Page 22 -

2 Historical Introduction 15

vidual fate of all the members of this population (sample) with the help of transition
probabilities empirically estimated from official statistics. These transitions could
be transitions between different jobs and educational levels, or they could represent
death or the birth of a child or marriage; these models have mainly been used for
predicting demographic changes and the effects of tax and transfer rules. Usually,
these models do not take into account that the overall changes of the aggregated
variables of the population (or the sample) affect the individual behaviour. Thus in
the sense of Coleman (1990, p. 10), these models neglect the “downward causation”
(i.e. the influence of the aggregate on the individual) and focus only the “upward
causation”, namely, the changes on the macro level which are the result of the
(stochastically simulated) behaviour of the individuals.
The fluoridation referendum campaign model already mentioned above (Abelson
and Bernstein 1963) was one of the first models that can be classified as an
early predecessor of today’s agent-based models. It consisted of a large number of
representatives of people living in a community faced with the option of compulsory
fluoridation if drinking water—an issue often discussed in the 1960s—which they
would have to vote upon at the end of a longish campaign in which the media and
local politicians were publishing arguments in favour of or against this issue. In this
model, 500 individuals are exposed to information spread by several communication
channels (or sources), and additionally, they also exchange information among
themselves. It depends on their simulated communication habits to which extent
they actually receive this information and, moreover, to which extent this leads
to changes in their attitudes towards the referendum issue. Abelson and Bernstein
defined 51 rules of behaviour, 22 of which are concerned with the processing of
the information spread over the communication channels, and 27 rules are related
to the information exchange among the individuals; another 2 determine the final
voting behaviour at the end of the referendum campaign. The rules for processing
the information from the public channels and those for processing the information
exchanged among the individual citizens are quite similar, one of these rules—
A3 and B2, respectively—is, for instance, “Receptivity to [source] s is an inverse
function of the extremity of [individual] i’s attitude position”.
This early model did, of course, not endow the model individuals with an
appropriate repertoire of behaviours, but nevertheless it displays a relatively broad
range of communication possibilities among the model individuals which was
neither aimed at in the classical microanalytical simulation approach nor in the
cellular automata approach adopted in the early 1970s in Thomas Schelling’s
seminal paper on segregation. One of the shortcomings of Abelson’s and Bernstein’s
model in the eyes of its critics was the fact that it “has never been fully tested
empirically” (Alker 1974, p. 146), and another is the fact that one never knows “how
adequate are the static representations of citizen belief systems defined primarily
in terms of assertions held, assertions acceptance predispositions, with associated,
more general, conflict levels?” (Alker 1974, p. 146). And, moreover, the assertions
are modelled numerically (not a problem with the proponents of a mathematical
sociology who would even have used a large system of differential equations to
model the citizens’ attitude changes) where obviously real citizens’ attitudes were