Recent queries on graduate projects

["William Steinhurst"]

<snip>
It is not like word processing -- one can not
> just buy the software and read a few pages of the manual and begin to
> compose.

Absolutely. Owning a copy of AutoCad does not make one an architect.

Bill Steinhurst
wsteinhu@psd.state.vt.us

[WWGunn@aol.com]

Why shouldnt someone with business experience who, after reading about
system dynamics for the first time, and carefully reading a software manual
for iThink or Vensim, begin to build useful models?

What is the catch?

W.W. Gunn
WWGunn@aol.com

[Nordin Othman]

George Richardson wrote:
> I am becoming increasingly alarmed at messages to this list
>from people doing graduate work in system dynamics who do not...


I share somewhat similar concern. I like the way it was expressed by Dr.
Richardson.."I hope it is not rude but rather deep kindness..". Lacking
multidisciplinary knowledge and experience, successful SD practitioners
usually find that SD effort requires "team work".

Even working within a team work environment, modelling even a small part
of a problem, to a point where everyone find it "logical and make sense"
and "buy-in" to a model, is not a simple exercise.
--

Nordin Othman, Senior Consultant
SAPURA ADVANCED SYSTEMS,
18th Floor, Menara Tun Razak,
Jalan Raja Laut, 50350 Kuala Lumpur,
MALAYSIA
nothman@mail2.sapura.com.my

Tel: +6 03 294 3000
Fax: +6 03 294 6587
______________________________

[Alexandre Rodrigues]

I read and was very impressed with this message from George Richardson.
This is in my opinion a most critical issue in System Dynamics.

Does this topic relate to the fact that Universities, as institutions,
are inheriting the practices within business organisations? and that
this is then trasnferred to the students mentalities? Within business
organizations there is an ever more strong pressure to deliver products
quickly into the market, with the customer and the producer often not
fully understanding what they have developed. The products are sold
because the market is there to buy them, but the standards of quality
go down.

SD software packages allow the quick development of models and
production of fancy results, in a short periof of time. As we know, the
methodology (like many others), lacks a mechanism to test whether the
work developed is "valid" or not. It is therefore an obvious tentation
to develop SD work quickly, appearing to be highly professional, whilst
giving little attention to validity. This is particularly true whe the
models are developed in a context where they do not have to comply with
actual data from the real world.

It is fairly easy to learn to play a simple melody in a piano, even
though somewhat out of pace and tune. But it is extremely hard to
become a prefessional pianist. But what is the market interested in?
High quality symphonies, or just quick easy-to-sell but rough melodies?

I dont know the answer, but my natural view is to advocate quality
and serious scientific work. It may well be that if the market happens
to perceive poor quality in the SD practical initiatives, the
methodology will inevitably loose credibility.

Best regards,

Alexandre Rodrigues
alex@mansci.strath.ac.uk

[jm62004@Jetson.UH.EDU]

Greetings everyone

I will try to respond to Dr Richardsons (GR) and John Holdens (JH) emails
in my message here.

I believe GR was responding to some recent queries such as "how do i start
in system dynamics/Stella and how do I use it in my graduate work??". GRs
response is to work in this area, take a few courses, get familiar with
the SD lit. and **THEN** start working on your grad work
(thesis/dissertation). JH responded by saying that even tho courses exist
teaching STELLA-type *applications*, no courses on *system dynamics* exist
per se.

My comments:

Are we talking here about the "theory of system dynamics"? Based on my
limited understanding of SD, there is no theory of SD, as, say, exists in
microeconomics (hence the economic critique of SD) or as in
Newtonian/Einsteinian physics. We dont start from first principles, but
talk mainly of applications, taught through very user-friendly software
(Stella/powersim etc.). I realize there are "principles??" of feedback,
nonlinearity, delay, etc. but these are more descriptions of what kind of
real system behavior you can easily model using SD, and not principles of
SD as such.

The courses teaching the cleanest **theory**, in the classical sense, of SD
that I can think of are, well, courses in ordinary differential equations
(ODE) (linear and nonlinear, with twists such as delays, pulses, and
other many fancy functions thrown in), and courses in numerical solutions
of ODE. But that defeats the purpose of traditional SD modeling in the
first place.

To explain with an example, which is easier? taking World3 model, if I ask
you to give me the expanded ODEs describing population cohorts, it will be
EXTREMELY hard to come up with ODEs similar to the ones describing the
population submodel in World3. BUT, using STELLA, and the SD method, as
shown in the book Dynamics of Growth in a Finite World, it is MUCH EASIER
to build the model, understand and simulate it (I know the ODEs
corresponding to expanded forms of the World3 models are extremely large
(At times *hundreds* of pages of math text, as I discovered when I tried to
use them for my own dynamic game models, when I needed the expanded forms
for optimization).

My BOTTOM LINE point: SD is a *method* that works, without much of a
theory, but it is extremely useful, from pedagogical, communication, and
understanding points of view, and can be best learnt by studying many
applications and by continuous refinement of models. If JH were looking
for the theory of SD, then math (ODE) and CS (numerical analysis) courses
would be appropriate.

SD can have a theoretical framework if it steals it from the theory of
differential equations (the traditional electrical engineering approach of
describing the behavior of nonlinear plants). The problem is the economic
critique: the differential equations describing social systems (unlike
physical) are mostly out of the blue, and statistically untested. But how
much of that really is a problem: I think every SD model potentially
defines its own assumptions and logic, hence its own theory, and IF it
describes reality better than a simplistic, but statistically tested
economic model, I would still choose the SD models (of course wiith
limitations). I wonder if someone on this list knows of empirical success
rates of (neoclassical) economic versus SD models. For very complex
systems, specific & problem-oriented SD models may be the only way to go.

In this regard, Id end with a quote from "Fronties of Complexity" by Peter
Coveney and Roger Highfield, in which they are discussing theoretical
limitations to mathematics (cf Godel, Chaitin, Turing).
Discussing complexity of something as the size of the smallest program
which computes or describes it, and compressibilty as the measure of size
of that program, the authors say that if a property of the real world is
algorithmically incompressible, the most compact way of simulating its
behavior is to observe the process itself. I think SD, in the limiting
case, does exactly that with its continual model refinements and matchings
with reality.

Sorry for a rather long note - I hope it can generate some more discussion
in this hot and controversial area.

Regards,

Jaideep

Jaideep Mukherjee, Ph. D.
Research Associate
Department of Industrial Engineering
University of Houston
4800 Calhoun Road
Houston, TX 77204-4812
jm62004@Jetson.UH.EDU
Phone: 713 743 4181; Fax: 713 743 4190
****************************************

[Alexandre Rodrigues]

I believe that Richardsons point was that System Dynamics is not a
computer application, but rather a scientific modelling methodology. I
also believe there is nothing wrong with developing models as you
describe, but this should be seen as an intiation stage of the modellet.
Of course, a lot depends on what is the purpose of the models, their
complexity, and how the work developed is presented to its audience.

Best regards,

Alexandre Rodrigues

------------------------------------------------
Alexandre Jorge G. P. Rodrigues
Department of Management Science
Strathclyde Business School
The University of Strathclyde
Glasgow G1 1QE
United Kingdom

Tel: (0044) +141 552 4400 x4361
Fax:(0044) +141 552 6686
email: alex@mansci.strath.ac.uk
http://www.strath.ac.uk/Departments/MgtSci/alex.html

[Chris Schroeder]

I think that Richardson raises an important issue, but do not necessarily
think that it means that students should not use modeling in their
research. The important thing is that we all need to be sure to recognize
that all models have limitations. The relationships and assumptions used
for any model (whether developed by an experienced modeler or someone who
just cracked open a book) must be examined closely to be sure that it makes
sense. The day we start relying on models without taking the time to
understand the relationships and assumptions is the day that this whole
science will lose credibility.

Perhaps this suggests that the most important aspect of model building is
in the development of clear documentation of the relationships and
assumptions???

Regards,
Chris
========================================================
Chris Schroeder
rcs@centrec.com
AEC/Centrec
3 College Park Court
Savoy, IL 61874
Phone (217)352-1190 Fax (217)352-1425
http://www.centrec.com

[Alexandre Rodrigues]

I have found the comments below from Dr Mukherjee most interesting. They
move
the discussion further down to the deepest issues of what SD is supposed
to be (as initially proposed by proposed by Forrester), what it has
evolved to become, and what we want it to be in the future: a theory to
describe systems? a method to simulate systems as continuous processes
regulated by information feedback? or a simulation-based computer tool
that helps conceptualizing and solving complex differential equations?
Other views are, of course, possible.

As defined by Forrester in Industrial Dynamics, SD consists of
investigating the *feedback* character of complex (industrial) systems,
through the use of *computer-based simulation models*, with the aim of
helping to re-design and improve *organisational form* and *managerial
control policies*. This identifies the principles of SD:
(1) the behaviour of social systems can be explained, in great part, by
the information-feedback phenomenon, which occurs and results from the
many reciprocal internal interactions;
(2) models can be developed to *describe* these feedback mechanisms that
partially *explain* how the system behaves;
(3) these *explanatory* models can and are better implemented using
computer simulation. Feedback necessarily implies reciprocal
interactions *over time*. Simulation is the obvious candidate to
(re)create these interactions; digital computers appear to do the job
well;
(4) as a model eventually describes the feedback mechanisms embodied in
a real system, and further expains how these drive its behaviour, the
knowledge embodied in the model can be used as the basis to exercise
reasoning on how, if the system were different, its behaviour would
change. In this sense, a model works as an explanatory theory (as
discussed for example by Barlas and Carpenter 1990).

But is SD in itself a theory? SD certainly builds upon other theories,
but that does not prevent it to be a theory on its own. A theory can be
seen as an object of structured abstract knowledge which explains how
human-kind perceives how something works (including human abstract
creations). As described by Roberts (1978), and by many other authors,
there is no doubt that SD sought inspiration in the control theory and
principles of feedback, in the emergent studies of the human
decision-making processes, and in computer simulation (see also
Richardson 1991). It is also true that the Engineering field also sought
inspiration in the Control Theory (and helped to develop it) and found
critical support in digital computers. It did and continues to do so.
The term "System Dynamics" is often used in the engineering field (e.g.
see Shearer 1967, Seborg 1989, and Ogata 1993). The four points above
apply to any kind of system driven by feedback, and Engineers found that
too. Forrester made it clear that Engineering models provided the
appropriate precedent for the social systems.

Now, where do the mathematics of ODE come from? Mathematics seem to be
ultimate level of formality to where knowledge can be structured. In a
sense, it can be seen as the ultimate level of understanding. Although
this is debatable, one tends to push knowledge towards mathematical
structuring as one understands it better. Of course, mathematics is made
of relative conventions which over the years have proven to fit well in
the human framework of abstract reasoning; yet this requires thorough
education. The description of feedback mechanisms, if taken to this
level of formality, does lead to the mathematics of differential
equations.

SD has been criticised since the early days as representing no more than
a system of ordinary differential equations, often not properly
validated, and which aim at describing reality (see Berlinsky 1976, "On
Systems Analysis"). But I have serious doubts that Forresters initial
thoughts were: "The behaviour of social systems can clearly be described
by ordinary differential equations, but as these are so complex that
they are difficult to conceptualize and to solve analytically, let us
develop computer tools to do the job." Of course, after one has learnt
the mathematics of DE, one can anticipate when it can be applicable to
certain situations. But the process of developing a feedback-based
representation of a system, to decide what to include and what no
neglect, to identify and recognize the importance of the feedback
forces, to relate these forces to observed behaviour, to quantify the
individual interactions within the system, to continuously debate and
revise all this by analysing and simulating this representation, to
validate all this while moving towards the perceived reality, and
ultimately to learn more about the real system, and eventually identify
alternative structures which provide improved behaviour, all this looks
to me as going far beyond the theory of ODE. Differential equations
emerge as the ultimate level of formality to *define* (but a difficult
to read representation) our beliefs on how a feedback system works, and
the consequences of such beliefs. Yet, neither they are the objective
purpose of an SD study, nor they are the appropriate means to represent
and learn about the system, and find better solutions.

In my opinion, and while building upon other theories, SD provides a
rather novel way of studying social systems. A new way of structuring
knowledge about these systems, of explicating the feedback nature of
their behaviour, of helping to identify solutions for real problems in
the real world. Is it a theory on its own? The answer is debatable. It
is certainly much more than a computer application that helps developing
and solving ODEs about social systems. It raises critical philosophical
issues about our understanding of social systems; it raises critical
methodological issues about modelling in the wide field of OR/MS, like
"validation". To understand whether there is a theory of SD, and what
this theory is about, one needs to study the other uderlying theories
and find the steps given further by SD. This takes scientific
discipline, time, and effort. Of course, the same is not true about
developing models as an art. These "artistic" models may well prove
useful sometimes. But as Forrester argued, the way ahead from art to
science is to distill and structure knowledge gained from empirical
experience.

All theories take time to emerge, to prove themselves, and to be
recognised as such.

Apologies for such a long message, and thanks to JM for raising so
imporant questions and ideas.

Best regards,


Alexandre Rodrigues
alex@mansci.strath.ac.uk

[Anupam Saraph]

* George Richardson wrote
* I am becoming increasingly alarmed at messages to this list from people
* doing graduate work in system dynamics who do not appear to have had
* course work in the field or knowledge of system dynamics literature. =


Population---> SD aware---> SD student---> SD novice---> SD practitioner--->SD expert


Can we talk about criteria to recategorise the memberships of each
accumulator? Can we talk about the throughput? The cycle time? And most
importantly the PURPOSE of each accumulator? Can we REDESIGN this
process?

We ALL do mental modelling all the time. How many pause to refine and
understand the tools of our thought? Are we not guilty of using our
thought much the same as a wordprocessor most of the while? How open are
we? How rigorously can we question the assumptions we hold the deepest
and strongest?

Like good thinking practice, SD is a discipline of the mind. Like any
discipline can it come without practice? Does practice need to know of
the discipline of others? Does the discipline of the mind not stem best
when the mind can learn to watch itself and its design? Or does it stem
from more and more thinking on anything (applying thinking?)?

Are there programs to examine the DESIGN of SD? Who has observed the
design of SD? Do we all stress so much on applying SD that we forget to
watch SD at work?


Jaideep Mukherjee wrote

* Are we talking here about the "theory of system dynamics"? Based on my

Prof. Forresters "Principles of Systems" are generally regarded as the
theory behind SD. A valid viewpoint would be to regard them as
descriptions of system behaviour as they follow from the design. In the
language of theories regard them as theorems if you will.

SD is more than an operational method to represent the world using
integrations. It is an important and extremely practical theory about
feedback dynamics.

SD has been restated as a theory (Anupam Saraph, Toolbox for Tomorrow :
Exploring and Designing Sustainable Systems ISBN 81-7161-378-0).
[ send email to saraph@giaspn01.vsnl.net.in for details[

SD can therefore be learnt not just by studying applications but
learning the design of SD. I teach quantitative techniques to the
Masters level students and to teachers at the University of Pune who
learn SD (and other theories) by understanding the design of modelling
tools. The objective is not to generate experts in SD (or any other
theory), but rather generate

a) awareness of the designs of tools they will use in the future and=
b) produce novices who will try their hands at understanding the utility
of these theories in light of their design

Pre-PhD and PhD students would be expected to develop the ability to
redesign theories and then apply them. So would the teachers who hope to
use these in their research. As a believer in systems, evolution,
adaptation, and in reorganisation what is the point of producing the
expert as a conformist?

Regards,

Anupam Saraph

Anupam Saraph, Ph.D.
11 Mangalam Paud Road, Pune 411038, India=20
Tel/Fax: +91-212-346191 E-mail: saraph@giaspn01.vsnl.net.in
Consultancy/Training
esearch/Business Process Reengineering/Systems =
Thinking/Simulation/Modelling

[Phil Odence]

I agree with much of the discussion following Georges point: Learn SD
first, write dissertation second. I am concerned, however, about scaring
business people off, because I am firmly convinced that even people new
to SD can derive great value from SD modeling.

Wendell Gunn asked, "Whats the catch?" Chris Schreoder and Alexandre
Rodrigues touched upon it. Theres a danger in over-valuing output of a
model, without understanding the assumptions behind, so caution is
advisable. But, if you view the act of modeling as a learning process,
basic modeling skills are sufficient to enable you advance your
understanding of your business beyond current state. And thats pretty
good.

***************************************************************************
*
Phil Odence
podence@hps-inc.com
High Performance Systems
45 Lyme Road, Suite 300
Hanover, NH 03755
voice- 603 643 9636 x107, fx- 603 643 9502, web- http://www.hps-inc.com

[jm62004@Jetson.UH.EDU]

Thanks all for responding to my email. I have no problems with the new and
very user-friendly perspective that SD provides to understanding and
controlling complex systems (I am already sold, so you dont have to
convince me;-)). My concern is that one can teach a whole course in SD
without ever mentioning ODE. While this may be okay for simple models, for
larger and more complex models, I truly think that you cannot survive with
just the "Principles of Systems" that some have mentioned (it is no
surprise that many good system dynamicists have been engineers, including
the initial World3 team, and the guru of SD, Dr. Forrester, himself. Not
having any knowledge of the math behind SD can lead to sloppy, dangerous
mistakes - many times while modeling in SD (using Stella) myself, I have
had to go to basic principles, hence what I said about ODE as the "pure"
theory of SD).

I do feel comfort and safety in the fact that the whole mathematics of ODE
supports SD, so we cant quarrel about that. ODE theory is very
well-established - it is in SD theory that similar rigor is lacking. Once I
am comfortable that SD theory (based on established math) is sound, I can
direct my attention to the other criticisms of SD models, namely, do the
relationships, assumptions etc. make sense or not (as Chris mentioned too)?

An example - Input-Output analysis: there are books describing the methods,
interpretations of various coefficients, etc. in great detail, and all
these are quite useful in their own right. However, as soon as you start
asking for the theory of IO analysis, Id say well go and take a course in
linear algebra. How can one make theoretical advancements in IO without
understanding matrices and how can you make theoretical advances in SD
without understanding the math behind it (I am not talking about verbal
philosophical arguments on system dynamics)? Even the user-friendly
packages such as Stella/Powersim promote themselves by promoting their
mathematical powers. Similar arguments would apply to LP theory versus LP
applications and so on.

Am I still missing the point?

(Thanks again for giving me food for thought)

Regards

Jaideep
jm62004@Jetson.UH.EDU

*******************************************************
Jaideep Mukherjee, Ph. D.
Research Associate
Department of Industrial Engineering
University of Houston
4800 Calhoun Road
Houston, TX 77204-4812, USA

http://www.uh.edu/~jm62004/jm97.html
Office Phone: 713 743 4181; Fax: 713 743 4190
*******************************************************