Page 134 -

7 Checking Simulations: Detecting and Avoiding Errors and Artefacts 131

the formal model is a legitimate particularisation of the non-formal model created
by the thematician, the conclusions obtained for the modeller’s formal model can
be interpreted in the terms used by the non-formal model. Furthermore, if the
modeller’s formal model is representative of the thematician’s model, then there is
scope for making general statements on the behaviour of the thematician’s model.
Finally, if the thematician’s model is satisfactorily capturing social reality, then the
knowledge inferred in the whole process can be meaningfully applied to the target
system.
In the following section, we use our extended framework to identify the different
errors and artefacts that may occur in each of the stages of the modelling process
and the activities that can be conducted to avoid such errors and artefacts.

7.4 Errors and Artefacts

7.4.1 Definition of Error and Artefact and Their Relevance for
Validation and Verification

Since the meanings of the terms validation, verification, error, and artefact are not
uncontested in the literature, we start by stating the meaning that we attribute to
each of them. For us, validation is the process of assessing how useful a model is
for a certain purpose. A model is valid to the extent that it provides a satisfactory
range of accuracy consistent with the intended application of the model (Kleijnen
6
1995; Sargent 2003). Thus, if the objective is to accurately represent social reality,
then validation is about assessing how well the model is capturing the essence of
its empirical referent. This could be measured in terms of goodness of fit to the
characteristics of the model’s referent (Moss et al. 1997).
Verification—sometimes called “internal validation”, e.g. by Taylor (1983),
Drogoul et al. (2003), Sansores and Pavón (2005), or “internal validity”, e.g. by
Axelrod (1997a)—is the process of ensuring that the model performs in the manner
intended by its designers and implementers (Moss et al. 1997). Let us say that
a model is correct if and only if it would pass a verification exercise. Using our
previous terminology, an expression of a model in a language is correct if and only
if it is the same model as the developer’s specifications. Thus, it could well be the
case that a correct model is not valid (for a certain purpose). Conversely, it is also
possible that a model that is not correct is actually valid for some purposes. Having
said that, one would think that the chances of a model being valid are higher if
it performs in the manner intended by its designer. To be sure, according to our
definition of validation, what we want is a valid model, and we are interested in its
correctness only to the extent that correctness contributes to make the model valid.

6 See a complete epistemic review of the validation problem in Kleindorfer et al. (1998).