Page 27 -

20 K.G. Troitzsch

Table 2.1 Overview of important approaches to computational social science
Approach Used since Characteristics
System dynamics Mid-1950s Only one object with a large number of attributes
Microsimulation Mid-1950s A large number of objects representing individuals
that do not interact, neither with each other nor
with their aggregate, with a small number of
attributes each, plus one aggregating object
Cellular automata Mid-1960s Large number of objects representing individuals
that interact with their neighbours, with a very
restricted behaviour rule, no aggregating object,
thus emergent phenomena have to be visualised
Agent-based models Early 1990s with Any number of objects (“agents”) representing
some forerunners individuals and other entities (groups, different
in the 1960s, kinds of individuals in different roles) that interact
afterwards heavily with each other, with an increasingly rich
discontinued repertoire of changeable behaviour rules
(including the ability to learn from other, to
change their behavioural rules and to react
differently to identical stimuli when the situation
in which they are received are different

with static equilibria—and claim “that the methodology developed [in Sugarscape]
can help to overcome these problems” (Epstein and Axtell 1996,p.2).
To complete this overview, Table 2.1 lists the approaches touched in this
introductory paper with their main features.
As one easily sees from this table, only the agent-based approach can “cover
all the world” (Brassel et al. 1997), as only this one can include the features of
all the others, and only this one can meet the needs of social science, as social
science cannot content itself with models of individuals which cannot exchange
symbolic messages that have to be interpreted by the recipients before they can
take effect. If social science deals with large numbers of individuals in comparable
situations, then microsimulation, cellular automata, sociophysics models and even
systems dynamics can be a good approximation to what happens in human societies.
But if we deal with small communities, including the local communities Abelson
and Bernstein analysed, then the process of persuasion—which needs at least one
persuasive person and one or more persuadable persons—has to be taken into
account, and this calls for a richer structure of agents than the early approaches
could provide.
Most of the literature suggested for further reading has already been mentioned.
Epstein’s and Axtells’s (1996) work on generating societies gives a broad overview
of early applications of agent-based modelling; Epstein (2006) goes even further
as he defines this approach as the oncoming paradigm in social science. For the
state of the art of agent-based modelling in the social sciences at the onset of this
approach, the proceedings of early workshops and conferences on computational
social science are still worth reading (Gilbert and Doran 1994; Gilbert and Conte
1995; Conte et al. 1997; Troitzsch et al. 1996). And a very wide overview of topics