Page 126 -

7 Checking Simulations: Detecting and Avoiding Errors and Artefacts 123

Simplicity

Generality

Specificity (makes
precise predictions)

Lack of error
(accuracy of results)
Fig. 7.1 The trade-off between various desirable features depends on the specific case and model.
There are not general rules that relate, not even in a qualitative fashion, all these features. The
figure shows a particular example from Edmonds (2005) that represents the possible equilibrium
relationships between some features in a particular model

sensitivity to assumptions that are allowed to change within the boundaries of the
model (i.e. parameter values and nonstructural assumptions).
However, the process of reducing social reality to formal models is not exempt
from disadvantages. Social systems can be tremendously complex, so if such
systems are to be abstracted using a formal language (e.g. mathematical equations),
we run the risk of losing too much in descriptiveness. To make things worse, in
those cases where it appears possible to produce a satisfactory formal model of
the social system under investigation, the resulting equations may be so complex
that the formal model becomes mathematically intractable, thus failing to provide
most of the benefits that motivated the process of formalisation in the first place.
This is particularly relevant in the domain of the social sciences, where the systems
under investigation often include non-linear relations (Axtell 2000). The usual
approach then is to keep on adding simplifying hypotheses to the model—thus
making it increasingly restrictive and unrealistic—until we obtain a tractable model
that can be formally analysed with the available tools. We can find many examples
of such assumptions in economics: instrumental rationality, perfect information,
representative agents, etc. Most often these concepts are not included because
economists think that the real world works in this way, but to make the models
tractable (see, for instance, Conlisk 1996;Axelrod 1997a; Hernández 2004;Moss
2001, 2002). It seems that, in many cases, the use of formal symbolic systems
tends to increase the danger of letting the pursuit for tractability be the driver of
the modelling process.