Page 71 -

66 E. Norling et al.

Chapter 4 in this volume (Edmonds 2017) goes into five common modelling
purposes in more detail, with analyses of the particular risks for each kind of
purpose, and the basic steps to mitigate these risks.

5.3 Modelling Assumptions

Whilst the available evidence will directly inform some parts of a model design,
other parts will not be so well informed. In fact, it is common that a large part of a
complex simulation model is not supported by hard evidence. The second source
for design decisions is the conceptions of the modeller, which may have come
from ideas or structures that are available in the literature. However, this is still
not sufficient to get a model working. In order to get a simulation model to run
and produce results, it will be necessary to add in all sorts of other details: these
might include adding in a random process to “stand in” for an unknown decision
or environmental factor, or even be a straight “kludge” because you don’t know
how else to program something. Even when evidence supports a part of the design,
there will necessarily be some interpretation of this evidence. Thus, any model is
dependent upon a whole raft of assumptions of different kinds.
If a simulation depends on many assumptions that are not relatable to the object
or process it models, it is unlikely to be useful. However, just because a model has
some assumptions in it, this does not mean it should be disregarded. Any modelling
is necessarily a simplification of reality, done within some context or other. Hence,
there will be the assumption that the details left out are not crucial to the aspect of
the results deemed important, as well as those assumptions that are inherent in the
specification of the context. This is true for any kind of modelling, not just social
simulation. It is not sufficient to complain that a model has assumptions or does
simplify, since modelling without this is impossible; one has to argue or show why
the assumptions included will distort the results. Equally, the author of a model
should be able to justify the assumptions that have been made.
However, the use that is made of a simulation will be limited by the strength or
weakness of the assumptions taken as a whole. If, for example, the model is going
to be used in a policy process that will impact on many people’s lives, then a high
level of evidential support and validation will be required. If the model is more
exploratory—for example, to suggest unconsidered risks or new hypotheses—then
more assumptions with weaker evidence might well be acceptable. Chapter 29 in
this volume (Edmonds et al. 2017) looks at the dangers when models are used to
inform issues of policy importance.
What one can do is to try to make the assumptions as transparent, as clear and
as explicit as possible. Thus, future researchers will be better able to judge what
the model depends upon and adapt the model if any of the assumptions turns out to
be considered bad. The most obvious technique is to try to document and display
the assumptions. This not only helps to defend the model against criticism but also
helps one to think more clearly about the model.