REPLY Structural changes and validity (SD6660)

[SDMAIL Jay Forrester]

Posted by  Jay Forrester <jforestr at MIT.EDU>

On Oct 4, 2007, at 6:27 AM, SDMAIL Monte Kietpawpan wrote:
> Structual validation is key to the success of model development.
>   However, we have no test to determine whether a given model structure
>   is stable over the time horizon of interest.

A model is a hypothesis (a theory) about a structure that one proposes is 
causing behavior of interest.  The model is a theory in the same sense that 
Ohm's Law in electricity is a theory or Einstein's Theory or Relativity.  
There are no proofs of the general validity of such theories that are 
proposed to represent the real world, one only builds confidence in the 
reasonableness and usefulness of such theories within some domain of 
applicability.  For a system dynamics model, part of the asserted theory 
relates to the time dimension and the space dimension of applicability.  
By the space dimension, I mean the degree to which a model fits, given 
parameters appropriate to the specific situations, all members of the 
class of systems to which the system of interest belongs; is the model 
generic with broad applicability across systems as well as across time.

>  It is therefore
>   hard believe that SD models are theories, unless the structures
>   of the aspects being modeled are stable.

The duration of stability of structure is determined by how deeply one 
chooses to dig into the decision-making processes.  Let me illustrate with 
an example.  In ordering to maintain an inventory, there is a goal as to 
the proper amount of inventory.  But what is that goal, how is it defined, 
and over what time horizon should it be valid?  In the short run, the goal 
might be to keep 1000 units in inventory, but that would not be responsive 
to rising or falling demand for the product.  So, one could go further 
behind the scenes and say that the inventory should be equal to four weeks 
worth of sales and then the inventory would rise and fall with longer-term 
average sales.  But in the still longer term. why should the inventory be a 
fixed number of weeks of sales because that might change with the state of 
technology and with the speed and adequacy of communications.  What is a 
deeper objective?  It might be to be able to fill a given fraction, like 
95%, of incoming orders from stock.  So, the history of order fulfillment 
might be invoked to alter the number of weeks of sales to maintain.

Time horizon is important, but it is incorporated as part of the modeling 
process and as part of the alleged domain of usefulness of the model.

> SD is not effective in dealing with social problems, of which
>   the dynamics of the system structures are not yet predictable.

At a deep enough level, the social structures change slowly, or not at all.  
We can still draw relevant insights about human behavior from the writings 
of the ancient Greeks.  Histories of the rise and fall of civilizations show 
a repeating dynamic of growing beyond the capacity of the local natural 
environment and then collapsing, the deep human structure of behavior 
survives for centuries to produce repeated disasters.  The issue here does 
not arise from impossibility of modeling to fit a desired time horizon, but 
rather depends on the skill of the person doing the modeling and the 
correctness of the assertion about the domain of usefulness of a particular 
model.

_______________________________________________

Jay W. Forrester
Professor of Management
Sloan School, MIT
Posted by  Jay Forrester <jforestr at MIT.EDU>
posting date  Fri, 5 Oct 2007 21:47:01 -0400