Early adolescent alcohol use is a major public health challenge. One quarter of all adolescents begin drinking alcohol by 13
years of age [1
]. Drinking before the 14th birthday is associated with a fourfold increase in risk of alcohol dependence in adulthood [2
]. Early alcohol initiation is linked to many risky adolescent behaviors, including marijuana and cocaine use, having sex with multiple partners, and academic underperformance [3
A wide body of literature indicates that adolescents and their friends exhibit more similar alcohol use behavior than would be expected by chance alone [4
]. Drinking by a best friend has been tied to alcohol initiation among middle and high school students [6
]. In a systematic review of longitudinal studies on adolescent drinking, alcohol-using peers were consistently predictive of an adolescent’s own drinking behavior at a later wave [8
]. However, there is debate over the mechanism by which friends come to resemble one another over time. One possible explanation is that similarities occur as a result of peer influence, or the spread of behaviors and behavioral norms through social ties. In this manner, the behavior of an individual would move toward the average behavior of one’s friends over time. Another pathway to homogeneity within friendships is that friends may be similar due to social selection, or homophily, the tendency for similar people to be attracted to and form friendships among one another.
Understanding the mechanism by which similarity in alcohol use behavior among adolescent friends occurs is an important clinical matter. Lacking clear direction on the causal pathways between peer interactions and alcohol use, adolescent brief alcohol interventions have produced, at best, mixed results [9
]. As an example, influence-driven contagion in adolescent groups would lend itself to peer-to-peer methods of alcohol use intervention. It would also allow for the possibility of multiplier or spillover effects from targeted individuals to a larger network of friends. Selection-driven behavior patterns in adolescent groups would lend themselves to interventions that target alcohol use in adolescent friendship groups identified through social network characteristics.
Previous studies have attempted to disentangle selection from influence effects in adolescent alcohol use [8
]. Analytic techniques employed have included structural equation modeling [11
], latent growth models [14
], instrumental variables [19
], and fixed effects [20
]. These studies have relied on lagged indicators of alcohol use and friendship connections in an attempt to isolate the selection effect from influence. The results indicate that both selection and influence are occurring, but the relative contribution of the two factors cannot be determined.
A major limitation in prior studies on selection and influence effects in adolescent alcohol use is the failure to account for the co-evolution of network ties and alcohol use behavior. Assuming network ties are fixed while estimating changes in drinking produces biased parameter estimates. In a similar manner, modeling alcohol use as constant when estimating friendship tie formation can lead to systematic error in results. In addition, previous studies are limited in their control for social network elements beyond the unidirectional dyadic relationship which could drive friendship formation and behavior change. These factors include reciprocity (i.e., the likelihood to reply to friendship with friendship) and transitive closure (i.e., the likelihood of friends of friends to become friends). Furthermore, previous modeling strategies fall short in accounting for the dependent nature of social network ties data. Peer interactions that encompass interdependent selection and influence effects violate the notion of independence required by traditional modeling techniques such as structural equation or fixed effects modeling. The complexity of longitudinal social network data necessitates more advanced statistical methods than were used in previous studies of adolescent alcohol use.
A new analytical approach to the analysis of the co-evolution of social network ties and behavior is stochastic actor-based modeling [21
], which provides a powerful new tool to simultaneously model an agent’s selection of friends based on alcohol use and changes in an agent’s alcohol use behavior over time. The primary assumptions of the actor-based model are that individuals choose their friendship ties and their behaviors in one step-at-a-time micro-steps. At each micro-step, an agent maximizes a personal utility function for surrounding network and relative friend behavior. At that time point, the agent only considers the current network characteristics in deciding whether a change in behavior or network tie is preferable to the current state. In this manner, the process of co-evolution of network and behaviors from one wave of data to another is simulated as a result of a potentially large number of individually unobserved micro-step changes, and the network and behavior preferences parameters can be estimated. The actor-based model can disentangle selection and influence and determine their relative contribution to similarities in alcohol use behavior among friends [23
Recent studies have begun to use actor-based modeling to examine the relative contribution of selection and influence effects on adolescent alcohol use [26
]. In one such study, a sample of 1,204 7th
graders in Finland were followed for 30
months to determine the degree to which the children selected or were influenced by friends based on alcohol use. The results indicated both selection and influence played a role in alcohol use similarities among friends, although influence was stronger at younger ages and selection became stronger as students aged [30
]. Another study followed cohorts of 4th graders, 7th graders, and 10th graders in Sweden for 2
years. The findings indicated that selection based on alcohol use was strongest in early adolescence, while both influence and selection effects contributed to alcohol use similarities during later adolescence [29
]. More research is needed to determine the relative role of selection and influence in alcohol use homogeneity within adolescent friendships. The current study will add to this body of literature by examining alcohol use behavior in a large sample of 7th through 11th grade U.S. adolescents.
Without clear guidance on the causal pathways between peers and alcohol use, adolescent alcohol interventions may be incomplete [9
]. To fill this gap in the literature, the present study will investigate the selection and influence processes as they relate to peer friendship formation and alcohol use behavior in the larger context of adolescent friendship networks. Specifically, the study will address the following research questions:
Selection Research Question #1 Do adolescents select friends with similar alcohol use?
Influence Research Question #2 Do adolescents adjust their alcohol consumption in correspondence with the alcohol consumption level of their friends?
We hypothesize that both selection and influence effects will be present in the network model of adolescent alcohol use.