Statistics Senior Thesis Defense by Joseph LaRocca '22

Homophily, Assimilation, and Drug Use in Rural Eastern Kentucky: An Analysis Using Stochastic Actor-Oriented Models by Joseph LaRocca ’22, Statistics Senior Thesis Defense, Tuesday, May 17, 1:10 – 1:50 pm, North Science Building 017, Wachenheim.

Abstract:  The United States is currently in the midst of a nationwide drug epidemic; the Centers for Disease Control and Prevention (CDC) estimated that nearly 100,000 Americans died from drug overdoses in 2020 alone.  We sought to investigate the co-evolution of social network ties and drug use behaviors from the Social Networks among Appalachian People (SNAP) dataset, which contains five observations of 503 individuals, all of whom used drugs at the beginning of the study, over a two-year period.  Using the stochastic actor-oriented model (SAOM), we examined the relationship between drug, sex, and social support ties within the SNAP network and the use of injection drugs, heroin, methamphetamines, cocaine, and illegal prescription opiates, with a specific interest in identifying whether homophily (the tendency to create or maintain network ties with those who exhibit similar behavior) or assimilation (the tendency to adjust one’s behavior to that of one’s network ties) was at play with respect to each type of drug. SAOMs are named as such because they are stochastic in nature (specifically, they follow a Markov process) and simulate actors, or the people in the network, making their own autonomous decisions. SAOMs are powerful tools for analyzing longitudinal social network data, since they can account for the simultaneous influence of networks and behaviors on each other using two separate but concurrent Poisson processes that determine when each actor has the opportunity to make changes to their network ties or behavior.  Using four separate SAOMs, we found very strong evidence of homophily (p < 0.001) and assimilation (p < 0.001) with respect to injection drug use and moderately strong evidence of assimilation with respect to methamphetamine/cocaine use (p < 0.05), but did not find any other significant evidence of either effect.  Therefore, we concluded that there is a particularly strong connection between the SNAP network and injection drug use when compared to other types of drug use.  We hypothesized that the strength of this association is attributable to the fact that injection drug use is an inherently social activity that often involves the sharing of needles.