Predicting the Diffusion of Intervention Effects Using Social Network Analytic Measures

Introduction: Many evaluation studies assess how interventions impact individuals, but there is an increasing interest in clarifying how interventions can impact larger social settings. One process that can lead to these setting-level effects is diffusion, in which intervention effects spread from participants to non-participants. Diffusion may be particularly important when intervention participation rates are low, as they often are in universal family-based interventions. Notably, some networks may be more likely than others to support diffusion. Some network-level features that may support diffusion can be assessed using traditional analytic measures, but other features must be measured using tools from the field of social network analysis (SNA). In this paper, we tested the extent to which 10 SNA measures predicted diffusion over and above the effect of traditional analytic measures.

Method: Data were from 42 networks involved in the PROSPER intervention trial (n = 5,784 students; M = 11.8 years; 50% Female; 82% White, 6% Hispanic, 12% Other). All families of 6th graders were invited to participate in a family-based substance use intervention, and <20% of families attended any sessions. All students completed a pretest survey in Fall of 6^th grade and surveys in Spring of 6^th (posttest), 7^th (1-year follow-up), and 8^th grade (2-year follow-up). As part of the surveys, students reported their own substance use and named up to 7 friends. We measured diffusion with Cohen’s D, comparing substance use between participants and non-participants (lower Cohen’s D = more evidence of diffusion as diffusion leads participants and non-participants to become more similar over time). We calculated 10 SNA measures of diffusion potential from the friendship nominations.

Results: We correlated each SNA measure with diffusion at 1- and 2-year follow-up, controlling for network size, survey participation rate, and traditional analytic measures of diffusion potential. Diffusion was greater in highly connected networks, less clustered networks, more hierarchical networks, and in networks where a higher proportion of non-participants were within 2 steps of participants. Contrary to hypotheses, the relative social status of participants did not predict diffusion. The SNA measures were more predictive at the 2-year follow-up (suggesting that diffusion is a slow process). Finally, the posttest SNA measures were more predictive than pretest measures (suggesting that interventions may impact network structure).

Conclusions: These SNA measures provide one way to assess how a family-based intervention may change social processes within schools. We end by recommending which SNA measures may be the most promising for studying the diffusion of intervention effects.