REPLY Structural changes and validity (SD6660)

[SDMAIL Tom Fiddaman]

Posted by  Tom Fiddaman <tom at ventanasystems.com>


>No existing validity tests serve to check whether the relationships in the real system
>will not significantly change in the future--within the time horizon.

I don't believe that this is strictly true.

For example, as a matter of habit I include a first-order negative feedback constraint 
on inventories, which guarantees that the outflow (sales) is zero when the contents of 
the inventory is zero. I don't recall ever working with data that demonstrated that 
constraint in action, but its existence is common sense. Extreme conditions tests and a 
variety of other sources of knowledge can reveal similar latent feedback loops that 
might become active at some point in the future.

Perhaps my example is most relevant to physical systems, where conservation laws and so 
forth somewhat constrain the possible structural changes. Still, even in models of 
social systems, it is possible to speculate about feedbacks that might emerge. I could 
imagine testing a model by randomly adding terms to every rate equation that created 
arbitrary perturbations of the system structure, in order to discover which, if any, 
might be interesting or important.

We may now be witnessing the effects of climate feedbacks that represent changes in 
loop dominance, e.g. rapid progression of ice-albedo effects in polar ice. The outcome 
of those feedbacks is surprising in the context of global models, which are constructed 
with a bias toward avoiding making concrete statements about ill-parameterized effects. 
Yet the presence of novel feedbacks as the system is pushed outside its historical 
envelope is hardly surprising. The particulars of the social response to climate news 
is probably more unpredictable, but we could similarly expect the emergence of novel 
behavior.

The point is well taken that some changes in system structure cannot be anticipated. All 
theories, whether formalized as mathematical models or not, are subject to such 
surprises. This should not stop us from modeling social systems. The real trouble with 
surprise arises from a forecasting approach to problem solving: the fallacy that "if I 
know what will happen I'll know what to do." If one instead uses models to build robust 
feedback controls that work over a wide range of conditions, surprises from latent 
structure are not so problematic.

I think there are social phenomena that are difficult to represent in traditional 
aggregate SD models. Agent based models are a useful alternative for exploring their 
dynamics, but I don't think their track record for prediction or policy making is 
conspicuously better, nor are they somehow theories when other formalisms are not. 
I view the two approaches as complementary and frequently overlapping.

Tom

****************************************************
Tom Fiddaman
Ventana Systems, Inc. 
Posted by  Tom Fiddaman <tom at ventanasystems.com>
posting date  Fri, 12 Oct 2007 21:11:32 -0600