Geoscience Reference
In-Depth Information
As the complexity of the model dynamics increases, the recognition and characterisation of col-
lective phenomena also becomes more complicated. The three different types of patterns in Figure 9.7
cannot be fully distinguished with one segregation index; recognition of their homogeneous and het-
erogeneous parts (if they exist) demands special algorithms (Hatna and Benenson, 2011).
The Schelling model is more than 40 years old. Intensive studies with this model, however,
have only been undertaken during the last decade. Before then, researchers followed the scheme
of Thomas Schelling's chessboard experiments: that is, equal numbers of agents of two groups,
equal demand of agents of both types for friends and always sufficient number of vacant cells for
relocation (Schelling, 1969, 1971). What is the importance of a modeller's decision regarding the
behavioural rules of the residential agents as posed in Section 9.3.2? What would be the model
dynamics if Schelling's population consisted of a minority and a majority? What would happen in
an asymmetric situation where members of one group demand fewer friends within the neighbour-
hood than the members of the other group? Most of these questions have been posed only recently
and Table 9.2 presents some of the answers. These answers confirm the general property of the AB
models -
light
variation in the model rules and parameters can result in essentially different model
dynamics. I remind the reader that they can reproduce these results and make their own investi-
gations using the online version of the Schelling model at http://erezhatna.com/SchellingModel/
Model.html.
Note that some of the model versions in Table 9.2 consider satisficer and maximiser agents that
reflect important differences between the views of human behaviour typical for economics and
behavioural science (Simon, 1969). It is impossible to discuss this difference in this review paper
and I have already recommended Kennedy (2012) and Gigerenzer et al. (2011) for further reading.
With respect to the Schelling model, for a satisficer agent, all locations for which f > F holds are
equally good, while a maximiser agent will always choose the location with the maximum fraction
of friends.
Several qualitative extensions of the Schelling model are also considered. Just to mention a few:
In the model of Portugali and Benenson (1997), the population consists of agents who belong to
several ethnic groups. In Bruch and Mare (2006), the agents prefer neighbourhoods with some pres-
ence of strangers compared to neighbourhoods consisting of friends only. Agents can react not only
to their immediate neighbourhood but also to the ethnic structure of the residential block or larger
area (Fossett and Waren, 2005; Fossett, 2006a,b). Residential agents can communicate between
themselves and swap locations instead of searching for a residence over the common pool of vacan-
cies (Zhang, 2004). Furthermore, the majority of residential decisions are made in situations where
the ethnic factors are marginal and the residential decisions of families are determined by economic
factors. The step from an ethnic to an economic-based residential behaviour requires an essential
alteration of the Schelling model (Benenson et al., 2009) and would take us far beyond the illustra-
tion of an AB modelling approach.
9.3.5 r
eal
-w
orld
S
chelling
-l
ike
S
iMulation
As discussed earlier, a properly constructed AB model should only be loosely connected to the
spatial representation of the non-animated objects. The Schelling model, in particular, has to be
developed independently from the partitioning of space into
residential locations
, and the latter can
be easily achieved as presented in Figure 9.6. An arbitrary partitioning of space enables us to apply
Schelling's ideas to a real-world area populated by members of two population groups for which
ethnic-like preferences can be important. Our model of the Jew-Arab segregation in Yaffo, which
is a mixed Tel Aviv neighbourhood with a population of about 30,000 (Figure 9.8), can serve as an
example of such development (Benenson et al., 2002) and as a proof of concept that AB models can
imitate the real-world residential dynamics. Recent attempts of applying the Schelling model to
real-world settings include Jordan et al. (2012) and Moreno et al. (2009).
