[Swarm-Modelling] parsimony and usage

[glen e. p. ropella]

Marcus G. Daniels wrote:
> All I'm really saying is that, all things being equal, a parsimonious
> explanation is better than a complex one.

It depends on what the "explanation" is used for.  An explanation is a form of
description or representation.  So, to bubble up one notch, if a description is
being used to think or understand abstractly (with the concrete particulars
removed), then the more parsimonious one is definitely more useful.

If, however, the devil is in the details and abstraction prevents prediction
(which is often the case in even the most simple complex systems), then the more
parsimonious description is not the best.  Or in the case where the abstract
explanation teaches one solely about generalities without teaching us about the
impact of the detail, then the parsimonious one is not as useful as the
sprawling one.

> If an ABM doesn't get the big picture right, or does with lots of
> parameter settings that are hard to justify, but happens to have bubbles
> and crashes, that's no better either.

Even if an ABM fails to get the big picture right, it can often capture
mesoscopic lessons about component interaction that wouldn't be captured by a
more abstract (but correct) description.  And even if the assumptions
(parameters are one form) that go into the model are difficult to justify (as
assumptions often are by definition), an ABM can still capture lessons that
wouldn't be captured in a more abstract model.

Of course, I'm not saying anything the people on this list don't already know.
[grin]  But, the balance between "how abstract" and "how detailed" to make a
model _is_ the art that remains black in modeling.  And ultimately the answer to
where one should sit on that spectrum is "It depends."  Specifically, it depends
on the usage to which the model will be put.  Saying that one is better than the
other in all or even in most cases makes the decision appear simpler than it is.

I think all Pietro is saying is that sometimes the devil really does lie in the
details and you need to represent those details in order for the model to
exhibit the phenomena you're looking for.

-- 
glen e. p. ropella, 971-219-3846, http://tempusdictum.com
He who refuses to do arithmetic is doomed to talk nonsense. -- John McCarthy