SIMULATION AS A TOOL TO IMPROVE PROCESS MODELS

Process models of decision making usually give very good results when it comes to ''explaining'' afterwards how a concrete decision process was constructed (retrodiction). The situation, however, is not as satisfactory if we look at predictions. This relative weakness is not an inherent property of process models. The problem is that if one makes all the necessary specifications the model becomes so complex that it is practically impossible to work out predictions over a wider range without a tool such as simulation, even if the model is formulated in some kind of formal language. My argument is that simulation may offer a solution to the problem. The representation of a process model by a working computer program enables, relatively easily, the computation of broad range predictions, and forces the scientist to formulate the model precisely. In this article I use the Chunking-by-Similarity Model as an example. Following the introduction to this model I first present some theoretical considerations concerning the simulation of a psychological theory. Then I discuss some important steps in the simulation of a theory in greater detail. I conclude that in spite of some problems which must be dealt with when a psychological process model is simulated, simulation is a powerful tool which can be used to improve the work with process models.