Truth

Truth is a difficult philosophical concept that will not be addressed in this paper. Instead, we take a more pragmatic approach relating to model purpose and validation.

Social scientists following the methodological position of Popper or Friedman essentially identify the truth of a theory or model with its predictive power. Nuanced differences in their positions are not germane to the present discussion. Experience shows that social science models do not have predictive power whether in the short or the long run.

Other social scientists adopt a post-modernist stance which, following Derrida, asserts that any text has to be interpreted and subject to criticism entirely on its own terms. In social scientific terms this comes down to something like "reality is socially constructed" so that one cannot evaluate any social analysis without knowing the social basis of the analyst.

Evidence-driven agent based modelling with expert/stakeholder participation falls in neither of these methodological camps. The purpose of the modelling is to capture the perceptions and self-described behaviour of the stakeholders in a way that formalises their expressions of those perceptions and descriptions. The validation the models capturing those perceptions and descriptions is achieved by

• implementing agent processing as a rule based implementation in which the rules are logic-like if-then statements couched in the linguistic terms used by the stakeholders and experts themselves
• presenting suitable formulations of the rules to stakeholder/experts for confirmation that the rules suitably capture their views
• implement the models to produce traces of agent behaviour and social interactions together with descriptions of the reasons for such behaviour and interactions
• present suitable formulations of the traces to the stakeholders and experts for confirmation of their plausibility, implausibility or simply their unexpected features.

Evidently, the purpose of the model is not to provide some objective or external truth but to capture actors' and experts' perceptions in ways that they find appropriate and enlightening.

This statement of ``truth'' implies an appropriate constraint on simplicity - but it is a constraint on maximal complication rather than maximal simplicity. In relation to the obiter dictum of William of Ockham, the number of entities in the model is determined by evidence and available computational resources. The modeller specifies the types of individuals and such groupings as households and churches. Agents are designed to represent individuals in these groupings. The number of agents is the lesser of the size of the population being modeled and the number of agents that can be implemented within the processing and memory constraints of the hardware and the time constraints of the modeller or model user.