System dynamics and the credibility syndrome
M Govindara jan
Dept. of Management Studies
College of Engineering
Anna University, Madras-600 025
India
ABSTRACT
Classical system dynamics has been presented as a paradigm
in its own right. Causal explanations and repeated
simulations are the main forte of the method. However, the
perceptions on the nature and purpose of the method are so
varied that many researchers cannot even place system
dynamics in a taxonomy of modelling methods. It is
difficult to assess the value of system dynamics and
justify the choice of such a method. Does the method lack
real life significance and suffer from a credibility
crisis? The paper looks at the credibility problem from
both a philosophical and researcher's perspective.
THE SITUATION
The author pursued his research work in a_ leading
institute of technology in India. It was decided to adopt
the system dynamics technique to solve a marketing problem
involving the diffusion process of a technology-based
product. It was, as earlier’ said, a technological
institute where people were interested in specific results
and not in a plethora of scenarios. The subjective nature
of the method with its educated guesses, the supposed
non-linearity introduced by TABHL functions and the
sluggishness of the model to respond to parametric changes
drew fire and the author faced an identity crisis with
regard to his work vis-a-vis that of others who treaded the
well-trodden path.
And now for an introspection.
THE METHOD
In most cases the existing system is the subject of
investigation. The sequence is somewhat like this
(Willard, 1980):
~ 199 -
* Defining problems dynamically, in terms of graph over
time.
* Looking for an endogenous, behavioural view of the
significant dynamics of the system.
* Dealing with real system as continuous, consisting of
multiple nonlinear feedback loops linked together in
a complicated manner.
* Representing all systems that change through time by
varying levels and rates.
* Obtaining data for the equations by the combination
of available data and educated guesses.
* Constructing computer simulation models to rationally
analyse the structure, interactions and modes of
behaviour and thereby.
* Providing a framework for policy testing.
* Synthesizing changes in the feedback structure of the
model that will improve system performance.
THE KNOWLEDGE
The kinds of knowledge (Margaritha, 1990) involved in SD
model building process are:
Mental Models - Expressed in ordinary language the
knowledge about the. interactions that
produce known behaviour.
Reference Models - The empirical knowledge about the
system.
Operational - The knowledge to simulate the dynamic
Models phenomena of the modelled system.
THE PARADIGM
The system dynamics has traditionally been presented as a
paradigm. The perceptions on the nature and purpose of the
paradigm are so varied that many researchers cannot even
place SD in a taxonomy of modelling methods. The central
idea of the paradigm is to conceive of societal phenomena
as feedback or closed system. The second fundamental idea
is the examination of the long-term dynamic behaviour of
the system (Bernard, 1980). In SD model building process
the absence of a precise and established theoretical
knowledge is conspicuous.
THE CONFLICT
The classical system dynamics method heavily relies on
expert opinion, intuition and personal acquaintance with
the real system as information base for model specification
(Mansfred, 1984). SD models need data primarily for
initial, level values and for parameter measurement.
The main thrust of SD is towards improving the mental
models of corporate managers which, in turn, could
translate into effective organisational learning. This is
possible only when the decision makers themselves build the
model. Mere passive consumption of model results does not
transfer systems thinking ability into organisation.
SD models address policy level issues. Policy analyses
could be either system-specific or generic. Due to the
presence of nonlinear elements closed-form solutions are by
passed in favour of a simulation methodology. The
simulation aims to demonstrate the characteristic behaviour
of the system rather than to predict specific events.
Simulation is neither duplication (as the genesis of that
behaviour is not repeated) nor explanation (not knowing
what is happening and why).
Dynamics do not simply happen, but that the states of the
system, depends on the system history (expressed by its
current states), any exogenous input and the policies by
which the system attempt to regulate its own behaviour.
Policies are evaluated by observing the effect of changes
in model parameters (input) and model structure (equations)
on the simulated behaviour of model variables (output). A
policy which is not designed to be consistent with the type
of shock the system could encounter and the particular
structure of the system could make matters worse.
The dynamic behaviour of a system is insensitive to
fluctuations of many system parameters and structural
variations of equations.
THE SCENARIO
A scenario does not: seek numerical precision. It provides
a more qualitative and contextual description of how the
present will evolve into the future and is dependent on the
structure (and its delays) of the system. The policy
simulation modelling emphasises a continuing interaction
with the model. Comparisons are made across possible
futures, each reflected in a separate run of the model. No
attempt is made to produce sophisticated measurement of the
variables of interest. Multiple scenarios are all
plausible futures for the system but none would be
assured,
THE ‘ism's
In the perspective of Hilary Putnam's ‘internal realism',
realistic compromises have to be adopted only from the
inside of conceptual schemas (Margaritha, 1990). Mental
models can discriminate the realistic context of SD models.
The conceputal schemas can be diverse and have objective
truth criteria inside them, but they do not lead to radical
relativism. Relativism rests on continuity and change.
There is always continuity in some form or other after a
change.
The Systems Science covers the whole spectrum from
refutationistic to holistic approaches (Raimo, 1990).
Refutationistic approach insist on causal explanations. SD
models have greater potential for refutability than other
models,
All these '‘isms' rooted in philosophy have, instead of
clarifying the issues, have really made the SD philosophy
more abstract. This tends to lend an air of respectability
to SD models, just to differentiate them from the
disparaging remark ‘yet another simulation’.
THE TOOL
For the analysis of structural characteristics of SD
models, a variety of tools have been used - causality
loops, flow diagrams and structural equations. However ,
these methods are lacking in objectivity and precision.
Presently, problem solving seems to be the principal use of
SD - either for real or hypothetical situations.
THE LOOP
The traditional definitions of positive and negative links
in causal loop diagrams hide the fact that definitions and
characterization in terms of dynamic behaviour are not
possible (Richardson, 1986). The deceptive simplicity of
the causal loop diagrams lead to misunderstandings as they
make no distinction between information links and
rate-to-level links. Polarities cannot be defined in terms
of dynamic behaviour alone.
THE STRUCTURE
The SD method offers no guidance about how to move from a
group of case-specific models to generic structures (Mark,
1985). There are no procedures or methods for synthesizing
a mass of case-specific analyses into something more
general. The transferability of structures across fields
may not be appropriate for solving specific problems.
THE RATIONALE
System dynamics paradigms usually are so broad in scope, in
time and space, and so complex in logic that it is
difficult to analyse the entire framework to know the
working of the system. An important task in the analysis
of a behavioural simulation model is to explain clearly how
the models' assumptions lead to its simulated behaviour.
In an article, Morecroft (1983) has proposed two methods of
analysis - Premise Description and Partial Model Testing
respectively, to examine the bounded rationality of
policies in the light of cognitive limitations and to
expose the intended rationality of a policy mix whether
they produce sensible actions with respect to their
premises. They may serve as diagnostic methods for
simulation modelling.
THE MODEL
A generic system dynamics model has been developed to
understand the diffusion process of a new product-based
technology - Personal Computer in the Indian Context
(Govindarajan, 1990a, 1990b, 1991, 1992). The model
endogenously generates a variety of decisions and outcomes
that can directly or indirectly affect the diffusion
process. Several subsectors like buyer pool, company
sales, marketing effort, product development, user
perception, post-adoption behaviour, capacity, competition,
business, customer and new product have also been defined.
For instance the buyer pool is represented as
PBPOO1.K=PBPOOL.J+(DT) (D1-D2)
where D1=NRADVT+POSINT
D2=FORGET+NEGINT+AADOPT
and POSINT.KL=(ALPHA1) (PBPOOL.K+PSBUYA.K)
POSINT=POSITIVE INTERACTION (BUYERS/YEAR)
ALPHA1=POSITIVE INTERACTION COEFFT (DIMENSIONLESS)
Similarly ALPHA2=NEGATIVE INTERACTION COEFFT (DIMENSIONLESS)
THETA=FORGETTING COEFFT (DIMENSIONLESS )
Needless to say, that the values for the above coefficients
are purely subjective and the simulation is ‘trial and
error’, The question is how to really assess the values
for dimensionless quantities?
Another instance: The Sales performance is defined as
SALPER. K=EXP (S*LOGN(FMKTLNC.K) )
- 203 -
where S is the Sales Performance Coefficient.
FMKTLNC=Fraction of Market lost to New Competiton.
The assumption is that the sales performance follows a log
function. Another way of calculating is to take the ratio
of current to long term averaged sales and the effective
sales performance is a clipped value of the above two.
Extremely difficult to justify. The model, however, works.
Similarly it is difficult to give values (with confidence)
to 'indicated' variables. So are the behavioural variables
like acceptance, user perception and functional capability.
The model should adopt to variability in the environment.
Hence constant coefficients yield wrong signs for the
parameters. Hence the concept of feedback and this results
in 'indefensible' values of constants. The only way out is
that the model can be tested in various ways by adducing
evidence in favour of initial assumptions or by testing
conclusions drawn from the model.
CONCLUSIONS
A model is a set of assumptions about the factors which are
relevant to a given situation and the relationships which
exist between them. The SD models make sweeping
assumptions leading to ‘leap of logic' between equations
and consequences. Hence, the credibility syndrome.
REFERENCES
Benard.S, 1980, 'Paradigmatic Barriers to System Dynamics’,
IEEE, 682-688.
George Richardson.P, 1986, ‘Problems with Causal Loop
Diagrams', System Dynamics Review, (2), 158-170.
Govindarajan.M and Ramaswamy.N, 1990a, ‘Discontinuous
Innovation Diffusion Analysis', International System
Dynamics Conference, Boston, 458-471.
Govindarajan.M and Ramaswamy.N, 1990b, ‘Diffusion Analysis
using System Dynamics', IV National Conference on System
Dynamics, Tirupathi, 87-98
Govindarajan.M and Ramaswamy.N, 1991, ‘Diffusion Analysis
in Marketing', International System Dynamics Conference,
Bangkok, 219-227.
Govindarajan.M and Ramaswamy.N, 1992, ‘Behavioural Dynamics
of Marketing Technology-based Products', International
System Dynamics Conference (forthcoming), Netherlands.
John DW Morecroft, 1983, ‘Rationality in the Analysis of
Behavioural Simulation Models', Management Science, 31,
(7), 900-915.
Mansfred Sommer, 1984, 'The Econometric Challenge to System
Dynamics. and vice-versa: Some Future Perspectives', TFSC,
25, 263-280.
Mark Parich, 1985, ‘Generic Structures', System Dynamics
Review (1), 126-132.
Margaritha Vazanez et.al, 1990, ‘Some Conceptual Problems
in the System Dynamics Model Building Process',
International System Dynamics Conference, Boston,
1175-1186.
Raimo Keloharju, 1990, ‘Semihard Systems Thinking: A
System of an Emerging Metaphor', IV National Conference
on System Dynamics, Tirupathi, 21-30.
Willard Fey, 1980, 'The Philosophy of System Dynamics
Method', IEEE, 689-694,
a
