Page 188 -

9 Verifying and Validating Simulations 185

been inhered from simulation in engineering and computer science, particularly
from the reviews of validation and verification in engineering by Sargent (2013).
All techniques that we describe can be found in the literature, but it would be rare
to find a model in which only one technique was used, consistent with the fact that
the validation process should be diverse. Also, there are no standard names in the
literature and some techniques overlap with others.

9.3.3.1 Face Validity

Face validity is a general kind of test used both before and after the model is put
to use. During the model development process, the various intermediate models are
presented to persons who are knowledgeable about the problem in order to assess
whether they are compatible with the expert’s understanding and reasonable for their
purpose (Sargent 2013). Face validity may be used for evaluating the conceptual
model, the components thereof, and the behaviour of the computational models
in terms of categorical outcomes or direct input/output relationships. This can be
accomplished via documentation, graphing visualisation models, and animation of
the model as it moves through time. Visualisation of model outputs (including a
brief look at model animation) is analysed in Chap. 10 of this volume (Evans et al.
2017). Insofar as this is a general kind of test, it is used in several iterations of the
model.

9.3.3.2 Turing Tests

People who are knowledgeable about the behaviour of the target system are asked
if they can discriminate between system and model outputs (Sargent 2013;Law
2015). The logic of Turing tests is the following: If the outputs of a computational
model are qualitatively or quantitatively indistinguishable from the observation of
the target system, a substantial level of validation has been achieved.
Note that the behaviour of the target system does not need to be observed
directly in the cases where a computational direct representation is available. For
example, suppose that videos of car traffic are transformed into three-dimensional
scenes, whereby each object in the scene represents a car following the observed
trajectory. If an independent investigator is not able to distinguish the computational
reproduction from an agent-based simulation of car traffic, then a substantial level of
validation has been obtained for the set of behaviours represented in the simulation
model.

9.3.3.3 Historical Validity

Historical validity is a kind of retrodiction where the results of the model are
compared with the results of previously collected data. If only a portion of the