Re: Regressions and Rules of behavior

[Rick Riolo]

I think marcus and paul's answers have sketched
out the two alternative approaches, which in large
part depend on what one is trying to do with the model.

In one approach, as mentioned by Marcus,
one has some observations from the real world,
e.g., some regression-able relationships.
In this case one would like the model to generate those
relationships "from the bottom up", i.e., one models
the agents' decision rules in some "realistic" way
(the assumption being that people don't do a regression
in their heads and then decide),
and hope that generates the regression relationship
in your model just as it is observed in the world.
Its not so impressive if you had your agents use some regression-based
rule to make their decisions, and then you observe
the regressionable relationship in the model behavior.

But in another approach, one may not being
trying to build a model to predict/explain
the regression relationship in question.
Thus in the stock market sims mentioned they aren't trying to 
explain why some decision rules seem to be describable
by a simple regression.   They are trying to predict/understand
some more aggregate level phenomena.
Thus they model their agents by just building the regression
relationship into the rules---in effect saying "we don't know
how this regression comes about, but the details of how
it is generated are not important
for what we are trying to show from this model." (Of course
this assumption may or may not be right, but its often
a reasonable place to start.)
Then they see what kind of aggregate behavior results.

- r

Rick Riolo                           address@hidden
Center for Study of Complex Systems (CSCS)
4477 Randall Lab                
University of Michigan         Ann Arbor MI 48109-1120
Phone: 734 763 3323                  Fax: 734 763 9267
http://www.pscs.umich.edu/PEOPLE/rlr-home.html

On 17 Sep 2000, Marcus G. Daniels wrote:

> Date: 17 Sep 2000 20:37:41 -0700
> From: Marcus G. Daniels <address@hidden>
> Reply-To: address@hidden
> To: address@hidden
> Subject: Re: Regressions and Rules of behavior
> 
> >>>>> "RM" == Rayman Mohamed <address@hidden> writes:
> 
> RM> can an agent say, for example, decide whether to go live
> RM> somewhere based on a regression where some index is the dependent
> RM> variable and there are a number of independent variable?  Then, if
> RM> the dependent variable is within a certain range, the agent does
> RM> one thing but if it is in another range, the agent does another?
> 
> Usually, you'd do it the other way around, i.e. use the simulation to
> generate data from which you'd run a regression.
> 
> Usually, with agent-based modeling, you want agents described in such
> a way they are comprehensible and tweakable as programs, otherwise you
> may not have a clear picture of how many specific and individual
> agents behavior(s) generate a global behavior.
> 
> So you might run a regression to identify gross variable relationships
> and then use minor changes to agent rules and parameters to map out a
> more-detailed texture.
> 
> 
>                   ==================================
>    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.
>                   ==================================
> 



                  ==================================
   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.
                  ==================================