System Dynamics Applications / 2 / 2.4

2.4 MODEL AS A NECESSARY AND SUFFICIENT REPRESENTATION

There will always be temptation to build elegant models containing large numbers of variables and information links, very closely representing the real-world situation. Unfortunately, building large and elegant models brings significant risks. There are risks of errors in design and structure, and coding errors. Large models can be exceedingly difficult to debug and almost impossible to test comprehensively. If these models contain structural or logical flaws they will not replicate real-world behaviour. It can be exceedingly difficult to identify why the model does not function as intended, and to make the necessary corrections. The likelihood of errors increases exponentially with the size of the model.

At the other end of the scale, overly simplistic models can be made error free but we may learn little from them, and they do not represent the real world. The challenge here is to build models that are necessary to inform our understanding and sufficient representations of the real world to be realistic.

The Occam’s Razor principle is a useful guide to prevent us from building unnecessarily complicated models. Williams (2002:43) explains that this principle expresses the idea that if a few entities or reasons are sufficient to explain a phenomenon, this is a preferable explanation to one using many entities or reasons—application of Occam’s Razor to our analysis and modelling will lead us to using the simplest models appropriate to our task.

Of course, this has to be balanced by the need not to build overly simplistic models. The model must have sufficient requisite variety, Ashby (1956: 202-218): there is a finite (minimum) limit to the extent to which we might simplify a problem. This has a number of implications. For example, if any model we might build does not represent sufficient features of the problem situation being examined, it will not permit the properties of the system under study to emerge. Models must contain sufficient requisite variety in their elements to make it possible and, indeed, highly likely to discover the causes of dynamic real-world behaviour and emergent properties produced through correct combining of the structural elements of the systemic problem.