Identifying Social Network Conditions that Facilitate Sedentary Behavior Change: The Benefit of Being a “Bridge” in a Group-based Intervention

Using data from one of the first trials to try to leverage social networks as a mechanism for obesity intervention, we examined which social network conditions amplified behavior change. Data were collected as part of a community-based healthy lifestyle intervention in Nashville, USA, between June 2014 and July 2017. Adults randomized to the intervention arm were assigned to a small group of 10 participants that met in person for 12 weekly sessions. Intervention small group social networks were measured three times; sedentary behavior was measured by accelerometry at baseline and 12 months. Multivariate hidden Markov models classified people into distinct social network trajectories over time, based on the structure of the emergent network and where the individual was embedded. A multilevel regression analysis assessed the relationship between network trajectory and sedentary behavior (N = 261). Being a person that connected clusters of intervention participants at any point during the intervention predicted an average reduction of 31.3 min/day of sedentary behavior at 12 months, versus being isolated [95% CI: (−61.4, −1.07), p = 0.04]. Certain social network conditions may make it easier to reduce adult sedentary behavior in group-based interventions. While further research will be necessary to establish causality, the implications for intervention design are discussed.

The following individual-level network statistics [25] were computed to capture the extent to which participants were central and connected in their group social network:  Indegree centrality: the number of incoming ties (i.e., nominations) a participant received from other group members, normalized by the maximum theoretical indegree (group n-1). This measures a participant's popularity regarding providing advice to other group members.  Outdegree centrality: the number of outgoing ties (i.e., nominations) made by a participant to other group members, normalized by the maximum theoretical outdegree (group n-1). This measures how actively a participant seeks advice from other group members.  Betweenness centrality: the number of advice tie paths that travel through a participant, (akin to the idea of "six degrees of separation") normalized by the theoretical maximum given the observed size of the group. Participants with higher betweenness centrality ("fewer degrees of separation from many people") can be important conduits for advice, because they can influence the flow of information and resources through the network. They may have access to more diverse knowledge or resources from different segments of the network.
The following group-level network statistics [25] were computed to reflect the global structure of participants' social networks:  Density: the proportion of ties observed among all group members, relative to possible ties among group members (i.e., group n*(n-1)). High density means that groups had many participants seeking advice from one another; low density reflects sparsely connected groups with little advice seeking and little potential for social influence and support.  Transitivity: the number of triad structures (where node A is connected to node B, node B is connected to node C, and node C is connected to node A) in the network, divided by the total number of triad structures possible given the number of 3-node sets in the group. It reflects a social process where "I seek advice from you, and from the person from whom you seek advice", and globally represents clustering of ties into smaller subgroups or "communities" likely to establish stronger norms and identities that could support healthy behavior change.

File S2:
Multivariate Hidden Markov Models MHMM can be considered an extension of the latent class analysis to longitudinal setting. Like latent class and latent profile analysis, MHMM conceptualizes a causal process in which an underlying, unobserved (hidden or latent) construct is driving multiple manifest variables called indicators (hence deemed multivariate). In the case of our study, the unobserved construct is network position, whereas the indicators form the behavioral profile of an individual. The goal of MHMM is to (1) delineate clusters of individuals that have similar behavioral profiles, and (2) estimate the dynamic of change over time in which individuals may stay as a member of a specific cluster or transition from being a member of one cluster to being a member of another. The clusters are called states in the current paper. The optimal number of states for the model is derived from goodness-of-fit indexes calculated from fitting models of different number of states to the data. Transition follows the Markov assumption or the memoryless principle -namely the transition probabilities for the possible next states are only dependent on the current state and not farther back. The MHMM analysis results in a set of parameters that include the behavioral profile of each state, the initial condition of the states, and the transition probabilities between the states. It also produces a trajectory of the most probable states for each individual. State 1. Adults in State 1 had an average number of connections (indegree centrality and outdegree centrality), although they had lower than average betweenness centrality, and they were embedded in networks of average density and transitivity. This indicates that they held average positions in average networks.
State 2. Adults in State 2 had higher than average indegree centrality and outdegree centrality and especially high betweenness centrality, indicating that they were often "bridges" among group members and had a high potential to influence others. They were embedded in networks of average density and transitivity. This indicates that they were "bridging" actors in average networks.
State 3. Adults in State 3 had lower betweenness centrality than State 2 but very high indegree centrality and outdegree centrality, and were embedded in networks with high density and transitivity. This suggests they had many ties and were in highly connected networks but were only average in "bridging".
State 4. Adults in State 4 had lower than average indegree centrality, outdegree centrality, and betweenness centrality, and they were embedded in groups with low density and transitivity. This suggests they were isolated in sparse networks.

Prevalence of the Four Latent Network States at Each Timepoint
States 1 and 4 were the most prevalent across the timepoints (Figure 2). The number of adults in State 4 declined from 60% at week 3 to 30% at week 12. In contrast, the number of adults in State 1 increased from 15% at week 3 to 35% at week 12. State 3 was not present at week 3, early in network formation, but 7% of adults were in State 3 at week 6, and 20% were in State 3 at week 12.

Transition Probabilities Between States
Adults in States 3 and 4 had ≥ 60% probability of staying in the same state over time, whereas adults occupying State 2 had a 40% probability of staying in that state over time ( Table 1). Few adults transitioned into State 4.