In our work on behaviour change interventions among adolescents, we used influence-agent selection based on network centrality (closeness) derived from friendship networks, and then trained these agents via smartphone to promote physical activity. Although the intervention did not produce the expected effect on overall physical activity, this study advanced the field in three ways:

• it used a greedy search algorithm to identify influence agents by network centrality rather than simpler heuristics;
• it demonstrated feasibility of smartphone-based training of network agents;

it used multilevel modelling to explicitly account for nested network/individual structure.

These methodological advances show how network methods can (and should) be integrated into behaviour-change research, even if outcome results are mixed.