on using cellular automata for understanding social phenomena

queries on this paper:
Rainer Hegselmann and Andreas Flache (1998) Understanding Complex Social Dynamics: A Plea For Cellular Automata Based Modelling, Journal of Artificial Societies and Social Simulation vol. 1, no. 3, http://www.soc.surrey.ac.uk/JASSS/1/3/1.html

1. How to map social dimensions to a low-dimensional CA-based model?
CA-based models assume that individuals perform actions in a low-dimensional space. However, social processes usually take place in a high-dimensional space. For example, although the choice of partner can be affected by geographical distance, it is also related to friendship, similarity and disparity of individuals’ attributes, and other factors. The concept of social dimensions refers to an interesting phenomenon, small-world effect, which is known that individuals interact not only with people near by but also those far away. An explanation for such effect is as follows: the distance between two individuals may be long in terms of one social relation but short in another. If we are unclear about what kinds of social relations are important for the phenomenon under consideration, it would be difficult to choose some of the relations, or map multiple relations, to a low-dimensional CA-based model.

2. The strength of interaction is usually invisible in CA-based models
An individual may actively interact with one but less actively with another. The interactivity between two individuals may also vary over time. When we consider the structural changes in a society, the strength of interaction is usually critical. For example, we notice that a community is dispersing when the interactions among community members become sparse. However, it is difficult to represent the strength of interaction in CA-based models, since only the states of cells are recorded, not the interactions between cells.

3. Who are the key players?
Some individuals play important roles in certain social phenomena. For example, a famous person is much more likely to be referred to than ordinary people, which has been studied as so-called preferential attachment. It is also likely that a person is not the key player in the beginning but become influential in a dynamical social process. Though a CA-based model can initiate cells under different conditions according to the predetermined roles assigned to them, as in the support network model described in the paper, it might not be able to identify the key players who are unknown before the process, and to design local rules for cells to interact with the key players.

4. Consider multiple attributes of cells/agents
An individual may choose a partner to interact with not only because of some local rules, but also different attributes of the candidates, such as race, age, gender, education, vocation, interests, and so on. A CA-based model can categorize cells in terms of one or two attributes, but it is less feasible to model cells with more attributes and to design a variety of local rules as well. On the other hand, agent-based models allow agents with unlimited number of attributes, which can be considered as generalized CA-based models.

In summary, CA-based models take advantage of two important features: (1) discretisation of time, space and states, and (2) local rules for cells to change states. The discretisation in low-dimensional space makes CA-based models effective to produce recognizable patterns, which help deduce the underlying rules behind complex social phenomena. The second feature, local rules, based on the assumption that a macro-level effect emerges due to the interactions at micro-level, focuses on the interaction among individuals. This assumption is useful to simplify social process, while finding a macro-level factor from a complicated system is usually intractable. However, there are limitations within the CA framework, such as: (1) mapping from multi-dimensional social relations to a low-dimensional CA-model, (2) representation of the strength of interaction, (3) identification of key players, and (4) limited variation of individuals. In this regard, agent-based models can be considered as a more flexible representation for modeling and understanding micro/macro relationship.