Problems for the Future of System Dynamics

George P. Richardson
Rockefeller College of Public Affairs and Policy
University at Albany — State University of New York

The field of system dynamics first envisioned by Forrester in 1956 is approaching its fortieth
anniversary. It has experienced two significant revolutions: changes in practice wrought by the
development of iconographic software1 and the shift from modeling to improve a process to
modeling for the purpose of improving people’s mental models (Forrester, quoted in Keough and
Doman 1992). Its third revolution — expanding the pool of practicing modelers far beyond the
nucleus who have received university training — has been underway since STELLA hit the
schools in the late 1980s. A fourth revolution — moving to the forefront qualitative modeling
based on word-and-arrow archetypes — began quietly in the mid 1980s (e.g., Wolstenholme and
Coyle 1983) and exploded on the scene with the publication of The Fifth Discipline (Senge
1990). These recent and ongoing revolutions carry promise and potential pitfalls for the field.

The potential pitfalls of our current successes suggest the time is right to sketch a view of
outstanding problems in the field of system dynamics, to focus the attention of people in the field
on especially promising or especially problematic issues. The list that follows should be thought
of as the opening remarks in what I hope becomes an increasingly interesting conversation among
practitioners. It stems from reflections on practice in the field during my tenure as editor of the
System Dynamics Review, but other than the view from that mountaintop I can claim no
particular expertise for the task. It seems vital for the field, however, that we begin the
conversation, and so I offer the following initial statement of Problems for the Future of System
Dynamics.

There is a double meaning here, which is intended: the problems we list should be deep and
difficult problems worthy of attention now and in the future, and they should be of such magnitude
that if not addressed they will pose problems for the field’s future. In two senses, then, we are
discussing Problems for the Future of the Field.

The initial list contains eight problem areas:

Understanding model behavior
Accumulating wise practice
Advancing practice
Accumulating results
Making models accessible
Qualitative mapping and formal modeling
Widening the base
Confidence and validation

The goal of the list is to initiate discussion of an agenda for concerted work in the field of system
dynamics. If indeed these are the deep, difficult, and important problems facing the field, then we
should begin to address them in collaborative and productively competitive ways.

1 STELLA (High Performance Systems), and later Vensim (Ventana Systems) and PowerSim (ModelData).

-1-
Understanding model behavior.
The principal outstanding technical problem in simulation modeling is the development of
tools to aid understanding model behavior.

For more than thirty five years practitioners have relied on a time-consuming and often incomplete
process that iterates from formulation to parametrization, testing, observation, hypothesizing and
back again. (See, for example, Richardson and Pugh 1981, 267-276.) Understanding connections
between complex model structure and behavior comes, if one is skillful and/or lucky, after a
prolonged series of model tests of deepening sophistication and insight.

Alternatively, and undoubtedly preferably, understanding connections between model structure and
behavior comes from a sequential modeling process that moves from simpler formulations to more
complex structure. [Richmond has urged, for example, that the modeler first build a complete,
simulated model that is understandably simple, and then move from that toward complexity; he
argues that attempts to build a complex model at the outset are likely to be doomed to failure.] At
each stage, understandings from the previous simpler stage guide the path to understandings of the
current structure. The process at each stage is still the time-consuming iteration of tests,
observations, and hypotheses, but it is undoubtedly speeded by understandings codified at the
previous stages of model complexity.

Technical support for understanding the connections between model structure and behavior is weak
to the point of being almost nonexistent. In most of the software in use in system dynamics, there
is no such support. The most sophisticated tools we have at hand in software are the causal tracing
tools and graphs in Vensim.2 Using these tools one can quickly move backward or forward in
causal chains in a model to track down the source of intriguing behavior in a particular variable.
While causal tracing capabilities are a major step forward in simulation-based tools available to us,
they are still inadequate to the task of understanding the behavior of complex models. We are still
forced to operate as we have for almost forty years, in a time-consuming and only gradually
illuminating iterative process of hypotheses and tests.

There are some potentially promising directions. The dissertations of Nathan Forrester (1982) and
Robert Eberlein (1984) contain approaches to understanding model behavior that rest, respectively,
on eigenvalue elasticities and participation factors (see Richardson 1986 for a summary). In those
works, formal analyses based on eigenvalues were used to identify dominant structure and, in
Eberlein’s work, to reduce a complex model to a simpler structure while preserving particular
behavior modes of interest. In a sense, simplifying a complex structure while preserving a
behavior of interest is one way of saying what we mean by understanding the connections
between structure and behavior. However, both of these approaches rest on mathematics
appropriate for linear systems and their applicability to nonlinear systems has not been tested in our
literature beyond the work of these authors.3

Richardson (1984/1995) contains a simpler approach that directly addresses nonlinearity and the
shifts in loop dominance that nonlinearities can cause. Mojtahedzadeh (1995) is pursuing a
2 Ventana Systems, 149 Waverley Street, Belmont, MA 02178.
3 It is undoubtedly significant that Eberlein has not embedded eigenvalue-based model analysis tools in Vensim.

-2-
generalization of this approach which appears to be a promising step forward and in the process is
striving to clarify what we mean by understanding. Kim (1995) is engaged in similar researches.
But clearly more effort is in order. If wider audiences are to build and understand models to aid
their own thinking or the thinking of others, then our software must do more than provide model
diagrams to help with the building. Of the three tasks — model conceptualization, model
formulation, and model understanding — unquestionably the two most difficult are the two that are
least formal — conceptualization and understanding. The future of the field needs software
support for understanding the links between stock-and-flow/feedback structure and dynamic
behavior.

Accumulating wise practice

The field of system dynamics needs improved mechanisms for accumulating and promulgating
insights about wise modeling practice.

In forty years of research and applications, system dynamics practitioners have solved many
practical modeling problems. Wise practice includes insightful statements about
• building blocks of system structure: stock-and-flow structures, and elementary generic
processes
• molecules of generically useful model structure — more complex than building blocks
• consulting wisdom
• group modeling principles
• teaching practice that accelerates growth in modeling capabilities
• wisdom about problem definition and system conceptualization
• wisdom about building confidence in models for policy analysis
Yet it is hard to locate the wisest statements about modeling practice. Searchers would begin, of
course, with Forrester’s classics (1961, 1968, 1969, 1971) and with the award-winning books
and articles in the field.4 But one should not miss the lesser known classics, such as Meadows
(1982), Ford and Gardiner (1979), Homer (1983), and others.

In addition to the great works in the field, embedded throughout our literature are
• solutions to modeling puzzles, such as the formulation and initialization of a coflow for years of
experience of a work force; a robust formulation for the fraction of a resource like petroleum
that can flow from a level like proven reserves; formulations for energy retrofits, managerial
recruitment, training, and attrition; continuous (aggregate) logic; statistical measures in
continuous models; and so on.
• wisdom about model-based consulting practice from the decades of experience of the great
system dynamics consulting firms
• practical wisdom about teaching the modeling arts, including exercises that can actually
accelerate learning the extremes of conceptual and technical skills modeling requires and projects

4 The Forrester awards include Roberts et al. (1983), Richardson and Pugh (1981), Aracil (1981) and Mosekilde (for
advances in complex nonlinear dynamic systems, e.g., Mosekilde et al.1985), Morecroft (1985), Richmond (for
STELLA and its manuals and teaching materials), Sterman (1989a,b), Dennis Meadows (for simulation-based games
such as Stratagem), Senge (1990), Richardson (1991), Abdel-Hamid and Madnick (1991), and Saeed (1990). Other
award-winning system dynamics works include Cooper (1980), Roberts and Dangerfield (1990), and Reagan-
Cirincione et al. (1991).

-3-
 that are so well designed that they move people rapidly toward self-sufficiency
• group model building and model use techniques that can move modeling out of the closet and
directly involve policy makers

New practitioners would find the wisdom and best practice in the field hard to find. Armed with
the apparent power of deceptively simple iconographic modeling software, they might not even
know to look. And sadly, there is no sure sense of agreement on what constitutes great work in
the field.

The difficulties of identifying and locating the best practice is potentially a serious problem that
threatens the future of the field, as the best practice is swamped by the enthusiastic work of legions
of new practitioners unfamiliar with the field’s forty years of experience. We must also grant a flip
side to the problem: the difficulty of identifying and locating the best innovative practice achieved
by people who bring new perspectives, skills, and applications to the field.

The field needs a way of identifying, accumulating, and promulgating best system dynamics
practice. Efforts in this direction include the edited collections of Roberts (1980) and Richardson
(1996). Better, one could imagine an edited volume on the state-of-the-art that is compiled under
the direction of the the System Dynamics Society and is rewritten, say, every five years in a way
that updates and accumulates the field’s best practice and wisdom. A journal captures the flow of
best written practice as currently perceived; we require a way of capturing the stock of best
practice as perceived over time.5

Advancing practice
The field of system dynamics abounds in good introductions to modeling; it lacks second and
third level workshops and texts that move practitioners to the most advanced levels of
knowledge and skill.

It has been repeatedly observed that we have a number of reasonably good introductory texts in the
field, but no text goes much beyond introductory material. Yet introductory texts continue to be
produced. And unfortunately, all current texts are tied to particular simulation languages. The
impression of our published teaching materials is that there is not much to know beyond stocks,
flows, feedback loops, and software details, and whatever there is to know is tied to particular
software, not to general insights and skills about modeling.

It seems obvious that the lack of advanced texts and workshops in modeling for policy analysis is a
serious threat to the future of the field. Practitioners who manage to advance to extremely high
levels of expertise do so either in the few University contexts where advanced courses are taught or
colleagues collaborate, or they reinvent themselves and the field on their own. Reinvention is time-
consuming and fraught with peril. Stinchcomb (1968, 143) said it best: “Explicit knowledge
accessible to intelligent beginners is obviously more efficient for a science than knowledge
perceived by the intuition of its geniuses.”

5 The increasing use of the Internet for scholarly conversation and communication vividly illustrates the flow of
ideas and the difficulty of accumulating a stock of the best ones.

-4-
The solution is clear, even if untried: we require advanced modeling texts, preferably written
collaboratively to capture rich pictures of the arts of model-based policy analysis. Essential
bibliographies are a start (e.g., Sastry and Sterman 1993), but the future of the field requires the
integration of advanced skills and understanding in excellent teaching materials.

Accumulating results
How can work in the field be organized so as to accumulate and promulgate widely accepted
understandings of the structural causes of dynamic behavior in particular systems?

In addition to shared principles of wise practice, one of the hallmarks of a science is the
accumulation of shared understandings with predictive power. After forty years it is fair to
observe that the field of system dynamics has not come far in accumulating such results. For
example, research and model building in economic dynamics have been ongoing in the field since
the mid 1970s, yet how many scholars and practitioners who identify with the field can list the
structural mechanisms underlying the various business cycles? Who has replicated the work of the
MIT group in this area? Work in project management, ranging from Roberts (1964) to Abdel-
Hamid and Madnick (1991), fares somewhat better in this regard, but what is the list of dynamic
insights about project management that the field has accumulated? Work in economic development
has been ongoing for more than a decade in the field (see, e.g., Saeed 1990), but again it would be
difficult to list the insights or identify scholars who have worked to test and replicate results. On a
different scale, many of us have simulated the Lotka-Volterra equations in introductory work, yet
how many know Volterra’s principle that stems from this simple dynamic system and how many
of those can say how far the insight can be pressed? Finally, a number of dissertations have made
significant advances in particular areas of application, but have failed to become common
knowledge among scholars and practitioners in the field. (Any list of examples I would write
would prove the point two ways: by illustrating some the reader is not familiar with and by leaving
out some the reader believes should be listed but I do not know about.)

To advance beyond a craft, to approach the rigors of a science, the field of system dynamics needs
to find ways to facilitate the accumulation of reliable results. Accumulating wise practice is crucial
for maintaining high quality work in the field as practitioners enter and leave, but accumulating
wise, replicated, and accepted results is essential for growth of understanding.

The future of the field requires mechanisms that can help avoid reinventing previous work or
previous insights, and that can help scholars and practitioners to build on previous work and add to
previous insights. Part of what is required is a culture of accumulation accepted by those who
identify with the field. Much like a well formulated dissertation, every article and book that can be
linked to previous work should trace its lineage and identify the foundation that it builds upon. In
a spirit of cooperative, creative scholarship, published work should be replicated — tested in
various ways for consistency, reliability, and extensibility — and the results added to a known and
growing body of literature in each applied area.6
6 We must acknowledge that there are forces inherent in scholarly activity that work against such cooperative
creative scholarship, in particular the scholarly need to carve out areas of individual expertise that identify one as the
expert in an area. Replication of previous work that “merely” adds confidence to the work and results of others is
often not valued in the social sciences.

-5-
The accumulation of results in the field has been helped greatly by the system dynamics
bibliography available on disk from the System Dynamics Society.7 Accumulation of results is so
crucial for the future of the field that I am prompted to suggest that the bibliography ought to be a
major ongoing work of the Society and the disks should be sent free to every member of the
Society every one or two years. The bibliography is not evaluated, however; it indiscriminately
mixes great and not-so-great work which may not stand up under deep scrutiny or replication.
Perhaps the bibliography could begin to include evaluative comments by people who have
reviewed or tried to replicate the work.

The mechanisms for accumulating results could include requiring students to replicate and if
possible improve existing work and publish the results as Notes in the System Dynamics Review.

But in addition, we require more attention to the packaging of results so that they can be easily
assimilated and passed along. The systems archetypes are a brave attempt in this direction, but
more is required, particularly in formally linking the structure, dynamics, and policy insights
embedded in results in particular applied areas.

Making models accessible

How can formal models be designed, formulated, and presented so that they are accessible to
the widest possible audience?

It has been said that an advantage of formal models is that they are explicit (Forrester 1961,
Richardson and Pugh 1981). In contrast, it is said, mental models contain unstated assumptions
and leaps of logic and intuition that force important aspects to be implicit. It has also been said that
most formal models of any significance are too complex to be accessible to all but a tiny minority
of modeling experts (e.g., Greenberger, Crenson, and Crissy 1976). We must admit the latter
claim is probably true: as an illustration, recall that a technically gifted critic of Forrester’s World
Dynamics (1971) found it necessary to simplify the model to understand it sufficiently to argue his
criticisms (Nordhaus 1973). The difficulties authors of articles in the System Dynamics Review
have in presenting complex models in understandable ways provide further evidence for the
significance and extent of the problem. We must acknowledge that in practice formal models, like
mental models, have a necessary implicitness which stems from their size, complexity, and
algebraic formalisms.8

There have been instructive, even noble efforts to make complex models accessible: Forrester’s
Urban Dynamics (1969) and World Dynamics (1971), differed markedly in size and complexity
but were presented in much the same way -- important stocks and flows, examples of crucial
feedback loops, usually the positive ones, and examples of typical structure, followed eventually
7 Write to Ms. Julia S. Pugh, System Dynamics Society, 49 Bedford Road, Lincoln, MA 01773.
8 The simulation methods and tools we use undoubtedly also impose hidden assumptions. What assumptions are
hidden, for example, in the fact that our simulation languages are structured to simulate systems of the form dx/dt =
f(x,p) but do not handle easily systems involving partial derivatives of variables other than time? What
assumptions are hidden in the fact that we can not simulate a loop without a level in it or that graphical functions in
past and current software are limited to functions of one variable?

-6-
by a detailed, equation-by-equation explanation/justification.

Another approach involves the use of sector overviews keyed to model equations and coupled with
diagrams detailing important microstructure; a typical example is Sterman and Richardson (1985).
The use of sector overviews seems certain to become more widespread now that STELLA/iThink
contains an overview layer that links in a meaningful way to the layer containing detailed model
structure. Yet STELLA/iThink overviews do not easily allow the labeling of elements of structure
so they, too, quickly become complex diagrams conveying little more than gross connections
among amorphous sectors.

In a computing environment one could imagine model documentation that is organized as a

collection of hypertext and hyperdiagram documents. The reader could click to activate a hyperlink
from one level of documentation to the next, enabling an orderly exploration from overviews to
model detail at the level of equations and even to explanations of items within equations. While
clearly possible with today’s technology, it is not clear what circumstances would call for such
hypertext documentation nor what principles should guide its construction. We have only intuitive
notions of what makes technical presentations accessible, and undoubtedly what works for one
reader could fail for another.

Our intuitions about presentations that make models accessible needs to move toward a science of
model documentation. It is an important area: Inaccessible models are barriers to understanding
and will become barriers to both the growth of the field and the spread of important model-based
policy insights.

Qualitative mapping and formal modeling

What are the wise uses of qualitative mapping approaches, and what are the conditions that
require formal, quantitative modeling?

Literature in the history of feedback thought repeatedly contains the assertion that formal,
quantitative models are essential for understanding the dynamics of complex systems (see
Richardson 1991 for examples). At the foundation of the field of system dynamics is the
presumption, which seems to be repeatedly reaffirmed in the practice of applied policy modeling,
that the dynamic implications of even moderately complex circular causal structures are impossible
to discern without the aid of formal models. Sterman (1995) provides richly documented evidence
that learning in and about complex systems without simulation is, in fact, impossible.

Yet the field is experiencing the increasing use of qualitative tools — systems archetypes, word-
and-arrow diagrams under various labels (causal-loop diagrams, influence diagrams, cognitive
maps), and other approaches and techniques that fall under the general rubric of qualitative systems
thinking. The relationships between these qualitative practices and the quantitative core of the field
of system dynamics are unclear. Some hold that qualitative systems thinking tools have their place
as precursors of formal modeling . Others contend that various qualitative approaches can stand by
themselves and lead to reliable policy insights without formal, quantitative modeling.9
9 For a range of examples and opinion, see “Systems Thinkers, Systems Thinking,” special issue of the System
Dynamics Review (10,2-3, Summer-Fall 1994), G.P. Richardson, E.F. Wolstenholme, and J.D.W. Morecroft, eds.

-7-
The history of feedback thought in the social sciences (Richardson 1991) does not hold out much
hope for this latter point of view. The cybernetics movement began in the 1940s with considerable
emphasis on circular causal processes, expressed qualitatively in maps and words, but rather soon
turned its focus away from feedback loops to questions of coding, languages, and computing
machinery. Initially subtitled “control and communication in the animal and the machine,” the field
of cybernetics came to see its greatest promise in the study of communication. Feedback loops
were largely lost in the shuffle, to the extent that most modern social scientists and folk in the street
do not make use of circular causal patterns of inference. Given the difficulty of making reliable
qualitative inferences from moderately complex circular causal maps, it is reasonable to suggest
that the movement away from feedback analyses stemmed at least in part from the lack of what
Forrester later supplied — tools and techniques (and a perspective) for building mathematical
models of realistically complex circular causal systems and inferring their behavior through
computer simulation.

Armed with almost forty years of experience simulating circular causal systems, some practitioners
are embracing qualitative approaches. Their work holds promise because it is based on insights
and patterns of thought that have been formed by years of modeling experience. Senge’s system
archetypes, for example, include insights that were first written down by Forrester in Urban
Dynamics (1969) and stemmed from his model-based work up to that point. Most of the
archetypes are model-based (although many practitioners would find it difficult to build formal
models of the archetypes — see Dowling, MacDonald, and Richardson 1995).

Yet readers of this qualitative work do not know of its quantitative foundation. They are
encouraged to apply qualitative feedback maps to understanding puzzling and complex
management problems. Given the history of feedback thought in the social sciences, that trend is
worrisome. The history suggests that inferring dynamic behavior from word-and-arrow maps
involving circular causality comes only in two situations, when the map is a recognized structure
previously verified by simulation, and when the map is extremely simple. Yet modelers and
nonmodelers are encouraged by our recent qualitative trend to try to learn from the application of
new qualitative maps to new situations.

The trend toward the use of qualitative word-and-arrow maps for dynamic analysis raises several
research questions whose answers may be crucial for the future of the field. What are the system
conditions that suggest that a qualitative mapping approach can produce reliable inferences? What
are the conditions under which a qualitative mapping may yield unreliable or false inferences? Are
word-and-arrow maps showing explicit stocks and flows more reliable, although less accessible,
for various practitioners or audiences? What are the implications of packaging systems insights in
systems archetypes? Do archetypes limit or expand people’s capabilities to reason in circular
causal settings? Finally, is it possible to state conditions under which qualitative maps are best
employed and other conditions which require quantitative modeling?

The field must address the relationships between qualitative mapping and quantitative modeling —
in short, when to map and when to model. To advance in this area the field requires both
academic research and reflective, constructively self-critical practice.

-8-
Widening the base
The future of the field of system dynamics rests on our abilities to widen its base, the
population understanding the significance of feedback and circular causality in living systems.

Ever since Forrester (1971b), it has been common within the field to assert that the behavior of
complex social systems is counterintuitive. Critics have noted with considerable offense that it
sounds like system dynamics devotees are saying “counterintuitive to everyone except us.” That is
a fair criticism: deep down, the counterintuitive claim rests on the often puzzling dynamics that a
nonlinear feedback system can exhibit and which can only be reliably drawn out and understood
with the aid of simulation models of the sort that system dynamicists build. To increase the
acceptability of results from excellent and often counterintuitive system dynamics studies, it is
essential for the field to widen the base of people prepared to receive difficult messages about the
behavior of living systems.

Toward this end, two kinds of current work are exemplary: the journalism of Donella Meadows
(1989, 1991) and bringing systems thinking and system dynamics modeling to school children
(see, e.g., Brown 1992 and Systems Thinking in Education, special issue of the System
Dynamics Review 9,2(summer 1993)). Meadows’s writings are the best work I know of related
to the field of system dynamics striving to reach a lay audience. She aims not at modelers or
scholars but at thoughtful readers of all sorts and backgrounds and educational levels. She is
almost alone in this endeavor, however.10 To widen the base for system dynamics work, more
efforts like hers must appear.

Bringing systems thinking and system dynamics in accessible forms into elementary and high
schools is also a major step forward for the future of the field of system dynamics. Those at the
forefront of this effort in the United States are school teachers, aided and abetted richly by Barry
Richmond and colleagues at High Performance Systems, a few other system dynamics
professionals, and the Creative Learning Exchange in Acton, Massachusetts, under the direction of
Lees Stuntz. Scholars and practitioners in the field need to recognize the importance of these
efforts and support them actively. If extended, they promise to widen dramatically the population
ready to hear, understand, and act on messages like “worse before better behavior,” and to accept
the policy implications of “tragedies of the commons.”

These efforts to widen the base deserve the help of scholars and practitioners in the field. Far from
being diversions, these efforts to expand the dynamically wise population are crucial for the future
of the field of system dynamics.

Confidence and Validation

What are appropriate procedures and standards for establishing user confidence in system
dynamics models in various decision environments?

In 1980 Forrester and Senge published a classic statement on model validation and the
development of user confidence. Examining the triple relationships of model structure, model
10 Jørgen Randers in Norway has made similar journalist efforts but I do not know any details.

-9-
behavior, and linkages between structure and behavior, they proposed a suite of seventeen separate
tests that can be used ensemble to establish user confidence in system dynamics models.11 More
importantly, they confronted the notion that one or several simple tests, such as classic statistical
goodness of fit tests, could not serve as a single and uniform path to model validation. While a
number of specific papers have followed up on aspects of this work (see, e.g., Sterman (1984),
Barlas and Carpenter (1990), Lane (1995)), it seems sadly clear that the most comprehensive
statement on model validation and user confidence in system dynamics models is still a single
paper that is nearly twenty years old. Furthermore, Scholl’s benchmarking survey of practice in
the field suggests that many of the tests Forrester and Senge described are not used by current
practitioners.

The field must return to the central issue of model evaluation and unpack it into its several
constituent parts. System dynamics models are now being constructed to support planning and
decision making in a wide range of decision environments. Models built to support testimony in
court (e.g., Cooper 1980) should surely adhere to different confidence and validity standards than
those used to inform the strategic thinking of a small corporate board of directors. Similarly,
models that support public infrastructure planning efforts will meet a different level of scrutiny and
review than will those supporting policy development in a legislative environment (see Andersen
1991). Hence, it seems that a necessary first task in advancing our thinking about model
confidence and evaluation is to craft a fairly comprehensive taxonomy of decision environments so
that a more detailed discussion of validation procedures and tests can become more focused and
fine-grained.

As a next step, the field needs to achieve greater consensus concerning what types of confidence
building and validation procedures and tests are more appropriate in what types of decision
environments. For example, Forrester and Senge’s "surprise behavior" test or the "mistaken
identity" behavioral test may be an extremely powerful way to build confidence when working
with a small executive group who share a point of view about the problem being studied but may
have little power at all in more public decision making environments. We need to accumulate
wisdom about under which conditions do various types of tests and procedures appear to be most
appropriate.

Finally, our list of appropriate tests for building confidence in system dynamics models needs to be
expanded and software and other more automatic procedures need to be invented to implement our
most commonly used evaluation tests, such as the reality check feature of Vensim (Peterson and
Eberlein 1994). For example, Reagan-Cirincione et al. (1991) proposed a simple "complete the
graph" exercise in which a group of managers were asked to sketch the completion of an
interrupted time series simulation under various policy changes. By comparing what managers
thought would happen in the system to what the simulation predicted would happen, managers
were able to engage the relationships between model behavior and model structure (and hence
policy changes) at a deeper level. This proved to be a powerful test that could be more formally
developed and software could be easily constructed to generalize the application of a test such as
11 See Richardson and Pugh (1981, 310-320), which organizes a number of the Forrester and Senge tests into tests
focusing on structure and on behavior. The question of model validity is cast in terms of the model’s suitability for
its purpose and the problem it addresses, its consistency with the slice of reality it tries to capture, and its utility
and effectiveness. Lane (1995) begins with and significantly extends this framework.

-10-
this one.

The use of formal models to aid policy understanding requires a rich and diverse mix of procedures
to establish confidence in model-based analyses. The future of the field requires ongoing attention
to the question of model validity.

Epilogue
This note has identified eight problems for the future of the field, ranging from problems specific
to the field of system dynamics to problems of more general concern in any scientific discipline.
The identified problems and questions are:

1. Understanding model behavior: The principal outstanding technical problem in simulation

modeling is the development of tools to aid understanding model behavior.

2. Accumulating wise practice: The field of system dynamics needs improved mechanisms for
accumulating and promulgating insights about wise modeling practice.

3. Advancing practice: The field of system dynamics abounds in good introductions to

modeling; it lacks second and third level workshops and texts that move practitioners to the
most advanced levels of knowledge and skill.

4. Accumulating results: How can work in the field be organized so as to accumulate and
promulgate widely accepted understandings of the structural causes of dynamic behavior in
particular systems?

5. Making models accessible: How can formal models be designed, formulated, and presented
so that they are accessible to the widest possible audience?

6. Qualitative mapping and formal modeling: What are the wise uses of qualitative mapping
approaches, and what are the conditions that require formal, quantitative modeling?

7. Widening the base: The future of the field of system dynamics rests on our abilities to widen
its base, the population understanding the significance of feedback and circular causality in
living systems.

8. Confidence and validation: What are appropriate procedures and standards for establishing
user confidence in system dynamics models in various decision environments?

No doubt these are not the only problems in the field. Certainly, for example, the difficulties of
system conceptualization place that topic high on anyone’s list of significant problems facing both
new and experience practitioners. And although I have offered arguments for the significance of
these particular eight problems for the future of the field, I acknowledge that some may be able to
identify other, more pressing problems or in other ways extend and improve this list. In fact, I
invite readers to do just that.

-11-
But a list is a start and the battles can now be joined. For the goal of such a list of outstanding
problems for the future of the field is a concerted effort by many system dynamicists around the
world to address and solve these problems. As we continue to employ our simulation arts and a
dynamic, feedback perspective to address the most significant problems the real world presents us,
let us also address these outstanding problems our field presents us.

-12-
References

Abdel-Hamid, T. K., and S. E. Madnick. 1991. Software Project Dynamics. Englewood Cliffs,
New Jersey, USA: Prentice Hall. Aracil, J. 1981. Structural Stability of Low-Order System
Dynamics Models. International Journal of Systems .
Andersen, D. F. 1990. Analyzing who gains and who loses: the case of school finance reform in
New York State. System Dynamics Review 6,1 (winter): 21-43.
Aracil, J. 1981. Structural Stability of Low-Order System Dynamics Models. International
Journal of Systems.
Barlas, Y., and S. Carpenter. 1990. Philosophical roots of model validation: two paradigms.
System Dynamics Review 6(2): 148-166.
Brown, G. S. 1992. Improving education in public schools: innovative teachers to the rescue.
System Dynamics Review 8(1): 83-90.
Cooper, K. G. 1980. Naval Ship Production: A Claim Settled and Framework Built. Interfaces
10(6): 20-.
Dowling A.M., R.H. MacDonald, and G.P. Richardson. 1995. Simulation of Systems
Archetypes. Proceedings of the 1995 International System Dynamics Conference, Tokyo.
Lincoln, MA: System Dynamics Society.
Eberlein, R.L. 1984. Simplifying Dynamic Models by Retaining Selected Behavior Modes.
Ph.D. dissertation, Sloan School of Management, M.I.T., Cambridge, MA 02139.
Ford, A., and P. Gardiner. 1979. A New Measure of Sensitivity for Social System Simulation
Models. IEEE Transactions on Systems, Man and Cybernetics (March).
Forrester, J. W. 1961. Industrial Dynamics. Cambridge MA: MIT Press; reprinted by
Productivity Press, Portland, Oregon.
Forrester, J. W. 1968. Market Growth as Influenced by Capital Investment. Industrial
Management Review. (MIT) 9(2): 83-105.
Forrester, J. W. 1969. Urban Dynamics. Cambridge MA: MIT Press; reprinted by Productivity
Press, Portland, Oregon.
Forrester, J. W. 1971a. World Dynamics. Cambridge, MA: MIT Press; reprinted by
Productivity Press, Portland, Oregon.
Forrester, J. W. 1971b. Counterintuitive Behavior of Social Systems. Technology Review
73(3): 52-68.
Keough, Mark and A. Doman. 1992. The CEO as Organization Designer: An Interview with
Professor Jay W. Forrester, the Founder of System Dynamics. McKinsey Quarterly 2: 3-30.
Forrester, J. W., and P. M. Senge. 1980. Tests for Building Confidence in System Dynamics
Models. In A. A. Legasto Jr. and e. al., eds. System Dynamics, TIMS Studies in
Management Science, vol. 14: 209-228. New York: North-Holland.
Forrester, N. 1982. A Dynamics Synthesis of Basic Macroeconomic Theory: Implications for
Stabilization Policy Analysis. Ph.D. dissertation, Sloan School of Management, M.I.T.,
Cambridge, MA 02139.
Homer, J. B. 1983. A Dynamic Model for Analyzing the Emergence of New Medical
Technologies. Ph.D. dissertation, Sloan School of Management, M. I. T., Cambridge, MA.
Lane, D. 1995. The Folding Star: A Comparative reframing and Extension of Validity Concepts
in System Dynamics. Proceedings of the 1995 International System Dynamics Conference:
Vo. I, Tokyo. Lincoln, MA: System Dynamics Society.
Meadows, D.H. 1982. Whole Earth Models and Systems. Coevolution Quarterly (Summer): 98-

-13-
 108.
Meadows, D.H. 1989. System Dynamics Meets the Press. System Dynamics Review 5(1): 68-
80.
Meadows, D.H. 1991. The Global Citizen. Washington, D.C.: Island Press.
Mojtahedzadeh, M. 1995. Structural Analysis of the Dynamics of URBAN1. Presentation at the
1995 International System Dynamics Conference, Tokyo, Japan, August 1995. Available
from the author, Rockefeller College, University at Albany – SUNY, Albany, NY 12222.
Morecroft, J. D. W. 1985. Rationality in the Analysis of Behavioral Simulation Models.
Management Science 31(7): 900-916.
Mosekilde, E., S. Rasmussen, H. Joergensen, F. Jaller, and C. Jensen. 1985. Chaotic Behavior
in a Simple Model of Urban Migration. In Proceedings of the 1985 International Conference of
the Systems Dynamics Society, (pp. 575). Keystone, Colorado: International System
Dynamics Society.
Nordhaus, W. D. 1973. World Dynamics: Measurement Without Data. The Economic Journal
83(332): 1145-1183.
Peterson, D. W. and R. L. Eberlein. 1994. Reality check: a bridge between systems thinking and
system dynamics. System Dynamics Review 10,2-3: 159-174.
Reagan-Cirincione, P., S. Schuman, [Link], and S. Dorf. 1991). Decision Modeling:
Tools for Strategic Thinking. Interfaces 21,6 (Nov-Dec 1991).
Richardson, G. P. 1984/1995. Loop Polarity, Loop Dominance, and the Concept of Dominant
Polarity. System Dynamics Review 11,1 (spring 1995).
Richardson, G. P. 1986. Dominant structure. System Dynamics Review 2(1): 68-75.
Richardson, G. P. 1991. Feedback Thought in Social Science and Systems Theory.
Philadelphia: University of Pennsylvania Press.
Richardson, G.P. 1996. Modelling for Management: Simulation in Support of Systems
Thinking. Aldershot, England: International Library of Management; Dartmouth Publishing
Co. Ltd.
Richardson, G.P. and A.L. Pugh III. 1981. Introduction to System Dynamics Modeling with
DYNAMO. Cambridge: MIT Press; reprinted by Productivity Press, Portland, Oregon.
Roberts, C. A., and B. C. Dangerfield. 1990. Modelling the Epidemiological Consequences of
HIV Infection and AIDS: a contribution from Operational Research. Journal of the
Operational Research Society (UK) 41(4): 273-289.
Roberts, E. B. 1964. The Dynamics of Research and Development. New York: Harper and
Row.
Roberts, E. B. 1980. Managerial Applications of System Dynamics. Cambridge: M.I.T.
Press; reprinted by Productivity Press, Portland, Oregon.
Roberts, N., D. F. Andersen, R. M. Deal, M. S. Grant, and W. A. Shaffer. 1983. Introduction
to Computer Simulation: The System Dynamics Modeling Approach. Reading, MA:
Addison-Wesley; reprinted by Productivity Press, Portland, Oregon.
Saeed, K. 1990. Towards Sustainable Development: Collected Papers by Khalid Saeed.
Bangkok: IE & M, Asian Institute of Technology.
Sastry, A. and J.D. Sterman. 1993. Desert Island Dynamics: An Annotated Survey of the
Essential System Dynamics Literature. Proceedings of the 1993 International System
Dynamics Conference, Cancún, Mexico. Lincoln, MA: System Dynamics Society: 466-475.
Senge, P. M. 1990. The Fifth Discipline: the Art and Practice of the Learning Organization.
New York: Doubleday/Currency.

-14-
Sterman, J. D. 1984. Appropriate Summary Statistics for Evaluating the Historical Fit of System
Dynamics Models. Dynamica 10(2): 51-66.
Sterman, J. D. 1989a. Misperceptions of Feedback in Dynamic Decision Making. Organizational
Behavior and Human Decision Processes 43(3): 301-335.
Sterman, J. D. 1989b. Modeling Managerial Behavior: Misperceptions of Feedback in a Dynamic
Decision Making Experiment. Management Science 35(3): 321-339.
Sterman, J. D. 1994. Learning In and About Complex Systems. System Dynamics Review
10,2-3: 291-330.
Sterman, J. D., and G. P. Richardson. 1985. An Experiment to Evaluate Methods for Estimating
Fossil Fuel Resources. Journal of Forecasting 4(2): 197-226.
Stinchcomb, A.L. 1968. Constructing Social Theories. New York: Harcourt, Brace, and
World.
Wolstenholme, E. F., and R. G. Coyle. 1983. The Development of System Dynamics as a
Methodology for System Description and Qualitative Analysis. Journal of the Operational
Research Society (UK) 34(7): 569-581.

-15-