CitationMa, Hannan & Chen, Chien-fei (2012). Bayesian network structure learning for analyzing the impact of social networks on adolescent heavy drinking. 2012 Add Health Users Conference. Bethesda, MD.
AbstractWe use data from Wave1 and Wave2 of the National Longitudinal Study of Adolescent Health to study the problem of alcohol use and negative consequences among adolescents using Bayesian network and statistical inference method. We constructed a Bayesian network to study both the factors associated with underage alcohol use, joint occurrence of drinking and risk behaviors, as well as the friendship influence on individual’s drinking behaviors via social network based on the friendship nomination provided in the in-school survey from Wave1. Finally, concurrent negative consequences of alcohol use and the longitudinal effects over time were shown by inference over the structure. Bayesian network is a probabilistic graphical model with connected nodes and directed edges. It models the causalities among nodes by the interconnected structure, and numerically quantifies the conditional probability of factors using statistical inference. We model each selected survey question as a factor and represent it as a node in Bayesian Network. Then the directed edges showing the cause-effect relationships are determined by Bayesian Structure Learning (BSL) algorithm. The structure is learned from training data (which is the selected no-missing data in Add Health) and is tested for high inference accuracy. The BSL algorithm proposed in this paper is a novel algorithm proved to have better performance than previous structure learning algorithms.
Reference TypeConference proceeding
Book Title2012 Add Health Users Conference