Page 181 -

178 N. David et al.

purpose programming languages. This is not a black or white choice, since several
simulation toolkits offer the option of developing models using general-purpose
programming languages (e.g. Repast Simphony), and/or provide high-performance
and scalable workflows, with Repast HPC (Collier and North 2013) being a case in
point.
When the direct use of general-purpose programming languages is involved, the
adoption of good programming practices for designing and implementing the model
is fundamental. Techniques such as object-oriented design, modularity and encapsu-
lation not only simplify testing and debugging, but also promote incremental model
development and the mapping of programming units (e.g. classes or functions) to
model concepts, thus making computational models easier to understand, extend and
modify. Additionally, defensive programming methodologies, such as assertions and
unit tests, are well suited for the exploratory nature of simulation, making models
easier to debug and verify.
Two important verification methods, traces and structured walk-throughs, com-
plement the techniques discussed thus far. The former entails following a specific
model variable (e.g. the position of an agent or the value of a simulation output)
throughout the execution of the computational model, with the goal of assessing
whether the implemented logic is correct and if the necessary precision is obtained.
Modelling toolkits and programming language tools typically offer the relevant
functionality, making the use of traces relatively simple (Sargent 2013). In turn,
structured walk-throughs consist of having more than one person reading and
debugging a program. All members of the development team are given a copy of a
particular module to be debugged and the module developer goes through the code
but does not proceed from one statement to the next until everyone is convinced that
a statement is correct (Law 2015).
Nevertheless, and while the techniques described here are an important part of
the verification process, a computational model should only be qualified as verified
with reasonable confidence if it has been successfully replicated and/or aligned with
a valid pre-existing model. We will return to this topic in greater detail in Sect. 9.4.

9.2.2 What Does It Mean to Validate a Model?

Model validation is defined as ensuring that both conceptual and computational
models are adequate representations of the target. The term “adequate” in this sense
may stand for a number of epistemological perspectives. From a practical point of
view we could assess whether the outputs of the simulation are close enough to
empirical data.
Alternatively, we could assess various aspects of the simulation, such as if the
mechanisms specified in the simulation are well accepted by stakeholders involved
in a participative-based approach. In Sect. 9.3 we will describe the general idea of
validation as the process that assesses whether the pre-computational models—put