Methodological Problems of Computer Simulation in the Social Sciences

Abstract

Modelling — which includes computer simulation — is a scientific activity which has the aim of creating a replication of some real system such that this replication will react to inputs in a manner that resembles the reaction of the real system to the same inputs. The first step in this modelling activity is thus the identification of some part of reality as a ‘real system’ consisting of elements, of relations defined on these elements, and of relations defined on the elements of the system and its environment (Bunge,1979). We believe that the properties of the system, of its elements, and of its environment will change due to some deterministic or stochastic laws, and we make our (mathematical or computer simulation) model follow the same laws which we believe will control that part of reality which is to be modelled. Thus the second step of any modelling activity will be the detection — or rather the reconstruction — of the laws governing that part of reality we are about to model. Kreutzer (1986,2) calls this step ‘system representation’. In a third step we shall try to combine our notions of the laws governing reality into a model which may be real by itself (as is the case in animal experiments used to detect primary and secondary effects of drugs), iconic, verbal, or formal, i.e. written down in a formal language like the language of mathematics or like a computer programming language. In each of the four cases we try to draw our inferences about model behaviour to be expected from the premises incorporated into the model. In the cases of the informal models — real, iconic, verbal — these inferences are weak because real, iconic, and verbal models abound in (‘natural’) properties of their own that are not shared by the part of reality to be modelled (the ‘abundancy class’ of Stachowiak (1973, 155 – 157), whereas in the case of formal models — mathematical and computer simulation — the inferences are just as certain as our assumptions about the laws governing reality are completely incorporated into our model. In most instances, of course, they are not completely incorporated, there are rather a lot of properties of the real system that we do not know or that we cannot measure or that we cannot or do not want to formalize (the ‘preterition class’ of Stachowiak).