Table of Contents
MODEL C ONCEPTUALIZATION IN GROUP MODEL BUILDING:
A REVIEW OF THE LITERATURE EXPLORING THE TENSION BETWEEN
REPRESENTING REALITY AND NEGOTIATING A SOCIAL ORDER
Aldo A. Zagonel
Doctoral student - University at Albany, SUNY
Rockefeller College of Public A ffairs and Policy
11 Pheasant Ridge Dr. - Loudonville, NY 12211
Phones: (518) 439-4183 and 446-9651
E-mail address: zagonel@ aol.com
Proceedings of the 2002 International Conference of the System Dynamics Society.
Palermo, Italy, July 28-August 1. (Draft dated May 2002)
There is a growing practice of building system dynamics models directly with groups. This paper
traces a genealogy of group model building (GMB) along two stream of thought. It focuses upon
exploring the tension between modeling as a representation of reality, and modeling as a tool for
negotiating a social order. The literature is organized into five clusters that roughly represent
the members of a genealogy tree. A description of GMB is developed to fit an ideal type
conceptual dichotomy. Findings are summarized in tables, mostly quoting directly from surveyed
authors. The paper offers supporting evidence to the thesis that there are two intertwined threads
in the group approach to system dynamics modeling. GMB interventions strive both to create a
shared understanding of an interpersonal or inter-organizational problem, in the form of a
“boundary-object” model, and to build a “micro-world” type model that is useful in terms of
organizational redesign.
Key words: Group model building, system dynamics, decision conferencing, competing values
approach, problem identification, problem definition, system conceptualization,
micro-world, boundary- object.
Introduction!
Approaches to systems thinking (Richmond 1987/97, Morecroft and Sterman 1994, Richardson
et al. 1994-A, Kim and Senge 1994), strategic planning (Eden 1989, Carper and Bresnick 1989,
Quaddus et al. 1992, Bryson 1995), decision analysis (Adelman 1984, Buede and Bresnick
1992), decision support (DeSanctis and Gallupe 1987, Phillips 1988, Vennix et al. 1992/94), and
decision conferencing (Weiss and Zwahlen 1982, Reagan et al. 1991, Schuman and Rohrbaugh
1991) are increasingly coming to rely upon the practice of building models directly with
management teams and decision-making groups. The objectives of these researchers and
practitioners are manifold, ranging from improving group decision-making processes to
enhancing group, team and organizational effectiveness and productivity (Andersen et al. 1997).
'T’d like to thank David Andersen, George Richardson and John Rohrbaugh for introducing me to the intricacies of
system dynamics, decision conferencing and group model building, and for their continued support. The opportunity
to work as a student, and hands-on apprentice, with these wise masters and generous friends has been of
immeasurable value to me. To David my special thanks for suggesting that I structure my confusing thoughts as a
dichotomy. Hopefully my research theme has become a little more tangible as a result of writing this paper.
The importance of involving the clients in the process of model building has been
recognized early on in the field of system dynamics. Forrester (1961) stated that the power of
system dynamics lies in the ability to use information obtained from the clients, and to portray
more usefully its implications (p. 117). He also emphasized the relevance of the clients in terms
of establishing model validity, as a measure of confidence in the model’s correspondence to the
clients’ actual system (Chapter 13). Roberts (1978-B, 1978-C) made more explicit claims
through a series of recommendations, such as: realize an opportunity important to the client,
maximize in-house involvement as a means to secure implementation, and gear tests of validity
to clients’ assurance criteria, among other thoughts on the significance of client involvement.
A recent development in the field of system dynamics involves more active client
engagement especially but not exclusively in the conceptual phase of model building, in the form
of group meetings or conferences. This line of research and practice has been termed group
model building (Richardson et al. 1992, Vennix et al. 1997). Richardson (1999) defines it as “the
processes and techniques designed to handle the tangle of problems that arise in trying to involve
a large number of people in model construction” (p. 375). Vennix (1996) characterizes it as a
kind of group decision support for helping teams tackle strategic problems (p. xi). Gradually, a
unified body of knowledge containing methodological guidelines to develop group model
building procedures is flourishing (Richardson and Andersen 1995, Vennix 1996, Andersen and
Richardson 1997). As they are experimented with, these procedures are also being extensively
evaluated and tested (Rouwette et al. 1999, 2002).
While group model building is based essentially in the system dynamics model building
method, deeply involving a client group in the process of model construction has required
theoretical and applied input from other fields, such as sociology, social psychology, and small-
group research (Vennix 1999, p. 379). In the applied-research in group model building that is
being conducted in Albany, this influence has been filtrated and boosted in terms of a framework
called decision conferencing (Rohrbaugh 2000).
While one could probably apply group model building to modeling any sort of dynamic
problem, it seems that particular kinds of problems or situations “attract” the application of this
approach. Group model building interventions will often address problems that involve multiple
stakeholders that contribute with partial views of the system, but who are affected by the system
as a whole (Huz et al. 1997, Rogers et al. 1997). They are also particularly useful in situations in
which there is strong inter-personal disagreement in the client group, regarding the problem
and/or regarding the policies that govern system behavior. Vennix (1999) refers to the latter as
messy problems, i.e., “a situation in which opinions in a management team differ considerably”
(p. 379).
Thus, this approach to system dynamics modeling, referred to as group model building, is
a result of specific, and probably identifiable, idiosyncrasies. These are related both to the
theoretical contributions from the several domains of knowledge that are shaping it (system
dynamics, small group research, etc.), and to the nature of its application (messy problems, inter-
organizational problems, etc.). If we were to think of group model building as an entity, we
might choose to understand it in terms of its background (or genealogy) and nature (or
personality). This is the intent of this literature review. To understand and describe group model
building, both in terms of its ancestry and persona.
Tracing a genealogy of group model building
The genealogy of group model building will be traced from the point of view of the approach
used by a research group working at the University at Albany. There are at least two schools of
thought contributing to group model building practice in Albany, as illustrated in Figure 1.
thread Decision thread
Servomechanisms
Decision
engineering Connpuler toon itp enalysis
ability dynamics Decision
Sg support
G) Micro
System computing ———» Decision G3)
dynamics conferencing
@) Portable computing ()
Direct SD modeling and client friendly sp modeling used in
with clients | decision conferences
Group model
G) building
Figure 1. Tracing a genealogy of group model building.
System dynamics” is at the root of the policy stream. The system dynamics model
building method can be described in phases that begin with a clear definition of the problem of
interest, and end with a conclusive statement about this problem, containing policy
recommendations aimed at its solution or mitigation (Richardson and Pugh 1981, pp. 15-17).
This method is based upon an endogenous feedback view of system causes and effects. Solutions
to the perceived problem are revealed through feedback thinking, the key expertise offered by
system dynamicists (Forrester 1961, Sterman 1994).
° The Encyclopedia of Operations Research and Management Science contains an elaborate statement defining and
explaining the system dynamics method, tracing its roots to servomechanisms engineering (Richardson 1996-A).
Simply stated, Richardson claims that “system dynamics is a computer-aided approach to policy analysis and
design” (p. 656).
The second stream, called the decision stream, is formed by a confluence of schools that
gave shape to the decision conferencing framework. Those are group dynamics, decision analysis
and decision support (Rohrbaugh 2000). People who conduct decision conferences consider
themselves technique/process experts, and they focus upon the appropriate techniques and the
best processes used to arrive at decisions (Reagan and Rohrbaugh 1990, Reagan et al. 1991,
Nunamaker et al. 1991). They help structure problem solving while focusing upon facilitation
and elicitation strategies and techniques (Phillips and Phillips 1983, Moore and Feldt 1993,
Griffith et al. 1998).
A purposive sample of the literature was chosen to represent each of the five clusters
depicted in Figure 1:
1. Classic system dynamics
2. Direct system dynamics modeling with clients
3. Decision conferencing
4. System dynamics modeling used in decision conferences
5. Group model building
These key references are listed in the Appendix, by cluster, in chronological order. These
references not only help us understand the origins of group model building, but they also serve as
the source of information for revealing and understanding its main features and characteristics.
Revealing the characteristics of group model building
The first major characteristic of group model building is its diversity in objectives and
expectations, resulting most likely from the confluence of the diverse influences giving shape to
it (system dynamics, small-group dynamics, decision support, etc.). A superficial examination of
its genealogy alone will reveal a tension between policy versus decision, and between content
versus process. To some extent these tensions overlap.
The decision conferencing influence emphasizes a decision to be made, and focuses upon
the processes that lead up to this decision. Decision or process oriented objectives in group
model building may be stated as accelerating a management team’s work (Vennix et al.
1992/94), problem structuring and classification schemes (Eden et al. 1983; Vennix et al. 1988,
1990), generating commitment to a decision (Rohrbaugh 1992, Vennix et al. 1993), creating a
shared vision and promoting alignment (Huz et al. 1997), and creating agreement or building
consensus about a policy or decision (Winch 1993, Vennix 1994)
Alternatively, policy or content oriented objectives may be stated as improving shared
understanding regarding the system or problem at hand (Eden and Ackermann 1992, Bryson and
Finn 1995), system improvement, and system process and outcome change (Richmond 1987/97,
Cavaleri and Sterman 1997). These involve changing the mental models of individuals in the
group or organization, guided by insights produced using the modeling tools and methods
(Richardson and Senge 1989).
Ideally, as Eden (1990) appropriately points out, astute analysis (content) and skillful
facilitation (process) should be combined: “within the context of group decision support it may
be suggested that the two skills can become integrally tied together so that they are fully
interdependent” (p. 49). In the context of group model building, this may be stated as promoting
organizational learning (Senge 1990, Vennix and Scheper 1990, Morecroft and Sterman 1994)
and organizational change (Akkermans et al. 1993, McCartt and Rohrbaugh 1995, V ennix et al.
1996), or promoting collaboration and cooperation amongst interdependent stakeholders
(Kraemer and King 1988).
The combination of adequate analytical tools that appropriately address the content of a
problem, with careful facilitation and elicitation procedures, resulting in an effective
intervention, is at the heart of the decision conferencing framework. Nevertheless, it involves a
great deal of tension reflected in pursuing competing values (Reagan and Rohrbaugh 1990).
However, Forrester (1987-B) warns that emphasis on decision-making can obscure attention that
ought to be placed in policy-making:
A number of “obvious truths” seem to have been accepted in varying degrees as
the philosophical guidelines for much of the search for a scientific foundation
underlying management and economics. All of the following appear to be given at
least some credence, and all seem to me to be misleading: ... That emphasis in
models should be on decision making. The sharp distinction between policy and
decision has been obscured. Too much attention has been concentrated on the
individual decisions and not enough on the policy that governs how the decisions
are made. Models ... should be directed toward policy. In other words, what are
the rules by which information sources are converted into a continuous flow of
decisions? (p. 159).
Much can be learned from contrasting these two roots of group model building. But, for
the purpose of this paper, I’ve chosen to probe the existing tensions in group model building in
terms of a dichotomy between building models to represent a reality as opposed to building
models to construct a socially negotiated order. Whereas the genealogy of group model building
reveals much insight about its origins and contributing traces, I believe contrasting these two
views of model building will be even more revealing in terms of its present characterization.
A dichotomous view of group model building
My thesis on group model building is that it is a multithread approach to team learning, decision
making, and policy change. While there is a host of technologies and techniques that give shape
to the group model building portfolio, analyzing them separately does not necessarily yield the
best understanding of the method. Instead, I propose examining it from the point of view of a
simple dichotomy. On the one hand, the model built by the group is perceived as a “micro-
world”? representation of reality. On the other, it is understood to be a “boundary-object”® to
arrive at a negotiated view of the group’s social order. Table 1 distinguishes conceptually these
two proposed views of the model.
3] use these terms metaphorically, without much concern regarding their meanings as social-scientific concepts.
Table 1. “Definition” of the dichotomous view of models in group model building.
Models as “micro-worlds”: Models as “boundary- objects”:
Problems are preexistent in the system we’ re Problems emerge from debate and discussion.
modeling. We do our best work to get the We do our best work to come upon a shared
important elements of the problem and the understanding regarding what we think our
facts right, to create a realistic representation of | problems are, and how we might best tackle
this policy system, and to accurately address them. We strive to understand our
the content of the issue we’ re modeling. We complementary and sometimes competing
strive to find the “correct” solution to the views, to build ajoined picture that we can
problem. We're focused upon the results and understand and share. We're concerned with
outcomes of this group model building reconciling our different views and opinions so
intervention, in terms of the answers it will that we may proceed toward a better solution to
provide to the questions we have about our our problem. The process we use to “negotiate”
reality. Therefore, our group process needs to | this model is as important, if not more
be effective at getting at the answers we need. | important, than the accuracy of the model as a
We need clarity in both purpose and problem _| representation of our reality. Therefore, our
in order to proceed efficiently. group process needs to be open and fair.
While this dichotomy represents an ideal type of sorts, it is useful, however, to examine
the higher purposes for which models are used in group model building. Furthermore, it helps us
to understand how the multiple technologies and techniques are combined, and more or less
emphasized, in the process of building models with groups. This dichotomy artificially separates
the pursuit of truth from consensus building. It servers to highlight many of the tensions found in
group model building theory and practice. One of the key objectives of this research is to
understand how these two threads are coming together in the form of group model building, and
the implication thereof. I argue that good group model building involves understanding and
balancing these two views. Ideally, group model building interventions will result in consensual
learning, most commonly referred to as team learning. The alternatives are mistaken consensus,
or groupthink*, and insightful models that have little or no impact in the lives of people.
In this paper, I will survey the literature identified in the genealogy of group model
building, focusing upon two phases of the system dynamics model building method: problem
identification and definition, and model conceptualization. I will map this literature into the
proposed dichotomy, and I will argue that there is a close fit. Prior to that, I will introduce the
competing values approach framework, distinguish the phases of model building, and provide an
illustration of the dichotomy.
‘ “Where members of a group mutually reinforce their current beliefs, suppress dissent, and seal themselves off from
those with different views or possible disconfirming evidence (Janis 1982)” (Sterman 2000, p. 33).
The competing values approach to group decision process effectiveness
There exists a theoretical foundation related to the proposed dichotomous view of models in
group model building. It can be extracted from the competing values approach to group decision
process effectiveness, found in the decision conferencing literature (Quinn and Rohrbaugh 1983,
Quinn et al. 1985, McCartt and Rohrbaugh 1989, Rohrbaugh 1989, Rohrbaugh and Eden 1990,
Reagan and Rohrbaugh 1990, Rohrbaugh 1992, McCartt and Rohrbaugh 1995).
Although the proposed dichotomy was conceived independently, and not derived from
this existing work, the process of reviewing the literature denounced its pertinence to the
dichotomy between the “micro-world” and “boundary-object” views of group model building.
Furthermore, I found a close parallel to the dichotomy in one specific article (Quinn et al. 1985).
This section will provide a short review of the theoretical and empirical basis of the competing
values approach framework, and refer to other similar frameworks found in the organization
behavior and theory literature. It will also illustrate the parallel between the dichotomy that I’m
creating, and the one discussed in Quinn et al. (1985). Finally, I’ll briefly comment on the
usefulness of this framework in terms of evaluating the effectiveness of decision conferences, as
well as group model building interventions.
THE COMPETING VALUES APPROACH FRAMEWORK. The competing values approach (CVA) isa
theoretical framework to organizational analysis that has been empirically uncovered, and
confirmed, through the factor extraction statistical method of analysis (McCartt and Rohrbaugh
1989, p. 246). The theory proposed that there are four models of organizational analysis:
an open systems model focuses on flexibility and readiness as the means by
which resource acquisition and growth can be increased as primary
organizational objectives;
arational goal model focuses on planning and setting objectives as the means
by which productivity and efficiency can be improved ... ;
an internal process model focuses on information management and
coordination as the means by which stability and equilibrium can be
developed ... ;
a human relations model focuses on cohesion and morale as the means by
which the value of human resources can be made greater... (p. 246)
[Emphasis added, order altered]
When this theoretical framework was applied to the process of group decision making,
factor analysis revealed, empirically, four corresponding perspectives conceming the
effectiveness of group decision processes. Figure 2, copied from McCartt and Rohrbaugh (1995,
p.574), contains a synthesis of the results. While the theory drew a parallel with Parson’s (1959)
four functional prerequisites of any system of action, the empirical results mirrored Taggart and
Robey’s (1981) four dominant decision-making styles (McCartt and Rohrbaugh 1989, pp. 246-
247). Table 2 provides a contrast between the labels given to the four quadrants in each of these
frameworks.
Figure 2. The competing values approach to group decision process effectiveness
(Copied from McCartt and Rohrbaugh 1995, p. 574)
Instrumental
CONSENSUAL PERSPECTIVE
“Pattern Maintenance”
POLITICAL PERSPECTIVE
“Adaptation
Effectivencss criteria:
internal,
Effectiveness criteria:
EMPIRICAL PERSPECTIVE
Participatory process
Supportability of decision
Effectiveness criteria:
Adaptable process
Legitimacy of decision
External
“Integration™
Data-based process
Accountability of decision |
Goal-centered process
Efficiency of decision
"Goal Attainment’
RATIONAL PERSPECTIVE
Consummetory
Table 2. The four quadrants of competing values contrasted across frameworks
(Derived from McCartt and Rohrbaugh 1989, pp. 246-247)
Empirically derided Competing values Parson’s (1959) Taggart and
perspectives on approach (CVA): theory of functional Robey’s (1981)
effectiveness of decision (Theoretical prerequisites of any decision-making
making processes: framework of models system of action: styles:
of organizational
analysis)
Political perspective Open systems model | Adaptive function Insightful style
Factor 1: Realism and
resources
Rational perspective Rational goal model | Goal attainment Logical style
Factor 2: Subjective function
rationality
Empirical perspective | Internal process Integrative function Matter of fact style
Factor 3: Information model
utilization
Consensual perspective | Human relations Pattern maintenance | Sympathetic style
Factor 4: Feelings and
social compromise
model
function (tension
management)
It is important to note that the factor extraction statistical method of analysis used by
Milter (1986) and Rohrbaugh (1987), described in McCartt and Rohrbaugh (1989) revealed,
more precisely, three dimensions (not two), resulting in eight (not four) distinct performance
criteria by which to judge effectiveness in group decision processes. The last dimension is
characterized by a distinction between ends versus means; i.e. the nature of the process versus
the ends achieved (p. 247). Therefore, as depicted in Figure 2, two criteria of effectiveness are
associated with each quadrant. In terms of ends achieved: 1) legitimacy of the decision, 2)
efficiency of the decision, 3) accountability of the decision, and 4) supportability of the decision.
In terms of means-to-an-end: 1) adaptable process, 2) goal-centered process, 3) data-based
process, and 4) participatory process.
TWO APPROACHES TO DECISION MAKING. Quinn et al. (1985) proposed a new approach to
organizational decision making, called automated decision conferencing (ADC), later to be
referred as, simply, decision conferencing. They argued that this approach allowed executives
“to integrate quantitative analysis and subjective intuition” (p. 49). They proposed this approach
as an integrative alternative to more traditional approaches to decision making that focused upon
only partial needs of decision makers:
[W]e establish a framework that clarifies the value differences between the two
most general approaches to decision making: the “hard” management science or
operations research view and the “soft” group process or organization
development view. We will then show how ADC integrates and unifies the values
reflected by these very distinct approaches. (p. 49)
In essence, in their article, Quinn et al. use the CVA framework to characterize a
dichotomy between management science and organization development. Figure 3, copied from p.
51, contains an illustration of how these two approaches would occupy the four-quadrant space
described above. There is an obvious parallel between the dichotomy illustrated in this figure,
and the one that I describe in this paper. I recognize this fact. I also take advantage of it, by
drawing upon this literature throughout the paper, as a source of insight to the existing tensions
between using a model to represent a reality, and as an instrument to negotiating a social order.
For instance, the authors indicate that the criteria for effectiveness in any given quadrant
“tend to complement somewhat the criteria in neighboring quadrants” but “stand in sharp
contrast to criteria in the opposite quadrant” (p. 50). They also argue that there are several
reasons why some quadrants may be more or less emphasized than others: 1) disciplinary and
methodological biases, 2) personal values, and 3) situational demands (p. 51). With respect to the
latter, they hypothesize:
When time pressures are high, little emphasis will be placed on the consensual
and empirical approaches. Instead, emphasis will shift to [political and rational]
criteria... When time horizons are long, the opposite shift may occur. When
uncertainty is high, tightly regulated, analytical methods are less likely to be
used... When certainty increases, the emphasis will shift toward more empirical
and rational approaches. (p. 51)
Flexible,
Implicit,
Collective
Consensual Supportability
Perspective of decision
Adaptable Political
process Perspective
Organization
Participatory Development
process
Legitimacy
of decision
External focus.
Less information,
Greater speed,
Concern with impact
Internal focus,
More information,
Less speed,
Concern with process ~
Data-based Management a
process Science ¢ of decision
Empirical ieeulc int Rational |
Perspective Accountability I Goal-centered Perspective
of decision | process
Regulated,
Explicit,
Individual
Figure 3. Two approaches to decision making
(Copied from Quinn et al. 1985, p. 51)
Similarly to these authors, this paper describes a vision for group model building that also
integrates these competing values. The contrast I describe is between the “hard” approach to
model building represented by the “micro-world” view, and the “soft” approach represented by
the “boundary-object” view.
The CVA framework has been extensively used in terms of evaluating the effectiveness
of decision conferences (Reagan and Rohrbaugh 1990, Rohrbaugh 1992, McCartt and
Rohrbaugh 1995). The sort of evaluation proposed is based upon assessing the processes, not the
outcomes, of interventions (Rohrbaugh 1989). It has also been used to “understand the mix of
method, consultant style, and client setting that in combination define [the] ‘ways of working’”
between consultants and clients (Rohrbaugh and Eden 1990, p. 40). It is important to consider
this knowledge base, when attempting to improve the effectiveness of group model building.
The phases of the system dynamics modeling method
Several authors have found useful to describe the system dynamics method in terms of its phases
(see Table 3). Andersen and Richardson (1979/80) proposed a seven-phase iterative’ process
consisting of both conceptual and technical phases. This framework was also adopted in
Richardson and Pugh (1981), and Roberts et al. (1983). Except for some small variations, these
phases can be specified as: 1) problem definition, 2) system conceptualization, 3) model
5 Randers (1980-B) afirms that “no amount of prior lessons will transform modeling into a sequential execution of a
set of activities requiring no repetition” (p. 130). See also Homer (1996).
formulation, 4) model behavior, 5) model evaluation, 6) policy analysis, and 7) model use or
implementation.°
Table 3. Phases, stages or steps of the system dynamics model building method.
Andersen and
Richardson (1979/80, p.
93)
Richardson and Pugh
(1981, p. 16)
Roberts et al. (1983, p.
8)
Sterman (2000, p. 87)
Problem recognition
Problem identification
and definition
Problem definition
Problem articulation
(boundary selection)
System
System
System
Dynamic hypothesis
conceptualization conceptualization conceptualization
Model representation Model formulation Model representation
Formulation
Analysis of model r
Model behavior . Model behavior
behavior
Model evaluation Model evaluation Model evaluation Testing
Policy formulation
Policy analysis (design) and evaluation
Policy analysis
Policy analysis and
model use
Decisions
Model use or e
Model use . (organizational
implementation 2
experiments)
Sterman (2000) uses a slightly different framework, based upon five phases only: 1)
problem articulation, 2) dynamic hypothesis, 3) formulation, 4) testing, and 5) policy formulation
and evaluation (p. 87). Except for the added emphasis in developing a dynamic hypothesis, as
opposed to system conceptualization in general, Sterman’s approach simply collapses model
formulation and model behavior into the formulation phase. Implicit in Sterman’s framework is a
sixth phase called “decisions”, where the results of the modeling effort are to be implemented (p.
88).
In this paper, I will distinguish the phases of the system dynamics modeling method as
follows:
Problem identification and definition
Model conceptualization
Model formulation and simulation
Model testing and evaluation
Poo ho
5 Andersen and Richardson (1980), p. 93; Richardson and Pugh (1981), p. 16; Roberts et al. (1983), p. 8.
5. Model based problem analysis and policy experimentation
6. Understanding and discernment
7. Policy implementation (action) and outcomes
This framework also collapses model formulation and model behavior into a single phase.
However, it makes explicit the transition from the modeling work to model use or
implementation by including a new phase called “understanding and discernment”. As a matter
of personal choice, these are the seven phases that will be used to hold the discussion regarding a
dichotomous view of models in group model building, beginning with the inspection of 1) the
problem identification and definition phase, in the next section, and 2) the model
conceptualization phase, in the second half of this paper. Figure 4 illustrates this alternative view
of the system dynamics model building method.
Action and
outcomes
Understanding
> and
discemment
Quantitative inquiry
Model based problem
-—> analysis and policy
experiments
Model testing
[ and evaluation “*
Model formulation
and simulation
-—_»-
Model
i conceptualization
~~
Problem
identification and~«q——I
definition
Qualitative reflection
Figure 4: Steps of the system dynamics model building method.
This figure depicts the phases of the method as a sequence of iterative steps, as in
climbing up and down a ladder. The first two steps have to do with a qualitative reflection
involving problem definition and model conceptualization. The next three steps relate to a
quantitative inquiry based upon model formulation and simulation, model testing and evaluation,
and model based problem analysis and policy experimentation.’ The iteration happens both
7 Randers (1980-B) distinguishes these two clusters (qualitative reflection v. quantitative inquiry) in terms of model
conceptualization v. formulation (p. 130): “The goal of the conceptualization stage is to arrive at a rough conceptual
within each cluster of steps, and across clusters, as desired or needed. At any point in the
process, there exists some degree of understanding and discernment regarding the problem and
the system under study.
It is assumed that as one climbs toward the higher steps, from qualitative analysis to
quantitative inquiry, and from formulation to testing, to model based analysis, the level of
understanding and discernment improves and gains accuracy." At some point in this process, if
the model building effort is to be successful, the insights generated will result in decisions and
actions in the form of new policy implementation. Those, in turn, will lead to new outcomes.
A detailed list and discussion of the specific ingredients of each phase (or step) involved
in the system dynamics process can be found in several sources.’ A selected set of these
ingredients will be addressed in the discussion of the dichotomy to follow. Table 4 is a creative
illustration of the dichotomy as I expect it to unfold across all phases of the model building
method.
1. The dichotomy in problem identification and definition
The first step to building a system dynamics model is problem identification and definition. In
this phase of the model building process, several important elements of the model building effort
need to be addressed. Some of these are: a) identifying the problem/issue to be modeled; b)
establishing the purpose of the modeling effort; c) specifying the audience interested in the
results of the work (sometimes a client); d) revealing the time-horizon involved in the unfolding
of the problem, and in the quest for a solution to it; e) identifying the key variables; and f)
eliciting or otherwise obtaining reference modes for the problem variables (Richardson and Pugh
1981, Chapter 2; Randers 1980-B; Sterman 2000, pp. 89-94). Key to understanding the
dichotomous view of models in group model building are the issues related to establishing the
focus of the intervention: identifying the problem to be modeled, the purpose of the modeling
effort, and the audience (or interested or otherwise affected parties).
Tables 5a, 5b and 5c, found attached in the Appendix, lay out the organization of this
discussion, and contain an epitome of my findings while mapping the literature into the
dichotomy, mostly quoting directly from the authors surveyed. The left-hand column portrays the
micro-world view, inherited from the system dynamics tradition. The right-hand column portrays
the boundary-object view, extracted from the group model building literature, and from the
literature in system dynamics modeling used in decision conferencing. The elements in the left-
hand column can also be found in the group model building literature, but I chose to quote from
the original authors. Thus, both sides of the dichotomy can be found in the group model building
literature.
model capable of addressing a relevant problem. The formulation stage should embrace two processes: the test of
the dynamic hypothesis ... and model improvement...” (pp. 130-131).
® The extent to which qualitative analysis alone can lead to understanding and discemment or, alternatively stated,
the extent to which quantitative inquiry is essential, is a highly controversial topic in system dynamics. The
arguments and counter-arguments in this ongoing discussion are most recently summarized in Coyle (2000, 2001)
and Homer and Oliva (2001).
° Richardson and Pugh (1981), Chapters 2, 4, 5 and 6; Roberts et al. (1983), pp. 8-10; Sterman (2000), pp. 85-104.
Table 4. A creative illustration of the dichotomous view of models in group model building.
Question: How do intervenors and participants view the model they are building?
Steps of the SD method:
Model as “micro-worlds”:
Model as “boundary- objects” :
. Problem identification and
definition
Monolithic client
Preexisting problem
The modeling purpose is
= Multiple constituencies
= Socially constructed
problems
= Multiple purposes, starting
How can we fix it?
to identify and solve a with negotiating a shared
problem view
. Model conceptualization Getting at the facts = Agreeing upon “reality”
Envisioning the causal = Model is a synthesis of the
feedback structures group’ s negotiated view of
capable of reproducing the “reality” (issues of scope
problematic behavior and level of aggregation)
. Model formulation and Build a quantifiable model |= Should we even bother
simulation and test the dynamic building a quantifiable
hypothesis model?
Modeler’s ownership of = Group’s ownership of the
the model model should not be
threatened
. Model testing and Organized approach to = Group judges model for
evaluation model testing and structural and behavioral
evaluation correspondence, mostly in
terms of face-validity
Modeler is free to review |= Significant changes in
and adjust model conceptualization
conceptualization and and formulation need to be
formulation checked with the group
. Model based problem Structural analysis of the |= Strategic analysis of
analysis and policy problem interrelated problems
experimentation Experimentation with new |= Experimentation with
causal structure and/or alternative strategies and
decision rules scenarios
. Understanding and What's causing the = Do we agree on the
discemment problem? problem? Do we share a
view of the system?
= Are we ready to make a
decision
. Policy implementation
(action) and outcomes
Structural change
Change resulting from
“new” understanding
regarding relationship
between structure and
behavior
= Changes in goals,
objectives and strategy
= Change resulting from
agreements in goals,
objectives and strategies
Tables 5a, 5b, and 5c will be discussed in the following order. First, I will address the
micro-world view (left-hand column). I extract two related underlying assumptions from it. First,
that while problems can be complex, they are still “preexistent” and can be clearly specified.
Second, we know or we can find out sufficient information about the problem to model it.
Sterman (2000) argues that while natural and human systems have high levels of dynamic
complexity (p. 21), the most complex behaviors normally arise from the interactions among the
components of the system, and not from the complexity of the components themselves (p. 12).
Forrester (1987-A) proposes that the components of the system (causal structure and decision
policies) can be reliably extracted from the mental database of the people who experience the
system, and from other available information. Moreover, he states that, from the mental database,
consensus usually emerges that is useful and sufficiently correct (p. 144).
The field of system dynamics contains countless examples of complex issues that have
been successfully modeled, many of which addressing complexity in social systems (Forrester
1961, Part III; Forrester 1969, 1971-B; Roberts 1978-A; Meadows et al. 1992; Richardson 1996-
B; Ford 1997; Sterman 2000, Chapter 2; to cite but a few).!° But, can it be assumed that these
two assumptions are always true? For example, Richardson and Senge (1989) contrast two
independent system dynamics studies in which the rising costs of liability insurance are modeled.
In one case, while a sophisticated model was built to assess the effectiveness of alternative policy
options, the problematic behavior driving the system could not be endogenously modeled.
Because, “at the aggregate level of regulatory politics, no one [was] confident they [knew] why
settlement awards [were] growing at 20-to-25 percent per year” (p. 16).
The boundary-object view will be discussed second. In it I’ll survey less clear and
specific problems and realities, and I’ll report on how the above-mentioned elements of the
model building process (problem, purpose, and client) might be interpreted somewhat differently
in the modeling process.
View of models as “micro-worlds”
The problem
Classic system dynamics offers clear guidelines as to establishing the focus of a model building
project. Forrester (1961) argues that a model should be designed to answer a specific, tangible,
and meaningful question, or set of questions (p. 449). This implies that questions should be
precisely and explicitly stated, and they should relate to real and actual phenomena. It is
impractical (and impossible) to model a system (Sterman 2000, pp.89-90). In order to build a
model, one must draw a boundary, deciding what are important elements to include in the
analysis, and leaving out non-essential elements in the system (Sterman 2000, pp. 79-80). The
choice of the problem defines this boundary (Richardson and Pugh 1981, pp. 42-43). For this
reason, system dynamicists emphasize that models should be developed to address a particular
© System dynamics models are not without criticism. A mong the most controversial work are Forrester’s Urban
Dynamics (1969) and World Dynamics (1971-B). For example, see Brewer and Hall (1973) and Nordhaus (1973).
Criticism of World Dynamics has been rebutted in Forrester et al. (1974).
problem, as opposed to modeling the system (Richardson and Pugh 1981, p. 18; Roberts et al.
1983, p. 167; Sterman 2000, pp. 79 and 89-90).
A meaningful system dynamics problem is a relevant and dynamically complex problem.
Sterman (2000) states that worthy problems are those in which the modeling work can have
lasting benefits (p. 84). Ultimately, the client and/or the audience should perceive the problem as
relevant (Stenberg 1980, p. 308; Sterman 2000, p. 85). Reagan et al. (1991) propose that the
primary strength of system dynamics models is fostering understanding of complicated
interrelationships and feedback-rich systems that make policy-problems complex (p. 62). System
dynamics applies to problems that are dynamic and arise in feedback systems (Richardson and
Pugh 1981, p. 19).
The purpose
Next to defining the problem, defining the purpose of the modeling effort is the most critical part
of the undertaking. Forrester (1961) argues that the seasoned modeler knows that a systems study
must be for a purpose if it is to be productive, pointing out that “the beginner tends to forge
ahead into detailed construction of a model before its purpose has been adequately defined” (p.
449). Richardson and Pugh (1981) add, “a model without a purpose is like a ship without a sail”
(p. 38). Classic system dynamics tends to favor understanding of key dynamics, for the goal of
improved system performance, as the fundamental purpose of building models:
The goal of a modeling effort is to improve understandings of the relationships
between feedback structure and dynamic behavior of a system, so that policies for
improving problematic behavior may be developed. (Richardson and Pugh 1981,
p. 38)
The Claims Learning Lab described in Richardson and Senge (1989) serves as a good example
of a model built for the purpose of fostering understanding of the key dynamics in insurance
claims operations, particularly the dynamics pertaining to rising insurance costs.
Thus, Sterman (2000) defines system dynamics as a method to enhance leaming in
complex systems (p. 4). With understanding comes the desire to “fix” the problem:
The goal is to improve performance of the system... The real value of the process
comes when models are used to support organizational redesign... “The goal
should be to find management policies and organizational structures that lead to
greater success” (Forrester 1961, p. 449). (Sterman 2000, pp. 80 and 84)
In general, classic system dynamics has placed its emphasis, in terms of modeling purpose, in the
goal of policy analysis and improvement, as found in, for example, Richardson and Pugh (1981).
In this tradition, the purpose is normatively clear: to identify and solve a problem.
The client/audience
The last element providing the focus of a model building effort is the client or audience
interested in the work. Classic system dynamics highlights three critical aspects related to client
involvement in specific and to the audience in general. If the work is to be done for a client, to be
effective the modeling process should be focused on the clients’ needs (Sterman 2000, p. 85).
Roberts (1978-B) recommends trying to solve a real problem that presents an opportunity
perceived as important to the clients (pp. 78-79). If it is simply a research effort, then the
audience of interest for the study must replace the client in terms of defining the purpose of the
effort (Richardson and Pugh 1981, pp. 45 and 50). So, first and foremost, the clients/audience are
essential in defining the purpose of the modeling effort, and in identifying the problem of
interest.
Second, the clients are also a very important source of information in the modeling effort.
They are the first source for both qualitative and quantitative information pertaining to the
problem. They enrich the empirical basis of the analysis, and open up channels for the exchange
of insights (Stenberg 1980, pp. 299 and 303). In the study of policy options for the Scandinavian
forestry and forest industry, in the absence of a clearly defined client, Stenberg (1980) assembled
a “reference group” as “a kind of mini-universe of the part of the real world under study” (p.
303). He also conducted additional empirical research drawing upon decision makers and outside
researchers, as well as historical records and statistics, to arrive at a “richer and more accurate
picture of those aspects of the real world that contribute to the dynamic behavior of the ...
model” (pp. 309-310).
Forrester (1987-A, 1992/94) identifies three sources of information for building system
dynamics models: the mental, the written and the numerical databases. He argues that the written
and numerical databases contain progressively less information for building a model, particularly
about causal structure and decision policies (Forrester 1994, p. 72). He suggests that the
dominant significance of information from the mental database is not sufficiently appreciated in
the social sciences (Forrester 1987-A, p. 143). He concludes that if the mental database is so
important in the understanding of social systems, then system dynamics models should draw
primarily upon the mental database to reflect knowledge of organizational policies and system
structure (Forrester 1994, p. 73).
A third reason why the clients are perceived as important actors in the modeling effort
has to do with the implementation of modeling results. Implementation was an implicit concern
in Industrial Dynamics (Forrester 1961), but Roberts (1978-B/C) made it explicit. He
summarized the importance of active client involvement not only in terms of ensuring interest in
the modeling work, and adequacy and accuracy of model formulation with respect to reality. But
also in terms of providing the basis for implementation of the recommended policy changes
derived from the modeling effort (1978-C, p. 156).
The dynamic hypothesis
System dynamicists often synthesize problem definition in the form of a dynamic hypothesis’
(Stenberg 1980, pp. 307-308; Richardson and Pugh 1981, pp. 55 and 63; Sterman 2000, pp. 94-
102). The dynamic hypothesis in a system dynamics study is a statement of the feedback
structures in a system that are hypothesized to generate or contribute to the problem behavior
(Richardson and Pugh 1981, pp. 55 and 63), as it is depicted in the reference modes (Stenberg
1980, p. 300). According to Randers (1980-B), the belief that the basic feedback structure can
actually reproduce the reference modes remains an assumption until the model is formulated and
simulated, and the output of the simulation proves the dynamic hypothesis to be correct - that is
the actual behavior of the model resembles the reference modes (p. 131 and 134).
Sterman (2000) feels so strongly about the importance of the dynamic hypothesis concept
that he decided to give this label to the conceptualization phase of the model building process
(pp. 86-87):
Once the problem has been identified and characterized over an appropriate time
horizon, modelers must begin to develop a theory, called a dynamic hypothesis, to
account for the problematic behavior... A dynamic hypothesis is a working theory
of how the problem arose... Much of the remainder of the modeling process helps
you to test the dynamic hypothesis... (pp. 94-95)
Richardson and Pugh (1981) add that while a dynamic hypothesis should be sketched early on in
the modeling process, a well-focused, consistent and clear statement may not be possible until
the model is formulated, simulated and evaluated extensively (pp. 56 and 63).
Stenberg (1980) reported spending as much as six months exploring problem definition
prior to engaging in a particular model building effort. In retrospect, he found it to be a wise
decision because it gave the rest of the project the necessary direction. He wams: “The model
builder much too easily loses sight of the objectives of his work, and begins to develop a general
purpose model that aspires to answer all questions but in the end yields disappointingly few
insights” (p. 300).
View of models as “boundary-objects”
The analysis of complex problems can be difficult because critical information is lacking or
because decision-makers lack the ability to effectively integrate and utilize the information that
is available (Reagan et al. 1991, p. 53). For this reason, Simon (1957) proposed that decision-
makers make bounded rather than optimal decisions. This is where system dynamics can be
useful, by helping guide the selection, and by efficiently and effectively integrating and
1 The earliest citation I found to the concept of a dynamic hypothesis was in Stenberg (1980), p. 300/312, endnote
number 3, referring to J. Randers’ Ph.D. dissertation, p. 54: “Conceptualizing Dynamic Models of Social Systems:
Lessons from a Study of Social Change,” Alfred P. Sloan School of Management, MIT (September 1973),
Cambridge, Massachusetts.
processing information that is interrelated in complex ways.'” However, some would argue that
problems are interrelated, and that there is room for ambiguity in problem selection and analysis
(Reagan et al. 1991, p. 52).8
Stakeholders and multiple constituencies
The primary source for ambiguity lies in the fact that much too often multiple constituencies use
multiple criteria, and multiple resources and constraints when thinking about and addressing
complex problems. Thus, problem definition is perception-dependent and value-laden.
Consequently, different people define and give shape to problems differently. This adds an
additional layer of complexity to already complex situations (V ennix 1996, p. 1). When dealing
not with one client, and not with a very specific and tangible issue, problem definition and policy
analysis will most likely have to emerge from some sort of discussion. For example, Reagan et
al. (1991) report using decision conferencing, based upon multiple technologies,’* to help a
client assess policy changes and their utility to various stakeholders.
In the forest study conducted by Stenberg (1980), at the start of the intervention the
research team had a list of emerging problems that were interrelated. In deciding which problem
to focus upon:
They discussed in meetings with the reference groups what might become the
most important problems... and how those problems could be dealt with. The
team had to sort out temporary changes from persistent trends, attempt to explain
the forces behind the trends, and then hypothesize about what kind of future
would emerge... they would present theories and receive criticism or support. (p.
300)
We can clearly see how in these two cases, the problem and the analysis emerged and/or were
given shape through discussion. The discussion not only involved the clients and/or stakeholders,
but also the research team and its facilitator(s), who can be conceivably very influential in the
whole process. Had the participants been different, would these groups have traveled the same
paths, and arrived at similar conclusions? How robust were these processes in terms of resulting
in the same outcomes? Could minor changes in these interventions, such as the use of an
alternative facilitator, for example, have produced significantly different results and findings?
This paragraph should be revisited to avoid referring to Reagan et al. when should really be quoting from Simon
directly, and to do a better job in specifying Simon's contribution to the SD paradigm, extracting quotes from classic
system dynamics texts referring to Simon’s work (for example, Sterman’s link between bounded rationality and
misperceptions of feedback).
Bas argued in Eden et al. (1983), among others.
Tn this particular case, a system dynamics model was combined with both a multi-attribute utility (MAU) model
and a process designed to uncover and challenge the key assumptions on which policies and strategies rested (SAST
- Strategic Assumption Surfacing and Testing). This involved identifying all of the important stakeholders (and
listing their assumptions) and weighing their relative importance (p.58).
“Messy” problems
The fact that strategic problems may be interrelated, thus increasing the complexity of the
problems, has led to the concept of “messy” problems (Vennix 1996, p. 1; Vennix 1999, p.
380)*5, ie., an ill-defined problem resulting from a “situation in which opinions in a management
team differ considerably” (Vennix 1999, p. 379). According to Vennix (1999), such situations
arise from individual and social deficiencies in perception, memory, and communication (pp.
383-389).
Individual sources of messy problems are related to selective perception and selective
memory based upon personal experience and formed expectations (V ennix 1999, pp. 383-384).
This process results in forming illusions which, in tum, construct realities: “Everyday reality
presents itself as an inter-subjective world which is shared with others (Berger and Luckmann
1966)” (Vennix 1999, p. 383). Social sources of messy problems are related to deficient patterns
of social interaction and communication, which fail, in and of themselves, to demystify the
illusions formed in the mental models of individuals (Vennix 1999, pp. 385-388). This process
results in a ‘reality of multiple realities’: “Humans not only construct reality in their minds; their
behavior also causes this reality in their minds to become reality in their environment... ‘If men
define situations as real, they are real in their consequences’ (Thomas and Thomas 1928, p.
572)” (Vennix 1999, pp. 386 and 387).
According to Vennix (1996), sometimes people will not even agree that there is a
problem, much less what it is (p. 13). Either way, in these situations, problems are quite
ambiguous and intangible, as opposed to the idealized problem statement pursued in the micro-
world view of model building. In this sense, one may argue that “there are no ‘objective’
problems, only situations defined as problems by people” (Vennix 1996, p. 13)'6,
Evolving problem definition and choice
The above discussion implies that when we are dealing with messy problems, it is, by definition,
difficult to get agreement among different people on what is the problem to be modeled.
Different participants will see the problem differently, and hold different priorities as to what are
the most important issues. In a group model building intervention, the modeling-team will strive
to move the group toward an agreed-upon modeling exercise, beginning with a particular focus
on one issue, which may then evolve to link with or change to other, more pressing, important,
central or dynamically relevant problems. As Andersen et al. (1997) indicate:
The outcome of a group model-building process may differ considerably from
what was expected at the outset... [This] results from the difficulty to diagnose
readily and fully a client’s problem in advance of the group model-building
intervention. Sometimes the “real” problem does not emerge until the group
model-building process is underway. (p. 194)
5 Citing Ackoff (1974, 1979).
© Citing among others Eden, Jones and Sims (1983).
In some cases, problem definition will rest simply in a matter of choice. This excerpt from
Vennix (1996), reporting on the first session of an actual intervention, illustrates this situation of
an evolving problem definition:
After a brief introduction to the topic of system dynamics ... the discussion was
started with the identification of the problem to be modeled. Participants initially
disagreed about the problem to be addressed with the model. One group member
emphasized the new financial situation for housing associations... Another
member of the group disagreed and explained the dangers in focusing on financial
issues... A third view on the matter was... These three different purposes were
discussed at length and it was difficult to arrive at an immediate choice for one of
these three issues... it tumed out that the third issue ... would most probably lead
to a very complex model... As a result, the group felt that it would be better to
first focus on the other two goals. And rather than selecting one of these two, it
was agreed that it might be interesting to use the model to try to find out how
competitive these two objectives are by incorporating both into one single
model... (pp. 205-206)
Obviously, in these situations, stakeholders are not simply identifying a “real” problem, and
providing information or securing implementation of the modeling results. They are indeed
defining and shaping the problem and the system, as they come to jointly perceive it. That is, the
participants of the group model building intervention are constructing a socially negotiated order
that can be best understood, in the form of this boundary- object called the model.
In working with social issues, do we know in advance when we're dealing with
preexisting as opposed to messy problems? To what extent are we modeling reality, as opposed
to a socially constructed order? If we knew, it would be easy to decide what to do. Unless we're
working on simple problems, prior to having worked extensively on the problem, we probably
don’t know the answer to this question. Thus, important stakeholders are key to the modeling
effort. This is because the modeling exercise becomes a venue for negotiation and alignment to
occur. The way the problem gets defined depends on who’s in the room. The elements in the
model depend upon how the participants perceive and negotiate their reality. The model becomes
a boundary-object in this negotiation. The model reflects what the group perceives as important
elements to depict in the system, and to describe and tackle “their” problem.
Lack of agreement and the need for consensus building
Vennix (1996) argues that wide discrepancies among individual mental models of problems have
detrimental effects upon organizational effectiveness:
All else being equal, the larger the discrepancies between managers’ mental
models in an organization the more lack of shared vision, the more divergence in
behavior and the higher the dispersion of organizational energy. This in tum
impedes the effective operation of the organization, because it will induce a lack
of cooperation. (p. 24)
Based upon the research of others,’” he recommends that the most important goal in dealing with
messy problems is the creation of a shared reality and problem definition among what he calls
problem owners (Vennix 1996, p. 24).
In building system dynamics models, the lack of agreement on the problem being
modeled, and the purpose of the modeling effort, will lead to not building a model at all, or
building models of systems, as opposed to problems. Either outcome is less than ideal. Building
a qualitative model (in the form of mapping and diagramming), or building a quantitative model
of a system (but without a clear problem to solve) may still be worthwhile. Because it provides
the participants the opportunity to lear from each other's perspectives, thus aligning their
mental-models. While, at the same time, adding rigor to the discussion, providing them with
means to keep track of complex causal structures, and serving as a group memory of their
understanding (V ennix 1999, p. 382).
Some would argue that it is simply just useful to think about the problem in new ways,
particularly if this new way of thinking can provide an added value in terms of problem
structuring and/or perspective. In using system dynamics models in decision-conferencing,
Reagan et al. (1991) indicated:
Because the problem was complex and the implications of any policy change
were uncertain, [we] constructed a system dynamics simulation model... [We]
selected this modeling technique because it would expose the nature of the ...
system, make controversial assumptions explicit, and provide a common
framework that would help policy makers develop a shared understanding of the
problem... [However] the value of decision modeling to strategic thinking is
primarily in the cognitive, social, and political activity of building the model,
rather than in the completion of the model. (pp. 55 and 63)
Indeed, Stenberg (1980) suggests that once the problem has been defined, it is important that the
client group perceives the problem as sufficiently relevant to warrant further modeling analysis
(p. 308).
Difficulties in understanding social systems
Forrester began to investigate less tangible social systems (1969, 1971-B) shortly after
conceiving the tools and methodology to model industrial dynamic systems (1961). He
envisioned that system dynamics would be a useful tool to advance the knowledge of social
systems, by exploring their dynamic nature (Forrester 1987-A, p. 136). Forrester indicated that
social science had not advanced in step with natural science. He quoted Skinner (1971):
Twenty-five hundred years ago it might have been said that man understood
himself as well as any other part of his world... Today he is the thing he
understands least. Physics and biology have come a long way, but there has been
no comparable development of anything like a science of human behavior...
Aristotle could not have understood a page of modem physics or biology, but
1” Citing among others Eden, Jones and Sims (1983).
Socrates and his friends would have little trouble in following most current
discussions of human affairs. (p. 3)
Forrester believed that system dynamic models would raise the quality of the debate (1987-A, p.
147), System dynamics capable of tracing the complexity of social systems, would provide a
means for improved communication and testing of people’s mental models.'®
Stenberg (1980) pointed out that this change in the field of application of system
dynamics (to public policy) would have to be accompanied by an evolution in methodology (p.
292). He added, “the problems of integrating information gathering, modeling, and
implementation are accentuated when we move into the area of public [social] policy” (p. 294).
When Stenberg (1980) anchored his modeling work on the thinking of a group of key
stakeholders, he scratched the surface in terms of identifying disagreements, potential conflicts
of interest, and problems of communication in building a model that belonged to multiple
constituencies. A fter quite a bit more experience with group model building, Richardson (1999)
concludes:
We know that building insightful system dynamics models is difficult and
requires advanced skills in the modelers’ arts and sciences; building insightful
models with groups is made even more difficult by the intricacies of interpersonal
communications, group process, and human relations. (p. 375)
This approach to system dynamics modeling is revealing the fragility of our premises in
trying to understand social systems. Before we can set out course to solve “real” problems, we
have to struggle upon a shared understanding of what real is. Also, for this very reason, group
model building practice has resulted in deviation from classic system dynamics, in terms of
modeling purpose.
The multiple purposes of group model building
While the main purpose of system dynamics, as argued by Forrester himself in Industrial
Dynamics (1961), is to aid in designing better management systems, its application in group
model building can be best understood when problems are perceived as emerging from debate
and discussion, as opposed to preexistent. Therefore, group model building has also been useful
in helping to create a shared perspective and understanding of the clients’ issue. The model built
by the group is viewed as a boundary-object subject to negotiation, and it is useful to the extent
that it helps them reach agreement regarding the problem. The model also serves as a tool to
investigate potential lines of action (Richardson and Senge 1989, Reagan et al. 1991). The model
is useful to the extent that it helps the client group reach a consensual decision about what to do
(Winch 1993).
From this point of view, a group model building intervention is a team-leaming or
organizational-learning activity in the sense that it seeks to achieve the above objectives by
promoting alignment, and pursuing a shared-vision for the group, team or organization (V ennix
18 Both his methodology and findings, however, have been subject of controversy (Brewer and Hall 1973, Nordhaus
1973).
1994, Huz et al. 1997). If successful, the intervention can lead to commitment to and, eventually,
organizational change (A kkermans et al. 1993, Vennix et al. 1993, Vennix et al. 1996). The new
“reality” that is created can then be observed and assessed, i.e., the process starts all over again.
2. The dichotomy in model conceptualization
The second step of the system dynamics model building process is model conceptualization. I'll
illustrate and discuss the dichotomy in model conceptualization in terms of four main issues: a)
the role of the structuring- framework; b) knowledge elicitation and mental models; c) delineation
of model boundary; and d) the role of the modeler/facilitator. Tables 6a, 6b, 6c and 6d, found
attached in the Appendix, summarize my findings while mapping the literature into the
dichotomy.
View of models as “micro-worlds”
The goal is to test the dynamic hypothesis
As previously stated, one of the end products of the problem identification and definition step is
a preliminary theory accounting for the cause of the problematic behavior, called the dynamic
hypothesis. According to Forrester (1961), the first objective of building a model is to test the
dynamic hypothesis (pp. 56-57). In other words, the experimental world (the model) is designed
to confirm or disconfirm the initial hypothesis, at least in the experimental setting: “We build a
model to see if the mode of behavior could exist and whether or not it can result from the initial
assumptions” (p. 450).
Randers (1980-B) suggests that the major creative step in model conceptualization is
using the reference mode as “a catalyst in the transition from general speculation about a
problem to an initial model” (p. 136). The goal of the conceptualization stage is envisioning the
causal structure capable of reproducing the problematic behavior, as depicted in the reference
mode (pp. 130-131). The actual test of the dynamic hypothesis is carried out when the model is
formulated and simulated. But, model conceptualization involves explicit identification of the
key variables, and of the key interrelationships among these variables, responsible for the
observed problematic behavior.
The power of the system dynamics framework comes from its ability to examine the
causes of behavior endogenously. Therefore, problematic dynamic behavior should be addressed
with an endogenous theory capable of explaining the dynamics of a system through the
interaction of the key variables represented in the model. Once the first objective -testing the
dynamic hypothesis- is fulfilled (with an endogenous explanation), the model becomes useful in
its second and main objective -system redesign for system improvement:
By specifying how the system is structured and the mules of interaction (the
decision rules in the system), you can explore the pattems of behavior created by
those rules and that structure and explore how the behavior might change if you
alter the structures and rules. (Sterman 2000, p. 95)
Sterman warns that the dynamic hypothesis is always provisional, subject both to revision or
even abandonment (p. 95). In other words, it is a “working” theory that captures a present state
of knowledge. Our knowledge, and its articulation in the form of a working theory, is subject to
change due to learning from the experimental world, as well as the real world (pp. 88-89).
A rather simplistic yet useful way of characterizing the micro-world view of model
conceptualization is describing it in terms of a top-down approach that seeks to conceive the key
pieces of causal structure capable of reproducing key reference modes of dynamic behavior. The
reference modes are the starting point to theory building. The dynamic hypothesis is a working
theory of the feedback structures that supposedly will reproduce the reference modes of
behavior. Once the model is formulated, the simulated behaviors are contrasted against the
reference modes. The working theory is evaluated in terms of the closeness of fit between
simulated and actual modes of behavior. The dynamic hypothesis is either confirmed or (partially
or totally) rejected. In the latter case, model based learning suggests reviewing the working
theory, and revising or reformulating the dynamic hypothesis, until a limited set of key pieces of
causal structure are indeed capable of reproducing the key reference modes of behavior (based of
course in a logical real world explanation).
Eliciting prospective theories and facts
Simulation models are conceptualized based upon information gleaned from the real world
(Sterman 2000, p. 88), synthesized in the form of a mental model, that is, “an understanding of
the operation of the real world” (Randers 1980-B, p. 119). According to Forrester (1961), active
practitioners possess sufficient information to serve the model builder in conceptualizing an
initial model:
Searching questions, asked at points throughout the organization under study by
one skilled in knowing what is critical in system dynamics, can divulge far more
useful information than is apt to exist in recorded data. (pp. 58-59)
In group model building, the information upon which the model will be built has to be
elicited from the multiple mental models of the client team. Sterman (2000) cautions that
different members of the client team may hold different theories about the causes of a problem
(p. 95). Forrester (1961) warns of the danger of the participants’ “wishful thinking” and
“strongly formed past prejudices”, as hazards to successful model conceptualization (p. 452).
The latter can be regarded as the problem of the difference between espoused theories and
theories in use (Argyris 1999); and the issues previously referred to as selective perception and
selective memory.
The micro-world view of model conceptualization stresses the importance of a factual
based and empirically accountable model. If the basic assumptions built into the model are
derived from the mental models of people, then:
A good modeling process challenges the clients’ conception of the problem.
Modelers have a responsibility to require their clients to justify their opinions,
ground their views in data, and consider new viewpoints. (Sterman 2000, p. 85)
Parsimony and the dynamic hypothesis guide model boundary decisions
The reference mode(s) the model is meant to portray determines what to include or exclude from
the boundary of the model's causal structure: “The reference mode helps the modeler focus on a
specific phenomenon instead of ending in diffuse mapping of a system” (Randers 1980-B, p.
131). Thus, “the behavior of interest must be identified before the boundary can be determined”
(Forrester 1975, p. 112). Also, the questions to be addressed in the model control the content of
the model (Forrester 1961, p. 60), further shaping its boundary. Therefore, together, the
definition of the problem and the purpose of the model, initially synthesized in the form of a
dynamic hypothesis, should guide decisions regarding the boundary and scope of the conceptual
model (Sterman 2000, p. 98). As stated by Forrester (1961):
The initial hypothesis is part of the establishment of the initial questions and goals
for the study. Without this initial mental and verbal model of the dynamic
behavior being studied, there is no basis for deciding what factors might be
important and which ones could be neglected. (p. 450)
For this reason, Forrester (1961) wams that lack of clarity of the dynamic hypothesis will subject
the modeler to vulnerability to unessential complexity and detail (p. 453)
Classic system dynamics emphasizes (particularly at the stage of initial model
development) the importance of parsimony, guided by choices based upon the dynamic
significance of variables, made in the context of the study’s purpose and problem. Richardson
and Pugh (1981) advise to begin simply, containing complexity, and including in the model’s
boundary only those quantities that are perceived as dynamically significant for the purposes of
the model, until a simple causal structure is well understood (pp. 43 and 61). Forrester (1975)
recommends defining the boundary in terms of the smallest numbers of components needed to
capture the essential dynamics and purpose of the study:
One asks not if a component is merely present in the system. Instead, one asks if
the behavior of interest will disappear or be improperly represented if the
component is omitted. If the component can be omitted without defeating the
purpose of the system study, the component should be excluded and the boundary
thereby made smaller. (p. 112)
In Industrial Dynamics (1961), Forrester hypothesized that the novice modeler includes
too much detail in the model, because he/she lacks the ability to discriminate if a particular factor
is indeed necessary. While, alternatively, experience in building models leads to discovery of
how much simplification is possible. He concluded that this problem boils down to a matter of
degree (p. 453). And, he recognized that “defining the system boundary and the degree of
aggregation are two of the most difficult steps in successful modeling” (in Sterman 2000, p.
100).
Nevertheless, classic system dynamics highlights the importance of parsimony in initial
model development. By admitting only in the model improvement phase -i.e., “after the initial
model passed generalized testing at an acceptable level”- that the model be extended and
elaborated “to increase richness and realism through changes in system boundary, level of
aggregation, or detailed formulation” (Randers 1980-B, p. 135). The most recent text in system
dynamics reiterates this position:
The art of model building is knowing what to cut out, and the purpose of the
model acts as the logical knife. It provides the criteria to decide what can be
ignored so that only the essential features necessary to fulfill the purpose are
left... [W ]ithout a clear purpose, there is no basis to say “we don’t need to include
that” when a member of the client team makes a suggestion. (Sterman 2000, pp.
89-90)
In fact, Sterman strongly advises:
Modelers should not automatically accede to clients’ requests to include more
detail or to focus on one set of issues while ignoring others, just to keep the
clients on board” (p. 85).
Forrester (1961) concluded that the key to success in determining the boundary and scope of a
model lies in the modeler (p. 450). He suggested the modeler should be bold, yet fit the
conceptual work of model development to his/her own skill, time, and experience.
Regardless the aptitude of the modelers, they should not disguise the limitations of their
work (Sterman 2000, p. 98). An essential instrument, “surprisingly useful and shockingly rare”,
to reveal the boundary and scope of a model is a model boundary chart (Sterman 2000, p. 97-99).
This is a three-column table that explicitly recognizes the results of discriminating thinking
(dynamic-, problem- and purpose-based) regarding decisions about which key variables to model
endogenously (first column), to model exogenously (second column), and to altogether exclude
from the model boundary (third column). The model boundary chart allows model users to
“decide for themselves whether the model [is] appropriate for their purpose” (p. 98). According
to Sterman:
Without a clear understanding of the boundary and assumptions, models
constructed for one purpose are frequently used for another for which they are ill-
suited [or even totally inappropriate]. (pp. 98-99)
Another important instrument to convey information regarding the boundary and level of
aggregation in a model is the subsystem diagram (Sterman 2000, pp. 99-102). This is a macro
view of the model showing the number and type of different organizations (agencies or sectors)
represented, and how they are interrelated in the model. The subsystem diagram will only
implicitly reveal information regarding endogenously, as opposed to exogenously modeled
variables, and it says nothing about variables that have been excluded from the model. But it
does convey a system’s view of the model, that is absent in the three-column model boundary
chart. Together, the model boundary chart and the subsystem diagram reveal the results of the
modeler’s systematic decisions regarding the boundary and scope of the model.
The modeler as an expert in the technology
The modeler brings to the group model building effort technological skills that must be exercised
diligently and smartly. First, the modeler should view the problem and the system from the
proper perspective: not too far, not too close (Forrester 1961, p. 451). To regard it from too great
a distance is to neglect essential decision points, nonlinearities, and interconnections. To
approach it too closely is to include too much detail, and to place too much importance on
individual decisions as opposed to decision rules and policies.
Second, the choice of the time horizon of the simulation has significant influences upon
the definition of the problem, and the evaluation of the policies under consideration (Sterman
2000, pp. 90-94):
The time horizon should extend far enough back in history to show how the
problem emerged and describe its symptoms. It should extend far enough into the
future to capture the delayed and indirect effects of potential policies. (p. 90)
Sterman suggests that clients tend to underestimate time delays, think of cause and effect as local
and immediate, and therefore propose time horizons that are far too short (p. 90-91). He
concludes that the modeler must “guard against accepting the clients’ initial assessment of the
appropriate time frame” (p. 94).
Third, the modeler’s expertise and experience in dynamically complex systems, and in
modeling and simulation technology, are key to the development of a useful model. According to
Forrester (1961):
The skill of the person who undertakes to use a model is tested immediately -his
first decision is to ask pertinent questions having important answers. (p. 60)
Fourth, the modeler should distinguish the actual state of affairs from mistaken or
idealized perceptions of it, based upon the clients’ biases and normative standpoints. In other
words, the modeler needs to observe first-hand the system to distinguish espoused theories from
theories in use (Forrester 1961, p. 452).
The list of potential contributions of the modeler as an expert in the technology is vast,
and beyond the purpose of this section. The point to be made is that a smart system dynamics
modeler can build a more insightful model than the client- group:
The exploring (problem solving) task is both most central in the model-building
process and least well-developed in the psychological literature. Some evidence
suggests that well-trained or knowledgeable individuals can perform as well as or
even better than groups. Simply put, a well-trained model builder can do as well
as a group of model builders in tasks such as proposing formulations or designing
feedback structures. (V ennix et al. 1992, p. 33)
For this reason, it is argued that the modeler has ethical and professional responsibilities, above
and beyond his/her desire to work with and please the clients:
The political context of modeling and the need to focus on the clients’ problem
does not mean modelers should be hired guns, willing to do whatever the clients
want... As a modeler you have an ethical responsibility to carry out your work
with rigor and integrity. Y ou must be willing to let the modeling process change
your mind. You must “speak truth to power,” telling the clients that their most
cherished beliefs are wrong, if that is what the modeling process reveals, even if it
means you will be fired. (Sterman 2000, p. 85)
The technical expertise of the modeler should not be confused with substance-matter or
content knowledge of the subject under study. While the modeler may know a great deal about
the problem at the start of the intervention (or become very knowledgeable in the subject as a
result of learning through the modeling process), the modeler should be regarded only as an
expert in the technology (Reagan et al. 1991, p. 63). For this reason, particularly in the start of
the intervention, the modeler is hard-pressed to learn as much as possible, and very quickly, from
the client group, about the problem and its context. In the words of Sterman (2000):
Early in the modeling process, the modeler needs to act as a facilitator [in the
discussion among the client group], capturing [their] mental models without
criticizing or filtering them. Clarifying and probing questions are often useful, but
the modeler’s role during this early phase is to be a thoughtful listener, not a
content expert... Your goal is to help the client develop an endogenous [and
valid] explanation for the problematic dynamics. (p. 95) [Emphases added]
When models are viewed as micro-worlds, the definition of “The Client” takes on a
peculiar meaning:
The client is not the person who brings you in to an organization or champions
your work, nor even the person who pays for the modeling study, though it is
helpful to have contacts, champions, and cash. Your clients are the people you
must influence for your work to have impact. They are those people whose
behavior must change to solve the problem... If your [paying] clients push you to
generate a result they've selected in advance or that is not supported by the
analysis, push back. If your clients’ minds are closed, if you can’t convince them
to use modeling honestly, you must quit. Get yourself a better client. (Sterman
2000, p. 84-85) [Emphasis added]
View of models as “boundary- objects”
In the previous sections, I described the micro-world view of model building in terms of an
intelligent and skilled modeler, who synthesizes in the form of a mental model information
derived from the client group. A problem statement, reference modes, and a dynamic hypothesis
provide the structuring-framework and guide the modeler, while he/she filters the information
gathered from the clients, adjusting it with respect to potential problems from the participants’
biases in perception, and wishful thinking. Prospective theories are contrasted, and facts are
elicited. Theoretical inconsistencies and judgment errors on the part of the participants are
handled by pursuing logical coherence, and correspondence with observed data, and concrete
behaviors. Based upon the best mental model of the problem -that the modeler is capable of
envisioning- a conceptual model is conceived. This model serves as the basis for model
formulation, for the purpose of testing the dynamic hypothesis. In other words, for the purpose of
testing the theoretical explanation for the problematic behavior.
In the following sections, I'll shift attention to the boundary-object view of model
building, previously discussed in terms of the existence of multiple constituencies and socially
constructed problems. I’ll begin by collecting the evidence that while we may be interested in
eliciting theories and facts, we may actually be eliciting merely views and opinions. Second, I'll
raise the issue of the role of structuring-frameworks in knowledge elicitation. What kind of, and
how much structuring should there be in the elicitation process? Third, I'll address the
consequences of the answer to this latter question, in terms of issues dealing with model scope
and boundary. I'll conclude the illustration of the dichotomy in model conceptualization by
discussing the role of the modeler in the boundary-object view of model building, which is
related to the issue of group-ownership, as opposed to modeler- ownership of the model.
Eliciting views and opinions
Roberts et al. (1983) argued that it is impossible to identify the components of any system
without a clear idea of what the problem is, and who is interested in the problem (p. 26). There is
more to this argument than the previously discussed idea that the clients define the problem.
Who’s interested in the problem also determines the knowledge base to tackle the
problem, and has a great deal of influence in shaping the boundary of the model. According to
Morecroft (1994), individual mental models play two roles in the modeling intervention. They
are the source of knowledge and information in environmental scanning, and they give shape to
the groups debate and dialogue (pp. 7-8). Consequently, varied mental models based upon
whatever knowledge the participants have - “real or imaginary, naive or sophisticated”- enter the
debate and give final shape to the group’s collective view of the system, determining their future
actions (p. 7). In group model building, it has been acknowledged that the intervention depends
“on the thoughts and agendas the client group itself brings to the workshop” (Richardson and
Andersen 1995, p. 133). [Emphasis added]
Figure 5, copied from Morecroft (1994, p.10), underscores the fragility of the notion that
one can readily elicit theories and facts from the mental models of participants. On the top, right-
hand comer of the picture, a deck of cards illustrates the knowledge base and mental models of
the participants. Each dot represents a fact that a participant carries in his/her head. Small
rectangles represent learned concepts or perceptions of social and political factors. An
individual’s knowledge base of facts and concepts is extensive, and contingent upon the
individual’s singular collection of experiences. The network of facts and concepts, as illustrated
in a single card, composes the individual’s mental model. The mental models are activated in the
process of recognition of strategic issues, or problem issues. They shape debate and dialogue,
and are in tum shaped by the exchange (of knowledge, facts, concepts, and networks of facts and
concepts) with other participants, facilitated by the modeling team. (Therefore, these individual
mental models are not static, but changing over time.) The structuring-framework adopted in the
intervention influences the exchange among participants. A modeling team or facilitator
intermediates the whole process (pp. 5-11).
changing business environment
——
recognized strategic issue
a
knowledge base & mental
‘executive debate and dialogue
=: concepts and
=" theory plus
facilitation
action plans and chang
diagrams
e 1
tI words
HH friendly algebra
facilitation
simulation
models &
gaming
TecRonoRts simulators
FRAMEWORK
Figure 5. Knowledge base and mental models, debate and dialogue
(Copied from Morecroft 1994, p. 10)
As a result, the development of a shared mental model depends, quite literally, upon
“who’s in the room”. The replacement of the deck of cards (participants), the choice of
structuring-framework, a change in the modeling team or of facilitator, all have potential
implications in terms of the construction of a problem definition, of knowledge elicitation, and of
agreed-upon causal explanations. In conceptualizing a model based upon this shared mental
model developed during the intervention, the resulting model can not be seen as cast in stone.
If we accept the notions proposed by Lane and Morecroft:
If the model is indeed a representation of a client’s idea on how the world
functions, then this microcosm, or microworld, is the transitional object upon
which the experimentation is performed. (Lane 1994, p. 100) [Emphasis added]
And
[Pleople leam effectively when they have transitional objects to play with in
order to develop their understanding (or refine their mental models) of a particular
subject or issue. The combination of transitional objects, leamer, and learning
process is what Papert [1980] calls a microworld. (Morecroft 1994, p. 11)
[Emphasis added]
Then, we may extend this notion by implying that a transitional object that is found acceptable to
a group of people becomes a boundary-object that reflects the group’s negotiated representation
of reality (i.e., a socially negotiated order). This is why it is so consequential that the key
members involved in an important decision-making process be in attendance of the intervention
or workshop. Quinn et al. (1985) add, “[a]ttendance of all key members ensures not only
appropriate expertise and input, but also understanding and commitment” (Quinn et al. 1985,
p.53). [Emphases added]
When a group of people assembles, as depicted in Figure 5, in a group model building
intervention, to make a decision, or to develop an action plan, around a problem or issue, there
are many sources of, and reasons for disagreement. If they are conceptualizing a model, they
may disagree in several points of the conceptualization process: problem definition, selection of
key variables, interpretation of variables (concepts and constructs), appropriateness of reference
modes (particularly if they are not based upon time-series data), causal theories, inputs to rate
equations, and parameter values. The participants may disagree for several reasons: selective
perception/memory, different political points of view, wishful thinking, uncertainty and/or
ignorance.
The way to resolve disagreement within the client group may be contingent to the locus
of, and reasons for disagreement. A political disagreement around problem definition may be
resolved through negotiation. While, different causal theories may be contrasted and tested using
the model as a laboratory. But, because often people simply don’t know how some processes
function, ’’ it is easier to accept the notion that we’ re eliciting views and opinions, as opposed to
theories and facts. If this is the case, and if participants of the intervention hold quite different
views and opinions, rather than eliciting people’s theories as a starting point, it may be more
useful to elicit their views of the system. This will be discussed in the next section, in terms of a
bottom-up approach to model conceptualization.
19 For example, in a model conceptualization exercise, Vennix and Gubbles (1994) admitted:
Although there was consensus on many issues, it also became clear that several processes in health
care are poorly understood. Here the knowledge elicitation process was arrested at the point where
there were only vague conjectures. (p. 138)
In eliciting parameter values, Richardson and Andersen (1995) found:
The parameter elicitation exercise was surprisingly crucial, not only providing input to the
modeling effort but revealing areas of uncertainty, disagreement, and actual ignorance among
these experts on foster care, which pointed toward the need for further work. (p. 117)
What kind of structuring-framework, and how much structuring?
This section builds upon an assumption, that instead of being told or taught, people learn through
discovering for themselves (Morecroft 1994, p. 4). In other words, people make up their own
minds (de Geus 1994, p. xiv). It discusses how a structuring-framework to learning (an example
of which is the system dynamics model building process) imposes constraints and biases, leading
people to learn particular things, in particular ways (Morecroft 1994, Lane 1994). I review
authors who argue that these biases need to be minimized, particularly early on in the
intervention (Lane 1993, Richardson and Andersen 1995).
Building a model around a dynamic hypothesis introduces a strong bias in model
conceptualization. Therefore, I suggest that some modelers often adopt a bottom-up approach to
model building, constructing (with the client group) a broader shared view of the system. Rather
than holding a narrow focus (in model conceptualization) on the dynamic hypothesis (Vennix et
al. 1988, 1990; Vennix and Gubbles 1994; Morecroft et al. 1991; Richardson et al. 1992;
Richardson and Andersen 1995; Lane 1994; Wolstenholme 1994). I explore the role of the model
in these cases (if it is not to test the dynamic hypothesis! ).
Toward the end of the section, I survey a view of the role of structuring- frameworks that
suggests that they are necessary, but they need to be used selectively. Different cognitive tasks
require different structuring-frameworks (Richardson et al. 1989, Vennix et al. 1992/94). A
group model building intervention is composed of a repertoire of sub-frameworks, wisely used,
embedded within the larger method, called system dynamics model building (Lane 1994, Vennix
1996, Andersen and Richardson 1997). An effective intervention is one that appropriately
matches the series of model building tasks with the best structuring-procedures for knowledge
elicitation and group dialogue.
In general, this section is about how models can be useful in a dialogue, and how people
can learn through building a model together. It addresses the complex issue of providing the
client group with frameworks that are helpful in carrying this dialogue, without biasing it. I argue
that the central idea is that the model serves as a boundary-object in the dialogue. While the
model may be more or less useful in informing a policy-making context, the goal is not
necessarily to find the answer to a problem. The goal is having the client group share a common
language (Lane 1994), a common view of the system (Morecroft 1994), generate and test ideas
and scenarios (Morecroft et al. 1991), and build consensus and support around an action plan
(Vennix 1994).
WHAT IS A STRUCTURING-FRAMEWORK? Morecroft (1994) distinguishes models in terms of
three attributes: 1) maps that capture and activate knowledge, 2) frameworks that filter and
organize knowledge,”° and 3) microworlds for experimentation, cooperation, and learning (p. 3).
°° The use of the word framework in different contexts and with different meanings throughout this paper is bound
to create some confusion. Because of Morecroft’s (1994) use of the term, distinguishing frameworks from maps, I
chose to refer to the general class of frameworks - approaches to organizing knowledge- as structuring-frameworks.
The latter includes research methods (e.g., system dynamics), approaches (e.g., group model building), and
frameworks (e.g., feedback-loop causal diagrams), as well as mapping tools (e.g., influence diagrams), lists, and
other less structured (or naturally structured) means of organizing knowledge (e.g., Hodgson’ s [1994] hexagons). A
top-down approach to organizing knowledge filters and fits information into a method or framework for a given
He proposes that each supports different cognitive tasks and group processes. The most obtrusive
use of a model is as a framework. They combine maps with concepts and theories. They add
structure imposing logical constraints:
Whereas a simple list just captures items of knowledge, a framework packages
and organizes knowledge. A framework also filters knowledge because some
ideas won't easily fit within the constraints of the framework... So although
modelers often say nowadays that they are mapping mental models, really they
are not. They are filtering and organizing from mental models to fit the modeling
framework. (pp. 9 and 11)
Because frameworks can introduce bias in knowledge elicitation from, and discussion among the
client group. First and foremost, it is important to establish which framework will best fit the
cognitive needs of the client group (Morecroft 1994, p. 11; Lane 1994, p. 104).
Lane (1993) recommends using “flexible” approaches to generate, select and study the
clients’ issues, with the goal of reducing any bias in the elicitation process toward the system
dynamics modeling method (p. 239). He argues that this allows the participants to frame their
problem in the most appropriate structuring technique (p. 240). This is particularly important in
the early stages of an intervention. Richardson and Andersen (1995) describe how simply
“explaining the mysteries of system dynamics or of a particular model formulation can get in the
way of uninhibited group discussion focused on the problem independent of approach or
formulation” (p. 132).
Among group model building practitioners, the mere use of a simple “concept model” as
a starting point of a conceptualization session already draws suspicion of introduction of bias in
knowledge elicitation and model conceptualization:
One might also question the extent to which the concept model driven by three
time series ... biased the group in the main two-day workshop toward exogenous
formulations. (Richardson and Andersen 1995, p. 135, footnote number 10)
It is important to highlight that the role of “concept models” is simply to introduce the system
dynamics framework (and its icons), to demonstrate the connection between causal structure and
system behavior, and to initiate discussion regarding the “real” system (Richardson and
Andersen 1995, p. 130). These models are not intended as preliminary versions of (endogenous)
causal structures addressing a dynamic hypothesis! Interestingly, I could not find evidence in the
group model building literature,”’ of models built around a 1 dynamic hypothesis. If this was the
case, the authors may have unconsciously omitted this fact.”
purpose. A bottom-up approach begins with instruments (e.g., hexagons) that allow for more spontaneous surfacing
and organization of knowledge (generating lists and naturally forming clusters), and gradually builds in the direction
of the most appropriate framework to tackle a given problem.
*! Group model building cluster of readings listed in the Appendix (number 5).
22 T suspect this to be the case, and it may have to do with the issue of client-ownership, discussed toward the end of
this paper.
A BOTTOM-UP APPROACH TO MODEL BUILDING. I characterize the boundary-object view of
model conceptualization as a bottom-up approach to model building, and I take the liberty of
drawing upon published work in system dynamics to illustrate it (Vennix et al. 1988, 1990;
Vennix and Gubbles 1994; Morecroft et al. 1991; Lane 1994; Wolstenholme 1994; Richardson
and Andersen 1995). I do not mean to imply that these authors take a pure boundary- object (or
bottom-up) approach to model building. But, this work does provide a sharp contrast with the
pure micro-world (or top-down) view, as I have defined it, in terms of a narrow problem
statement, a set of key reference modes, and -most importantly- a dynamic hypothesis, guiding
the elicitation and model conceptualization phase of the modeling process.
The bottom-up view of model conceptualization does not ignore reference modes, and
their critical role in model building and validation. In fact, reference modes are an integral part of
the problem definition phase. But, it tends to put them aside for a moment, in the model
conceptualization phase, and it concentrates first upon building a shared view of the system (in
terms of its conceptual structure). It de-emphasizes the (endogenous) model as a structuring-
framework, and uses it more simply as a mapping tool. As an example, I quote Morecroft’s
(1994) description of Wolstenholme’s approach to model conceptualization:
The modeler collects fragments of structure that, to begin with, are just lists of
key resources, states and resource flows. Lists are a good way to capture
managers’ [own] categories and concepts. These particular lists also generate raw
material for an influence diagram. Wolstenholme’ s [1994] approach gently shapes
a discussion first into a list and then into a diagram that eventually shows
feedback loops, delays, and organizational boundaries. (p. 23-24)
An alternative approach can found in Lane (1994, pp. 107-114).
Once a model is formulated, the simulated behaviors can be contrasted against the
reference modes. The dynamic hypothesis plays a minor role in model conceptualization, and
inferences regarding the causes of particular behavior surface from exploring the consequences
of the causal structure that was created (e.g., Richardson and Andersen 1995). Initial causal
theories are revisited in light of the conceptual model (e.g.: Lane 1994, p. 111). This approach
often results in a different understanding and definition of the problem (e.g.: Vennix et al. 1990,
p. 204-205; Vennix and Gubbles 1994, p. 140; Lane 1994, p. 112).
In practice, this approach is exercised in many different ways, with more or less emphasis
upon the role of the reference modes, and of a dynamic hypothesis, in model conceptualization.
But the general idea is that less emphasis on modeling as a framework allows for a richer, less
filtered elicitation and discussion, thus limiting the bias of the method in the results of the work.
This approach focuses upon the model as a tool for group dialogue and alignment. It recognizes
that a bulk of the problems confronting managers and policy makers are political in nature (Lane
1990, p. 93):
The reality is that any problem is embedded in a network of political, cultural and
power relationships. It is naive and futile to imagine that these can all be cut
through because a solution is known to be mathematically optimal. Any solution
that requires action to be taken will need to address the relationships of those
involved, account for them, and take time to organize their re-configuration. (p.
90)
Therefore, problem solving requires among other things creativity, in the form of idea and
scenario generation, and exploration (de Geus 1994, p. xv; Lane 1994, pp. 110-113; Morecroft et
al. 1991).
This approach to building a conceptual model generates useful policy relevant
information (Vennix and Gubbles 1994, p. 139; Eden et al. 1983), and it also enables the client-
group to share their mental models. It helps them develop a common language and a shared
understanding. As illustrated in Lane (1994):
The tool was found useful for analyzing ideas and generating insight... One of the
team members ... commented that the discussions had allowed him to produce
much information that might otherwise not have been captured in such an
organized form. As a result, he believed, the team would be able to use its shared
understanding [of their problem of interest] much more effectively as they had a
common language in which to describe it. (p. 110)
It helps them know better what they already knew:
By expressing such a mental model in some external form, we can help a client
use effectively a much greater proportion of the knowledge that they possess...
The most widely used reasons for creating an extemal representation of mental
models is the great benefit that can be gained by [naturally] structuring and
sharing information. (p. 100)
According to Morecroft (1994), the most informative work about the process of mapping in the
context of groups originates in the group decision support literature (p. 5). The decision
conferencing tradition is related to, and draws extensive upon, the group decision support
literature (Rohrbaugh 2000). In this vein, Milter and Rohrbaugh (1985) synthesize the role of the
model as a “decision accounting system” (p. 221). This means that the model captures and
reflects the results of the series of systematic decisions made by the group during the process of
elicitation and discussion (Quinn et al. 1985, p. 55). Thus, the group uses the intervention to give
form to a problem/system. The “working” model serves as a boundary- object for discussion and
negotiation. The “final” model reflects the result of the group's structuring-decisions. The model
is the closest thing to a concrete reality shared by the group. It is their representation of their
reality. Nevertheless, it is the group's socially negotiated order.
Taken literally, this may suggest that in some cases there is only marginal gain in actually
formalizing and simulating the model. Most of the benefit from model building results simply
from model conceptualization:
[T]he process of model building is frequently more important then the resulting
model. Model building itself is largely a learning process about the problem. Most
insights about the characteristics of an ill-structured problem are gained during
the iterative process of designing a computer model, rather than after the model is
finished. (Vennix and Gubbles 1994, p. 122) [Emphasis added]
SCRIPTS TO GROUP MODEL BUILDING. While it is important to use flexible approaches to study
the clients’ issues, thereby reducing bias associated with forcing the problem through a particular
frame. One must not forget that there are good reasons why research methods and analytical
frameworks are brought to bear upon problems. Reagan et al. (1991) explain why a host of
modeling techniques are used in decision conferences:
Decision models are intellectual tools that have been developed to make unwieldy
problems more manageable by structuring thought processes, clarifying
interrelationships, and handling complex data. These tools make the policy-
making process more efficient by enabling policy makers to rapidly integrate and
analyze information and options and make it more effective by enabling them to
examine policies and their implications thoroughly. (p. 53)
System dynamics, in specific, is perceived as particularly useful in exploring and understanding
the endogenous causes of problematic dynamic behavior, embedded in feedback rich, complex
systems (p. 54).
Hence, if the appropriate modeling technique is chosen to handle a problem, too much
flexibility may get in the way of learning about the problem. For instance, Richardson and
Andersen (1995) recognize that being too careful about the group process may have yielded
disappointing analytical results in a particular intervention:
The modeling team pressed for some causal feedback views but did not force an
endogenous dynamic feedback view. In the end, the [client] team was left with
few insights about the causal structure of critical parts of the system... [This]
model-based group work might be faulted for trying to be too responsive to the
group, and for failing to do a good job presenting and motivating the system
dynamics approach. (p. 133)
Research in the field of cognitive psychology revealed that knowledge elicitation and
problem analysis involve distinct cognitive processes related to three general types of tasks:
eliciting information, exploring courses of action, and evaluating situations (Richardson et al.
1989, pp. 346-347; Vennix et al. 1992, pp.29-30; Andersen and Richardson 1997, pp. 111-112).
Eliciting information, also referred to as intelligence, production, or conceptual behavior,
is best accomplished using a divergent structuring-framework. Divergent thinking is useful in
system dynamics modeling, for example, when exploring problem definition and alternative
causal explanations (Richardson et al. 1989, p. 346; Lane 1994, p. 104). Exploring courses of
action, also referred to as problem solving or design, is a different cognitive task requiring a
convergent structuring-framework. This type of structuring-framework is useful in revealing
feedback paths and formulating rates (Ibid.). Finally, the two forms of evaluation are judgment
and choice. Judgment has to do with assessment on a scale, as in the case of parameter
estimation. Choice has to do with selecting one or more options from a set, as in the case of
assessment of the performance of different policies (Richardson et al. 1989, p. 347).
Therefore, Andersen and Richardson (1997) argue that the key to successful group model
building “is selecting the most appropriate type of group structure and group task for each point
in time in the modeling conference” (p. 111; see also Vennix et al. 1992/94). For this reason,
these authors have begun to develop “scripts” for group model building - “sophisticated pieces of
small group process” (Andersen and Richardson 1997, p. 107), “planned and rehearsed for
accomplishing subgoals in the course of a group model building workshop” (Richardson and
Andersen 1995, p. 130).
Andersen and Richardson (1997) suggest a number of scripts for problem definition,
system conceptualization, parameterization, data estimation, idea generation (policy
alternatives), and model refinement. In these scripts, they’ve tried to match the nature of the
cognitive tasks with the most useful and least obtrusive structuring-frameworks. A sequence of
scripts thoughtfully used in an intervention generates useful products to the client-team, such as
“a stakeholder analysis, a precise description of a problem to be solved, a sketch of model
structure, or the determination of a set of actions to be taken” (p. 108).
Dealing with scope and level of aggregation
In the micro-world view of models, the issue of delineation of model boundary has been
characterized as a difficult task to be handled by the modeler, guided by as clear as possible
dynamic hypothesis, and driven by parsimony. In contrast, in the boundary-object view, the
minor role of a dynamic hypothesis in model conceptualization, coupled with a decision to depict
the system in richer detail, and deference to the client-group, raises some critical issues regarding
how to make judgments about model scope and level of aggregation.
Forrester (1961) conceded that some detail, “even when it does not affect system
performance, is justified in order to provide apparent reality and easier communication” with the
modeler’s client or audience (p. 453). Vennix and Gubbels (1994) considered an improvement in
quality in their initial model -in terms of reduction of ambiguity- the fact that the number of
variables in the model doubled, when the model was worked on by the client group:
[Building a conceptual model often generates very useful policy relevant
information... In our case several tangible results materialized from this
conceptual model building stage. These are related to the quality of the conceptual
model, the definition of the policy problem, and the structuring of future research
efforts... [T]he quality of the conceptual model was increased drastically on a
number of aspects, first, with regard to the number of variables included...
Although a larger conceptual model is not necessarily better, the increase was
primarily caused by refinement of the concepts and relationships... (pp. 139-140)
The willingness to depict the system in richer detail, guided by the desire to have the client group
share a common language, and a common understanding of the problem, is also characteristic of
Lane’s (1994) “Modeling as Learning” approach:
[A]s models are revisited, variables are seldom dropped, it being far more likely
that intermediate variables are added to clarify the nature of the causality. (pp.
102-103)
In a similar vein, the interventions described in Richardson and Andersen (1995) portray an
evolution in model structure, driven by the client group, to incorporate more and more structural
detail to the conceptual models (pp. 116-129). The authors noted:
The obvious malleability of the models, and their partial fit to the mental models
of the participants, led to a laundry list of concepts and variables the group wished
to see incorporated into a full model useful for forecasting and policy. (p. 121)
So, if the client group is not somehow contained, they are likely to push the modeling team in the
direction of incorporating into the model every institutional actor and relationship perceived as
even of modest importance in their system. Instead of acceding to modeling one problem, the
client group will wish to examine several issues. The client group will also be interested in how
these issues are interrelated.
In addition to willingness to depict the system in richer detail, the second important
characteristic of the boundary-object view is greater deference to the client group, in terms of
decisions regarding model scope and boundary. Lane (1994), for instance, requires the clients to
choose one issue to be addressed in the model, but then gives the clients considerable discretion
in deciding what issue they want to address, and what’ s important in addressing this issue:
[W]e should not try to model a system since there is no end to the effects that
should be included... [I]n order to put a boundary on the effects to be included ...
we model only one issue. We place the issue in the context of a system and then
include only those aspects of the system that the client considers to be important
or that they wish to concentrate their study on. There is no a priori requirement of
certainty regarding quantification, or even cause and effect. The very discussions
that take place around such points are part of the process, part of the deliverable.
(p. 96)
If the dynamic hypothesis does not act as the “logical knife” in the hands of the “savvy”
modeler, in the delineation of the model’s boundary, then, how are judgments made about model
scope and level of aggregation? Well, first, it is probably not true that modelers give up, fully
and willingly, their ability to use their system dynamics’ expertise, and their modeling
experience, in favor of the clients, in makings judgments about model boundary. More likely, the
existing literature does not reveal the subtleties in this modeler-client relationship, in which the
modeler certainly exercises a great deal of influence upon the client group, nurturing and guiding
the group’s decisions. While at the same time, providing the clients with a sense of power and
ownership about the decisions that are being made.
Nevertheless, in contrast with the micro-world approach, which recommends attention
primarily (or even solely) to the behavioral implications of the structural elements in the model
(Forrester 1961, p. 112). The boundary-object approach relies also upon a mix of group
techniques to force the client group to make simplification decisions on model scope and level of
aggregation. For instance, the separation of the roles of modeler from group facilitator allows the
modeler to take the onerous task of proposing simplifications to the group. While the facilitator
gives the group the option to take the modeler’s advice, or not, but urges the group to listen to
the modeler, because he/she is an expert in the technology, and a smart systems thinker
(Richardson and Andersen 1995).
There are several examples in the literature, where the Delphi and the Nominal Group
techniques are used to induce the group to make tradeoff decisions, ranking in importance, for
instance, key variables to be included in the model (Vennix and Gubbels 1994, Rohrbaugh
2000). Of course, these techniques are useful in several other situations, such as deciding which
strategic issue to focus upon, which policy alternatives to explore, parameter estimation, model
calibration, and evaluation of policy outcomes (Rohrbaugh 1979, 1981; Richardson et al. 1989;
Vennix et al. 1992/94).
While the “art” in group model building, in the micro-world side of the dichotomy, lies in
large part around the issue of conceiving the dynamic hypothesis, departing from problem
identification and model purpose. It seems to me that, in the boundary-object side, it is related to
this issue of delineation of model boundary. If experienced micro-world modelers explained how
they leap from problem definition to model conceptualization, and if experienced boundary-
object modelers described how to deal with scope and level of aggregation, then “more science
would be added to the group model building craft,” to paraphrase A ndersen et al. (1997).
The tension between the micro-world and the boundary-object views, regarding the
delineation of model boundary, is closely related to a more significant tension having to do with
what one considers a “good” model. Is a good model one that parsimoniously captures the
dynamic behavior of the key variables in a system, based upon an endogenous causal structure?
Or, is it a model in which the participants see themselves, and share an understanding of the
nature of the interrelationships of key variables in their system?
The modeler as facilitator; the issue of ownership
In “Modeling for Learning Organizations,” Morecroft and Sterman (1994) argue there is now a
“modem” view of modeling that repositions the role of the model and of the modeler. Models are
“owned” by policymakers, not by technical experts. They are created in a process in which the
modeler is, in part, a facilitator, “one who designs and leads group processes to capture team
knowledge” (p. xviii). This view is based upon the recognition of the fact that while the model is
an intellectual and analytical tool, the process of modeling is social and political (Reagan et al.
1991, p. 53).
A pragmatic understanding -that insights from even the most well developed models are
unlikely to be implemented- supports it:
I have not met a decision maker who is prepared to accept anybody else’s model
of his/her reality, if he knows that the purpose of the exercise is to make him, the
decision maker, make decisions and engage in action for which he/she will
ultimately be responsible. People (and not only managers) trust only their own
understanding of their world as the basis for their actions. “I'll make up my own
mind” is pretty universal principle for everyone embracing the responsibility of
their life... (de Geus 1994, p. xiv)
According to this view, the consultant modeler should avoid wearing the “expert hat”.
Instead, he/she should act as a “facilitation consultant,” offering “a process in which the ideas of
the team are brought out and examined in a clear and logical way (Lane 1994, p. 93). His/her
role is to “activate” the participant's knowledge and mental models (Morecroft 1994, pp. 8-9).
This role:
. is simply to encourage clients to put forward their ideas, to clarify them if
necessary, and to record them in a form that is both permanent and transferable.
(Lane 1994, p. 96)
Lane’s (1994) “Modeling as Learning” approach supports client ownership of all
analytical work, modeler acting as a facilitator, and predomination of “soft” analysis. The model
is labeled as an “articulated model” (p. 96). Perhaps, a close parallel to the “boundary-object”
type model. Here, even well known generic feedback structures need to be used with care, to
avoid loss of client ownership:
[W]e do not suddenly produce these large structures and give them to clients; this
would be against the whole philosophy of the approach. Instead we read around
the client’s problem to check whether there are any useful structures in existence
and, if so, slowly introduce helpful pieces to the client during the process of
model building. This process may not be as fast as just conjuring up a large
model, but it does ensure ownership and the benefits that flow from it. (p. 106)
TEAMWORK IN GROUP MODEL BUILDING. The dual role of the consultant as modeler and
facilitator has long been recognized in the decision conferencing literature (Milter and
Rohrbaugh 1985, p. 222; Quinn et al. 1985, p. 53). A way to balance the modeler and facilitator
roles of the consultant is to assign them to different members of a modeling-team. The decision
conferencing tradition influenced group model building, leading Richardson and Andersen
(1995, see also Richardson et al. 1992) to identify five roles in group model building. (In
addition to the facilitator and modeler, they also highlight the roles of the process coach, the
recorder and the gatekeeper.)
These authors hypothesized that all five roles must be present for effective group support.
While some of these roles may be combined, their explicit recognition and deliberate assignment
to skilled practitioners can significantly accelerate the group’s modeling effort (p. 115). They
warn, however, against combining the roles of facilitator and modeler:
[T]he more powerful minimum is not one person enlightened by perceiving
several essential roles but at least two people in a group modeling team, one
focusing on group facilitation, knowledge elicitation, and initial drafts of
structure, and the other focusing on the problem, the system being conceptualized,
real-time refinements of structure, and emerging insights. (p. 129)
Richardson and Andersen’s approach suggests that in resolving facilitator/modeler
conflicts, the bias is in favor of the facilitation role. The facilitator has the lead role. He/she is the
organizer and conductor of the group process, and “on stage and vulnerable” for most of the
group meeting time. The modeler and the process coach serve as content and process coaches
respectively:
We have chosen the word coach advisedly -a coach does more than diagnose
problems; a coach suggests plays. And great coaches make their suggestions with
deep knowledge of the situation in the game and all the players’ strengths and
weaknesses. (p. 132)
THE MODELER AS REFLECTOR. “Teamwork in Group Model Building” (Richardson and
Andersen, 1995) polishes the role of the system dynamics modeler:
[T]he modeler/reflector [acts] not as a master modeler but more as a reflector on
the group’s discussion, a “contemplator” whose job [is] to refine and crystallize
the thinking of the group. We came to understand that the role of the
modeler/reflector is more general than that of modeler and that there is great value
to having a person reflecting on the group’s thinking and reflecting it back to
them. The modeler/reflector can perceive subtleties the facilitator might miss, can
identify linkages and systems insights that emerge only from reflection, and can
punctuate the discussion with points of important emphasis. (p. 124)
This approach does not allow for the modeler to claim ownership of (or responsibility
over) the client’s model. As described in Reagan et al. (1991):
The analyst [modeler] function[s] as a critical outsider whose role [is] to ask
penetrating questions, show the decision makers how to think about the problem
in new ways, discover and resolve inconsistencies, and enhance the decision
makers’ emerging understanding (p. 55).
They add:
It is often quite difficult for those decision makers to place their trust in an
unfamiliar modeling technique and depart from their accustomed ways of thinking
about the problem. Y et it is this process of rethinking the all-too-familiar problem
in an unconventional way that contributes substantially to improved decision
making. (p. 63)
The role of the modeler/reflector is to best fit the problem at hand so that it may be
analyzed within the chosen method of analysis. The role of the facilitator, on the other hand, is to
make sure that problem identification and elicitation are not biased in the direction of a particular
method, but accommodated into the most adequate framework. In this balance, the skilled
modeler is “the one who can best merge problem definition and specification assumptions [e.g.,
selection of levels, identification of causal paths, formulation and parameterization of rate
equations] so as to capture the underlying social reality in an insightful and useful manner”
(Andersen 1980, pp. 63-64). The skilled facilitator is the judge who draws the line beyond which
the work of the modeler becomes intrusive (biasing the analysis or alienating the client). The
facilitator prevents the modeler from taking model ownership away from the client.
Preliminary findings and discussion
In mapping the literature in the group model building genealogy into the conceptual dichotomy
proposed and defined in this paper, I’ve found supporting evidence to the thesis that there may
be two intertwined threads in this new approach to system dynamics modeling involving a group
of people in model construction. Group model building interventions strive both to create a
shared understanding of an interpersonal or inter-organizational problem, in the form of a
“boundary-object” model, and to build a “micro-world” type model that is useful in terms of
organizational redesign.
In the problem identification and definition phase of the modeling process, the extent to
which the modeling team moves from a boundary-object to a micro-world type model depends
upon the clarity of the focus of the intervention. This, in turn, is shaped by the degree of
convergence in the management team, regarding the problem to be modeled and the purpose of
the intervention.
Drawing upon the terminology of the CVA framework, I’d argue that at the start of the
intervention, one should regard the model as a boundary-object, and should stress the
instrumental dimension (see Figure 2), balancing the political with the consensual perspectives.
If there is convergence within the group, leading to clarity and focus in problem definition, one
should begin to regard the model as a micro-world, and shift attention to the consummatory
dimension, balancing the rational with the empirical perspectives. The point of transition
between building a boundary-object or a micro-world model may be best understood in terms of
the parturition of a clear statement of the dynamic hypothesis.
The use of the dichotomy in looking at the problem definition phase of the modeling
process also served to explain the reason for the multiple purposes in group model building. The
“disconnect” between the classic system dynamics approach, emphasizing policy change and
organizational redesign, and the varied purposes of group model building interventions
(alignment, building commitment, decision making) is not in the goals themselves, but in the
nature of the client/audience. Before a client group is ready to engage in informed policy change
and organizational redesign, multiple constituencies and messy problems require preliminary and
intermediate steps in problem definition and analysis. Perhaps, prior to changing the decision
tules and causal structure of a system, the group needs to agree on shared strategies, goals and
objectives.
In the model conceptualization phase of the modeling process, the “bridge” between the
two modes of operation appears to be related to the discussion regarding knowledge elicitation
and mental models. This, in turn, may be shaped by several factors, among which those
identified, by Quinn et al. (1985), as: 1) disciplinary and methodological biases, 2) personal
values, and 3) situational demands (p. 51). For example, a “hard” system dynamicist, and/or a
clearly defined problem and focused purpose for the intervention may shift the model building
task toward a micro-world type model. This implies eliciting theories and facts from the
participants, and using a top-down approach to model building. Altematively, a “soft” system
dynamicist, and/or a messy problem embedded in an environment characterized by high degree
of uncertainty may shift the model building task toward a boundary-object type model. In this
case, the modeling team searches for the group’s views and opinions using a bottom-up approach
to model building.
More likely, there is no pure top-down or bottom-up approach. Instead, a mix of the two
is used in ways that are not clearly explained or understood. Experienced modelers probably
draw upon both approaches. They probably attempt to elicit theories and facts, while being
skeptical about views and opinions. They probably strive for the most parsimonious endogenous
feedback-rich model, while capturing the client’s system with as rich as necessary view that the
participants can agree upon and share. In essence, they strive to draw upon their technological
knowledge and modeling experience to build an insightful model, while at the same time keeping
the clients on board and shifting as much model ownership (and leaning) to the client group as
they possibly can. This is the “art” of model building, a craft perfected with training and
experience. These are also the issues that we must shed light upon, to “add more science to the
group model building craft” (Andersen et al. 1997).
In order to accomplish this, I believe modelers have to be more forthright, and extemalize
how it is that some “leaps” are accomplished. How was it that a particular dynamic hypothesis
was crafted? How did they deal with model scope and boundary issues when conceptualizing the
model with the clients? How do they balance professional and personal ethics with consultant-
client relationship ethics? Where do you draw the line between loyalty to the system dynamics
method, and loyalty to the client’ s requests and needs?
I also believe that these issues should be resolved on a case-by-case basis, with the clients
aware of the tradeoffs, and engaged in the decision-making process. I speculate that awareness of
the distinction between viewing the model, as a boundary-object, as opposed to a micro-world is
beneficial both to the modeling and management teams. Perhaps both pre-intervention
questionnaires (to reveal the client’s expectations from the intervention) as well as post-
intervention questionnaires (to evaluate the process used and the results achieved) should be
used.
Figure 6 contains an illustration of two profiles of decision conferences. Clearly, one
profile portrays a more effective intervention than does the other. But, wouldn't it be important
to contrast before and after profiles? Wouldn’t it be important to discuss with the clients the pre-
intervention expectations, before designing and implementing the intervention? Using the CVA
framework, Rohrbaugh and Eden (1990) propose the need to match the client’s setting with the
consultant’s style and method (pp.45-47). Would it be wise to confront the client with respect to
their expectations? Would it be possible (and advisable) for the modeling-team to adapt to the
clients needs (or desires)?
Consensual
Perspective
Political
Perspective
Rational
Consensual
Perspective
Empirical
Political
Perspective
Rational
Empirical
Perspective Perspective
Perspective Perspective
Figure 6. Two profiles of decision conferences
(Copied from Rohrbaugh 1989, pp. 126-127)
While group model building is deviating from classic system dynamics, this is neither
necessarily good nor bad. It will depend upon how effectively the tension points resulting from
competing values are handled. A gain, borrowing from the CVA framework, I list the competing
values as:
1. An adaptable process leading to a legitimate (representative) model;
2. A goal-centered process leading to an efficient (parsimonious) model;
3. A data-based process leading to an accountable (valid) model; and
4. A participatory process leading to supportability (of implementation) of model (results).
The system dynamics tradition has exercised the balance of the rational and empirical
perspectives (items 2 and 3, respectively), highlighted in the micro-world view. The decision
conferencing tradition has noted the importance of incorporating the political and consensual
perspectives (items 1 and 4, respectively), when involving a group in a decision-making process.
Existing theoretical and applied work in group model building provides evidence that the latter
may be perceived as a boundary-object view of model building. Good group model building
theory and practice should provide the rationale and the guidelines for making this whole
package work for the client.
While the decision conferencing tradition has helped in introducing important elements to
the group model building approach developed in Albany, it may have shifted the attention of the
modeling team to decision-making, as opposed to policy-making. Awareness of the
idiosyncrasies of group model building -the nature of its application and its theoretical
foundations- will help us develop a better canon. Bringing people together, and providing them
with adequate tools and proficient means to understand and work on their problems -effective
group model building practice- may be just what we need to advance knowledge of social
systems.
As a final note, it’s worth pointing out that the concept of a “messy” problem, while key
in understanding the nature of interpersonal disagreements may not be sufficient to explain the
motivation behind group model building. The Albany experience has yet to be explicitly
articulated. The motivation for group model building in Albany is not only related to the issue of
disagreement in the client group, but also lack of knowledge of, and appreciation for,
interdependencies among organizations. The fragmented nature of American government has
created many opportunities and much need for integration of services, and for networking,
collaboration and cooperation among government agencies and nonprofit organizations in
policy-making and implementation.
List of references
Ackoff, R.A. 1974. Redesigning the Future: A Systems Approach to Societal Problems. New
Y ork: Wiley.
Ackoff, R.A. 1979. The future of Operational Research is past. Journal of the Operational
Research Society 30 (2): 93-104.
Adelman, L. 1984. Real-time computer support for decision analysis in a group setting: another
class of decision support systems. Interfaces 14 (2): 75-83.
Akkermans, H.A. and J.A.M. Vennix. 1997. Clients’ opinions on group model-building: an
exploratory study. System Dynamics Review 13 (1): 3-31.
Akkermans, H.A., J.A.M. Vennix and E.A,J.A. Rouwette. 1993. Participative modelling to
facilitate organizational change: a case study. In Proceedings of the 1993 International
Conference of the System Dynamics Society. Cancun, Mexico: 1-10.
Allers R., R. Johnson, D.F. Andersen, T.P. Lee, G.P. Richardson, J. Rohrbaugh and A.A.
Zagonel. 1998. Group model building to support welfare reform: part II, Dutchess
county. In Proceedings of the 1998 International Conference of the System Dynamics
Society. Québec City, Canada.
Andersen, D.F. 1980. How differences in analytic paradigms can lead to differences in policy
conclusions. In J. Randers (Ed.). 1980. Elements of the System Dynamics Method: 61-75.
Cambridge, MA: Productivity Press.
Andersen, D.F. and G.P. Richardson. 1979. “A core curriculum in system dynamics.” Graduate
School of Public Affairs. SUNY . Albany, NY. Published in 1980 as Toward a pedagogy
of system dynamics. In A.A. Legasto Jr., J.W. Forrester and J.M. Lyneis (Eds.). System
Dynamics. TIMS Studies in the Management Sciences 14: 91-106. New Y ork: North-
Holland.
Andersen, D.F. and G.P. Richardson. 1997. Scripts for group model building. System Dynamics
Review 13 (2): 107-129.
Andersen, D.F., G.P. Richardson and J.A.M. Vennix. 1997. Group model building: adding more
science to the craft. System Dynamics Review 13 (2): 187-201.
Argyris, C. 1999. On Organizational Learning. 2" Edition. Malden, MA: Blackwell Publishers.
Bell, J.A. and P.M. Senge. 1980. Methods for enhancing refutability in system dynamics
modeling. In A.A. Legasto Jr., J.W. Forrester and J.M. Lyneis (Eds.). System Dynamics.
TIMS Studies in the Management Sciences 14: 61-73. New Y ork: North-Holland.
Berger P. and T. Luckmann. 1966. The Social Construction of Reality: A Treatise in the
Sociology of Knowledge. New Y ork: Doubleday.
Brewer G.D. and O.P. Hall, Jr. 1973. Policy analysis by computer simulation: the need for
appraisal. Public Policy 21 (3): 343-365.
Bryson, J.M. 1995. Strategic planning for public and nonprofit organizations: A guide to
strengthening and sustaining organizational achievement. San Francisco: Jossey-Bass.
Bryson, J.M. and C.B. Finn. 1995. Creating the future together: developing and using shared
strategy maps. In A. Halachmi and G. Bopuckaert (Eds.). The Enduring Challenges of
Public Management: 247-280. San Francisco, CA: Jossey-Bass.
Buede, D.M. and T.A. Bresnick. 1992. Applications of decision analysis to the military systems
acquisition process. Interfaces 22: 110-125.
Cavaleri, S. and J.D. Sterman. 1997. Towards evaluation of systems thinking interventions: a
case study. System Dynamics Review 13 (2): 171-186.
Carper, W.B. and T.A. Bresnick. 1989. Strategic planning conferences. Business Horizon
(September- October): 34-40.
Coyle, G. 2000. Qualitative and quantitative modelling in system dynamics: some research
questions. System Dynamics Review 16 (3): 225-244.
Coyle, G. 2001. Rejoinder to Homer and Oliva. System Dynamics Review 17 (4): 357-363.
de Geus, A.P. 1994. Foreword: Modeling to predict or to learn? In Morecroft, J.D.W. andJ.D.
Sterman (Eds.). Modeling for Learning Organizations: xiii-xvi. Portland, OR:
Productivity Press.
Decision Techtronics Group - DTG. 1984. “Design of a system dynamics model: the
implications of a dues increase at the National Association of Social Workers.” Report.
Rockefeller College, SUNY, Albany, NY 12222, USA.
Decision Techtronics Group - DTG. 1985. “Addressing alcoholism treatment program needs in
New Y ork State: a service delivery model.” Report. Rockefeller College, SUNY, Albany,
NY 12222, USA.
Decision Techtronics Group - DTG. 1987. Medical malpractice insurance: policy implications
and evaluations. Volumes I (Conference results) and II (Technical appendices). Report.
Rockefeller College, SUNY, Albany, NY 12222, USA.
DeSanctis G. and R.B. Gallupe. 1987. A foundation for the study of group decision support
systems. Management Science 33: 589-609.
Eden, C. 1989. Strategic options development and analysis. In J. Rosenhead (Ed.). Rational
Analysis in a Problematic World. London, England: John Wiley & Sons.
Eden, C. 1990. The unfolding nature of group decision support: two dimensions of skill. In C.
Eden and J. Radford (Eds.). Tackling Strategic Problems: The Role of Group Decision
Support: 48-52. London, England: SAGE Publications.
Eden, C. and F. Ackermann. 1992. Strategy development and implementation: the role of a
group decision support system. In R.P. Bostrom, R.T. Watson and S.T. Kinney (Eds.).
Computer Augmented Teamwork: A Guided Tour. New Y ork: Van Nostrand Reinhold.
Eden, C., S. Jones and D. Sims. 1983. Messing About in Problems: An Informal Structured
Approach to their Identification and Management. Oxford, England: Pergamon Press.
Ford, A. 1997. System dynamics and the electric power industry. System Dynamics Review 13
(1): 57-85.
Ford, D.N. andJ.D. Sterman. 1998. Expert knowledge elicitation to improve formal and mental
models. System Dynamics Review 14 (4): 309-340.
Forrester, J.W. 1961. Industrial Dynamics. Portland, Oregon: Productivity Press.
Forrester, J.W. 1969. Urban Dynamics. Waltham, MA: Pegasus Communications.
Forrester, J.W. 1971(A). Principles of Systems. Portland, Oregon: Productivity Press.
Forrester, J.W. 1971(B). World Dynamics. Waltham, MA: Pegasus Communications.
Forrester, J.W. 1975. Market growth as influenced by capital investment. In J.W. Forrester.
Collected Papers: 111-132. Portland, OR: Productivity Press.
Forrester, J.W. 1987(A). Lessons from system dynamics modeling. System Dynamics Review 3
(2): 136-149.
Forrester, J.W. 1987(B). 14 “obvious truths”. System Dynamics Review 3 (2): 156-159.
Forrester, J.W. 1992. Policies, decisions, and information sources for modeling. European
Journal of Operational Research 59 (1): 42-63. Also published in Morecroft, J.D.W. and
J.D. Sterman (Eds.). 1994. Modeling for Learning Organizations: 51-84. Portland, OR:
Productivity Press.
Forrester, J.W. 1994. System dynamics, systems thinking, and soft OR. System Dynamics Review
10 (2-3): 245-256.
Forrester, J.W., NJ. Mass and G.W. Low. 1974. The debate on world dynamics: a response to
Nordhaus. Policy Sciences 5 (2): 169-190.
Forrester, J.W. and P.M. Senge. 1980. Tests for building confidence in system dynamics models.
InA.A. Legasto Jr., J.W. Forrester and J.M. Lyneis (Eds.). System Dynamics. TIMS
Studies in the Management Sciences 14: 209-228. New Y ork: North-Holland.
Gardiner, P.C. and A. Ford. 1980. Which policy run is best, and who says so? In A.A. Legasto
Jr., J.W. Forrester and J.M. Lyneis (Eds.). System Dynamics. TIMS Studies in the
Management Sciences 14: 241-257. New Y ork: North-Holland.
Graham, A.K. 1980. Parameter estimation in system dynamics modeling. In A.A. Legasto Jr.,
J.W. Forrester and J.M. Lyneis (Eds.). System Dynamics. TIMS Studies in the
Management Sciences 14: 125-142. New Y ork: North-Holland. Also in J. Randers (Ed.).
1980. Elements of the System Dynamics Method: 143-161. Cambridge, MA: Productivity
Press.
Griffith, T.L., M.A. Fuller and G.B. Northcraft. 1998. Facilitator influence in group support
systems: intended and unintended effects. Inform Sys Res 9: 20-36.
Hodgson, A.M. 1994. Hexagons for systems thinking. In Morecroft, J.D.W. and J.D. Sterman
(Eds.). Modeling for Learning Organizations: 359-374. Portland, OR: Productivity Press.
Homer, J. 1996. Why we iterate: scientific modeling in theory and practice. System Dynamics
Review 12 (1): 1-19.
Homer, J. and R. Oliva. 2001. Maps and models in system dynamics: a response to Coyle.
System Dynamics Review 17 (4): 347-355.
Huz, S., D.F. Andersen, G.P. Richardson and R. Boothroyd. 1997. A framework for evaluating
systems thinking interventions: an experimental approach to mental health system
change. System Dynamics Review 13 (2): 149-169.
Janis, I. 1982. Groupthink: Psychological Studies of Policy Decisions and Fiascoes, 2" ed.
Boston: Houghton Mifflin.
Kim, D.H. and P.M. Senge. 1994. Putting systems thinking into practice. System Dynamics
Review 10 (2-3): 277-290.
Kraemer, K.L. andJ.L. King. 1988. Computer-based systems for cooperative work and group
decision making. ACM Comput Survey 20: 115-146.
Lane, D.C. 1993. The road not taken: observing a process of issue selection and model
conceptualization. System Dynamics Review 9 (3): 239-264.
Lane, D.C. 1994. Modeling as leaming: a consultancy methodology for enhancing learning in
management teams. In J.D.W. Morecroft and J.D. Sterman (Eds.). Modeling for Learning
Organizations: 85-117. Portland, Oregon: Productivity Press.
Legasto Jr., A.A., J.W. Forrester and J.M. Lyneis (Eds.). 1980. System Dynamics. TIMS Studies
in the Management Sciences 14. New Y ork: North-Holland.
McCartt, A.T. andJ. Rohrbaugh. 1989. Evaluation of group decision support effectiveness: a
performance study of decision conferencing. J ournal of Decision Support Systems 5: 243-
253.
McCartt, A.T. and J. Rohrbaugh. 1995. Managerial openness to change and the introduction of
GDSS: explaining initial success and failure in decision conferencing. Organization
Science 6 (5): 569-584.
Meadows, D.H., D.L. Meadows and J. Randers. 1992. Beyond the Limits. Post Mills, VT:
Chelsea Green Publishing Company.
Milter, R.G. 1986. An exploration of criteria for effective decision making in organizations and
an evaluation of a decision support system: a case study approach. Doctoral dissertation.
University at Albany, State University of New Y ork.
Milter R.G. and J. Rohrbaugh. 1985. Microcomputers and strategic decision making. Public
Productivity Review 9 (Summer/Fall): 215-226.
Moore A.B, andJ.A. Feldt. 1993. Facilitating Community and Decision-Making Groups.
Malabar, FL: Krieger.
Mooy, R., E. Rouwette, GJ. Valk, J.A.M. Vennix, and A. Maas. 2001. Quantification and
evaluation issues in group model building: an application to human resource management
transition. In Proceedings of the 2001 International Conference of the System Dynamics
Society. Atlanta, U.S.A.
Morecroft, J.D.W., D.C. Lane and P.S. Viita. 1991. Modeling growth strategy in a biotechnology
startup firm. System Dynamics Review 7 (2): 93-116.
Morecroft, J.D.W. andJ.D. Sterman (Eds.). 1994. Modeling for Learning Organizations.
Portland, OR: Productivity Press.
Nordhaus, B. 1973. Wold dynamics: measurement without data. Economic Journal (December).
Nunamaker, J.F., A.R. Dennis, J.S. Valacich, D.R. Vogel and J.F. George. 1991. Electronic
meeting systems to support group work. Commun ACM 34: 40-61.
Papert, S. 1980. Mindstorms. New Y ork: Basic Books.
Parsons, T. 1959. General theory in sociology. In R. Merton, L. Broom and L.S. Cottrell, Jr.
(Eds.). Sociology Today: Problems and Prospects. New Y ork: Basic Books.
Phillips L.D. 1984. Decision support for managers. In H.J. Otway and M. Peltu (Eds.). The
Managerial Challenge of New Office Technology. London, England: Butterworths.
Phillips L.D. 1985. Systems for solutions. Datamation Business (April): 26-29.
Phillips L.D. 1986. Computing to consensus. Datamation International 68 (October): 2-6.
Phillips L.D. 1988. People-centered group decision support systems. In G. Doukidis, F. Land and
G. Miller (Eds.). Knowledge-Based Management Support Systems. London, England:
Harwood.
Phillips LD and Phillips MC. 1983. Facilitated work groups: theory and practice. J ournal of the
Operations Research Society 44: 533-549.
Quaddus M.A., DJ. Atkinson and M. Levy. 1992. An application of decision conferencing to
strategic planning for a voluntary organization. Interfaces 22: 61-67.
Quinn R. and J. Rohrbaugh. 1983. A spatial model of effectiveness criteria: towards a competing
values approach to organizational analysis. Management Science 29: 363-377.
Quinn R., J. Rohrbaugh and M.R. McGrath. 1985. Automated decision conferencing: how it
works. Personnel 62 (6): 49-55.
Randers, J. (Ed.). 1980(A). Elements of the System Dynamics Method. Cambridge, MA:
Productivity Press.
Randers, J. 1980(B) Guidelines for model conceptualization. In J. Randers (Ed.). Elements of the
System Dynamics Method: 117-139. Cambridge, MA: Productivity Press.
Reagan, P. and J. Rohrbaugh. 1990. Group decision process effectiveness: a competing values
approach. Group Organizational Studies 15: 20-43.
Reagan-Cirincione, P., S. Schuman, G.P. Richardson, G.P. and S.A. Dorf. 1991. Decision
modeling: tools for strategic thinking. Interfaces 21 (6): 52-65.
Richardson, G.P. 1996(A). Definition of system dynamics. In S.I. Gass and C.M. Harris (Eds.).
Encyclopedia of Operations Research and Management Science: 656-660. Boston, MA:
Kluwer Academic Publishers.
Richardson, G.P. (Ed.). 1996(B). Modelling for Management (2 volumes). Aldershot, UK:
Dartmouth Publishing Co.
Richardson, G.P. 1999. Citation for winner of the 1999 Jay Wright Forrester A ward: Jac A.M.
Vennix. System Dynamics Review 15 (4): 375-377.
Richardson, G.P. and D.F. Andersen. 1995. Teamwork in group model building. System
Dynamics Review 11 (2): 113-137.
Richardson, G.P., D.F. Andersen, J. Rohrbaugh and W. Steinhurst. 1992. Group model building.
In Proceedings of the 1992 International Conference of the System Dynamics Society.
Utrecht, The Netherlands: 595-604.
Richardson, G.P. and A.L. Pugh III. 1981. Introduction to System Dynamics Modeling with
DYNAMO. Cambridge, MA: Productivity Press.
Richardson, G.P. and P.M. Senge. 1989. Corporate and statewide perspectives on the liability
insurance crisis. In Proceedings of the 1989 International Conference of the System
Dynamics Society. Stuttgart, Germany.
Richardson, G.P., J.A.M. Vennix, D.F. Andersen, J. Rohrbaugh and W.A. Wallace. 1989.
Eliciting group knowledge for model-building. In Proceedings of the 1989 International
Conference of the System Dynamics Society. Stuttgart, Germany: 343-357.
Richardson, G.P., Wolstenholme, E.F. and Morecroft, J.D.W. (Eds.). 1994(A ). Systems Thinkers,
Systems Thinking. System Dynamics Review, 10 (2-3).
Richmond, B. 1987. “The Strategic Forum: from vision to strategy to operating policies and back
again”. Lyme, NH: High Performance Systems. Published in 1997. The Strategic Forum:
aligning objectives, strategy and process. System Dynamics Review 13 (2): 131-148.
Roberts, E.B. (Ed.). 1978(A). Managerial Applications of System Dynamics. Cambridge, MA:
The MIT Press.
Roberts, E.B. 1978(B). Strategies for effective implementation of complex corporate models. In
Roberts, E.B. (Ed.). Managerial Applications of System Dynamics: 77-85. Cambridge,
MA: The MIT Press.
Roberts, E.B. 1978(C). Some insights into implementation. In Roberts, E.B. (Ed.). Managerial
Applications of System Dynamics: 155-161. Cambridge, MA: The MIT Press.
Roberts, N., D. Andersen, R. Deal, M. Garet and W. Shaffer. 1983. Introduction to Computer
Simulation: A System Dynamics Modeling Approach. Portland, Oregon: Productivity
Press.
Rogers J., R. Johnson, A.A. Zagonel, J. Rohrbaugh, D.F. Andersen, G.P. Richardson and T.P.
Lee. 1997. Group model building to support welfare reform in Cortland county. In
Proceedings of the 1997 International Conference of the System Dynamics Society.
Istanbul, Turkey: 605-612.
Rohrbaugh, J. 1979. Improving the quality of group judgment: social judgment analysis and the
delphi technique. Organizational Behavior and Human Performance 24: 73-92.
Rohrbaugh, J. 1981. Improving the quality of group judgment: social judgment analysis and the
nominal group technique. Organizational Behavior and Human Performance 28: 272-
288.
Rohrbaugh, J. 1987. Assessing the effectiveness of expert teams. In J.L. Mumpower, L.D.
Phillips, O. Renn and V.R.R. Uppuluri (Eds.). Expert judgment and expert systems.
Berlin: Springer-Verlag.
Rohrbaugh J. 1989. Demonstration experiments in field settings: assessing the process, not the
outcome, of group decision support. In I. Benbasat (Ed.), The Information Systems
Research Challenge: Experimental Research Methods, Vol 2. Cambridge: Harvard
Business School Publishing.
Rohrbaugh, J. 1992. Cognitive challenges and collective accomplishments. In R.P. Bostrom,
R.T. Watson and S.T. Kinney (Eds.). Computer Augmented Teamwork: A Guided Tour:
299-324. New Y ork: Van Nostrand Reinhold.
Rohrbaugh, J. 2000. The use of system dynamics in decision conferencing: implementing
welfare reform in New Y ork State. In G. David Garson (ed.). Handbook of Public
Information Systems: 521-533. New Y ork: Marcel Dekker.
Rohrbaugh, J. and C. Eden. 1990. Using the competing values approach to explore ‘ways of
working’. In C. Eden and J. Radford (Eds.). Tackling Strategic Problems: The Role of
Group Decision Support: 40-47. London, England: SAGE Publications.
Rouwette, E., J.A.M. Vennix, and T. van Mullekom. 1999. Group model-building: a review of
assessment studies. In Proceedings of the 1999 International Conference of the System
Dynamics Society. Wellington, New Zealand. Published in 2002. Group model-building
effectiveness: a review of assessment studies. System Dynamics Review 18 (1): 5-45.
Senge, P.M. 1990. The fifth discipline: the art and practice of the learning organization. New
Y ork: Doubleday.
Schuman, S. and J. Rohrbaugh. 1991. Decision conferencing for systems planning. Information
and Management 21: 147-159.
Simon, H. 1957. Administrative Behavior: A Study of Decision-Making Processes in
Administrative Organizations, 2™ ed. New Y ork: Macmillan.
Skinner, B.F, 1971. Beyond Freedom and Dignity. New Y ork: Bantam Books.
Stenberg, L. 1980. A modeling procedure for public policy. In J. Randers (Ed.). Elements of the
System Dynamics Method: 292-312. Cambridge, MA: Productivity Press.
Sterman, J.D. 1994. Learning in and about complex systems. System Dynamics Review 10 (2-3):
291-330.
Sterman, J.D. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World.
Boston, MA: The McGraw-Hill Companies, Inc.
Taggart, W. and D. Robey. 1981. Minds and managers: on the dual nature of human information
processing and management. Academy of Management Review 6: 187-195.
Thomas W.I and D.S. Thomas. The Child in America: Behavior Problems and Programs. New
Y ork: Alfred Knopf.
Vennix, J.A.M. 1994. Building consensus in strategic decision-making: insights from the process
of group model building. In Proceedings of the 1994 International Conference of the
System Dynamics Society. Stirling, Scotland.
Vennix, J.A.M. 1996. Group Model Building: Facilitating Team Learning Using System
Dynamics. London, England: John Wiley & Sons.
Vennix, J.M.A. 1999. Group model-building: tackling messy problems. System Dynamics
Review 15 (4): 379-401.
Vennix, J.A.M , H.A. Akkermans and E.A.J.A. Rouwette. 1996. Group model building to
facilitate organizational change: an exploratory study. System Dynamics Review 12 (1):
39-58.
Vennix, J.M.A., D.F. Andersen and G.P. Richardson (Eds.). 1997. Group Model Building.
System Dynamics Review 13 (2): 103-106.
Vennix, J.A.M., D.F. Andersen, G.P. Richardson and J. Rohrbaugh. 1992. Model-building for
group decision support: issues and alternatives in knowledge elicitation. European
Journal of Operational Research 59 (1): 28-41. Also 1994. InJ.D.W. Morecroft and J.D.
Sterman (Eds.). Modeling for Learning Organizations: 29-49. Portland, Oregon:
Productivity Press.
Vennix, J.A.M. andJ.W. Gubbels. 1994. Knowledge elicitation in conceptual model building: a
case study in modeling a regional Dutch health care system. In J.D.W. Morecroft and J.D.
Sterman (Eds.). Modeling for Learning Organizations: 121-145. Portland, Oregon:
Productivity Press.
Vennix, J.A.M., J.W. Gubbels, D. Post and HJ. Poppen. 1988. A structured approach to
knowledge acquisition in model development. In Proceedings of the 1988 International
Conference of the System Dynamics Society. La Jolla, U.S.A. Published in 1990. A
structured approach to knowledge elicitation in conceptual model building. System
Dynamics Review 6 (2): 194-208.
Vennix, J.A.M. and W. Scheper. 1990. Modeling as organizational learning: an empirical
perspective. In Proceedings of the 1990 International Conference of the System
Dynamics Society. Boston, Massachusetts.
Vennix, J.A.M., W. Scheper and R. Willems. 1993. Group model-building: what does the client
think of it?. In Proceedings of the 1993 International Conference of the System Dynamics
Society. Cancun, Mexico: 534-543.
Weil, H.B. 1980. The evolution of an approach for achieving implemented results from system
dynamics projects. In J. Randers (Ed.). Elements of the System Dynamics Method: 271-
291. Cambridge, MA: Productivity Press.
Weiss, J.J. and G.W. Zwahlen. 1982. The structured decision conference: a case study. Hospital
Health Service Administration 27: 90-105.
Winch, G.W. 1993. Consensus building in the planning process: benefits from a “hard” modeling
approach. System Dynamics Review 9 (3): 287-300.
Wolstenholme, E.F. 1994. A Systematic approach to model creation. In J.D.W. Morecroft and
J.D. Sterman (Eds.). Modeling for Learning Organizations: 175-194. Portland, Oregon:
Productivity Press.
Appendices
Key references organized according to the clusters identified in the genealogy
1. System dynamics
Y ear: Author(s): Title:
1961 Forrester Industrial Dynamics
1971-A_| Forrester Principles of Systems
1978-A_| E. Roberts (Ed.) Managerial Applications of System Dynamics
1978-B | E. Roberts Strategies for effective implementation of complex corporate
models
1978-C_| E. Roberts Some insights into implementation
1979/80 | Andersen A core curriculum in system dynamics (Toward a pedagogy of
Richardson system dynamics)
1980 Legasto System Dynamics
Forrester
Lyneis (Eds.)
1980 Bell Methods for enhancing refutability in system dynamics
Senge modeling
1980 Forrester Tests for building confidence in system dynamics models
Senge
1980 Gardiner Which policy run is best, and who says so?
Ford
1980 Graham Parameter estimation in system dynamics modeling
1980-A | Randers (Ed.) Elements of the System Dynamics Method
1980-B | Randers Guidelines for model conceptualization
1980 Stenberg A modeling procedure for public policy
1980 Weil The evolution of an approach for achieving implemented results
from system dynamics projects
1981 Richardson Introduction to System Dynamics Modeling with DYNAMO
Pugh
1983 N. Roberts Introduction to Computer Simulation: A System Dynamics
Andersen Modeling Approach
Deal
Garet
Shaffer
1987-A_| Forrester Lessons from system dynamics modeling
1987-B | Forrester 14 “obvious truths”.
1992/94 | Forrester Policies, decisions, and information sources for modeling
1994 Sterman Learning in and about complex systems
1996-A | Richardson Definition of system dynamics
2000 Sterman Business Dynamics: Systems Thinking and Modeling for a
Complex World
Key references organized according to the clusters identified in the genealogy (cont.)
2. Direct system dynamics modeling with clients
Y ear: Author(s): Title:
1980 Stenberg A modeling procedure for public policy
1987/97 | Richmond The Strategic Forum: from vision to strategy to operating
policies and back again (aligning objectives, strategy and
process)
1988/90 | Vennix A structured approach to knowledge acquisition in model
Gubbels development; A structured approach to knowledge elicitation in
Post conceptual model building
Poppen
1989 Richardson Corporate and statewide perspectives on the liability insurance
Senge crisis
1990 Senge The fifth discipline: the art and practice of the learning
organization
1991 Morecroft Modeling growth strategy in a biotechnology startup firm
Lane
Viita
1993 Lane The road not taken: observing a process of issue selection and
model conceptualization
1993 Winch Consensus building in the planning process: benefits from a
“hard” modeling approach
1994 Morecroft Modeling for Learning Organizations
Sterman (Eds.)
1994 Morecroft Executive knowledge, models, and learning
1994 Lane Modeling as learning: a consultancy methodology for enhancing
learning in management teams
1994 Vennix Knowledge elicitation in conceptual model building: a case
Gubbels study in modeling a regional Dutch health care system
1994 Wolstenholme A systematic approach to model creation
1994 Richardson Systems Thinkers, Systems Thinking
Wolstenholme
Morecroft (Eds.)
1994 Forrester System dynamics, systems thinking, and soft OR
1994 Kim Putting systems thinking into practice
Senge
1997 Cavaleri Towards evaluation of systems thinking interventions: a case
Sterman study
1998 D. Ford Expert knowledge elicitation to improve formal and mental
Sterman models
2000 Sterman Business Dynamics: Systems Thinking and Modeling for a
Complex World
Key references organized according to the clusters identified in the genealogy (cont.)
3. Decision conferencing
Y ear: Author(s): Title:
1979 Rohrbaugh Improving the quality of group judgment: social judgment
analysis and the delphi technique
1981 Rohrbaugh Improving the quality of group judgment: social judgment
analysis and the nominal group technique
1983 Eden Messing About in Problems: An Informal Structured Approach
Jones to their Identification and Management
Sims
1984 Phillips Decision support for managers
1984 Adelman Real-time computer support for decision analysis in a group
setting: another class of decision support systems
1985 Phillips Systems for solutions
1985 Milter Microcomputers and strategic decision making
Rohrbaugh
1985 Quinn Automated decision conferencing: how it works
Rohrbaugh
McGrath
1986 Phillips Computing to consensus
1988 Phillips People-centered group decision support systems
1989 Carper Strategic planning conferences
Bresnick
1989 McCartt Evaluation of group decision support effectiveness: a
Rohrbaugh performance study of decision conferencing
1989 Rohrbaugh Demonstration experiments in field settings: assessing the
process, not the outcome, of group decision support
1990 Eden The unfolding nature of group decision support: two dimensions
of skill
1990 Reagan Group decision process effectiveness: a competing values
Rohrbaugh approach
1991 Schuman Decision conferencing for systems planning
Rohrbaugh
1992 Rohrbaugh Cognitive challenges and collective accomplishments
1995 McCartt Managerial openness to change and the introduction of GDSS:
Rohrbaugh explaining initial success and failure in decision conferencing
Key references organized according to the clusters identified in the genealogy (cont.)
4, System dynamics modeling used in decision conferences
Y ear: Author(s): Title:
1984 DTG Design of a system dynamics model: the implications of a dues
increase at the National Association of Social Workers
1985 DTG Addressing alcoholism treatment program needs in New Y ork
State: a service delivery model
1987 DTG Medical malpractice insurance: policy implications and
evaluations
1989 Richardson Corporate and statewide perspectives on the liability insurance
Senge crisis
1991 Reagan-Cirincione | Decision modeling: tools for strategic thinking
Schuman
Richardson
Dorf
2000 Rohrbaugh The use of system dynamics in decision conferencing:
implementing welfare reform in New Y ork State
Key references organized according to the clusters identified in the genealogy (cont.)
5. Group model building
Y ear: Author(s): Title:
1988/90 | Vennix A structured approach to knowledge acquisition in model
Gubbels development; A structured approach to knowledge elicitation in
Post conceptual model building
Poppen
1989 Richardson Eliciting group knowledge for model-building
Vennix
Andersen
Rohrbaugh
Wallace
1990 Vennix Modeling as organizational learning: an empirical perspective
Scheper
1992/94 | Vennix Model-building for group decision support: issues and
Andersen alternatives in knowledge elicitation
Richardson
Rohrbaugh
1992 Richardson Group model building
Andersen
Rohrbaugh
Steinhurst
1993 Akkermans Participative modelling to facilitate organizational change: a
Vennix case study
Rouwette
1993 Vennix Group model-building: what does the client think of it?
Scheper
Willems
1994 Vennix Knowledge elicitation in conceptual model building: a case
Gubbels study in modeling a regional Dutch health care system
1994 Vennix Building consensus in strategic decision-making: insights from
the process of group model building
1995 Richardson Teamwork in group model building
Andersen
1996 Vennix Group Model Building: Facilitating Team Learning Using
System Dynamics
1996 Vennix Group model building to facilitate organizational change: an
Akkermans exploratory study
Rouwette
1997 Akkermans Clients’ opinions on group model-building: an exploratory study
Vennix
Key references organized according to the clusters identified in the genealogy (cont.)
5. Group model building (continued)
Y ear: Author(s): Title:
1997 Vennix Foreword: Group model building, art, and science. Group Model
Andersen Building
Richardson (Eds.)
1997 Andersen Scripts for group model building
Richardson
1997 Huz A framework for evaluating systems thinking interventions: an
Andersen experimental approach to mental health system change
Richardson
Boothroyd
1997 Andersen Group model building: adding more science to the craft
Richardson
Vennix
1997 Rogers Group model building to support welfare reform in Cortland
Johnson county
Zagonel
Rohrbaugh
Andersen
Richardson
Lee
1998 Allers Group model building to support welfare reform: part II,
Johnson Dutchess county
Andersen
Lee
Richardson
Rohrbaugh
Zagonel
1999 Richardson Citation for winner of the 1999 Jay Wright Forrester A ward: Jac
A.M. Vennix
1999 Vennix Group model-building: tackling messy problems
1999/02 | Rouwette Group model-building effectiveness: a review of assessment
Vennix studies
van Mullekom
2000 Rohrbaugh The use of system dynamics in decision conferencing:
implementing welfare reform in New Y ork State
2001 Mooy Quantification and evaluation issues in group model building: an
Rouwette application to human resource management transition
Valk
Vennix
Maas
Table 5a. A dichotomous view of models in problem identification and definition.
The Problem
Question: How do intervenors and participants of group model building interventions view the
model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
Preexisting problems:
A model should be designed to answer a
specific, tangible and meaningful question, or
set of questions. (Forrester 1961, p. 449)
Develop a model to solve a particular problem,
not to model the system. (Sterman 2000, p. 79)
A meaningful system dynamics problem is a
relevant and dynamically complex problem,
embedded in a feedback-rich system. (Stenberg
1980, Richardson and Pugh 1981, Reagan et
al. 1991, Sterman 2000)
Socially constructed problems:
Problems are interrelated and, given multiple
constituencies, there is room for ambiguity in
problem selection and analysis. (Reagan et al.
1991, p. 52)
Sometimes people will not even agree that
there is a problem, much less what it is.
(Vennix 1996, p. 13)
Sometimes the “real” problem does not emerge
until the group model-building process is
underway. (Andersen et al. 1997, p. 194)
Interrelation and ambiguity in problems adds
an additional layer of complexity to already
complex situations. (Vennix 1996, p. 1)
Table 5b. A dichotomous view of models in problem identification and definition.
The Purpose
Question: How do intervenors and participants of group model building interventions view the
model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
“The” Purpose:
A systems study must be for a purpose if it is
to be productive. (Forrester 1961, p. 449)
A model without a purpose is like a ship
without a sail. (Richardson and Pugh 1980, p.
38)
The goal is to improve performance of the
system. (Sterman 2000, p. 80)
The main purpose of system dynamics
modeling is to aid in designing better
management systems. (Forrester 1961)
Multiple purposes:
The purpose of the intervention is to provide a
venue for negotiation and alignment to occur,
adding rigor to the discussion, and providing
participants with means to keep track of
complex causal structures, and serving as a
group memory of their understanding. (Huz et
al. 1997, Vennix 1999)
Modeling helps to create a shared perspective
and understanding of the clients’ issue. (Lane
1994, p. 110)
In modeling messy problems, the most
important goal is the creation of a shared
reality and problem definition among problem
owners. (Vennix 1996, p. 24)
The model becomes a boundary- object in this
negotiation.
The model can also serve as a tool to
investigate potential lines of action.
(Richardson and Senge 1989, Reagan et al.
1991)
The model-based analysis is useful if it helps
the group reach a consensual decision about
what to do. (Winch 1993)
Table 5c. A dichotomous view of models in problem identification and definition.
The Client/A udience
Question: How do intervenors and participants of group model building interventions view the
model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
A monolithic client/audience:
The modeling process should be focused on the
clients’ needs. (Sterman 2000, p. 85)
Solve a real problem that presents an
opportunity perceived as important to the
clients. (Roberts 1978-B, pp. 78-79)
Active client involvement is essential to ensure
adequacy and accuracy of model formulation
with respect to reality, and to provide a basis
for implementation of resulting recommended
changes. (E. Roberts 1978-C, p. 156)
Stakeholders/multiple constituencies:
Different people define and give shape to
problems differently. Multiple constituencies
using multiple criteria, and multiple resources
and constraints, cause ambiguity in problem
selection and analysis. (Reagan et al 1991)
Stakeholders define and give shape to a
socially constructed problem that emerges as
an agreement from discussion and negotiation.
The way the problem gets defined depends on
who’s in the room.
Social sources of messy problems are related to
deficient patterns of social interaction and
communication, which fail, in and of
themselves, to demystify the illusions formed
in the mental models of individuals. (V ennix
1999, pp. 386 and 387)
Before we can set out course to solve “real”
problems, we have to struggle upon a shared
understanding of what real is.
Table 6a. A dichotomous view of models in model conceptualization.
The role of the structuring-framework
Question: How do intervenors and participants of group model building interventions
view the model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
The goal is to test the dynamic hypothesis:
What kind of structuring-framework? How
The beginner usually fails to realize the
importance of an initial hypothesis. There is
often a feeling that to propose modes of
dynamic behavior before a system model is
constructed is to prejudge the answers. This is
exactly what is needed. We start with a
hypothesis for behavior. We build a model to
see if the mode of behavior could exist and
whether or not it can result from the initial
assumptions. (Forrester 1961, p. 450)
The goal of the conceptualization stage is to
arrive at a rough conceptual model capable of
addressing a relevant problem. The reference
mode acts as a catalyst in the transition from
general speculation about a problem to an
initial model. This transition is the major
creative step in modeling. (Randers 1980-B,
pp. 130 and 136)
The first task in formulating a model is the test
of the dynamic hypothesis, which is a
preliminary check to see that the basic
mechanism included in the conceptual model
actually reproduce the reference modes.
(Randers 1980-B, pp. 130-131)
Classic system dynamics advocates a top-down
approach to model conceptualization, that
seeks to conceive the key pieces of causal
structure capable of reproducing key reference
modes of dynamic behavior.
(Continued in the next two pages... )
much structuring?
Groups are more likely to use models when it
is clear to them that their ideas and knowledge
is represented in the model, and when models
do not seem to overly restrict team thinking.
(Morecroft 1994, p. 4)
People learn through discovering for
themselves. People make up their own minds.
(Morecroft 1994, p. 4; de Geus 1994, p. xiv)
Whereas a simple list just captures items of
knowledge, a framework packages and
organizes knowledge. A framework also filters
knowledge because some ideas won't easily fit
within the constraints of the framework. So,
although modelers often say that they are
mapping mental models, really they are not.
They are filtering and organizing from mental
models to fit the modeling framework.
(Morecroft 1994, pp. 9 and 11)
It is important to establish which type of
discussion framework will suit the client best.
(Lane 1994, p. 104)
Flexible approaches should be used to
generate, select, and study the issues -
particularly in the early stages of
interventions- since these reduce any biasing
of the elicitation toward system dynamics, and
also allow the participants to take up the most
appropriate problem structuring approach.
(Lane 1993, pp. 239-240)
Table 6a. A dichotomous view of models in model conceptualization.
The role of the structuring-framework (continued, p. 2/3)
Models as “micro-worlds”:
Models as “boundary- objects”:
Decision models are intellectual tools that have
been developed to make unwieldy problems
more manageable by structuring thought
processes, clarifying interrelationships, and
handling complex data. These tools make the
policy-making process more efficient by
enabling policy makers to rapidly integrate and
analyze information and options and make it
more effective by enabling them to examine
policies and their implications thoroughly.
(Reagan et al. 1991, p. 53)
System dynamics, in specific, is particularly
useful in exploring and understanding
endogenous causes of problematic dynamic
behavior, embedded in feedback rich, complex
systems. (Reagan et al. 1991, p. 54)
The modeling team pressed for some causal
feedback views but did not force an
endogenous dynamic feedback view. In the
end, the client team was left with few insights
about the causal structure of critical parts of the
system. This model-based group work might
be faulted for trying to be too responsive to the
group, and for failing to do a good job
presenting and motivating the system dynamics
approach. (Richardson and Andersen 1995, p.
133)
(More in the next page... )
Explaining the mysteries of system dynamics
or of a particular model formulation can get in
the way of uninhibited group discussion
focused on the problem independent of
approach or formulation. (Richardson and
Andersen 1995, p. 132)
Some modelers often adopt a bottom-up
approach to model building, constructing (with
the client group) a broader shared view of the
system. Rather than holding a narrow focus (in
model conceptualization) on the dynamic
hypothesis.
The modeler collects fragments of structure
that, to begin with, are just lists of key
resources, states and resource flows. Lists are a
good way to capture manager’s own categories
and concepts. These lists generate raw material
for an influence diagram. Wolstenholme’ s
(1994) approach gently shapes a discussion
first into a list and then into a diagram that
eventually shows feedback loops, delays, and
organizational boundaries. (Morecroft 1994,
pp. 23-24)
The most widely used reasons for creating an
external representation of mental models is the
great benefit that can be gained by (naturally)
structuring and sharing information. (Lane
1994, p. 100)
The working model serves as a boundary-
object for discussion and negotiation. The final
model reflects the result of the group’s
structuring- decisions.
Table 6a. A dichotomous view of models in model conceptualization.
The role of the structuring-framework (continued, p. 3/3)
Models as “micro-worlds”:
Models as “boundary- objects”:
The model provides an organizing and
coordinating framework, structuring the
group’s thinking and encouraging them to
make a series of systematic decisions. The
model serves as a decision accounting system.
(Quinn et al. 1985, p. 55; Milter and
Rohrbaugh 1985, p. 221)
The process of model building is frequently
more important then the resulting model.
(Vennix and Gubbles 1994, p. 122)
Different cognitive tasks require different
structuring-frameworks. (Richardson et al.
1989, Vennix et al. 1992/94)
A group model building intervention is
composed of a repertoire of sub-frameworks,
wisely used, embedded within the larger
method, called system dynamics model
building. An effective intervention is one that
appropriately matches the series of model
building tasks with the best structuring-
procedures for knowledge elicitation and group
dialogue. (Lane 1994, Vennix 1996, Andersen
and Richardson 1997)
Table 6b. A dichotomous view of models in model conceptualization.
Knowledge elicitation and mental models
Question: How do intervenors and participants of group model building interventions
view the model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
Eliciting prospective theories and facts:
Eliciting views and opinions:
Simulation models are informed by our mental
models and by information gleaned from the
real world. Strategies, structures, and decision
rules used in the real world can be represented
and tested in the virtual world of the model.
(Sterman 2000, p. 88)
The modeler strives toward a “mental model,”
that is, an understanding of the operation of the
real world. (Randers 1980-B, p. 119)
A mathematical model should be based on the
best information that is readily available, but
the design of a model should not be postponed
until all pertinent parameters have been
accurately measured. In general sufficient
information exists in the descriptive knowledge
possessed by the active practitioners to serve
the model builder in all his initial efforts.
(Forester 1961, p. 58)
The micro-world view of model
conceptualization stresses the importance of a
factual based and empirically accountable
model.
A good modeling process challenges the
clients’ conception. Modelers have a
responsibility to require clients to justify their
opinions and ground their views in data.
(Sterman 2000, p. 85)
We base our models on whatever knowledge
we have -real or imaginary, naive or
sophisticated. The client team may carry
around quite different mental models. It is
these varied models that enter the debate.
(Morecroft 1994, p. 7)
The group model building effort depends on
the thoughts and agendas the client group
brings to the workshop (Richardson and
Andersen 1995, p. 133)
Figure 5, copied from Morecroft (1994, p. 10)
underscores the fragility of the notion that one
can readily elicit theories and facts from the
mental models of participants. The
development of a shared mental model
depends, quite literally, upon “who's in the
room.”
A transitional object that is found acceptable to
a group of people becomes a boundary- object
that reflects the group’s negotiated
representation of reality (i.e., a socially
negotiated order).
Often people simply don’t know how some
processes function. (Vennix and Gubbles 1994,
p. 138; Richardson and Andersen 1995, p. 117)
Rather than eliciting people’s theories as a
starting point, it may be more useful to elicit
their views of the system.
Table 6c. A dichotomous view of models in model conceptualization.
Delineation of model boundary
Question: How do intervenors and participants of group model building interventions
view the model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
Parsimony and dynamic hypothesis quide
Dealing with scope and level of aggregation:
model boundary decisions:
The reference mode indicates the necessary
level of aggregation and the extent of the
system boundary. The modeler should select
and describe the smallest set of feedback loops
considered sufficient to generate the reference
mode. (Randers 1980-B, p. 131)
The behavior of interest must be identified
before the boundary can be determined. Define
the boundary that encloses the smallest number
of components. Ask not if a component is
merely present. Instead, ask if the behavior of
interest will disappear or be improperly
represented if the component is omitted. If the
component can be omitted without defeating
the purpose of the study, the component should
be excluded and the boundary thereby made
smaller. (Forrester 1975, p. 112)
Without the initial hypothesis regarding the
dynamic behavior under study, there is no
basis for deciding what factors might be
important and which ones could be neglected.
(Forrester 1961, p. 450)
The art of model building is knowing what to
cut out, and the purpose of the model acts as
the logical knife. Modelers should not
automatically accede to their clients’ requests
to include more detail or to focus on one set of
issues while ignoring others, just to keep them
on board. (Sterman 2000, pp. 89 and 85)
Some detail is justified in order to provide
apparent reality and easier communication with
others less skilled in model building. (Forrester
1961, p. 453)
The quality of the conceptual model was
increased drastically with the inclusion of more
variables. Although a larger conceptual model
is not necessarily better, the increase was
primarily caused by refinement of the concepts
and relationships in the model. (Vennix and
Gubbels 1994, pp. 139-140)
As models are revisited, variables are seldom
dropped, it being far more likely that
intermediate variables are added to clarify the
nature of causality. (Lane 1994, p. 102-103)
The malleability of models, and their partial fit
to the mental models of participants, leads to a
laundry list of concepts and variables the group
wishes to see incorporated into the full model.
(Richardson and Andersen 1995, p. 121)
In order to put a boundary on the effects to be
included, we model only one issue. We place
the issue in the context of a system and then
include only those aspects of the system that
the client considers to be important or that they
wish to concentrate their study on. There is no
a priori requirement of certainty regarding
quantification, or even cause and effect. (Lane
1994, p. 96)
Table 6d. A dichotomous view of models in model conceptualization.
The role of the modeler/facilitator
Question: How do intervenors and participants of group model building interventions
view the model they are building?
Models as “micro-worlds”:
Models as “boundary- objects”:
The modeler as an expert in the technology; the
The modeler as facilitator; the issue of
“smart” systems thinker:
The modeler brings to the group model
building effort technological skills that must be
exercised diligently and smartly.
The modeler should view the problem and the
system from the proper perspective: not too far,
not too close. The modeler needs to observe
first-hand the system to distinguish espoused
theories from theories in use. (Forrester 1961,
pp. 451 and 452)
Modelers must guard against accepting the
Client’s initial assessment of the appropriate
time frame. (Sterman 2000, p. 94)
A well-trained model builder can do as well as
a group of model builders in tasks such as
proposing formulations or designing feedback
structures. Involving a group may have an
apparent purpose of designing model structure,
but have as a real purpose developing
understanding of the system under study.
(Vennix et al. 1992, p. 33)
Modelers should not be hired guns. Modelers
have ethical responsibilities. Modelers should
“speak truth to power.” The clients are the
people the modeler must influence for his/her
work to have impact. If necessary, the modeler
must quit and get a better client. (Sterman
2000, p. 85)
ownership:
A “modem” view of modeling repositions the
role of the model and the modeler. Models are
“owned” by policymakers, not by technical
experts. They are created in a group process.
The modeler is, in part, a facilitator, one who
designs and leads group processes to capture
team knowledge. (Morecroft and Sterman
1994, p. xvii-xviii)
While the model is an intellectual tool, the
process of modeling is social and political.
(Reagan et al. 1991, p. 53)
I have not met a decision-maker who is
prepared to accept anybody else’s model of
his/her reality. “I’ll make up my own mind” is
pretty universal principle for everyone
embracing the responsibility for his/her
actions. (de Geus 1994, p. xiv)
Rather than attempting to take the position, “I
am an expert in techniques that will teach you
about your business,” the modeler should act
as a facilitation consultant, offering a process
in which the ideas of the team are brought out
and examined in a clear and logical way. (Lane
1994, p. 93)
(More in the next page... )
Table 6d. A dichotomous view of models in model conceptualization.
The role of the modeler/facilitator (continued, p. 2/2)
Models as “micro-worlds”:
Models as “boundary- objects”:
We came to understand that the role of the
modeler/reflector is more general than that of
modeler and that there is great value to having
a person reflecting on the group’s thinking and
reflecting it back to them. The modeler/
reflector can perceive subtleties the facilitator
might miss, can identify linkages and systems
insights that emerge only from reflection, and
can punctuate the discussion with points of
important emphasis. (Richardson and Andersen
1995, p. 124)
The role of the consultant is simply to
encourage clients to put forward their ideas, to
clarify them if necessary, and to record them in
a form that is both permanent and transferable.
We use the term “articulated model.” (Lane
1994, p. 96)
The modeler/reflector acts not as a master
modeler but more as a reflector on the group's
discussion, a “contemplator’ whose job is to
refine and crystallize the thinking of the group.
(Richardson and Andersen 1995, p. 124)
The modeler functions as a critical outsider
whose role is to ask penetrating questions,
show the decision makers how to think about
the problem in new ways, discover and resolve
inconsistencies, and enhance the decision-
makers’ emerging understanding. (Reagan et
al. 1991, p. 55)
Back to the Top
