiving
behavioral cues of reality, generate actions that change behavior. These actions would be of the type that
Richardson (2002b) calls parametric changes that do not change the structure. This loop can be related to
what Argytis and Schon (1996) call “single-loop leaming”. Also, thestructure-influencing-processes loop
telates to exactly the same as the behavioral influencing- processes loop but associated to structural cues
of the system instead of behavioral ones. This way of leaming about the system and changing the system
would be a ‘structural’ one. This two different processes presented can help us identify two different
“single-loop leaming” processes: the behavioral single-loop leaming and the structural single-loop
leaming. Even thou this two leaming processes would be considered “single-loop” type, the main source
of leaming would be quite different behavioral vs. structural.
The changing-behavior-via-structure loop, which identifies the path in which actors change the
behavior of the system through changes in the structure, is related to what Argyris and Schon (1996) call
“double-loop leaming’. In the representation presented in Figure 5, actors in the system judge perceptions
of the behavior and of the structure of the system (against goals or standards) in order to generate actions
that can modify the reality perceived The action will be triggered when the existence of a difference or
gap between the perception of the real system and the goal is larger than the tolerance that the actors have
to the existence of this gap. The greater the tolerance, the larger the gap must be to actually trigger actions
geared towards changing the real system. Theoretically, a ‘supertolerant’ actor would never act in
response to any gap because no gap would be large enough to be considered reason for action, and a
compleely ‘intolerant’ actor would generate a continuous stream of action in response to the smallest
perceived difference. Additionally, these perceptions— behavioral and structural—can, and most
probably will, be biased and/or mistaken (Levinthal and March, 1993).
Page 18
Changing Behavior
via Structure
System's S S
Structure [“* oe) 2 > sional
Change in System's Change in System's |_Behavi
Structure Behavior
Structure-influencing Behavior-influencing
Perception of Processes ; Processes fon ok
System Structure Actor's Actions Ss - ici
Tolerance to ee
Diff ein St Structure-Validating
ce Processes System Processes
Conceptualization
Representation of Deduction of Model
Model Structure Behavior
OOS ee Moc Jel ee od
Fonnulatiort
Figure 5— Structure- Behavior
Page 19
The interpretation of the cues by means of the judgment functions—cognitive structures—of the actors of
the system can have important conceptual problems as well (Mumpower, 1991). All of these possible
misunderstandings can cause important differences for managers when acting in complex systems
(Mumpower and Stewart, 1996) because, in a way, all of those elements are part of the supre-structure
that is actively contributing to changes in the structure of interest associated to the problematic behavior
under study.
Conclusion and Future Research
One of the basic assumptions of system dynamicists, is that when we hit the <1un> button we have, in
the model, an ‘accurate’ representation of the real world that can be considered a deductive mechanism
for behavioral evolution analysis. An implicit assumption is that during the time horizon of the model nn,
or the simulation, the operating noms and physical constraints of the real world will remain constant.
However, it could be that the technology would change and the norms would change altering the structure
of the real world. If this were the case, the output of the model would not reflect the behavior of the real
world unless we were to change the assumptions and relationships of the model for that point in time on.
The task of modelers trying to understand behavior based on the ‘right’ structure for the ‘right’
reasons as opposite to the ‘wrong’ structure or reasons have been identified in the system dynamics
literature (see Oliva, Forthcoming, for a detailed explanation). If there are several different structures that
can generate the behavior. How would you discriminate among them? As argued in other section of this
paper, (Oliva, Forthooming) addresses this problem by treating the altemative structures as altemative
hypotheses that we should be able to discard using scientific procedures. Others, like Richardson (1985),
suggests that it does not matter as long as the chosen structure resembles the real world. However, how
about if the difference in representations is minor between the ‘right’ and the ‘wrong’ representation and,
by not identifying it, you generate misleading recommendations? At the end of the day, are system
dynamicists trying to understand systems or are they trying to solve problems? Isn't the former a
prerequisite of the latter?
Page 20
All the dements discussed in this paper have led me to think about the importance to find a way to
understand them better, through an integrative framework or theory, how reality, perception, tolerance,
gaps, goals, action, structure, and behavior, relate to each other in order to create the realities that we
know. Richardson and others (1994) have thought really hard about the foundations of mental model
research and how these might influence the way system dynamics interventions can change to change the
word. This seems important because, as Sterman (2002, p. 527) says, “the real purpose of system
dynamics [is]: To create the future we truly desire—not just in the here and now, but globally and for the
long tem.”
Further research can be done in trying to explore new sources of leaming— individual and
organizational—hby exploring the differences that arise when individuals and organizations leam primarily
from behavior or primarily from structure. In addition, it is needed to explore the research methods
needed for the analysis of the dynamic nature of the structure: longitudinal studies and specialized
mathematics might be needed. Lastly, and most inportantly than anything else here presented, it is
needed to explore the essential mechanisms to be able to change from M1 to M2—see Figure 1 at the
beginning of the paper—both in a fixed word (R1) and a changing word (moving from R1 to R2) inan
efficient and systematic way. This change in paradigm is needed to move forward the modeling practice
and to be able to add a little more science to the craft of modeling.
Page 21
Appendix 1—Equations for Simple Model
Number | Equation
Area Area=1000 Units: acres
2 BC Business Construction=BS Business Structures*BCN Business
Construction Nonmal*BLM Business Land Multiplier Units: units/year
3 BCN Business Construction Normal=0.07 Units: fraction/year
4 BLM Business Land Multiplie=BLM F Business Land Multiplier
Function (LFO Land Fraction Occupied) Units: Dnml
BLM F Business Land Multiplier Function ([(0,0)-
5 (1,1.5)1,(0,1),(0.1,1.15),(0.2,1.3),(0.3,1.4),(0.4,1.42105),(0.5,1.36184),(0.
6,1.19737),(0.7,0.907895),(0.8,0.539474),(0.9,0.25),(1,0)) Units: Dmnl
6 BS Business Structures=INTEG (BC Business Construction, 1000)
Units: units
7 LBS Land per Business Structure=0.2 Units: acres/units
8 LFO Land Fraction Occupied={BS Business Structures*LBS Land per’
Business Structure)/Area Area Units: fraction
FINAL TIME =24 Units: year
INITIAL TIME =0 Units: year
Control | SAVEPER =1Units: year
TIME STEP =0.125Units: year
Page 22
References
Argytis, C. and D. A. Schon (1996). Organizational Leaming II: Theory, Method, and Practice. New
York, Addison-Wesley Publishing Company.
Fey, W. R. (2002). Organizational Change froma New Perspective: Pattem Feedback Control in Human
Systems. Proceedings of the 20th Intemational Conference of the System Dynamics Society,
Palermo. Italy.
Forrester, J. W. (1968a). "An Appreciation of Industrial Dynamics." Management Science 14 (7): 383-
397.
. (1968b). "Industrial Dynamics-A fter the First Decade." Management Science 14 (7): 398-415.
. (1968c). "Market Growth as Influenced by Capital Investment." Industrial Management Rev.
MIT) 9(2): 83-105.
. (1969). "A Deeper Knowledge of Social Systems." Technology Review 71 (6): 2-11.
. (1979). "System Dynamics-Future Opportunities." MIT D-Memo Files D-3108-1: 1-24.
. (1985). "'The' Model Versus a Modeling ‘Process’." System Dynamics Review 1 (1): 133-134.
. (2001). System Dynamics. Decision and Policy Sciences Brown-bag Lunch Seminar Series.
Albany, N-Y., University at Albany.
Lane, D. C. and R. Oliva (1994). The Greater Whole: Toward a Synthesis of SD and SSM. 1994.
Intemational System Dynamics Conference, Sterling, Scotland, System Dynamics Society.
Levinthal, D. A. andJ. G. March (1993). "The Myopia of leaming." Strategic Management Jourmal
14 (Special Issue: Organizations, Decision Making and Strategy): 95-112.
Mojtaheczadeh, M. (1996). A Path Taken. Unpublished Doctoral Dissertation. School of Public
Administration and Policy. Albany, NY, University at Albany.
. (2001). Digest: A New Tool for Creating Insightful System Stories. Proceedings of the 19th
Intemational Conference of the System Dynamics Society, Atlanta, GA USA.
Mumpower, J. (1991). "The Judgment Policies of Negotiators and the Structure of Negotiation
Problems." Management Science 37: 1304-1324.
Mumpower, J. L. and T. R. Stewart (1996). "Expert judgment and expert disagreement." Thinking And.
Reasoning 2 (2/3): 191-211.
Oliva, R. (Forthcoming). "Model Calibration as a Testing Strategy for System Dynamics Models."
European Joumal of Operational Research
Richards, I. A. (1936). The Philosophy of Rhetoric. London, Oxford University Press.
Richardson, G. P. (1985). "Introduction to the System Dynamics Review." System Dynamics Review 1 (1):
13.
Page 23
. (1996). System Dynamics. Encyclopedia of Operations Research and Management Science. C.
M. Hanis. Boston, MA, Kluwer Academic Publishers: 656-660.
. (2002a). Definition of Structure provided at the presentation of: Structure and Behavior:
Exploring System Dynamics Terminology. DAPS Brown-bag Lunch Seminars. Albany, NY,
Univesity at Albany.
ee . (2002b). Exploring System Dynamics Temminology--Some comments and definitions.
PAD824-Advanced-Topics-in- System: Dynamics-Course-Fall-2002. Albany, NY, University at
Albany.
Richardson, G. P., D. F. Andersen, T. Maxwell and T. Stewart (1994). Foundations of Mental Models
Research. Albany, New Y ork.
Richardson, G. P. and A. L. Pugh, III (1981). Introduction to System Dynamics Modeling with
DYNAMO. Cambridge MA, Productivity Press.
Saeed, K. (1992). "Slicing a complex problem for system dynamics modeling." System Dynamics Review
8(3): 251-261.
Senge, P. M. (1990). The Fifth Discipline: the Art and Practice of the Leaming Organization. New Y ork,
Doubleday/Cunrency.
Sterman, J. D. (1994). "Leaming in and about Complex Systems." System Dynamics Review 10(2/3):
291-330.
. (2000). Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston
MA, Irwin McGraw-Hill.
. (2002). "All models are wrong; reflections on becoming a systems scientist." System Dynamic
Review 18(4): 501-531.
Wolf, F. A. (1989). Taking the Quantum Leap: The New Physics for Non- Scientists. New Y ork,
Page 24.
Back to the Top
CC BY-NC-SA 4.0