Introduction

As in any human endeavour, social modellers tend to conform to a range of social norms. These include constraining model design by appeal to some social theory, designing models to have the smallest possible number of degrees of freedom and producing results that can be interpreted as useful indications about the nature of social processes and/or useful guides to public policy. It is by no means unusual for social theories and analytical techniques to be closely related. Properties and implications of n-person games and general economic equilibrium models, for example, are derived by appeals to fixed point theorems; macroeconomic models are closely bound up with econometrics. Moreover, it is not uncommon for assumptions to be made ``for the sake of simplicity'' which means that the problem being addressed is modified to render it amenable to analysis with the mathematical or statistical technique associated with the chosen theory. The point about degrees of freedom is that the fewer the degrees of freedom -- the smaller the number of ``free'' variables -- the more definite are the social and policy implications of the model.

An alternative approach is evidence-driven agent based social simulation, sometimes called companion modelling. The norm in this tradition is to engage with stakeholders and independent domain experts to identify behaviours of the relevant actors and their social context. Models designed and implemented within the constraints of this evidence are used to generate formal scenarios that are presented to the relevant stakeholders and used to inform the development of narrative scenarios for the exploration of alternative social policies or strategies. Such evidence-based modelling has been undertaken at a range of scales such as within river catchments in Senegal (Barreteau et al., 2001) and Thailand Becu et al. (2003), within a large enterprise (Moss, 1998), across southern England (including both rural areas and London) Downing et al. (2000).

Evidence-based social simulation modelling with stakeholder participation is a recent development without established social norms. Different practitioners use different programming languages and development environments, different representations of individual cognition and different modelling paradigms. There is no explicit guidance on such issues as the value of model simplicity. Generality has not been much of an issue precisely because the models are developed to describe and simulate specific, grounded social contexts and behaviour. Nor are these models used to forecast future outcomes but rather they are used to explore and discuss possible futures contingent on various policy or strategic actions.

The purpose of this paper is to explore what modelling norms would be functional and how these are different from the observed norms of conventional social modelling.