Geoscience Reference
In-Depth Information
Computer simulation has known a growing success among geographers,
archaeologists, economists and sociologists for nearly two decades [BAT 05,
KOH 00, EPS 96, GIL 99, BEN 04, PHA 07, HEP 12]. Although the success of this
form of modeling is increasing, it is more relevant to consider these two approaches
(statistics and computer simulation) as complementary rather than competitive. A
somewhat systematic analysis of their differences and respective advantages shows
this well.
4.1.1.
From covariation to interaction, from differentiation to emergence
The statistical approach is centered on highlighting similarities between objects
(as statistical individuals) according to their characteristics and the
interrelations
existing among their different attributes. The agent approach
1
, for its part, focuses on
the
interactions
; interactions among agents, on the one hand, and between agents
and environment, on the other hand (in the example of Schelling's model, the
environment is simply composed of the organization of the other agents in the
immediate vicinity). Thus, the key concepts are clearly distinct in the two
approaches. In the first case, the issue can be expressed in terms of differentiation
and covariation, and in the second case in terms of interaction and emergence.
These differences are visible in the symbolic representation of these methods: a
data table comparing statistical individuals with variables in the statistics case
(Figure 4.3(a)); following a social metaphor in the case of MAS (Figure 4.3(b)),
with agents located inside an environment, having a representation of this
environment and acting on it according to their interactions with the other agents
[FER 95]. Figure 4.3 illustrates this difference, where the interest lies in the change
of the social composition of the high schools of a city or a region. These high
schools can be considered as the “individuals” in a statistical approach and as the
agents in a MAS. The attributes (public/private status, number of students, number
of incoming students, number of outgoing students, etc.), contained columnwise in
the statistical table, are strongly associated to the individuals. Considered as
properties in the agent approach, they are able to influence the behavior of agents,
but they do not appear as central in this formalization and, therefore, are not
explicitly represented in the classic diagram proposed by Ferber [FER 95]. This
representation highlights the interaction between agents, in this example, by the flow
of pupils changing school.
1 In this comparison, we will consider especially, with regard to computer simulation, multi-
agent systems (MAS) that we qualify sometimes, in a more general manner, as agent
approach, and cellular automata (CA). More precise definitions are given further in the text.
For a very thorough presentation of this field, we refer to [PHA 07].
