Page 193 -

190 N. David et al.

In the remainder of this section we discuss replication and comparison of
simulation models under three different perspectives. First, in Sect. 9.4.1,we
distinguish the terminology and origins of the different goals related to model
replication and comparison. Next, in Sect. 9.4.2, we go over the best practices in
developing models so that they may be replicated by other researchers in the future.
Finally, in Sect. 9.4.3 we discuss a number of model comparison techniques.

9.4.1 Model Replication, Model Alignment or Submodel
Comparison?

A model replication study commonly assesses the extent to which building com-
putational models that draw on the same conceptual, usually published, model give
results compatible with the ones reported for the latter. If the new results are similar
to the published results, then the confidence in the correspondence between the
computational and the conceptual models is increased. Replication is represented
in Fig. 9.3.
The work of Edmonds and Hales (2003) is particularly informative and worthy
of reference. Edmonds and Hales performed two independent replications of a
previously published model involving co-operation between self-interested agents.
Several shortcomings were found in the original model, leading the authors to
conclude that unreplicated simulation models and their results cannot be trusted.
The issue was found to be a subtle difference in one of the submodels, which lead
to different conclusions about the functioning of the overall model.
The term model alignment is frequently used as a synonym for model replication.
However, its meaning is somewhat more subtle, as it is more related with the
extent to which models can be coupled or docked so that their consequences and
results are consistent with each other. In its most general form, this concerns both
to V&V. After Axtell et al. (1996), the term became associated with the process
of determining whether different published models describing the same class of
social phenomena produce the same results. Usually the alignment or docking of

Computational
model A’

Published results compare
conceptual model A results
and results

Computational
model A”

Fig. 9.3 Model replication