Using an Agent-Based Model to Simulate the Development of Risk Behaviors During Adolescence

Adolescents tend to adopt behaviors that are similar to those of their friends, and also tend to become friends with peers that have similar interests and behaviors. This tendency towards homogeneity applies not only to conventional behaviors such as working for school and participating in sports activities, but also to risk behaviors such as drug use, oppositional behavior or unsafe sex. The current study aims at building an agent model to answer the following related questions: How do friendship groups evolve and what is the role of behavioral similarity in friendship formation? How does homogeneity among peers emerge, with regard to conventional as well as risk behaviors? On the basis of the theoretical and empirical literature on friendship selection and influences on risk behavior during adolescence we first developed a conceptual framework, which was then translated into a mathematical model of a dynamic system and implemented as an agent-based computer simulation consisting of simple behavioral rules and principles. Each agent in the model holds distinct property matrices including an individual behavioral profile with a list of risky (i.e., alcohol use, aggressiveness, soft drugs) and conventional behaviors (i.e., school attendance, sports, work). The computer model simulates the development, during one school year, of a social network (i.e. formation of friendships and cliques), the (dyadic) interactions between pupils and their behavioral profiles. During the course of simulation, the agents' behavioral profiles change on the basis of their interactions resulting in individual developmental curves of conventional and risk behaviors. These profiles are used to calculate the (behavioral) similarity and differences between the various agents. Generally, the model output is analyzed by means of visual inspection (i.e., plotting developmental curves of behavior and social networks), systematic comparison and by calculating additional measures (i.e., using specific social analysis software packages). Simulation results conclusively indicate model validity. The model simulates qualitative properties currently found in research on adolescent development, namely the role of homophily, the appearance of friendship clusters, and the increase in behavioral homogeneity among friends. The model not only converges with empirical findings, but furthermore helps to explain social psychological phenomena (e.g. the emergence of homophily among adolescents).