Social network analysis involves the collection and analysis of survey data about with whom individuals interact and their relationships, so that social or information networks can be specified using lines (relationships) and nodes (individuals/ organisations) (Knoke and Yang, 2019). At the same time behavioural outcomes of interest can be collected, such as alcohol consumption. The impact of the social networks upon behaviours can then be assessed.
It may be that intervention studies are undertaken within a population or community where the impacts of social networks are already (implicitly) captured, for example studies assessing the smoking ban in public places. Assuming additional impacts of the interventions due to social networks in this situation could lead to double counting if study follow up is sufficient. It is therefore important to understand the sample included within the intervention studies of interest; if it reflects the population and follow up is substantial then it may not be appropriate or worthwhile incorporating social networks explicitly. However, if social influences on behaviour would not have been captured by the intervention studies, then social network analysis could be considered. Undertaking social network analysis has the advantage that the relationship between the social network and the behaviour(s) of interest can be informed by the data collected.
Collecting data from the full network is ideal; however, practically this may not be possible. Egocentric network analysis, which can be done by collecting relationship data from individuals (or ‘egos’) from a sample of the population who may or may not be connected, may be more feasible and has been shown to provide reasonable results compared with a full network (Smith et al., 2018). Most social network analysis assumes that social relationships are constant. Stochastic actor-based models for network dynamics allow social relationships to evolve over time as they would in practice (Snijders, 2017). This approach could be considered when an intervention might change social relationships, for example, if university students are given interventions to reduce binge drinking. It requires social network data for at least two time points, although preferably more.
A literature search for social network analysis associated with the behaviours of interest is recommended to first assess the benefits of including social network analysis and second for model parameterisation. Studies exist for some behaviours and outcomes which could be utilised (Choi et al., 2020; Storey et al., 2021) if the population is relevant. If there are no existing analyses and there are insufficient resources to undertake social network analysis, random networks, small world networks (all individuals linked by a small number of nodes) or scale free network (most people have a small number of connections, whilst a few have a large number) could be used within an ABM, and there are existing software packages to do this. However, these networks make ties at random or conditional on some characteristics and might not reflect the real social network, and assumptions are required about how individuals influence each other’s behaviours.