Influence of community design on user behaviors in online communities

Journal of Business Research 67 (2014) 2258–2268

Contents lists available at ScienceDirect

Journal of Business Research

Influence of community design on user behaviors in online communities☆ Marina Fiedler a,⁎, Marko Sarstedt b,c a b c

University of Passau, Innstraße 27, 94032 Passau, Germany Otto-von-Guericke-University Magdeburg, Universitätsplatz 2, 39106 Magdeburg, Germany University of Newcastle, Newcastle Business School, Faculty of Business and Law, University Drive Callaghan, NSW 2308, Australia

a r t i c l e

i n f o

Article history: Received 1 October 2013 Received in revised form 1 March 2014 Accepted 1 May 2014 Available online 7 July 2014 Keywords: Online communities Community design Design principles Member attachment

a b s t r a c t While the question of how community design influences user behavior in online communities has recently attracted considerable research, few studies have empirically evaluated the influencing factors of specific user behavior. Building on a conceptual framework of identity-based vs. bond-based attachment in online communities, this study evaluates the influence of several antecedents on user attachment as well as attachment's mediating role for explaining consumer behavior. Results of a survey reveal that network effects, intergroup comparison, and social categorization have a positive and significant effect on common identity attachment, whereas this is not the case with in-group interdependence. Conversely, collectivism, interpersonal similarity, and social interaction drive common bond attachment, while personal information has no effect. Most importantly, the results show that common identity attachment is the primary driver of user behavior in online communities. © 2014 Published by Elsevier Inc.

1. Introduction Online communities are a relatively new way to build relationships among physically remote individuals. They enable people to meet friends or to make new acquaintances and to exchange opinions and media such as texts, videos, and pictures. Armstrong and Hagel (1996) were among the first to propagate the benefits of this new form of community for business applications. Today, many companies—such as Microsoft, P&G, Sony, IBM, and Ducati—acknowledge the potential of online communities. They use this channel to recognize consumers' varying needs and wants, or to allow their knowledge to be incorporated into the product design and the service delivery process (e.g., Casaló, Flavián, & Guinalíu, 2013; Nambisan & Watt, 2011). The number of online communities is continually increasing. With annual growth rates of more than 100%, online communities such as Facebook and MySpace are competing with established Internet portals such as Google and Yahoo. However, some communities are more successful than others. For every flourishing community, a moribund social networking site exists

☆ The authors thank Martina Angelova, Elisa Barskaia, Torsten Clauß, Silvia Heer, Silke Schuppler, Timo Thoennissen, and Fei Wu for their excellent research assistance. A prior version of this paper was presented at the 2010 International Conference on Information Systems (ICIS), St. Louis, USA. The order of authors is alphabetical; both authors contributed equally to this work and should be considered corresponding authors. ⁎ Corresponding author. E-mail addresses: fi[email protected] (M. Fiedler), [email protected] (M. Sarstedt).

http://dx.doi.org/10.1016/j.jbusres.2014.06.014 0148-2963/© 2014 Published by Elsevier Inc.

that attracts too few members to remain viable. Furthermore, numerous online communities have attracted many members but have failed to maintain active relationships (Lee, Lee, Taylor, & Lee, 2011). For example, across a wide range of Usenet groups, more than 60% of the newcomers who post a message in any given month in a group are never seen again (Arguello et al., 2006). All online communities embody technical and social choices that influence how members interact with information as well as community members. Community designers make several large and small design decisions regarding a community's site features (e.g., concerning interactions, organization structures, and policies) that influence how a community motivates participants (e.g., Ren, Kiesler, & Kraut, 2007; Ren and Kraut, 2012). Consider, for example, Harley–Davidson's Harley Owners Club, a prototypical example of a brand community cited in many marketing research studies (e.g., Algesheimer, Dholakia, & Herrmann, 2005; Bagozzi & Dholakia, 2006). Should this website's design—through its moderation, policies, or structure—discourage members from having off-topic discussions? Such discussions can be distracting, and they may be especially offputting to newcomers whose initial expectations are likely to be violated. On the other hand, off-topic discussion provides opportunities for self-disclosure and friendship. As Ren et al. (2007, p. 279) point out, “if designers discourage off-topic discussion, they might lose people who would like to talk with others like themselves. Discouraging off-topic discussion also may annoy old-timers, who have gotten to know each other. Thus, the choice that community designers make about off-topic discussion can influence who joins

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

the community and who stays.” Similar trade-offs occur when designers determine whether to limit an online community's size or allow unlimited growth, whether to cluster users into communities of interest or provide unstructured access to all content, or whether to require members to register with a verifiable identity or allow them to participate in the community anonymously (Ren et al., 2007). As such, the environment of an online community is similar to the atmosphere of retail stores, with its emphasis on layout, merchandise displays, lighting, or signage (Mummalaneni, 2005). The question of how online community design can trigger certain user behavior has attracted considerable attention in marketing research, especially in terms of describing brand community member behavior (e.g., Algesheimer et al., 2005; Bagozzi & Dholakia, 2006; Schau, Muñiz, & Arnould, 2009; Zaglia, 2013; Zhou, Zhang, Su, & Zhou, 2012). However, the literature provides little theorybased knowledge to predict how or why specific policies and features make online communities successful in engaging and retaining members (Ren et al., 2012). An exception is Ren et al.'s (2007) comprehensive framework based on the social psychological theories of common identity and common bonds. Common identity and common bond attachment refer to the different reasons for being in a community, that is, because users like a community as a whole (identity-based attachment), or because they like individuals in a group (bond-based attachment). Specifically, common identity attachment implies that members feel a commitment to an online community's purpose, whereas common bond attachment implies that members feel socially and emotionally attached to specific community members. Hence, if community members experience common identity attachment, belonging to this community is very important to them. They feel good when being described as a member of this community, and they feel a strong attachment to the community as a whole. If community members experience common bond attachment, they primarily feel very close to specific members of this community. Recently, Ren et al. (2012) found that online community features intended to foster identity-based attachment have stronger effects on participation and retention than features intended to foster bond-based attachment. Furthermore, the authors found that removing off-topic messages increases members' information benefits at the expense of their opportunities to develop online relationships. While these results provide support for several tenets of Ren et al.'s (2007) model, most of the original specified relationships remain research hypotheses. Specifically, the relationship between identitybased and bond-based attachment and their numerous antecedents, as well as the mediating role of attachment types for explaining user behavior, have not yet been empirically evaluated as an entire model. The central aim in this article is to understand how community design influences user behavior in online communities, with a special focus on the influence of bond-based and identity-based attachments. To simultaneously evaluate the complex cause-andeffect relationships, the study draws on data from two large-scale surveys gathered at two different points in time. This allows for understanding the influence of a number of theoretically validated antecedents, such as in-group interdependence, intergroup comparison, and social interaction on common identity and common bond attachment in online communities. In so doing, this study supplements Ren et al.'s (2007) original model by introducing two additional antecedents of attachment: network effects and collectivism. This study also examines the effects of the two different attachment styles on user behaviors including on-topic and off-topic discussion, generalized and direct reciprocity, and membership robustness and in-group loyalty. The study provides valuable insights into the complex relationships that underlie online community user behavior and yields important implications for community design.

2259

2. Theoretical framework As a basis for understanding the antecedents and consequences of identity-based and bond-based attachments, this study primarily draws on Ren et al. (2007) and augments their model with economic and social psychological literature. The following sections present a series of hypotheses regarding the relationship between (1) potential antecedents and identity-based as well as bondbased attachments, and (2) the two types of relationship attachments and potential behavioral consequences.

2.1. Antecedents of common identity attachment Online communities allow for large network sizes and effects owing to spontaneous, interactive exchange among community members across geographic distances (McAfee, 2006). The literature on network effects indicates that network size has a considerable impact on market share, shaping customer choices and technology competition (Lee, Lee, & Lee, 2006). Studies on social networks find that open networks with many weak ties and social connections are more likely to introduce new ideas to their members than closed networks with many redundant ties (e.g., Scott, 1991). Several studies also emphasize the relevance of network size and effects for information sharing (e.g., Tsai, 2002). Network theory suggests that an online community needs a critical mass of users who generate content and canvass new members in order to be successful (e.g., Preece & Maloney-Krichmar, 2003). If the member base grows, the potential contact possibilities increase and the number of potential contributions multiplies; in turn, this can increase the community's usefulness to members (O'Murchu, Breslin, & Decker, 2004) and, since the establishment of bondbased attachment seems more dependent on interpersonal social relationships than network size, this fosters identity-based rather than bond-based attachment with the community (Ren, Kraut, Kiesler, & Resnick, 2011). Hence: H1a. Network effects relate positively to identity-based group attachment. Sherif and Sherif (1953) first analyzed the relationship between ingroup interdependence and social identity, and concluded that a shared fate strengthens group identity awareness. In-group interdependencies most often emerge when a common goal, common reward, or an external threat exists (Flippen, Hornstein, Siegal, & Weitzmann, 1996). Therefore, group members behave cooperatively because they depend on one another concerning the results of their actions inasmuch as they have a common destiny (e.g., Cheng, 1996). Thus: H1b. In-group-interdependence relates positively to identity-based group attachment. Social identity theory implies that group membership induces identity (Tajfel, Billic, Bundy, & Flament, 1971). Specifically, group members' self-categorization triggers the development of social identity (Tajfel et al., 1971). Members evaluate organizations through intergroup comparisons (Brickson, 2007). Group comparisons are therefore decisive for individuals' identification with social groups to which they belong (Yuki, 2003). Whether or not the comparison group is physically present is not important for identification with a group (Utz, 2003). According to Ren et al. (2007), intergroup comparison in online communities often emphasizes competitive elements by demonstrating a group's superiority. For example, postings on the frequently asked questions (FAQ) sections of apache.org, home of the Apache web server open-source development project, compare the Apache server's speed, performance, and market share with other commercial servers, fostering the common identity of those who work on Apache software. Wikipedia

2260

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

uses a similar technique, for example by highlighting competition with other encyclopedias. Jointly, these examples underline the importance of intergroup comparisons for establishing social identity. Thus: H1c. Intergroup comparison relates positively to identity-based group attachment. According to self-categorization theory, an individual has not only a personal identity, but also several other social identities that may come to the fore in different situations (Rogers & Lea, 2005). Research suggests that social identity can be fostered by objective similarity (e.g., regarding age, education, and gender), perceived similarity (e.g., regarding interests, convictions, and attitudes) or degree of familiarity (e.g., quantity and quality of communication) (e.g., Mollica, Gray, & Trevino, 2003). Tajfel and Turner (1986) conclude that intergroup categorization leads to in-group favoritism and discrimination against the out-group. Hence: H1d. Social categorization relates positively to identity-based group attachment.

The pursuit of harmony, group orientation, and a strong social interdependence between individuals shape collectively oriented individuals' group attachment process. Individuals characterized by collectivism therefore tend to view interpersonal relationships as the basis for the formation of bond-based attachment. Hence, H2c. Collectivistic orientation relates positively to bond-based attachment. Social interactions are crucial in establishing social cohesiveness, shared values, and expectations of collaborative interpersonal relationships (e.g., Jarvenpaa & Leidner, 1999). Social interaction can increase so-called “other regarding preferences” such as altruism and fairness, which in turn can lead to increased bond-based attachment (Buchan et al., 2006). Similarly, social interaction increases common bond attachment through opportunities for self-disclosure and friendship (Preece & Maloney-Krichmar, 2003). Lastly, social interaction can appeal to a person's mood and feeling of self-content (Keltner & Bonanno, 1997), which—in turn—can lead to positive emotions about themselves and others, increasing their bondbased attachment. Hence, H2d. Social interaction relates positively to bond-based attachment.

2.2. Antecedents of common bond attachment Research into interpersonal relationships assumes that direct contact with other members leads to stronger bonds between them, reinforcing group cohesion (Buchan, Croson, & Johnson, 2006). Collins and Miller (1994) find that people who reveal personal information are more popular with others and that the self-revelation process leads to mutual sympathy, which is an important precondition for social bonds. Members of social networking communities could, for instance, discover commonalties by revealing personal interests or hobbies, which in turn increases the probability of social interaction (Ren et al., 2007). Members can also easier trust other community participants when they know more about their personalities (Ridings, Arinze, & Gefen, 2002). Therefore: H2a. Personal information relates positively to bond-based group attachment. Robust findings from psychology indicate that the desire for similarity strongly influences choices (e.g., Mollica et al., 2003; Sommers, 2006). From an economic perspective, this can be explained with reduced costs of adaptation of one's mental models and representations, as well as increased ease of communication. Furthermore, similarity is rewarding because it positively confirms one's self-image and supports the need to belong. Similarity between an online community's members thus may create sympathy and thereby provides the basis for social relationships with other community members. Therefore: H2b. Interpersonal similarity relates positively to bond-based attachment. One fundamental component of interpersonal cooperation is the degree of individualism or collectivism internalized by a person as a guiding norm (e.g., Zhang, Fu, Lowry, & Zhou, 2007). Individualism relates to the extent to which persons view a situation as competition and place emphasis on maximizing their individual payoff. Collectivism reflects both a cooperative understanding of a situation and seeks to prioritize maximizing group payoffs. According to Frank (1988), two individual disposition types exist: cooperative individuals cooperate, even in anonymity, while individualistic persons seek to maximize their own payoffs. Buchan et al. (2006) find evidence that collectively oriented participants show more trust and trustworthiness than individualistically oriented participants. Similarly, Yuki (2003) shows that social categorization processes hardly play a role for attachment in individuals with a more collectivist orientation. Intact social bonds with other people are much more important for collectivists than for individualists.

2.3. Behavioral effects of common identity and bond attachment Communication is at the core of online communities, determining their long-term success (Casaló et al., 2013). The nature of communication exchanges is likely to depend on the type of attachment that members have to a group (Ren et al., 2007). Back (1951) found that identity-based groups primarily engage in discussions relevant to achieving a certain goal, whereas bond-based groups engage in longer conversations on a broader range of topics. Several decades later, members of online communities that engaged in discussions on a narrow range of topics reported high group identity, while those who discussed a wide variety of topics stated that other members of the group are more likeable (Sassenberg, 2002). Based on these findings, Ren et al. (2007) conclude that members who experience common identity with their community will be more likely to engage in on-topic discussions. Conversely, community members that feel bond-based attachment will be more tolerant of off-topic discussions. Hence, H3. Common identity attachment relates positively to on-topic discussion. H4. Common bond attachment relates positively to off-topic discussion. The norm of reciprocal behavior is considered universally applicable in all cultures (Gouldner, 1960). This norm is constructed from the following two behavior rules: (1) people should not harm those who have helped them, and (2) people should help those who have helped them. Since common identity relates to being committed to a community's purpose, this attachment type should positively influence generalized reciprocity (Ren et al., 2007). Conversely, members with bond-based attachment to a community are more likely to exchange help only with particular others (Ren et al., 2007). Hence: H5. Common identity attachment relates positively to generalized reciprocity. H6. Common bond attachment relates positively to direct reciprocity. Prentice, Miller, and Lightdale (1994) argue that the group attachment type can influence a group's stability. They expect that common identity attachment concerning membership change leads to more stability over time than common bond attachment. They assume that a common bond group will quickly become uninteresting when friends leave,

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

2261

Network effects In-group Interdep.

H1a H1b H1c

Common identy

H1d

Intergroup comp. Social categor.

H3 H8

Personal informaon Interpers. similarity

On-topic discussion Off-topic discussion Generalized reciprocity Direct reciprocity Robustness In-group loyalty

H2a H2b H2c

Common bond

H2d

Collecvism Social interacon

Antecedents of aachment

Types of aachment

Behavioral consequences of aachment

Fig. 1. Theoretical framework of antecedents and behavioral consequences of attachment types.

whereas common identity attachment is independent of specific bonds. Hence, H7. Common identity attachment relates positively to robustness with respect to membership turnover. Finally, since both group attachment types lead members to perceive a group as cohesive and to evaluate their group more favorably than other groups, Ren et al. (2007) assume that common identity attachment and common bond attachment are related to in-group loyalty. Ren et al. (2011, p. 81) provide a more differentiated description of these relationships, underlining that “when people identify with a community or group as a whole, they tend to perceive other members in the group as interchangeable… One consequence is that their commitment to the group is stable in the face of turnover in membership, at least in comparison to bond-based attachment.” Hence, H8. Common identity attachment exerts a stronger influence on ingroup loyalty than common bond attachment. Fig. 1 summarizes the hypotheses on the antecedents and behavioral consequences of identity-based and bond-based attachments to online communities.

3. Measures and data The operationalization of the constructs primarily draws on measures commonly used in prior research. Since all measures denote manifestations of the underlying construct, this study uses reflective items—as opposed to formative ones—in all measurement models (Diamantopoulos, Riefler, & Roth, 2008). Table A1 in the Appendix A provides an overview of all construct measures. To account for common method bias, the measurement of the independent and dependent variables was done at two different points in time. Whereas the measures for the criterion variables stem from the first survey, measures for bond-based and identity-based attachments as well as their antecedents derive from the second survey.

The first survey, which took place in October to November 2007, relied on a list of 470 online communities, drawn up by means of an Internet search. This list represented a wide selection of all possible online communities concerning their community focus, size, and geographical distribution. Where possible, potential participants were contacted by posting a link to these online communities. Alternatively, online community members were contacted directly via email. Of the 1900 persons who started the survey, 1013 (53%) completed their questionnaires. Respondents who made their email addresses available were invited in December 2007 to January 2008 to participate in the follow-up study. This procedure yielded a dataset with 147 respondents who answered all questions related to antecedents and attachment types as well as the behavioral consequences across the two surveys. Contingency tests in respect of different socio-demographic characteristics (e.g., age, gender, and education) as well as behavioral characteristics (e.g., hours spent in online communities) did not indicate any significant effects (Mooi & Sarstedt, 2011). The comparison of early and late respondents did not indicate any significant differences concerning demographic variables, suggesting that nonresponse bias is of no concern (Armstrong & Overton, 1977). The percentage of women and men remained relatively balanced, with approximately 45% female and 55% male participants in both the main and follow-up investigation. The average interviewee age was close to 26, with the youngest being 15 and the oldest 65. Most respondents (approximately 60%) were between 20 and 30 years of age. Most of the survey participants had completed a higher education degree: 41.6% had a university degree, 38.3% had a university-entrance qualification, and 20.1% had a lower educational level or did not indicate their education level. Working persons (24.9%) and students (49.0%) constituted the major part of the sample survey. 40.3% of the interviewees reported that they were from China, 25.5% from Germany, and 22.1% from Bulgaria, while the remaining participants were from the U.S. (4.7%), France (3.4%), or other countries (3.4%). The respondents also had to indicate which community they were most actively engaged in. Overall, 24.5% of the communities can be classified under social networks (e.g., Facebook, MySpace), and 23.8% under

2262

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

(product category-related) discussion forums (e.g., hardwareBG, teenzona). Surprisingly, the majority of communities (approximately 40%) do not involve a specific purpose or activity (e.g., Kaiyuan, 6park, and Net-iris). Such general interest communities offer the opportunity of starting personal relationships and talking about general topics of interest. Given the difficulty of clearly distinguishing between communities' aims, the analysis uses all respondents—rather than engaging in group comparisons between common bond and common identity communities—in the evaluation of the antecedents and consequences of common bond and common identity attachments. However, to ascertain that (unobserved) heterogeneity does not bias the results, the analysis also uses a latent class detection approach on the data (Hair, Sarstedt, Ringle, & Mena, 2012). The average length of the interviewees' membership of a community is approximately two years. Approximately 22% of the study participants constitute inexperienced users with a membership of less than six months. Around 45% of the interviewees have been members of an online community for one to two years. The remaining 33% are experienced users, since they have belonged to a community for at least three years. On average, the respondents spend approximately 10 h per week in their most active community.

4. Analysis and results 4.1. Methodology and models To test the hypotheses, this study uses structural equation modeling (SEM) on the data. When estimating structural equation models, researchers can employ either covariance-based methods or the variance-based partial least squares (PLS-SEM) approach. Drawing on the findings of comparisons of the two approaches by Hair, Ringle, and Sarstedt (2011), Henseler et al. (2014) as well as Reinartz, Haenlein, and Henseler (2009), for example, PLS-SEM was applied because of the research model's complexity in relation to the number of observations. Since the PLS-SEM algorithm consists of several ordinary least squares regressions for separate subparts of the focal path model, the model's overall complexity hardly influences sample size requirements. Consequently, PLS-SEM is most beneficial when applied to models with complex cause-effect relationships—such as in this study—because the approach avoids improper or nonconvergent solutions (e.g., Hair et al., 2011). Reinartz et al. (2009) show that that PLS-SEM is an appropriate methodological choice if sample size is low, since the method achieves higher levels of statistical power compared to its covariance-based counterpart. To compare the model setups, the analysis examines separate path models for each criterion variable. Specifically, the analysis comprises six path models, with on-topic discussion (Model 1), off-topic discussion (Model 2), generalized reciprocity (Model 3), direct reciprocity (Model 4), robustness (Model 5), and in-group loyalty (Model 6) as the criterion variables. For parameter estimation, this study draws on

Table 1 Reliability and convergent validity estimates.

Network effects In-group interdependence Intergroup comparison Social categorization Personal information Interpersonal similarity Collectivism Social interaction Common identity Common bond

Min. (α)

Min. (ρc)

Min. (AVE)

0.73 0.86 0.88 0.83 0.84 0.83 0.81 0.87 0.88 0.77

0.84 0.91 0.91 0.88 0.90 0.90 0.87 0.90 0.91 0.85

0.64 0.77 0.62 0.59 0.76 0.75 0.63 0.65 0.62 0.59

SmartPLS software (Ringle, Wende, & Will, 2005), applying the path weighting scheme (Hair et al., 2012) and bootstrapping, using the no sign change option, 5000 bootstrap samples and a number of cases equal to the sample size used in each model. Table A2 in the Appendix A provides an overview of descriptive statistics and correlations among the constructs. 4.2. Results Methodological considerations relevant to the PLS-SEM analysis include the assessment of the measures' reliability, convergent and discriminant validity, as well as determining the models' predictive capabilities (Hair, Hult, Ringle, & Sarstedt, 2013). Table 1 shows the minimum reliability and validity estimates for all constructs related to antecedents and attachment types across the six models. The related results for the behavioral consequences of attachment are shown in Table 3, which also includes the parameter estimates for the structural models. Cronbach's α and composite reliability ρc facilitate the assessment of the construct measures' reliability. All constructs achieve α and ρc values higher than the generally stipulated threshold of 0.70 (Hair et al., 2012). The results thus indicate a high degree of reliability. All indicators' loading estimates achieve high levels across all model setups (i.e., above 0.70) and are significant. Accordingly, the average variance extracted (AVE) values across all models are acceptable. The analysis results (Tables 1 and 3) reveal that (minimum) AVE scores calculated across all model setups are above 0.50, thus providing support for the construct measures' convergent validities (Hair et al., 2012). Two approaches facilitate the assessment of the constructs' discriminant validity. The first analysis involves examining the indicators' cross-loadings, which reveals that no indicator loads higher on any opposing construct (Hair et al., 2013). Second, the Fornell and Larcker (1981) criterion compares the square root of the constructs' AVEs with the construct correlations. This analysis shows that across all model setups and constructs, each latent variable shares more variance with its own block of indicators than with another latent variable representing a different block of indicators. Thus, both approaches provide support for the constructs' discriminant validities, indicating that there are no collinearity issues among the constructs. Criteria applied for evaluating the models' predictive capabilities include R2 and the cross-validated redundancy measure Q2 (Hair et al., 2012). The analysis shows that the R2 values of the two mediating constructs—common bond and common identity—are acceptable. Likewise, all Q2 values are significantly larger than 0, providing support for the exogenous constructs' predictive relevance (Table 2). With respect to the structural model, results are highly consistent across all model setups, providing support for the research model's robustness. Specifically, network effects, intergroup comparison, and social categorization have a positive and significant (p ≤ 0.01) effect on common identity, whereas this is not the case with in-group interdependence. Thus, the results provide support for hypotheses H1a, H1c, and H1d. Common bond is significantly (p ≤ 0.05) driven by interpersonal similarity, collectivism, and social interaction, whereas personal information has no significant positive effect on the interpersonal bonds in online communities, which supports hypotheses H2b, H2c, and H2d (Table 2). Table 3 provides path coefficient estimates for the relationships between common identity/common bond and the behavioral consequences, reliability, validity, R2, and Q2 measures for the criterion variables. Again, all Q2 values are above 0, indicating that common identity and common bond have predictive relevance for the criterion constructs (Hair et al., 2012). While in-group loyalty has a very high R2 value, all other criterion variables achieve considerably lower values. However, this is unsurprising, considering the numerous potential direct determinants of the criterion variables. Most importantly, the results clearly show that common identity is the primary driver for the consequences of attachment in online

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

2263

Table 2 Model estimates for relationships between antecedents and attachment types. Hypothesis tested

Relationship tested

Model 1: on-topic

Model 2: off-topic

Model 3: GR

Model 4: DR Model 5: robustness

Model 6: loyalty

H1a H1b H1c H1d H2a H2b H2c H2d

Network effects → common identity In-group interdependence → common identity Intergroup comparison → common identity Social categorization → common identity Personal information → common bond Interpersonal similarity → common bond Collectivism → common bond Social interaction → common bond R2 common bond R2 common identity Q2 common bond Q2 common identity N

0.23⁎⁎⁎ 0.07 0.34⁎⁎⁎ 0.32⁎⁎⁎ 0.07 0.22⁎⁎ 0.18⁎⁎⁎ 0.35⁎⁎⁎

0.24⁎⁎⁎ 0.06 0.33⁎⁎⁎ 0.31⁎⁎⁎ 0.07 0.23⁎⁎ 0.17⁎⁎ 0.35⁎⁎⁎

0.23⁎⁎⁎ 0.08 0.33⁎⁎⁎ 0.33⁎⁎⁎ 0.06 0.22⁎⁎ 0.21⁎⁎⁎ 0.34⁎⁎⁎

0.23⁎⁎⁎ 0.06 0.34⁎⁎⁎ 0.33⁎⁎⁎ 0.07 0.22⁎⁎ 0.18⁎⁎⁎ 0.36⁎⁎⁎

0.26⁎⁎⁎ 0.04 0.34⁎⁎⁎ 0.32⁎⁎⁎ 0.06 0.22⁎⁎ 0.19⁎⁎ 0.35⁎⁎⁎

0.22⁎⁎⁎ 0.07 0.33⁎⁎⁎ 0.34⁎⁎⁎ 0.07 0.23⁎⁎ 0.17⁎⁎⁎ 0.35⁎⁎⁎

0.39 0.54 0.21 0.32 146

0.38 0.53 0.20 0.32 140

0.39 0.55 0.22 0.32 141

0.39 0.54 0.21 0.31 144

0.400 0.54 0.16 0.25 140

0.39 0.55 0.21 0.33 144

GR: generalized reciprocity, DR: direct reciprocity. ⁎⁎⁎ p ≤ 0.01. ⁎⁎ p ≤ 0.05. ⁎ p ≤ 0.10.

communities, exerting a positive and significant effect on the criterion variables across almost all models. For example, common identity exerts a positive and significant (p ≤ 0.01) influence on in-group loyalty (0.68), on-topic discussion (0.37), membership robustness (0.30), and generalized reciprocity (0.28). Only off-topic discussion is not positively affected by common identity. In contrast, common bond has a fairly limited influence on attachment consequences. Specifically, only direct reciprocity is significantly (p ≤ 0.05) predicted by common bond. In summary, the results provide support for hypotheses H3, H5, H6, and H7. To test whether common identity attachment exerts a significantly stronger influence on in-group loyalty than common bond attachment (H8), the analysis draws on a permutation-based multigroup comparison approach for paired samples (Sarstedt & Wilczynski, 2009). The analysis shows that common identity's effect on in-group loyalty is significantly stronger than that of common bond (t143 = 4.09; p ≤ 0.01), thus providing support for H8. Following Hair et al. (2012), the final step was to evaluate the total effects of the attachment types' antecedents on the criterion variable(s). The analyses reveal that for most types of criterion variables (except offtopic discussion), intergroup comparison and social categorization exert the greatest influence on user behaviors, with total effects ranging between 0.03 and 0.22 (intergroup comparison), as well as 0.08 and 0.23 (social categorization).

4.3. Additional analyses 4.3.1. Model extensions As the current study supplements Ren et al.'s (2007) original model by introducing two additional antecedents of attachment (network effects and collectivism), further analyses extended the original model setup by relating these two antecedents to both common identity and common bond. The analyses show that across all six models, collectivism exerts a positive and significant (p ≤ 0.01) influence on common identity, with path coefficients ranging between 0.15 (Model 1: on-topic) and 0.17 (Model 3: generalized reciprocity, and Model 5: robustness). Contrary to this, the network effects construct does not significantly influence common bond. The additional analyses also involve estimating full models in which all antecedents relate to both common identity and common bond. Table 4 presents the results for the criterion variable on-topic discussion. The results show that the central tenets of Ren et al.'s (2007) model remain unaffected by an extension of the model. Since PLS is a regression-based approach to SEM, the increase in the number of predictor constructs deflates the path coefficients. However, with merely one exception (interpersonal similarity → common bond), all path relationships in the original model remain significant when the model is extended, thus providing support for the robustness of the results. Similarly, only one path (personal information → common bond) that

Table 3 Model estimates for relationships between attachment types and behavioral consequences. Relationship tested

Model 1: on-topic (H3) Model 2: off-topic (H4) Model 3: GR (H5) Model 4: DR (H6) Model 5: robustness (H7) Model 6: loyalty (H8)

Common bond → criterion variable Common identity → criterion variable Cronbach's α criterion variable ρc criterion variable AVE criterion variable R2 criterion variable Q2 criterion variable N

0.03 0.37⁎⁎⁎

0.12 0.11

0.12 0.28⁎⁎⁎

0.27⁎⁎ 0.23⁎

0.15 0.30⁎⁎⁎

0.14 0.68⁎⁎⁎

0.89 0.93 0.82 0.15 0.11 146

0.70 0.80 0.58 0.04 0.01 140

0.80 0.88 0.72 0.13 0.09 141

0.81 0.88 0.64 0.20 0.11 144

0.70 0.77 0.57 0.17 0.04 140

0.89 0.91 0.64 0.60 0.36 144

GR: generalized reciprocity, DR: direct reciprocity. ⁎⁎⁎ p ≤ 0.01. ⁎⁎ p ≤ 0.0. ⁎ p ≤ 0.10.

2264

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

Table 4 Model estimates for full model (Criterion variable: on-topic discussion). Relationship tested Network effects → common bond Network effects → common identity In-group interdependence → common bond In-group interdependence → common identity Intergroup comparison → common bond Intergroup comparison → common identity Social categorization → common bond Social categorization → common identity Personal information → common bond Personal information → common identity Interpersonal similarity → common bond Interpersonal similarity → common identity Collectivism → common bond Collectivism → common identity Social interaction → common bond Social interaction → common identity Common bond → on-topic discussion Common identity → on-topic discussion R2 common bond R2 common identity R2 on-topic discussion

Estimates: full model

Original estimates

0.14 0.15⁎⁎⁎ 0.03 0.06 0.17 0.36⁎⁎⁎

– 0.23⁎⁎⁎ – 0.07 – 0.34⁎⁎⁎

0.00 0.30⁎⁎⁎ 0.20⁎⁎

– 0.32⁎⁎⁎

−0.04 −0.02 0.22⁎⁎⁎ 0.07 −0.03 0.28⁎⁎⁎ 0.11⁎ 0.02 0.27⁎⁎⁎ 0.41 0.57 0.15

0.07 – 0.22⁎⁎ – 0.18⁎⁎⁎ – 0.35⁎⁎⁎ – 0.03 0.37⁎⁎⁎ 0.39 0.54 0.15

⁎⁎⁎ p ≤ 0.01. ⁎⁎ p ≤ 0.05. ⁎ p ≤ 0.10.

was nonsignificant in the original model becomes significant in the more complex model. Most importantly, no sign changes occur that might lead to a trade-off situation when making community design decisions. Overall, the results provide a strong empirical support for Ren et al.'s (2007) model.

4.3.2. Heterogeneity check Although the analysis on the aggregate data level can provide insight into the complex relationships between the two attachment types and their various antecedents and behavioral consequences, this type of analysis still makes the simplifying assumption that all observations come from a single, homogeneous population. However, individuals may be heterogeneous in their perceptions and evaluations of unobserved constructs, which might account for distinct group-specific structural model path coefficients (Sarstedt, 2008). This especially holds for this study, since the data can be viewed as a hierarchical system of communities and community types, defined at separate levels of this hierarchical system. Specifically, data structures might be heterogeneous owing to similar response patterns among respondents from one community, comparable community types, or any other (combination of) respondent characteristics. To ensure that the aggregate data results are not biased by (unobserved) heterogeneity, the finite mixture PLS (FIMIX-PLS) procedure (Hahn, Herrmann, Huber, & Johnson, 2002) was applied to the data. FIMIX-PLS is the most commonly used approach for carrying out latent class analyses in PLS-SEM (e.g., Ringle, Sarstedt, Schlittgen, & Taylor, 2013). The approach combines the PLS method's strengths with the advantages of maximum likelihood estimation when deriving groups of observations with the help of finite mixture models. FIMIX-PLS assumes that the data derive from a source with several subpopulations; each subpopulation is modeled separately, and the overall population is a mixture of these subpopulations. The procedure calculates the probabilities of segment membership in respect of each observation so that the data fits into a predetermined number of segments. To decide on the number of segments, the analysis draws on AIC3 and CAIC, which have been shown to work well in a FIMIX-PLS context (Sarstedt, Becker, Ringle, & Schwaiger, 2011). While both criteria indicate a two-segment solution, the corresponding entropy

value is only 0.32. This result indicates that the probabilities of segment membership for a two-segment oscillate around 0.50 (as opposed to a situation where these probabilities are close to 0.90 for one segment and 0.10 for the other), suggesting fuzzy segment membership (Ringle, Sarstedt, & Mooi, 2010). Thus, the global model represents all observations well (e.g., Hair et al., 2011; Ringle et al., 2010). 4.3.3. Robustness check To provide further support for the findings' robustness and to mitigate the potential biasing effect of the previously-used data collection approach in terms of self-selection bias, new data from students was gathered during a lecture at a German university. All but four students who attended the lecture participated in the survey, yielding a dataset of 46 responses (92% response rate). Furthermore, and different from the original study, participants could not choose the community to evaluate but had to assess one specific community: Facebook. The results confirm common identity's primary role in explaining the consequences of attachment in online communities. Compared to common bond, common identity attachment has a stronger influence on on-topic discussion (common identity: 0.27; common bond: 0.06), generalized reciprocity (common identity: 0.48; common bond: 0.27), and loyalty (common identity: 0.85; common bond: 0.03). As was the case in the initial study, neither common identity, nor common bond attachment significantly influence off-topic discussion (common identity: 0.18; common bond: 0.17). The only difference emerges for common bond's role in explaining direct reciprocity. While the results parallel those from the initial study regarding the stronger influence of common bond on direct reciprocity, this effect is much more pronounced in the follow-up study (common identity: 0.23; common bond: 0.50). Note that the robustness construct had to be excluded, since the analysis did not yield a reliable construct measure. These results highlight the role of common identity vis-à-vis common bond attachment and provide ample support for the previous findings' robustness. 5. Discussion 5.1. Contributions of the study Based on Prentice et al.'s (1994) work, many researchers find that the attachment type an individual has towards a community, whether bond-based or identity-based, is one of the most important factors that influences his or her membership in a community (e.g., Dholakia, Bagozzi, & Pearo, 2004; Postmes et al., 2002; Ren et al., 2007; Sassenberg, 2002; Utz, 2003). However, very few studies empirically consider antecedents that influence these two attachment types in an online community and how these attachment styles in turn influence different user behaviors. Addressing this gap in research, this study sets out to assess the relationships between a number of theoretically validated antecedents (network effects, in-group interdependence, intergroup comparison, social categorization, personal information, interpersonal similarity, collectivism, and social interaction) as well as common identity and common bond attachment in online communities and how the two different attachment styles influence a wide variety of user behaviors in online communities (on-topic and off-topic discussion, generalized and direct reciprocity, membership robustness, and loyalty to an online community). Most importantly, the study shows that common identity attachment is the primary driver for user behavior in online communities. In contrast, common bond attachment has a fairly limited influence. These findings parallel the findings from Ren et al.'s (2012) study, in which the authors show that community features intended to foster identity-based attachment had stronger effects

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

on participation and retention than features intended to foster bond-based attachment. In terms of attachment drivers, network effects, intergroup comparison, and social categorization have positive and significant effects on common identity attachment, while this is not the case with in-group interdependence. Conversely, interpersonal similarity, collectivism, and social interaction primarily drive common bond attachment, while personal information has no effect. Additional analyses of the antecedents' total effects reveal the increased importance of intergroup comparison and social categorization on user behaviors. The results for in-group interdependence and personal information contrast with the predictions found in the literature, potentially because of the sample composition, which draws on users from voluntary online communities. Voluntary online communities seldom demand real in-group interdependence (i.e., users of voluntary online communities rarely work on team products dependent on each other's contributions). Even most open-source software communities only depend on one project manager and face a fairly inactive community (Krishnamurthy, 2002). However, a potential way to analyze the importance of in-group interdependence for the evolution of common identity attachment in voluntary online communities would be to consider highly rated virtual world guild members, for instance, in World of Warcraft. Members of a highly rated guild heavily depend on each other, since they must cooperate in order to advance in the game. In sum, this study makes a number of contributions. First, the study tests a comprehensive cause-and-effect model that has not been empirically evaluated before in this breadth and coverage. Second, the study contributes to the discussion in the literature on online community management by testing whether established findings from social sciences apply in the context of online communities. Third, the result regarding the influence of common identity and common bond attachments on user behavior answers the call for research by Ren et al. (2007, 2012). The model also adds two additional constructs (network effects and collectivism) to the original model of Ren et al. (2007), which have significant positive relationships with common identity and common bond attachments. 5.2. Implications for online community design In light of the many online communities, the relevance of design principles increases. From the results, the following recommendations for community designers emerge. Online communities that value on-topic information exchange, generalized reciprocity norms, robustness to member turnover, and in-group loyalty should focus on establishing common identity attachment to the community. Community designers can foster common identity by creating a large community with many connection possibilities, a positive intergroup comparison, and superior social categorization compared to other communities, for example, by emphasizing the community's uniqueness compared to other communities or by only allowing a specific group to participate in the community that fulfills a signal others do not. In fact, many communities use hidden clues that outsiders are unlikely to recognize, unlike attentive aspiring members. On the other hand, online communities that value off-topic discussion and the building of specific reciprocity should emphasize common bond attachment, for example, by fostering social interaction, interpersonal similarity and collectivism. Hence, members should have the opportunity to learn more about similar interests and social interaction. In this regard, a large choice of possible functions is available in online communities, for example, the use of liking and knowledge tests, etc. In this way, members learn about similar interests without revealing too much personal information. The strong influence of collectivism on common bond attachment could have an interesting implication for online communities'

2265

strategies. Some online communities in the U.S. only became successful when an increasing number of users from collectivist countries began to participate. Orkut, for example, was long unknown in the U.S. until the community made a breakthrough in Brazil (Boyd & Ellison, 2007). Finally, the increased importance of common identity attachment for predicting criterion variables indicates that online community moderators should strengthen member identification with the community. 5.3. Limitations of the study This study contains some limitations that suggest grounds for further research. Whenever researchers investigate questions for which no sufficient archived data are available—as is the case in this study—they must resort to subjective measures (Kumar, Stern, & Anderson, 1993). In this context, some operationalizations might not fully capture the construct's theoretical domain, but partially focus on one potential outcome of the theoretical construct (e.g., the measures of intergroup comparison cover aspects such as perceived superiority of a group as well as interpersonal similarity with perceived friendship). Similarly, the measurements used for social categorization and common identity overlap from a conceptual perspective. Since this study draws on a field survey rather than an experimental setup, which would allow for controlling the degree of social categorization, this is a study limitation. Lastly, in terms of data collection, respondents who identify with their online community might be more willing to respond than others. This self-selection could bias the analysis by inflating the effect of common identity attachment. While the robustness checks provide support for the findings, this type of self-selection among respondents cannot fully be ruled out. 5.4. Avenues for future research An especially fruitful avenue for further research on the design of online communities relates to the interplay between common bond and common identity attachments. This aspect is even more relevant since many communities' goals are blurry, as evidenced in the difficulty to categorize communities as common identity or common bond groups. In fact, many community designers routinely implement multiple features that allow for both attachment types. Further research should examine how other concepts such as utilitarian motives of community members or organizational identity of a community guide behavior in a social network. Following Brickson (2005, 2007, 2013) communities could develop different types of organizational identities (i.e., individualistic, relational, and collectivistic). Specifically, an individualistic organizational identity generally forges relationships based on instrumentality and maintains relationships to the extent that they enhance the organization's own aims. A relational organizational identity, on the other hand, will forge stakeholder relationships based on dyadic concern and trust. A collectivistic organizational identity will forge external and internal stakeholder relationships based on a common purpose. Re-examining the results regarding the influence of in-group interdependence on common identity attachment, for example, by specifically analyzing highly rated virtual world guild members (e.g., in World of Warcraft) would also be of interest. Another area that merits further investigation is the examination of the relationship between personal information and liking, trust, and respect since opportunities for self-disclosure do not seem to be related to common bond attachment. Likewise, given that common identity attachment seems more important regarding a wide range of user behavior, future studies should consider more antecedents of common identity attachment.

2266

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

Appendix A

Table A1 Construct measures. Construct

Items

Antecedents of common identity Network effects I joined this community because of its large membership Membership increases lead directly to more benefits for me I have the feeling that I communicate with more and more people because of an increasing membership In-group interdependence All members need to contribute to achieve the community's goals This community accomplishes things that no single member could achieve In this community, members need to cooperate to complete community tasks Intergroup comparison Compared to other similar communities, this community is much more tolerant Compared to other similar communities, this community is much more flexible Compared to other similar communities, this community is much more creative Compared to other similar communities, this community is much more intelligent I think this community has much to be proud of People in other similar communities generally admire this community Social categorization I often think about the fact that I am a member of this community In general, being a member of this community is an important part of my self-image The fact that I'm a member of this community often enters my mind I am a typical member of this community Most people in this community have similar values Most people in this community have similar preferences Most people in this community behave similarly Antecedents of common bond Personal information

Interpersonal similarity

Collectivism

Social interaction

The people I interact with in this community are personally identifiable to me I believe that other community members can identify me This community offers me the opportunity to get more information about other members I think I could become friends with some members of this community Members of this community would fit into my circle of friends I sometimes feel I know some members of this community personally When I meet someone from my nation or group, I know we will have common goals and aspirations If I were to lose touch with my group, I would be a different person I generally accept the decisions made by my group When I meet someone from my own nationality or religion, I know we will have common goals and interests In general, I actively interact with other members of this community In this community, I share information about a particular subject with other members In this community, I share my skills and abilities with other members I can always count on getting many responses to my posts/messages I can always count on getting responses to my posts/messages fairly quickly

Common identity and common bond Common identity Belonging to this community is very important to me It would feel great to be described as a typical member of this community I often mention this community when I first meet someone I feel a strong attachment to this community I often acknowledge the fact that I am a member of this community I am a typical member of this community Common bond I feel very close to the other members of this community Many members of this community have influenced my thoughts and behaviors Many of my friends are from this community Criterion constructs On-topic discussion

Off-topic discussion Generalized reciprocity

Direct reciprocity

Robustness

In-group loyalty

I come to the community website to get information on a particular topic I use the community website when I want advice on how to carry out certain tasks I come to the community website when I need facts about a particular subject The posts/messages on the community website often contain private topics Members of this community seem very willing to disclose private information about themselves To help somebody is the best policy, to be certain that s/he will help you in the future In this community, most of the members comply with the existing rules and norms In this community, I help other members, so I will be helped in the future I am ready to incur personal costs to help someone who has helped me before If someone helps me, I am pleased to help him/her When someone does me a favor, I feel committed to repay him/her Members of this community generally like to answer posts/messages If my friends were to leave this community, I would also leave (reversed item) I would remain active in this community even if my friends in the community were to become inactive This community would still be attractive to me if my friends were to leave it I talk up this community to my friends as a great community to be a part of I feel very loyal to this community I am proud to tell others that I am part of this community I really care about the fate of this community I am willing to put in much effort to help this community be successful To me, this is the best of all possible communities to be a part of

Note: All items use 7-point Likert-type scales (1 = strongly disagree, 7 = strongly agree).

References Sahay and Riley (2003)

Henry, Arrow, and Carini (1999) Utz (2003) Ellemers, Kortekaas, and Ouwerkerk (1999) Yuki (2003)

Cameron (2004) Utz (2003) Yuki (2003)

Postmes, Lea, and Spears (2002) Pinsonneault and Heppel (1998) McCroskey and McCain (1974)

Oyserman (1993)

Ridings et al. (2002)

Prentice et al. (1994)

Prentice et al. (1994)

Ridings et al. (2002)

Ridings et al. (2002) Gallucci and Perugini (2003)

Gallucci and Perugini (2003)

Ren et al. (2007)

Mowday, Steers, and Porter (1979)

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268 0.49⁎⁎⁎ 0.37⁎⁎⁎ 0.52⁎⁎⁎

15

0.51⁎⁎⁎ 0.30⁎⁎⁎ 0.40⁎⁎⁎

14

−0.06 0.05 −0.16⁎⁎ −0.2

13

−0.07 0.44⁎⁎⁎ 0.47⁎⁎⁎ 0.50⁎⁎⁎ 0.55⁎⁎⁎

12

0.25⁎⁎⁎ 0.12 0.28⁎⁎⁎ 0.33⁎⁎⁎ 0.13 0.50⁎⁎⁎

11

0.49⁎⁎⁎ 0.49⁎⁎⁎ 0.41⁎⁎⁎

0.00 0.24⁎⁎

0.09 0.15⁎ 0.20⁎⁎

0.13 0.35⁎⁎⁎ 0.49⁎⁎⁎ 0.38⁎⁎⁎ 0.69⁎⁎⁎ −0.03 0.40⁎⁎⁎ 0.32⁎⁎⁎ 0.28⁎⁎⁎ 0.57⁎⁎⁎

0.10 0.40⁎⁎⁎ 0.39⁎⁎⁎ 0.38⁎⁎⁎ 0.70⁎⁎

0.07 0.25⁎⁎⁎ 0.18⁎⁎ 0.28⁎⁎⁎

0.27⁎⁎⁎ 0.59⁎⁎⁎ 0.43⁎⁎⁎ 0.44⁎⁎⁎ 0.17⁎⁎ 0.180⁎⁎⁎ 0.10 0.20⁎⁎ 0.20⁎⁎

0.09 0.24⁎⁎⁎ 0.10 0.33⁎⁎⁎ 0.10 0.28⁎⁎⁎ 0.21⁎⁎⁎ 0.22⁎⁎⁎

0.71⁎⁎⁎ 0.34⁎⁎⁎ 0.58⁎⁎⁎ 0.39⁎⁎⁎ 0.60⁎⁎⁎ 0.63⁎⁎⁎ 0.45⁎⁎⁎ 0.38⁎⁎⁎ 0.55⁎⁎⁎ 0.58⁎⁎⁎ 0.19⁎⁎ 0.33⁎⁎⁎ 0.32⁎⁎⁎ 0.46⁎⁎⁎ 0.49⁎⁎⁎ 0.44⁎⁎⁎ 0.47⁎⁎⁎

0.37⁎⁎⁎ 0.57⁎⁎⁎ 0.50⁎⁎⁎ 0.54⁎⁎⁎ 0.63⁎⁎⁎ 0.46⁎⁎⁎ 0.45⁎⁎⁎

0.48⁎⁎⁎ 0.13 0.38⁎⁎⁎ 0.28⁎⁎⁎ 0.27⁎⁎⁎

0.07 0.19⁎⁎ 0.41⁎⁎⁎ 0.13 0.43⁎⁎

0.27⁎⁎⁎ 0.49⁎⁎⁎ 0.39⁎⁎⁎ 0.33⁎⁎⁎ 0.15⁎ 0.34⁎⁎⁎ 0.32⁎⁎⁎ 0.15⁎ 0.47⁎⁎

−0.04 0.30⁎⁎⁎ 0.54⁎⁎⁎ 0.31⁎⁎ 0.55⁎⁎

0.55⁎⁎⁎ 0.37⁎⁎⁎ 0.13 0.33⁎⁎⁎ 0.39⁎⁎⁎ 0.29⁎⁎⁎ 0.75⁎⁎

10 9 8 7 6

⁎⁎⁎ p ≤ 0.01. ⁎⁎ p ≤ 0.05. ⁎ p ≤ 0.10.

5 4 3 1

2

1.60 1.80 1.38 1.38 1.68 1.41 1.41 1.36 1.41 1.24 1.72 1.37 1.44 1.07 1.33 1.36

SD Mean

References

3.90 4.23 4.88 3.96 4.69 4.88 3.87 4.86 4.10 3.92 4.63 4.05 4.73 5.37 4.82 4.40 1. Network effects 2. In-group interdependence 3. Intergroup comparison 4. Social categorization 5. Personal information 6. Interpersonal similarity 7. Collectivism 8. Social interaction 9. Common identity 10. Common bond 11. On-topic discussion 12. Off-topic discussion 13. Generalized reciprocity 14. Direct reciprocity 15. Robustness 16. In-group loyalty

Table A2 Descriptive statistics and construct correlations.

2267

Algesheimer, René, Dholakia, Utpal M., & Herrmann, Andreas (2005). The social influence of brand community: Evidence from European car clubs. Journal of Marketing, 69(3), 19–34. Arguello, Jaime, Butler, Brian, Joyce, Elisabeth, Kraut, Robert, Ling, Kimberly S., & Wang, Xiaoqing (2006). Talk to me: Foundations for successful individual-group interactions in online communities. Proceedings of the ACM conference on human-factors in computing systems (pp. 959–968). New York, NY: ACM Press. Armstrong, Athur, & Hagel, John (1996). The real value of on-line communities. Harvard Business Review, 74(3), 134–141. Armstrong, J. Scott, & Overton, Terry S. (1977). Estimating nonresponse bias in mail surveys. Journal of Marketing Research, 14(3), 396–402. Back, Kurt W. (1951). Influence through social communication. Journal of Abnormal and Social Psychology, 46(1), 9–23. Bagozzi, Richard P., & Dholakia, Utpal M. (2006). Antecedents and purchase consequences of customer participation in small group brand communities. International Journal of Research in Marketing, 23(1), 45–61. Boyd, Danah M., & Ellison, Nicole B. (2007). Social network sites: Definition, history and scholarship. Journal of Computer-Mediated Communication, 13(1), 210–230. Brickson, Shelley L. (2005). Organizational identity orientation: Forging a link between organizational identity and organizations' relations with stakeholders. Administrative Science Quarterly, 50(4), 576–609. Brickson, Shelley L. (2007). Organizational identity orientation: The genesis of the role of the firm and distinct forms of social value. Academy of Management Review, 32(3), 864–888. Brickson, Shelley L. (2013). Athletes, best friends, and social activists: Accounting for the role of identity in organizational identification. Organization Science, 24(1), 226–245. Buchan, Nancy R., Croson, Rachel T. A., & Johnson, Eric J. (2006). Let´s get personal: An international examination of the influence of communication, culture and social distance on other regarding preferences. Journal of Economic Behaviour & Organization, 60(3), 373–398. Cameron, James E. (2004). A three-factor model of social identity. Self and Identity, 3(3), 239–262. Casaló, Luis V., Flavián, Carlos, & Guinalíu, Miguel (2013). New members' integration: Key factors of success in online travel communities. Journal of Business Research, 66(6), 706–710. Cheng, Xi-Ping (1996). The group-based binding pledge as a solution to public goods problems. Organizational Behavior and Human Decision Processes, 66(2), 192–202. Collins, Nancy L., & Miller, Laurie C. (1994). Self-disclosure and liking: A meta-analytical review. Psychology Bulletin, 116(3), 457–475. Dholakia, Utpal M., Bagozzi, Richard P., & Pearo, Lisa K. (2004). A social influence model of consumer participation in network- and small-group-based virtual communities. International Journal of Research in Marketing, 21(3), 241–263. Diamantopoulos, Adamantios, Riefler, Petra, & Roth, Katharina P. (2008). Advancing formative measurement models. Journal of Business Research, 61(12), 1203–1218. Ellemers, Naomi, Kortekaas, Paulien, & Ouwerkerk, Jaap W. (1999). Self-categorisation, commitment to the group and group self-esteem as related but distinct aspects of social identity. European Journal of Social Psychology, 29(2–3), 371–389. Flippen, Annette R., Hornstein, Harvey A., Siegal, William W., & Weitzmann, Eben A. (1996). A comparison of similarity and interdependence as triggers for in-group formation. Personality and Social Psychology Bulletin, 22(9), 882–893. Fornell, Claes, & Larcker, David F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. Frank, Robert (1988). Passions within reason. The strategic role of the emotions. New York, NY: WW Norton & Co. Gallucci, Marcello, & Perugini, Marco (2003). Information seeking and reciprocity: A transformational analysis. European Journal of Social Psychology, 33(4), 473–495. Gouldner, Alvin W. (1960). The norm of reciprocity: A preliminary statement. American Sociological Review, 25(2), 161–178. Hahn, Carsten H., Herrmann, Andreas, Huber, Frank, & Johnson, Michael D. (2002). Capturing customer heterogeneity using a finite mixture PLS approach. Schmalenbach Business Review, 54(3), 243–269. Hair, Joseph F., Hult, G. Tomas M., Ringle, Christian M., & Sarstedt, Marko (2013). A primer on partial least squares structural equation modeling (PLS-SEM). Thousand Oaks, CA: Sage. Hair, Joseph F., Ringle, Christian M., & Sarstedt, Marko (2011). PLS-SEM — Indeed a silver bullet. Journal of Marketing Theory & Practice, 19(2), 139–151. Hair, Joseph F., Sarstedt, Marko, Ringle, Christian M., & Mena, Jeannette A. (2012). An assessment of the use of partial least squares structural equation modeling in marketing research. Journal of the Academy of Marketing Science, 40(3), 414–433. Henry, Kelly B., Arrow, Holly, & Carini, Barbara (1999). A tripartite model of group identification. Small Group Research, 30(5), 558–581. Henseler, Jörg, Dijkstra, Theo K., Sarstedt, Marco, Ringle, Christian M., Diamantopoulos, Adamantios, Straub, Detmar W., et al. (2014). Common belies and reality about PLS: Comments on Rönkkö and Evermann (2013). Organizational Research Methods, 17(2), 182–209. Jarvenpaa, Sirkka, & Leidner, Dorothy E. (1999). Communication and trust in global virtual teams. Organization Science, 10(6), 791–815. Keltner, Dacher, & Bonanno, George (1997). A study of laughter and dissociation: The distinct correlates of laughter and smiling during bereavement. Journal of Personality and Social Psychology, 73(4), 687–702. Krishnamurthy, Sandeep (2002). Cave or community?: An empirical examination of 100 mature open source projects. First Monday, 7(6) (http://ojphi.org/htbin/cgiwrap/bin/ ojs/index.php/fm/article/view/960/881). Kumar, Nirmalya, Stern, Louis W., & Anderson, James C. (1993). Conducting interorganizational research using key informants. Academy of Management Journal, 36(6), 1633–1651.

2268

M. Fiedler, M. Sarstedt / Journal of Business Research 67 (2014) 2258–2268

Lee, Eocman, Lee, Jeho, & Lee, Jongseok (2006). Reconsideration of the winnertake-all hypothesis: Complex networks and local bias. Management Science, 52(12), 1838–1848. Lee, Hyun J., Lee, Doo-Hee, Taylor, Charles R., & Lee, Jung-Ho (2011). Do online brand communities help build and maintain relationships with consumers? A network theory approach. Journal of Brand Management, 19(3), 213–227. McAfee, Andrew P. (2006). Enterprise 2.0: The dawn of emergent collaboration. MIT Sloan Management Review, 47(3), 21–28. McCroskey, James C., & McCain, Toni A. (1974). The measurement of interpersonal attraction. Speech Monographs, 41(3), 261–266. Mollica, Kelly A., Gray, Barbara, & Trevino, Linda K. (2003). Racial homophily and its persistence in newcomers' social networks. Organization Science, 14(2), 123–136. Mooi, Erik A., & Sarstedt, M. (2011). A concise guide to market research. The process, data, and methods using IBM SPSS Statistics. Berlin: Springer. Mowday, Richard, Steers, Richard, & Porter, Lyman W. (1979). The measurement of organizational commitment. Journal of Vocational Behavior, 14(2), 224–247. Mummalaneni, Venkatapparao (2005). An empirical investigation of Web site characteristics, consumer emotional states and on-line shopping behaviors. Journal of Business Research, 58(4), 526–532. Nambisan, Priya, & Watt, James H. (2011). Managing customer experiences in online product communities. Journal of Business Research, 64(8), 889–895. O'Murchu, Ina, Breslin, John G., & Decker, Stefan (2004). Online social and business networking communities. Digital Enterprise Research Institute Technical Report (http:// www.deri.org). Oyserman, D. (1993). The lens of personhood: Viewing the self and others in a multicultural society. Journal of Personality and Social Psychology, 65(5), 993–1009. Pinsonneault, Alain, & Heppel, Nelson (1998). Anonymity in group support systems research: A new conceptualization, measure and contingency framework. Journal of Management Information Systems, 14(3), 89–108. Postmes, Tom, Lea, Martin, & Spears, Russell (2002). Intergroup differentiation in computer-mediated communication: Effects of depersonalization. Group Dynamics: Theory Research and Practice, 6(1), 2–16. Preece, Jenny, & Maloney-Krichmar, Diane (2003). Online communities. In Julie A. Jacko, & Andrew Sears (Eds.), Handbook of human–computer interaction (pp. 596–620). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Publishers. Prentice, Deborah A., Miller, Dale T., & Lightdale, Jennifer R. (1994). Asymmetries in attachment of groups and to their members: Distinguishing between common-identity and common-bond groups. Personality and Social Psychology Bulletin, 20(5), 484–493. Reinartz, Werner, Haenlein, Michael, & Henseler, Jörg (2009). An empirical comparison of the efficacy of covariance-based and variance-based SEM. International Journal of Research in Marketing, 26(4), 332–344. Ren, Yuqing, Harper, Maxwell, Drenner, Sara, Kiesler, Sara, Terveen, Loren, Riedl, John, et al. (2012). Building member attachment in online communities: Applying theories of group identity and interpersonal bonds. MIS Quarterly, 36(3), 841–864. Ren, Yuqing, Kiesler, Sara, & Kraut, Robert E. (2007). Applying common identity and bond theory to design of online communities. Organization Studies, 28(3), 377–408. Ren, Yuqing, & Kraut, Robert E. (2012). A simulation for designing online community: Member motivation, contribution, and discussion moderation. Working Paper (http://kraut.hciresearch.org/sites/kraut.hciresearch.org/files/articles/ren07SimulatingOnlineCommunit-v4.6-rek.pdf). Ren, Yuqing, Kraut, Robert E., Kiesler, Sara, & Resnick, Paul (2011). Encouraging commitment in online communities. In Robert E. Kraut, & Paul Resnick (Eds.), Building successful online communities. Evidence-based social design (pp. 77–123). Cambridge, MA: MIT Press. Ridings, Catherine M., Arinze, Bay, & Gefen, David (2002). Some antecedents and effects of trust in virtual communities. Journal of Strategic Information Systems, 11(4), 271–295.

Ringle, Christian M., Sarstedt, Marko, & Mooi, Erik A. (2010). Response-based segmentation using FIMIX-PLS. Theoretical foundations and an application to ACSI data. Annals of Information Systems, 8, 19–49. Ringle, Christian M., Sarstedt, M., Schlittgen, R., & Taylor, Charles R. (2013). PLS path modeling and evolutionary segmentation. Journal of Business Research, 66(9), 1318–1324. Ringle, Christian M., Wende, Sven, & Will, Alexander (2005). SmartPLS 2.0 (Beta). (Hamburg, http://www.smartpls.de). Rogers, Paul, & Lea, Martin (2005). Social presence in distributed group environments: The role of social identity. Behaviour & Information Technology, 24(2), 151–158. Sahay, Alok, & Riley, Debra (2003). The role of resource access, market considerations, and the nature of innovation in pursuit of standards in new product development process. Journal of Product Innovation Management, 20(5), 338–355. Sarstedt, Marko (2008). A review of recent approaches for capturing heterogeneity in partial least squares path modelling. Journal of Modelling in Management, 3(2), 140–161. Sarstedt, Marko, Becker, Jan-Michael, Ringle, Christian M., & Schwaiger, Manfred (2011). Uncovering and treating unobserved heterogeneity with FIMIX-PLS: Which model selection criterion provides an appropriate number of segments? Schmalenbach Business Review, 63(1), 34–62. Sarstedt, Marko, & Wilczynski, Petra (2009). More for less? A comparison of single-item and multi-item measures. Business Administration Review, 69(2), 211–227. Sassenberg, Kai (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), 27–37. Schau, Hope Jensen, Muñiz, Albert M., & Arnould, Eric J. (2009). How brand community practices create value. Journal of Marketing, 73(3), 30–51. Scott, John (1991). Social network analysis. London: Sage. Sherif, Muzafer, & Sherif, Carolyn W. (1953). Groups in harmony and tension. Harper Bros: New York, NY. Sommers, Samuel R. (2006). On racial diversity and group decision making: Identifying multiple effects of racial composition on jury deliberations. Journal of Personality and Social Psychology, 90(4), 597–612. Tajfel, Henri, Billic, Michael G., Bundy, Richerd P., & Flament, Claude (1971). Social categorization and intergroup behaviour. European Journal of Social Psychology, 1(2), 149–178. Tajfel, Henri, & Turner, John (1986). The social identity theory of intergroup behavior. In Stephen Worchel, & William G. Austin (Eds.), The social psychology of intergroup relations (pp. 7–24). Chicago, IL: Nelson-Hall. Tsai, Wenpin (2002). Social structure of coopetition within a multiunit organization: Coordination, competition and intraorganizational knowledge sharing. Organization Science, 13(2), 179–190. Utz, Sonja (2003). Social identification and interpersonal attraction in MUDs. Swiss Journal of Psychology, 62(2), 91–101. Yuki, Mika (2003). Intergroup comparison versus intragroup relationships: A crosscultural examination of social identity theory in North American and East Asian cultural contexts. Social Psychology Quarterly, 66(2), 166–183. Zaglia, Melanie E. (2013). Brand communities embedded in social networks. Journal of Business Research, 66(2), 216–223. Zhang, Dongsong, Fu, Xiaolan, Lowry, Paul B., & Zhou, Lina (2007). The impact of individualism–collectivism, social presence, and group diversity on group decision making under majority influence. Journal of Management Information Systems, 23(4), 53–80. Zhou, Zimin, Zhang, Qiyuan, Su, Chenting, & Zhou, Nen (2012). How do brand communities generate brand relationships? Intermediate mechanisms. Journal of Business Research, 65(7), 890–895.