Page 136 -

7 Checking Simulations: Detecting and Avoiding Errors and Artefacts 133

still includes all the core assumptions) and obtain significantly different results.
When this occurs, we run the risk of interpreting the results obtained with the
(nonrepresentative) model beyond its scope (Edmonds and Hales 2005). Thus, to
the extent that identifying causal links in the model’s referent is part of the purpose
of the modelling exercise, the presence of artefacts decreases the validity of the
model. In any case, the presence of artefacts denotes a misunderstanding of what
assumptions are generating what results.

7.4.2 Appearance of Errors and Artefacts

The dynamics of agent-based models are generally sufficiently complex that model
developers themselves do not understand in exhaustive detail how the obtained
results have been produced. As a matter of fact, in most cases if the exact results
and the processes that generated them were known and fully understood in advance,
there would not be much point in running the model in the first place. Not knowing
exactly what to expect makes it impossible to tell whether any unanticipated results
derive exclusively from what the researcher believes are the core assumptions in
the model or whether they are due to errors or artefacts. The question is of crucial
importance since, unfortunately, the truth is that there are many things that can go
wrong in modelling.
Errors and artefacts may appear at various stages of the modelling process (Galán
and Izquierdo 2005). In this section we use the extended framework explained in the
previous section to identify the critical stages of the modelling process where errors
and artefacts are most likely to occur.
According to our definition of artefact—i.e. significant phenomena caused by
accessory assumptions that are not considered relevant—, artefacts cannot appear
in the process of abstraction conducted by the thematician, since this stage consists
precisely in distilling the core features of the target system. Thus, there should not
be accessory assumptions in the thematician’s model. Nevertheless, there could still
be issues with validation if, for instance, the thematician’s model is not capturing
social reality to a satisfactory extent. Errors could appear in this stage because the
thematician’s specifications are usually expressed in natural language, and rather
than being written down, they are often transmitted orally to the modeller. Thus, an
error (i.e. a mismatch between the thematician’s specifications and the non-formal
model received by the modeller) could appear here if the modeller misunderstands
some of the concepts put forward by the thematician.
The modeller is the role that may introduce the first artefacts in the modelling
process. When formalising the thematician’s model, the modeller will often have
to make a number of additional assumptions so the produced formal model is
fully specified. By our definition of the two roles, these additional assumptions
are not crucial features of the target system. If such accessory assumptions have
a significant impact on the behaviour of the model and the modeller is not aware
of it, then an artefact has been created. This would occur if, for instance, (a) the