Ventana software support forum

Posted: **Fri Jun 04, 2004 10:47 am**

Dear Members,

if SD modeling can be understood as part of inquiry cycles (for instance
the figure in the introductory part of Sterman 2000, p.34), then it
might well be that a model ""M"" that has been developed and used (as
""valid"") at a point in time ""P"" has to be revised and possibly
modified. I believe that this must happen sometimes, like in the car
market example given in Sterman (2000, p.34), and then we may think om
model ""M"" in successive versions.

One might want to measure how much the resources, the sectors, the
decisions and the loops have changed between two such versions, since
this will tell something about the learning that is going on between
them.

I have defined some indicators in order to do this (I'll even have the
chance to present this in the Oxford conference); however I do not have
a model with successive versions to apply this measures to.

If someone out there knows of such a model or has some he or she would
be willing to ""lend"" me in order to try out the idea, this would be very
kind of you.

Thanks,

Martin Schaffernicht
University of Talca
Talca - CHILE
From: Martin Schaffernicht <martin@utalca.cl>

Posted: **Sun Jun 06, 2004 12:14 am**

I nominate the hog commodity market model originally published by Meadows in
his classic book on commodity cycles, and partially updated by Sterman. I
tried and tried to make either model work with current hog market numbers
and failed to get it to behave -- probably due to me ineptitude. (However,
the original model was not very robust with respect to numbers out of
expected range, and those numbers have occurred.)

From: ""John Gunkler"" <jgunkler@sprintmail.com>

Posted: **Sun Jun 06, 2004 5:31 pm**

Hi Martin

This question is not only fundamental to SD modelling but to any method that
is systematic,

as long as the environment has not changed too much between the two
successive models.

I would be much interested to know if the first model had any influence on
the decisions taken and if after a period of time, on can evaluate if this
influence had any beneficial effect. If it had no beneficial effect, is it
possible to learn something from it, so as to be able to construct a second
model more efficient.

I have started many models that influenced more or less my decisions, but
having no method to revise them after a while and to measure the utility of
their influence, I have left them aside and they are now waiting an eventual
awakening in a problematic future.

In his book Sterman has just exposed the problem (page 20 and 34,35).

In another part of the Business Dynamic Sterman writes about replicability
page 855 and especially page 857 beginning of page about the few cases where
researchers build about the original work of their predecessors. It is
certainly because first the original work were not build to ease the
continuation of the research and that for some reasons, people lack a method
to evaluate what is good and bad in any study AFTER A PERIOD OF TIME, and
what to do to change it.

I am greatly interested to learn something about the subject.

Jean-Jacques Laublé
From: =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <JEAN-JACQUES.LAUBLE@WANADOO.FR>
Allocar, Rent a Car
Strasbourg France

Posted: **Mon Jun 07, 2004 10:33 am**

Martin,

If I understand you correctly, you are trying to model the change of a
model as a result of learning.

George Richardson presented a paper on this subject in Norway (2000:
Can System Dynamics Models Learn?).

Bottom line of his approach - as I understand it - is that SD models
can only use other structures if these
structures are allready present in it.
Subsequently, (this type of) learning implies the answering of the
reperating question which structure is
best given a number of criteria. And that leads to two problematic
questions:
1) which structure should (also) be added to the model?
2) which criteria should be used to determine when and how a model has
to adopt another structure?

I tried this approach in modelling the Dutch asylum procedure over a
period of 25 years.
I defined different procedural routes as different structures.
Subsequently, I tried to model the triggering
event that implied a change of structure, i.e. using a different route
through the asylum procedure.

Anything beyond this, implies invoking other disciplines in model
building such as Artificial Intelligence.

Carolus Grütters
From: ÿSO-8859-1?Q?Carolus_Grütters?ÿc.grutters@jur.kun.nl>
Law & IT,
Faculty of Law
University of Nijmegen,
The Netherlands

Posted: **Mon Jun 07, 2004 1:43 pm**

Martin Schaffernicht is right on when he suggests that models are
always provisional and subject to change as conditions change. The
best and shortest statement on this principle is from 1970, when Jay
Forrester wrote a memo entitled "" 'The' Model Versus a Modeling
'Process' "", which was subsequently reprinted in the first issue of
the SD Review:

Forrester, J. W. (1985). ""'The' Model Versus a Modeling 'Process'.""
System Dynamics Review 1(1): 133-134.

John Sterman
From: John Sterman <jsterman@MIT.EDU>

Ventana software support forum

Successive versions of a model

Successive versions of a model

Successive versions of a model

Successive versions of a model

Successive versions of a model

Successive versions of a model