Re: [Swarm Modelling] Re: The "Art" of Modeling

[Steve Railsback]

Jason Alexander wrote:
> 

I'm finding this discussion very fascinating because my colleague Volker
Grimm and I are writing about the same issues but from a completely
different perspective: a very practical, how-to book on using
agent-based models in ecology. (If all goes well, Princeton University
Press will have it out by the end of the year.)

Volker's thoughts about these issues of explaining vs. reproducing, how
do you know how much detail must be in a model, etc., are based around
the concept that you start the whole process by defining some observed
phenomena that you want to explain. Usually these are system-level
patterns that you assume a priori emerge from individual-level traits
(which is why you use an agent-based model).

We are also using the word "theory" (please please don't shoot us!) for
a model of an individual trait that explains the system phenomena of
interest. But we're saying that there are several ways you can "test"
such a theory, and the ability to reproduce the system-level pattern is
just one. 

We are focusing on two ways especially for testing how explanatory an
ABM is- how well its rules for individual behavior explain real systems.
First, can it explain a wide range of observed patterns instead of just
one or two? Second, what evidence is there that the real organisms
you're modeling have traits similar to your model's rules?

An example: Jason mentions "boids" and the very interesting question of
whether the 3 rules driving boids just reproduce something that looks
lifelike, or are they really the rules that explain flocking in real
birds (or schooling behavior in fish). Turns out that some of Volker's
colleagues in Germany did a number of detailed experiments using both
agent-based simulation and real live fish and found out that the real
fish have physiological mechanisms producing behavior that is, in fact,
very similar to the rules in boids. And they used modeling to test
subtle alternative rules to find the ones that best reproduce schooling
in real fish.  

Perhaps a subtle lesson (also learned from my own experience trying to
apply "theory" from behavioral ecology to fairly realistic agent-based
models) is that dealing with a real system provides a lot of structure:
it certainly makes it easier to reject bogus models but also provides a
lot of information for finding and testing good models for what
individuals do.

Steve R.



-- 
Lang Railsback & Assoc.
250 California Ave.
Arcata CA  USA 95521
707-822-0453; fax 822-1868