Social Impacts on Consumer Behavior .

er behaIntern. J. of Research in Marketing 21 (2004) 241 – 263 www.elsevier.com/locate/ijresmar

A social influence model of consumer participation in network- and small-group-based virtual communities
Utpal M. Dholakiaa,*, Richard P. Bagozzia, Lisa Klein Pearob a Rice University, Jesse H. Jones Graduate School of Management, 6100 Main Street, 314 Herring Hall-MS 531, Houston, TX 77005, USA b Cornell University, Cornell School of Hotel Administration, Ithaca, NY 14853, USA Received 8 May 2003; received in revised form 1 September 2003; accepted 5 December 2003

Abstract We investigate two key group-level determinants of virtual community participation—group norms and social identity—and consider their motivational antecedents and mediators. We also introduce a marketing-relevant typology to conceptualize virtual communities, based on the distinction between network-based and small-group-based virtual communities. Our survey-based study, which was conducted across a broad range of virtual communities, supports the proposed model and finds further that virtual community type moderates consumers’ reasons for participating, as well as the strengths of their impact on group norms and social identity. We conclude with a consideration of managerial and research implications of the findings. D 2004 Elsevier B.V. All rights reserved.
Keywords: Virtual communities; Internet marketing; Consumer behavior; Electronic commerce; We-intentions

A web of glass spans the globe. Through it, brief sparks of light incessantly fly, linking machines chip to chip, and people face to face (Cerf, 1991, p. 72) 1. Introduction Marketers have become more and more interested in learning about, organizing, and managing
* Corresponding author. Tel.: [phone]; fax: [phone]. E-mail addresses: [e-mail] (U.M. Dholakia)8 [e-mail] (R.P. Bagozzi)8 [e-mail] (L.K. Pearo). 0167-8116/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ijresmar.2003.12.004

virtual communities on their internet venues (Bagozzi & Dholakia, 2002; Balasubramanian & Mahajan, 2001). Such an interest stems not only from their ability to influence members’ choices, and to rapidly disseminate knowledge and perceptions regarding new products (e.g., Dholakia & Bagozzi, 2001), but also from the numerous opportunities to engage, collaborate with, and advance customer relationships actively in such forums. In the current research, consistent with the prevailing view (e.g., Rheingold, 2002; Wellman & Gulia, 1999), virtual communities are viewed as consumer groups of varying sizes that meet and

242

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

interact online for the sake of achieving personal as well as shared goals of their members. Researchers have employed various theories such as social network analysis (e.g., Wellman & Gulia, 1999), life cycle models (e.g., Alon, Brunel, & Schneier Siegal, 2004), and motivational theories (e.g., Bagozzi & Dholakia, 2002) for studying virtual communities, examining such issues of marketing relevance as what draws participants to such communities, what they are used for, and how they influence the subsequent knowledge, opinions, and behaviors of participants. A common theme underlying many of these investigations is to better understand the nature and role of the social influence exerted by the community on its members (Alon et al., 2004; Postmes, Spears, & Lea, 2000; see Dholakia & Bagozzi, 2004 for a review). Bagozzi and Dholakia’s (2002, hereafter B&D) study provides a useful starting point for framing our discussion since it adopted a marketing lens to identify two key social influence variables, group norms, and social identity, impacting virtual community participation. Using the social psychological model of goal-directed behavior (e.g., Perugini & Bagozzi, 2001) and social identity theory (e.g., Tajfel, 1978) as underlying frameworks, B&D conceptualized participation in virtual chat rooms as bintentional social actionQ involving the group. They modeled participants’ bwe-intentions,Q i.e., intentions to participate together as a group, to be a function of individual (i.e., attitudes, perceived behavioral control, positive, and negative anticipated emotions) and social determinants (i.e., subjective norms, group norms, and social identity). Despite the insights derived from their theorizing and empirical analysis, the following two limitations of the B&D framework are noteworthy and provide the motivation for the present research. First, B&D viewed the social influence variables to be exogenous constructs in their framework, i.e., they did not consider the antecedents of either group norms or social identity, two important predictors in their model. Understanding the antecedents of social influence is important since it is likely to provide significant managerial guidance regarding how to make virtual communities useful and influential for their participants. Second, B&D’s empirical study was limited to virtual chat rooms and did not consider or

elaborate on the distinctions between different types of virtual communities or their implications for marketers. Indeed, marketers have narrowly conceived of virtual communities as commercially sponsored bulletin-boards or chat rooms on company websites (e.g., Thorbjørnsen, Supphellen, Nysveen, & Pedersen, 2002; Williams & Cothrel, 2000; cf. Catterall & Maclaran, 2001). Addressing these limitations, our objectives in the present research are three-fold. First, building upon the B&D (2002) framework, we develop a social influence model of consumer participation in virtual communities. Like B&D (2002), the central constructs in our model are group norms and social identity, but unlike B&D, we not only consider the antecedents of social influence, but also include such mediating constructs as mutual agreement and accommodation among group members. We draw upon existing communication research regarding the motivational drivers of media use (e.g., Flanagin & Metzger, 2001; McQuail, 1987), philosophical writings on group action (Bratman, 1997; Tuomela, 1995) and social psychological research on social identity (e.g., Ellemers, Kortekaas, & Ouwerkerk, 1999; Tajfel, 1978) to develop our theoretical model. Second, we present a marketing-relevant typology to conceptualize virtual communities within a firm’s internet venues that makes and elaborates on the distinction between network- and small-group-based virtual communities. In doing so, we also make the conceptual distinction between the venue where the virtual community meets, and the networks or small groups of individuals constituting the community. In our survey-based study conducted across a broad range of virtual communities, our proposed model is supported. We also find virtual community type— network- or small-group-based—to be a moderator, influencing both, the reasons why members participate, and the strengths of their impacts on group norms and social identity. Finally, we consider the implications of our framework and the distinction made between network- and small-group-based virtual communities, for marketing practice. We elaborate on some of the trade-offs that may be involved, and on issues that must be considered, when organizing and managing these two types of virtual communities effectively.

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

243

Our objective in doing so is not only to provide guidance to marketers in managing their internet venues, but also to stimulate academic researchers to consider these issues in depth.

2. Theoretical background and hypothesis In developing a theory of consumer participation in virtual communities, one approach has been to postulate that a number of individual-level and group-level variables act separately to influence the consumer’s desires, we-intentions, and ultimately his or her participation in the community (B&D; see also Bagozzi, 2000). An alternative perspective, which builds upon this view, and one that we adopt in this article, is that, whereas both individual-level and group-level variables are important drivers of virtual community participation, at least some of the individual-level variables are antecedents to group-level variables, which in turn influence participation. Such a perspective is consistent with social identity theory

(Hogg & Abrams, 1988) as well as recent research on online social interactions (McKenna & Bargh, 1999) and views group influences on the participant to stem from an explicit understanding that group membership yields beneficial outcomes. Using this approach, we start with a set of individual-level motives that help explain why consumers participate in virtual communities. To the extent that these motives can be satisfied through participation, the community should exert influence on its members. Our theoretical model (see Fig. 1) is developed in detail next. 2.1. Individual motives for participation in the virtual community To understand the motives of virtual community participants, we draw upon the well-established uses and gratifications paradigm, originally developed and employed by communications researchers to understand people’s motivations for using different media (e.g., Flanagin & Metzger, 2001; McQuail, 1987). This research has shown that individuals often seek

Fig. 1. A social influence model of virtual community participation.

244

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

out media in a goal-directed fashion to fulfill a core set of motivations, which are also helpful in understanding why consumers might participate in virtual communities. Of special relevance from a marketing perspective, informational value is one that the participant derives from getting and sharing information in the virtual community, and from knowing what (presumably credible) others think. We also included instrumental value that a participant derives from accomplishing specific tasks, such as solving a problem, generating an idea, influencing others regarding a pet issue or product, validating a decision already reached or buying a product, through online social interactions (e.g., Hars & Ou, 2002; McKenna & Bargh, 1999). These objectives are all instrumental in the sense that they are usually defined prior to participation and facilitate achievement of specific end-state goals (Bagozzi & Dholakia, 1999). Although informational and instrumental values tend to be viewed as distinct by communication researchers (e.g., Flanagin & Metzger, 2001), it is perhaps more appropriate to view them as constituents of a single purposive value construct from a marketing perspective, which we define as bthe value derived from accomplishing some pre-determined instrumental purposeQ (including giving or receiving information) through virtual community participation. Indeed, the empirical analyses reported below support this reformulation. The second type of value, self-discovery, involves understanding and deepening salient aspects of one’s self through social interactions. One aspect of selfdiscovery is to interact with others so as to obtain access to social resources and facilitate the attainment of one’s future goals (McKenna & Bargh, 1999). Another aspect of self-discovery is that such interactions may help one to form, clearly define and elaborate on one’s own preferences, tastes, and values. Whereas purposive value relates to utilitarian concerns connecting one’s self to external objects or issues, self-discovery focuses on intrinsic concerns, constituted by or embedded in the self itself. But both these values are self-referent, i.e., they primarily involve and refer to one’s personal self. The next two values we included have more to do with others, i.e., other members of the virtual community. Maintaining interpersonal connectivity

refers to the social benefits derived from establishing and maintaining contact with other people such as social support, friendship, and intimacy. Several studies have shown that many participants join such communities mainly to dispel their loneliness, meet like-minded others, and receive companionship and social support (e.g., McKenna & Bargh, 1999; Wellman & Gulia, 1999). Social enhancement is the value that a participant derives from gaining acceptance and approval of other members, and the enhancement of one’s social status within the community on account of one’s contributions to it (Baumeister, 1998). Studies have shown that many participants join virtual communities mainly to answer others’ questions and to provide information, for recognition by peers (Hars & Ou, 2002). Maintaining interpersonal connectivity and social enhancement both emphasize the social benefits of participation, and are group-referent, i.e., the referent of these values is the self in relation to other group members. This distinction between self- and groupreferent values is important, since later on, we develop the idea that the type of virtual community dictates which values are more influential in predicting social influence and participation therein. Finally, the last value we included is entertainment value, derived from fun and relaxation through playing or otherwise interacting with others. Studies have shown that many participants do so for entertainment through exploring different fictional identities, encountering, and solving virtual challenges, etc. (McKenna & Bargh, 1999). 2.2. Social influences on member participation in the virtual community In their model, B&D (2002) hypothesized that three group-level influences drive virtual community participation: compliance (i.e., normative influence of others’ expectations), internalization (i.e., congruence of one’s goals with those of group members), and identification (i.e., conception of one’s self in terms of the group’s defining features). B&D found that internalization and identification were significant predictors of participation, but compliance was not. This non-significant result for compliance is not surprising since participation in virtual communities is usually voluntary and anonymous, and members are

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

245

able to leave without much effort. So most members may not feel the need to comply with others’ expectations. We did not include compliance influences in our model, instead viewing identification and internalization to be the two salient social influences of the virtual community on member participation. Such a two-factor view of social influence is favored by existing sociological research as well (e.g., McMillan & Chavis, 1986; Postmes et al., 2000; Wellman, 1999). For instance, Etzioni (1996) suggests that two characteristics are necessary for a social grouping to be considered a community. First, a community requires an understanding of, and a commitment by the individual to, a sense of values, beliefs, and conventions shared with other community members, i.e., internalization. Second, it entails a web of affect- and value-laden relations (of varying strengths) among a group of individuals, often reinforcing one another, and going beyond the immediate utilitarian purpose of a particular interaction, i.e., identification with the group. 2.2.1. Social identity in the virtual community Social identity captures the main aspects of the individual’s identification with the group in the sense that the person comes to view himself or herself as a member of the community, as bbelongingQ to it. This is a psychological state, distinct from being a unique and separate individual, conferring a shared or collective representation of who one is (Hogg & Abrams, 1988), and involves cognitive, affective, and evaluative components (e.g., Bergami & Bagozzi, 2000; Ellemers et al., 1999). In a cognitive sense, social identity is evident in categorization processes, whereby the individual forms a self-awareness of virtual community membership, including components of both similarities with other members, and dissimilarities with non-members (Ashforth & Mael, 1989; Turner, 1985). Belonging to a virtual community also has emotional and evaluative significance (Tajfel, 1978). In an emotional sense, social identity implies a sense of emotional involvement with the group, which researchers have characterized as attachment or affective commitment (e.g., Bagozzi & Dholakia, 2002; Ellemers et al., 1999). Emotional social identity fosters loyalty and citizenship behaviors in group settings (e.g., Bergami & Bagozzi, 2000; Meyer,

Stanley, Herscovitch & Topolnytsky, 2002), and is useful in explaining consumers’ willingness to maintain committed relationships with firms in marketing settings (Bhattacharya & Sen, 2003). Finally, since the definition of one’s identity influences one’s sense of self-worth (e.g., Blanton & Christie, 2003), social identity also entails an evaluative component. Evaluative social identity is measured as the individual’s group-based or collective self-esteem and is defined as the evaluation of self-worth on the basis of belonging to the community. In our model, the cognitive, affective, and evaluative elements are components of a second-order social identity construct (see Fig. 1). Identifying with a virtual community that one has chosen volitionally stems from an understanding that membership entails significant benefits. Consistent with this view, social identity theorists posit that identification with social groups is derived, first and foremost, from their functionality—groups are identified with to the extent that they fulfill important needs of the member (Hogg & Abrams, 1988). While some needs may concern the self alone, others may also be group-referenced. Based on this discussion, we hypothesize that: Hypothesis 1. Higher levels of value perceptions lead to a stronger social identity regarding the virtual community. 2.2.2. Group norms in the virtual community Internalization, operationalized here by group norms, refers to the adoption of common self-guides for meeting idealized goals shared with others, because they are viewed as coinciding with one’s own goals. It may therefore be defined as an understanding of, and a commitment by, the individual member to a set of goals, values, beliefs, and conventions shared with other group members. Group norms are especially relevant for virtual communities since they are perhaps the most readily accessible (for instance, through FAQs) or inferable (from archives of previous interactions, for example) elements of grouprelated information available in many communities (Postmes et al., 2000) and regulating interactions among members over time (Alon et al., 2004). Group norms become known to members in different ways. One occurs upon joining the community, where the new participant actively seeks out the

246

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

group’s goals, values, and conventions. In other cases, the participant slowly comes to discover the community’s norms through socialization and repeated participation therein, over a period of time. A third possibility is that the individual learns of the community’s norms beforehand and joins the community on account of one’s perceived overlap with the community’s norms. In order to be influential, group norms should be volitionally accepted by members as congruent to their own motives (Postmes et al., 2000). An understanding of what one seeks to gain from participation should be a crucial antecedent to group norms. Therefore, Hypothesis 2. Higher levels of value perceptions lead to stronger group norms regarding the virtual community. In addition to providing knowledge regarding what the community’s objectives are and how it interacts together, an understanding and acceptance of its group norms by itself allows the individual to consider oneself as its full-fledged member. Because of this, once the member has learnt and accepted the virtual community’s norms, he or she will identify with the community more. In this regard, Hogg and Abrams (1988) note that cooperative interdependence resulting from the pursuit of shared goals results in the establishment of a well-defined group structure— which in turn leads its members to identify with it. Similarly, in a virtual community context, Alon et al. (2004) postulate that instrumental behaviors and the understanding of each others’ goals precede the establishment and propagation of the community’s identity, in their model of community life cycles. Hence, Hypothesis 3. Stronger group norms lead to a stronger social identity regarding the virtual community. Hypothesis 3 implies that value perceptions influence social identity in two ways: directly and also through their impact on group norms. Next, it is useful to consider the specific processes by which group norms advance the individual’s desires for participation. At one level, strong group norms implicitly generate consensus among members regarding when and how to engage in online social interactions. In this respect, group norms promote mutual agreement among group members regarding the specific details

of participation itself. In a second sense, research on group negotiation has shown that group norms facilitate a cooperative motivational orientation among group members (Weingart, Bennett, & Brett, 1993). Philosopher Bratman (1997) similarly notes that shared intentional activity is preceded by associated forms of mutual responsiveness on the members’ parts to do whatever it takes to be able to complete their own parts in enabling joint action to occur. Group norms should therefore increase participants’ inclinations to mutually accommodate their schedules, preferences and commitments with others’ in order to be able to engage in group action. Thus, Hypothesis 4. Stronger group norms lead to stronger mutual agreement to participate in the virtual community. Hypothesis 5. Stronger group norms lead to a stronger willingness to mutually accommodate each other to enable participation. Both mutual agreement and mutual accommodation represent mechanisms through which the participant moves from rather general and broadly defined goals and conventions of the group, toward actualizing specific episodes of online social interactions. In this sense, they serve as mediators by which group norms influence the individual’s participation desires in our model. Both provide the potential for deciding to engage in virtual community activities but do not, in and of themselves, necessarily provide the motivation to do so. The transformation of mutual agreement and accommodation into intentions to engage in virtual community activities is hypothesized to be provided by felt desires to engage in these activities. Desires provide the motivation to decide in favor of acting as part of a virtual community. Therefore, Hypothesis 6. Stronger mutual agreement leads to stronger desires to participate in the virtual community. Hypothesis 7. Stronger mutual accommodation leads to stronger desires to participate in the virtual community. At the same time, we posit that participation desires are also influenced by social identity. Since identification renders a person to maintain a positive self-defining relationship with other virtual community members, he or she will be motivated to engage in

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

247

behaviors needed to do so (Hogg & Abrams, 1988). An important part of maintaining this relationship with the group is to actively participate in online social interactions. In this respect, social identities prescribe and instigate group-oriented behaviors. As examples, Ellemers et al. (1999) studied experimentally formed groups and found that aspects of social identity influenced acts of in-group favoritism, whereas Bergami and Bagozzi (2000) found that social identity led to performance of organizational citizenship behaviors by firm employees. Based on this discussion, Hypothesis 8. Stronger social identity leads to stronger desires to participate in the virtual community. Consistent with the B&D model, we view desires as mediators of the influence of individual and grouplevel antecedents on we-intentions. Since we study intentional social action, the referent of the participant’s actions is the virtual community rather than one’s self. A we-intention is defined as a bcommitment of an individual to engage in joint action and involves an implicit or explicit agreement between the participants to engage in that joint actionQ (Tuomela, 1995, p. 9; see B&D, 2002 for a detailed discussion). We note here that such joint action may not necessarily be contemporaneous; members can perform their respective parts at different times. Nevertheless, joint actions entail coordinated endeavors between group members. The role played by desires is to transform the multiple reasons for acting found in the antecedents, which in our model are individual and social reasons for participating, into an overall motivation to act. Since such behavior is effortful, involving a greater or lesser degree of effort (e.g., remembering when to meet or respond to a group member, adjusting other engagements in one’s schedule to interact online, etc.), desires are necessary precursors to we-intentions in performing such actions (Perugini & Bagozzi, 2001). Based on this discussion, Hypothesis 9. Stronger desires lead to higher levels of we-intentions to participate in the virtual community. Further, we posit that the mediation of desires in the effects of the social influence variables on weintentions will be partial. This is because, participation in virtual communities, although goal-directed, involves both effortful as well as habitual

components. The habitual aspects of such actions are relevant since many members may have belonged to the virtual community for a long time beforehand, having developed routines of participation therein. For many participation episodes, behavior may be automatic, as in checking periodically to see if new messages have been posted on a bulletinboard to which one belongs. For such habitual participation, group norms and social identity should influence we-intentions directly, rather than through desires, depending on the strength of one’s habit. Therefore, Hypothesis 10. Stronger group norms lead to higher levels of we-intentions to participate in the virtual community. Hypothesis 11. A stronger social identity leads to higher levels of we-intentions to participate in the virtual community. Finally, it is important to stress that whereas the B&D (2002) analysis ended with we-intentions, we also measured participants’ behaviors in a second wave, expecting we-intentions to significantly predict subsequent participation, in accordance with standard attitude-theoretic formulations (Eagly & Chaiken, 1993). Hence, Hypothesis 12. Higher levels of we-intentions lead to higher levels of participation in the virtual community. 2.3. Network-based and small-group-based virtual communities In the literature on virtual communities (and especially so within marketing), they have tended to be construed as vast, vaguely defined, social spaces comprised of ever-changing congregations of participants (e.g., B&D, 2002; Wellman et al., 1996; Williams & Cothrel, 2000). The implicit assumption in such construals is that this abstract social category, the community as a whole, is the salient basis of social identity and group norms for all its members. Such a view also does not allow one to distinguish between different types of communities that might meet in different internet venues (see Section 3.1 below), nor does it allow for the possibility that the nature of the community may change over time as repeated

248

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

interactions among members result in the formation of interpersonal relationships (Alon et al., 2004). While internet venues such as bulletin-boards and chat rooms may be unambiguous to organizers or outside observers, their participants may have starkly different views regarding who belongs to the virtual communities located therein, what their values are, and how central they are for its members. In exploring this issue further, the distinction made by sociologists (e.g., Wellman, 1999) between neighborhood solidarities, defined as tightly bounded, densely knit groups with strong relationships between members, and social networks, defined as loosely bounded, sparsely knit networks of members sharing narrowly defined relationships with one another, is useful. Whereas neighborhood solidarities are geographically conjoint groups, where each member knows everyone else and relies on them for a wide variety of social support, social networks are usually geographically dispersed groups that interact with one another for a specific reason, and usually without prior planning (Wellman, 1999). Social psychologists similarly distinguish between common bond and common identity groups (Prentice, Miller, & Lightdale, 1994; Sassenberg, 2002). Whereas bonds between members are the glue holding the group together in common bond groups, such attachment is dependent on identification to the whole group, in common identity groups. Common bond groups therefore correspond to neighborhood solidarities, whereas common identity groups correspond to social networks. This distinction, of viewing the community as either the same group of individuals with each of whom the person has relationships, or viewing it as a venue where people (strangers or acquaintances) with shared interests or goals meet, provides a useful typology of marketing relevance to classify virtual communities. In some instances, the member’s definition of the virtual community may primarily be in terms of the venue, and only superficially associated with any particular individuals within it. For instance, a person may log into a bulletin-board on gardening, and participate because he is interested in the subjectmatter, but have no expectation or inclination to meet, chat or communicate with any particular individual therein. Similarly, an engaged Amazon. com customer may read and benefit from reviews

offered by other customers, without any personal knowledge of, or relationships with, the reviewers. We call a virtual community defined this way, i.e., a specialized, geographically dispersed community based on a structured, relatively sparse, and dynamic network of relationships among participants sharing a common focus, to be a network-based virtual community. In other cases, the member may identify primarily with a specific group (or groups) of individuals, rather than with the online venue itself. For example, a software developer may log on to a messaging system specifically to chat with her geographically distant bbuddy groupQ of software developers every Wednesday night to trade ideas, learn new concepts, and socialize. Here, the developer’s focus is on communication with peers that she knows personally, rather than on the venue of the AOL messaging system. We call such a virtual community, constituted by individuals with a dense web of relationships, interacting together online as a group, in order to accomplish a wider range of jointly conceived and held goals, and to maintain existing relationships, to be a smallgroup-based virtual community. These are virtual communities because they meet through online venues for a significant proportion (but not necessarily all) of their overall interactions together, as a group. Moreover, they often have commercial focuses. For example, within such company-sponsored organizations as Harley Owners Groups (HOGs), many small-group-based virtual communities exist that participate extensively in internet-based activities, which are augmented by face-to-face interactions periodically. 2.4. The moderating role of community type in the social influence model We first consider how members’ motivations for participation might vary between these two virtual communities. To do so, it is useful to better understand how small-group-based and network-based communities differ from each other. One important difference between them is that the specific group with which the member interacts holds greater importance for those belonging to small-group-based when compared to network-based communities. This is because the individual knows everyone else personally, and may

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

249

often have special shared histories and close personal relationships with them. As a result, relationships between group members are likely to be stronger, more resilient, and more stable than those in network-based communities, where members are more likely to participate primarily to achieve functional goals (e.g., to learn how to install a software program) and may have tenuous, short-lived, and easily severed relationships with others. Accentuating the importance of the group for small-group-based virtual community members is also the fact that the particular virtual community is often only one of a number of places where such groups meet. Online social interactions are often supplemented by face-to-face and other offline forms of interactions. For instance, a small group of HOG members may not only chat online with one another periodically in the course of a week, but meet on weekdays for coffee and fellowship, and on weekends for group outings. In contrast, network-based virtual community members are more likely to interact with each other exclusively online. These differences all point to the greater importance of group-referent values, for small-group-based community members and self-referent values for network-based virtual community members. As a result, we expect that: Hypothesis 13. Purposive and self-discovery value perceptions will be stronger for network-based when compared to small-group-based virtual community members. Hypothesis 14. Maintaining interpersonal connectivity and social enhancement will be stronger for smallgroup-based when compared to network-based virtual community members. These posited distinctions in strength of value perceptions should manifest themselves in differences in the expressed mean levels of value perceptions by members of the two virtual communities. Further, we expect that, since these different motivations—selfreferent for network-based and group-referent for small-group-based virtual communities—provide the impetus for participation, they should also influence the social influence variables, group norms, and social identity to a much greater extent, respectively. Specifically, we expect that,

Hypothesis 15. The impact of purposive and selfdiscovery values on group norms and social identity will be stronger for network-based than for smallgroup-based virtual community members. Hypothesis 16. The impact of maintaining interpersonal connectivity and social enhancement on group norms and social identity will be stronger for small-group-based than for network-based virtual community members. Taken together, all the above hypotheses provide an understanding of why consumers participate in virtual communities, the bases of the community’s social influence, as well as differences between smallgroup-based and network-based virtual communities.

3. Empirical study 3.1. Finding members of small-group-based and network-based virtual communities As noted, online venues offer a useful starting point for finding both types of virtual community members. For the sake of generalizability, we included virtual communities from seven different types of internet venues (e.g., Catterall & Maclaran, 2001) in this study. The first type, email lists, refers to specialized mailing lists organized around particular topics of interest, and are widely used by firms to maintain customer relationships. A message posted to the list by one member is generally transmitted to all members, with or without editing by a list moderator. Among email lists included in our study were those of the Lord of the Rings enthusiasts and the bDisneyDollarLessQ Club for budget-minded tourists. The second type, website bulletin boards, is company-sponsored venues, where participants can post and read messages about the firm’s products and services. An example in our study included the bAdvanced Squad Leader1 Q website. The third type was Usenet newsgroups, each having a specific focus of interest such as technical issues (e.g., Linux installation), hobbies (e.g., Pokemon), and specific products and brands (e.g., Ford
1 This website was sponsored and maintained by Multi-Man Publishing, publisher of the Advanced Squad Leader video game.

250

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

Mustang cars). Among others, our study included members from the alt.marketing.ebay and alt.guitar.amps newsgroups. The fourth venue was real-time online-chat systems, such as ICQ and AOL instant messenger, both of which were represented in our study. These venues allow participants to chat with others in real time. The fifth type of venue was web-based chat rooms such as those on the AOL and MSN websites. Examples in our study included the AOL Word Haven chat room and the NHB chat room on pork.com. The sixth type of venue we included was multiplayer virtual games, wherein gamers can play as a group by simultaneously logging online together, through wired or wireless interfaces.2 Examples of networked video games in our study included Diablo II, Dungeon Siege, and Neverwinter Nights. Finally, the seventh venue included in the study was multi-user domains (MUDs). MUDs are a special form of real-time computerized conferencing, where participants don pseudonymous personas and role play in quests, masquerades, games, and also in work-related communal interactions (Wellman et al., 1996). Among examples of MUDs included in our study were Avatar, Wheel of Time, and Xyllomer. 3.2. Pre-test To better understand whether these seven venues harbor small-group-based communities, networkbased communities or both, we conducted a pretest with 240 regular participants in these venues. Participants were first asked to choose the venue that they participated in most often, and then to describe their interactions therein in detail. These descriptions were content-analyzed by two coders. Specifically, each response was coded into one of the following three categories: (1) the respondent usually interacts with the same group of people; (2) the respondent usually interacts with different individuals or groups of people; and (c) unable to determine the type of interaction. Of the 240 decisions made, the 2
During game-play, players normally engage in spirited conversations regarding the game as well as other topics. For instance, in describing his experience playing such a game, one of our participants noted, bUsually my group opts for less distracting, less roleplay intensive games so we can converse more freely.Q
2

coders agreed on 213 (or 89%) decisions. The remaining decisions were resolved after comparison and discussion. After eliminating those descriptions which could not be gauged by the coders for interaction type, the final classification can be found in Table 1, which provides the proportion of respondents by type of venue indicating that they participated either with the same group or with different groups every time. The results showed that most participants of the first three venues—email lists, website bulletin boards, and Usenet newsgroups—engaged in interactions with different individuals or groups on each occasion. In contrast, a vast majority of participants in the remaining four venues interacted with the same group on most occasions. Based on these results, we concluded that the first three venues would be the most suitable for finding network-based virtual communities, whereas the last four venues would be appropriate for finding small-group-based virtual communities for our study. 3.3. Method of main study We then collected data from regular participants in the seven venues by conducting an internet-based
Table 1 Proportion of participants by virtual community venue and type of social interaction in pre-test Virtual community venue Email listsa Website bulletin boardsa Usenet newsgroupsa Real-time online-chat systemsb Web-based chat roomsb Multiplayer virtual gamesb Multi-user domainsb a b

Proportion participating with same group every time 3.4% 17.6%

Proportion participating with different groups every time 96.6% 82.4%

Sample size

29 34

3.0% 97.2%

97.0% 2.8%

33 34

90.0% 100% 85.2%

10.0% 0% 14.8%

20 6 27

Determined as suitable for finding networks. Determined as suitable for finding small groups.

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

251

survey in the Spring of 2002. The survey was publicized by contacting a significant number of organizers of popular online venues of each type. These organizers informed their membership about the survey, and encouraged their members to participate by visiting a website where the survey was made available. The study was introduced to participants as an bopinion survey regarding group interactions on the internet.Q Participants were asked to select the venue that they most frequently visited when online, giving them the opportunity to complete the survey regarding the type of virtual community with which they were most familiar. After this selection was made, participants described their chosen interaction in some detail such as the name of the venue, the date when they first joined, whom they normally interacted with, details regarding their interactions, what they liked about their online group, etc. Based on our pretest results, participants of the four venues corresponding to small-group-based virtual communities were then branched to another section of the survey, where they were told: bImagine that you are logging on to the internet to engage in the group interaction that you described above. You have a number of friends within that group that you regularly interact with. Please picture briefly in your mind the name and image of each online friend. Then write your first name and their first names/handles in the table below. You may include up to, but not necessarily, five group members. Please be sure to include only friends that are part of the group you regularly interact with on the internet.Q Similarly, since our pretest results indicated that most participants of the remaining three venues interacted with whomever was online, they were then branched to a section where they described their last online interaction in detail. These respondents were then told to visualize up to five average members, using them as representatives of the other virtual community members. All participants, regardless of the venue selected, responded to the same set of measures. 3.4. Sample characteristics and measures A total of 545 participants representing 264 different virtual communities completed the survey.

Of the entire sample, 41.8% were female, 54.3% were male, while 3.9% did not disclose their gender. Respondents ranged in age from 18 to 79, with a mean age of 33.1 years (median=30, S.D.=13.43). While 387 (71%) were US residents, the other 29% belonged to a total of 27 other countries. Canada (n=42, 7.7%), Australia (n=23, 4.2%), and Germany (n=21, 3.9%) were the three next largest subgroups, represented in the sample. On average, respondents had been online for 7.53 years (S.D.=3.57), suggesting a high level of experience. All of the measures used in the survey are provided in Table 2. The value perception measures were the same as those used by Flanagin and Metzger (2001), and were introduced with the following preface, bHow often do you use your online group (as identified above) for satisfying the following needs?Q The measures of group norms, social identity, desires, and we-intentions were similar to those used by B&D (2002). Because of the large number of different virtual communities (264) involved, we measured participation behaviors through self-reports, rather than other means such as observation. Participants were contacted through a follow-up email approximately a month later to obtain this information, with two reminders to encourage responses given thereafter. A total of 465 (or 85.3%) participants responded to this second-wave of questions regarding participation behavior with response rates ranging from 80.9% to 91.4% depending on venue type. 3.5. Preliminary analysis Our full sample model includes participants of all seven venues and is used to test Hypotheses 2–12, and our network-based and small-group-based subsamples are used to test the moderation (Hypotheses 13–16). All of the models (CFA and SEM) described below were run using the LISREL 8.52 program (Joreskog & Sorbom, 1999). The goodness-of-fit of ¨ ¨ ¨ the models was assessed with chi-square tests, the root mean square error of approximation (RMSEA), the non-normed fit index (NNFI), and the comparative fit index (CFI). Discussions of these indices can be found in Bentler (1990), Browne and Cudeck (1993), Marsh and Hovecar (1985), and Marsh, Balla, and Hau (1996). Satisfactory model fits are indicated by non-

252

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 Table 2 (continued) Constructs and measures q q= 0.84b, AVE=0.75c Evaluative social identity (two measures) bI am a valuable member of the groupQ (seven-point bagree–disagreeQ scale) bI am an important member of the groupQ (seven-point bagree–disagreeQ scale) Group norms (two measures) bInteracting together sometime within the next 2 weeks with your online group can be considered to be a goal. For each of the people listed below, please estimate the strength to which each holds the goalQ (five-point bweak–strongQ scales) Strength of self’s goal Average of the strength of group members’ goal Mutual agreement (two measures) bHow strong would you say the explicit or implicit agreement is among each of the following to interact with on the internet as a group sometime during the next 2 weeks? (five-point bweak–strongQ scales) Strength of self’s agreement Average of the strength of group members’ agreement Mutual accommodation (two measures) bHow willing are each of the following to accommodate or adjust to the needs of the others in the group so as to choose a time and place to interact together on the internet sometime during the next 2 weeks? (five-point bnot at all willing–very willingQ scales) Strength of self’s willingness to accommodate Average of the strengths of group members’ willingness to accommodate Desires (three measures) bI desire to interact with the group sometime during the next 2 weeksQ (seven-point bagree–disagreeQ scale) bMy desire for interacting together with the group can be described as:Q (seven-point bvery weak desire–very strong desireQ scale) bI want to interact together with my group during the next 2 weeks.Q (seven-point bdoes not describe me at all–describes me very muchQ scale) q q= 0.97, AVE= 0.84

Table 2 Details of measures in the main study Constructs and measures Purposive value a (nine measures) To get information To learn how to do things To provide others with information To contribute to a pool of information To generate ideas To negotiate or bargain To get someone to do something for me To solve problems To make decisions Self-discovery value a(two measures) To learn about myself and others To gain insight into myself Maintaining interpersonal interconnectivity a (two measures) To have something to do with others To stay in touch Social enhancement value a (two measures) To impress To feel important Entertainment value a (four measures) To be entertained To play To relax To pass the time away when bored Cognitive social identity (two measures) Please indicate to what degree your selfimage overlaps with the identity of the group of friends as you perceive it (seven-point bnot at all–very muchQ scale) How would you express the degree of overlap between your personal identity and the identity of the group you mentioned above when you are actually part of the group and engaging in group activities? (eight-point bnot at all–very muchQ scale) Affective social identity (two measures) How attached are you to the group you mentioned above? (seven-point bnot at all–very muchQ scale) How strong would you say your feelings of belongingness are toward the group you mentioned above? (seven-point bnot at all–very muchQ scale)

q q= 0.92, AVE= 0.71

q q= 0.89, AVE= 0.69

q q= 0.94, AVE= 0.76

q q= 0.93, AVE= 0.72

q q=0.89, AVE=0.66

q q= 0.90, AVE= 0.71

q q= 0.97, AVE=0.80

q q=0.87, AVE=0.68

q q= 0.94, AVE=0.93

q q= 0.87, AVE= 0.68)

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 Table 2 (continued) Constructs and measures We-intentions (two measures) bI intend that our group (i.e., the group that I identified before) interact on the internet together sometime during the next 2 weeksQ (five-point bagree–disagreeQ scale) We (i.e., the group that I identified above) intend to interact on the internet together sometime during the next 2 weeks.Q (five-point bagree–disagreeQ scale) Participation behavior (one measure) Product of: bHow many times did you chat online with your group within the last 2 weeks?Qd (venue-appropriate question customized for each venue) How much time did you spend on average when you chatted with your group? Questions for the other venues are available from the authors. a All value perceptions measures used five-point scales. b q q is the value of composite reliability. c AVE is the average variance extracted. d This version was used for web-based chat-rooms and realtime online chat systems. q q=0.92, AVE=0.71

253

3.6. Results 3.6.1. Measurement model evaluation We evaluated the internal consistency and discriminant validity of model constructs. Given space considerations, the results for only the full sample are reported here in detail. The results for the subsamples were substantively similar and are available from the authors. 3.6.2. Internal consistency We used two measures to evaluate internal consistency of constructs. The composite reliability (q q) is a measure analogous to coefficient a (Bagozzi & Yi, 1988; Fornell & Larcker, 1981, Eq. (10)), whereas the average variance extracted (q VC(n)) estimates the amount of variance captured by a construct’s measure relative to random measurement error (Fornell & Larcker, 1981, Eq. (11)). Estimates of q q above 0.60 and q VC(n) above 0.50 are considered supportive of internal consistency (Bagozzi & Yi, 1988). The q q and q VC(n) values for all constructs in the model (provided in Table 2) were significantly higher than the stipulated criteria, and therefore indicative of good internal consistency. 3.6.3. Discriminant validity Discriminant validity of the model constructs was evaluated using three different approaches. A confirmatory factor analysis model was built with 14 latent constructs and a total of 29 measures. Results showed that the model fit the data well. The goodnessof-fit statistics for the model were as follows: v 2(287)=1010.83, pc0.00, RMSEA=0.07, SRMR= 0.04, NNFI=0.95, CFI=0.96. The /-matrix (correlations between constructs, corrected for attenuation) is provided in Table 3. As a first test of discriminant validity, we checked whether the correlations among the latent constructs were significantly less than one. Since none of the confidence intervals of the /-values (Ftwo standard errors) included the value of one (Bagozzi & Yi, 1988), this test provides evidence of discriminant validity. Secondly, for each pair of factors, we compared the v 2-value for a measurement model constraining their correlation to equal one to a baseline measurement model without this constraint. A v 2-difference test was performed for each pair of factors (a total of 91

significant chi-square tests, RMSEAV0.08, and NNFI and CFI valuesz0.90. Two indicators were used to operationalize each latent construct in the CFA and the SEM. For latent constructs where more than two items were available (informational value, instrumental value, entertainment value, and desires), these were combined to produce two indicators according to the so-called bpartial disaggregation modelQ (Bagozzi & Edwards, 1998). Compared to models where every item is a separate indicator, this yielded models with fewer parameters to estimate, and reasonable ratios of cases to parameters, while smoothing out measurement error to a certain extent. All analyses were performed on covariance matrices (Cudeck, 1989). An initial exploratory analysis and examination of the correlation matrix showed that the correlations between the measures of informational value and instrumental value were very high. Consequently, and because such a combination is theoretically justifiable (see our earlier discussion), these two values were treated as a single construct labeled bpurposive valueQ with four measures, two each of informational and instrumental value.

254

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

Table 3 /-matrix of latent constructs for full sample BEH BEH WE DES PUR SD MII SE ENT CSI ASI ESI GN AG AC 1 0.44* 0.36* 0.10* 0.23* 0.35* 0.21* 0.44* 0.12* 0.30* 0.32* 0.36* 0.36* 0.27* WE 1 0.60* 0.39* 0.34* 0.32* 0.20* 0.18* 0.35* 0.51* 0.35* 0.62* 0.53* 0.46* DES PUR SD MII SE ENT CSI ASI ESI GN AG AC

1 0.30* 0.40* 0.34* 0.20* 0.28* 0.32* 0.51* 0.27* 0.59* 0.43* 0.33*

1 0.51* 0.34* 0.35* 0.05* 0.32* 0.37* 0.24* 0.33* 0.30* 0.34*

1 0.60* 0.45* 0.45* 0.38* 0.47* 0.32* 0.35* 0.37* 0.31*

1 0.38* 0.65* 0.25* 0.50* 0.36* 0.33* 0.33* 0.34*

1 0.43* 0.37* 0.25* 0.29* 0.21* 0.15* 0.16*

1 0.16* 0.38* 0.29* 0.18* 0.21* 0.19*

1 0.48* 0.25* 0.31* 0.29* 0.32*

1 0.70* 0.50* 0.43* 0.47*

1 0.33* 0.37* 0.35*

1 0.65* 0.57*

1 0.55*

1

BEH=behavior, WE=we-intentions, DES=desires, PUR=purposive value, SD=self-discovery value, MII=maintaining interpersonal interconnectivity, SE=social enhancement, ENT=entertainment value, CSI=cognitive SI, ASI=affective SI, ESI=evaluative SI, GN=group norms, AG=mutual agreement, AC=mutual accommodation. All correlations are significantly less than 1.00. * Significant at a=0.05 level.

tests in all), and in every case resulted in a significant difference, again suggesting that all of the measures of constructs in the measurement model achieve discriminant validity. Third, we performed a test of discriminant validity suggested by Fornell and Larcker (1981). This test is supportive of discriminant validity if the average variance extracted by the underlying construct is larger than the shared variance (i.e., the / 2 value) with other latent constructs. This condition was satisfied for all of the 91 cases. In sum, internal consistency and discriminant validity results enabled us to proceed to estimation of the structural model. 3.7. Structural model estimation Structural models were built separately for the full sample (to test Hypotheses 1–12), as well as for the network and small group subsamples (to test Hypotheses 13–16). Table 4 provides the goodness-of-fit statistics for these models and R 2 values of the endogenous constructs. Tests of mediation and comparisons with rival models were conducted on the full sample to test its robustness. Using multiple-sample analyses in LISREL, structured means analyses were conducted to test Hypotheses 13 and 14, and tests of

moderation were conducted to test Hypotheses 15 and 16. 3.7.1. Full sample model Considering the fit-statistics from Table 4, the chisquare is significant ( pb0.05), which is usually the case for large sample sizes. All the other statistics are within the acceptable ranges for the full model, indicating a good fit to the data. Considering social identity first, both purposive (c=0.15, S.E.=0.07) and entertainment (c=0.21, S.E.=0.07) values are significant predictors of social identity, whereas the other three value perceptions are not, supporting Hypothesis 1. Examining the antecedents of group norms next, two of the five value perceptions, purposive (c=0.34, S.E.=0.10) and self-discovery (c=0.19, S.E.=0.09) values, have significant paths to group norms, whereas the other three do not. Twenty-four percent of the variance in group norms is explained by value perceptions, supporting Hypothesis 2. Fig. 2 summarizes these and subsequent results. In examining Hypotheses 3–5 which explicate the associations between group norms and its consequences, we find that group norms influences social identity (b=0.23, S.E.=0.07), mutual agreement (b=0.83, S.E.=0.07), and mutual accommodation (b=0.77, S.E.=0.07), providing support to all three

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 Table 4 Goodness-of-fit statistics for structural models Statistics v2 RMSEA SRMR NNFI CFI Full sample v 2(342)=927.77, pb0.001 0.072 0.061 0.95 0.96 Small-group-based virtual communities subsample v 2(342)=854.25, pb0.001 0.074 0.067 0.95 0.95

255

Network-based virtual communities subsample v 2(342)=824.73, pb0.001 0.074 0.066 0.94 0.95

R 2 values for endogenous variables Group norms 0.24 Social identity 0.62 Mutual accommodation 0.43 Mutual agreement 0.57 Desires 0.37 We-intentions 0.54 Participation behavior 0.24

0.28 0.85 0.35 0.42 0.48 0.58 0.33

0.22 0.47 0.32 0.56 0.40 0.66 0.21

hypotheses. Sixty-two percent, 57% and 43% of variance in social identity, mutual agreement, and mutual accommodation are explained by their antecedents, respectively. Considering whether these variables influences desires to participate next, we find that mutual agreement (b=0.34, S.E.=0.07) and social identity (b=0.59, S.E.=0.13) do influence desires, but mutual

accommodation does not (b=À0.05, S.E.=0.05). Thus, Hypotheses 6 and 8 are supported, but Hypothesis 7 is not. Thirty-seven percent of the variance in desires is explained by these antecedents. On hindsight, the non-significant effect of mutual accommodation on desires is perhaps not surprising, since for many of the participants belonging to network-based communities and interacting with different members

Fig. 2. Parameter estimates for final structural model. Unstandardized coefficients and standard errors in parentheses; insignificant paths are omitted for ease of exposition.

256

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

every time, given the mutual agreement to participate, the necessity for mutual accommodation to adjust to the needs of others may not be an issue. Instead, they may be willing to interact with whomever is online. Considering the direct impact of group-influence variables on we-intentions, the path from group norms (b=0.43, S.E.=0.06) is significant, but that from social identity (b=0.16, S.E.=0.10) is not. Thus, Hypothesis 9 is supported but Hypothesis 10 is not. Finally, supporting Hypotheses 11 and 12, the paths from desires to we-intentions (b=0.19, S.E.=0.04), and from we-intentions to behavior (measured in the second-wave; b=1.43, S.E.=0.19) are both significant. Fifty-four percent of the variance in we-intentions and 17% of the variance in behavior is explained by their antecedents. 3.7.2. Tests of mediation To obtain further support for the validity of the model, rather than using a saturated model where beverything is related to everythingQ as the baseline, we performed formal tests of mediation for all possible paths in our model. This was done to check if additional direct paths not included in the model were significant. Specifically, we conducted 7 tests to check for the significance of a total of 32 potential paths.3 As an example, to check if the direct paths from the five value perceptions to desires were significant, we compared the model described above with a model in which five additional direct paths were added from the five value perceptions to desires. The difference in chi-square values between the two 2 models (v d (5)=6.37), with five degrees of freedom, is a test of the significance of these added paths. Since this difference is not significant ( pN0.27) and none of the individual paths is significant, we concluded that the direct paths from the value perceptions to desires are insignificant, and therefore group norms and social identity mediate all of the effects of value perceptions value on desires, as hypothesized. Of 32 potential paths tested, results show that only 3 of these were significant. The direct paths from entertainment value to behavior, entertainment value to mutual agreement, and mutual agreement to

we-intentions were all significantly greater than zero (dashed arrows in Fig. 2). The goodness-of-fit statistics for a model including these three paths were as follows: v 2(339)=884.21, pc0.00, CFI= 0.96, NNFI=0.95, RMSEA=0.069, and SRMR= 0.058. The R 2 values of the three endogenous variables after incorporating these additional significant paths were mutual agreement (0.57 vs. 0.57 before), we-intentions (0.54 vs. 0.54 before), and behavior (0.24 vs. 0.17 before). The other 29 paths were not significant, providing additional evidence that our proposed model is robust, and suggesting that the social influence variables mediate most of the effects of value perceptions on participation in virtual communities. 3.7.3. Moderating effects of virtual community type (test of Hypotheses 13–16) We conducted multiple sample analyses (Joreskog ¨ & Sorbom, 1999) for the network and small-group ¨ ¨ subsamples to test the hypotheses regarding the moderating role of virtual community type. Table 5 provides the means, standard deviations, and the Cronbach a reliabilities of the constructs for the subsamples. As is evident, the reliabilities are good overall. Hypothesis 13 posited that the self-referent values, i.e., purposive and self-discovery values, would be greater for the network-based relative to the smallgroup-based subsample, whereas Hypothesis 14 posited that the group-referent values, maintaining interpersonal connectivity and social enhancement, would be greater for the small-group-based relative to the network-based virtual communities. To test these hypotheses, we conducted a structured means analysis in LISREL, using the following model of means structures (Joreskog & Sorbom, 1999): x (g)=s x + ¨ ¨ ¨ K x n (g)+d (g), where g =small-group and network, x (g) is a vector of input variables, s x is a vector of constant intercept terms, K x is a matrix of coefficients of the regression of x on n, n is a vector of latent independent variables, d is a vector of measurement errors in x and the means of the n (g)=j (g). We set the j (small-group)=0 to define the origin and units of measurement of the n-factors and computed j (network), and then determined whether the differences in the factor means of the two groups were significantly different from each other. Table 6 provides the results.

3 The detailed results are available from the authors upon request.

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 Table 5 Means, standard deviations, and reliabilities for construct measures Scale Small-group-based virtual community subsample (N=278) No. of measures Purposive value Self-discovery Maintain inter. interconnectivity Social enhancement Entertainment value Group norms Cognitive SI Affective SI Evaluative SI Mutual agreement Mutual accommodation Desires We-intentions Participation behavior 9 2 2 2 4 2 2 2 2 2 2 3 2 1 Mean 25.43 5.49 7.01 4.03 15.88 7.07 6.97 10.16 10.25 7.56 6.78 15.89 7.71 6.38 S.D. 7.25 2.26 2.08 2.11 2.93 1.92 3.45 2.89 3.19 2.10 2.00 3.52 1.76 2.61 a 0.89 0.85 0.85 0.78 0.74 0.83 0.82 0.77 0.92 0.88 0.86 0.84 0.85 – Network-based virtual community subsample (N=265) No. of measures 9 2 2 2 2 2 2 2 2 2 2 3 2 1 Mean 26.74 5.14 5.79 3.39 11.79 6.90 7.14 9.58 8.74 7.23 6.55 15.75 8.00 4.56 S.D. 5.39 2.21 2.38 1.67 4.22 2.30 3.42 3.07 3.73 2.30 2.47 3.76 1.70 2.86

257

a 0.80 0.77 0.84 0.81 0.80 0.83 0.78 0.83 0.89 0.78 0.91 0.84 0.80 –

As can be seen, the factor mean of purposive value was significantly higher for the network subsample, but that of self-discovery was not different between the two groups. These results partially support Hypothesis 13. Considering the two group-referent values, from Table 6, we find that both, maintaining interpersonal connectivity and social enhancement factor means were higher for the small-group relative to the network subsample, providing support to Hypothesis 14. Interestingly, entertainment value although not characterized as either self- or groupreferent, was also significantly higher for the small group subsample, suggesting that this value perception might have a group-referent basis. These results provide evidence that the motivations of participants in the two virtual communities have different bases of reference.

To test Hypotheses 15 and 16, we conducted tests of moderation to determine whether the strengths of the paths from value perceptions to social identity and group norms were different between the small-groupbased and the network-based subsamples. Table 7 summarizes the analyses and results. Consider the first test presented in Table 7. To test Hypothesis 15, that the purposive value to group norms path is stronger for network-based relative to small group-based virtual communities, we ran two multiple-sample models. In the first model, all paths were unconstrained between the two groups. This is the bno constraintsQ or the baseline model in the first row of Table 7. In the second model, the purposive value to group norms path was constrained to be equal for both subsamples. This is the bequal paths model.Q The difference in chi-square values between the two

Table 6 Test of factor mean differences between network and small group subsamples for the five value perception constructs (Hypotheses 13 and 14) Value perceptions Small-group-based subsample factor meana j (small-group) 0 0 0 0 0 Network-based subsample factor mean j (network) 0.29 À0.16 À0.61 À0.31 À1.01 t-value (significance level) 4.21 ( pb0.001) À1.76 (ns) À6.23 ( pb0.001) À4.03 ( pb0.001) À12.53 ( pb0.001) Result of statistical est.

Purposive value Self-discovery value Maintaining interpersonal interconnectivity Social enhancement value Entertainment value a j (small-group)bj (network) j (small-group)=j (network) j (small-group)Nj (network) j (small-group)Nj (network) j (small-group)Nj (network)

Reference value: factor means set to zero.

258

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

Table 7 Results of multiple-sample moderation analyses to test Hypotheses 15 and 16 Hypothesis Baseline model Hypothesis 15 Purposive valueYgroup norms is greater for network than for small group. Self-discovery valueYgroup norms is greater for network than for small group. Purposive valueYsocial identity is greater for network than for small group. Self-discovery valueYsocial identity is greater for network than for small group. Omnibus test: for all four paths together Path coefficients in unconstrained model Chi-square statistics test results No constraints model: v 2(496)=1675.51

c (network)=0.32a (0.14b), c (small-group)=0.01 (0.12) c (network)=0.38 (0.12), c (small-group)=0.04 (0.11) c (network)=0.37 (0.15), c (small-group)=À0.03 (0.15) c (network)=0.33 (0.13), c (small-group)=0.17 (0.13)

Equal paths model: v 2(497)=1678.31; test of H 1 : v 2(1)=2.80, pc0.09; d difference marginally significant Equal paths model: v 2(497)=1679.72; test of H 1 : v 2(1)=4.21, pb0.05; d difference is significant Equal paths model: v 2(497)=1679.29; test of H 1 : v 2(1)=3.78, p=0.05; d difference is significant Equal paths model: v 2(497)=1676.31; test of H 1 : v 2(1)=0.80, pN0.37; d difference not significant Equal paths model: v 2(500)=1692.51; test of H 1 : v 2(4)=17.00, pb0.001; d all four paths combined are different between the subsamples

Hypothesis 16 Maintaining interpersonal connectivityYgroup norms is greater for small group than for network Social enhancement valueYgroup norms is greater for small group than for network. Maintaining interpersonal connectivityYsocial identity is greater for small group than for network. Social enhancement valueYsocial identity is greater for small group than for network. Omnibus test: for all four paths together

c (network)=À0.11 (0.11), c (small-group)=0.15 (0.10) c (network)=À0.02 (0.13), c (small-group)=0.19 (0.10) c (network)=À0.06 (0.12), c (small-group)=0.46 (0.13) c (network)=0.02 (0.14), c (small-group)=0.30 (0.12)

Equal paths model: v 2(497)=1678.30; test of H 1 : v 2(1)=2.79, pc0.09; d difference marginally significant Equal paths model: m2(497)=1676.53; test of H 1 : m2(1)=1.02, pN0.31 d Difference not significant Equal paths model: v 2(497)=1684.41; test of H 1 : v 2(1)=8.90, p b0.05; d difference is significant Equal paths model: v 2(497)=1677.42; test of H 1 : v 2(1)=1.91, pN0.16; d difference not significant Equal paths model: v 2(500)=1687.35; test of H 1 : v 2(4)=11.84, pb0.01; d all four paths combined are different between the subsamples

a b

Unstandardized coefficient. Standard error.

2 models (v d (1)=2.80) with a single degree of freedom, provides a test of the equality of the path for the two groups. Since this difference approaches significance at the pc0.09 level, we may conclude that the direct path between purposive value and group norms is marginally greater for network- when compared to small-group-based virtual communities. Other tests were conducted similarly. As can be seen from Table 7, in testing Hypothesis 15, three of the paths are significantly greater for the

network- relative to the small-group-subsample, and one path (self-discovery value to social identity) is not. These findings are supportive of Hypothesis 15. The results for Hypothesis 16 are mixed. The paths from maintaining interpersonal connectivity to group norms and social identity are significantly higher for small-group-based versus network-based virtual communities. However, the paths from social enhancement to neither of these variables are significantly different for the two subsamples. Hypothesis 16 thus

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

259

receives support for maintaining interpersonal connectivity but not for social enhancement value.

4. General discussion Our empirical survey-based study, conducted across a variety of different virtual communities, found overall support for our proposed social influence model of virtual community participation. The findings suggest that an appropriate conceptualization of intentional social action in virtual communities is one where the community’s influence on members stems from an understanding or expectation of various benefits that participants seek to attain from social interactions therein. Further, we also found that there are interesting differences between network-based and small-group-based virtual community participants in both levels of selfreferent and group-referent motives, as well as their impacts on the social influence variables. These findings raise several interesting issues discussed below. 4.1. Understanding how to deliver value desired by virtual community participants For participants of network-based virtual communities, purposive value was found to be a key driver of participation. From a managerial perspective, such purposive motives can be characterized as complementary to each other. For instance, in measuring informational value, one item that we used was bto get information,Q whereas another one was bto provide information to others.Q It can be argued that an information-seeker will find the virtual community useful only if he or she can find another participant with the complementary motive of providing that information. As a result, an important task of networkbased virtual community managers may be defined in terms of matching of participants’ complementary motives effectively and maintaining a balance so that the purposive goals of most participants are achieved. The finding that social benefits such as maintaining interpersonal connectivity and social enhancement are significant drivers of participation in small-groupbased virtual communities is also noteworthy. Since it suggests that many participants in such communities

are interested in engaging in social interactions together, as a group, the marketer’s objective may be defined in terms of matching group members’ preferences to interact together. These differences also imply that virtual community organizers will need to thoughtfully decide on which tools and functionalities to provide in their venues. In network-based virtual communities, members may find bapplications of purposeQ—tools, application, and content that enables them to achieve their goals successfully—to be valuable. Examples of such applications include comprehensive FAQs lists, organization of past responses from community members in transparent and easily accessible hierarchies, query-tools to match information-seekers to information providers, and so on. In contrast, in small-group-based communities, bapplications of processQ that enable uninterrupted, vivid and enjoyable group interaction may be more valuable. Aesthetic and easily learnt user interfaces, the ability to engage in interactive communication (see Section 4.2), and tools allowing members to contact and solicit the participation of group members, are all examples of applications of process. Understanding the relative importance and effectiveness of different applications of process and applications of purpose that enable self- and group-referent motives to be attained through future research seems especially important in future research. 4.2. Venue characteristics and virtual community type In our pretest, we found internet venues to strongly favor either network- or small-group-based virtual communities, suggesting that managers might be able to influence the type of communities that they organize by offering specific venues to their consumers. A closer examination of these venues suggests that the type of communication processes therein— whether interactive or non-interactive—play an important role in determining whether network- or small-group-based virtual communities dominate. The degree of interactivity of communication processes within a virtual community is influenced by both, the synchronicity of communication—the capability of enabling a participant to formulate and deliver a response in real time, allowing a real-time dialogue to occur (Hoffman & Novak, 1996), and by

260

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

the number and range of inputs that the participant can provide, such as text, audio, video, etc. (Lombard, 2001). Whereas communication researchers have studied the impacts of synchronicity and input attributes on the outcomes of communication processes (e.g., Lombard, 2001), we know relatively little about the marketing-relevant impacts of these variables for virtual communities. There are several reasons why higher levels of interactivity may be more suitable for small-groupbased and lower levels of interactivity more conducive to network-based virtual communities. First, researchers have shown that a high level of interactivity generally entails a higher level of involvement on the participant’s part (Hoffman & Novak, 1996). This implies that in interactions involving high interactivity, stronger relationships between participants may be necessary; in addition, participants should be responsive and engaged throughout the duration of the interaction—such requisites all the more characteristic of small groups. Second, the greater the interactivity afforded by the venue, the higher the likelihood of spontaneity between the participants, the more the possibility of interruption or preemption, and the greater the mutuality and patterns of turn-taking (Brown & Yule, 1983; Zack, 1993). Again, interactions of this type are possible when participants know and understand each other well. Third, it is more likely that interactive exchanges will continue and be repeated at future times when participants can engage in many different topics of conversation, move easily from one topic to the next, and have at least some shared history or knowledge base, all requiring broad-based relationships. Confirming this prediction, as well as studying the specific influences of interactivity and its constituents on the economic activities within virtual communities, are promising future research issues. 4.3. Understanding how to convey member information to other participants The mode by which the identity and information about a member is conveyed to others is also likely to be influenced by community type. In networkbased communities, because members don’t know each other at first in most cases, and their motives

are self-referent, a member’s reputation is likely to be crucial as a means of establishing trust and status and for fostering social interactions (Resnick, Zeckhauser, Friedman & Kuwabara, 2000; Rheingold, 2002). Reputation mechanisms considering contribution frequency and quality made may therefore need to be carefully and elaborately designed for such communities. Communities such as slashdot also choose to leave a visible trail of each member’s contribution history for other members to see and judge them. On the other hand, because small-group-based community members know each other well and participate to achieve group-referent goals, reputation systems may not be required or may be less essential. Instead, in this case, it may prove more useful to enable members to share a detailed personal, selfcomposed history with other members—for instance, through bAbout MeQ web-pages or in-depth member profiles. On the whole, we know relatively little about the importance of different information elements of a participant’s reputation or other identifying information, and when or how such measures are used by participants to make interaction decisions within the virtual community (see Rheingold, 2002 for a detailed discussion). These questions offer interesting future research opportunities. 4.4. Concluding thoughts Through studying the antecedents of social influence, and making and elaborating on the distinction between network- and small group-based communities, our broad objective in this exploratory study was to stimulate thinking among practitioners and researchers regarding the scope of virtual communities for marketing applications. Our contention is that marketers for the most part have tended to view virtual communities narrowly, focusing entirely on network-based communities. Through our presentation, we defined and elaborated on a second type, the small-group-based virtual community, found empirical differences, and developed some practically useful distinctions between them. The following two issues deserve more elaboration. First, more development is needed into expanding the conceptual difference between venues and virtual communities. For example, it is important to note that

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263

261

the same venue, such as a slashdot bulletin-board, may possibly host both network- and small-groupbased virtual communities at the same time. Indeed, even a particular person may belong to both virtual communities within this venue as when she exchanges ideas frequently with her regular group of friends weekly, yet sometimes reads messages on bulletinboards to update her knowledge on current software issues. But as we noted, high levels of interactivity and other features such as applications of process, imply that certain venues are more conducive for small-group-based virtual communities, and others for network-based virtual communities. Indeed, marketing managers may be able to influence the type of communities that are built within their venue through an informed selection of the venue characteristics discussed above. Second, as noted before, it is possible, indeed very likely that some groups within network-based virtual communities may over time evolve into small-groupbased virtual communities, as frequent interactions among the same individuals result in greater knowledge and the building of interpersonal relationships (see Alon et al., 2004 for a detailed discussion). But until we learn more about the conditions leading to such transitions, it will be difficult to draw conclusions or make inferences regarding a particular online group’s future type, based on its current type. Building on our consumer-centric definition presented earlier, the monitoring and management of a virtual community is best viewed as an ongoing task by its organizers. A final point we wish to make is to summarize the differences between our model and the B&D (2002) model. Decision making in our model is a direct function of social influence and an indirect function (through social influence) of value perceptions, whereas decision making in the B&D model is a direct function of both social influence and individual-level variables. Our model therefore makes stronger predictions in the sense that a particular sequence is specified among social- and individual-level antecedents, whereas these are left as exogenous predictors in the B&D model. Second, our model proposes five specific categories of value perceptions derived from the communications literature, whereas B&D rely upon general, summary variables derived from the theory of planned behavior (i.e., attitudes, subjective norms,

perceived behavioral control) and the model of goaldirected behavior (i.e., positive and negative anticipated emotions). Our antecedents have more managerial relevance than those found in B&D. Third, unlike B&D, we developed explanations based on contingencies inherent in different types of virtual communities. Finally, our tests of the model went a step beyond B&D’s tests by including participation behavior as a dependent variable. In spite of these contributions, it is important to recognize the exploratory nature of this research, and its attendant limitations. For instance, of the five hypothesized benefits, two—maintaining interpersonal interconnectivity and social enhancement—did not have significant effects on any of the variables. This suggests that more research is needed to determine all of the benefits, and differences between the two communities in this regard. In conclusion, it seems important to echo the optimism expressed by marketing scholars studying virtual communities (e.g., Balasubramanian & Mahajan, 2001), and suggest that these online forums are only likely to grow in importance, influence, and the activities for which they are used, as consumers become more comfortable and acclimatized with these environments and marketers learn how to forecast, monitor, and design their communication programs to take advantage of such opportunities. They merit continued and increasing attention from both practitioners and academicians.

Acknowledgements We would like to thank the participants of the bDefining the Value of Virtual CommunitiesQ special session at the 2002 AMA Summer Educators Conference for their comments. Special thanks to the editor and three anonymous reviewers.

References
Alon, A., Brunel, F. B., & Schneier Siegal, W. L. (2004). Ritual behavior and community life cycle: Exploring the social psychological roles of net rituals in the development of online consumption communities. In C. Haugvedt, K. Machleit, & R.

262

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 Dholakia, U. M., & Bagozzi, R. P. (2004). Motivational antecedents, constituents and consequents of virtual community identity. In S. Godar, & S. Pixie-Ferris (Eds.), Virtual and collaborative teams: Process, technologies, and practice (pp. 252 – 267). London7 IDEA Group. Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. Forthworth, TX7 Harcourt Brace Jovanovich. Ellemers, N., Kortekaas, P., & Ouwerkerk, J. W. (1999). Selfcategorization, commitment to the group, and group self-esteem as related but distinct aspects of social identity. European Journal of Social Psychology, 29, 371 – 389. Etzioni, A. (1996). The responsive community: A communitarian perspective. American Sociological Review, 61(1), 1 – 11. Flanagin, A. J., & Metzger, M. J. (2001). Internet use in the contemporary media environment. Human Communication Research, 27(1), 153 – 181. Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39 – 50. Hars, A., & Ou, S. (2002). Working for free? Motivations for participating in open-source projects. International Journal of Electronic Commerce, 6(3), 23 – 37. Hoffman, D. L., & Novak, T. P. (1996, July). Marketing in hypermedia computer-mediated environments: Conceptual foundations. Journal of Marketing, 60, 50 – 68. Hogg, M. A., & Abrams, D. (1988). Social identifications: A social psychology of intergroup relations and group processes. London7 Routledge. Joreskog, K., & Sorbom, D. (1999). Lisrel 8: User’s reference ¨ ¨ ¨ guide. (2nd ed.), Chicago7 Scientific Software International. Lombard, M. (2001). Interactive advertising and presence: A framework. Journal of Interactive Advertising, 1(2) (http:// www.jiad.org/vol1/no2/lombard/). Marsh, H. W., Balla, J. R., & Hau, K. (1996). An evaluation of incremental fit indices: A clarification of mathematical and empirical properties. In G. A. Marcoulides, & R. E. Schumacker (Eds.), Advanced structural equation modeling: Issues and techniques (pp. 315 – 353). Mahwah, NJ7 Erlbaum. Marsh, H. W., & Hovecar, D. (1985). Application of confirmatory factor analysis to the study of self-concept: First and higher order factor models and their invariance across groups. Psychological Bulletin, 97(3), 562 – 582. McKenna, K. Y. A., & Bargh, J. A. (1999). Causes and consequences of social interaction on the internet: A conceptual framework. Media Psychology, 1, 249 – 269. McMillan, D. W., & Chavis, D. M. (1986). Sense of community: A definition and theory. Journal of Community Psychology, 14, 6 – 23. McQuail, D. (1987). Mass communication theory: An introduction. (2nd ed.), London7 SAGE. Meyer, J. P., Stanley, D. J., Herscovitch, L., & Topolnytsky, L. (2002). Affective, continuance, and normative commitment to the organization: A meta-analysis of antecedents, correlates, and consequences. Journal of Vocational Behavior, 61, 20 – 52. Perugini, M., & Bagozzi, R. P. (2001). The role of desires and anticipated emotions in goal-directed behaviors: A model of

Yalch (Eds.), Online consumer psychology: Understanding how to interact with consumers in the virtual world. Hillsdale, NJ7 Erlbaum. Ashforth, B. E., & Mael, F. A. (1989). Social identity theory and the organization. Academy of Management Review, 14, 20 – 39. Bagozzi, R. P. (2000). On the concept of intentional social action in consumer behavior. Journal of Consumer Research, 27(3), 388 – 396. Bagozzi, R. P., & Dholakia, U. M. (1999). Goal setting and goal striving in consumer behavior. Journal of Marketing, 63, 19 – 32. Bagozzi, R. P., & Dholakia, U. M. (2002). Intentional social action in virtual communities. Journal of Interactive Marketing, 16(2), 2 – 21. Bagozzi, R. P., & Edwards, J. R. (1998). A general approach for representing constructs in organizational research. Organizational Research Methods, 1, 45 – 87. Bagozzi, R. P., & Yi, Y. (1988). On the evaluation of structural equation models. Journal of the Academy of Marketing Science, 16(1), 74 – 94. Balasubramanian, S., & Mahajan, V. (2001). The economic leverage of the virtual community. International Journal of Electronic Commerce, 5(3), 103 – 138. Baumeister, R. F. (1998). The self. In D. T. Gilbert, S. R. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (pp. 680 – 740). New York7 McGraw-Hill. Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107, 238 – 246. Bergami, M., & Bagozzi, R. P. (2000). Self-categorization, affective commitment, and group self-esteem as distinct aspects of social identity in an organization. British Journal of Social Psychology, 39(4), 555 – 577. Bhattacharya, C. B., & Sen, S. (2003). Consumer–company identification: A framework for understanding consumers’ relationships with companies. Journal of Marketing, 67(2), 76 – 88. Blanton, H., & Christie, C. (2003). Deviance regulation: A theory of action and identity. Review of General Psychology, 7(2), 115 – 149. Bratman, M. E. (1997). I intend that we J. In G. Homstrom¨ Hintikka, & R. Tuomela (Eds.), Contemporary action theory, vol. 2 (pp. 49 – 63). Dordrecht, the Netherlands7 Kluwer. Brown, G., & Yule, G. (1983). Discourse analysis. Cambridge, UK7 Cambridge University Press. Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen, & J. S. Long (Eds.), Testing structural equation models (pp. 136 – 162). Newbury Park, CA7 Sage. Catterall, M., & Maclaran, P. (2001). Researching consumers in virtual worlds: A cyberspace odyssey. Journal of Consumer Behaviour, 1(3), 228 – 237. Cerf, V. (1991). Networks. Scientific American, 265(3), 72 – 81. Cudeck, R. (1989). Analysis of correlation matrices using covariance structure models. Psychological Bulletin, 317 – 327. Dholakia, U. M., & Bagozzi, R. P. (2001). Consumer behavior in digital environments. In J. Wind, & V. Mahajan (Eds.), Digital marketing: Global strategies from the world’s leading experts (pp. 163 – 200). New York7 Wiley.

U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 goal-directed behavior. British Journal of Social Psychology, 40, 79 – 98. Postmes, T., Spears, R., & Lea, M. (2000). The formation of group norms in computer-mediated communication. Human Communication Research, 26(3), 341 – 371. Prentice, D., Miller, D. T., & Lightdale, J. R. (1994). Asymmetries in attachments to groups and to their members: Distinguishing between common identity and common bond groups. Personality and Social Psychology Bulletin, 20, 484 – 493. Resnick, P., Zeckhauser, R., Friedman, R., & Kuwabara, K. (2000). Reputation systems. Communications of the ACM, 43(12), 45 – 48. Rheingold, H. (2002). Smart mobs: The next social revolution. Cambridge7 Perseus. Sassenberg, K. (2002). Common bond and common identity groups on the internet: Attachment and normative behavior in on-topic and off-topic chats. Group Dynamics, Theory, Research, and Practice, 6(1), 21 – 37. Tajfel, H. (1978). Interindividual behavior and intergroup behavior. In H. Tajfel (Ed.), Differentiation between groups: Studies in the social psychology of intergroup relations (pp. 27 – 60). London7 Academic Press. Thorbjbrnsen, H., Supphellen, M., Nysveen, H., & Pedersen, P. E. (2002). Building brand relationships online: A comparison of two applications. Journal of Interactive Marketing, 16(3), 17 – 34. Tuomela, R. (1995). The importance of us: A philosophy study of basic social notions. Stanford, CA7 Stanford University Press.

263

Turner, J. C. (1985). Social categorization and the self-concept: A social cognitive theory of group behavior. In E. J. Lawler (Ed.), Advances in group processes (pp. 77 – 122). Greenwich CT7 JAI Press. Weingart, L. R., Bennett, R. J., & Brett, J. M. (1993). The impact of consideration of issues and motivational orientation on group negotiation process and outcome. Journal of Applied Psychology, 78(3), 504 – 517. Wellman, B. (1999). The network community: An introduction. In B. Wellman (Ed.), Networks in the global village: Life in contemporary communities (pp. 1 – 48). Boulder, CO7 Westview Press. Wellman, B., & Gulia, M. (1999). Net-surfers don’t ride alone: Virtual communities as communities. In B. Wellman (Ed.), Networks in the global village: Life in contemporary communities (pp. 331 – 366). Boulder, CO7 Westview Press. Wellman, B., Salaff, J., Dimitrova, D., Garton, L., Gulia, M., & Haythornthwaite, C. (1996). Computer networks as social networks: Collaborative work, telework, and virtual community. Annual Review of Sociology, 22, 213 – 238. Williams, R. L., & Cothrel, J. (2000). Four smart ways to run online communities. Sloan Management Review, 81 – 91. Zack, M. H. (1993). Interactivity and communication mode choice in ongoing management groups. Information Systems Research, 4(3).

References: Alon, A., Brunel, F. B., & Schneier Siegal, W. L. (2004). Ritual behavior and community life cycle: Exploring the social psychological roles of net rituals in the development of online consumption communities. In C. Haugvedt, K. Machleit, & R. 262 U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 Dholakia, U. M., & Bagozzi, R. P. (2004). Motivational antecedents, constituents and consequents of virtual community identity. In S. Godar, & S. Pixie-Ferris (Eds.), Virtual and collaborative teams: Process, technologies, and practice (pp. 252 – 267). London7 IDEA Group. Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. Forthworth, TX7 Harcourt Brace Jovanovich. Ellemers, N., Kortekaas, P., & Ouwerkerk, J. W. (1999). Selfcategorization, commitment to the group, and group self-esteem as related but distinct aspects of social identity. European Journal of Social Psychology, 29, 371 – 389. Etzioni, A. (1996). The responsive community: A communitarian perspective. American Sociological Review, 61(1), 1 – 11. Flanagin, A. J., & Metzger, M. J. (2001). Internet use in the contemporary media environment. Human Communication Research, 27(1), 153 – 181. Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39 – 50. Hars, A., & Ou, S. (2002). Working for free? Motivations for participating in open-source projects. International Journal of Electronic Commerce, 6(3), 23 – 37. Hoffman, D. L., & Novak, T. P. (1996, July). Marketing in hypermedia computer-mediated environments: Conceptual foundations. Journal of Marketing, 60, 50 – 68. Hogg, M. A., & Abrams, D. (1988). Social identifications: A social psychology of intergroup relations and group processes. London7 Routledge. Joreskog, K., & Sorbom, D. (1999). Lisrel 8: User’s reference ¨ ¨ ¨ guide. (2nd ed.), Chicago7 Scientific Software International. Lombard, M. (2001). Interactive advertising and presence: A framework. Journal of Interactive Advertising, 1(2) (http:// www.jiad.org/vol1/no2/lombard/). Marsh, H. W., Balla, J. R., & Hau, K. (1996). An evaluation of incremental fit indices: A clarification of mathematical and empirical properties. In G. A. Marcoulides, & R. E. Schumacker (Eds.), Advanced structural equation modeling: Issues and techniques (pp. 315 – 353). Mahwah, NJ7 Erlbaum. Marsh, H. W., & Hovecar, D. (1985). Application of confirmatory factor analysis to the study of self-concept: First and higher order factor models and their invariance across groups. Psychological Bulletin, 97(3), 562 – 582. McKenna, K. Y. A., & Bargh, J. A. (1999). Causes and consequences of social interaction on the internet: A conceptual framework. Media Psychology, 1, 249 – 269. McMillan, D. W., & Chavis, D. M. (1986). Sense of community: A definition and theory. Journal of Community Psychology, 14, 6 – 23. McQuail, D. (1987). Mass communication theory: An introduction. (2nd ed.), London7 SAGE. Meyer, J. P., Stanley, D. J., Herscovitch, L., & Topolnytsky, L. (2002). Affective, continuance, and normative commitment to the organization: A meta-analysis of antecedents, correlates, and consequences. Journal of Vocational Behavior, 61, 20 – 52. Perugini, M., & Bagozzi, R. P. (2001). The role of desires and anticipated emotions in goal-directed behaviors: A model of Yalch (Eds.), Online consumer psychology: Understanding how to interact with consumers in the virtual world. Hillsdale, NJ7 Erlbaum. Ashforth, B. E., & Mael, F. A. (1989). Social identity theory and the organization. Academy of Management Review, 14, 20 – 39. Bagozzi, R. P. (2000). On the concept of intentional social action in consumer behavior. Journal of Consumer Research, 27(3), 388 – 396. Bagozzi, R. P., & Dholakia, U. M. (1999). Goal setting and goal striving in consumer behavior. Journal of Marketing, 63, 19 – 32. Bagozzi, R. P., & Dholakia, U. M. (2002). Intentional social action in virtual communities. Journal of Interactive Marketing, 16(2), 2 – 21. Bagozzi, R. P., & Edwards, J. R. (1998). A general approach for representing constructs in organizational research. Organizational Research Methods, 1, 45 – 87. Bagozzi, R. P., & Yi, Y. (1988). On the evaluation of structural equation models. Journal of the Academy of Marketing Science, 16(1), 74 – 94. Balasubramanian, S., & Mahajan, V. (2001). The economic leverage of the virtual community. International Journal of Electronic Commerce, 5(3), 103 – 138. Baumeister, R. F. (1998). The self. In D. T. Gilbert, S. R. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (pp. 680 – 740). New York7 McGraw-Hill. Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107, 238 – 246. Bergami, M., & Bagozzi, R. P. (2000). Self-categorization, affective commitment, and group self-esteem as distinct aspects of social identity in an organization. British Journal of Social Psychology, 39(4), 555 – 577. Bhattacharya, C. B., & Sen, S. (2003). Consumer–company identification: A framework for understanding consumers’ relationships with companies. Journal of Marketing, 67(2), 76 – 88. Blanton, H., & Christie, C. (2003). Deviance regulation: A theory of action and identity. Review of General Psychology, 7(2), 115 – 149. Bratman, M. E. (1997). I intend that we J. In G. Homstrom¨ Hintikka, & R. Tuomela (Eds.), Contemporary action theory, vol. 2 (pp. 49 – 63). Dordrecht, the Netherlands7 Kluwer. Brown, G., & Yule, G. (1983). Discourse analysis. Cambridge, UK7 Cambridge University Press. Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen, & J. S. Long (Eds.), Testing structural equation models (pp. 136 – 162). Newbury Park, CA7 Sage. Catterall, M., & Maclaran, P. (2001). Researching consumers in virtual worlds: A cyberspace odyssey. Journal of Consumer Behaviour, 1(3), 228 – 237. Cerf, V. (1991). Networks. Scientific American, 265(3), 72 – 81. Cudeck, R. (1989). Analysis of correlation matrices using covariance structure models. Psychological Bulletin, 317 – 327. Dholakia, U. M., & Bagozzi, R. P. (2001). Consumer behavior in digital environments. In J. Wind, & V. Mahajan (Eds.), Digital marketing: Global strategies from the world’s leading experts (pp. 163 – 200). New York7 Wiley. U.M. Dholakia et al. / Intern. J. of Research in Marketing 21 (2004) 241–263 goal-directed behavior. British Journal of Social Psychology, 40, 79 – 98. Postmes, T., Spears, R., & Lea, M. (2000). The formation of group norms in computer-mediated communication. Human Communication Research, 26(3), 341 – 371. Prentice, D., Miller, D. T., & Lightdale, J. R. (1994). Asymmetries in attachments to groups and to their members: Distinguishing between common identity and common bond groups. Personality and Social Psychology Bulletin, 20, 484 – 493. Resnick, P., Zeckhauser, R., Friedman, R., & Kuwabara, K. (2000). Reputation systems. Communications of the ACM, 43(12), 45 – 48. Rheingold, H. (2002). Smart mobs: The next social revolution. Cambridge7 Perseus. Sassenberg, K. (2002). Common bond and common identity groups on the internet: Attachment and normative behavior in on-topic and off-topic chats. Group Dynamics, Theory, Research, and Practice, 6(1), 21 – 37. Tajfel, H. (1978). Interindividual behavior and intergroup behavior. In H. Tajfel (Ed.), Differentiation between groups: Studies in the social psychology of intergroup relations (pp. 27 – 60). London7 Academic Press. Thorbjbrnsen, H., Supphellen, M., Nysveen, H., & Pedersen, P. E. (2002). Building brand relationships online: A comparison of two applications. Journal of Interactive Marketing, 16(3), 17 – 34. Tuomela, R. (1995). The importance of us: A philosophy study of basic social notions. Stanford, CA7 Stanford University Press. 263 Turner, J. C. (1985). Social categorization and the self-concept: A social cognitive theory of group behavior. In E. J. Lawler (Ed.), Advances in group processes (pp. 77 – 122). Greenwich CT7 JAI Press. Weingart, L. R., Bennett, R. J., & Brett, J. M. (1993). The impact of consideration of issues and motivational orientation on group negotiation process and outcome. Journal of Applied Psychology, 78(3), 504 – 517. Wellman, B. (1999). The network community: An introduction. In B. Wellman (Ed.), Networks in the global village: Life in contemporary communities (pp. 1 – 48). Boulder, CO7 Westview Press. Wellman, B., & Gulia, M. (1999). Net-surfers don’t ride alone: Virtual communities as communities. In B. Wellman (Ed.), Networks in the global village: Life in contemporary communities (pp. 331 – 366). Boulder, CO7 Westview Press. Wellman, B., Salaff, J., Dimitrova, D., Garton, L., Gulia, M., & Haythornthwaite, C. (1996). Computer networks as social networks: Collaborative work, telework, and virtual community. Annual Review of Sociology, 22, 213 – 238. Williams, R. L., & Cothrel, J. (2000). Four smart ways to run online communities. Sloan Management Review, 81 – 91. Zack, M. H. (1993). Interactivity and communication mode choice in ongoing management groups. Information Systems Research, 4(3).