Modeling Deviance using Empirical Ego-Networks

Introduction The role of social networks in the proliferation of deviant behavior, including commissions of individual acts of crime, remains a largely unexplored area of research. The study of social networks as a specialty in sociology is nascent and has generally focused on relatively easily observable and measurable networks, such as those in the workplace or small networks of friends. The import of the social environment in the formation of any attitudes or behaviors, including those labeled as deviant, is self-evident. For instance, numerous studies have shown that the social environment, in the form of the immediate family during infancy and early childhood, and the community at large through the later years are strong determinants of criminal behavior. A complication to this simplistic portrait is introduced when the individual becomes purposive in selecting the members of his or her network. The evolution of one’s social network is ultimately a feedback process in which outcomes can become contingencies: a classic chicken-and-egg problem. Furthermore, the evolution of social networks involving stigmatic behavior often differs from that of networks of common relationships such as friendship (Erickson, 1981)(Lee, 2000). This work focuses on the role of social networks in the proliferation of illicit drug use. Networks of drug use don’t necessarily follow well-behaved patterns of diffusion. Their structures can be contingent on the stage of the diffusion process and evolve in erratic ways resulting in varying response patterns to intervention policies (Behrens et al, 1999). Archival data of largely static measures drug behavior and social networks are used to establish a pattern of correlation between the composition of one’s network and one’s risk of engaging in the use of illegal drugs such as marijuana or cocaine. This task, whether conclusive or not, is the first step in constructing a model best informed by the data. The second step is to construct virtual agents reflecting the real individuals profiled in the data archives and to allow these agents to interact in an idealized virtual world, idealized because 1) a model mimicking the real world behavior is an unrealistic task, and 2) the current data does not allow us to construct such an emulative model. Theoretical Basis Individuals’ social networks neither develop nor exist in isolation. These ego-nets generally overlap with those within social and geographic proximity. Particular patterns in the way these social circles mesh are inevitable; homophily is ubiquitous, and thus, we should find many homogenous groups. Alternatively, the means through which social circles extend beyond their majority group is a persistent research question, and especially relevant to the diffusion of an illicit good like drugs. The idea of crosscutting social circles (Blau, 1984) refers to these links that cross the boundaries, which differentiate allegedly discrete social groups. Looking at the evolution and maintenance of a behavior within a single group, however interesting and important, will not tell us about how the behavior manifests in a separate group. The unit of analysis must correspond to the distribution of behavior; while this is trivially obvious, the selection of the proper unit of analysis is not a trivial task. Current Data Sources 1985 NHSDA (National Household Survey on Drug Abuse) (NHSDA, 1985) Provides basic predictors/profiles of illicit drug users and is the most comprehensive survey on drug use behavior. 1985 GSS (General Social Survey, 1985, includes Social Networks Module) (GSS, 1985) Provides network structure of friendship and confidant networks of various individuals. Auxiliary information includes basic demographics, such as sex, age, race, and religion. Nature of Data The data come in several different types; these categories determine how they are used and incorporated into the model. Furthermore, these categories are not specific to a single data archive; many of these span different surveys allowing for multiple imputation: Behavioral: Current or recent drug use. Demographic: Sex, age, race, education, marital status, number of children. Social network: 1985 GSS Confidant and Friendship Networks Attitudinal: Happiness, Propensity to obey the law. Linking the Datasets The lack of complete data in a single archive necessitates the inference to span multiple data archives. A formal strategy is ‘multiple imputation’ in which missing data items are inferred from distributions of variables, which are assumed to be correlates. Specifically, we use the demographic and attitudinal items in both the GSS and NHSDA to impute network measures for those individuals in the NHSDA. Subsequently, the analysis is performed only for the NSHDA respondents for whom network measures have been imputed.