Though I have certainly not simulated on the scale of millions of
agents, I have found that beyond some fairly low scale it is not the
number of agents but the population density (in the sense of how many
other agents any one agent could in principle observe of communicate
with) that matters. For example, in the markets model described in the
Moss-Edmonds paper, I specified the radius within which agents could
observe and communicate with one another and the agents were allocated
randomly over a grid. By reducing the size of the grid but keeping
everything else the same, the sort of turbulence described by Norman
Johnson emerged.
I conclude from these results that the importance of scale is not
independent of other factors such as (but probably not exhaustively) the
intensity of social interaction within the population of agents.
On 08/11/2008 10:07 AM, Peer-Olaf Siebers wrote:
> Here are my random thoughts...
>
> In order to study the real effects of policy changes one needs to get the
> response variability of the model system to the same level than the real
> system. Therefore, you need the right level of heterogeneity of the agents
> but also a scale model would help to account for unforeseen effects that you
> do not model explicitly, at least that is my experience. Often the size of
> the modelled population regulates the effects of policy changes. It is also
> important to remember that a group is not just a collection of individuals
> but develops its own dynamics. These might be different for different group
> sizes. Bigger groups might create additional dynamics and consequently the
> levels of variability and stochasticity within the simulation model will
> change even if the proportions of stereotypes within the groups are kept
> constant (is that what people often call emergence?). Therefore, I think,
> size matters!
>
> Another aspect to consider is validation. Previously I initiated a SIMSOC
> discussion about validation of ABMS models. From there I learned that AMBS
> models are often validated quantitatively on a micro level and qualitatively
> on a macro level (see for example Moss and Edmonds, 2005). I guess that
> having a scale model of the real system definitely supports the qualitative
> validation on a macro level and might even allow some (limited) form of
> quantitative validation on the macro level.
>
> Regards
> Peer-Olaf
>
> Reference:
> Moss, S. and Edmonds B. (2005). Sociology and simulation: statistical and
> qualitative cross-validation. American Journal of Sociology 110:1095-1131.
>
>
--
Scott Moss
Professor of Social Simulation
Centre for Policy Modelling
Manchester Metropolitan University Business School
Aytoun Building
Manchester M1 3GH
United Kingdom
t: +44 (0)161 247 3886
m: +44 (0)776 968 9991 or +44 (0)7733 484991
f: +44 (0)161 247 6802
|