Model construction in multilevel-multivariate analysis

Conclusion

The simulation models discussed here are all of the type that is relevant to the main problem of this paper. They all start with the behavior of individuals in a social system and the behavior of the system is derived from the behavior of the individuals determined by the situation in the system in conjunction with their individual attributes. By combining the individual behaviors in suitable ways the behavior of the system is arrived at.

It is not claimed that these examples exhaust the possibilities or even that they are very good examples. What they do show is that it is possible and even easily possible with simulation techniques to build up a social system from individual behavior with quite complicated assumptions about the structure of the system.

This is a clear advantage of simulation models in comparison with ordinary mathematical or other formal models. The capacity of existing mathematics to deal with the kind of structures encountered in social systems is limited. Mathematical models require simplifications that make them unrealistic. On the other hand, mathematical models often allow explicit solutions and exact determination of other characteristics which cannot be achieved by simulation. But simulation models have in common with formal theories the exact formulation of attributes and relations. For simulation this follows from the requirements of the machines.

The difficulties in using simulation for theory testing which were outlined above and which were evident from the examples discussed have perhaps been instrumental in delaying the use of simulation in sociology. It is my impression that we do not use simulation by far as much as should be done. If we cannot test theories in an exact sense we should use simulation to force researchers to formulate their theories precisely enough to allow simulation, to test the theories for consistency and reasonableness, and to show the kind of data that are necessary for a better testing of the theories.

Personally, I believe that in the long run the construction of synthetic theories will be the most important use of simulation. This presupposes that we have reliable and valid relations between the elements to start with which are not available in general at the present time. However, attempts to make synthetic theories would show what kinds of relations are needed and thus presumably accelerate the development of the science in the desirable direction.