Page 200 -

9 Verifying and Validating Simulations 197

become accepted by a scientific community, their influence seems to reach several
other disciplines and contexts. Perhaps for this reason, these kinds of models are the
most popular in social simulation, and some models are able to reach a considerable
impact in many strains of social science.

9.5.2 Context-Specific Agent-Based Models

It would be simplistic to say that models in social simulation can be characterised
according to well-defined categories of validation strategies. Even so, the capacity
to describe social complexity, whether through simplicity or through rich detail and
context, is a determining factor for a catalogue of modelling strategies.
We cannot hope to model general social mechanisms that are valid in all
contexts. There are many models that are not designed to be markedly general
or metaphorically general, but to stress accurateness, diversity, and richness of
description. Instead of using possible worlds representing very arbitrary contexts,
models are explicitly bounded to specific contexts. Constraints imposed on these
models can vary from models investigating properties of social mechanisms in
a large band of situations which share common characteristics, to models with
the only ambition of representing a single history, like Dean’s retrodiction of the
patterns of settlement of the Anasazi in the southwestern United States, household
by household (Dean et al. 2000).
Constructing and validating a model of this kind requires the use of empirical
knowledge. They are, for this reason, often associated with the idea of “Empirical
Validation of Agent-Based Models.”
What is the meaning of empirical in this sense? If the goal is to discuss empirical
claims, then models should attempt to capture empirically enquired characteristics
of the target domain. Specifying the context of descriptions will typically provide
more ways for enquiring quantitative and qualitative data in the target, as well
as using experimental and participative methods with stakeholders. In this sense,
empirical may be understood as a stronger link between the model and a context-
specific, well-circumscribed problem domain.
The Anasazi model by Dean et al. (2000) is a well-known and oft-cited example
of a highly contextualised model built on the basis of numerous sources, from
archaeological data to anthropological, agricultural and ethnographic analyses, in
a multidisciplinary context.
Given the higher specificity of the target domain, the higher diversity of ways for
enriching the model as well as the increased semantic specificity of the outputs
produced by the model, context-specific models may be more susceptible to be
compared with empirical results of other methods of social research. On the other
hand, comparison with other simulation models is complex and these models are
more difficult to replicate and compare.