شما شبکه تان را انتخاب کنید: ترجیحات اجتماعی در جامعه سلامت آنلاین

Abstract A growing number of online health communities offer individuals the opportunity to receive information, advice, and support from peers. Recent studies have demonstrated that these new online contacts can be important informational resources, and can even exert significant influence on individuals' behavior in various contexts. However little is known about how people select their health contacts in these virtual domains. This is because selection preferences in peer networks are notoriously difficult to detect. In existing networks, unobserved pressures on tie formation – such as common organizational memberships, introductions to friends of friends, or limitations on accessibility – may mistakenly be interpreted as individual preferences for interacting/not interacting with others. We address these issues by adopting a social media approach to studying network formation. We study social selection using an in vivo study within an online exercise program, in which anonymous participants have equal opportunities for initiating relationships with other program members. This design allows us to identify individuals' preferences for health contacts, and to evaluate what these preferences imply for members' access to new kinds of health information, and for the kinds of social influences to which they are exposed. The study was conducted within a goal-oriented fitness competition, in which participation was greatest among a small core of active individuals. Our results show that the active participants displayed indifference to the fitness and exercise profiles of others, disregarding information about others’ fitness levels, exercise preferences, and workout experiences, instead selecting partners almost entirely on the basis of similarities on gender, age, and BMI. Interestingly, the findings suggest that rather than expanding and diversifying their sources of health information, participants' choices limited the value of their online resources by selecting contacts based on characteristics that are common sources of homophily in offline relationships. In light of our findings, we discuss design principles that may be useful for organizations and policy makers trying to improve the value of participants' social capital within online health programs.

Introduction Over the last decade, the Internet has become an increasingly important domain for health (Fogel et al., 2002, Thackeray et al., 2008, Chou et al., 2009, Hawn, 2009, McNab, 2009, Pampel et al., 2010 and Salathe and Khandelwal, 2011). Recent surveys of Internet use for health estimate that 23% of US patients living with chronic illnesses, such as high blood pressure, diabetes, heart conditions, or cancer, use peer-to-peer online resources to help support their medical treatment and discovery processes (Fox, 2011). Even more striking, among populations with chronic diseases who are seeking “practical advice for coping with day-to-day health situations,” patients were overall more likely to seek out informal sources of peer-to-peer assistance than consult with medical professionals (Fox, 2011). As this trend increases, social scientists interested in the social dimensions of health are increasingly concerned with characterizing the online social networks that people use. In particular, recent research has begun to explore the question of how online social networks influence the spread of health information and behavior change (White and Dorman, 2001, Japuntich et al., 2006, Hawn, 2009, Centola, 2010 and Centola, 2011). Centola, 2010 and Centola, 2011 uses controlled online experiments to demonstrate the effects of both network structure and homophily in promoting the contagious spread of health behaviors. However, relatively little is known about how online health communities form, and what kinds of networks people “create” in these often anonymous environments (Wellman and Hampton, 1999 and Wellman, 2001). Given the variety of online health contexts for information exchange and influence (Fox, 2011), we focus our study on the increasingly popular domain of online fitness programs, which are designed to promote exposure to health information and increased fitness through peer to peer interaction (Centola, 2013). The popular bromide that people select ties “homophilously” – i.e., based on preferences for others with similar characteristics – was formally introduced over a half a century ago by Lazarsfeld and Merton (1954). The goal of their study was to determine why strong correlations were regularly observed between people with specific demographic characteristics and those exhibiting certain beliefs, attitudes and behaviors. Their explanatory strategy was first to show that people with similar demographic traits selectively formed ties to one another, and then to show that people who were socially connected influenced each other's beliefs. However, while they found that friends influence friends, support for homophily in tie formation (henceforth “choice homophily”) was variable, occurring in some situations, but not in others. As Lazarsfeld and Merton put it, “[T]he problem of selection [is] not adequately formulated by the familiar and egregiously misleading question: When it comes to close friendships, do birds of a feather actually flock together? Rather it is a more complex problem of determining the degree to which such selectivity varies for different kinds of social attributes, how it varies within different kinds of social structure, and how such selective patterns come about.” (Lazarsfeld and Merton, 1954:18). A large literature has since emerged on homophily in social relations. As the terminology has evolved, the term “homophily” has now come to refer to the observed population-level regularity that people within a community tend to be socially connected to others who are more similar to themselves than would be expected by random chance (Coleman, 1958). Researchers in this tradition have identified several, very different, mechanisms that can generate this regularity. The most obvious mechanism, initially identified by Lazarsfeld and Merton (1954), is “choice homophily”: People preferentially make ties to others who are similar to themselves. However, inferring individual choice homophily from population level homophily risks running afoul of the ecological fallacy since choice homophily can be completely absent at the individual level even when populations exhibit high levels of observed homophily. This disjuncture between individual behavior and collective outcome is due to the variety of other mechanisms that can produce similar population-level patterns. For instance, a second mechanism, which has recently been widely discussed in the literature on networks and health is the process of social influence (McPherson and Smith-Lovin, 1987, Popielarz and McPherson, 1995 and Christakis and Fowler, 2007). While homophily on some traits, like race and gender, cannot emerge through social influence, interpersonal correlations on other health characteristics, such as obesity, heart disease, or smoking, can be linked to social influences between contacts (Christakis and Fowler, 2007). Recent research has emphasized these mechanisms as competing explanations for patterns of observed homophily on obesity, giving rise to a dichotomization of the literature on homophily and health into the competing positions of “social influence” vs. “choice homophily”. However, the scope of the problem of the origins of interpersonal correlations on health characteristics is actually much broader. Other explanatory mechanisms, which operate at the level of social structure rather than at the level of the individual or the dyad, are equally important factors in the emergence of correlations in social networks. For instance, organizational and institutional sorting processes at schools and workplaces typically determine the set of potential social contacts that an individual is exposed to within a given context (Feld, 1982, McPherson et al., 2001, Moody, 2001, Ruef et al., 2003 and Bertrand and Mullainathan, 2004). These structures often implicitly “preselect” individuals into homophilous groups (by race, class, gender, educational background, and so forth), thereby eliminating opportunities for heterophilous tie formation (Blau, 1977, Blau and Schwartz, 1984 and McPherson and Smith-Lovin, 1987). These social processes can force homophilous tie formation even when the members of a population lack any particular preference for homophilous ties (McPherson and Smith-Lovin, 1987). Similarly, homophily can also emerge from the process of friends introducing friends to one another, or “triadic closure” in social networks (Kossinets and Watts, 2009). For instance, if a pair of friends, A and B, are homophilous, and B also has a friend C with whom she is similarly homophilous, then A may become friends with C by virtue of B's introduction. A homophilous tie between A and C can thus form by virtue of social structure, without A having any particular interest in “finding” someone similar to herself. More importantly, homophily can emerge in social networks even when individuals consciously prefer heterophily. In friendship networks, competitive preferences to form ties with the most healthy, most physically attractive or most successful individuals can create patterns of observed homophily via the endogenous exclusion of low-health or low-attractiveness members of the population, who are then forced to form ties with one another (Ali et al., 2012). Crosnoe et al. (2008) show that this mechanism of social exclusion can generate explicit patterns of homophily on obesity. More generally, across a broad array of social characteristics in which actors have “aspirational” preferences to form ties to “desirable” alters, patterns of systematic exclusion of the less desirable individuals can lead to the false appearance of choice homophily in domains such as health (Ali et al., 2012), online dating (Hitsch et al., 2010), marriage markets (Mare, 1991 and Kalmijn, 1994), scientific collaboration (Dahlander and McFarland, 2013), and residential segregation (Van de Rijt et al., 2009). Finally, selection on an unobserved trait may be mistaken for a selective preference for a correlated trait that is observed (Yamaguchi and Kandel, 1993 and Kalmijn and Vermunt, 2007). For example, as fitness is related to age, a tendency for individuals to choose ties to others of a similar fitness observed in a study that measures subject fitness but not age may in actuality represent an unobserved tendency for subjects to select on the basis of age. Consequently, in evaluating the implications of social networks for health communications, observed patterns of homophily on health characteristics do not provide clear evidence for individuals' selective preferences for health contacts. These issues become particularly salient in contexts where the selections that people make are typically sought after as informational or motivational resources. Within online fitness programs, the selection of health contacts explicitly serves the goal of providing a reference point for achievement within the program, and establishing a standard against which to evaluate success. Our goal is to determine how people select ties in these contexts, and thereby to understand how social selection both frames the scope of participants' exposure to novel and productive health information, and provides a motivational frame for future health. In particular, we are interested in whether participants select online health contacts who have levels of fitness and “status” on health characteristics that suggest aspirational goals in establishing ties, or whether ties are formed primarily to contacts with similar levels of fitness as themselves. This difference between “aspirational” tie formation, vs. “homophilous” tie formation is important for understanding the ultimate impact of online health networks on participants' health. One of the primary incentives for forming contacts within an online health program is because they provide a means for discovering new ways to lead a healthier lifestyle by providing exposure to new health information. Another reason that participants form ties is because they are seeking connections with health “leaders,” whom they may not have contact with in their day-to-day routines, but who can provide peer-guidance on improving their fitness and lifestyle. However, these goals are primarily served only if the ties participants make actually connect them to people who extend their informational and motivational exposure. Thus, we emphasize that our goal is not to explore the familiar tension between homophily and contagion as competing explanations for observed correlations between network ties and individual traits. Rather, we are interested in people's selection patterns on relatively stable health characteristics, which determine the kinds of informational and motivational exposure that the members of these online communities receive. The goal of the present study is to clearly identify individuals' preferences in forming fitness-specific online health contacts. There are important new methods (Steglich et al., 2010) that have been developed for identifying selection behavior in complex observational datasets (Mercken et al., 2009, Wimmer and Lewis, 2010 and Lewis et al., 2012). Each of these methods is designed to solve problems of causality and identification that are caused by uncontrolled factors such as endogeneity, unobserved heterogeneity, and exogenous influence (Aral et al., 2009 and Shalizi and Thomas, 2011). Our study was developed to eliminate these factors at the outset by using a controlled, randomized design, implemented within an existing online health program. Recently, many scholars have used randomization and experimental controls to eliminate the large number of factors that can prevent the identification of social influence in network contexts (Centola, 2010 and Centola, 2011; Bond et al., 2012). By contrast, our design eliminates social influence, as well as the confounding factors of organizational grouping, hierarchical exclusion, friends introducing friends and exogenous influence, in order to isolate and identify individual preferences in tie selection. Our approach to studying the process of network formation follows that of Lazarsfeld and Merton (1954), who argue that the problem of emergent patterns of association in social networks is not one of determining a general model of choice dynamics. Rather, in different contexts, different selection criteria guide individuals' preferences. Motivated by the growing importance of social media in peer-to-peer informational exchange and health related decision-making (Fox, 2011), this study shows how individual selection can shape the active communication channels in an online fitness program, and what this implies for participants' access to health information and social influence in this domain.1

Conclusion In Lazarsfeld and Merton's (1954) study of tie selection, they distinguish between two basic kinds of choice homophily: value homophily, based on similar attitudes, beliefs, and behaviors, and status homophily, based on nominal status characteristics, such as class, gender, or race ( McPherson et al., 2001). Our results suggest that attitudinal factors, such as aspirational interests (i.e., “goals” for the fitness program) and health attitudes (i.e., “diet preferences”), were not primary considerations for tie selection among the subjects in our study. This is perhaps explained by the fact that the focus of the fitness program was particularly tailored to achieving weekly exercise goals. Yet, participants could also have selected on health-based ranking (i.e., “fitness level”), or a number of behavioral factors that were specifically relevant to the goals of the program, such as exercise minutes, exercise intensity, and favorite exercise. Value-based homophily on any of these factors might be motivated by participants' interest in finding relevant comparisons ( Festinger, 1954) for evaluating their behavior against others with similar exercise routines and health habits. By connecting to people with similar minutes, intensity, fitness level, or exercises, participants could establish a benchmark with peers whose fitness profiles were similar to their own, and whose exercise goals and aspirations would also be similar. Yet, our results indicate that participants did not select ties based on any of these characteristics. Rather, they seem to have mostly ignored value homophily, and selected ties overwhelmingly based on status characteristics ( McPherson et al., 2001). In offline health networks, where participants have no pre-existing relationships, status characteristics, like age, gender, and BMI are intuitive selection factors because they are readily observable features, which can easily be used to infer a potential partner's relevance for one's own exercise behavior. They provide a simple and effective heuristic for selecting health contacts in the absence of easily identifiable traits, such as regular exercise intensity, minutes, or activities. However, to facilitate participants' ability to make the most relevant connections, our program explicitly revealed these fitness characteristics (i.e., intensity, minutes, and activities), which were specifically targeted to the task of increasing and maintaining participants' exercise levels. Participants' choices to instead select homophilously on familiar demographic and health traits suggests that not only did individuals not select “health leaders” or “desirable individuals,” but they did not even select the individuals who might form the most apt comparison group for evaluating their weekly progress toward the program's goals. At both the high and low scales of health status, participants reproduced the basic forms of status homophily that might otherwise be created by social exclusion and institutional sorting (McPherson and Smith-Lovin, 1987 and Crosnoe et al., 2008). The endogenous, choice-based emergence of these status preferences raises the curious question of whether individuals chose these characteristics simply because they are already familiar with these attributes from their offline experiences, or whether they “intrinsically” prefer connecting to others with these characteristics. While individuals may indeed prefer homophily along observable characteristics, a longstanding sociological observation suggests that social structures frame individual expectations (Marx, 1977 [1867], Weber, 1978 [1922] and Berger et al., 1977). Our results may thus suggest that the freedom of the online space may be bounded in distinct ways by the social traditions that precede it. Thus, while our fitness study was strategically constructed to eliminate all the constraints on tie choice that normally limit opportunities for interaction across social categories, participants nonetheless deliberately avoided selecting alters with significant differences on these categories. The conclusion from our findings is that in the online fitness context, people prefer to make ties to the “devil they know.” By selecting ties based on familiar social characteristics, participants may unintentionally limit their available social capital, and restrict their opportunities for finding new health information from sources that they are not normally exposed to. For organizations interested in using online health networks to promote informational access and greater social exposure, this suggests that they may need to provide participants with an interest in forming ties that cut across traditional boundaries. For instance, postings that encourage heterophilous or activity-specific ties may increase participants' likelihood of making these connections. Further, promoting tie formation across traditional status boundaries may require incentivizing higher health-status people to initiate tie formation, or perhaps providing program goals that are particularly tailored to encouraging participants to make social ties to health buddies whom they would not otherwise meet. Introducing these goals and incentives can both increase people's awareness of potential contacts, and highlight their salience for those outside their status community. Both of which may be necessary to achieve the goal of increasing participants' exposure to new sources of health information and influence. Finally, these incentives may also increase participation in health programs overall by highlighting the value of online social capital for discovering health resources that are less easily found in contexts with less transparent information, and higher barriers to introduction.