Evaluating peer-influence processes in a prison-based therapeutic community: a dynamic network approach

•Dynamic network analysis in a prison-based therapeutic community.•Residents clustered on treatment engagement, primarily due to racial homophily.•Residents perceived as role models were more engaged with treatment.•No evidence of peer influence on treatment engagement in community network.•Dynamic network models are a promising approach for evaluating TC processes. The Therapeutic Community (TC) is a common treatment modality for incarcerated individuals with substance use disorders. TCs rely on peer group processes to promote lasting behavioral and identity change, yet prior research has not adequately tested the peer influence mechanisms underlying the theoretical model. This study applied dynamic network analysis to estimate peer influence processes central to TC philosophy. A stochastic actor-oriented model (SAOM) was applied to ten months of social network data collected from prisoner surveys within a TC unit (N = 62) in a medium-security Pennsylvania prison. Respondents (N = 177, 84% of unit) completed at least one prison survey and provided network and community role model nominations. Although residents’ levels of treatment engagement were significantly correlated with their nominated peers, estimates of peer influence for treatment engagement were non-significant in longitudinal network models. Nor were estimates of peer influence significantly greater for peers perceived as community role models. Rather, inmates connected with peers who were of similar treatment engagement as themselves (i.e., a peer selection process), and the latter primarily resulted from racial homophily in the TC social network. Inconsistent with the desired treatment model, treatment engagement diffusion was not evident in the sampled TC. Results suggested that highly-engaged residents clustered together at the center of the TC’s social structure but had little impact on less-engaged and peripheral inmates. The relatively short (i.e., four-month) program length and moderate-to-low treatment fidelity likely contributed to the lack of peer influence processes.