Re: Advantages of agent-based modeling

[M Lang / S Railsback]

ceoji singh wrote:
> 
> Hi
> 
> i have just started out with agent based
> modeling...and have a number of questions..some of
> which i am are
> 1. what are the assumptions on which a typical agent
> based model rests.

Ah, I'm currently writing a book on this topic. The primary assumption
is of course that system behavior can be explained by individual traits
as the individuals interact and adapt to each other and their
environment. The modeler's art is then to (1) define an appropriate
modeling context- model of the environment, spatial and time scales,
etc.; then (2) find those individual traits (assumptions for agent
behavior) that explain the system behaviors of interest. So an important
difference from other types of models is that you have to be looking at
two levels: the individuals and the system. 

A paper that may help you understand this approach to ABMs is called
'Getting "results": The pattern-oriented approach ...' which you can get
here: http://math.humboldt.edu/~simsys/Products.html

> 2. how does the results from this technique compare
> with other simulation techniques.
> 3. what are the advantages and disavantages of
> agent-based modeling...in other words what is it that
> is so great and bad about agent-based modeling...

It would take quite an effort to answer your question completely, but
below is a bit of text I recently wrote for my field, modeling animal
populations and their response to various kinds of stressors. 

A widely cited paper on the (mainly, anticipated) benefits of
agent-based simulation in ecology is: Huston, M., D. DeAngelis and W.
Post (1988). New computer models unify ecological theory. BioScience 38:
682-691. 

However, this book may be of more interest and certainly is one of the
best for understanding when/why to use ABMs: Axelrod, R. (1997). The
complexity of cooperation: agent-based models of competition and
collaboration.
__________________________________

Advantages of ABMs for modeling animal population response to stressors:
· The ability to predict the complex, nonlinear ways that stressors
interact with each other and natural processes to affect population
properties. In ABMs, complex interactions are emergent, whereas in other
models the types of interactions must be foreseen and written into the
model.
· ABMs are naturally able to predict population-level effects of
stressors whose impacts are known at the individual level. Laboratory
studies, for example, can be used to model how a contaminant affects
individuals, and an ABM can predict the population-level consequences of
the effects on individuals.
· The ability to make many kinds of testable predictions that are
directly relevant to management decision-making. ABMs can, for example,
predict how population abundance and production, habitat preferences,
and age structure change with stressors.
· A good ABM is a very general model, representing individuals in such a
way that little, if any, revision is needed in applying the model to new
sites. Instead, only input to characterize habitat is needed at a new
site.
· ABMs are built and tested using a wide variety of information, not
just calibrated to field data sets. The often vast literature on
individual physiology, behavior, stress response, etc. can all be used
to design and test components of an ABM. In contrast, conventional
population-level models and statistical models face a tradeoff between
being too simple to model the processes of interest and having more
parameters than can be fit using field data, a problem that has
undermined other regional modeling efforts.
· ABMs readily model spatial and temporal variation in stressors and
their effects. Unlike many other classes of models, they can easily
predict transient responses to stressors that vary over time and space.

Steve Railsback

-- 
address@hidden
Lang, Railsback & Assoc.
250 California Ave., Arcata CA 95521
707-822-0453; Fax 822-1868


                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================