Agent-Based Models of Culture: Growing Group Patterns From Individuals’ Motives

One of the perennial problems faced by cross-cultural psychologists is the link between individuals’ motives and group behaviours. Agent-based modelling is a technique that allows to “grow” a social system to investigate this relationship between individual level and social system level. Each individual is typically modelled as an agent, and groups of agents create emergent behavioural patterns at system level. In this way, an agent-based model can function as a hypothesis about causal links between empirically obtained agent data and observed system regularities. One can manipulate input parameters, or run the same model with different agent populations, and thus obtain data about the model’s behaviour under different circumstances. This flexibility, lacking in experimental research, can advance cross-cultural psychology. The paper discusses how agent-based models work. These models require capacities for the agents such as perception, interpretation and action. They also require mechanisms that make the agents coordinate. The paper reviews the mechanisms used by agent-based models regarding two issues that are important in group coordination: affiliation and hierarchy. These issues are also crucial to culture: in-group versus out-group and leader versus follower are central role distinctions in social life, and differently shaped across cultures. A more in-depth discussion of a single agent-based model of the emergence of gender differences in groups of playing children closes the article. This model shows how existing social scientific theory can be used for the design of agent mechanisms. It uses the status-power theory of Kemper for agent motivation, and the dimensions of culture by Hofstede to model differences in the social system surrounding the agents. Empirical studies on gender and child play are used for details of motivation. The article is also used to point to some methodological points inherent in working with agent-based models, notably, issues of validation.