Page 128 -

7 Checking Simulations: Detecting and Avoiding Errors and Artefacts 125

To summarise, a computer program is a formal model (which can therefore be
expressed in mathematical language, e.g. as a set of stochastic or deterministic
equations), and computer simulation is a tool that enables us to study it in ways
that go beyond mathematical tractability. Thus, the final result is a potentially more
realistic—and still formal—study of a social system.

7.3 Agent-Based Modelling

7.3.1 Concept

As stated before, modelling is the process of building an abstraction of a system for
a specific purpose—see Chap. 4 in this volume (Edmonds 2017; Epstein 2008)for a
list of potential modelling goals. Thus, in essence, what distinguishes one modelling
paradigm from another is precisely the way we construct that abstraction from the
observed system.
In our view, agent-based modelling is a modelling paradigm with the defining
characteristic that entities within the target system to be modelled—and the
interactions between them—are explicitly and individually represented in the model
(see Fig. 7.2). This is in contrast to other models where some entities are represented
via average properties or via single representative agents. In many other models,
entities are not represented at all, and it is only processes that are studied (e.g. a
model of temperature variation as a function of pressure), and it is worth noting
2
that such processes may well be already abstractions of the system. The specific
process of abstraction employed to build one particular model does not necessarily
make it better or worse, only more or less useful for one purpose or another.
The specific way in which the process of abstraction is conducted in agent-based
modelling is attractive for various reasons: it leads to (potentially) formal yet more
natural and transparent descriptions of the target system, provides the possibility
to model heterogeneity almost by definition, facilitates an explicit representation
of the environment and the way other entities interact with it, and allows for the
study of the bidirectional relations between individuals and groups, and it can
also capture emergent behaviour (see Epstein 1999; Axtell 2000; Bonabeau 2002).
Unfortunately, as one would expect, all these benefits often come at a price: most of
the models built in this way are mathematically intractable. A common approach to
study the behaviour of mathematically intractable formal models is to use computer
simulation. It is for this reason that we often find the terms “agent-based modelling”
and “agent-based simulation” used as synonyms in the scientific literature (Hare and
Deadman 2004).

2
The reader can see an interesting comparative analysis between agent-based and equation-based
modelling in Parunak et al. (1998).