896THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
A FUNDAMENTAL PHILOSOPHY OF SCIENCE QUESTION
AND VALIDATION GF SYSTEM DYNAMICS MODELS
Yaman Barlas, Ph.D.
Systems Analysis Department
Miami University
Oxford, OHIO 45856
Stanley Carpenter, Ph.O.
School of Social Sciences
Georgia Institute of Technology
Atlanta, Georgia 30332
ABSTRACT
System Dynamics models, being causal simulation models, are in this
sense very much like scientific theories. Hence, there is a relationship
between validation of such models and verification of scientific
theories. In evaluating System Dynamics models, we naturally apply our
implicit “norms of scientific inquiry". Most criticisms of such models
hold that System Dynamics does not employ formal “objective”,
quantitative model validation procedures. We show through a historical
review of Philosophy of Science, that this type of criticism presupposes
the traditional losical-empiricist philosophy of science. This
Philosophy assumes that Knowledge is entirely "objective representation*
of reality, and that theory justification can be an entirely objective,
formal, “atomistic" process. According to the more recent “relativist*
philosophy of science, on the other hand, Knowledge is aunt “entirely
objective Truth", but it is relative to a given culture, epoch, and
scientific worldview. Theories can not be verified (falsified? by
entirely formal, reductionist, “confrontational” methods. Completely
objective Ctheory-free> observation is impossible. The act of observing
itself requires an assumed theory. Theory justification is therefore a
semi-formal’, holistic, social, "conversational" process.
ide discover that these twe opposing philosophies of science correspond
to two oprosing philosophies of model validation. Most critics of System
Dynamics seem te assume the traditional empiricist philosophy of
science, whereas System Dynamicists mostly agree with the recent
relativist phiplosophy on the question of model validity. We show that
these philosophics! results do have practical implications for both the
System Dynamicists and their critics. Finally, having shown that the
relativist philosophy is consistent with System Dynamics practice, we
emphasize that such # philosophy of madel validity should not lead to a
total rejection of formal quantitative teocls of model validation. Gn the
cortrery, we arsue that such tools, appropriately chosen, are most
useful when irterpreted with the relativist philosophical perspective.
I- INTROQUCTION
Roth in natural sciences and in social sciences, the question of how
models should te validated has been a most controversial issue for many
veers. Especially in social sciences, this controversy has become more
and more crucial as new and complex modeling tools have emerged in
recent years. System Dynamics (SD) methadology constitutes one such
toal, and net surprisingly, SD model validation practices have been
subject to close scrutiny.
In the last 20 vears, there have been numerous reviews Cpositive and
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 897
nesative> of SD models and we have witnessed a heated debate on
validation of such models. (For example see Ansoff and Slevin (1968),
Forrester (1968), Nordhaus (1973), Forrester et al.¢ 1974), Forrester
©1980) and Zellner ¢1982)). Throughout this long debate, critiques of
SD methodology have had one common general theme! SD does not employ
formal, “objective", rigorous quantitative model validation procedures
<which are supposed to be fundamental to scientific inquiry>. The
imelication of this type of criticism is that SD models are not “quite
scientific enough". System Dynamicists have responded to this, by
stating that model validity is strongly tied to the nature of the
problem, the purpose of the model, the background of the user, the
backsround of the analyst etc. Accordingly, model validation is
inherently a social, judgemental, highly "qualitative" process! Models
can not be "proven" to be valid but they can be “judged" to be so.
We see that there are some fundamental differences in the worldviews of
the two sides in the SD validity debat The issue is complicated by
the fact that certain concepts such as "model", "reality", “truth",
"validity" that are central to the debate are understood and used
differently by authors of different worldviews. Uniess we are explicit
and clear about what we mean by these terms, the question “is SD
methodology scientific 7" is not meaningful. Furthermore, it is
impossible to answer this Question without first stating what exactly
makes an inquiry "scientific" Cor "unscientific"). In this article we
will try to clarify the fundamental differences in the two opposing
worldviews involved in the validity debate. We will show that the
validity debate is strongly tied to a fundamental Philosophy of Science
problem. After reviewing this philosophy problem in its historical
development, we will derive its implications for SD model validation
TI- MODELS AND MODEL VALIDITY
In order to see the connection between Philosophy of Science and SD
model validation, we must first define what we mean by “models” and by
"SD models". Then, we will see that validation of SD-type of models, by
their very nature, involves some fundamental Philosophy of Science
questions.
"Models" are used in most disciplines! natural sciences, engin
architecture, computer science, social sciences, philosophy...
impossible te give a single and specific definition of "model", because
its usage greatly varies across diverse disciplines. Quite broadly
though, a model might be defined as “a substitute for some aspects of
reality". Thus, whether we have a scale model of a submarine, a
collection of balls describing the movement of gas molecules, a set of
mathematical equations to predict demand for a product, or even an
entirely verbal description of the major factors involved in drug
addiction, all these models are "substitutes for some aspects of
reality". The medels mentioned above are different from one another in
many different respects! Physical (eg. model ofsubmarine) vs. conceptual
Ceg. mathematical equations); dynamic (collection of balls) vs. static
(model of submarine)? quantitative «mathematical equations) vs.
qualitative (verbal model) etc. For our purpose, the category of
"conceptual models" is important because SD models belong to this
category. "Concertual models" are comprised of thoughts, expressions,
symbols and diagrams, rather than "physical objects". A mathematical
model is one type of conceptual model where the model is constructed by
898THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
means of mathematical symbols and expressions. SD models dre examples of
mathematical models.
In this article, we must further distinguish beatuwean tuo fundamentally
different types of mathematicz1 models! 1- Causal (theory-l ike)
mathematical models, @- Non-causal (statistical-correlational)
mathematical models. Causal models base their mathematical expressions
on postulated causal relationships within the modeled system. They are
collections of mathematical statements describing how the modeled system
works -in some respects- in real Life. Thus, by making causal claims
about how certain aspects of a real system function, they become
theories about that system. Therefore, such models can be used for both
prediction and explanation. Non-causal mathematical models on the other
ohand, simply express observed’ associations Cin form of statistical
correlations)? among various elements of a real system. Such models ara
purely-empirical (correlational), their mathematical relationships not
be ing gee on. any..theorized causal mechanism. These models are used
purposes, on the assumption that "they just. mork*
uithin-a certain range of values. of variables. They can not be
considered theories. since they make no causal claims.
SD models ‘belong to the. class..of causal mathematical models. A SD model
consists ofa set of ‘mathematital equations that attempts to descr ibe
the causal relationships existing in.the real system. Hence, a SD model
is at the same time a theo bout ‘how a system actually works in
certain respects... This ucial-propérty of: SD models with respect
to the problem of validations: — ps aividual.model equations claim to
be causal statements about ‘system el. nships, every. individual
statement Cequation) must. be de: ahd justified as an indispensable
Part of model validation. That is probably why SD. models are usually
very closely scrutinized. If a critic can show that one -of the model
equations does not make sense (does not. agree #ith an obvious
causality), then the model is refuted even if. the. agoresate model output
agrees well with observed data. The same is not: true for purely
correlational models. In such models, since no. claim of causality is
made, every equation is not subject to criticism and justification. What
matters is only the final output of the model. lf .the.model output
matches the observed data with a certain degree of accuracy, the model
is validated. For SD models, in addition to individual statement.
justification, the overall output behavior of the model mst.also. be’
evaluated against available output. data. Hence, there are two conditions
for SD model validity, both necessary but neither of them by itsel#
sufficient.
We now turn to the crucial property of SD models that makes them
different from some other quantitative models of social systems! This is
the principle of Causal explanation. A SD model consists of “causal
mathematical statements” that must be justified individually for the
model to be valid. In this respect, SD models are very much like
scientific theories. Thus, whether we are System Dynamicist or critic,
we tend to apply our accepted norms of scientific theory testing to
SO model validation. This is where one faces fundamental Philosophy of
Science questions! “What constitutes justification of a proposition 7?"
“Is it possible to completely confirm the truth of a statement ?" “How
are theories verified in mature (natural) sciences ?" Answering these
questions will provide a reference point in discussing the validation of
SD models. More specifically, it will set an upper bound on the
formalism to be expected from SD validation procedures. It will be an
upper bound because SO models have certain properties Cuncertainties
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 899
inherent in human systems, complexity and dimensions of typical SD
models, impossibility of controlled experimentation, unavailability of
data, too much noise buried in observed data...) which make them more
difficult to validate than theories of natural sciences. (We will not
discuss these properties in this article. Interested reader may r
Barlas ¢1985) Chapters II and IV and Forrester (1961) Chapter 13>.” In
the next section we take a fundamental Philosophy of Science problem,
termed “justification of Knouledge-claims" or “yerification of
propositions".
One technical point needs clarification before we start the following
discussion # As a philosophical term, “validation” refers to @ purely
logical problem, dealing with the internal consistency of a set of
propositions with respect to a set of logic rules. The philosophical
problem of “verification” on the other hand, deals with “justification
of Knowledge claims" and corresponds to "validation" as used in modeling
Literature. “Verification"*in modeling literature deals with the
internal consistency of a computer program. One must be careful in
interpreting these two terms, as they "switch" meanings from one
literature to the other. We will adopt the usage of “validation" common
in modeling literature. Readers with philosophical background should
read this to mean "verification".
III- A FUNDAMENTAL PHILOSOPHY OF SCIENCE QUESTION
Once SD models are considered as theories, their validation bears direct
relation to the fundamental Philosophy of Science question! “Under what
conditions should a scientific theory be regarded as having been
confirmed?" Philosophy of Science has emerged as a distinct
philosophical discipline in the late nineteenth and early twentieth
century, but it is strongly related to a much older philosophical
subject! Epistemology (Theory of Knowledge). The purpose of epistemology
is to find out the "conditions that make Knowledse possible". Since
scientific theories consist of knowledge-claims, it is very natural that
Philosophy of Science encompasses epistemology. In the following
section, we give a brief historical overview of epistemology before we
go on to discuss the fundamental Philosophy of Science question.
III.1, Epistemology
The idea of developing a coherent “theory of Knowledge" can be traced
back to Rene Descartes (1596-1650). Descartes believed that philosophy
needed a neu method, the deductive reasoning of mathematics, because the
only truths that can be accepted without any doubt were the ones
revealed by this method. He claimed that such a purely deductive
reasoning was possible because the ideas of such reasoning were innate,
prior to all experiencet He was a pure rationalist. In his famous
Meditations on First Philosophy (1641), Descartes uses his “method of
doubt" and deductive reasoning in order to find out what we can believe
with certainty and what we must doubt. He concludes that the "Mind*
("Thinking Self") exists with certainty €"I think therefore I am"), and
that the existence of the "things out there" ¢ “corporeal objects") must
be doubted. But Descartes does not claim that the corporeal objects are
aon-existent. He reasons that external objects “must exist", yet we
rould never be sure of their existence since our Knowledge about them is
incertain. For him, the only true Knowledge is the Kind revealed by
Jeductive reasoning, from self-evident propositions.
The other important source of modern theories of Knowledse is John
900 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
Locke's empiricism. Locke (1632-1784) can be considered the founder of
the empirical theory of Knowledge. In @p Essay Concerning Human
Understandine (1748), Locke hopes to discover where the ideas and
Knowledge come from, what we are capable of Knowing and how certain
Knon@edge can be. Locke disagrees sharply with Descartes by believing
that none of our ideas are “innate*. According to him, our mind is a
"blank tablet* C"tabula rasa") uhen we are born. All Knowledge is the
result of experience. Locke believes that external objects do exist, but
agrees with Descartes that our Knowledge about them is uncertain. But
Locke's doubt comes from his extreme emeiriciam! When we see an object,
we must be satisfied of its existence as long as we look at it. But the
moment we stop looking at the object, we have no_Knowledge as to whether
it still exists. According to Locke, “ideas” are caused directly by the
physical world, and Knowledge is a result of the mind's “acquaintanc:
with the ideas. Although Knowledge acquisition also involves the mind's
manipulation of the ideas (termed “description"), “acquaintance” is
prior to "description": The ideas are first put in the passive mind, and
then the mind starts manipulating them. This model of Knowledse
acquisition will be, as we shall see, very influential in the mainstream
Philosophy of Science.
In the Eighteenth Century, Immanuel Kant (1724-1804) defined the
epistemological problem as a search for the “principles of thinking"
©1833). Kant had been influenced by the two most important philosophical
schools of his time: Descartes’ rationalism and Locke's (and David
Hume's) Empiricism. From Descartes, he took the concept of the “active
mind", and from Locke the role of sensations <experience? in Knowledse
acquisition, According to Kant, ideas are caused by experience, but
having ideas does not mean Having Knowledges the latter is not by mere
"acquaintance", but it is by "description". The mind does not just
receive the Knowledge, but it actively produces it. The ideas are
organized according to some "a priori forms of intuitions" and processed
according to the “principles of thinking". Thus, the “essence" of
Knowledge is not to be found in a special Kind of relationship between
the external objects and the mind, but in the necessary "non-empirical
rules of understanding". This is the fundamental difference between Kant
and Locke. In Kant, the mind is not a "blank tablet", It has certain
“ideas of reason" which are “a priori", not warranted by experience.
Such a priori ideas regulate the operations of understanding. According
to Kant, there are three types of statements: 1- "Analytic a priori",
which are warranted by definitions and rules of logic, 2- “Synthetic a
posteriori!, which are warranted by experience, and 3- "Synthetic a
priori", which are warranted by an internal organizing principle of the
mind. ® crucial characteristic of Kant's philosophy is its acceptance of
“synthetic a priori” statements. According to Kant, the general
principles of all sciences (such as "every effect has a cause") and
smathematical judgements" ("straight line between tuo points is the
shortest") are ¢ thetic a priori. Kant believed that such statements
-synthetic, yet prior to experience- were not only legitimate, but also
essentia! for Knowledge to be possible.
Let us now observe an assumption common to the theories of Knowledges
Knowledge is seen as entirely objective, asocial, acultural, ahistorical
"Truth® Crather than "socially justified belief"). It follows that
Knouledse acquisition can be understood by "pure" philosophical
analysis, an analysis inderendent of all the sociai, cultural,
historical conditions of particular era. For instance, Kantian
philosophy attemets to "ground" all possible Knowledge in a description
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 901
of "Mind", a frame independent of all social and historical factors. In
his recent book, Philosopher Richard Rorty (1979) calls this ongoing
search for “neutral* foundations of Knowledge the "foundational ist"
philosophy. According to Rorty, this attempt to find the foundations of
"Truth" in something permanent, neutral (entirely objective) goes as far
back. as the ancient Greek philosophy. In Descartes/Kant tradition, the
permanence is sought in the "Mind"; in the “linguistic philosophy® of
the Twentieth century, ‘language" replaces the "Mind". But one
commitment has persisted for over three hundred years! the effort to
construct a timeless, neutral framework of inquiry relevant for all
times, for all culture. All mainstream philosophies agreed on one thing!
Knowledge is a result of some “privileged relationships", and once ve
understand them, we can tell exactly which statements are “objectively
true", independent of all cultural, historical factors. Knowledge is
entirely objective representation of reality. Rorty uses the metaphor
"Mirror of Nature" to explain this "foundationalist" view! Knowledge is
the reflection of nature on an °unclouded mirror" (the "Mind", later the
“language"). Thus, Knowledge is imposed via a privileged relationship.
The philosopher's task is to see that the mirror is being used properly,
because if it is, it will automatically deliver the "Truth".
An alternative view of Knowledge, which emerged in the i195@'s is that
Knowledge is "socially justified belief". It is aot a result of
"mirroring" the nature. A Knowledge-claim is true not because of some
"privileged" way it was acquired, but because of the arguments given to
support it, Knowledge is socially, culturally and historically
dependent. Accordingly, there are no “neutral foundations" of Knowledge,
and entirely objective verification of Knouledge-claims is not possible.
Knowledge justification is a relative, social, external process, rather
than an absolute, representational, internal one. Wa shail focus on this
recent philosophical trend later in the article. But first, the
mainstream ("foundationalist*) philosophy of science movements of the
Twentieth Century.
IIT.@. Mainstream Philosophies of Science
In the Twentieth Century, epistemology took in general an anti-Kantian
character by rejecting the legitimacy of Kant's “synthetic a priori"
statements. Inspite of this anti-Kantian trend, almost all philosophers
of science have been attracted to Kant's problematique of discovering
the neutral “foundations” of Knowledge. Bertrand Russell was one of the
first and most influential of such philosophers. Russell explicitly
rejected the existence of “innate ideas" and the legitimacy of
"synthetic a priori® propositions. He believed that all ideas come from
sense experience. He revived the Lockean thesis «that Knowledge by
"acquaintance" is prior tc Knowledge by “description”. In this respect,
Russell is anti-Cartesian. Russell's philosophy is an important revision
of Kant's epistemological program. The foundations of Knowledge are no
longer to be found in the mind, but rather in those propositions that
come from "direct acquaintance® with objects. Russell argues that
statements about physical world could be translated into statements
about "sense data", data of immediate experience (Russell (1949)). This
reductionist claim that statements can be categorized according to the
degree of their empirical content has been very influential in- the
development of philosphy of science. Az we shall see, philosophers have
assumed that propositions could be separated inte empirical and
non-empirical components and the empirical components could then be
902 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
isolated and “verified" against empirical data.
Another important work that influenced the Twentieth Century is Ludwig
Wittgenstein's early book Iractatus (1922). Like Russell,the youns
Wittgenstein had strong reductionist and empiricist views (which, he
abandoned in his later years). In Tractatus, Wittgenstein attempts to
show how a meaningful language system ought to be formulated. He states
that an analytic a priori statement, that says “nothing new about the
world", is not empirically verifiable. A synthetic statement, on the
other hand, does say something new, and must correspond to empirical
“atomic facts". Therefore, any synthetic statement that is not
empirically verifiable (which Kant called “synthetic a priori") is
meaningless. ¢This category would include value judgements, ethical
arguments, most philosophical inauiries>. Tractatus argues that people
frequently talk nonsense because of the deficiency of the ordinary daily
language. An ideal language system (*logical symbolism") mould prevent
nonsense by excluding those statements that are more than logical
deductions and at the same time not empirically verifiable. This thesis
has been very influential in the philosophy of science, especially in
the development of “logical empiricism" which has been the most
widespread philosophy of science until the 1950's.
Logical empiricism Cor logical positivism) is the name given to the
philosophical movement emanating from the “Vienna Circle", a discussion
group of famous philosophers who met between the early 1920's and mid
1938's at the University of Vienna. Originally, the most important
topics involved the possibility of reducing all synthetic statements to
direct observational statements, settings up a rigorous criterion of
meaningfulness and designing an ideal meta-languase for philosophical
analysis of scientific language systems. As a general philosophical
movement, logical positivism became very influential although not all
philosphers associated with it agreed on all issues involved. Among the
most prominent logical empiricists were Rudolf Carnap, Moritz Schlick,
Otto Neurath, Car! Hempel, Richard Von Mises and Ernest Nagel. If
logical empiricism is taken in its narrow sense as it originated in the
Vienna Circle, some of the above philosophers would not be strictly
called logical empiricist. But we will use the term in a wider sense to
imply an agreement on the following points at least: 1- Rational
discourse can have only two types of statements? Analytic a priori
(definitions and purely logical deductions) having no empirical content,
and synthetic a posteriori (statement of facts) that must be empirically
verifiable. All synthetic statements that are not empirically verifiable
must be excluded from rational discourse. 2- Philosophy must reshape the
general structure of scientific statements so that they become free from
ambiguity, vagueness and inconsistencies. The ideal would be to reduce
all scientific languages into one unified canonical form ¢ “unity of
science"). 3- The context of scientific discovery can and must be
totally separated from the context of scientific iustification.
Discovery is a historical, social, psychological process and lies
completely outside the domain of philosophical analysis. (Justification
consists of the verifiability,@¢ propositions and deductive validity of
the arguments). Logical empiricism, taken in this mider sense, comprises
the great majority of the early philosophies of science. .
One of the major flaws in logical empiricism was a logical problem
involved in the principle of "verification". Karl Popper ¢1959) analyzed
this “problem of induction" and suggested his oun solution. To see the
problem, consider a theory T and its conclusion C. C is derived from T
according to the deduction:
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA903
If the theory T is true, then the conclusion C follows.
Now to verify T, according to *verificationism", one tries to observe C.
But the verifying argument "C is observed, therefore T is true" is
logically incorrect since in reality C may occur as a result of a
process different from the one hypothesized in T. Thus, statements of
general nature (scientific theories) can never be fully verified by
observat fon. Popper's solution to this problem is the principle of
“falsification® (Popper, 1959). Accordingly, the following argumant is ©
always logically validt
If. Tis true; then C follows
*Not-G* is observed, therefore T is false.
Thus, Popper. argues that the requirement of falaifiability must replace
verifiability! Scientific theories must be required to be falsifiable.
The credibility of a theory increases as more and more non-falsifying
observations are found. Thus, theory verification is replaced by a
gradual process of "corroboration"
The wide acceptance of this principle of "falsification" can be seen as
a sign of "mellowing” for the hard-line logical empiricism. Yet, like
verifiability, falsifiability too has strong logical empiricist
elements. It assumes too, that theories can be totally separated into
their analytic and synthetic components and that for every synthetic
component it is possible to find a corresponding observation.
Furthermore, falsification assumes that although theories gain
credibility gradually, they are thrown away at once, upon a falsifying
instance. But in reality, this idealized scenario does not happen
because a typical theory is always presented with a set of assumptions:
If assumptions A and T hold, then C follows.
Now if C is not observed, it is not always clear whether it is due to
wrong theory or invalid assumptions. It is always possible to Keep the
theory by stating that "the assumptions did not hold*.°@nother practical
problem is that an observation rarely ever comes as @ er: Cor “non-C",
but mostly as a complex data subject to interpretation. 11 & largely up
to the scientist to organize and. interpret the data and to decide
whether the observation actually constitutes a "falsifying instance*
For these -and other reasons that we shall see later- Popper's original
principle of falsifiability was actually a logical empiricist thesis.
nother major problem with the early logical empiricism was its
insistance on predictive ability as the enly criterion for theory
dustification. Since, according to logical positivism, the content of a
scientific theory is irrelevant to the philosophical problem of
verification, explanatory power is not a criterion for justification.
ficcording to the principle of verifiability Cor falsifiability>, the
only criterion for justification is whether ‘the observations match with
the predictions Cimplications> of. the theory. According to this view,
explanation may be quite important in other activities ‘such as
construction of new theories, but has nothing to,d0,with justification.
Stephen Toulmin £1977), reviewing the last fifty years of the philosophy
of ‘science, explains the absurdity of relying merely on predictions, by
noting that we would then consider “horserace tipsters as scientists"
and evolutionary biology as "non-scientific". Faced with this
difficulty, many empiricists had to accept the importance of
"explanation" as evidence of Knowledge. This acceptance, Toalmin
observes, "... began to undercut the formalist approach at its very
foundations..." because explanation necessitates "... a shift to quite
another conceptual level, involving a-Kind of theoretical
reinterpretation whose merits can rot be captured in a merely formal
904THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
algorithm’ (Toulmin, 1977). .
There were criticisms of traditional epistemology as early as in the
Nineteenth Century, but the bulk of consistent criticism came in the
second half of the Twentieth Century. Richard Rorty (1878) mentions two
very important works that questioned the basic assumptions of
» epistemology taken granted since Kant. One of these assumptions holds
that Knowledge acquisition consists of two separate and distinct forms
of representations! What is "siven* to us from the outside, and what is
"added" by our mind. This fundamental distinction between the
*given* and the "added* is challenged by Wilfrid Sellars in Science.
Perceetion and Reality (1963). The other crucial assumption of logical
empiricism is the claim that propositions can be separated into their
analytic (true by meaning) and synthetic (true by virtue of experience)
components and that every synthetic statement must correspond to a
unique sense experience ( “reductionism">. This assumption is challenged
by W. ¥. Quine in "Two Dogmas of Empiricism" (1953). We shall very
briefly summarize the main ideas of these two works , as they constitute
major steps towards the construction of a new philosophy of science.
Traditional epistemology assumes that tuo essentially different sorts of
ideas ("given from the outside” and “added by the mind") come together
to produce Knowledge. Wilfrid Sellars (1953) tries to show that this
given/added distinction is not an inevitable, “essential” one, but
merely a convention of the reductionist, atomistic theories of
Knowledge. According to Sellars, it is impossible to draw an absolute
line between the "given" and the "added". Knowledge acquisition is
holistic rather than atomistic. The empiricist's assumption that
‘learning of the "particulars" constitutes the basis of Knowledge is
misguided. Even awareness of particulars is a Linguistic ‘social>
affair. We can not define the "awareness* of a machine, an insect or a
new-born baby, because none of these can play our “language game". fs we
cunderstand awareness, “being aware of things" makes no sense prior to
language acquisition ( "language" defined in its most ganeral sense of
“symbol manipulation"). Thus, according to Sellars, Knowledge is
socially iustified belief.-By opposing the “given/added" distinction,
Sellars does net try to develop a new theory of how mind works. On the
contrary, he claims that such a theory could net possibly account for
why Knowledge is possible, because the latter is socially justified
belief and occurs in a social, conversational domain. Once we
acknowledge that Knowledge is social and temporal, then we do not need a
Kantian theory of how the mind works in order to find the necessary
conditions for Knowledge. Rorty (1879) observes that, Kant had made the
sgiven/added distinction, nat because he had discovered something
fundamental about how mind acquires Knowledge, but because such a
distinction was needed for Nis philosophical program of finding the
objective, neutral foundations of Knowledge. Once the given/added
distinction is abandoned, Knowledge acquisition becomes naturally
holistic and developing an atomistic theory of how and why Knowledge is
Possible becomes hopeless
In "Tuo Dogmas of Empiricism" €1953),-W. V. Quine attacks tuo important
assumptions of empiricism. First, Quine shows that the analytic/
synthetic distinction'is not absolute or essential, but it is merely
conventional. Quine asserts the impossibility of defining “anaiyticity"
except in extremely unimportant and trivial cases like "no unmarried man
is married". Quiné shows that, in its more general and frequent usage,
it is impossible to define “analyticity"” without assuming some
"synthetic" Cempirical> facts. Thus, it becomes impossible to define an
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 905
essential "philosophical" analytic/synthetic distinction. (See Quine
©1953) for his lengthy argument on the topic>. According to Quine,
certain statements are appropriately called "analytic", because there is
virtually total consensus about the meanings of the terms involved, and
given our linguistic rules, it becomes very easy to reach an agreement
on the truths of such statements. Quine is not against such a
distinction as a useful "convention". Quine's criticism is the way
philosophers have been using the distinction in order to construct a
“reductionist™ theory of verificationism. Thus, the “second dogma* that
Quine attacks is the reductionist claim that for every synthetic
statement, there must be a unique set of observations the occurences of
which would help confirm that statement, an a unique set of observations
the occurences of which would decrease the likelihood of its truth.
Quine shows the problems involved in trying to test individual
statements in isolation from the accompanying ones. According to him,
statements can only be tested as a corporate body. Quine argues that in
a scientific theory, the analytic and synthetic components can not be
entirely separated. Furthermore, he claims that science is like a *...
field of force, like a fabric which impinges on experience only along
the edges", but "... no particular experiences are really linked with
any statement in the field except indirectly through considerations of
equilibrium affecting the field as a whole" (1953). Accordingly, there
are many ways of accomodating a theory to an “abnormal experience". be
choose a particular way of doing it, not not due toe some absolute
scientific principle, but because it is convenient, causing small
disturbance in the existing theory. Thus, Guine's view of justification
is holistic and conversational-as opposed to reductionist and
confrontational.
Guine's and Sellars' criticisms of the two fundamental assumptions of
logical empiricism were imeortant steps towards the formation of an
anti-positivist philosophy of science. In the meantime, Thomas Kuhn
published his extremely influential anti-positivist work, Ihe Structure
of Scientific Revolutions (1962). Kuhn attempts a historical analysis of
how science progresses. He argues that, at any given epoch, the rules to
be followed by science are dictated by the "ruling paradigm". During the
periods of “normal science", the paradigm is accepted without any
questioning of the underlying assumptions! "In its normal state, then, a
scientific community is an immensely efficient instrument for solving
the problems or puzzles that its paradigms define” (1970, p.4®).
Eventually comes a period when the ‘ruling paradigm” can not solve
certain problems, and scientists start questioning the paradigm's
fundamental assumptions. When enough scientists become convinced that it
is impossible to solve the “anomaly* within the framework of the ruling
paradism, and only if an alternative paradigm is already available, then
a "scientific revolution" takes place. The old assumptions are abandoned
and replaced by new ones. Kuhn shows by historical examples that a
scientific revalution involves a fundamental shift in the scientific
worldview so that new problems are defined by the new paradigm. The
perspective, the methods and rules to be followed, and even the “norms
of rationality" are restated. What is rational in one epoch may be
considered irrational in another epoch. In short, it is as if the
scientist's world has totally changed. After the revolutionary paradigm
establishes itself, it becomes the ruling paradigm for next generations
to come, and the process repeats itself. Kuhn sees this process as
“scientific progress". Kuhnian progress is not directed towards an
objective and absolute "Truth", it is simply “successful creative work".
906THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
A scientific theory is accepted not because it is true in any absolute
sense, but because it proves to be useful for the advancament of science
in a particular era. The crucial anti-positivist elemant in Kuhn's
thesis is that everything a scientist does depends on the dominant
scientific worldview. Accordingly, “theory-frae observation” is simply
not possible. Everything from the initial formulation of the problem to
the interpretation of the results is shaped by the dominant worldview.
Richard Rorty illustrates this idea by stating that Newton did not
necessarily give “right answers to the questions to which Aristotle had
given wrong answers", because they were not necessarily asKing the same
questions (1973, p.266). According to Kuhn, ruling paradigms of
different. epochs are “incommensurable" because they do not even deal
with the same probiems. Kuhn's thesis that there can be no “neutral
observations" has done considerable damage to logical empiricism,
because the entire verification (falsification) theory assumes the
possibility of neutral observations.
After the 1960's, faced with the Kinds of criticisms illustrated in the
previous paragraphs, logical empiricism has had to acknowledge the
impossibility of purely formal, ahistorical, acultural analysis of
scientific inquiry. Karl Popper recognized the importance of
understanding the. “internal history" of science, though he still tried
to exclude sharply the "external" factors influencing scientific
inquiry. His view of history of science was a “rational reconstruction"
of history under the principles of “scientific rationality". His student
Imre Lakatos holds an even less positivist view of science. In his vieu,
the history and psychology of science are important in understanding how
science progresses. Lakatos also acknowledges that entirely rational
reconstruction of history is impossible, that studies of both internal
and external histories are necessary. He rejects “naive
falsificationism", having observed that “no experiment, exper imental
report ... alone can lead to falsification” (Lakatos 197@).
In the 1978's, philosophers and scientists have increasingly
acknowledged the inadequacies of logical empiricism. Today, logical
empiricism has lost its prestigious place it held in the first half of
this century. The purely formal, algorithmic, abstract "organon" of
logical empiricism has proven inadequate for the practhcal questions
facing the studies of science. Many Philosophers now hold that it is
impossible to explain the scientific change as an entirely objective
process. Stephen Toulmin describes this tendency as "From Form to
Function" €1977). Thus, the "doors of history, psychology and sociology"
have opened one by one to the philosophy of science (Toulmin 1977).
Toulmin observes that after the 196@'s, terms like “historicism",
*‘relativism" or “psychologism" were not anymore being used to discredit
those who "mixed" history, sociology or psychology in their
- philosophical works. AS a consequence of this, Toulmin notes, “These
days, we are all prepared to be ‘interdisciplinary'" (1977). The pursuit
of timeless and absolute truths has become out of fashion. "Practical
use“ has taken the place of formal rigor, “truth and “excellence".: In
short, “formal® was being replaced by “functional" ¢Toulmin 1977).
This brief historical review of epistemology and. philosophy of science
shous that there exists tuo opposing philosophies? The traditional
formalist/absolutist camp and the new functional/relativist camp. In the
following sections, we shall see the implications of both philosophical
positions for model validation controversy.
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 907
IV- IMPLICATIONS FOR SD MODEL VALIDATION
If one adopts a logical empiricist, reductionist, formalist philosophy
of model validation, then validation is seen as a strictly formal,
algorithmic, "atomistic" and “confrontational” process. Since the model
is assumed to be an objective and absolute representation of the real
system, it can be either true or false. And given that the analyst uses
‘the proper validation algorithms (and (s)he is honest), once the model
“confronts the empirical facts, its “truth* Cor falsehood) is
automatically revealed. Validity becomes a matter of “formal accuracy"
rather than practical use.
If one takes a relativist, holistic, functional philosophical approach
to the validity problem, then validation becomes a semi-formal,
conversational process. A valid model is assumed to be only one of many
possible ways of describing a real situtation. No particular
representation is superior to others in any absolute sense. No model can
be entirely objective, for every model carries in it the modeler's
worldview. Models are not either true or false, but lie on the continuum
of usefulness. Model validation is a process of building confidence in
the usefulness of the model. Such a process. is inherently gradual and at
best partly algorithmic. Validity does not reveal itself automatically
as a result of some formal tests, but it builds gradually as a result
of a social process. Validation is a matter of social conversation,
because establishing model usefulness is a conversational matter. This
is especially true when the model user is not the model builder, in
which case the user must be convinced about the usefulness of the
model.
Thus, we see that the two opposing schools of philosophy of science
imply two opposing philosophies of model validation. In the following
sections, we shall ilfustrate this observation by referring to specific
articles. Although our main topic is SD model validation, we shall also
present examples of non-SD articles addressing some fundamental issues
of model validation. (Our. intent is by no means to give an extensive
literature review. For a quite complete review of validation literature,
the reader is referred to Wright and Shahin ¢1980)).
1V.1. Relationships with Non-SD Modeling Literature
One cf the early and important non-SD articles dealing with
philosophical aspects of validation is Naylor and Finser‘s “Verification
of Computer Simulation Models" (1968). The authors discuss..some basic
philosophical positions in validation controversy ® 1- "Rationalism", 2-
"Empiricism"\and 3- Milton Friedman's. “positive economics* which asserts
that assumptions of a hypothesis should net be required to be verified,
that the only criterion of confirmation isthe model's predictive
ability. Naylor and Finger argue that in practice, these three views
need not be mutually exclusive, and try to combine the three in a
"multi-stage" verification program. Although Naylor and Finger take an
eclectic approach, their fundamental. asumetion is actually empiricist’
"s+. @ simulation model the validity of which has not been ascertained
by empirical observation, may prove to be of interest for expository or
pedagogical purposes eg. to illustrate a particular simulation
technique), such a model contributes nothing to the understanding of the
system being simulated" (1968). The article also holds the view that a
model is either true or false, rather than viewing validity as a ‘degree
of usefulness '.
fnother article, published about the same time, but closer to the
opposite philosophical view is Mitroff's “Fundamental Issues in the
908THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
Simulation of Human Behavior* (1969). Mitroff argues for
C. W. Churchmann's “experimentalism". This view holds that reality can
not be Knoun as an isolated objects it is not a “starting point", but a
"process" of going back and forth between the world and the model.
According to experimentalism, Knowledge is holistic and social, and both
model building and model validation are inevitably subjective, by being
aspects of one's theory of scientific inquiry. Mitroff (1969) notes that
those elements we choose as "essential" and include in our model are
probably also chosen as "essential" in validating the model.
Milton Friedman's "positive economics* discussed briefly in Naylor and
Finger (1968) is analyzed by Cyert and Grunberg (1963) at more length.
According to Friedman, the assumptions of a hypothesis need net be
realistic. A hypothesis is confirmed only by its predictive success.
Given that such a success is achieved, the validity of the assumptions
is irrelevant. (This sounds very much like early logical empiricism). In
Friedman's example of the “expert billiard player", the hypothesis he
considers ist “the player solves the formal mathematical problem of the
path of the balls required for success". Now, this hypothesis is based
on the assumption that the player has the mathematical Knowledge to
solve such a complex mathematical problem. It is easy to disconfirm this
assumption by testing the player for his mathematical skills. But for
Friedman, such disconfirmation is irrelevant to the verification of the
hypothesis! If the latter predicts that the player will make certain
shots on certain situations, and if the player does make the predicted
shots in all those situations, then the hypothesis is confirmed. Cyert
and Grunbers criticize this view. They point out that Friedman's first
mistake is his belief that conclusive empirical confirmation is
possible. They take the Popperian view that hypotheses can only be
disconfirmed. The second -and fundamental- problem with Friedman's
theory is that, followed literally, it would lead to the acceptance of
hypotheses without any critical appraisal or discussion. His theory
implies that “explanatory power“ has no role in hypothesis confirmation.
Cyert and Grunberg propose that we give much more emphasis to the
explanatory ability of models. They make the important observation that
acceptance of “billiard player's Knowledge of advanced math" comes from
an unwillingness to study his actual decision-making process. If we take
the alternative approach of trying to model his decision making process
and incorporate it in our hypothesis, then, the authors state, “we can
not only join our Knowledge with that of other disciplines studying
similar behavior, but we will gain explanatory value for our models as
well as predictive ability" ¢1963>.
A very gocd overview of the problem of validating “large scale models”
is provided by House and McLeod (1977). The authors appreach the problem
of validity from a very practical perspective, by considering what a
"businessman would be willing to spend" for a model! “The businessman
can not afford te disccunt a 'hoped~for' infinite return as the result
of an unknenn expenditure for a near-perfect model today. Gur business
world exists in the present, so the businessman will be satisfied to buy
a somewhat less than a perfect model for a Known cost* (1977). ‘Perfect
validity’ is an unrealizable, ideal concept which implies that a model
is an exact duplicate of the real system. Interestingly, the authors
reject the desirability of ‘perfect models' even as an ideal concept,
because understanding them would be as difficult as understanding the
real system!
This brief review of literature on validation illustrates how different
views of model valida’ n assume différent philosephies of scientific
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA909
inquiry.
IV.@. Relationships with SD Validation Literature -
The first exposition of the vieus of SD paradigm on the question of
model validity was given in chapter 13 of Industrial Dynamics by Jay
Forrester (1961). Forrester argues that validity of a model can not be
discussed without reference to a specific purpose: Model validity is a
relative concept. He makes the stronger claim that “the validity of a
model should not be separated from the validity and the feasibility of
the goals themselves". Since reaching an agreement on the feasibility of
the goals can not be achieved through an entirely formal algorithmic
process, validation becomes very much a matter of social discussion.
ficcording to Forrester, "any ‘objective’ model validation procedure
rests eventually at some lower level on a judgement or faith that either
the procedure or its goals are acceptable without any objective proof"
©1961). Forrester also criticizes the illusion that using fixed
statistical ‘significance levels’ brings objectivity to the validation
procedure. His point is that the selection of the significance level
must ultimately be tied to our goals. Another non-traditional view of
Forrester is his willingness to accept non-quantitative model
validation. He argues that a negative attitude towards ‘qualitative’
validation procedures is not justifiable, since “... a preponderant
amount of human Knowledge is in non-quantitative form" (1961). Finally,
Forrester sees explanatory power as important as predictive power in
model validation. Forrester's views on model validity correspond to the
relativist, holistic philosophy of science. We shali see in the
following sections, that this is true for System Dynamicists in general.
Seven years after its publication, one of the most well-known and
representative reviews of Industrial Dynamics was given by Ansoff and
Slevin ©1868). Ansoff and Slevin criticize -among other ideas of
Industrial Dynamics- Forrester's views on model validation. First, they
object to Forrester's claim that model validation need not be entirely
quantitative. They quote from another critic of Industrial Dynamics,
Harvey M. Wagner? "Does Industrial Dynamics represent a truly scientific
approach? Or does it represent the judgemental approach of a particular
scientist"? (1968). The authors admit that such a criticism should be
directed not only to SD, but to the ‘management science' in general.
This implies that management science is not “truly scientific", because
it is “qualitative and judgemental". This view assumes a utopic concept
ef science. Like logical empiricism, it assumes that there can be an
entirely objective, "non-judgemental" method of inquiry. Ansoff and
Slevin point out that Forrester is not as much concerned mith the
quantitative predictive validity as an econometrician is. In Industrial
Ovnamics, the authors state, emphasis is placed on "making models ‘true
to life’ the first timé, on observing carefully, on testing boundaries,
on testing the internal logic of the model, on obtaining parameters from
real-life situations" ©1968). The authors complain that neither a clear
criterion cf validity, nor the degree of “correspondance sought" is
specified by Forrester, rendering the validation process not only
aualitative but also gubiective. They add that seeking “objective
validity" does not necessarily mean seeking “absolute accuracy".
According to this view, "absolute truth” is unattainable due to the
imperfections of the inductive method, but not due to the subjective
elements inherent in all inquiry. ficcording to this “naively realist"
view, scientific method has its Limitations, yet it can be entirely
objective. Ansoff & Slevin overemphasize "quantitative", "formal"
validation. Towards the end of the article, they state the first
910THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
condition a theory must meet! “It should embrace a well-defined body of
observable variables (emphasis added)" (1968). Overall, Ansoff & Slevin
defend a philosophy of mode! validation that has strong logical
empiricist elements
In his response to Ansoff and Slevin, Forrester (1968) articulates his
relativist ideas of model validity presented in Industrial Ovnamics. He
reemphasizes the role of ‘explanation’ in model validation by stating
that a model may well replicate the observed behavior “for the wrong
reasons". Forrester also asserts that validation is ultimately an
"agreement" and not a proof. Thus, although the question of validity has
no definite ansuer “in the abstract", he states he has “never personally
encountered a situation where, in the context of a specific system, a
particular model and a clear purpose, there was a continuing
disagreement about validity" (1968). Once again, Forrester argues for a
“conversationalist", "functional" philosophy of model validation.
Another strong criticism of SD method is given by Nordhaus (1973).
Nordhaus’ paper mostly consists of specific technical criticisms of a
specific SD model, namely Forrester's World Dynamics. The technical
criticisms are naturally beyond the scope of our article (a detailed
technical response is provided by Forrester, Low and Mass 1974). But a
few general assertions made by Nordhaus on the question of model
validity are pertinent to our discussion. The author states that ‘the
treatement of empirical relations in World Dynamics can be summarized as
*, “ss. as not a single relationship is drawn
from empirical studies”. To “what extent these criticisms. are valid
depends on what the author means by “empirical studies", on the purpose
and intended use of the model, none of which specified in the article.
But beyond the technicalities, the author does hold an empiricist
philosophy of science quite incommensurable with that of System
Dynamics. Quoting from Naylor and Finger, he claims that a model not
subjected to empirical validation is "void of meaning". Such a
“criterion of meaning" is reminiscent of the extreme logical empiricism
of the 1930's.
fn important philosophically oriented SD article is Donella Meadou's
"The Unavoidable A Priori" (1988), The central idea of the article is
the Kuhnian thesis that every modeling school inevitably has biases that
influence the selection of problems, solution methods and evaluation
criteria. Meadows compares the major assumptions of tuo specific
modeling schools! System Dynamics and Econometrics. Accordingly, the
major assumption of SD is that the behavior of a complex system arises
from its causal structure, that people do things for some reason
(whether Known or not). The process of modeling consists of writing
causal equations that in some way describe the system's structure. To be
able to explain the behavior by the system's internal structure (rather
than by external influences), the modeling approach must be extremely
‘holistic’ and ‘interdisciplinary'. The approach is non-empirical in its
classical sense, not requiring strict numerical empirical validation.
Many, of the equations may be derived by “conversations with people
involved*, Meadows next takes the Econometric modeling and states that
in such models causality is not a major concern. The model equations,
mostiy dictated by data, do not make an explicit claim of causality. The
crucial criterion is that the model predicts; ‘causal explanation’ is
not sought for. The approach is empiricist, highly "atomistic" and
"non-interdisciplinary". Next, comparing the SD and Econometrics
paradigms, Meadows asks "hill one competing paradigm eventually
eliminate the other completely Cas a Kubnian position would imply)*?
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS. SOCITY. CHINA 911
Meadows, while admitting that the tuo disciplines can not be mixed,
states that the two can co-exist because they do nat compete to solve
the same type of problems (long-term perspective vs. short-term
forecasting).
‘Finally, the most complete discussion of model validation in SD is
given by Forrester (1973) in an unpublished research paper. In an
attempt to clarify the issues underlying tha model validity debat
Forrester asks how and why the concept of validity is interpreted
differently by different groups of people. He observes that most
professionals (managers, engineers, doctors) take validity as "relative
usefulness", whereas most literature on social systems modeling sees it
as a “formal logical concept rather than a pragmatic issue“. Forrester
calls the two groups the “operators* and the “observers” respectively.
How “operators” see validity is very similar to House and McLeod's
description of how businessmen see validity. An operator sees a model as
an incomplete, imperfect theory about his reality, which is valid if it
proves to be a useful tool in making decisions. Forrester stresses that
an operator “seeks shared confidence" because he is “seldom a secure and
absolute dictator. He must persuade, he must explain, he must lead"
€19873>. For an operator, model validation is very much a “public
process". To illustrate the viewpoint of an “observer", Forrester refers
to the notion of “logical validity of an inference* (described earlier
in this article as the "philosophical problem of validity of an
argument"). Forrester claims that many “observers* have such a concept
of "validity" when they seek absolute and objective model validity
tests. Now such tests will tell us whether a logical mistake is made in
deriving model implications from its assumptions, but nothing about the
relevance of the model to a real-life problem. Such tests are necessary,
but insufficient to establish model credibility. Forrester seems to
suggest that many "observers", not having to make real-life decisions,
are confused about the two aspects of model validity . According to
Forrester, such “observers” fail to see the impossibility of model
Justification by entirely formal objective tests: For them, “the
appropriateness of the assumptions is not a part of the validity issue’.
cM. Friedman's "positive economics" described earlier is an illustration
of this view). Thus, Forrester argues that models built by such
Observers become “collector's items*, having no purpose of practical
use. And he concludes that since for most observers practical use is not
important, rather than seeking “shared confidence and consensus",
observers would seek debate! "The observer aims not to create public
constituency, but instead to display individual effort, diligence and
virtuosity" ©1973).
This brief survey of literature shows that the views of System
Dynamicists on validation are in the direction of the “*relativist"
Philosophy of science. SD practitioners see the validation problem much
the same way the neu philosophy of science sees the problem of “theory
verification". Accordingly, validation ("verification") is inevitably
relative. It is a matter of social conversation, rather than objective
confrontation, It is holistic, rather than reductionist, practical,
rather than formal, Having seen the connections between model validation
and the tuo opposing philosophies of science, we can repeat the question
posed at the beginning of this article! *Is System Dynamics method truly
scientific?" The answer is obvious! “It depends on one's philosophy of
science". If one adopts the traditional formalist, empiricist
philosophy, then SD method does not sound entirely scientific. We showed
in this article that this type of philosophy underlies most of the
912THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
empiricist criticisms of SD methodology. If on the other hand,.one
adopts the recent relativist philosophy of science, then there is
nothing "unscientific" in the way SD treats the question of model
validity.
V- CONCLUSIONS
In this paper, we started by stating that since System Dynamics type of
(causal) models are very much like scientific theories, we tend to ask
that validation of such models conform with our norms of “scientific
inquiry". But then we showed that the philosophy of science does not
Present a unique view about the nature of scientific inquiry. ble
described tuo fundamental and opposing philosophies of science! The
traditional logical empiricist view which holds that scientific theories
can be verified (falsified) by entirely opdective formal methods. The
natural implication of this view is that model validation can be
carried out by entirely objective formal, “confrontational” methods?
validity means “truth". The modern, relativist philosophy of science, on
the other hand, holds that scientific theories can only be verified by
graduel, conversational, semi-formal methods. This means that model
velidetion toc can erly be carried out by semi-formal, holistic,
conversational methods! validity means relative usefulness. Then, we
showed that the System Dynamics paradigm sides with this recent
relativist philosepy of science on the issue of model validity.
Similarly, most criticisms of System Dynamics methodology are based on
the opposing logical empiricist philosophy of science. These conceptual
links between the philosophies of science and the views of model
validation have important practical implications for both the System
Dynamicists and their critics.
First, the implications for the critic's position. The critic who
accuses SD for being “unscientific” because SD validation procedures are
not "objective, formal and quantitative enough” should know that his/her
view of "scientific objectivity" and "formalism" represents only one
side of the fundamental philosophy of science debat Sthe) should take
into account the fact that there is an alternative widely held
philesaphy of science that is in agreement with how System BDynamicists
view the model yvalicits question. This being so, critics should try to
aveid criticizing SD medel validation based on such general
characterizations as "not objective", “not empirical", “not formal
enough". Such criticisms will never be persuasive for the System
Dvnamicist who haprens to hold just the opposite philosophical view on
such igsues. To be constructive, critics should take specific SD model
validation techniques and applications and explain whythey think these
are wes. tcols of medel validation. Critics must be able to say “the
follewing specific validation tools you.are using are not convincing for
the fcllowing reasons". Then, they must suggest alternative more
“abjective and formal" methods, and state why these alternative methods
would help increase the validity of the model.
Our corclusians have some practical implications for the System
Oynamicist's pesition as well, Real-life experience has taught most SD
Cor other causal) modeling practitioners that models are inherently
incomplete, relative and partly subsective and that model validity is
really usefulness with respect to a specific purpose. But at the same
time, most practitioners unaware of the recent relativist philosophical
developments, would think that their own view of model validity is not
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA913
really ‘quite scientific". These practitioners have been influenced by
the established traditional philosophy of science that requires a utopic
objectivity and formalism for an inquiry to be “scientific". Thus, many
practitioners, while experiencing that validation is bound to be a
relative, semi-formal and conversational process, at the same time see
this as a weakness of their modeling fields. We show in this article
that the recent relativist philosophy of science claims just the
opposite: Accordingly, even the scientific theories of natural sciences
are justified much the same way as models of social systems are
validated. There is no qualitative difference between the twot They are
both semi-formal, relative, holistic, social processes. SD-type modelers
owe no apology for not meeting an outmoded and utopic criterion of
scientific inquiry.
Finally, we must point out that the relativist philosophy of model
validation does not imply that pursuit of formal quantitative validation
tools be abandoned. On the contrary, such tools are most useful when
they are used with the relativist philosophical perspective.
‘Accordingly, formal tools can not be complete tests of model validity
and they can not turn the overall validation problem into a purely
objective formal, algorithmic process. But, these tools are very
effective ways of organizing, summarizing and communicating information.
Formal tests can not automatically determine the validity of a model,
but they can provide valuable information in judging and communicating
the usefulness of a model, Since the relativist philosophy emphasizes
that validation is a matter of social conversation, System Dynamicists
should be the first to appreciate the role of formal quantitative tools
in summarizing the information pertinent to model validity and
communicating it to the interested community. (See Sterman 1984 and
Barlas 1985). The challenge is to design quantitative measures that
capture information pertinent to the model's usefulness with respect to
its purpose.
REFERENCES
Ansoff, H.I. and Slevin, 0.P.¢196@) “An Appreciation of Industrial
Dynamics". Manasement Science 14: 383-397
Barlas, Yaman (1885) i i
i ra ur. 7 1 i . Georgia
Institute of Technology, Atlanta! Unpublished Ph.D. Dissertation.
Cyert, R. and Grunberg, E.(1962> "Assumption, Prediction and Explanation
in Economics" in Cyert and March, eds., & Behavioral Theory of the Firm.
N.J.tPrentice-Hall
Descartes, Rene’! (1931) "Meditations on First Philosophy" in E.S.
Haldane and G.R.T. Ross, Trans., The Philosophical borks of Rene’
Descartes. Cambridge! Cambridge University Press
Forrester, Jay We. €19681) L : . Cambridge: MIT Fress
Forrester, Jay W. (1968) "A Response to Ansoff anc Slevin". Management
Science !41 6Q1-E15
Forrester, J.W., Low, G.W. and Mase, N.J. (1974) "The Debate on World
Dynamics! A Response to Nordhaus”. Bolicy Sciences 5S: 169-190
Forrester, J.W. C1972) "Confidence in Models of Social Behavior with
Emphasis on System Dynamics Models". MIT System Dynamics Group.
Cambridge! Unpublished Research Paper
Forrester, J.W, €198@> "Information Sources for Modeling the National
Economy". Journal cf the smsrican Statistical Assaciation 75:555-574
914THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
House, Peter W. and McLeod, J. (1977) Large Scale Models for Policy
Evaluation. NewYorktJohn Wiley and Sons
Kant, Immanuel €1933) Critique of Pure Reason, N.K. Smith, trans.
London? St. Martin's Press Inc.
Kuhn, Thomas €197@) The Structure of Scientific Revolutions. Chicagot
University of Chicago Press
Lakatos, Imre (1978) "Falsification. and the Methodology of Scientific
Research Programmes" in I, Lakatos and A. Musgrave, eds., Criticisms and
. Cambridge? Cambridge University Press
the Growth of Knowledae.
Locke, John (1894) @n_Essay Concerning Human linderstandina. A.C. Fraser,
ed. Oxford? Clarendon Press
Meadows, Donella (1988) "The Unavoidable A Priori" in J. Randers, ed.,
Elements of the System Dynamics Method. Cambridge! MIT Press .
Mitro##, Ian €1969) "Fundamental Issues in the Simulation of Human
Behavior". Management Science 15: 6351649
Naylor, T.H. and Finger, J.M., "Verification of Computer Simulation
Models". Management Science 14:92-101 .
Nordhaus, W.0.€1973) "World Dynamics! Measurement Without Data‘. Ihe
Economic Journal 83:1156-1183
Popper, Karl. (1959) The Logic of Scientific Discovery. NewYork Basic
Books
Quine, W.V. ©1853) "Two Dogmas of Empiricism" in Erom a Logical Point of
View. Cambridge: Harvard University Press
Rorty, Richard €1973) Philosophy and the Mirror of Nature. N.J.!
Princeton University Press
Russell, Bertrand (1949) Qur Knowledae of the External bMorld. Londont
Allen and Unwin
Sellars, Wilfrid (1963) Science, Perception and Reality. NewYork:
Humanities Press
Sterman, John D. (1984) "Appropriate Summary Statistics for Evaluating
the Historical Fit of System Dynamics Models". Ovnamica, 10! 51-66
Toulmin, Stephen (1877) "From Form to Function! Philosophy and History
of Science in the 195@'s and Now". Daedalus 106:143-162
Wittgenstein, Ludwis (1922) Tractatus Logico-Philosophicus. C.K Ogden,
trans. London! Routledge : e
bright; R.D. and Shahin, G. (1880) "A Bibliography of Simulation
Evaluation": Proceedings of the iith Annual Pittsburah Conference:
541-s46
Pellner, Arnold ¢1S@@) "Comment on Forrester's ‘Information Sources for
Modeling the National Economy'". Journal of the Gmerican Statistical
@zsociation 75: S67-562.
