The 24" International Conference of the System Dynamics Society
Nijmegen, The Netherlands
July 23 -27, 2006
Fresh Insights into System Dynamics Methodology- Developing
an abductive inference perspective
John Barton
John Barton C onsulting &
Department of Management
Monash University, Melbourne, Australia
bartcons@ bigpond.net.au
john.barton@ BusEco.monash.edu.au
Tim Haslett
Department of Management
Monash University, Melbourne, Australia
Abstract:
Issues relating to System Dynamics (SD) method and the validation of SD models are an
important preoccupation of SD practitioners. It is argued that these issues are debated
within the framework set by deductive logic which is appropriate for closed systems, but
not for open systems as typically found in management decision making. Using the early
Forrester- Ansoff and Slevin debate as a prime example (Forrester, 1968; Ansoff and
Slevin, 1968), it is shown that while A nsoff and Slevin argue from the position of
deductive logic which assumes certainty and no environmental change, Forrester is
arguing from an abductive inference framework in which action results from a best
available hypothesis resulting from the development and use of an SD model within a
broader learning-decision making framework.
In addition, it is argued that the familiar events-patterns-structure tool used in SD is a
structured approach to the abduction process. An implication of these arguments is that
debates relating to SD methodology need to shift emphasis from the validation of models
to debates on evaluation of the model development process, the implementation of
strategies based on model-based thinking, and the associated outcomes.
Introduction.
The relevance of abductive inference- the process of forming hypotheses- to System
Dynamics (SD) methodology has been raised previously by Ryan (1996) and Barton
(1999).
This paper provides a further explication of this relevance. Specifically, it provides:
e An introduction to the logic of abductive inference and its relevance to describing
the management process.
e A re-interpretation of the methodological debate between Forrester and Ansoff
and Slevin (Forrester, 1968; Ansoff and Slevin, 1968)
e An interpretation of the Events- Pattems of Events- Structure framework used in
SD as an application of abductive inference.
e Implications for interpreting the role of the SD model within a complete learning
structure with an increased emphasis on evaluation.
Abductive Inference.
Abductive inference is a mode of inference which, along with deduction and induction
dates back to Aristotle but was largely overlooked by Western philosophers, and
generally confused with induction, until the late 19" century. At this time, the founder of
American pragmatist philosophy, Charles Sanders Peirce (1839 -1914), started to
establish abduction as a comerstone of his philosophical framework:
“ Abduction consists in studying facts and devising a theory to explain them” .
(CP 2: 270).
For Peirce, abduction represented a highly creative and perceptual act, not to be confused
with induction:
Deduction proves that something must be; Induction shows that something actually is
operative; Abduction merely suggests that something may be.
(CP 5.171)
In this sense abduction bears a strong resemblance to the “speculative leap” in Einstein's
model for constructing a scientific theory (Figure 1). (Holton, 1998: 28-56).
Einstein described the jump from the observed facts to the set of “Axioms of
Fundamental Principle’- the fundamental hypothesis, as a “Speculative leap based on
hunch, conjecture, inspiration, and guesswork... .We are dealing, after all, with the
private process of theory construction or innovation, the phase not open to inspection by
others and indeed perhaps little understood by the originator himself. But the leap to the
top of the schema symbolizes precisely the precious moment of great energy, the
response to the motivation of “wonder” and the “passion of comprehension”.” (Holton,
1998: 31)!. It is this type of “speculative leap that sets abduction apart from induction.
A; Axiom of fundamental principles
Logical path to give assertions/
predictions
Jy st s2 s3
Speculative leap
Testing against experience
Observable events
af
E v yey
E (chaotic)
1 E2 E3
Schema Cy
Figure 1. Einstein’s Model for C onstructing a Scientific Theory
In addition to Einstein’ s theory of relativity, this “speculative jump” is what characterizes
the great developments in science such as Euclid’s elements, Galileo’ s re-conception of
the planetary system, the Newtonian model, and Descartes’ framework of thought.
While not at such a grand level (but some might debate this!), such “speculative leaps”
are what characterize the deep insights (Senge, 2006) that help conceptualize the seminal
SD models like Forrester’s models of corporate growth, urban dynamics, and world
resource dynamics. (Forrester, 1975).
But in a rather unique manner, SD’s events- pattern- structure methodology and the use
of simulation modeling to guide the search for a causal explanation for an observed
dynamic phenomena represented by reference modes, provides a framework that supports
this search process for deep insights. We can identify this process of hypothesis
formation as abduction.
While the origins of abduction can be traced to Greek dialectic, it was revived by Peirce
(1877, 1878) who used it with deductive and inductive inference to develop a theory of
inquiry. Peirce recognized abduction as the most important of the three modes of
inference and central in his attempt to develop a complete philosophical architectonic’.
1 Consequently, abduction is associated with the process of synthesis, a foundation stone of
systemic thought.
Peirce never completed a final statement of his architectonic but several researchers have
attempted to construct one from Peirce’s extensive writings. For example, Hausman (1993)
argues that Peirce’s pragmaticist architectonic provides:
e A theory of meaning- the pragmatic maxim.
“There are in science three fundamentally different kinds of reasoning, Deduction (called
by Aristotle {synagégé} or {anagogé}), Induction (Aristotle's and Plato's {epagdgé}) and
Retroduction (Aristotle's {apagdgé}, but misunderstood because of corrupt text, and as
misunderstood usually translated abduction). Besides these three, Analogy (Aristotle's
{paradeigma}) combines the characters of Induction and Retroduction.” (CP 1: 65)
Indeed, Peirce later identified abduction as being at the heart of pragmatism and reflected
on his fascination with the (cognitive) process by which we are capable of isolating a
relatively small number of plausible hypotheses to account for observable facts.
While in his earlier writings, Peirce seemed to use abduction and retroduction as
synonyms, he later articulated abduction as “hypothesis formulation and selection” and
retroduction as “hypothesis testing and elimination” (Rescher, 1978: 41). Rescher
describes the taxonomy of Peirce’s overall inductive conception of science as shown in
Figure 2 and identifies it with Popper’ s (later) refutationist model of scientific inquiry.
oan Abduction-
Inductive Hypothesis
methodology formulation and
of science selection
Qualitative
induction
Retroduction-
hypothesis
testing and
elimination
Figure 2: Peirce’s Taxonomy of Inductive Methodologies (Rescher, 1978:41)
Abductive inference is most concisely described along with deductive inference and
inductive inference as one of three possible variations to the “modus ponens” argument.
Deductive Inference:
e A method of inquiry acknowledging the role of a “community of inquiry” and applying
three rules of inference- abduction, deduction and induction.
e A phenomenology consisting of three categories that provide the basis of semiotics.
e A theory of continuity which Peirce (1892) in Houser and Kloesel (1992: 312-313) called
“synechism and tychism” and which Hausman (1993) describes as “evolutionary realism”.
P>Q
P is true
«'. Qis true
This is the most familiar form of inference and is accepted as the most rigorous form of
argument. For example, if we assume the premise that: “contracting reduces costs”, and
we contract, then costs will be subsequently reduced. In practice, such an argument will
raise an immediate objection from the observer who will note that this premise is overly
simplistic and that, in particular, several enabling conditions are necessary before the
hypothesis could be deemed true. That is, P is a conditional (Bayesian) statement.
Furthermore, both P and Q are likely to conjunctions of several statements (vectors).
Modus ponens also extends to the most rigorous form for testing hypotheses using proof
by contradiction. This form is known as “modus tolens”:
P>Q
Q is false
.*. Pis false
Inductive Inference:
P is true
Q is true
“*, Po Q
In this case, we are asserting a conclusion based on a pattern of data relating to P and Q.
For example, if we observe that cost reductions appear to follow contracting, we might
conclude that contracting causes the cost reduction. In fact, the cost reduction might have
more to do with increased productivity of computers, than the advent of contracting.
Nevertheless, induction is a vital process for attempting to empirically support
hypotheses.
Abduction:
.*. Pis true
While this is the least rigorous form of inference, it is the only form that can generate
new knowledge.
The following abstracts detail how Peirce uses the three modes of inference to constitute
a “logic of inquiry”. It is this logic that forms the basis of Dewey’s experiential learning
model (Dewey, 1910) and its extant versions including, for example, Kolb (1984),
Shewhart (1939) and Deming (1950), and (Argyris, 1985)
Peirce starts by describing abduction as:
“the provisional adoption of a hypothesis, because every possible consequence of it is
capable of experimental verification, so that the persevering application of the same
method may be expected to reveal its disagreement with facts, if it does so disagree. For
example, all the operations of chemistry fail to decompose hydrogen, lithium, glucinum,
boron, carbon, nitrogen, oxygen, fluorine, sodium, ... gold, mercury, thallium, lead,
bismuth, thorium, and uranium. We provisionally suppose these bodies to be simple; for
if not, similar experimentation will detect their compound nature, if it can be detected at
all. That I term retroduction.”
(CP1: 68)
But Peirce wars:
“Retroduction does not afford security. The hypothesis must be tested.
This testing, to be logically valid, must honestly start, not as Retroduction starts, with
scrutiny of the phenomena, but with examination of the hypothesis, and a muster of all
sorts of conditional experiential consequences which would follow from its truth. This
constitutes the Second Stage of Inquiry. For its characteristic form of reasoning our
language has, for two centuries, been happily provided with the name Deduction” . (CP 2:
470)
The purpose of Deduction, that of collecting consequents of the hypothesis, having been
sufficiently carried out, the inquiry enters upon its Third Stage, that of ascertaining how
far those consequents accord with Experience, and of judging accordingly whether the
hypothesis is sensibly correct, or requires some inessential modification, or must be
entirely rejected. Its characteristic way of reasoning is Induction. This stage has three
parts. For it must begin with Classification, which is an Inductive Non-argumentational
kind of Argument, by which general Ideas are attached to objects of Experience; or
rather by which the latter are subordinated to the former. Following this will come the
testing-argumentations, the Probations; and the whole inquiry will be wound up with the
Sentential part of the Third Stage, which, by Inductive reasonings, appraises the
different Probations singly, then their combinations, then makes self-appraisal of these
very appraisals themselves, and passes final judgment on the whole result’. (CP 6: 472)
The final sentence has been made bold to emphasise the importance of “appraisals” using
what we can now identify as practices of single and double-loop learning (Argyris and
Schon, (1974). This can be enhanced to include Flood and Romm’s (1996) “triple loop”
learning which adds consideration of “power relationships”, and to include ethical and
aesthetical considerations (for example, unintended consequences).
In summary, Figure 3 describes Peirce’s model of inquiry as conducted by a “community
of inquiry”.
Observe outcomes
(Induction)
Evaluation
Monitor
implementation Test
hypotheses
Observe
events | |
Support hypothesis
(retroduction- Develop testable
triangulation) hypotheses
Form hypothesis (Deduction)
(Abduction)
Figure 3. Peirce’s System of Inquiry
Management as A bduction:
Forrester’ s early work identified the shortcomings of management science and operations
research as it was being practiced in the 1950s. For example, Forrester (1961) described
the search for optimal solutions as “misleading” and “often results in simplifying the
problem until it is devoid of practical interest’. Management science “must accept the
world as it is, not as an idealized abstraction that fails to be meaningful. It must search for
improvement, not hold out for the optimum and perfection. It must use the information
that is available, all that is pertinent, but, like the manager, it cannot wait for
measurement of everything that one might like to know. It must be willing to deal with
“intangibles” where these are important. It must speak in the language of the practicing
manager’.
These sentiments are supported by the decline in rational approaches to problem solving
such as those proposed by Kepner and Tregoe (1965). Despite an apparent rationality,
these approaches have lost out to the “alternate approaches actually employed by
managers on the job” Wagner (2002:45). On a broader front the feasibility and
desirability of rationality and certainty has been fundamentally questioned by Toulmin
(2001), Searle (2001) and others.
In management, it is becoming increasingly acknowledged that people make decisions on
the basis of their “best” hypothesis. Of course, what is meant by “best” is subjective.
From studies of decision making under extreme pressure as occurs with emergency
services, Klein (1998) concludes that:
“We have found that people draw on a large set of abilities that are sources of power.
The conventional sources of power include deductive logical thinking, analysis of
probabilities, and statistical methods. Yet the sources of power that are needed in natural
settings are usually not analytical at all- the power of intuition, mental simulation,
metaphor, and storytelling. The power of intuition enables us to size up a situation
quickly. The power of mental simulation lets us imagine how a course of action might be
carried out. The power of metaphor lets us draw on our experience by suggesting
parallels between the current situation and something else we have come across. The
power of story-telling helps us consolidate our experiences to make them available in the
future, either to ourselves or to others. These areas have not been well studied by
decision researchers’ *.
This supports the contention that experience with the use of “micro worlds” (Senge,
2006) may prove to be effective management training.
Klein's conclusions also support the importance of better understanding how hypotheses
are formed leading to action- the abductive process. Already, there is a growing
recognition of the role of abduction in decision making:
e Abduction forms the basis of artificial intelligence methodology (Josephsen and
Josephsen, 1996)
e Abduction has been proposed as the philosophical basis to strategic thinking
(Powell Thomas, 2001, Powell, 2002, Powell, 2003, Powell, 2006)
e Abduction has been associated with clinical judgment and decision making in
medicine (Montgomery, 2006).
In AI work in areas like medicine, hypotheses need to be formed based on the best
available evidence and within a prescribed time frame. A ppropriate action is then taken
on the basis of this hypothesis and outcomes observed. In medicine this corresponds to
the adoption of an appropriate treatment regime and seeing whether or not the patient
recovers. (Josephsen and Josephsen, 1996). In this context Josephsen and Josephsen
define abduction as “..inference to the best explanation..a form of inference that goes
from the data describing something to a hypothesis that best explains or accounts for the
data. Thus abduction is a kind of theory-forming or interpretative inference” and “the
basis to diagnostic reasoning”.
Josephsen and Josephsen quote Chamiak and McDermott (1985) as “characterizing
abduction as variously modus ponens turned backward, inferring the cause of something,
generation of explanations for what we see’ around us, and inference to the best
explanation. They write that medical diagnosis, story understanding, vision, and
understanding natural language are all abductive processes”. Josephsen and Josephsen
take abduction to be “a distinctive type of inference that follows this pattern pretty nearly:
D is acollection of data (facts, observations, givens),
H explains D (would, if true, explain D),
No other hypothesis can explain D as well as H does.
Therefore, H is probably true.
3 The authors are indebted to Dr Geoff McDonnell for introducing them to the work of Klein and to
the later reference to Montgomery.
The core idea is that a body of data provides evidence for a hypothesis that satisfactorily
explains or accounts for that data (or at least it provides evidence if the hypothesis is
better than explanatory alternatives)”.
These themes are further articulated in clinical practice by Montgomery (2006).
Powell (2001), Powell (2002), Powell (2003), Powell et al (2006) examine the logical and
philosophical foundations of the hypothesis that competitive advantage leads to superior
performance. Powell finds that even this widely accepted pillar of strategic thinking has
many interpretations and ambiguities. He concludes, however, that “contemporary
theories of competitive advantage may find justification in the epistemologies of
abductive inference and a pragmatic, instrumentalist theory of truth”.
On a lighter side, abduction has also been recognized as the logic of detective work as
practiced by Sherlock Holmes (Copi, 1953).
Ata more serious level, abduction, if applied inappropriately, can lead to gross error as
described by Argyris’ “Ladder of Inference” (Ross, 1994). In this case, a (false)
assumption is continuously reinforced by what you observe to the extent that you block
out other possible explanations. As a consequence you take actions which you believe are
soundly based, but are in fact wrong. (Such reasoning can also be used to explain the
careless adoption of management “fads” and their subsequent failure).
By demonstrating how different policy decisions can result from using dynamic,
compared to static decision-making frameworks, Andersen (1980) emphasizes the
importance of declaring the world view that frames the abductive process. A gain refer to
Figure 1 (above).
To minimize the likelihood of errors arising from narrow perspectives and incorrect
interpretations of data, it is important to attempt to validate the hypothesis using as many
approaches as possible (triangulation). These may typically include interviews, case
studies, cognitive mapping, and, of course simulation modelling. Simon and Sohal (1996)
refer to this process as being “generative” research.
While management might aspire to base action on testable hypotheses of the type
associated with deductive inference, the reality is that simple inferences of the type P >
Q do not adequately reflect the complexity of human and social systems and of the
fallible behaviour of individuals.
In fact, management is about taking action based on a “best hypothesis”, at a point in
time, which may reflect great urgency. The manager, having taken action, then intervenes
in the resulting outcomes to make any corrections necessary to achieve the desired goals.
Indeed, these goals may be unclear at the outset and only gain clarity through on-going
experience.
Consequently, it is observed that management relates most strongly to abductive
inference, with deduction and induction providing secondary roles- deduction in
transforming hypotheses into their logical consequences, and induction as a means of
empirical support.
The Validity of SD Models- The Forrester- A nsoff/ Slevin Debate.
Richardson (2006) provides an excellent summary of the meaning of “validation” and
validation processes. Significantly, Richardson titles his presentation: “Model Validation
as an Integrated Social Process” (our emphasis) and cites the definition established by
Forrester (1973), and Forrester and Senge (1980):
“Validation is a process of establishing confidence in soundness and usefulness of a
model” .
It is contended that Richardson’ s account is in agreement with the application of
retroductive inference. However, it is argued that critics of such frameworks are in fact
arguing from a position of deductive logic.
Consequently, the debate is at cross purposes. This can be demonstrated by reference to
the classic debate in 1968 between Forrester, and Ansoff and Slevin. While Ansoff and
Slevin (1968) argue from the perspective of deductive logic, Forrester (1968), although
presumably not aware of the abductive framework, argues from an abductive logic point
of view. This observation is further strengthened from later contributions, particularly
Forrester and Senge (1980).
Following Forrester’ s publication of the article “Industrial Dynamics- A major
breakthrough for decision makers” (Forrester, 1958), (and the subsequent publication of
the book Industrial Dynamics (Forrester, 1961), Ansoff and Slevin (1968) (A&S)
published “An Appreciation of Industrial Dynamics”. A fter outlining the method of
Industrial Dynamics, A&S conclude that “(T)o this point the approach would raise few
objections from a majority of practicing management scientists interested in simulation.
They would cheerfully admit to being “industrial dynamicists””. But from that point on,
A&S become less supportive noting the following areas of discomfort:
e The use of descriptive data within the context of a completely quantitative model
e The use of the model as a “tool for enterprise engineering” and not as an
instrument for forecasting
e Anapparent paradox to a models implementation in whereby “(W)hile insisting
on reduction of model content to fully quantitative terms, he argues that model
validation should not meet this requirement”.
e The possibility that any two modelers coming to different conclusions in answer
to the same strategic problem.
e Problems with “quality assurance” in the construction of models.
e Establishing “dynamic validity” with historical time series, but with no objective
measure of what constitutes “good fit”.
10
e Anassumed ability for the model to cover all “facets’ of reality and to quantify all
related variables and a reliance on the “properties” of the Dynamo compiler.
e A perception that Forrester failed to “ formalize the processes of abstraction of
data from managers and to provide tests of validity of the information obtained”.
e The possibility that the “information feedback viewpoint’ may be more
appropriate for some areas of business (such as production and distribution) and
less appropriate to areas like marketing. Consequently, there is a possibility that
the problem is adjusted to fit the modeling approach and not the reverse.
e How can an Industrial Dynamics model be judged as being more beneficial than
any other quantitative method?
Finally, A&S pose the question of whether or not Industrial Dynamics constitutes a
feedback “theory” of the firm.
Ina later issue of Management Science, Forrester (1968) addressed each of these points
under the headings:
What is Industrial Dynamics?
Areas of Usefulness
Structure
Feedback Loops
Quantification in Models
Sources of Information.
Validity of Models
Time and Cost.
At this point only Forrester’s discussion of validity will be considered, although his
discussion of the importance of the theory of structure is of particular significance to the
more complete learning structure discussed later in this paper.
Forrester argues that controversy over validity “seems to arise from confusion about the
nature of proof and about the avenues available for establishing confidence in a model”.
He stresses two points: firstly, the importance of linking validity to “purpose”, and
secondly, “to realize the impossibility of proof... . There is no absolute proof but only a
degree of hope and confidence that a particular measure is pertinent to linking together
the model, the real system, and the purpose” (Forrester, 1968: 614). This statement
supports his earlier argument (Forrester, 1961: 123) that “Any “objective” model
validation procedure rests eventually at some lower level on a judgment or faith that
either the procedure or its goals are accessible without objective truth’”.
In the terms of logical inference, it becomes increasingly clear that Forrester is presenting
an abductive argument, that is, forming a hypothesis that constitutes a best “theory”, and
acting on it, while A&S are talking from a perspective defined purely within the realm of
‘This quotation was brought to the authors’ attention in a question from Tim Quinn to the SD
Society's list serve on March 1, 2006
11
deductive inference. That is, A&S were basing their theory validation process on the
logic of modus tolens. Testing validity on the basis of making correct forecasts is a logic
appropriate to closed systems in which agents are not purposeful. But management is
about working in purposeful open systems (Ackoff and Emery, 1972). In such systems
agents endogenise the information provided by forecasts and adjust their behaviors
accordingly, either to meet the forecast (for example, meeting sales “forecasts”, or to
ensure that the predictions are not met (for example, if you continue to not observe the
traffic when crossing the road, I might forecast that you will get run over! So what do you
do?).
It is now a matter of history that each of the points raised by A&S has been addressed
many times within the SD literature (recent examples include Barlas, 1996; and Homer
1996, 1997). Unfortunately, much of this literature continues to debate the issues within
a frame set by deductive logic. Consequently, despite some excellent arguments, they
never seem to quite escape the inevitable consequences that deductive logic sets for
validation. Reframing the debate using abductive logic changes this.
An Abductive View of SD Method.
SD modelling has traditionally been expressed as a form structuralism, in which an
underlying structure is sought that explains a pattern of events, which in tum has been
brought to our attention as a single event. This description of SD method clearly aligns
with one of Peirce’s most often quoted descriptions of abductive inference: (CP 5: 181)
“The surprising fact, C, is observed.
Butif A were true, C would be a matter of course.
Hence, there is reason to suppose that A is true”.
In this instance, C is the pattern of events (the “surprising fact”) drawn to our attention
from an initial event, A is an expression of a causal hypothesis obtained by developing
an SD model representing the structure that best describes a pattern of events (A > C)
and A is the basis for possible future action.
The SD model constitutes “our best hypothesis” upon which we take action. In this sense,
various inputs to the modeling process plus simulation experiments constitute the
triangulation process for building confidence in the hypothesis. None of these processes
constitutes a validation of the model in the sense of deductive logic and modus tolens.
Consequently, the recognition that SD modeling is part of an abductive process, and that
the model represents the hypothesis consequent upon the abductive process, places a new
level of support for arguments against a refutationist stance in which it is deemed
possible to formulate a hypothesis that is capable of being refuted through empirical
testing. (See Bell and Bell, 1980).
Furthermore, as Emery and Emery (1997) explain in great detail, abduction is founded in
“ecological leaning” where “ecological learning and retroduction define the logic of
12
discovery”. These are ideas associated with open systems thinking, and not as Jackson
and Keys (1984) argue, partly on the basis of highly flawed definitions of simple and
complex systems, as a technique for “simple-unitary” (closed) systems. That is, situations
in which “the problem solver can easily establish objectives in terms of system(s) in
which it is assumed a problem resides... (and where)... it is also taken for granted that
there is little or no dispute about these”. (Flood and Jackson, 1991: 37).
Accepting the argument that SD modeling is an abductive process raises the question of
how this relates to the rest of SD methodology. Forrester (1992) provides an insight into
what constitutes an effective methodology. In his review of System Dynamics after 35
years:
“The ultimate success of a system dynamics model investigation depends on a clear
initial identification of an important purpose and objective. Presumably a system
dynamics model will organize, clarify, and unify knowledge. The model should give
people a more effective understanding about an important system that has previously
exhibited puzzling or controversial behavior. In general, influential system dynamics
projects are those that change the way people think about a system. Mere confirmation
that current beliefs and policies are correct may be satisfying but hardly necessary,
unless there are differences of opinion to be resolved. Changing and unifying viewpoints
means that the relevant mental models are being altered. But whose mental models are to
be influenced? If a model is to have impact, it must couple to the concerns of a target
audience. Successful modeling should start by identifying the target audience for the
model” .
Although Forrester does not explicitly mention “action”, presumably, it is implied that
changing mental models will present itself in changed behaviour (or intended behaviour).
Elsewhere, Forrester states that the “purpose of SD is to enable managers to take more
informed action”.
This suggests any System Dynamics methodology must cover the following bases:
Definition of problem/ purpose (related to ‘puzzling or controversial behavior)
Identification of stakeholders
Development of model that identifies feedback behavior
Learning (single loop learning)
Changing mental models (double-loop learning)
Taking action
Expressions of SD methodology including Richardson and Pugh (1981), Wolstenholme
(1990), Lyneis (1999), and Sterman (2000) illustrate the type of processes currently used
to meet Forrester’ s goals. (See Figures 4 to 6)
13
Policy
implementation:
Understanding.
Soy 7. hy
Problem
Policy analysis definition
System
Simulation conceptualisation
\ se
formulation
Figure 4: Richardson and Pugh (1981)
Richardson and Pugh’s model is stronger in its articulation of the model building and
simulation phases with a repeated cycling back to improvements in “understanding the
system”. But little detail is shown regarding the policy analysis and policy
implementation phases except to emphasize that (successful) policy implementation
requires both sound policy analysis and a good understanding of the system.
14
PLANNING
ANALYSIS Refining
Organizationaf® Organizational Goal:
Goals
Sensitivity
Modeling testing 1
structuring th Identification o Evaluation of Selection & Expected
Paton Potential [———} Altemative Implementation of =
P| Strategies Strategies Desired Strategy
\ . \
Refining Existing
Testing the Strategy The Ree word
Problem
Actual
SENCHte \ ite
Refining the 4 _—
Problem Structure
Figure 5: An Iterative View of
Lyneis (1999) defines a four-phased approach:
iS jected VS
ctaul
CONTROL
Strategy (Lyneis, 1999)
Phase Description
Main Objective
1 Business structure analysis
Clearly define problem of interest
2 Development of a small, insight-
based model
To understand the dynamics of the business
by exploring the relationship between the
system structure and behaviour over time &
educate client
Development of a detailed, calibrated
model
The purpose of this phase is to:
Assure that the model contains all of
the structure necessary to create the
problem behaviour
Accurately price out the cost-benefit
of alternate choices
Facilitate strategy development and
implementation
Sell the results to those not on the
client's project team.
e
e
On-going strategy management
system and organizational learning
Develop an iterative view of strategy,
compared to the traditional episodic view
(that only involves analysis and planning).
15
This structure emphasizes the iterative (leaming) nature of analysis, planning, and
control, where the (reflexive) learning is driven by the gap between actual and desired
performance.
Wolstenholme (1990: 4) summarises his methodology under the headings of Qualitative
and Quantitative System Dynamics as follows:
Qualitative SD Quantitative SD
(Diagram construction & (Simulation phase)
analysis phase)
Stage 1 Stage 2
Purpose: Purpose: Purpose:
e To create and examine e To examine the e To design alternative
feedback loop quantitative behaviour of system structures and
structure of systems all system variables over control strategies based
using resource flows, time. on:
represented by level ¢ To examine the validity o Intuitive ideas
and rate variables and
information flows,
represented by
auxiliary variables.
e To providea
qualitative assessment
of the relationship
between system.
processes (including
delays), information,
organizational
boundaries and
strategy.
e To estimate system
behaviour and to
postulate strategy
design changes to
improve behaviour
and sensitivity of system
behaviour to changes in
0 Information
structure
0 Strategies
0 Delays/uncertain
ties.
0 Control theory
analogies
0 Control theory
algorithms.
In terms of non-optimising
robust policy design.
e To optimize the
behaviour of specific
system variables.
Table 1. Wolstenholme’s (1990: 4) Methodology
Again there is an emphasis on model building and the analysis of system behaviour, but
again, very little on implementation.
16
Decisions
(organizational
gee
Real world
a
Information
feadhack
Figure 6: Sterman’s Version of SD Methodology (Sterman, 2000).
Sterman’s framework shows the SD modeling activity embedded in a “real world”
system. It is arguable that his representation most faithfully captures the way in which the
modeling activity influences mental models and hence real word behaviour.
Taking these albeit abbreviated representations of SD methodology (and it is totally
unfair to separate them from more detailed descriptions!), it is reasonably easy to
correlate the model building steps with Peirce’s abductive stage of forming a hypothesis.
Similarly, those phases associated with simulation experiments can be identified with
deductive logic- outcomes resulting from the logic expressed by model are studied, and
Peirce’s inductive phase can be correlated with those steps in which policy outcomes are
studied.
On face value these expressions of SD method may seem to go far enough. But do they?
The critical point in Forrester’ s statement of desired outcomes is the need to change
“mental models” as the primary means of changing system behaviour. Senge’s (2006)
learning model attempts to address this, particularly by introducing Argyris and Schon’s
(1974) concept of single and double learning. And to this we really need to add Flood and
Romm’s (1996) “triple loop” learning to cover the power, ethical and aesthetic issues. As
17
indicated above, the importance of this stage is pre-empted in Peirce’s description of the
inductive phase:
... by Inductive reasonings, appraises the different Probations singly, then their
combinations, then makes self-appraisal of these very appraisals themselves, and passes
final judgment on the whole result” .
From Peirce’s perspective, changing “mental models” changes a person’s sense of reality
and hence, in accordance with his “pragmatic maxim”, that person’s possible actions
steps, either conscious or unconscious.
But in total, it is argued that this process constitutes the operation of a “community of
inquiry” in the sense described by Peirce and advocated in different terms by Forrester.
Sterman’ s (2000: 850) description of the process is most apt:
“Validation is intrinsically social. The goal of modeling, and of scientific endeavour
more generally, is to build shared understanding that provides insight into the world and
helps solve important problems. Modeling is therefore inevitably a process of
communication and persuasion among modelers, clients, and other affected parties. Each
party ultimately judges the quality and appropriateness of any model using his or her
own criteria” .
In other words, an SD model constitutes a synthesis created by an abductive process
performed by a “community of inquiry”.
A More Complete Description of SD Methodology?
The above discussion leads one to propose a description of the SD methodology that uses
Peirce’s system of inquiry (Figure 3) to better address Forrester’ s (1987) requirements:
Phase 1: Establishing the problem: Awareness/ scoping
Novel event is noticed and a pattern revealed
Establish importance of determining structural cause of this pattern
Identify stakeholder interests
Form a “community of inquiry” and a research team
Define strategic intent for project expressed as reference modes
Phase 2: Developing a hypothesis (abduction)
e Develop an SD model (s) and associated causal structure
e Use triangulation to build confidence in this “best hunch”
e Use simulations to identify most effective policy setting (Retroduction?)
Phase 3: Define strategies based on causal hypothesis (Deduction)
Phase 4: Implement strategies and monitor performance. Intervene to make
corrections as new data/information is revealed
18
Phase 5. Evaluation (Inductive phase)
e Use triple loop leaming to evaluate project
e Form recommendations for future inquiry
Phase 6: Iterate
These phases are further represented in Figure 5.
Monitor
Evaluate outcomes: <=> Implement Strategy:
© on *“Make it happen!”
Single, double, & triple loop learning
*Manage
stakeholders
Did we solve the Design implementation
right ponent // sty PeMenee
How well?
Problem Loosely Defined: Develop Policy Scenarios
«Identify stakeholder Y lterate 4 «Strategy formulation
interests * Strategy testing using model
* Form inquiry team
Dialogue Sensitivity tests &
strategy options
Articulate Problem : Triangulation of model Form Dynamic Hypothesis:
Actual & desired behaviors; ==> *Business model
reference modes SS *Causal map
« Specify strategic intent * Dynamic simulation model
+ Set boundaries; time horizons
Figure 5: An Enhanced SD Methodology
Conclusion.
It has been argued that interpreting the structural basis to SD modeling as an abductive
process sheds new light on SD methodological debates. Furthermore, when integrated
into Peirce’s system of inquiry, a generic learning structure can be proposed for SD
19
methodology which involves action steps taken on the basis of a “best” causal hypothesis,
and a renewed emphasis on evaluation.
References:
Ackoff, R, and Emery, F. (1972) On Purposeful Systems. London: Tavistock Publications.
Andersen, David F. (1980). How Differences in Analytic Paradigms Can Lead to Differences in Policy
Conclusions. Ch. 3 in Randers, Jorgen (Ed) Elements of System Dynamics Method. Cambridge, Mass.:
Productivity Press.
Ansoff H. Igor, and Slevin, Dennis, P. (1968) “An Appreciation of Industrial Dynamics” . Management
Science. Vol. 14, No 7.
Argyris C and Schon, D. (1974). Theory in Practice. San Francisco: Jossey-Bass.
Argyris, C, Putman, R and McLain Smith, D (1990). Action Science. San Francisco: Jossey-Bass.
Argyris, C. P., Robert; McLain Smith, Diana. (1985) Action Science, Jossey-Bass, San Francisco.
Barlas, Y (1996). “ Formal aspects of model validity and validation in system dynamics”. System
Dynamics Review, Vol 12, No 3. pp 183-210.
Barton, John. (1999). “Pragmatism, Systems Thinking and System Dynamics” . Proceedings, 17"
International Conference of the System Dynamics Society. Wellington, NZ.
Bell, J and Bell, J, 1980. System Dynamics and Scientific Method. In Randers, J. Elements of the
System Dynamics Method. Cambridge, Mass: Productivity Press.
Charniak, E and McDermott, D. (1985) Introduction to Artificial Intelligence. Reading Mass: A ddison-
Wesley
Copi, I. M. (1953) Introduction to Logic, Macmillan, New Y ork.
CP references refer to Volume and Paragraphs in Charles Hartshome & Paul Weiss. (1931). Collected
Papers of Charles Sanders Peirce. Cambridge Mass: HUP
Deming, W.E. (1950) Some Theory of Sampling. Wiley.
Dewey, John. (1910). How We Think. Lexington, Mass:Heath. (Prometheus, 1991)
Emery, F and Emery, M. (1997). “Towards a Logic of Hypotheses” . Concepts and Transformations
2:2, pp 119-144.
Flood Robert. L. and Jackson, Michael C. (1991) Creative Problem Solving. Chichester: John Wiley &
sons
Flood, R and Romm, N. (1996). Diversity Management. Triple Loop Learning. Chichester: Wiley.
Forrester, Jay W and Senge, P (1980). “Tests for Building Confidence in System Dynamics Models” in
Legasto, Forrester and Lyneis (Ed). System Dynamics. Studies in Management Sciences, Vol 14.
Amsterdam: North Holland.
Forrester, Jay W. (1968). “ Industrial Dynamics- a Response to Ansoff and Slevin”. Management
Science 14, No 9. Pp 601-618.
Forrester, Jay W. (1973) Confidence in Models of social Behavior- With emphasis on System
Dynamics Models. MIT System Dynamics Group.
Forrester, Jay W. (1975). Collected Papers of Jay W. Forrester. Cambridge, Mass.: Wright-Allen.
Forrester, Jay W. (1992) “System Dynamics and the Lessons of 35 Years” . Ch 7 in De Greene (1994) A
Systems-Based A pproach to Policy Making. Kluwer.
Forrester, Jay. (1961) Industrial Dynamics. Cambridge, Mass: MIT Press
Forrester, Jay.W. (1958) “Industrial Dynamics- A major breakthrough for decision makers”. HBR
Vol. 36 No 4. July - Aug:37 - 66.
Hausman, C. (1993). Charles S. Peirce’s Evolutionary Philosophy. Cambridge: CUP.
Holton, Gerald. (1998). The Scientific Imagination. Cambridge, Mass.: Harvard University Press.
Homer, J (1996). “ Why we Iterate: Scientific Modelling in Theory and Practice” . System Dynamics
Review 12(1): 1-19.
Homer, J (1997). “Structure, Data, and Compelling Conclusions: Notes from the Field” . System
Dynamics Review Vol 13, No 4: 293-309.
Houser, N and Kloesel, C.(Eds) (1992) The Essential Peirce, Vol 1 (1867 - 1893). Bloomington:
Indiana Univ Press.
Jackson, Michael, C. and Keys, Paul. (1984) “Towards a System of Systems Methodologies” . Journal
20
of the Operational Research. Vol 35,
Josephsen, John R. and Josephsen Susan G. (1994). Abductive Inference. New Y ork: Cambridge
University Press.
Kepner, C.H. and Tregoe, B.B. (1965) The Rational Manager. NY : McGraw Hill.
Klein, Gary (1998). Sources of Power. Cambridge, Mass.: MIT Press.
Kolb, David A. (1984) Experiential Learning. NJ: Prentice-Hall
Lyneis, J (1999). “System dynamics for business strategy: a phased approach” . System Dynamics
Review. Vol 15, No 1: 37 - 70.
Montgomery, K. (2006) How Doctors think: Clinical Judgment and the Practice of Medicine, Oxford
University Press, New Y ork.
Peirce Edition Project Eds) (1998) The Essential Peirce, Vol 2 (1893 - 1913). Bloomington: Indiana
Univ Press.
Peirce, C. (1877). “The Fixation of Belief” , in Houser, N and Kloesel, C, (1992). The Essential Peirce,
Vol 1. Indiana: Indiana Univ Press.
Peirce, C. (1878). “How to make Our Ideas Clear”, in Houser, N and Kloesel, C, (1992). The Essential
Peirce, Vol 1. Indiana: Indiana Univ Press.
Powell Thomas, C. (2001) Strategic Management Journal, 22, 875 - 888.
Powell, T. C. (2003) Strategic Management J ournal, 24, 285-291.
Powell, T. C., Lovallo, Dan, and Caringal, Carmina (2006) Academy of Management Review, 32, 175-
196.
Powell, T., C. (2002) Strategic Management J ournal, 23, 873-880.
Rescher, Nicholas. (1978) Peirce’s Philosophy of Science. Indiana: University of Notre Dame Press.
Richardson, George P. (2006). www.albany.edu/rockefeller/pad/faculty_1/richardson.htm
Richardson, George P. and Pugh, Alexander L (1981) Introduction to System Dynamics Modeling with
Dynamo. Cambridge, Mass. MIT Press.
Ross, Rick. (1994) “The Ladder of Inference” Ch 35 in Senge, Ross, Smith, Roberts, & Kleiner. The
Fifth Discipline Fieldbook. NY : Doublday.
Ryan, T (1996). “A Peircian Framework for using System Dynamics Modelling in Management
Inquiry”. In Richardson and Sterman (Ed), System Dynamics '96, Vol 2, System Dynamics Society.
Searle, John R. (2001). Rationality in Action. Cambridge, Mass.: MIT Press.
Senge, Peter M. (1990). The Fifth Discipline. NY : Doubleday.
Shewhart, W.A. (1939) Statistical method from the Viewpoint of Quality Control. Grad School of the
Deopt of Agriculture, Washington. (Dover, 1986)
Simon, Alan and Sohal, Amrik (1996) “Generative and Case Study Research in Quality
Management” . Part 1. IJQRM. Vol. 13. No.1, 32 - 42. Part 2.]JQRM. Vol. 13 No. 2, 75 - 87.
Sterman John D. (2000) Business Dynamics. Boston: Irwin McGraw-Hill.
Toulmin, Stephen (2001) Retum to Reason. Cambridge, Mass.: Harvard University Press.
Wagner, Richard, K. (2002) Smart people Doing Dumb Things. Ch 3 in Stemberg, Robert J. (Ed) Why
Smart People Can Be So Stupid. New Haven: Y ale University Press.
Wolstenholme, Eric F. (1990) System Enquiry Chichester: John Wiley and Sons..
21.
