Building Stochastic Actor-Oriented Models to investigate social network influences on youth nicotine product use transitions and to simulate different intervention effects across contexts - In the United States, Mexico, and in many countries around the world, teens’ combustible cigarette (CCs) use has declined in recent years, yet that progress is threatened by a dramatic rise in e-cigarette (EC) use. Prior research finds that teen CC&EC use is similar among best friends, yet the mechanisms leading to this clustering of behavior are unclear. Friends can spur changes in each other’s behavior (influence), but peers with similar risk for and use of CC&ECs can become friends (selection). Existing research on peer influence and CC use is almost exclusively based on best friendship networks, but emerging research suggests that more intimate connections like best friendship relations (strong ties) are empirically different from comparatively less intimate, and nowadays highly ubiquitous, online interaction networks (weak ties), such as people with whom teens interact via social media. Furthermore, recent school-based interventions have successfully leveraged interaction networks, not friendship networks, to reduce teen bullying and CC use. This study will investigate these network mechanisms by gathering data through 6 waves over 2.5 years from a cohort of Mexican high schoolers, where we will separately measure their (best) friendship and online interaction networks, measure teens’ preferences for CC&EC products that policy can influence (e.g., flavors) using discrete choice experiments (DCEs), and evaluate initiation and progression of CC&EC use. Stochastic Actor-Oriented Models (SAOMs), a family of Agent-Based Models specifically designed to permit statistical inference, will be used to analyze the co-evolution of social network dynamics and CC&EC use dynamics, while accounting for the interplay between online interaction and friendship networks, as well as the interplay between CC&EC use (e.g., exclusive vs. sequential vs concurrent CC&EC use). Our SAOMs will incorporate DCE-derived individual-level preferences (i.e., CC vs. EC; tobacco vs. other flavors), something the Agent-Based Modeling literature around health has not done to date. The resulting SAOMs will be used to empirically calibrate agent-based simulations that will serve as a virtual laboratory for evaluating the relative effectiveness of different network-based and other intervention strategies to reduce teen CC&EC use. Workshops with key stakeholder groups (e.g., students, school administrators, federal decision makers, advocates) will solicit feedback around the relative effects of intervention strategies, the feasibility of adoption and implementation of different strategies, and alternative interventions that our models can simulate. Finally, we will harmonize our surveys with data currently being collected in a cohort of high schoolers in the US (where network data are similar but less complete), so that we can compare models across countries to evaluate the consistency of network effects on CC&EC use and ensure the utility of our results for the US. Study results will not only help with efforts to promote and implement network and other interventions that aim to reduce CC&EC use; results will also inform efforts to extend theories around the importance of both friendship and online networks for influencing health behaviors, especially among teens.
