To our knowledge, this is the first formal analysis of a stable therapeutic online social network designed to assist health behavior change (smoking cessation and relapse prevention). We analyzed core characteristics of the entire network as well as subgroup composition. Critically, QuitNet’s metrics were similar to those of other known, characterized online networks,18-20,29,43
including characteristic graphical representations, a small number of well-connected actors with a power-law distribution of ties (i.e., a scale-free network), and identifiable subgroups. Because no standard criteria exist to validate the emergence of online social networks, this characterization derived from a persistent, stable network is a first step in the determination of the necessary and sufficient elements for future interventions and studies. These metrics represent a critical foundation for future basic and applied research to harness the full potential of social networks for population-level health behavior change. That these networks can be characterized has additional implications for advancing theory and informing interventions for behavior change in general.
We observed important similarities and differences among QuitNet’s core characteristics and those of other known networks. QuitNet’s clustering coefficient of 0.173 was within the range of other systems, including electronic messaging (0.13–0.3320
) and social network sites (0.16–0.2829
). Although the diameter of the network was similar to that of other systems, the mean degree was significantly lower at 13.69, indicating that on average QuitNet members formed fewer connections than users of other systems, whose average number of connections have been reported to range from 31 to 137.18,20
The mean can be deceptive because the distribution of tie counts did not follow a normal curve; many participants had only 1 tie, and a few had more than 1000. This scale-free pattern has been observed in many online networks,18,19,29,43
but not in all real-world networks (e.g., the Framingham Heart Study).30
In the case of QuitNet, the conformance was not perfect, the curves for Qmail and buddy lists (data not shown) displayed artifacts in their tails at between 100 and 150 ties. This is likely attributable to limited data at the higher degrees but is curiously close to Dunbar’s number, the hypothesized limit of relationships an individual can manage.44
On the other hand, the Qmail in-degree curve abruptly tailed off at 150 ties; the out-degree curve had an inflection at this point and continued with a small number with up to 1000 ties. The out-degree curve might have reflected a small number of individuals who spent large amounts of time on the site and served as unofficial welcomers and town criers, announcing events such as the anniversaries of members’ quit dates. Why certain individuals amass so many connections and spend such concentrated time within the community has not been explored thoroughly but may be attributable to status seeking,45
similar to the phenomenon that drives unpaid labor in open source software development networks.46
These similarities and differences add to our growing understanding of the nature of online social networks and highlight potential theoretical mechanisms for future study.
Participation of more women in smoking cessation programs is common. In our analysis, all subgroups of the network were predominantly female, with an increasing likelihood as ties and network density grew stronger. Similarly, age increased as ties grew stronger and network density increased: the mean age of the weakest subgroup was 42 years, and the average age of the most active participants (the key players) was 49 years. This tendency of network members to be female and older is noteworthy in light of the conventional wisdom that the population of Internet users skews in the opposite direction.
The maintenance of behavior change is as crucial—and as difficult—as the induction of the change. Other investigators have hypothesized that recent quitters may be the most likely to participate in social support systems,47
but we found a fairly equal representation within the network of abstinent smokers and those in the early stages of quitting, as well as marked heterogeneity of time on site within the network, with many who had been members for a year or longer. This is particularly important because our study found that although overall degree was negatively correlated with smoking, increasing numbers of smokers in an ego’s local network were positively correlated with smoking. Because maintaining abstinence after cessation is so difficult and because successful quitters provide valuable information and normative influence within a social network, these findings of heterophily and persistence are reassuring and suggest that evolving networks can become more effective over time.
Our key player analysis illustrated one mechanism of identifying subgroups within large networks for dissemination of information. In theory, identification of core groups such as key players could allow for more rapid and efficient dissemination of information.48
Other groups could be used to enhance network stability, growth, or density. In our QuitNet study, despite the presence of more than 7500 active members in the network, only a few (the highly active integrators) came into contact with new participants. Future research is needed to characterize network integrators and determine whether increasing their numbers or strengthening their role can effect more efficient behavior change.
We adhered to a traditional view of social networks, in which a relationship that is inferred from communications data is considered to be present throughout the observation period. In reality, the network was dynamic, and traditional network metrics may have overestimated the diffusion capacity of the network.36
We also derived information regarding smoking abstinence from participant-provided quit dates of unknown validity.
We used a limited selection of ties to define the network. Many participants appeared to be lurkers, who did not actively communicate but may have been exposed passively to normative influences such as blog postings or the profile information of other members. Finally, we know little about communications and ties between individuals that did not occur through the QuitNet system (e.g., regular e-mail, preexisting friendships, the use of other social networking systems), which may have resulted in underestimation of the strength of some ties or the omission of others.
More research is needed to determine the mechanisms and the effectiveness of persistent therapeutic networks. Our analysis provides a starting point, pointing to the challenges and potential opportunities to improve understanding of the ways social networks can be harnessed to facilitate health behavior change. Studies are needed to elucidate the determinants of network growth, stability, and effect, including age and gender proportions, the predictors of participation and dropping out,49
and the effect of long-term superusers and key players.
Social support theory suggests that the mechanisms that induce behavior change are broad and include various forms of social influence by observation, modeling, and adjustment to community norms—mechanisms that do not necessarily require explicit communications and will particularly benefit from future dynamic analysis. Our findings illustrate the potential that future research has for the development and implementation of innovative social network interventions to enhance behavior changes that can dramatically improve our nation’s health and the health of the world in the Internet age.