Social Network Structure of a Large Online
Community for Smoking Cessation
Nathan K. Cobb, MD, Amanda L. Graham, PhD, and David B. Abrams, PhD
Despite decades of research, tobacco use re-
mains the most deadly of behaviors, causing 5
million deaths worldwide annually
1
and pro-
jected to cause 10 million per year by 2030.
2
The United States has an estimated 44.5 million
smokers, leading to 430000 premature deaths
annually.
3
Evidence-based cessation interven-
tions exist but are vastly underutilized by
smokers.
4
There is a pressing need to maximize
the population impact of cessation with innova-
tions that are attractive and accessible to con-
sumers.
3
One method is to leverage social
network effects, which play a prominent role in
the induction of smoking cessation and the
perpetuation of abstinence.
5
Observational studies support a robust re-
lationship between social support and positive
outcomes for smoking, other health behaviors,
and health status.
6,7
Higher levels of connect-
edness and positive social support are asso-
ciated with smoking cessation and relapse
prevention.
8–11
Negative social support (e.g.,
a spouse who smokes or is critical of attempts at
cessation) are barriers to cessation.
11
After these
associations were established, intervention stud-
ies manipulated supportive interactions outside
the context of cessation treatment as a means
to improve outcomes, with disappointing re-
sults.
8,10–15
Consequently, enthusiasm for social
support interventions waned,
16
and the focus
shifted to delivering the briefer treatments pre-
ferred by smokers.
17
Online social networks, which have prolif-
erated in the past decade, offer a novel way to
address the gap between observational data
and lackluster intervention effects. Social net-
work interventions may work through multiple
mechanisms, including social support, infor-
mation transfer, social influence, modeling, and
the transmission of social norms. Despite the
growth of online communities and networks,
few published reports describe their charac-
teristics.
18–21
Moreover, health behavior studies
containing social network features have not
documented the characteristics of the social
network itself.
22–25
Before network effects are
studied, it is critical to determine whether a true
social network has developed. Otherwise, efforts
to evaluate the efficacy of a social network
intervention may fail if researchers unwittingly
study a system that has not yet developed into
a functional, sufficiently heterogeneous, large,
and stable network or in which the ties between
participants are weak or insufficient. Finally, no
social network studies to our knowledge have
examined the mechanisms that might underlie
theireffectivenessinchangingbehavior.
We used formal network methods and ana-
lytic techniques to explore key structural and
functional characteristics of a large, known
online community for smoking cessation. Spe-
cifically, we sought to (1) characterize the social
network and participants of this community,
(2) describe its structure and establish that it
shared characteristics with other known online
networks, and (3) identify subgroups whose
existence and characteristics might inform the
design of cessation interventions. Our intent was
to establish the necessary foundation for sub-
sequent investigations into the effectiveness of
online social networks in influencing cessation
outcomes as well as to advance understanding
of social network effects in tobacco treatment.
METHODS
QuitNet (http://www.quitnet.com) is one of
the most popular, long-lived, and successful
continuously operating online social networks
focused on smoking cessation. For over 10
years it has enrolled individuals into a network
of current and former smokers seeking to quit
or stay abstinent and has provided multiple
mechanisms of social support and influence.
Characteristics of QuitNet’s users and details
regarding its development and evolution are
published elsewhere.
25–27
Since the inception
of its social network features in 1997, more than
800000 individuals have registered. In 2007,
QuitNet had approximately 1.2 million unique
visitors, of whom 123927 registered as new
members (L. Severtson, Healthways QuitNet,
personal communication, March 7, 2008).
QuitNet’s community features allow for
multiple forms of social support. Communi-
cation can occur through asynchronous chan-
nels (e.g., private internal e-mail [Qmail] or
Objectives. We evaluated the social network structure of QuitNet, one of the
largest online communities for behavior change, and compared its characteris-
tics to other known social networks.
Methods. Using modern network analysis methods, we identified QuitNet
members who were active during a 60-day period, along with their ties. We then
derived multiple subgroups, such as key players and integrators, from connec-
tions and communication patterns.
Results. Among 7569 participants, we identified 103592 connections to other
members. Metrics of social network integration were associated with increased
likelihood of being female, being older, having been in the system longer, and
not smoking.
Conclusions. The QuitNet community is a large-scale social network with the
characteristics required for sustainability of social support and social influence
to promote smoking cessation and abstinence. These characteristics include
persistence of members over time, heterogeneity of smoking status, and
evidence of rich, bidirectional communications. Some of the influential sub-
groups we identified may provide targets for future network-level interven-
tions. (Am J Public Health. 2010;100:1282–1289. doi:10.2105/AJPH.2009.
165449)
RESEARCH AND PRACTICE
1282 | Research and Practice | Peer Reviewed | Cobb et al. American Journal of Public Health | July 2010, Vol 100, No. 7

one-to-many messaging in the threaded fo-
rums) or through synchronous channels (such
as chat rooms). Users can self-affiliate into clubs
(user-initiated minisites, complete with dedi-
cated forums), and buddy lists allow individuals
to keep track of their friends. Social influence
regarding cessation is conveyed through profile
pages, journals (similar to a blog), anniversary
lists, and testimonials. Users are encouraged to
publicly share their quit dates, which are set
through a wizard tool, and users are prompted
for updates at each login.
QuitNet maintains a complete transactional
history of all events, including communications
that occur throughout the site. Active events
(e.g., sending internal e-mail, posting a public
message) and passive actions (e.g., reading
messages, viewing another individual’s profile)
are logged into a relational database. This
database provides a rich source of information
about social network ties (a connection be-
tween 2 actors, such as a communication or
friendship; also called an edge or a link)—literal
evidence of communications and links between
participants.
Data Extraction
We compiled data on registered QuitNet
members in the United States who indicated
during registration that they were looking for
smoking cessation help for themselves and who
logged in to the Web site during the 60-day
study period (March 1, 2007–April 30, 2007)
and completed 1 or more of the following
actions: (1) exchanged an internal message with
another participant, (2) posted a message
within the online forums, and (3) added, or was
added by, another QuitNet participant to
a buddy list. This data set included both new
users and QuitNet members who registered
before the study period. Individuals who met
the inclusion criteria formed the weakly con-
nected core. We delineated subsets of the
online community in this core.
We collected anonymized registration data
for all members of our core sample. We also
collected Web site utilization data, including
records of logins, message exchange time
stamps (internal e-mail sent and received,
forum posts), additions to buddy lists, initial
motivation to quit according to the stages of
change algorithm,
28
and subsequent recording
and changes to quit dates that occurred during
the 60-day window. We extrapolated smoking
status by carrying forward status at registration
and adjusting it according to user-provided quit
dates. Individuals who provided a quit date that
fell within or after the observation period (or
failed to provide a quit date) were coded as
smoking; individuals whose last known quit date
was prior to the observation period were
coded as abstinent. At the end of the 60-day
period, we calculated duration of participation
(time on site) by the number of days since
registration.
Because our initial data set (the weakly
connected core) was large and the number of
participants within this core who had few ties
was also large, we delineated 5 subsets of
participants. We first identified additional net-
work cores, subsets of the graph that were
connected with a relatively small diameter (the
longest path through the network when the
shortest possible path is selected for any 2
participants).
29
This is equivalent to the widely
disseminated concept of 6 degrees of separation,
where the maximum shortest pathway through
the theoretical world is 6 degrees. We defined
a strongly connected core of actors (individuals
having connections to other individuals; also
referred to as a node or a vertex) as individuals
connected by buddy nominations plus observed
communications and a densely connected core
as actors connected by symmetric buddy nomi-
nations plus a minimum of 5 communications
with at least 1 buddy during the observation
period. We chose symmetric buddy nominations
to differentiate the strongly connected core from
the densely connected core because previous
research in real-world networks indicated that
behavior change may be more likely when
nominations are symmetric.
30
We then delineated 3 additional subgroups
directly from the weakly connected core:
a group of new registrants from the initial 4-
week period (newcomers), their alters (integra-
tors; alters are actors with a tie to another actor
of interest, known as an ego), and key players,
a set of actors with high levels of connection to
the entire community.
Data Analysis
We examined demographic, smoking his-
tory, and Web site utilization characteristics
for the entire community and each sub-
group (Table 1). We used parametric and
nonparametric tests to determine the statistical
significance level. Members of the subgroup
were removed from the larger group (e.g., the
strongly connected core from the weakly con-
nected core) prior to analysis so that compar-
isons were between nonoverlapping groups.
We performed logistic regressions with cen-
trality measures (degree, the number of alters
to which an ego is connected, and Freeman’s
betweenness, the number of shortest paths that
include a given ego) as categorical data, zero as
the referent, smoking status as the dependent
value, and controls for age, gender, time since
registration, and number of logins. We used
SPSS for Windows version 17.0 for these
analyses.
31
We examined the community structure to
see whether it displayedcommoncharacter-
istics of social networks. For network ma-
nipulation, characterization, and statistical
analyses, we used the software program
ORA,
32
with the exception of Freeman’s be-
tweenness and the core–periphery correlation,
for which we used UCINet 6.
33
We created static
graphs and time-lapse animations with an itera-
tive spring-embedded algorithm to minimize
overlapping ties.
34
For static graphs we used the
Pajek program
35
and for animations, SoNIA
1.2 .
36
Although individuals with high degrees (high
numbers of alters—ties may link in one di-
rection but not the other [such as e-mail]; this is
referred to as the in-degree for ties to an actor
and the out-degree for ties emanating from
an actor) were easily identified, these individ-
uals often had significant overlap with other
well-connected individuals. A group of actors
that could reach the maximum proportion of
the rest of the network within a set maximum
path length was termed a key player set. We
derived key player sets from the weakly con-
nected core with the software program Key-
Player 1.4.
37
To derive the key player set, we
used diffusion measures, minimization of recip-
rocal distance, and the greedy algorithm as
software settings, with internal e-mail as the
primary tie.
RESULTS
The weakly connected core comprised
7569 QuitNet participants who met the in-
clusion criteria; these members had 103592
RESEARCH AND PRACTICE
July 2010, Vol 100, No. 7 | American Journal of Public Health Cobb et al. | Peer Reviewed | Research and Practice | 1283