Page 300 -

13 Combining Mathematical and Simulation Approaches to Understand... 299

Fig. 13.1 A computer model can be usefully seen as the implementation of a function that
transforms any given input into a certain probability distribution over the set of possible outputs

variable, and conveniently, if no seed is explicitly provided to the pseudorandom
number generator, most platforms generate a seed from the state of the computer
system (e.g. using the time). When this is done, the sequences of numbers obtained
with readily available pseudorandom number generators approximate statistical
randomness and independence remarkably well.
Given that—for most intents and purposes in this discipline—we can safely
assume that pseudorandom numbers are random and independent enough, we
dispense with the qualifier ‘pseudo’ from now on for convenience. Since every
random variable in the model follows a specific probability distribution, the
computer model will indeed generate a particular probability distribution over the
range of possible outputs. Thus, to summarise, a computer model can be usefully
seen as the implementation of a stochastic process, i.e. a function that transforms
any given input into a certain probability distribution over the set of possible outputs
(Fig. 13.1).
Seeing that we can satisfactorily simulate random variables, note that studying
the behaviour of a model that has been parameterised stochastically does not
introduce any conceptual difficulties. In other words, we can study the behaviour
of a model that has been parameterised with probability distributions rather than
certain values. An example would be a model where agents start at a random initial
location.
To conclude this section, let us emphasise an important corollary of the previous
paragraphs: any statistic that we extract from a parameterised computer model
follows a specific probability distribution (even if the values of the input parameters
6
have been expressed as probability distributions). Thus, a computer model can be
seen as the implementation of a function that transforms probability distributions

6
Note that statistics extracted from the model can be of any nature, as long as they are
unambiguously defined. For example, they can refer to various time-steps and only to certain agents
(e.g. ‘average wealth of female agents in odd time-steps from 1 to 99’).