Simulation and Reality

6 Conclusion

All of the models described here represented agents or their behaviour in ways which were validated independently of the models. The behaviour of consumers in the intelligent market modelling system conforms to the independent views of marketing practitioners. The representations of cognition in the critical-incident and transition-economy models conformed to important, common aspects of the main software architectures corresponding to experimentally verified theories of cognition. In all cases, the models captured empirically observed characteristics of the agents and their environments. The differences in the verification of each model was that we had exceptionally good statistical data with which to verify the market models by means of standard statistical measures whereas the other two models relied on stylized facts and statistical signatures which entail no clear formal criteria. The reasons were different in each case. In the critical-incidents model, the condensation of time gave us more useful results even though, with some expense, it would have been possible to obtain data about the actual duration of critical incidents. This would be a good case in which to develop techniques for mapping numerical social simulation output into actual statistical data. It has not, as far as I know, been attempted in the social simulation domain. In the case of the arrears model, there is no reliable statistical evidence so that all we have to rely on are stylized facts or statistical signatures. Validation of the model by appeal to domain experts and perhaps survey evidence concerning sources of information and the extent to which enterprise managers rely on different sources of information will doubtless help to validate our representations and verify qualitative model outputs.

Clearly the models discussed in this paper are distinguished by their role as representations of actual systems: markets for given sets of competing brands, a set of functions of a particular company, specified empirical problems facing a much-studied if little-understood economy. There a many models in this vein, though not all are concerned with management and economic issues. Anthropological simulations by Doran and his colleagues or models of emergent behaviour undertaken by Cress at Surrey, or the simulations of the emergence of altruistic behaviour by Pedone and Parisi are examples of relatively abstract social simulation models concerned with empirical phenomena not directly related to either economics or management issues.

Social simulation frequently entails emulation of real phenomena. The point of this paper has been partly to suggest some means of making the links between the models and their empirical referents clear and, where possible, measurable; but more importantly to put a case for the importance of relating social simulation models explicitly to the empirical phenomena. If this case were accepted, then the development of procedures for making those links in ways that avoid mere handwaving and unsupported assertions of relevance is by implication an important thread in the continuing development of the social simulation literature.

Simulation and Reality - 20 MAY 98

[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker