Page 70 -

5 Informal Approaches to Developing Simulation Models 65

• The author may have simply modelled without thinking about why (e.g. having
vague ideas about a phenomenon, the modeller decides to construct a model
without thinking about the questions one might want to answer about that
phenomenon).
• The model might have been developed for one purpose but is being presented as
if it had another purpose.
• The model may not achieve any particular purpose and so the author might be
forced into claiming a number of different purposes to justify the model.
The purpose of a model will affect how it is judged and hence should influence
how it is developed and checked.
The classic reason for modelling is to predict some unknown aspect of observed
phenomena—usually a future aspect. If you can make a model that does this for
unknown data (data not known to the modellers before they published the model),
then there can be no argument that such a model is (potentially) useful. Due to the
fact that predictive success is a very strong test of a model for which the purpose
is prediction, this frees one from an obligation as to the content or structure of the
1
model. In particular, the assumptions in the model can be quite severe—the model
can be extremely abstract as long as it actually does predict.
However, predictive power will not always be a measure of a model’s success.
There are many other purposes for modelling other than prediction. Epstein
(2008) lists 16 other purposes for building a model, e.g. explanation, training of
practitioners or education of the general public, and it is important to note that the
measure of success will vary depending on the purpose.
With an explanatory model, if one has demonstrated that a certain set of
assumptions can result in a set of outcomes (e.g. by exhibiting an acceptable fit to
some outcome data), this shows that the modelled process is a possible explanation
for those outcomes. Thus, the model generates an explanation, but only in terms of
the assumptions in the setup of the simulation. If these assumptions are severe ones,
i.e. the model is very far from the target phenomena, the explanation it suggests in
terms of the modelled process will not correspond to a real explanation in terms of
observed processes. The chosen assumptions in an explanatory model are crucial to
its purpose in contrast to the case of a predictive model—this is an example of how
the purpose of a model might greatly influence its construction.
It does sometimes occur that a model made for one purpose can be adapted for
another, but the results are often not of the best quality, and it almost always takes
more effort than one expects. In particular, using someone else’s model is usually
not very easy, especially if you are not able to ask the original programmer questions
about it and/or the code is not very well documented.

1
Of course a successfully predictive model raises the further question of why it is successful,
which may motivate the development of further explanatory models, since a complete scientific
understanding requires both prediction and explanation, but not necessarily from the same models
(Cartwright 1983).