Inferring Adolescent Social Networks Using Partial Ego-Network Substance Use Data

摘要

This dissertation explores the social network processes involved in adolescent substance use. Over the past three decades, researchers have focused on, with increasing clarity, the specific dynamics of peer selection and peer influence in their attempts to understand how adolescents first use a substance, what compels them to continue use, and why some of them quit. However, the exact nature of interplay between those two key social processes continues to be elusive, due to the lack of both robust longitudinal network data and sophisticated network methodologies capable of addressing such data; it is only in recent years that advances in the field have improved these deficiencies. The research presented here adopts an alternative approach using a large cross- sectional data set that is not without its limitations, but still manages to produce specific parameters for selection and influence some of which are surprisingly similar to those reported in some recent work on this topic. Inferences to describe adolescent networks are drawn from partially-formed ego network data contained in the 1998 and 1999 survey years of the National Survey on Drug Use and Health; a modest level of precision in these analyses is achievable thanks to the large sample size. A custom Poisson/binomial/multinomial mixture is employed to extract precise peer network properties from ordinal response data having categories of proportions which implicitly cover the (0,1) interval.