Our methodological foundation also draws on work in network science and algorithms. For example, the generative-model paper in Algorithms (2015) presented a model for random graphs with labelled vertices and weighted edges, enabling more realistic simulation of social-network data.

This paper underpins our ability to apply network methods in complex health behaviour contexts when network data has weights, labels, multiplex ties, or other complexities. By incorporating network analysis tools (centrality, community detection, generative models) we are better placed to design, evaluate, and interpret health intervention studies that operate through network channels.

 

Implications for practice and future work

There are a number of practical and methodological implications from this work:

1. Measure networks early: Collect data on social ties, structural positions, central actors, community boundaries, network density, bridging ties. This enables you to embed network metrics in the design.
2. Select and train network actors thoughtfully: Rather than just choosing high-degree nodes, use structural metrics (e.g., closeness, betweenness, bridging ties) and algorithmic selection where feasible (as our adolescent study did).
3. Account for nested and network-driven variance: Use multilevel and network-aware statistical methods to account for dependencies between units (for example, friends of friends, clusters of influence).
4. Conceptually rethink behaviour change frameworks: Move from individual-only models to “relational” or “networked” models: how do peers, networks, brokers, sub‐communities support or hinder change?
5. Integrate advanced network science methods: Use generative models or simulations to test how network structure might moderate intervention effects; detect communities, weighted ties, multiplex layers, and so on.
6. Recognise limitations and context specificity: Network methods add complexity, data collection is challenging, dependency assumptions must be addressed, and network structures vary dynamically across contexts (schools vs communities vs online).