programs have been developed for smoking cessation

Using Content and Network Analysis to Understand the Social Support Exchange Patterns and User Behaviors of an Online Smoking Cessation Intervention Program Mi Zhang and Christopher C. Yang College of Computing and Informatics, Drexel University, 3141 Chestnut Street, Philadelphia, PA 19104. E-mail: {mz349, chris.yang}@drexel.edu Informational support and nurturant support are two basic types of social support offered in online health communities. This study identifies types of social support in the QuitStop forum and brings insights to exchange patterns of social support and user behaviors with content analysis and social network analysis. Motivated by user information behavior, this study defines two patterns to describe social support exchange: initiated support exchange and invited support exchange. It is found that users with a longer quitting time tend to actively give initiated support, and recent quitters with a shorter abstinent time are likely to seek and receive invited support. This study also finds that support givers of informational support quit longer ago than support givers of nurturant support, and support receivers of informational support quit more recently than support receivers of nurturant support. Usually, informational support is offered by users at late quit stages to users at early quit stages. Nurturant support is also exchanged among users within the same quit stage. These findings help us understand how health consumers are supporting each other and reveal new capabilities of online intervention programs that can be designed to offer social support in a timely and effective manner. Introduction Tobacco use is the single largest preventable cause of death and disease in the United States. In 2011, an estimated 19.0% (43.8 million) of U.S. adults smoked cigarettes (Agaku, King, & Dube, 2012). Lung cancer, ischemic heart Received June 18, 2013; revised November 13, 2013; accepted November 14, 2013 2014 ASIS&T Published online 19 May 2014 in Wiley Online Library (wileyonlinelibrary.com)..23189 disease, and chronic obstructive pulmonary disease are highly related to smoking. At least 30% of cancer deaths are associated with smoking, and 440,000 U.S. citizens die from tobacco use every year (National Institute of Drug Abuse, 2012). To reduce the burden of tobacco use, many intervention programs have been developed for smoking cessation. Traditional intervention programs are based on face-to-face consulting. Although they help people quit smoking (Russell, Wilson, Taylor, & Baker, 1979), the scalability of these programs is limited. With the growth of the Internet and the popularity of social networking, more online intervention programs have been developed for smoking cessation in the recent years, which have the potential to reach a large population around the world. Some studies applied social network analysis to investigate user behaviors in these smoking cessation programs. Compared to peripheral users, it is shown that core users in these programs tend to be female, older and in the abstinent status of smoking (Cobb & Graham, 2006; Cobb, Graham, & Abrams, 2010). Various services are provided to increase users quit ratio in these online intervention programs (An et al., 2008; Cobb, Graham, Papandonatos, & Abrams, 2005; Graham, Cobb, Raymond, Still, & Young, 2007; Shahab & McEwen, 2009). QuitNet (http://www.quitnet.com), founded in 1995, is a popular website supporting smoking cessation that provides intervention services including interactive diagnostic tools, quit planning tools, online expert counseling, online communities, and one-to-one messaging (An et al., 2008). The quit rate of users with sustained use of QuitNet services is more than three times higher than that of other users (An et al., 2008). In health intervention programs, social support plays an important role in helping people achieve better intervention outcomes. Social support is an exchange of resources between two individuals perceived by the provider or the recipient to be intended to enhance the well-being of JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY, 66(3):564 575, 2015

the recipient (Shumaker & Brownell, 1984, p. 3). In traditional health intervention programs, social support is usually offered to patients through face-to-face communication. With a large number of communities built on the Internet for health discussion, social support is exchanged among Web users online. QuitNet developed 11 forums on its website, among which QuitStop (http://forums.quitnet.com/aspbanjo/message_list.asp?conference_id =10&Forum_ID=8&r=100777) is the most popular community, where users can discuss the tobacco quitting process, ask questions, and give or receive social support. In the QuitStop forum, both smokers and ex-smokers participate in discussions of smoking cessation and exchange social support. It provides an ideal platform on which smoking quitters can receive effective social support and achieve good intervention outcomes. However, previous studies of online smoking cessation programs did not analyze social support exchange from the perspective of user interactions. In this study, we analyze social support exchange in the QuitStop forum online community. The different types of social support are extracted, and the patterns of social support exchange are analyzed. There are two parts of this study: first, the types and frequencies of social support exchange and, second, the patterns and user interactions of social support exchange. Different types of social support are offered in health intervention programs (Chuang & Yang, 2010, 2011; Eichhorn, 2008), which can benefit support receivers in different aspects. Identifying types of exchanged social support could help us understand user features and interactions in online health intervention programs. However, to our knowledge, there are few studies investigating social support in online communities for smoking cessation. In the first part of this study, we analyze the types and frequencies of social support exchange on QuitStop forum, where two research questions are proposed: 1. What types of social support are exchanged on the QuitNet Forum? 2. What is the exchange frequency of each type of social support? Besides types and frequencies of social support, we are also interested in the interactions of users to exchange social support. Based on research into information-seeking behavior in everyday life, Savolainen (1995) proposed two information patterns, which are practical information and orienting information. Similar to information exchange, social support exchange is a social phenomenon. It is important to study who exchanges information with whom, about what, and by which media (Haythornthwaite & Wellman, 1998). Motivated by information patterns defined by Savolainen, we focus on the exchange patterns and user behaviors associated with social support. According to previous research, smoking quitters can be categorized into different quit stages according to their quit status and abstinence days (Velicer, Prochaska, Rossi, & Snow, 1992). We also analyze social support exchange between users at different quit stages. There are two research questions proposed for the second part of our study: 1. What are the exchange patterns of social support between user pairs on the QuitStop forum? 2. How do users at different quit stages interact with each other to exchange different types of social support? In the following sections, we first introduce related works in Literature Review. Then, our research structure and data set are described in the Research Design section. The two aspects of our study are described in the next two sections. Last, we summarize our work in the Conclusion section and list the limitations and outline future work. Literature Review Content Analysis for Online Health Discussions A large amount of research has investigated the discussion content of different health forums and online groups. Bender et al. (2011) analyzed 620 breast cancer groups on Facebook and found that the discussion themes include fundraising, awareness, product or service promotion related to fundraising or awareness, or patient and caregiver support. Greene, Choudhry, Kilabuk, and Shrank (2011) extracted five themes from diabetes discussion groups, which were advertisements, providing information, requesting information, support, and irrelevant content. In online health communities, people adopt different strategies to request social support. Eichhorn (2008) identified five strategies, including self-deprecating comments, shared experiences, requests for information, statements of personal success, and statements of extreme behavior. In our previous research, we found five discussion themes on QuitStop forum, including offering social support, requesting social support, receiving social support, other activities, and irrelevant content (Zhang, Yang, & Gong, 2013). Different types of social support are requested and offered in online health communities (Chuang & Yang, 2010, 2011; Eichhorn, 2008). Chuang and Yang (2010, 2011) extracted two main types of social support from discussions of an online alcoholism community: informational support and nurturant support. Informational support is described in many studies (Chuang & Yang, 2010, 2011; Cutrona & Suhr, 1992; Hwang et al., 2011; Mo & Coulson, 2008), which is similar to action-facilitating support and task-oriented support (Cutrona & Suhr, 1992; Finn, 1999). Informational support is given to help patients solve or eliminate health problems. Nurturant support, proposed by Cutrona and Suhr (1992), is similar to socioemotional support (Finn, 1999). It is given to comfort and console patients without direct efforts to solve problems. JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015 565

Benefit of Social Support Social support plays an important role in health intervention programs. It can directly contribute to the physical and psychological outcomes of health interventions (Wright & Bell, 2003). Social support can also shield patients from the negative effects of stress and help them fight against and control stress (Lichtenstein, Glasgowb, & Abrams, 1986; Wright & Bell, 2003). In traditional intervention programs for smoking cessation, it was found that social support was positively correlated with better outcomes, including successful cessation and maintenance (Lichtenstein et al., 1986). Enlisting a support person can increase the success quitting ratio (Pirie, Rooney, Pechacek, Lando, & Schmid, 1997). Traditionally, quitters can only get limited social support through face-to-face communication in pairs or small groups at specific times and locations. Although spouses or significant others have the potential to provide social support to quitters, it is difficult to change their behavior to offer effective support because smoking cessation may not be a topic of their concern in daily life (Lichtenstein et al., 1986). The wide use of web tools in the health area heralds the era of Health 2.0 (Hughes, Joshi, & Wareham, 2008), of which communication is an important feature (Eysenbach, 2008; Van De Belt, Engelen, Berben, & Schoonhoven, 2012). Large numbers of online communities have been developed for health issue discussions. Social support is exchanged online, where geographic and transportation barriers are absent. In online communities, the number of participants can be unlimited, so a large number of people at different locations can communicate with each other (Caplan & Turner, 2007; White & Dorman, 2001). People usually feel safer when discussing personal issues in online communities because of anonymity (Caplan & Turner, 2007). Caplan and Turner (2007) found that online social interactions can more easily achieve effective comforting communications. Social support may be exchanged smoothly among users in online communities for health issues, and in the area of smoking cessation intervention, numerous online communities have been created where people can give and receive social support. Murray, Johnston, Dolce, Lee, and O Hara (1995) found that in a traditional intervention program for smoking cessation, quitters supported by ex-smokers included in the same intervention program could achieve better quit outcomes. Information Behavior and User Interaction In studying information-seeking behavior, Savolainen (1995) defined two types of information: practical information and orienting information. When seeking practical information, people look for answers to discrete and specific information needs. When seeking orienting information, people do not have specific questions. They put themselves around the information neighborhood, in which there is information related to their ongoing interests and concerns (Burnett, 2000). Based on Savolainen s theory, Burnett (2000) observed and summarized two types of messages that indicate collaborative interactive behaviors in different online communities. The first type is the noninformational message, including neutral, humorous, and emotional messages. The second type is the informational message, including announcement pointers to information sources, announcement personal updates, query to group, group projects, response to queries, and queries to individuals. Social network analysis is an important method to analyze connections and interactions among people at a macro level. Cobb et al. (2010) built a social network on different types of user interactions on the QuitNet website. They extracted different subgroups and analyzed user features of each subgroup. Chang (2009) built social networks for different types of social support exchanges and calculated the basic network measurements for each network, including size, density, clique, and network centralization. Research Design A concurrent study is carried out in this research. On the QuitStop forum, there were 3,017 threads including 3,017 posts and 24,713 comments made between May 1, 2011, and May 31, 2011. We randomly selected 228 threads (228 posts and 1,672 comments) as a sample for analysis. All collected data are publically and freely accessible on the Internet, and the data are de-identified before conducting the qualitative analysis. Mixed methods are used in this study. We analyze the content of user discussions as well as the patterns of user interactions. Qualitative analysis is used to analyze the types of social support on QuitStop forum. It is also used to extract and define user interaction patterns of social support exchange. However, it is difficult to describe the full scale of user interactions only with qualitative analysis. Thus, we also adopt social network analysis and statistical analysis, which are quantitative approaches, to build user interaction models. We first analyze the content of messages (including posts and comments) and identify the types of social support given in the messages. Then, the givers and receivers of different types of social support are extracted, and the exchange patterns of social support are analyzed. Finally, social network analysis is employed to analyze user interactions with exchanges of different types of social support. Types and Frequencies of Social Support Exchange Not all messages on QuitStop involve social support exchange between users. In our previous study (Zhang et al., 2013), we summarized five themes in the same 566 JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015

TABLE 1. Numbers of posts and comments giving different types of social support. Posts Comments Posts Comments Informational support 51 371 Nurturant support 8 525 Advice 2 124 Esteem 2 363 Referral 6 19 Network 1 30 Fact 11 26 Emotional 6 182 Perceptual knowledge 26 38 Personal experience 13 178 Feedback/opinion 0 112 data sample, including offering social support, requesting social support, receiving social support, other activities, and irrelevant content. Messages with themes other activities and irrelevant content are not associated with social support exchange. For example, there are messages urging support for others, for example, to those of us that have been blessed to know PetroJMan (Bruce) please take a moment to send him a big hug as he has always done for us... and your prayers and good thoughts. Also some messages describe personal information without exchange of social support, for example, 7 days, 16 hours, 38 minutes, and 59 seconds smoke free. The theme of these messages is other activities, which are related to smoking cessation, but no social support is directly exchanged between users in the same thread. Some messages are irrelevant to smoking cessation, for example, Iced mochas are my favorite form of caffeine. In this study, we eliminate messages with themes other activities and irrelevant content, and retain other messages for analysis. To extract and understand types of social support, qualitative analysis is used to analyze the content of posts and comments in our data sample. We first developed a coding scheme, and two coders independently coded the content of posts and comments in the data sample to determine the types of social support given in these messages. For each post and comment that offers social support, the support types are not exclusive. Both informational support and nurturant support may be given in a same message. The inter-rater reliability was calculated for validation after coding process. Coding Scheme of Social Support Types To answer the first research question What types of social support are exchanged on QuitNet Forum, we developed a coding scheme to describe different types of social support on the QuitStop forum. In studies of online healthcare communities, different coding schemes have been developed for discussion topics (Ahmed, Sullivan, Schneiders, & McCrory, 2010; Bender, Jimenez-Marroquin, & Jadad, 2011; Eichhorn, 2008; Greene et al., 2011) and types of social support (Chuang & Yang, 2010, 2011; Cutrona & Suhr, 1992; Eichhorn, 2008; Hwang et al., 2011; Mo & Coulson, 2008). Informational support and nurturant support are two main types of social support described by Chuang and Yang (2010, 2011). To analyze the types of social support on the QuitStop forum, we first use deductive methods to apply codes developed by Chuang and Yang on the sample data. Then, inductive methods are used to revise and improve the existing coding system. Below is the description for our final coding scheme. We also list examples in our data sample for each category. The complete coding scheme is summarized in Table 1. Informational support is specific information about the disease, treatment, or coping (Chuang & Yang, 2011; Cutrona & Suhr, 1992). Its subcategories include: Advice: offering suggestions to solve specific problems according to a recipient s situation, for example, Too soon to go off the zyban. Stay on it until you feel like an ex-smoker than think about weaning off the zyban. See what the doctor says. Referral: referring recipients to other sources for further help, for example, The cognitivequitting.com website was very helpful. Fact: reassessing the situation and presenting facts, for example, Vitamin B12 and D3 work on the energy stores of your body. Perceptual knowledge: providing sensory information to reassess smoking cessation and help to establish confidence, for example, A crave is just a cramp. Not serious, it ll pass off soon. Personal experiences: telling stories about personal experiences for suggestions, for example, when I first started on the patch, my BP was a bit shaky for while too, even though I m on BP meds. Feedback/opinion: a judgment about a recipient s situation or idea, for example, sounds like your coming along very nicely. Nurturant support is expressing care or concern as well as expressing the importance of relationship (Chuang & Yang, 2011; Cutrona & Suhr, 1992). Subcategories include: Esteem: praising the achievement of support seekers, for example, Great job on your quit so far. Network: broadening support seekers networks to help them feel connected, for example, Reach out and grab someone else... stay close to the Q these first few days... stay busy helping others and let yourself be helped. Emotional: providing compassion and understanding, which helps support seekers building confidence, for example, When you get to that point, you will be fine. JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015 567

Frequencies of Different Types of Social Support According to content analysis of messages in the data sample, social support is offered in 54 posts and 714 comments. To answer the second research question, What is the exchange frequency of each type of social support, two coders independently assigned types of social support to these messages and calculated the frequency of each type of social support exchange. The probability of random agreement is 87.5%, and the Cohen s Kappa of all subcategories is 51.0%, which indicates moderate agreement (Viera & Garrett, 2005). As there are many exclusive subcategories, the probability of each message belonging to a certain category is low, which reduces the value of Cohen s Kappa. However, the high rate of agreement indicates a strong agreement between two coders. After independent coding, the two coders discussed messages with disagreements and made decisions to assign them certain codes. Table 1 shows the numbers of posts and comments offering different types of social support. There are many more posts offering informational support (51) than nurturant support (8), but more comments offer nurturant support (525) than informational support (371). Consistent with other health communities, users in the QuitStop forum offer informational and nurturant support that helps others achieve better intervention outcomes (Chuang & Yang, 2010, 2011; Cutrona & Suhr, 1992; Eichhorn, 2008; Finn, 1999). In posts, informational support is offered more frequently than nurturant support, but a larger number of comments give nurturant support than informational support. Similar to the findings of Caplan and Turner (2007), users on QuitStop are comfortable expressing personal narratives and disclosing themselves, which supports the idea that computer-mediated communication can establish comforting conversations. Esteem and emotional support are common types of nurturant support. Consistent with studies of other online health communities (Caplan & Turner, 2007), many support providers show their understanding, encouragement, and empathy for support receivers in the QuitStop forum. Patterns and User Interactions of Social Support Exchange As mentioned earlier, there are two research questions in this part of study. For the first research question, What are the exchange patterns of social support between user pairs on the QuitStop forum, we defined two patterns of social support exchange motivated by the theory of user information behavior and extracted different patterns for each type of social support from our data sample. To answer the second research question, How do users at different quit stages interact with each other to exchange different types of social support, we divided users into different quit stages according to their quit status and built social networks to model the interactions between different users. Patterns of Social Support Exchange In our previous study, we found that requesting, offering, and receiving social support are typical themes of posts and comments in the QuitNet forum (Zhang et al., 2013). Usually, one thread in the community includes one post and several comments with requesting, offering, and receiving social support by different users. Through analyzing these threads, different interactive patterns of user pairs can be extracted. Measurement for exchange patterns. In this study, the definition and extraction of social support exchange are motivated by studies of information exchange. As mentioned previously, practical and orienting information are two main types of information describing information exchange patterns (Savolainen, 1995). In this study, we focus on the exchange of social support, which is narrower in scope than information exchange. The naming and definition for practical and orienting information are based on user-seeking behavior. They reflect the information needs of information seekers (receivers). In this study, we define social support exchange patterns from the perspectives of both support givers and receivers. Similar to practical information seeking, some social media users explicitly express their need for social support (either informational support or nurturant support) and invite their peers to offer support to them. On the other hand, similar to orienting information seeking, some social media users do not have specific questions in mind but they participate in the community and interact with other users. Other users may provide social support to these users without being requested. Two basic patterns of support exchange are defined here in this study, and these are initiated support exchange and invited support exchange, as shown in Figure 1. In the initiated support exchange pattern, social support is offered voluntarily without request. Social support exchanged in this pattern is general and can benefit all users. For example, a user, marked as G, initiated a thread and wrote in the post that... smoking is such an anti-life activity. There really are no redeeming qualities of this activity. It depletes our wallets, it makes us stink and it causes all sorts of health problems.... The theme of this post is offering support. Two users, R 1 and R 2, commented on this post to indicate receiving support, saying very well said. This is going on the My Good Words list I keep in my purse. Thanks and I love the way you make us think. In this case, no user requested social support. The support was initially given by G and received by R 1 and R 2 respectively. In this pattern of initiated support exchange, social support flows from G to R 1, and from G to R 2. In the invited support exchange pattern, social support is requested before being offered. The support seeker actively requests support. Social support exchanged in this pattern is offered to an individual user. For example, a user, marked as R, initiated a thread to request support, writing that I no longer feel like I have to smoke and must say I 568 JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015

Two Pat terns Init iated Support Orienting Informat ion The support receivers do not ask for particular social support. They keep an eye on any information related to their general interests and concerns Invited Support Practical Informat ion With urgent and clear needs of informat ion, the support seekers ask quest ions to get specif ied answers Giver Active Giving Active Receiving Receiver offer support Four User Behaviors request support offer support Receiver Passive Receiving Passive Giving Giver FIG. 1. Exchange patterns and user behaviors associated with social support exchange. look and feel much better, but I still have that desire. Does it ever go away, or is this something I just have to deal with? This post is to request support. Three users, G 1,G 2, and G 3, commented on this post to offer support, writing that I m only at day 22 but can sympathize and send you a few hugs and smiles and your quit is looking great at 60 days,... You are doing excellent and will continue to do so and You are doing great by posting here! So BRAVO to you for that. In this case, the user R actively requests support. G 1,G 2, and G 3 offer support as R expected. In this invited support exchange pattern, social support flows from G 1 to R, from G 2 to R, and from G 3 to R. For each thread with social support interactions, a series of triples <U g,u r,p> are extracted, where U g is the individual user who gives social support, U r is the single receiver of social support, and P is the pattern of social support exchange (initiated support exchange or invited support exchange). For the two examples mentioned previously, five triples could be extracted, which are <G, R 1, Initiated>, <G, R 2, Initiated>, <G 1, R, Invited>, <G 2,R, Invited>, and <G 3, R, Invited>. By examining the social support interactions through all of the collected threads, a triple set is generated, which indicates the frequency of each social support exchange pattern. In this study, we also extract two subsets from the triplet set, one subset for each type of social support (i.e., informational support and nurturant support). In a triple set, a triple with the same elements could appear more than once, which indicates that different social support is exchanged with the same pattern between the same user pair several times. Savolainen (1995) pointed out that seeking practically effective information is active, and seeking orienting information is passive. As we explore the behaviors of both support givers and receivers, the active and passive behaviors associated with giving and receiving support are analyzed, and depicted in Figure 1 and Table 2. Concretely, TABLE 2. Exchange patterns and user behaviors associated with social support exchange. Pattern of social support exchange User behavior of social support exchange Giving behavior Receiving behavior Initiated support exchange Active giving Passive receiving Invited support exchange Passive giving Active receiving four user behaviors associated with social support exchanges are identified, which are active giving, passive giving, active receiving and passive receiving. The behaviors of active giving and passive receiving are extracted from the initiated support exchange pattern, and the behaviors of passive giving and active receiving are extracted from the invited support exchange pattern. Usually, initiated support is offered in posts. The support giver initially starts a thread and offers support to all users. Other users comment on the posts and acknowledge receiving the support in the same thread. In this case, the support giver initiates the support exchange process and is regarded as active. The receivers passively accept the support without requests. From the perspective of information seeking, initiated support is similar to orienting information (Savolainen, 1995). The support receivers do not ask for particular social support. They keep an eye on any information related to their general interests and concerns (Burnett, 2000). In the invited support exchange pattern, a support seeker starts a thread and writes a post to request social support. Other users comment on the post and give corresponding support to the support seeker. In this case, the support seekers actively request and receive social support, and the support givers passively offer social support in response. Invited support is similar to practical JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015 569

information (Savolainen, 1995). With urgent and clear information needs, the support seekers ask questions to get specified answers (Burnett, 2000). For the previous examples, G actively gives social support; R 1 and R 2 passively receive social support; G 1,G 2 and G 3 passively give social support; and R actively receives social support. Results relating to social support exchange patterns. From the data sample, 881 triples of <U g,u r,p> are extracted, which indicate 881 incidents of social support exchange between user pairs. Two hundred twenty-six triples have the P as initiated support exchange, and 655 triples have the P as invited support exchange. Each of the triples is associated with informational support exchange, nurturant support exchange, or both of the types of support exchanges. Five hundred eighty-three triples are associated with informational support exchange, and 497 triples are associated with nurturant support exchange. For informational support, there are 226 triples with the P of initiated-support exchange and 357 triples with the P as invited-support exchange. For nurturant support, there are 16 triples with the P as initiated support exchange and 481 triples with the P as invited support exchange. Counting all posts and comments in our data sample, there are more messages offering nurturant support (533) than informational support (422). However, fewer triples of <U g, U r,p> are extracted for nurturant support (497) than informational support (522), which means that nurturant support is exchanged more frequently between user pairs. The reason for this is that most nurturant support targets one receiver, so each message offering nurturant support can only be received by one user. However, informational support usually targets multiple users, thus each message offering informational support may be received by more than one user. From the results, for both informational support and nurturant support, invited support exchange is more prevalent. However, for informational support, 38.8% is initiated; whereas for nurturant support, only 7.8% is initiated. Comparatively, the pattern of initiated support exchange is more common for the dissemination of informational support. Online heath intervention programs may design different channels to deliver social support. Informational support could be offered and disseminated in an open platform, where a large number of people could receive the support and benefit from it. Nurturant support could be delivered through private channels, where people can more comfortably communicate and receive one-to-one messages. User Quit Status and Quit Stages Quit status is an important feature and health outcome of smoking quitters. In this study, we analyze how users at different quit stages interact with each other to exchange social support. On the QuitNet website, each registered user has a profile page that contains basic personal information. For some users, the quit date can be acquired from their profile pages. In most studies of online intervention programs for smoking cessation, there are only two types of quit status defined for users: smoking and abstinence. Auser is described as either smoking or abstinent in these studies for analysis (Cobb & Graham, 2006; Cobb et al., 2005, 2010). However, quitting smoking is a continuous process that may last for many years. It is difficult to develop a single measurement for all cases (Velicer & Prochaska, 2004). In our previous study, we used a continuous value to represent user quit status (Zhang, Yang, & Li, 2012). For each user, the quit status is defined as the number of days of abstinence. The continuous value of quit status could differentiate recent quitters and old quitters who have been abstinent for a long time. In this study, we use this continuous value to represent quit status. Concretely, the quit status of a user is the number of days that he or she has been abstinent from the selfreported day he or she stops smoking to the day he or she posts the last message on QuitStop in our data sample. We calculate the quit statuses of support givers and receivers identified from the data sample. Velicer et al. (1992) defined five stages of quitting smoking: precontemplation, contemplation, preparation, action, and maintenance. Once a quitter stops smoking, he or she enters the action stage, which is composed of two periods: an early action period and a late action period. In the early action period, the user has been abstinent for 0 to 3 months; in the late action period, the user has been abstinent for 3 to 6 months. After being abstinent for 6 months, the smoking quitter moves into the maintenance stage, which is suggested to be 6 to 60 months after quitting smoking. According to quit statuses, users are categorized into five groups of quit stages in this study. Based on stages since taking actions to quit smoking, the first group is composed of users at the early action stage with a quit status of 0 to 3 months. The second group is composed of users in the late action stage with a quit status of 3 to 6 months. Users in the third and fourth groups are at the early and late maintenance stages with quit statuses of 6 months to 2 years and 2 years to 5 years, respectively. The fifth group is composed of users who have been abstinent for more than 5 years. The quit status of 233 users can be accessed from their profile pages. These users are divided into five groups of different quit stages (Table 3). As shown in Table 3, more than half of the users are at the early action stage of smoking cessation, having been abstinent for less than 3 months. Usually, recent quitters have a strong determination and act more passionately in intervention programs of smoking cessation. They may also log on QuitNet to seek expert counseling. There are recent smoking quitters participating in discussions of QuitNet forum; TABLE 3. User groups of quit stages. Quit status (days) <90 91 180 181 720 721 1800 >1800 # of users 126 28 30 28 21 Percentage 54.1% 12.0% 12.9% 12.0% 9.0% 570 JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015

however, some older smoking quitters also participate in the community. These older smoking quitters offer support to others and receive nurturant support to maintain abstinence. User Interactions associated with Social Support Exchange Some studies built social networks to analyze user interactions in online health communities (Chang, 2009; Cobb et al., 2010). However, they only explored static indicators to measure entire social networks, such as network size, density, clique, and centralization. They did not identify actor positions or analyze interactions between different user groups in the network. In this study, we apply social network analysis to investigate user interactions based on the exchange of different types of social support. We consider user quit status in the network structure and explore interactions between users at different quit stages. As analyzed previously, we extracted 583 triples with informational support exchange and 497 triples with nurturant support exchange from the data sample (<U g,u r,p>). To analyze the social support exchange of users at different quit stages, we remove users whose quit status cannot be acquired and retain user triples with attainable quit statuses of U g and U r for social network analysis. For each type of social support, a directed and valued social network is developed on support exchange between user pairs. In the social network, users are represented as nodes (actors). Each tie connects two actors between whom the specific type of social support is exchanged. The direction of the tie is from the support giver to the support receiver based on extracted triples. The value of the tie is the number of corresponding triples, which indicates the frequency of support exchange between the two actors in different threads. For informational support exchange, we built a social network of 204 nodes and 397 directed ties. For nurturant support exchange, we built a social network of 171 nodes and 341 directed ties. By comparing these two social networks, the exchange patterns of different types of social support can be investigated. Network exposure and blockmodel based on quit stages are used for social network analysis. Network exposure model Development of the network exposure model. Network exposure models can be used to measure the extent to which an actor is exposed to neighbors with a specific behavioral attribute in a social network (Fujimoto, 2012; Valente, 2005). Generally, the network exposure E of an actor i in a social network is defined as: E i wy ij = w where w is the social network weight matrix, y is a vector of actors attributes, and j is a neighbor of actor i. E i considers ij j FIG. 2. Network exposure of users. the attributes of all neighbors of i. Usually, the weight matrix w can be built on different factors, including relation, position, and centrality (Valente, 2005). Network exposure has been applied in many studies of public health. It is calculated for a behavior attribute such as smoking, drinking, syringe sharing, and so on (Fujimoto, Unger, & Valente, 2012; Gyarmathy & Neaigus, 2006). Each element in the attribute vector y has a binary value of 1 or 0, indicating whether the corresponding actor had this behavior. For example, Gyarmathy and Neaigus (2006) analyzed a social network of injecting drug users. For every actor in the network, the personal network exposures for sharing cookers, sharing filters, receptive syringe sharing, distributive syringe sharing, and backloading were respectively calculated using closeness matrix (w). For each of the equipment sharing behaviors, the correlation between the actor s own behavior and his or her personal network exposure was investigated. In this study, we employ continuous values of quit status as elements in the vector y. The weight matrix w is built on values of ties in the network. As the social networks of informational and nurturant support are directed, we calculate two types of network exposure for the actors. For each actor i in a social network, E i(g) is defined as i s network exposure to its support givers. Only actors sending ties to i are selected as neighbors to calculate E i(g). E i(r) is defined as i s network exposure to its support receivers. Actors receiving ties from i are selected as neighbors to calculate E i(r). So, E i(g) represents the quit status of support givers that i is exposed to, and E i(r) represents the quit status of support receivers that i is exposed to. For example, in the network of Figure 2, U 1,U 2,U 3,U 4, and U 5 are neighbors who exchange social support with the user i. U 1,U 2, and U 3 are support givers to i, and U 4 and U 5 are support receivers from i. The weights of ties between i and U 1,U 2,U 3,U 4, and U 5 are w 1,w 2,w 3,w 4, and w 5, respectively. The quit statuses of U 1,U 2,U 3,U 4, and U 5 are q 1,q 2,q 3,q 4, and q 5, respectively. The network exposure to support givers of the user i JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015 571

is calculated based on quit statuses of support givers U 1,U 2, w1 q1+ w2 q2 + w3 q3 and U 3,as Ei ( G) =. The network w1+ w2 + w2 exposure to support receivers of user i is calculated based on quit statuses of support receivers U 4 and U 5, which is w4 q4 + w5 q5 Ei ( G) =. w + w 4 5 Results and analysis of network exposure model. As mentioned previously, two social networks are built: one for informational support exchange and one for nurturant support exchange. For actors in these two social networks, we compare their network exposures. Specifically, for all actors in the informational support network, we calculate their network exposures to givers, and represent them as a vector EG ( ) I. For all actors in the nuturant support network, their network exposures to givers are denoted as EG ( ) N. The mean values of EG ( ) I and EG ( ) N are 945.02 and 575.71 respectively, and a t test indicates that they are significantly different, p < 0.001. Similarly, we represent network exposures to receivers as ER ( ) I and ER ( ) N for actors in the informational support and nurturant support networks, respectively. The mean values of ER ( ) I and ER ( ) N are 105.74 and 207.43, and a t test shows that they are significantly different, p = 0.011. Comparatively, the support givers of informational support have been abstinent for a longer time than support givers of nurturant support. The reason may be that informational support is based on knowledge, advice and personal experiences of support givers. Older quitters are more knowledgeable at providing such information than new quitters. Comparatively, nurturant support does not provide information to help solving any health problem. Nurturant support givers show their concern and care for support recipients, and smoking quitters with any quit statuses can participate in this activity. Generally, users receiving informational support have quitted smoking more recently than users receiving nurturant support. Comparatively, informational support may be more important for recent quitters because information and suggestions are necessary for them to cope with problems. However, for people who have quit for a longer time, nurturant support may be more important because they need encouragement and esteem to retain abstinence. Blockmodel based on quit stages. For a social network, a blockmodel partitions actors into discrete subsets called positions. There are ties within or between positions to represent their relations (Wasserman & Faust, 1994). In this study, we build blockmodels for user quit stages. For each social network, users are partitioned into different positions according to their quit stages. There are five positions for actors in each social network, which are marked by B 1, B 2, B 3, B 4, and B 5. Corresponding to user groups of different quit stages, actors in B 1 are at early action stage of quitting smoking, actors in B 2 are at late action stage, actors in B 3 are at early maintenance stage, actors in B 4 are at late maintenance stage, and actors in B 5 have successfully quit smoking. For each social network in our study, the blockmodel is represented by a 5 5 matrix B = {b kl}, with entries b kl equaling 1 or 0, indicating the presence or absence of a tie from position B k to B 1. Each entry b kl represents a block in the blockmodel. It is a oneblock when b kl = 1, and it is a zeroblock when b kl = 0. To decide the value of b kl, we calculate the density Δ kl in this block (Wasserman & Faust, 1994) using the following formula: Δ kl = g i B k gb k i B k Bk w j B l gb l w g 1 j B k Bk ( ) ij ij, for k l, fork = l where w ij is the value of tie from actor i to actor j in the social network, gb k is the number of actors in the position B k, and gb 1 is the number of actors in the position B 1. Δ kl denotes the average value of ties from actors in B k to actors in B l. To define zeroblocks and oneblocks in the blockmodel, we adopt α density criterion (Wasserman & Faust, 1994), which is defined as: b kl = 0, if Δ < a 1, if Δ a In this study, we designate a =φ Δ max, where φ is the golden ratio, which approximate 0.618, and Δ max denotes the maximum value of elements in the density matrix Δ. For the social networks of informational support and nurturant support, the value is set as 0.0110 and 0.0064 respectively. According to the density matrix Δ and a, the block matrix B is built for each social network. Tables 4 and 5 present the density matrixes Δ and block matrixes B for the social networks of informational support and nurturant support. For social networks of informational support and nurturant support, we draw reduced graphs to present ties within and between positions as shown in Figures 3 and 4. In a reduced graph, positions are represented as nodes and ties between positions are represented as arcs. There is an arc between positions of B k and B l if b kl = 1, and there is no arc between B k and B l if b kl = 0. From the reduced graph of the informational support network in Figure 3, informational support flows from users at late quit stages to users at early quit stages. Users at early and late action stages (group 1 and group 2) only receive informational support from users at later quit stages. Users at late maintenance stage (group 4) and after the maintenance stage (group 5) only offer informational support to others. Users at early maintenance stage (group 3) give information to users at the early action stage (group 1), and also receive information from users who have completed smoking cessation (group 5). Informational support is usually offered by users at late quit stages that have 572 JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015

TABLE 4. Density matrix and block matrix of informational support network. Density matrix Block matrix B 1 B 2 B 3 B 4 B 5 B 1 B 2 B 3 B 4 B 5 B 1 0.0101 0.0017 0.0032 0.0014 0.0011 B 1 0 0 0 0 0 B 2 0.0088 0.0053 0.0048 0.0000 0.0000 B 2 0 0 0 0 0 B 3 0.0116 0.0036 0.0069 0.0024 0.0016 B 3 1 0 0 0 0 B 4 0.0133 0.0115 0.0071 0.0106 0.0034 B 4 1 1 0 0 0 B 5 0.0155 0.0170 0.0175 0.0051 0.0095 B 5 1 1 1 0 0 TABLE 5. Density matrix and block matrix of nurturant support network. Density matrix Block matrix B 1 B 2 B 3 B 4 B 5 B 1 B 2 B 3 B 4 B 5 B 1 0.0099 0.0031 0.0026 0.0031 0.0023 B 1 1 0 0 0 0 B 2 0.0077 0.0066 0.0024 0.0013 0.0000 B 2 1 1 0 0 0 B 3 0.0103 0.0048 0.0069 0.0024 0.0048 B 3 1 0 1 0 0 B 4 0.0074 0.0038 0.0024 0.0066 0.0085 B 4 1 0 0 1 1 B 5 0.0098 0.0051 0.0048 0.0017 0.0071 B 5 1 0 0 0 1 FIG. 3. Reduced graph of informational support network. FIG. 4. Reduced graph of nurturant support network. plenty of experiences and knowledge, and received by users who have been abstinent for shorter time. Figure 4 indicates that similar to informational support, nurturant support is offered by users at late quit stages (groups 2, 3, 4, and 5) to those at the early action stage (group 1). But differently, nurturant support is also frequently exchanged between users within the same quit stage. Offering nurturant support shows understanding and caring. Users with similar quit statuses are likely to better understand each other because they share similar experiences. Moreover, users at the same quit stage have been supporting each other because they started quitting at the same time. As a result, they maintain a supporting group of the same quit status, and other users who quit smoking earlier and later seem to be less active in interacting with them but find support from their own group. Conclusion In this study, we analyzed types and frequencies of informational and nurturant support exchanged in the QuitStop forum. The exchange patterns of social support and user exchange behaviors were extracted and analyzed. User interactions with informational and nurturant support exchange were explored within and between users in different groups of quit stages. Our results show that informational support is exchanged more frequently between user pairs, although there are a larger number of messages giving nurturant support in the QuitStop forum. Initiated-support exchange and invitedsupport exchange are defined as two basic exchange patterns of social support. Correspondingly, four exchange behaviors are extracted for users, including active giving, passive giving, active receiving, and passive receiving. When JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015 573

comparing social networks of informational support and nurturant support, it is found that givers of informational support have been abstinent for a longer time than givers of nurturant support, and receivers of informational support have been abstinent for a shorter time than receivers of nurturant support. Usually, informational support is offered by users at late quit stages to users at early quit stages. Nurturant support is also exchanged among users within the same quit stage. There are several limitations to this study. First, it only focuses on data in a fixed period, but the content of messages in the QuitStop forum may change over time. Second, some users do not provide quit statuses in their profile pages. Our study eliminates these users, which may limit the scope of the analysis. Third, the types and exchange patterns of social support may be different in other online communities for smoking cessation. In the future, we will apply data mining technology to analyze message content automatically, which will allow us to process a large amount of data. We will also study communities for smoking cessation in other social media channels, such as Facebook. Studies of social support and support exchange behavior in online heath communities could help us better understand user motivations, actions and smoking cessation outsomes in online intervention programs. By studying exchange patterns and user interactions in relation to social support in health community, we could build recommendation models based on message content and user health features. In the future, systems for social support delivery could make recommendations according to message content, user interests and user associations. As a result, improved services could be provided to achieve satisfactory intervention outcomes. References Agaku, I., King, B., & Dube, S.R. (2012). Current cigarette smoking among adults United States, 2011. Morbidity and Mortality Weekly Report: Centers of Disease Control and Prevention, 16, 889 894. Ahmed, O.H., Sullivan, S.J., Schneiders, A.G., & McCrory, P. (2010). isupport: Do social networking sites have a role to play in concussion awareness? Disability and Rehabilitation, 32(22), 1877 1883. An, L.C., Schillo, B.A., Saul, J.E., Wendling, A.H., Klatt, C.M., Berg, C.J.,... Luxenberg, M.G. (2008). Utilization of smoking cessation informational, interactive, and online community researches as predictors of abstinence: Cohort study. Journal of Medical Internet Research, 10, e55. Bender, J.L., Jimenez-Marroquin, M.-C., & Jadad, A.R. (2011). Seeking support on Facebook: A content analysis of breast cancer groups. Journal of Medical Internet Research, 13(1), e16. Burnett, G. (2000). Information exchange in virtual communities: A typology. Information Research, 5(4), paper82. Caplan, S.E., & Turner, J.S. (2007). Bringing theory to eesearch on computer-mediated comforting communication. Computers in Human Behavior, 23(2), 985 998. Chang, H.-J. (2009). Online supportive interactions: Using a network approach to examine communication patterns within a psychosis social support group in Taiwan. Journal of the Association for Information Sciences and Technology, 60(7), 1504 1517. doi:10.1002/asi.21070 Chuang, K., & Yang, C.C. (2010). Social support in online healthcare social networking. Paper presented at the iconference 2010, Urbana- Champaign, Illinois. Chuang, K.Y., & Yang, C.C. (2011). Helping you to help me: Exploring supportive interaction in online health community. Paper presented at the ASIS&T 2010, Pittsburgh, PA. Cobb, N.K., & Graham, A.L. (2006). Characterizing Internet searchers of smoking cessation information. Journal of Medical Internet Research, 8(3), e17. Cobb, N.K., Graham, A.L., & Abrams, D.B. (2010). Social network structure of a large online community for smoking cessation. American Journal of Public Health, 100, 1282 1289. Cobb, N.K., Graham, A.L., Papandonatos, G., & Abrams, D.B. (2005). Initial evaluation of a real-world Internet smoking cessation system. Nicotine and Tobacco Research, 7, 207 216. Cutrona, C.E., & Suhr, J.A. (1992). Controllability of stressful events and satisfaction with spouse support behaviors. Communication Research, 19(2), 154 174. doi:10.1177/009365092019002002 Eichhorn, K.C. (2008). Soliciting and providing social support over the Internet: An investigation of online eating disorder support groups. Journal of Computer-Mediated Communication, 14(1), 67 78. doi:10.1111/j.1083-6101.2008.01431.x Eysenbach, G. (2008). Medicine 2.0: Social networking, collaboration, participation, apomediation, and openness. Journal of Medical Internet Research, 10(3), e22. Finn, J. (1999). An exploration of helping processes in an online self-help group focusing on issues of disability. Health and Social Work, 24(3), 220 231. Fujimoto, K. (2012). Using mixed-mode networks to disentangle multiple sources of social influence. Paper presented at the Social Computing, Behavioral Cultural Modeling and Prediction, College Park, MA. Fujimoto, K., Unger, J.B., & Valente, T.W. (2012). A network method of measuring affiliation-based peer influence: Assessing the influences of teammates smoking on adolescent smoking. Child Development, 83(2), 442 451. doi:10.1111/j.1467-8624.2011.01729.x Graham, A.L., Cobb, N.K., Raymond, L., Still, S., & Young, J. (2007). Effectiveness of an Internet-based worksite smoking cessation intervention at 12 months. Journal of Occupational and Environmental Medicine, 49, 821 828. Greene, J.A., Choudhry, N.K., Kilabuk, E., & Shrank, W.H. (2011). Online social networking by patients with diabetes: A qualitative evaluation of communication with Facebook. Journal of General Internal Medicine, 26(3), 287 292. Gyarmathy, V.A., & Neaigus, A. (2006). The effect of personal network exposure on injecting equipment sharing among IDUs in Budapest, Hungary. International Network for Social Network Analysis, 27(1), 25 38. Haythornthwaite, C., & Wellman, B. (1998). Work, friendship, and media use for information exchange in a networked organization. Journal of the American Society for Information Science, 49(12), 1101 1114. Hughes, B., Joshi, I., & Wareham, J. (2008). Health 2.0 and Medicine 2.0: Tensions and controversies in the field. Journal of Medical Internet Research, 10(3), e23. Hwang, K.O., Ottenbacher, A.J., Green, A.P., Cannon-Diehl, M.R., Richardson, O., Bernstam, E.V., & Thomasa, E.J. (2011). Social support in an Internet weight loss community. International Journal of Medical Informatics, 79(1), 5 13. doi:10.1016/j.ijmedinf.2009.10.003 Lichtenstein, E., Glasgowb, R.E., & Abrams, D.B. (1986). Social support in smoking cessation: In search of effective interventions. Behavior Therapy, 17(5), 607 619. Mo, P.K.H., & Coulson, N.S. (2008). Exploring the communication of social support within virtual communities: A content analysis of messages posted to an online HIV/AIDS support group. CyberPsychology & Behavior, 11(2), 371 374. doi:10.1089/cpb.2007.0118 Murray, R.P., Johnston, J.J., Dolce, J.J., Lee, W.W., & O Hara, P. (1995). Social support for smoking cessation and abstinence. The Lung Health Study Addictive Behaviors, 20(2), 159 170. 574 JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015

National Institute of Drug Abuse. (2012). What are the medical consequences of tobacco use? Retrieved from http://www.drugabuse.gov/ publications/research-reports/tobacco-addiction/what-are-medical -consequences-tobacco-use Pirie, P.L., Rooney, B.L., Pechacek, T.F., Lando, H.A., & Schmid, L.A. (1997). Incorporating social support into a community-wide smokingcessation contest. Addictive Behaviors, 22(1), 131 137. Russell, M.A.H., Wilson, C., Taylor, C., & Baker, C. (1979). Effect of general practitioners advice against smoking. British Medical Journal, 2(6184), 231 235. Savolainen, R. (1995). Everyday life information seeking: Approaching information seeking in the context of way of life. Library & Information Science Research, 17(3), 259 294. Shahab, L., & McEwen, A. (2009). Online support for smoking cessation: A systematic review of the literature. Addiction (Abingdon, England), 104, 1792 1804. Shumaker, S.A., & Brownell, A. (1984). Toward a theory of social support: Closing conceptual gaps. Journal of Social Issues, 40(4), 11 36. Valente, T.W. (2005). Network models and methods for studying the diffusion of innovations. In P.J. Carrington, J. Scott, & S. Wasserman (Eds.), Models and methods in social network analysis (pp. 98 116). New York: Cambridge University Press. Van De Belt, T.H., Engelen, L.J., Berben, S.A., & Schoonhoven, L. (2012). Definition of health 2.0 and medicine 2.0: A systematic review. Journal of Medical Internet Research, 12(2), e18. Velicer, W.F., & Prochaska, J.O. (2004). A comparison of four self-report smoking cessation outcome measures. Addictive Behaviors, 29(1), 51 60. Velicer, W.F., Prochaska, J.O., Rossi, J.S., & Snow, M.G. (1992). Assessing outcome in smoking cessation studies. Psychological Bulletin, 111(1), 23 41. doi:10.1037/0033 2909.111.1.23 Viera, A.J., & Garrett, J.M. (2005). Understanding interobserver agreement: The kappa statistic. Family Medicine, 37(5), 360 363. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. New York: Cambridge University Press. White, M., & Dorman, S.M. (2001). Receiving social support online: Implications for health education. Health Education Research, 16(6), 693 707. Wright, K.B., & Bell, S.B. (2003). Health-related support groups on the Internet: Linking empirical findings to social support and computermediated communication theory. Journal of Health Psychology, 8(1), 39 54. doi:10.1177/1359105303008001429 Zhang, M., Yang, C.C., & Li, J. (2012). A comparative study of smoking cessation intervention programs on social media. Paper presented at the Proceedings of SBP2012, College Park, MD. Zhang, M., Yang, C.C., & Gong, X. (2013). Social support and exchange patterns in an online smoking cessation intervention program. Paper presented at the IEEE International Conference on Healthcare Informatics, Philadelphia, PA. JOURNAL OF THE ASSOCIATION FOR INFORMATION SCIENCE AND TECHNOLOGY March 2015 575