I think the discussion last week became too stark. Keith seems to be arguing
that theory should constrain (in Bruce's terminology) the models which should
then be applied in some way to observation. I think this is implicit in other
contributions as well. Bruce and I have been arguing that representational
modelling serves observation. My view in particular is that representational
social simulation modelling should be descriptive in the way that history and
journalism are descriptive.
There is plenty of scope here for the development of an analysis of when (and
whether) a particular theoretical approach will usefully constrain model
specification. There is also scope for inductive analysis of the model
specifications that are appropriate in the representation of classes of social
phenomena. When we have found a well verified modelling representation, it
would seem appropriate to consider whether such a specification amounts to (or
perhaps implies) an element of a social theory.
I find a modelling approach in which social norms emerge from agent behaviour
and interaction to be well verified and to enable me to explain observed social
phenomena. The scientifically open minded response from sociologists would
surely be to consider whether there is a general class of cases (of which,
presuming they are well validated, my models identify instances) in which
social norms and institutions are appropriately considered as the outcome of
individual behaviour and interaction. My objection is to the dismissal of
models that are well validated and plausible on the grounds that they are
methodologically individualist.
At the same time, scientific open mindedness should induce social simulation
modellers to constrain representational models by particular formalisms or
social theories and to seek to validate those models as tests of their
theories. Models that are validated in several ways -- e.g. as descriptions of
both observed individual behaviour and observed system properties -- will
clearly be better validated than models validated at only one of those levels.
The repreated validation of such theory-inspired models should give confidence
to the representational social simulation modeller in constraining model
specifications on the basis of the same theories.
Such open mindedness and interaction between theory and observation seems to me
to be the essence of good science. What I have described above (and what Bruce
argued for in the paper giving rise to this discussion) is modelling is (or can
be) relevant to that sort of good science.
Keith Sawyer wrote:
> >In the normal terminology, you get your model validated by comparing the
> behaviour
> >of agents and multi agent systems with the people, groups of people,
> organisations
> >or whatever your agents and multi agent systems are intended to represent.
>
> The scientific study of these things is called "sociology" (also
> "anthropology" "economics" and other social science disciplines). It
> sounds like you are accepting that sociology is relevant, thus I am puzzled
> by the following comment:
>
> >I just see these as pragmatic modelling decisions. If these decisions are
> >usefully to be informed by sociological knowledge, I would be very keen to
> know
> >how -- preferably by demonstration or concrete example.
--
Professor Scott Moss
Director
Centre for Policy Modelling
Manchester Metropolitan University
Aytoun Building
Manchester M1 3GH
UNITED KINGDOM
telephone: +44 (0)161 247 3886
fax: +44 (0)161 247 6802
http://www.cpm.mmu.ac.uk/~scott
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|