Re: Modeling the Emergence of Political Parties

[Paul Johnson]

"glen e. p. ropella" wrote:
> 
> Be careful saying stuff
> like "With simulation, the problem is that we often have no clear
> notion of what a result is..."
> 
> That is absolutely not true!  There is a whole body of knowledge
> on how to do simulation.  And it contains reams of instructions,
> methodologies, definitions, etc. on how to correctly simulate
> something.  Of course, you can't look to science for this info.
> It's all in engineering, which, I admit, is overcrowded with
> all sorts of technical mumbo-jumbo.  But, it is there.
> 
I think I am using the term result in a different sense than you are. Of
course, there are technical procedures to create and evaluate
simulations.  It it is technically possible to tell if two sets of runs
are significantly different, or if one distribution causes a set of
outcomes different from another.  

But in a substantive sense, we have trouble knowing which of those is a
"important results".  This is a problem in engineering and political
science, because the methodology is able to generate "internal validity"
(internal coherence and correctness) rather than theoretical knowledge.  

Perhaps a statistical analogy will help me.  We have tons of experience
with a t-test of the conjecture that two classes of students are drawn
from the same population.  We can t-test that hypothesis all kinds of
ways.  And, in the end we may reject the claim that they are identical. 
Now, is that an important result?  The methods don't tell us one way or
another.  Whether it is important or not depends on a bunch of
theoretical structure that has to exist in the substance of the research
topic and its community, not the method.  One may conclude that a
statistically significant difference between two classes is not an
important result, for example, if the difference is extremely small or
the cost of making them equal would be too great. 

Here is a political science example.  One result in multidimensional
voting is that there is no equilibrium point, generically.  We have tons
of theorems with all kinds of topologies and it is considered proven. 
Considering the space of all possible societies, except on a set of
measure zero, there is no majority rule equilibrium.  Now if you
simulate voting processes, you find not so much chaos as you might
expect from these theorems. In fact, I have been astonished how stable
majority rule seems. It wanders into the middle of the voter's
preferences and stumbles around in there.  

Now, is that a "result"?  Well, I think my colleagues who do formal
theory are hard to convince. They expect a result to be an existence
theorem which states the existence of equilibrium under a set of
conditions, or a proof that there is no equilibrium under those
conditions.  These kinds of things we find in simulations, such as "look
at this odd run, these guys wandered off and did a peculiar thing" are
tough to sell to the traditional modelers.  Part of their skepticism is
inspired by this "black art" characterization of simulation, but there
is more to it than that.  The math folks simply think they are working
on a more important problem by searching for characterization theorems.

I think the clearest, easiest to justify simulation result is obtained
like this. SOmebody writes a book using informal tools or differential
equations to make some claim that something is likely to happen.  That
gives you a nice null hypothesis, and you can then build a simulation
according to their story, and see what happens. If it does not match
their prediction, you have a great result. If it does match their
prediction, people might not be very interested, but it could still be
useful.  The key is that, whether or not you have a result depends on
the research community, their questions, claims, and your ability to
participate in their argument.

So, unless I misunderstand your comment, I still don't think I'm far
from the truth.

-- 
Paul E. Johnson                       email: address@hidden
Dept. of Political Science            http://lark.cc.ukans.edu/~pauljohn
University of Kansas                  Office: (785) 864-9086
Lawrence, Kansas 66045                FAX: (785) 864-5700


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