Page 199 -

196 N. David et al.

societies for modelling possible worlds that represent classes of social mechanisms,
while striving for maximal simplicity and strong generalisation power of the
representations used. Reasons for striving for simplicity include the computational
tractability of the model and to keep the data analysis as simple as possible.
Simplicity and generalisation power are often seen as elements of elegance
in a model. However, making the model simpler in the social sciences does
not necessarily make the model more general. More often than not this kind of
modelling only makes it metaphorically general, or simply counterfactual (with false
assumptions). For example, “What would happen if world geography is regarded
as a two-dimensional space arranged on a 10 10 grid, where agents are thought
of as independent political units, such as nations, which have specific behaviours
of interaction according to simple rules?” To assume that world geography is
one-dimensional, as Axelrod (1993) does in his Tribute Model, is clearly a false
assumption. Often these models are associated with a design slogan coined by
Axelrod (1997a), called the KISS approach—“Keep it Simple Stupid.” Despite their
simplicity, these kinds of models prove useful for concept formation and theoretical
abstraction. The emergence of macro regularities from micro-levels of interaction
becomes the fundamental source of concept formation and hypothesis illustration,
with the power of suggesting novel theoretical debates.
Given the tendency for simplification and abstraction, mechanisms used in these
models are normally described in a formalised or mathematical way. Axelrod’s mod-
els, such as the culture dissemination model, or Schelling’s residential segregation
model, are canonical examples. Their simplicity and elegance have been factors for
popularity and dissemination that span numerous disciplines and ease replication
and verification.
However, whereas simplicity eases verification, the use of metaphorical models
also brings disadvantages. Consider a word composed of several attributes repre-
senting an agent’s culture, such as in Axelrod’s culture dissemination model. The
attributes do not have any specific meaning and are only distinguishable by their
relative position in the word. Thus, they can be interpreted according to a relatively
arbitrary number of situations or social contexts. However, such a representation
may also be considered too simplified to mean anything relevant for such a
complex concept as a cultural attribute. As a consequence, verification is hardly
distinguishable from validation, insofar as the model does not represent a specific
context of social reality. In such a sense, the researcher is essentially verifying
experimentally whether his conceptions are met by an operationalisation that is
intentionally and computationally expressed (David et al. 2005). Nevertheless,
given their simplicity, subjunctive models can be easily linked and compared
to other models, extended with additional mechanisms, as well as modified for
model alignment, docking, or replication. Cross-element validation is a widely used
technique.
At any rate, the fact that these models are simpler to replicate and compare—but
hardly falsifiable by empirically acquired characteristics of social reality—stresses
their strong characteristic: when models based on strategies of maximal simplicity