Our experience has been that
> 1) size vs realisim tradeoff: the more veridical the agent model, the
> more detailed, the more sophisticated the slower it runs for the same
> number of agents. Consequently: for the same run time you might have
> several million simplistic agents - or a million more socially or
> cognitively realistic agents or a few thousand even more realistic
> agents or a handful of extremely realistic agents.
>
> 2) population size versus geo-spatial size: if spatial temporal factors
> are in the model, simply reducing the number of agents leads to
> different outcomes due having greater amounts of space among agents and
> more possible interactions with the same other in the same time period.
> Ratios here matter!
>
> 3) for agent based models at least these factors influence run length,
> storage constraints, and realism of results
> a) number of interaction partners
> b) size of agent's knowledge base (size of the knowledge network),
> resource base, or task assignments
> Often ratio's control outcomes, e.g., the ratio of amount of knowledge
> available to be communicated to the number of people who can communicate.
> What these factors mean, is that if the agents are in or interacting
> through networks such as a social network and a knowledge network -
> these ratios will control outcomes. As the size of the group/amt of
> knowledge ...increaes different ratios are possible. Consequently, you
> can and do observe qualitative and quantitative differences for
> different size networks.
>
> 4) for interaction based networks, size and density control outcomes.
> You do get scale effects because for different network topologies, some
> are not even be possible at certain sizes and densities.
>
> 5) We did a study on our model where we asked - as we increase the size
> of the population modeled how does the size of the peak, the timing of
> the peak, and the relative arrival of different peaks for
> epidemeological events. The short answer is that the size of the
> population modeled matters.
> Imagine that a peak is number infected, and another peak is in number
> who go to a hospital. We find both qualitative and quantiative
> differences as size increases.
>
> An aside - we've also developed and simulated millions of agents at very
> high fidelity in the bio warfare and epidemic area.
>
>
|