REPLY Structural changes and validity (SD6660)

[SDMAIL John Gunkler]

Posted by  "John Gunkler" <jgunkler at sprintmail.com>

Monte writes [numbers added]:

"(1) If the real structure changes, the model of past behavior would
  lack the ability to generate the future behavior. (2) It is therefore
  hard believe that SD models are theories, unless the structures
  of the aspects being modeled are stable.
   
  (3) SD is not effective in dealing with social problems, of which
  the dynamics of the system structures are not yet predictable."

Point (1):  Can't argue with this.  We build models, in part at least, to
understand why a system behaved the way it did in the past AND to predict
the way it will behave in the future ABSENT ANY FUNDAMENTAL CHANGES IN
STRUCTURE. We're saying, essentially, that "unless you change something
important to the system's dynamics, here's what you can expect."  Even the
theories of physics have (stated or assumed) limitations -- e.g., Newton's
laws of motion don't apply precisely where there is a very strong
gravitational field; Einstein's General Relativity doesn't apply in the
sub-atomic world; etc.  So, in SD models, our "limitation" is that the
fundamental causal dynamic structure stays intact.

Point (2):  My understanding of what is a scientific theory, from the
philosophy of science I studied, is that it is essentially a prediction of
future behavior of the entity under discussion.  How is an SD model (as
described in Point (1)) not a theory, then?

Point (3):  I would suggest that the referent of "the dynamics of the system
structures" has changed from what it was in Points (1) and (2).  In the
previous sentences we are discussing (I think) the fundamental structures
that are responsible for the dynamic behavior of the system.  In Point (3),
I suspect, we are now really talking about system behavior.  And, in fact,
SD models have a LOT to say about how systems will predictably behave -- how
various structures will interact, how those interactions will affect the
behaviors/outputs of the system as a whole, etc., etc.  One of the strengths
of an SD model is that it makes the complex "dynamics of system structures"
predictable.

Finally, I will admit that the issue of fundamental system structure changes
is very interesting to me.  In the current state of SD thinking, I believe
what we say about fundamental system structure changes can be boiled down to
these two things (oversimplifying somewhat):

1.  A System Dynamics model helps us understand what ARE the fundamental
causal structures in a system.  This, then, will allow us to monitor reality
and look for changes to those fundamental structures, which will then lead
us to change the model and, thus, be able to predict future behavior.  It
highlights what is important, and helps us to ignore what isn't --
tremendously useful in the complex situations for which we use SD.

2.  A System Dynamics model, by showing us what the fundamental causal
structures are, can help us make deliberate changes in those structures that
will lead to predictable changes in system behavior.  A good SD model can
show us where to "put the lever that will move the world."

Here's where I agree with Monte.  I would hope to see, within the SD method,
a better way to model structural changes.  For clarity, let's use Jay
Forrester term "policies" to describe a fundamental model structure.
"Policies" are represented by rate equations.  Policies change in the real
world, and there dynamic mechanisms behind (some of) these changes.  I wish
there were a better way in SD modeling to show how one policy (rate
equation) can become a different one.  We could then have models that
adapted to, or learned from, their environments.  While I can think of some
workarounds to do this (e.g., embedding an IF-THEN structure in a rate
equation), I can't think of ways that maintain the transparency that is the
strength of the SD method.

John Gunkler
Posted by  "John Gunkler" <jgunkler at sprintmail.com>
posting date  Thu, 4 Oct 2007 09:50:27 -0400