Proposing and Testing Computer-Mediated ...

This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder. This version of the referenced work is the post-print version of the article—it is NOT the final published version nor the corrected proofs. If you would like to receive the final published version please send a request to [email protected], and I will be happy to send you the latest version. Moreover, you can contact the publisher’s website and order the final version there, as well. The current reference for this work is as follows: Clay Posey, Paul Benjamin Lowry, Tom L. Roberts, and Selwyn Ellis (2010). “Proposing the online community self-disclosure model: The case of working professionals in France and the UK who use online communities,” European Journal of Information Systems, vol. 19(2), pp. 181–195 (doi:10.1057/ejis.2010.15). If you have any questions and/or would like copies of other articles I’ve published, please email me at [email protected], and I’d be happy to help. My vita can be found at http://www.cb.cityu.edu.hk/staff/pblowry Alternatively, I have an online system that you can use to request any of my published or forthcoming articles. To go to this system, click on the following link: https://seanacademic.qualtrics.com/SE/?SID=SV_7WCaP0V7FA0GWWx

Proposing the Online Community Self-Disclosure Model: The Case of Working Professionals in France and the UK Who Use Online Communities ABSTRACT The global use of online communities has exploded to involve hundreds of millions of users. Despite the tremendous social impact and business opportunities afforded by these communities, little information systems (IS) research has addressed them—especially in a crosscultural context. Our research proposes an online community self-disclosure model, tested in a cross-cultural setting using data provided by French and British working professionals. Our model is based on social exchange theory (SET) and social penetration theory (SPT), as well as on cross-cultural theory related to individualism-collectivism. SET explains that individuals engage in relationships when the perceived costs associated with the relationship are less than the expected benefits. SPT extends SET to explain that individuals participate in self-disclosure to foster relationships—reciprocation is the primary benefit of self-disclosure, whereas risk is the foundational cost of self-disclosure. Our study established several important findings: Positive social influence to use an online community increases online community self-disclosure; reciprocity increases selfdisclosure; online community trust increases self-disclosure; and privacy risk beliefs decrease self-disclosure. Meanwhile, a tendency toward collectivism increases self-disclosure. We further found that French participants had higher scores on horizontal individualism than British participants. Several other findings and their implications for practice are also discussed.

KEYWORDS Self-disclosure, social influence, trust, privacy, reciprocity, collectivism, individualism, culture

2

Proposing the Online Community Self-Disclosure Model: The Case of Working Professionals in France and the UK Who Use Online Communities

INTRODUCTION Online social networking and community sites have become an extremely popular social force. New registrations for Facebook have averaged 250,000 per day since 2007, totalling 250 million active users in 2009; the number of active users doubles every six months; more than half of the active users use the site daily; active users average 20 minutes per day (Anonymous, 2009b). In 2009, MySpace reported 130 million active users, with the virtual communities localized in and translated into 20 different countries and languages (Anonymous, 2009a). Online social networking does not simply represent an exploding social phenomenon; it is a trend that has many potential implications for business and information systems (IS). For example, there are more than one million Facebook developers and entrepreneurs who have developed more than 350,000 commercial applications for the 250 million active Facebook users (Anonymous, 2009b). These sites are increasingly being used as an ideal source of marketing research (Kozinets, 2002), as a way to enhance the impact of branding (Im et al., 2008) and to increase demand for products (Miller et al., 2009). The potential business benefits of self-disclosure in online communities are profound. Such disclosures are an ideal and unique source of marketing research (Kozinets, 2002). This opportunity to gather information is particularly salient to marketing research because in traditional market-research approaches, only a small percentage of people self-disclose desired information (Robertshaw & Marr, 2006). Furthermore, many market researchers now think of branding as invoking a relationship between the brand and a consumer that can be greatly 3

enhanced through online interactions—even though technical and non-interpersonal—through principles of interpersonal interaction that can be created by Web sites (Im et al., 2008). Meanwhile, recent strategic economics research shows that the participation of businesses in online communities can enhance demand for companies‘ products (Miller et al., 2009). Other innovative business uses of self-disclosure in online communities include energizing internal innovation among organizational employees, building brand ambassador programs and support forums, discovering the most enthusiastic customers and leveraging them, motivating customers, etc. (Bernoff & Li., 2008). A related recent development in which social networking sites and online communities can dramatically change the landscape of business is part of what is termed the ‗contribution revolution,‘ where firms create contribution sites for stakeholders interested in a particular business; they often act as advocates for the business (Cook, 2008). These sites are now being leveraged by leading firms to gain cost advantages and even strategic advantages over competitors through literally having volunteers contribute to the firms (Cook, 2008). Such sites have been used to acquire free raw materials, build customer service forums through online community member participation or develop volunteer-run marketing on social networking sites, capital resources, enhanced design and development and even production (Cook, 2008). What is shared between the pure social aspects of online communities and more businessoriented online communities is the desirability of open self-disclosure, which fosters social relationships and/or enhances business connections with people who share an affinity for a brand or company. Thus, an important contribution that IS research can make to online communities is to explain why online community users disclose or withhold information. In utilitarian systems, intentions to use and to continue to use a system are primarily based on the perceived usefulness 4

of the software and, to a lesser extent, perceived ease of use (Davis, 1989; Davis et al., 1989); however, research has shown that the reasons for using and self-disclosing in online communities are entirely different. The primary drivers of self-disclosure online are socially based factors. Specifically, people use social networking sites to form and foster relationships, disclosing and sharing information about themselves with others (Chang Lee & Kwon, 2008; Chiu et al., 2006). Self-disclosure research in online communities is just starting to emerge. One salient study applied social cognitive theory and social capital theory to predict factors that would encourage knowledge sharing in online communities (Chiu et al., 2006). Recently, a study explained how online communities can foster trust and self-disclosure in customers (Porter & Donthu, 2008). Another study found that more personal self-disclosure by online reviewers increased positive perceptions of the reviews and increased sales in online electronic markets (Forman et al., 2008). Though social networking has vast global and cultural implications, very little research has been conducted on cross-cultural online social networking. Most research has focused on homogeneous studies of either participants from the United States (US) (e.g., Ellison et al., 2007; Hargittai, 2007) or Asian participants (Ishii & Ogasahara, 2007). However, one study evaluated online relationship differences between Japanese and Korean online community members (Ishii & Ogasahara, 2007). Another study found that Argentineans self-disclose more than US participants, although the researchers did not consider electronic media (Horenstein & Downey, 2003). A more recent study looked at differences in social networking participation between the US and India (Marshall et al., 2008). To date, few researchers have studied European countries. Given these pressing opportunities, this research focuses on explaining and predicting the drivers by which individuals disclose personal and/or private information to others within an 5

online social network. We use the economics-based theory of social exchange theory (SET) (Thibaut & Kelley, 1959); its key extension, social penetration theory (SPT) (Altman & Taylor, 1973); and communication privacy management theory (Petronio, 2002) to ground our conceptual model and help to explain this self-disclosure process. We also consider individualand national-level cultural differences as drivers of online self-disclosure. Our research uses professional participants from the UK and France. We chose the UK and France because these two countries fit our criteria of cultural differences and high Internet usage. The Hofstede scores between the two countries were significantly different across power distance, individualism, uncertainty avoidance and masculinity (Hofstede, 1991). As explained by Hofstede (1991), power distance is the degree of inequality among individuals in the organization; individualism is the relationship between an individual and his or her collective group or, in this study‘s case, online social network; uncertainty avoidance is the acceptance or tolerance of uncertainty; finally, masculine social values indicate the importance of showing off, achieving something visible or making money predominate, while feminine social values focus upon quality of life and personal relationships. In addition, the UK (43.8 million users) and France (40.9 million users) are currently the exhibit the 2nd and 3rd highest levels of Internet usage in Europe (Anonymous, 2009c). The model that we developed is called the online community self-disclosure model. Our model has two major research questions: RQ1. What social and environmental factors best predict self-disclosure in online communities? RQ2. Do individual-level cultural differences affect self-disclosure in online communities, and do these individual-level differences extend to national-level 6

differences? Moreover, are there self-disclosure differences between the French and British national cultures? THEORETICAL MODEL In this section, we propose a new theoretical model, the Online Community SelfDisclosure Model, which predicts the key social factors that best predict why people self-disclose in online communities. Self-disclosure refers to what individuals voluntarily and intentionally reveal about themselves to others—including thoughts, feelings and experiences (Derlega et al., 1993; Pearce & Sharp, 1973). Self-disclosure is voluntary and purposeful (Pearce & Sharp, 1973). Moreover, self-disclosure is generally a positive experience that include benefits such as the formation of intimate associations (Altman & Taylor, 1973), reduced stress levels in the wake of traumatic experiences (Greenberg & Stone, 1992), social acceptance or approval for the individuals‘ ideas (Derlega et al., 1993) and reclaimed internal energy that was once devoted to holding the sensitive information within (Pennebaker, 1989). Self-disclosure has been conceptualized and measured along five separate dimensions: amount, depth, honesty, intent and valence (Wheeless, 1978; Wheeless & Grotz, 1976). Selfdisclosure amount represents the frequency and duration of an individual‘s disclosures. Depth reflects the degree of intimacy in the communication. Honesty refers to the accuracy with which one communicates information about oneself. Intent reflects an individual‘s control and awareness over his or her self-disclosures. Valence is the positive nature of the information being disclosed in communication. Self-disclosure plays an integral role in relationship development. Theorists have extended SET (Jarvenpaa & Staples, 2001; Thibaut & Kelley, 1959; Wasko & Faraj, 2005) as a foundation to explain the cognitive process that individuals engage in before self-disclosing 7

(Altman & Taylor, 1973), as well as to provide the requisite foundation to understand the cognitive processes individuals engage in before self-disclosing in online environments. SET posits that, before engaging in a relationship, individuals weigh the costs and benefits of the interaction, and decide whether to engage in the relationship. SET specifies that individuals actively assess the potential benefits and drawbacks of any activity before engaging in that behaviour. Using this foundation, individuals engage in activities that promote relationships when the perceived costs associated with that behaviour are less than the benefits expected from the action (Kankanhalli et al., 2005). Individuals engage in activities with the expectation that they will receive intangible benefits from the interaction (Gefen & Ridings, 2002). These perceived benefits are evaluated against perceived costs to act as a cognitive guide for the individual. Applied in an IS research context, research based on SET has shown that employees are more willing to exchange information and allow their organizations to claim ownership of that information if the employees believe the organization will reciprocate with increased recognition (Jarvenpaa & Staples, 2001). IS research has also indicated that professionals in electronic networks of practice tend to contribute more of their knowledge, as well as more helpful knowledge, to members in the network if the professionals perceive that the contribution will reciprocally increase their reputation in their profession (Wasko & Faraj, 2005). IS researchers have also applied SET to explain the cognitive process users engage in when deciding whether to contribute knowledge to knowledge repositories (Kankanhalli et al., 2005). Regarding the implementation of customer relationship management (CRM) systems, SET has been utilized to explain how implementation team responsiveness influences user evaluations and approval of the system (Gefen & Ridings, 2002). SET has also been applied to show the effects of trust on 8

knowledge sharing in virtual teams (Staples & Webster, 2008), the impact of knowledge sharing on IS outsourcing success (Lee, 2001) and the effect of disclosure on relationships with electronic trading partners (Son et al., 2005). While SET provides a relationship foundation for our model of self-disclosure, it is still more about relationships than self-disclosure. Extending SET, social penetration theory (SPT) (Altman & Taylor, 1973) applies the essential concepts of SET to interpersonal communications generally and to self-disclosure specifically. SPT explains and predicts relational closeness, which is seen in the superficiality or depth of the self-disclosures in a relationship. SPT posits that ‗people assess interpersonal rewards and costs, satisfaction and dissatisfaction, gained from interaction with others, and that the advancement of the relationship is heavily dependent on the amount and nature of the rewards and costs‘ (Altman & Taylor, 1973, p.6; Altman et al., 1981; Taylor & Altman, 1975). Such rewards may be exhibited in the form of reciprocal disclosures from relational partners and increased liking, while costs may take the form of increased vulnerability and risks related to others (Altman & Taylor, 1973). As long as the cost-benefit differential remains positive, relational engagement through disclosure is likely to progress. Reciprocation is the foundation of self-disclosure benefits, whereas risk is the foundation of selfdisclosure costs. To better understand self-disclosure, SPT provides a powerful metaphor: individuals are like onions in that they possess many layers that collectively form an individual‘s total personality (Altman & Taylor, 1973). The outer or peripheral layers store more visible information about the individual, which can be assessed quite easily by others without much probing (e.g., biographical items). As the layers progress toward the centre, they contain information about the individual that is of increasing vulnerability and/or social undesirability. 9

The deeper the characteristic resides in one‘s onion, the more the characteristic reflects one‘s total personality (Altman & Taylor, 1973). These more central layers are reached as relationships progress (Wolfe & Murthy, 2006), which acts as ‗a continuously widening and deepening ―wedge,‖…proceeding to more intimate layers of exchange but also expanding at prior levels of interaction‘ (Altman & Taylor, 1973, p. 39). People do not automatically self-disclose important information about themselves, despite their desire for acceptance and relational formation. Like onions, humans maintain protective outer layers that surround a delicate, central core representing the true, unadulterated self (Altman & Taylor, 1973). Such distal layers are initial impediments in the self-disclosure process; thus, they are not shed all at once. Rather, the segments of outer layers must first be exposed, experienced and peeled in succession before the inner, intimate layers are revealed. This gradual escalation of the revealing process is termed social penetration (Altman & Taylor, 1973) and provides the foundation for many communication studies in relational development (e.g., Gudykunst & Nishida, 1986; Hensley, 1996; VanLear, 1987, 1991; Walther & Burgoon, 1996). As individuals disclose more and more information regarding themselves (i.e., amount) to other members, it is likely that the disclosures will tend to reach toward the more central, more intimate cores (i.e., depth) of the relational partners during the progression of the relationships (Wheeless, 1978; Wheeless & Grotz, 1976). Aside from the traditional cost-benefit approach to predicting self-disclosure, we believe that these models underestimate the power an individual‘s cultural inclinations (e.g., which can be exhibited in terms of Hofstede‘s (1991) cultural dimensions) toward interacting and reciprocating with others. These cultural inclinations should directly affect the degree to which they feel comfortable with and inclined toward self-disclosure. The more that individuals are 10

inclined to reciprocate disclosures with others, the more they will want to self-disclose, and we submit that their cultural dimensions directly affect this inclination to reciprocate. Furthermore, we argue that social influence is a major factor, and provide additional support to fully account for this in a way that the previous models do not. Figure 1 summarizes our theoretical extension of SET and SPT, which we call the Online Community Self-Disclosure Model.

Figure 1. High-Level Online Community Self-Disclosure Model Our particular operationalization of self-disclosure theory focuses on French and British working professionals who disclose information about themselves in online communities (e.g., MySpace and Facebook). To further operationalize our self-disclosure model, we examined key measures in the literature that were deemed to best represent the constructs of social influence to use an online community, social benefits, social costs and inclinations toward reciprocity. We further expand the notions of social costs and risks into measures of online community trust, privacy risk beliefs and anonymity. Finally, countries with individualistic cultural tendencies tend to be less open with others and less prone to reciprocity than those with collectivistic 11

tendencies; thus, these measures are surrogates for cultural tendencies toward self-disclosure. Our extension of our underlying model‘s constructs into measures is depicted in Figure 2. Further justification and predictions are provided in the next section.

Figure 2. Extended Online Community Self-Disclosure Model SOCIAL INFLUENCE TO USE AN ONLINE COMMUNITY It was somewhat surprising to find that social influence has not been fully represented in previous self-disclosure models. Social influence is the degree to which an individual‘s beliefs, attitudes and/or behaviours are influenced by others in his or her environment (Deutsch & Gerard, 1955). The effects of social influence and socially accepted norms on others have been 12

well documented in academic research (Fishbein & Ajzen, 1975; Venkatesh et al., 2003). For instance, both the theories of reasoned action (Fishbein & Ajzen, 1975) and planned behaviour (Ajzen, 1991) posit that an individual‘s perception of acceptable group norms drives the intention to engage in a specific behaviour. Bandura‘s (1977) social learning theory specifies that individuals‘ behaviours are learned responses from the behaviours of other individuals within the environment. A recent examination of the power of social influence delineates various methods by which individuals are influenced or persuaded by others (Cialdini, 2001). Individuals have been shown to engage in an activity if they know or believe that others in their environment are also engaging in the activity. Cialdini (2001) refers to this principle as social proof, which can be utilized to begin the social-confirmation process. Individuals are drawn to others who are attractive, who share similarities and who easily give praise (Cialdini, 2001). When disclosing sensitive information, individuals who are easily influenced are likely to use these principles as a basis for their disclosure activity in electronic communities. Therefore, individuals may alter the frequency and nature of their disclosures to become more similar to those in their environment and thereby reach conformity by increasing their perceptions of attractiveness. Those who are susceptible to social influences may increase the rate of their disclosures and the level of honesty in their disclosures or vary positive and negative disclosures to make themselves more attractive and likable in their online community. For brevity, we do not consider all forms of social influence in this study; instead, we base our conceptualization on that by Venkatesh et al. (2003), who focused on social influence to use a system. We apply the same in our context of online communities and for further clarification we refer to this construct as social influence to use an online community. In sum, 13

H1. An increase in the social influence to use an online community increases selfdisclosure in online communities. RECIPROCITY Reciprocity is a special form of social influence that provides the key driving benefit for self-disclosure, per SPT. Reciprocity, also termed the dyadic effect, may best be explained as quid pro quo communication, synonymous with a ‗you tell me and I‘ll tell you‘ (Jourard, 1971, pp. 25-26) mentality. Feelings of reciprocity signal to an individual that his or her relational partners are willing to accept a certain level of vulnerability to continue the relationship, thereby increasing the individual‘s assessment of the relationship‘s worth and the need to maintain it via future disclosures. This signal of relationship worth from reciprocity is a very positive message that fosters social bonding and intimacy that can be very satisfying and drive several perceived benefits (Ellison et al., 2007; Ko & Kuo, 2009), including increased well-being due to increased social support and social integration, bonding social capital and bridge social capital. This reciprocal self-disclosure can be the core of building highly intimate relationships that are very rewarding and that even enhance social contact, satisfaction and one‘s overall quality of life. Reciprocity is not only a benefit but also drives further self-disclosure. SPT suggests that by increasing the perceived worth of an interaction, one will likely disclose more personal/private information to maximize the benefit of the interaction (Kankanhalli et al., 2005). For example, as disclosure recipients acquire their relational partners‘ personal information over time, the recipients feel indebted to respond to the received messages at a similar level of intimacy or a similar depth of their multi-layered self. This reciprocal communication allows individuals to successfully test deeper and deeper layers of partners to extract information residing at the centre core (Derlega et al., 1993) by further driving the communication wedge 14

(Altman & Taylor, 1973). In communication, ‗there is substantial evidence that people will engage in intimate self-disclosure—even with relative strangers—if they first become the recipients of such disclosure from their conversational partners‘ (Moon, 2000, p. 324). This has been demonstrated in online community self-disclosures where, essentially, by disclosing, the norm of disclosure is created, and the frequency of self-disclosure increases over time (DietzUhler et al., 2005)—especially when online self-disclosures are highly personal and involve emotional support (Barak & Gluck-Ofri, 2007). In summary, H2. An increase in perceived reciprocity increases self-disclosure in online communities. SOCIAL RISKS AND COSTS Trust in online community. In addition to reciprocity, researchers have noted the positive implication of trust in disclosure in a SET context (Lee, 2001; Staples & Webster, 2008). In fact, trust is the element that binds social exchanges (Pavlou & Gefen, 2005). Research has shown the relative importance of trust in forming behavioural intentions to engage in online shopping (Gefen et al., 2003) and in providing a foundation for effective online marketplace exchanges (Ba & Pavlou, 2002; Pavlou & Dimoka, 2006; Pavlou & Gefen, 2004). Trust is a necessity or prerequisite for honest communication in interpersonal communication (Emmert & Donaghy, 1981). We adapted Mayer et al.‘s (1995) conception of trust and define online community trust as the degree to which an individual believes that those within his or her selected online community are reliable and are trustworthy with information that makes the individual vulnerable. From a social exchange and social penetration perspective, individuals who perceive that their relational partners can be trusted tend to disclose more, as high levels of trust decrease the risks associated with releasing sensitive information. In virtual 15

communities, members have been shown to contribute knowledge of higher quality when they feel that they can trust the other members in their community (Chiu et al., 2006). Even trust in online brands has been shown to increase the propensity to disclose information (DelgadoBallester & Hernández-Espallardo, 2008; Lowry et al., 2008). Recently, a study showed how online communities can foster trust and that such trust increases self-disclosure in customers (Porter & Donthu, 2008). Therefore, we posit that trust in an online social network will provide a necessary decrease in perceived risks and will allow an individual to enjoy the benefits received from disclosing information about him or herself within that network of individuals, thereby adjusting the cost-benefit scales to his or her advantage. In summary, H3. An increase in online community trust increases self-disclosure in online communities. Privacy risk beliefs. Privacy concerns have long been a major factor that have held people back from releasing information online (Awad & Krishnan, 2006; Malhotra et al., 2004). These issues naturally extend to online self-disclosure. A recent extension to social penetration theory, called communication privacy management (CPM) theory (Petronio, 2002), specifically focuses the social exchange and penetration perspectives on the development and maintenance of individuals‘ communication privacy boundaries. CPM theory suggests that individuals maintain and coordinate many privacy boundaries with various communication partners depending on the perceived benefits and costs of self-disclosure. Individuals self-disclose in communication activities when the recipients and senders‘ privacy boundaries overlap, creating a mutual, collective boundary of privacy for disclosure. Such boundaries are desired because they lower the partners‘ privacy risk beliefs while simultaneously providing a channel for the benefits of disclosure to be attained. Privacy risk 16

beliefs are ‗the expectation that a high potential for loss is associated with the release of personal information‘ to others in their electronic communities (Malhotra et al., 2004, p. 341). Unless these beliefs are lowered, individuals may likely perceive the costs of disclosing to be too high, thus forcing the individuals to refrain from disclosing any sensitive information about themselves. The more a person believes that disclosing personal information online is risky, the less information he or she will likely disclose. Therefore, individuals who believe that their privacy boundaries effectively minimize the risks associated with self-disclosure will engage in personal communication activities with others in their electronic community. H4. An increase in privacy risk beliefs decreases self-disclosure in online communities. Anonymity. Our hypothesis on anonymity is a special extension of privacy risk beliefs. Anonymity, which on its most basic level can be defined as the lack of personal identification, has been shown to be a factor that decreases inhibition (e.g., decreased evaluation apprehension), allowing individuals to share information that they would not otherwise feel comfortable sharing (Connolly & Jessup, 1990; Lea et al., 2001; Nunamaker Jr. et al., 1991; Pinsonneault & Heppel, 1998). Disinhibition occurs when an individual feels free to perform public behaviours and is predicted by the degree to which he or she experiences public self-awareness and private selfawareness (Pinsonneault & Heppel, 1998). Public self-awareness ‗involves attention to oneself as a social object and concerns appearance and the impressions made in social situations‘ (Pinsonneault & Heppel, 1998, p. 94). Private self-awareness refers to ‗a focus on personal aspects of oneself, like perceptions, thoughts, and feelings‘ (Pinsonneault & Heppel, 1998, p. 95). Anonymity can significantly affect disinhibition and other behaviours only when social evaluation is an important source of inhibition (public self-awareness) (Pinsonneault & Heppel, 17

1998, p. 97). We posit that these social ties to anonymity provide a strong link to situations of self-disclosure, because self-disclosure involves people as social objects who are likely concerned about the impressions they make. Therefore, anonymity should provide more disinhibition to individuals in such social contexts, decreasing the risk of disclosing information and allowing individuals to feel more comfortable about self-disclosure. Notably, we posit that perceived anonymity is a more important influence than actual anonymity, as it is perceptions and beliefs that drive behaviours (Ajzen, 1991). Accordingly, less risk should increase selfdisclosure according to SET and SPT, and we predict: H5. An increase in perceived anonymity increases self-disclosure in online communities. INCLINATIONS TOWARD RECIPROCITY AND SOCIAL INFLUENCE Perceived Collectivism and Individualism. Finally, while several factors could account for an individual‘s inclinations toward reciprocity, we consider his or her cultural dimensions because these have a strong bearing on social influence, which is key to SPT. We submit that a theoretical basis for predicting this that will likely be useful is based on the theoretical cultural concepts of individualism and collectivism. Hofstede (1991) defines culture as the ‗collective programming of the mind which distinguishes the members of one group or category of people from another‘ (p. 5). Similarly, more recent research defines culture as ‗a system of implicit and explicit beliefs, values, norms, preferences, and behaviours that are stable over time, held in common by a group of people, and that distinguish one group from others‘ (Zhang & Lowry, 2008, p. 64). Individualism and collectivism are the most studied and common cultural dimensions in the IS literature (Shin et al., 2007). We leverage these concepts here. Individualism ‗describes 18

cultures in which the ties between individuals are loose,‘ and collectivism ‗describes cultures in which people are integrated into strong, cohesive groups that protect individuals in exchange for unquestioning loyalty‘ (Hofstede, 1991; Zhang & Lowry, 2008, p. 65). From individualismcollectivism, four types of cultural tendencies have been identified: (1) horizontal individualism (when people have a tendency to strive to be unique and do their own thing); (2) vertical individualism (when people want to do their own thing and strive to be the best); (3) horizontal collectivism (when people merge themselves with their in-groups); and (4) vertical collectivism (when people submit to the authorities of the group and are willing to sacrifice themselves for their group) (Triandis, 2001; Triandis & Suh, 2002). We submit that the cultural dimensions of individualism-collectivism likely have a large impact on whether a person is inclined to be socially influenced and to reciprocate in selfdisclosure. The key difference between these cultural dimensions is that those who are more collectivist tend to be more cohesive and integrated with other people in their interactions, whereas those with an individualistic inclination have much looser ties to people. Individualists emphasize uniqueness and independence in interpersonal interactions, while collectivists feel ‗duty to the in-group,‘ where ‗the in-group refers to a group of people sharing similar beliefs and interests and which typically excludes outsiders‘ (Husted & Allen, 2008, p. 295). Given these differences, which have been shown in a large body of literature, we submit that those with strong collectivistic tendencies are more prone to social influence and reciprocity, while those with strong individualistic tendencies are less likely to be prone to these influences. Hofstede‘s (1991) additional dimensions of power distance, uncertainty avoidance and masculinity were not included in our study because they do not directly relate to social networks, but are rather salient within particular, well-defined organizations. For example, power distance 19

relates to inequality among individuals; however, in the social network, individuals have equal access and a level playing field. Second, individuals participating in online social networks have already embraced uncertainty by joining the online social network. A key limiting assumption in this research is that while it is true that some cultures and people tend to lean toward either individualism or collectivism (Zhang et al., 2008), recent research shows that individualism and collectivism themselves are not dichotomous—they are independent dimensions. Thus, a person‘s score in one does not necessarily reflect his or her score in another (Husted & Allen, 2008; Oyserman, 2006; Triandis & Gelfand, 1998). Hence, a typical person will exhibit some degree of individualism and some degree of collectivism. In sum, H6. An increase in perceived collectivism increases self-disclosure in online communities. H7. An increase in perceived individualism decreases self-disclosure in online communities. Figure 3 depicts our full operational model. METHODOLOGY For our data collection, we hired a market research firm to provide us with randomly selected British and French participants from an online panel of working professionals who at least occasionally use online communities and social networking sites. Our collection was specifically intended to avoid traditional college students (ages 18-22) enrolled in universities, as they have been the focus of most online community research. There were 529 participants in total; 263 were from France and 266 were from the UK. Of the British participants, the average age was 36.0; the average level of Internet experience was 6.1 out of 7 (highly experienced). Of the participants, 43.4% were male, 54.4% 20

Figure 3. Operational Online Community Self-Disclosure Model were female, and 2.2% did not disclose gender; 55.9% carefully use privacy settings for online social networking; 52.2% frequently use Facebook; and 26.1% frequently use MySpace. In terms of British participants‘ use of any online community, 23.6% participate in an online community less than once a month, 15.1% participate a few times a month, 9.2% participate every week, 25.0% participate several times a week, 10.7% use their community once a day and 16.5% use their community several times a day. 21

Of the French participants, the average age was 33.6; average level of Internet experience was 5.9 out of 7 (highly experienced). Of the participants, 46.8% were male and 53.2% were female; only 27.0% carefully use privacy settings for online social networking; 22.4% frequently use Facebook; and 14.4% frequently use MySpace. In terms of French participants‘ use of any online community, 30.4% participate in an online community less than once a month, 25.5% participate a few times a month, 9.5% participate every week, 17.9% participate several times a week, 9.9% use their community once a day and 6.85% use their community several times a day. MEASURES As detailed later in the analysis section, we were careful to establish cross-cultural equivalence (that measures and constructs were equal in both data collection settings) in our measures and general research setting. This was particularly important in the present case because the measures had to be carefully translated and back-translated from English to French. Establishing cross-cultural equivalence is a way of confirming validity of measurement to ensure that differences in the results are not due to cultural differences in how respondents understand the underlying measurement. In other words, cross-cultural equivalence exists if respondents from both cultures and languages have the same understanding of what the measurement item means (Karahanna et al., 2002; Lowry et al., 2009b). Online Appendix 1 details the specifics of all measures that were used, with slight wording changes to focus on an online community context. Online community trust was adapted from Jarvenpaa and Leidner‘s (1999) online customer trust measure. Privacy risk beliefs were from Malhotra et al. (2004). Social influence was from Venkatesh et al. (2003), and specifically dealt with social influence to use a system. Reciprocity was taken from Kankanhalli et al. (2005) and Wasko et al. (2005). Anonymity was created only to represent lack of identification, as based on Pinsonneault and Heppel‘s 22

conceptualization (1998). The collectivism-individualism measures, including the horizontal and vertical dimensions, were from Triandis and Gelfand (1998). ANALYSIS Due to space limitations, most of the details of our analysis are captured in our supplementary online Appendix 1. As an overview, we performed a partial least squares (PLS) analysis, using PLS-GRAPH version 3.0. We first established factorial validity on our measures using the latest techniques for convergent validity and discriminant validity, as explained and demonstrated previously (Chin et al., 2003; Lowry et al., 2009a; Lowry et al., 2008), and we demonstrated high composite reliability, as summarized in Table 1. Next, we used a couple of techniques to establish that our study was not subject to common methods bias. We then carefully established the cross-cultural equivalence of our French and British samples. Table 1. Composite Reliability Construct Social influence to use an online community Reciprocity Trust in online community Privacy risk beliefs Anonymity Self-disclosure: amount Self-disclosure: valence Self-disclosure: intent Self-disclosure: honesty Self-disclosure: depth Horizontal individualism Horizontal collectivism Vertical individualism Vertical collectivism

Composite reliability 0.86 0.85 0.94 0.87 0.89 0.87 0.85 0.83 0.87 0.86 0.76 0.80 0.79 0.83

TESTING THE MODEL Given these background checks, we then tested the path model. Figure 4 summarizes the results of the interaction model. Variance explained is indicated for each construct as R 2. The 23

path coefficients, or betas (β‘s), are indicated on the paths between two constructs, along with their direction and significance. The significance of the path estimates was calculated using a bootstrap technique with 200 re-samples. The second-order, formative construct of selfdisclosure is composed of the first-order subconstructs of amount, depth, honesty, intent and valence. Table 2 summarizes the hypotheses, the path coefficients and the t-values for each path. We also confirmed that self-disclosure can be conceptualized as a second-order factor composed of amount, depth, honesty, intent and valence. Honesty, depth and amount were shown to be the strongest contributing factors.

24

Figure 4. Model Testing Results Table 2. Summary of Path Coefficients and Significance Levels Tested path

Hypotheses H1. Social influence to use an online community  selfdisclosure H2. Reciprocity  self-disclosure H3. Online community trust  self-disclosure H4. Privacy risk beliefs  (-) self-disclosure H5. Anonymity  self-disclosure H6. Collectivism  self-disclosure H7. Individualism  (-) self-disclosure

Path coefficient (β)

t-value (df = 529)

0.135

2.83**

0.183 0.270 (-0.268) 0.061 0.102 0.080

4.16*** 4.07*** 5.03*** 1.41 (n/s) 2.35* 1.79 (n/s)

(-0.019) (0.044)

0.52 (n/s) 1.05 (n/s)

Covariates Education  self-disclosure Age  self-disclosure

Confirmation of second-order factor of self-disclosure Amount is a first-order subconstruct of self-disclosure 0.374 Depth is a first-order subconstruct of self-disclosure 0.316 Honesty is a first-order subconstruct of self-disclosure 0.417 Intent is a first-order subconstruct of self-disclosure 0.208 Valence is a first-order subconstruct of self-disclosure 0.156 *p < 0.05, ** p < 0.01, ***p < 0.001

10.95*** 12.00*** 11.47*** 7.70*** 5.81***

EXPLORING DIFFERENCES BETWEEN FRANCE AND THE UK To further address RQ2, we statistically compared the major IVs in the study to see if there were any differences between the French and British participants (see Table A6).We first found that there was no statistical difference in overall individualism or collectivism between the countries; however, French participants did have higher scores on horizontal individualism than British participants. Overall, British participants had higher self-disclosure rates than French participants. Finally, we ran separate path models for each country to see if there were any differences between the two countries. These comparisons are summarized in Table 3. To be able to statistically compare the path coefficients between the two models required additional multi25

Table 3. Summary of France vs. UK Models France model:

Tested path

Path coefficient (β)

t-value (df = 263)

UK model: Path coefficient (β)

Hypotheses H1. Social influence to use an 0.232 2.84** 0.128 online community  selfdisclosure H2. Reciprocity  self-disclosure 0.187 2.68** 0.158 H3. Online community trust  self- 0.256 1.93 (n/s) 0.182 disclosure H4. Privacy risk beliefs  (-) self(-0.268) 1.93 (n/s) (-0.289) disclosure H5. Anonymity  self-disclosure 0.044 0.67 (n/s) 0.081 H6. Collectivism  self-disclosure 0.052 0.44 (n/s) 0.187 H7. Individualism  (-) self(-0.090) 0.63 (n/s) 0.166 disclosure *p < 0.05, ** p < 0.01, ***p < 0.001

t-value (df = 266)

Statistical difference in comparing the paths?

1.46 (n/s)

No

1.87 (n/s) 1.98*

No No

3.47**

No

1.23 (n/s) 2.40* 1.75 (n/s)

No No No

group analysis, as detailed in Table A7. To summarize, we found no statistical differences between the paths in the models, and the effect sizes in the differences in the paths were negligible. This allows us to predict one overall model for both cultures. DISCUSSION A key theoretical contribution of this study was the proposal of a theoretical model of online self-disclosure—the online community self-disclosure model—based on SET and SPT. SET explains that individuals engage in relationships when the perceived costs associated with the relationship are less than the expected benefits from the action (Kankanhalli et al., 2005). SPT extends SET to further explain that individuals participate in self-disclosure to foster relationships, reciprocation being the foundational benefit of self-disclosure while risk is the foundational cost of self-disclosure. To complete this model, we also included social influence to use an online community and cultural factors, both of which affect self-disclosure. The 26

remainder of this section summarizes additional results and contributions in testing our model, per the research questions that drove this study. RQ1. FACTORS PREDICTING SELF-DISCLOSURE Our study confirmed several important hypotheses related to our theoretical model of online self-disclosure. In our data sample, social influence to use an online community increases online community self-disclosure (H1). The primary self-disclosure benefit of reciprocity also increases self-disclosure (H2). Meanwhile, two risk factors in our model have an additional effect: Online community trust increases self-disclosure (H3), whereas privacy risk beliefs decrease self-disclosure (H4). However, the risk factor of anonymity—in terms of lack of identification—had no impact on self-disclosure (H5). RQ2. INDIVIDUAL-LEVEL AND NATIONAL-LEVEL CULTURAL INFLUENCES In terms of individual-level cultural differences, we found that a tendency toward collectivism increases self-disclosure (H6). A tendency toward individualism had no impact on self-disclosure (H7). The primary theoretical basis for this tie to our model is that a key factor and benefit in self-disclosure is reciprocity. Meanwhile, the tendency toward reciprocal behaviour and communication and relationships is much higher in collectivists; thus, they are more prone to self-disclose than individualists. We also found that there were no statistical differences in individualism or collectivism on the national levels of France or the UK. However, we did find that French participants had higher scores on horizontal individualism than British participants. Notably, the horizontal individualism scale has been associated with people who are sociable, have high family integrity, have high interdependence on others, yet are highly competitive and highly self-reliant (Triandis & Gelfand, 1998). 27

Moreover, British participants exhibited higher self-disclosure rates than French participants; however, this is potentially misleading, as there were stark differences in the underlying subdimensions of self-disclosure. In our study, British participants had higher scores on amount and valence, whereas French participants had higher scores on honesty and intent. No other statistical differences were seen between the countries‘ IVs. As noted in our demographic analysis, more of the UK participants were frequent involved in online community use than the French participants, which could also account for some of these differences. In our demographic data, we found that despite having virtually the same Internet experience, 55.9% of British participants used advanced privacy settings when using their online communities; this was true for only 27% of the French participants. Finally, we ran separate path models for each country to see if there were any differences between them. We found in the French model that social influence to use an online community and reciprocity had the highest influence on self-disclosure; in the British model, online community trust and privacy risk beliefs had the highest influence on self-disclosure. These differences suggest that there could be highly salient differences between the French and British samples affecting the kinds of self-disclosure, as well as differences in the factors that affect selfdisclosure. APPLICATION TO RESEARCH AND PRACTICE Our model and results indicate that privacy risk beliefs inhibit self-disclosure within online communities, and it appears these are stronger factors in British participants than in French participants. There may be several alternative explanations for these results, including different information laws between the countries and different identification requirements when opening accounts or establishing membership in social networks, identity theft awareness 28

programs, security education programs, the type and quality of government-issued identity documents and legal penalties for violating information laws. Therefore, such considerations may need to be localized to various national versions of online community Web sites. Naturally, those who worry about privacy do not want to reveal their information online. From a community-engagement perspective, it is critical to encourage participants to self-disclose in these communities so that they will build the relationships necessary for a satisfying experience and continuance with the communities. Non-disclosers are also particularly troublesome to marketers who need accurate information about target customers. We believe that less intrusive designs can lead people to shed additional layers of their personal ‗onions,‘ on the basis that lower levels of self-disclosure will lead to higher levels over time, as supported by the literature. The anonymity results are particularly interesting and are a key contribution of this study. Recall that anonymity helps limit evaluation apprehension and other factors that inhibit people from sharing information with others, and thus increases disinhibition. One might intuitively conclude that anonymity should foster more self-disclosure, but this was not the case in our study. We think this result can be better explained by looking at the subdimensions of selfdisclosure (amount, depth, honesty, valence and intent), which reveal an underlying tension with anonymity. While amount could be positively affected by not identifying oneself because one may be less inhibited, we realize that anonymity could be completely counter to the notions of honesty and depth. We think that anonymity could be counter to the need for intimacy, openness and honesty, which is the essence of self-disclosure. By understanding what fosters online self-disclosure in online communities of different cultural dimensions, sites can be better designed to improve relationship building and trust formation and to encourage self-disclosures that bring positive benefits to the participants and to 29

businesses. Designers and researchers should first look at manipulating the salient factors that we found have a strong impact on online-community self-disclosure: social influence to use an online community, reciprocity, online community trust, privacy risk beliefs, collectivism and horizontal-individualism. For example, providing active moderators or online activists in a professional community could be a positive factor for social influence to use an online community. LIMITATIONS AND FUTURE RESEARCH A limitation of this study is that there is more to the broader concept of anonymity than lack of personal identification; other factors can include diffused responsibility, proximity, confidence in the system being used and knowledge of others using the system (Pinsonneault & Heppel, 1998). The common thread between these factors is that they foster disinhibition (Pinsonneault & Heppel, 1998). Hence, it is possible that while lack of identification does not foster self-disclosure, there could be other elements of anonymity that foster disinhibition and thus encourage self-disclosure. Related to anonymity, there is also the potential issue of people acting in deceptive and malicious manners in order to manipulate or confuse an online community. Future research should further address these gaps. Our preliminary results suggest that the French participants are much more socially influenced and social benefits–oriented than the British participants, who are more concerned about self-disclosure risk factors. Interestingly, a separate recent study showed that the French are more than twice as likely to read and write blogs than the British; blogs can also involve intimate forms of self-disclosure (Bernoff & Li, 2008). Furthermore, in our study, French participants scored higher on the horizontal-individualism scale. However, we cannot definitively conclude that these differences alone can theoretically account for the differences in 30

these two national cultures. More theory and research need to be developed to examine other factors. For example, another potential cultural indicator that may prove beneficial in this regard is that of uncertainty avoidance, which is highly related to risk privacy beliefs. Perhaps the most exciting potential individual-cultural and cross-cultural research that can be done is to look at design factors that can better influence participants of different cultural inclinations to selfdisclose online. Another opportunity is figuring out how to foster collectivism. Preliminary research shows that although individualism and collectivism are cultural dimensions of one‘s likely behaviour, everyone has some degree of both factors, and these can be manipulated in an organizational setting to enhance the collectivistic tendencies of workers (Triandis & Gelfand, 1998). Hence, we believe that this is also possible in an online community setting. The key would be to provide rewards for collectivistic supportive behaviour, because such behaviour encourages not only one‘s own self-disclosures but also the self-disclosures of others. Rewards could include an increase in public ‘‗authenticity/honesty‘ and ‗helpfulness‘ measures, which could rate users on these items based on evaluations of their self-disclosures. Other limitations and research opportunities are that our data represented a ‗snapshot‘ of one point in time, and we did not have direct access to participants' actual self-disclosure on the social network to gauge the veracity of their responses related to their self-disclosure. Our method of data collection through Zoomerang insured our respondents‘ complete privacy, and their actual identities were not revealed. We did not observe the process of social penetration itself, which can be accomplished by conducting longitudinal studies (VanLear, 1987, 1991). Further, there are other potential elements of SPT that could be studied—including satisfaction, stability and security in a relationship—to predict self-disclosure. However, these are trickier to 31

study because researchers would have to look at particular dyadic relationships in a larger online community. Finally, while this research assumes that self-disclosure is somewhat of an unqualified ‗good‘ for building on online community, facilitating market research, and promoting business through the ‗contribution revolution,‘ it is important to emphasize that the latter two points have ethical considerations that need further examination. Clearly, marketing and business uses of self-disclosure can cross ethical guidelines and violate privacy and what is right and wrong in this regard may differ by culture and country. REFERENCES AJZEN I (1991) The theory of planned behavior. Organizational Behavior and Human Decision Processes 50(2), 179-211. ALTMAN I and TAYLOR D (1973) Social Penetration: The Development of Interpersonal Relationships. Holt, Rinehart & Winston, New York. ALTMAN I, VINSEL A and BROWN B (1981) Dialectic conceptions in social psychology: An application to social penetration and privacy regulation. Advances in Experimental Social Psychology 14(1981), 107-160. ANONYMOUS (2009a) Fact sheet. Myspace, Retrieved September 16, 2009, from http://www.myspace.com/pressroom?url=/fact+sheet/. ANONYMOUS (2009b) Press room: Statistics. Facebook, Retrieved September 15, 2009, from http://www.facebook.com/press/info.php?statistics. ANONYMOUS (2009c) Usage and population statistics. Internet World Stats, Retrieved March 15, 2009, from http://www.internetworldstats.com/. AWAD NF and KRISHNAN MD (2006) The personalization privacy paradox: An empirical evaluation of information transparency and the willingness to be profiled online for personalization. MIS Quarterly 30(1), 13-28. BA SL and PAVLOU PA (2002) Evidence of the effect of trust building technology in electronic markets: Price premiums and buyer behavior. MIS Quarterly 26(3), 243-268. BANDURA A (1977) Social Learning Theory. Prentice-Hall, Englewood Cliffs. BARAK A and GLUCK-OFRI O (2007) Degree and reciprocity of self-disclosure in online forums. CyberPsychology & Behavior 10(3), 407-417. BERNOFF J and LI. C (2008) Harnessing the power of the oh-so-social Web. MIT Sloan Management Review 49(3), 36-42. CHANG LEE K and KWON S (2008) A cognitive map-driven avatar design recommendation DSS and its empirical validity. Decision Support Systems 45(3), 461-472. CHIN WW, MARCOLIN BL and NEWSTED PR (2003) A partial least squares latent variable modeling approach for measuring interaction effects: Results from a Monte Carlo 32

simulation study and an electronic-mail emotion/adoption study. Information Systems Research 14(2), 189-217. CHIU C-M, HSU M-H and WANG ETG (2006) Understanding knowledge sharing in virtual communities: An integration of social capital and social cognitive theories. Decision Support Systems 42(3), 1872-1888. CIALDINI RB (2001) Influence: Science and Practice. (4th ed.). Allyn & Bacon, Boston. CONNOLLY T and JESSUP LM (1990) Effects of anonymity and evaluative tone on idea generation in computer-mediated groups. Management Science 36(6), 689-703. COOK S (2008) The contribution revolution: Letting volunteers build your business. Harvard Business Review 86(10), 60-66. DAVIS FD (1989) Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology. MIS Quarterly 13(3), 319-340. DAVIS FD, BAGOZZI RP and WARSHAW PR (1989) User Acceptance of Computer Technology: A Comparison of Two Theoretical Models. Management Science 35(8), 982-1003. DELGADO-BALLESTER E and HERNÁNDEZ-ESPALLARDO M (2008) Effect of brand associations on consumer reactions to unknown on-line brands. International Journal of Electronic Commerce 12(3), 81-113. DERLEGA VJ, METTS S, PETRONIO S and MARGULIS ST (1993) Self-Disclosure Sage Publications, Newbury Park. DEUTSCH M and GERARD HB (1955) A study of normative and information social influences upon individual judgment. Journal of Abnormal and Social Psychology 51(3), 629-636. DIETZ-UHLER B, BISHOP-CLARK C and HOWARD E (2005) Formation of and adherence to a self-disclosure norm in an online chat. CyberPsychology & Behavior 8(2), 114-120. ELLISON NB, STEINFIELD C and LAMPE C (2007) The benefits of facebook ―friends:‖ Social capital and college students' use of online social network sites. Journal of Computer-Mediated Communication 12(4), 1143-1168. EMMERT P and DONAGHY WC (1981) Human Communication: Elements and Contexts. Addison-Wesley Publishing Company, Reading, MA. FISHBEIN M and AJZEN I (1975) Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research. Addison-Wesley, Reading, MA. FORMAN C, GHOSE A and WIESENFELD B (2008) Examining the relationship between reviews and sales: The role of reviewer identity disclosure in electronic markets. Information Systems Research 19(3), 291-313. GEFEN D, KARAHANNA E and STRAUB DW (2003) Trust and TAM in online shopping: An integrated model. MIS Quarterly 27(1), 51-90. GEFEN D and RIDINGS CM (2002) Implementation team responsiveness and user evaluation of customer relationship management: A quasi-experimental design study of social exchange theory. Journal of Management Information Systems 19(1), 47-69. GEFEN D and STRAUB DW (2005) A practical guide to factorial validity using PLS-Graph: Tutorial and annotated example. Communications of the AIS 16(5), 91-109. GREENBERG MA and STONE AA (1992) Emotional disclosure about traumas and its relation to health: Effects of previous disclosure and trauma severity. Journal of Personality and Social Psychology 63(1), 75-84. GUDYKUNST WB and NISHIDA T (1986) The influence of cultural variability on perceptions 33

of communication behavior associated with relationship terms. Human Communication Research 13(2), 147-166. HARGITTAI E (2007) Whose space? Differences among users and non-users of social network sites. Journal of Computer-Mediated Communication 13(1), 276-297. HENSLEY WE (1996) A theory of the valenced other: The intersection of the looking-glass self and social penetration. Social Behavior and Personality 24(3), 293-308. HOFSTEDE G (1991) Cultures and Organizations: Software of the Mind. McGraw-Hill Book Company Europe, Berkshire, England. HORENSTEIN VD-P and DOWNEY JL (2003) A Cross-Cultural Investigation of SelfDisclosure. North American Journal of Psychology 5(3), 373-386. HUSTED B and ALLEN D (2008) Toward a model of cross-cultural business ethics: The impact of individualism and collectivism on the ethical decision-making process. Journal of Business Ethics 82(2), 293-305. IM S, LEE D-H, TAYLOR CR and D'ORAZIO C (2008) The influence of consumer selfDisclosure on Web sites on advertising response. Journal of Interactive Advertising 9(1), 87-106. ISHII K and OGASAHARA M (2007) Links between real and virtual networks: A comparative study of online communities in Japan and Korea. CyberPsychology & Behavior 10(2), 252-257. JARVENPAA SL and STAPLES DS (2001) Exploring perceptions of organizational ownership of information and expertise. Journal of Management Information Systems 18(1), 151183. JARVENPAA SL and TRACTINSKY N (1999) Consumer trust in an internet store: A crosscultural validation. Journal of Computer-Mediated Communication 5(2), 1-35. JOURARD S (1971) Self-Disclosure: An Experimental Analysis of the Transparent Self. Wiley & Sons, Inc., New York. KANKANHALLI A, TAN BCY and WEI K-K (2005) Contributing knowledge to electronic knowledge repositories: An empirical investigation. MIS Quarterly 29(1), 113-144. KARAHANNA E, EVARISTO R and STRITE M (2002) Methodological Issues in MIS CrossCultural Research. Journal of Global Information Management 10(1), 48-55. KO H-C and KUO F-Y (2009) Can blogging enhance subjective well-being through selfdisclosure? CyberPsychology & Behavior 12(1), 75-79. KOZINETS RV (2002) The field behind the screen: Using netnography for marketing research in online communities. Journal of Marketing Research 39(1), 61-72. LEA M, SPEARS R and DE GROOT D (2001) Knowing me, knowing you: Anonymity effects on social identity processes within groups. Personality and Social Psychology Bulletin 27(5), 526-537. LEE J (2001) The impact of knowledge sharing, organizational capability and partnership quality on IS outsourcing success. Information & Management 38(5), 323-335. LOWRY PB, ROMANO NC, JENKINS JL and GUTHRIE RW (2009a) The CMC interactivity model: How interactivity enhances communication quality and satisfaction in lean-media groups. Journal of Management Information Systems 26(1), 159-200. LOWRY PB, VANCE A, MOODY G, BECKMAN B, et al. (2008) Explaining and predicting the impact of branding alliances and Web site quality on initial consumer trust of ecommerce Web sites. Journal of Management Information Systems 24(4), 201-227. 34

LOWRY PB, ZHANG D, ZHOU L and FU X (2009b) Effects of culture, social presence, and group composition on trust in technology-supported decision-making groups. Information Systems Journal forthcoming(2009). MALHOTRA NK, KIM SS and AGARWAL J (2004) Internet users‘ information privacy concerns (IUIPC): The construct, the scale, and a causal model. Information Systems Research 15(4), 336-355. MARSHALL BA, CARDON PW, NORRIS DT, GOREVA N, et al. (2008) Social networking websites in India and the United States: A cross-national comparison of online privacy and communication. Issues in Information Systems 9(2), 87-94. MAYER RC, DAVIS JH and SCHOORMAN FD (1995) An integrative model of organizational trust. Academy of Management Review 20(3), 709-734. MILLER KD, FABIAN F and LIN S-J (2009) Strategies for online communities. Strategic Management Journal 30(3), 305-322. MOON Y (2000) Intimate exchanges: Using computers to elicit self-disclosure from consumers. Journal of Consumer Research 26(4), 323-339. MOORES TT and CHANG JC-J (2006) Ethical decision making in software piracy: Iniital development and test of a four-component model. MIS Quarterly 30(1), 167-180. NUNAMAKER JR. JF, DENNIS A, VALACICH J, VOGEL D, et al. (1991) Electronic meeting systems to support group work. Communications of the ACM 34(7), 40-61. OYSERMAN D (2006) High power, low power, and equality: Culture beyond individualism and collectivism. Journal of Consumer Psychology 16(4), 352-356. PAVLOU P, LIANG H and XUE Y (2007) Understanding and mitigating uncertainty in online exchange relationships: A principal-agent perspective. MIS Quarterly 31(1), 105-136. PAVLOU PA and DIMOKA A (2006) The nature and role of feedback text comments in online marketplaces: Implications for trust building, price premiums, and seller differentiation. Information Systems Research 17(4), 392-416. PAVLOU PA and GEFEN D (2004) Building effective online marketplaces with institutionbased trust. Information Systems Research 15(1), 37-59. PAVLOU PA and GEFEN D (2005) Psychological contract violation in online marketplaces: Antecedents, consequences, and moderating role. Information Systems Research 16(4), 372-399. PEARCE WB and SHARP SM (1973) Self-disclosing communication. Journal of Communication 23(4), 409-425. PENNEBAKER JW (1989) Confession, inhibition, and disease. In Advances in Experimental Social Psychology (Berkowitz L, Ed.), Vol 22, pp 211-244, Academic Press, New York. PETRONIO S (2002) Boundaries of Privacy: Dialectics of Disclosure. State University of New York Press, Albany. PINSONNEAULT A and HEPPEL N (1998) Anonymity in group support systems research: A new conceptualization, measure, and contingency framework. Journal of Management Information Systems 14(3), 89-108. PODSAKOFF PM, MACKENZIE SB, LEE JY and PODSAKOFF NP (2003) Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology 88(5), 879-903. PORTER CE and DONTHU N (2008) Cultivating trust and harvesting value in virtual communities. Management Science 54(1), 113-128. 35

RAYKOV T and GRAYSON D (2003) A test for change of composite reliability in scale development. Multivariate behavioral research 38(2), 143-159. ROBERTSHAW GS and MARR NE (2006) An empirical measure of the availability, completeness and reliability of voluntarily disclosed personal information for direct marketing purposes. Journal of Financial Services Marketing 11(1), 85-94. SHIN SK, ISHMAN M and SANDERS GL (2007) An empirical investigation of socio-cultural factors of information sharing in China. Information & Management 44(2), 165-174. SON J-Y, NARASIMHAN S and RIGGINS FJ (2005) Effects of relational factors and channel climate on EDI usage in the customer-supplier relationship. Journal of Management Information Systems 22(1), 321-353. STAPLES DS and WEBSTER J (2008) Exploring the effects of trust, task interdependence and virtualness on knowledge sharing in teams. Information Systems Journal 19(4), 475-495. STRAUB DW (1989) Validating instruments in MIS research. MIS Quarterly 13(2), 146-169. STRAUB DW, BOUDREAU MC and GEFEN D (2004) Validation guidelines for IS positivist research. Communications of the AIS 14(2004), 380-426. TAYLOR D and ALTMAN I (1975) Self-disclosure as a function of reward-cost outcomes. Sociometry 38(1), 18-31. THIBAUT JW and KELLEY HH (1959) The Social Psychology of Groups. John Wiley, New York. TRIANDIS HC (2001) Individualism-collectivism and personality. Journal of Personality 69(6), 907-924. TRIANDIS HC and GELFAND MJ (1998) Converging measurement of horizontal and vertical individualism and collectivism. Journal of Personality and Social Psychology 74(1), 118128. TRIANDIS HC and SUH EM (2002) Cultural influences on personality. Annual Review of Psychology 53(February), 133-160. VANLEAR CA (1987) The formation of social relationships: A longitudinal study of social penetration. Human Communication Research 13(3), 299-322. VANLEAR CA (1991) Testing a cyclical model of communicative openness in relationship development: Two longitudinal studies. Communication Monographs 58(4), 337-361. VENKATESH V, MORRIS MG, DAVIS GB and DAVIS FD (2003) User acceptance of information technology: Toward a unified view. MIS Quarterly 27(3), 425-478. WALTHER JB and BURGOON JK (1996) Relational communication in computer-mediated interaction. Human Communication Research 19(1), 50-88. WASKO MM and FARAJ S (2005) Why should I share? Examining social capital and knowledge contribution in electronic networks of practice. MIS Quarterly 29(1), 35-57. WHEELESS LR (1978) A follow-up study of the relationships among trust, disclosure, and interpersonal solidarity. Human Communication Research 4(2), 143-157. WHEELESS LR and GROTZ J (1976) Conceptualization and measurement of reported selfdisclosure. Human Communication Research 2(4), 338-346. WOLFE CJ and MURTHY US (2006) Negotiation support systems in budget negotiations: An experimental analysis. Journal of Management Information Systems 22(3), 351-381. ZHANG D and LOWRY PB (2008) Issues, limitations, and opportunities in cross-cultural research on collaborative software in information systems. Journal of Global Information Management 16(1), 61-92. 36

ZHANG D, LOWRY PB, ZHOU L and FU X (2008) 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.

37

ONLINE APPENDIX 1 Note to editors and reviewers: This appendix is intended for online distribution to supplement the published article; it is not intended to be published in the actual article. MEASUREMENT SCALES All scales were measured as 7-point Likert-like scales anchored on ―strongly disagree, moderately disagree, slightly disagree, neither disagree nor agree, slightly agree, moderately agree, strongly agree.‖ Questions were framed in the context of online communities. Construct Subconstruct Items Trust in Community N/A trust1. Overall, the people in my online social network are very trustworthy Adapted to online trust2. Members of my online social network are usually considerate of context from Jarvenpaa one another‘s feelings and Leidner (1999) trust3. The people in my online social network are friendly trust4. I can rely on those with whom I disclose personal information to in my online social network Privacy Risk Beliefs N/A prb1. In general, it is risky to give my private information to others Adapted to online online context from Malhotra, prb2. There is a high potential for loss associated with giving my Kim, and Agarwal personal information to others online (2004) prb3. There is too much uncertainty associated with giving my personal information to others online prb4. Providing others my private information online involves many unexpected problems *prb5. I feel safe giving my private information to others online Social Influence to use N/A si1. People who influence my behavior think that I should use an online an online community community si2. People who are important to me think that I should use an online Adapted to online community context from Venkatesh si3. I use an online community because of the proportion of my friends et al. (2003) that use it si4. People I am associated with who use an online community gain important benefits for using it si5. Having a personal profile on an online community is considered a status symbol Reciprocity N/A rec1. When others disclose personal information online, I believe that they expect me to do the same Based on Kankanhalli, rec2. Other online users trust me to return the favor of sharing personal Tan, and Wei (2005) information (Wasko & Faraj, 2005) rec3. I know that other users online disclose information about themselves, so it is only fair to do the same Self-disclosure Amount *amt1. I do not often talk about myself online amt2. I usually talk about myself for fairly long periods at a time Adapted to online *amt3. My conversation online lasts the least time when I am discussing context from (Wheeless, myself 1978; Wheeless & amt4. I often talk about myself online Grotz, 1976) amt5. I often discuss my feelings about myself online Depth dep1. I intimately disclose who I really am, openly and fully in my conversation online dep2. I often disclose intimate, personal things about myself without hesitation online

38

Honesty

Intent

Valence

Anonymity (Pinsonneault & Heppel, 1998)

n/a

Vertical and horizontal collectivism / individualism (Triandis & Gelfand, 1998)

n/a

dep3. I feel that I sometimes do not control my self-disclosure of personal or intimate things I tell about myself online dep4. Once I get started, I intimately and fully reveal myself in my selfdisclosures online hon1. I always feel completely sincere when I reveal my own feelings and experiences online hon2. My online self-disclosures are completely accurate reflections of who I really am * hon3. I am not always honest in my self-disclosures online Hon4. My statements online about my own feelings, emotions, and experiences are always accurate self-perceptions Hon5. I am always honest in my self-disclosures online int2. When I express my personal feelings online, I am always aware what I am doing and saying int3. When I reveal my feelings about myself online, I consciously intend to do so int4. When I am self-disclosing online, I am consciously aware of what I am revealing val1. I usually disclose positive things about myself online *val2. I normally reveal ―bad‖ feelings I have about myself online val3. I normally express my ―good‖ feelings about myself online val4. On the whole, my disclosures about myself online are more positive than negative Based on Pinsonneault‘s conceptualization of identification Id1. My identity can be distinguished from other users online Id2. Other users online know my real identity Id3. I believe none of the other users online know who I really am Id4. My identity is hidden from other users online 1. I‘d rather depend on myself than others. 2. I rely on myself most of the time; I rarely rely on others. 3. I often do ―my own thing.‖ 4. My personal identity, independent of others, is very important to me. 5. It is important that I do my job better than others. 6. Winning is everything. 7. Competition is the law of nature. 8. When another person does better than I do, I get tense and aroused. 9. If a coworker gets a prize, I would feel proud. 10. The well-being of my coworkers is important to me. 11. To me, pleasure is spending time with others. 12. I feel good when I cooperate with others. 13. Parents and children must stay together as much as possible. 14. It is my duty to take care of my family, even when I have to sacrifice what I want. 15. Family members should stick together, no matter what sacrifices are required. 16. It is important to me that I respect the decisions made by my groups. Items 1-8 (individualism); Items 1-4 (horizontal individualism); Items 5-8 (vertical individualism); Items 9-16 (collectivism); Items 9-12 (horizontal collectivism); Items 13-16 (vertical collectivism).

*=reverse-coded item

39

ESTABLISHING FACTORIAL VALIDITY Because our constructs are reflective, establishing factorial validity using the latest techniques with PLS was very straightforward. We performed confirmatory factor analysis (CFA) on the reflective constructs to establish factorial validity, following the procedures as outlined in (Gefen & Straub, 2005; Straub et al., 2004). We first established convergent validity. According to Gefen and Straub (2005), ―convergent validity is shown when each of the measurement items loads with a significant t-value on its latent construct‖ (p. 93). To do so, we generated a bootstrap with 200 resamples. We then examined the t-values of the outer model loadings; all of the outer loadings were significant at the .05 α level. These results indicate strong convergent validity in our model for the constructs (Table A1).

Table A1. Convergent Validity Results of Reflective Constructs Construct

Items

t-statistic

Social influence

soc1 soc2 soc3 soc4 soc5 rec1 rec2 rec3 trst1 trst2 trst3 trst4 trst5 priv1 priv2 priv3 priv4 priv5 sdam1 sdam2 sdam3 sdam4 sdam5 sdvl1 sdvl2 sdvl3 sdvl4 sdin1 sdin2

53.09*** 49.23*** 22.13*** 24.59*** 22.19*** 33.14*** 47.78*** 39.97*** 79.12*** 35.61*** 51.47*** 35.68*** 53.65*** 41.97*** 22.53*** 23.42*** 30.81*** 24.71*** 24.11*** 23.92*** 13.18*** 53.61*** 37.24*** 38.33*** 2.51** 21.62*** 28.66*** 41.50*** 17.80***

Reciprocity

Trust

Privacy

SD-Amount

SD-Valence

SD-Integrity

SD-Honesty

SD-Depth

Anonymity

H-Ind

V-Ind

H-Coll

V-Coll

sdin3 sdhn1 sdhn2 sdhn3 sdhn4 sdhn5 sddp1 sddp2 sddp3 sddp4 an1 an2 an3 an4 tria1 tria2 tria3 tria4 tria5 tria6 tria7 tria8 tria9 tria10 tria11 tria12 tria13 tria14 tria15 tria16

32.52*** 39.36*** 22.20*** 15.14*** 23.58*** 25.74*** 35.65*** 37.29*** 12.93*** 45.96*** 64.04*** 28.86*** 42.67*** 32.73*** 18.98*** 29.51*** 10.32*** 6.97*** 29.57*** 40.49*** 13.43*** 17.69*** 31.63*** 26.71*** 11.49*** 19.18*** 43.92*** 29.62*** 21.79*** 8.80***

***p < 0.001, ** p < 0.01

To establish discriminant validity of our indicators, we used two common techniques: (1) correlating the latent variable scores and (2) calculating the average variance extracted (AVE). The first approach requires one to generate correlations of the latent variable scores with all the measurement items. These correlations represent a confirmatory factor analysis where the correlations are the actual loadings. Exact guidelines on this have not yet been established, but the fundamental idea is that ―all the loadings of the measurement items on their assigned constructs should be an order of magnitude larger than any other loading‖ (Gefen & Straub, 2005, p. 93). Using latent variable scores, strong discriminant validity was established for all items with the exception of item 4 of vertical collectivism, item 3 of SD-amount, item 3 of SD-depth, and item 2 of SD-valence (Table A2). Because these items showed potential for overlap on other constructs, these items were removed to improve discriminant validity. The second approach that we used to establish discriminant validity was the AVE test. ―Conceptually, the

41

AVE test is equivalent to saying that the correlation of the construct with its measurement items should be larger than its correlation with the other constructs‖ (Gefen & Straub, 2005, p. 94), which is similar to correlation tests with multi-trait, multi-method matrices. The AVE is calculated through PLS-GRAPH by computing the variances shared by the items of a particular construct. See Table A3 (the AVE square roots are represented as the bold and underlined diagonal elements). Off-diagonal elements in the table represent the correlations between the constructs. To establish discriminant validity, the diagonal elements must be greater than the off-diagonal elements for the same row and column. The AVE analysis showed very strong discriminant validity for all subconstructs and thus further confirmed our choices of items to retain and drop. Finally, we established the reliability of the measures. Reliability refers to the degree to which a scale yields consistent and stable measures over time (Straub, 1989). PLS computes a composite reliability score (similar to Cronbach‘s α in that both are measures of internal consistency) as part of its integrated model analysis. Specifically, composite reliability is an index that reflects the impact of error on the measure (Raykov & Grayson, 2003). Each construct in our research model demonstrated high levels of reliability that exceeded the standard thresholds.

42

Table A2. Establishing Discriminant Validity of Reflective Constructs Using Latent Scores

triand1 triand2 triand3 triand4 triand5 triand6 triand7 triand8 triand9 triand10 triand11 triand12 triand13 triand14 triand15 triand16 sdamt1 sdamt2 sdamt3 sdamt4 sdamt5 sddep1 sddep2 sddep3 sddep4 sdhon1 sdhon2 sdhon3 sdhon4 sdhon5 sdint1 sdint2

H-Ind

V-Ind

H-Coll

0.765 0.782 0.633 0.548 0.239 0.182 0.212 0.017 0.102 0.087 0.012 0.164 0.107 0.156 0.081 0.084 -0.044 -0.037 -0.090 -0.020 -0.042 0.044 -0.061 -0.078 -0.037 0.148 0.181 0.136 0.185 0.223 0.222 0.122

0.199 0.134 0.186 0.161 0.743 0.779 0.590 0.670 -0.093 0.085 0.048 0.119 0.109 0.110 0.180 0.020 -0.028 0.137 -0.111 0.137 0.061 0.025 0.085 0.149 0.113 -0.016 -0.003 -0.176 0.016 0.044 -0.026 -0.021

0.037 0.044 0.181 0.181 0.133 0.058 0.141 -0.140 0.744 0.755 0.593 0.775 0.302 0.352 0.306 0.420 0.059 0.047 -0.022 0.070 0.171 0.117 0.071 0.066 0.088 0.212 0.184 0.127 0.233 0.245 0.204 0.150

V-Coll 0.114 0.118 0.151 0.108 0.165 0.151 0.163 -0.043 0.340 0.318 0.364 0.437 0.794 0.762 0.746 0.499 -0.029 0.004 -0.113 -0.002 0.037 0.052 -0.005 0.096 0.103 0.137 0.141 0.080 0.221 0.154 0.139 0.121

Amt

Depth

-0.100 -0.083 -0.049 0.004 0.022 0.060 0.002 0.094 0.078 0.035 0.071 0.048 -0.015 -0.076 -0.024 0.014 0.741 0.737 0.548 0.837 0.787 0.476 0.586 0.366 0.547 0.214 0.138 -0.022 0.187 0.102 0.004 0.149

-0.065 -0.039 -0.024 -0.057 0.044 0.112 -0.009 0.135 0.102 0.063 0.070 0.008 0.034 -0.007 0.069 0.086 0.386 0.602 0.224 0.564 0.626 0.762 0.799 0.586 0.827 0.303 0.187 0.067 0.252 0.169 -0.015 0.138

Honest 0.149 0.180 0.147 0.091 -0.016 0.039 0.015 -0.123 0.193 0.197 0.123 0.200 0.076 0.142 0.126 0.188 0.102 0.139 0.041 0.128 0.163 0.387 0.165 0.012 0.193 0.809 0.800 0.641 0.787 0.816 0.439 0.422

Integ

Val

Trust

0.143 0.131 0.152 0.120 0.020 -0.004 0.038 -0.087 0.195 0.161 0.026 0.134 0.057 0.104 0.085 0.140 0.082 0.029 0.009 0.111 0.103 0.198 0.026 -0.110 0.033 0.390 0.436 0.218 0.451 0.457 0.829 0.692

-0.081 -0.097 0.022 0.048 0.105 0.094 0.116 -0.027 0.125 0.071 0.077 0.201 0.061 0.028 0.113 0.117 0.071 0.034 0.007 0.159 0.165 0.130 0.072 -0.012 0.112 0.052 0.086 -0.004 0.087 0.071 0.127 0.051

0.024 0.057 0.026 0.100 0.021 0.039 0.020 -0.050 0.245 0.209 0.191 0.248 0.106 0.092 0.118 0.205 0.185 0.107 0.045 0.213 0.208 0.265 0.152 0.027 0.136 0.314 0.221 0.169 0.355 0.250 0.236 0.191

Reci -0.048 0.057 0.081 -0.022 0.043 0.006 0.039 0.028 0.151 0.080 0.069 0.090 0.095 0.132 0.141 0.080 0.134 0.226 0.108 0.221 0.315 0.247 0.247 0.179 0.277 0.158 0.147 -0.018 0.169 0.140 0.061 0.098

Privacy 0.093 0.092 0.063 0.030 0.098 0.047 0.064 -0.012 -0.032 0.013 0.095 0.070 0.119 0.154 0.121 0.048 -0.202 -0.174 -0.133 -0.201 -0.263 -0.260 -0.290 -0.103 -0.157 -0.122 -0.145 -0.092 -0.169 -0.160 -0.060 -0.140

Soc-in -0.054 -0.013 0.003 -0.065 0.092 0.109 0.022 0.158 0.044 0.023 0.057 0.012 0.069 0.012 0.075 0.011 0.135 0.265 0.009 0.261 0.260 0.186 0.206 0.278 0.281 0.044 -0.029 -0.126 0.011 -0.013 -0.073 0.037

anon -0.040 -0.042 -0.033 -0.050 -0.027 0.006 0.013 0.027 0.016 -0.029 0.039 -0.041 0.034 0.028 0.071 0.037 0.042 0.117 -0.034 0.079 0.126 0.099 0.112 0.117 0.124 0.007 -0.007 -0.061 0.007 -0.047 -0.029 -0.050

sdint3 sdval1 sdval2 sdval3 sdval4 trust1 trust2 trust3 trust4 trust5 recip1 recip2 recip3 privcy1 privcy2 privcy3 privcy4 privcy5r socinf1 socinf2 socinf3 socinf4 socinf5 anon1 anon2 anon3 anon4

0.177 -0.069 0.008 -0.039 0.068 0.091 0.165 0.095 0.087 0.119 0.073 0.054 -0.024 0.088 0.090 0.155 0.069 0.053 -0.062 -0.039 -0.085 0.056 0.007 -0.016 -0.095 -0.067 0.018

-0.002 0.087 -0.041 0.138 0.090 -0.024 0.045 0.047 0.045 0.018 0.041 0.026 0.037 0.103 0.055 0.062 0.145 0.008 0.129 0.139 0.077 0.141 0.158 0.038 -0.027 0.038 -0.041

0.143 0.165 0.034 0.197 0.166 0.313 0.301 0.302 0.270 0.320 0.107 0.145 0.125 0.073 0.027 0.089 0.120 -0.057 -0.025 0.015 0.073 0.082 0.082 -0.011 -0.023 -0.030 0.049

0.095 0.078 0.026 0.145 0.105 0.160 0.186 0.181 0.111 0.186 0.131 0.116 0.146 0.213 0.216 0.161 0.144 0.011 0.058 0.045 0.014 0.071 0.057 0.058 0.001 0.023 0.072

0.017 0.127 -0.350 0.277 0.064 0.187 0.166 0.202 0.119 0.143 0.117 0.198 0.296 -0.268 -0.157 -0.222 -0.256 -0.359 0.198 0.219 0.169 0.170 0.173 0.045 0.066 0.101 0.086

-0.031 0.087 -0.413 0.255 0.060 0.174 0.160 0.158 0.147 0.157 0.179 0.202 0.383 -0.276 -0.106 -0.201 -0.221 -0.410 0.248 0.243 0.192 0.271 0.233 0.133 0.125 0.160 0.089

0.384 0.036 -0.041 0.134 0.095 0.345 0.392 0.314 0.326 0.320 0.094 0.169 0.178 -0.175 -0.023 -0.055 -0.154 -0.241 -0.082 -0.017 -0.034 0.053 -0.042 -0.021 -0.032 -0.029 0.017

0.825 0.049 0.055 0.131 0.144 0.227 0.288 0.200 0.230 0.204 0.061 0.144 0.099 -0.124 -0.056 -0.010 -0.108 -0.192 -0.066 -0.048 -0.030 -0.024 -0.044 -0.047 -0.098 -0.042 0.053

0.096 0.832 0.253 0.791 0.791 0.141 0.075 0.167 0.076 0.112 0.016 0.114 -0.008 0.070 0.011 0.010 -0.028 -0.066 0.065 0.041 0.029 0.061 0.073 0.091 0.053 0.064 0.128

0.170 0.095 -0.069 0.191 0.129 0.898 0.809 0.836 0.847 0.887 0.023 0.187 0.152 -0.156 -0.066 -0.070 -0.135 -0.269 0.005 0.033 0.124 0.075 0.058 0.050 0.018 0.075 0.031

0.080 0.087 -0.261 0.147 0.026 0.176 0.133 0.165 0.131 0.133 0.799 0.812 0.805 -0.092 -0.059 -0.093 -0.054 -0.288 0.235 0.238 0.122 0.194 0.175 0.138 0.103 0.095 0.114

-0.054 0.014 0.109 -0.054 -0.054 -0.195 -0.166 -0.158 -0.211 -0.155 -0.067 -0.107 -0.262 0.792 0.684 0.751 0.735 0.673 -0.077 -0.148 -0.171 -0.090 -0.044 -0.010 -0.015 -0.062 0.030

-0.047 0.057 -0.210 0.140 0.063 0.070 0.050 0.115 0.065 0.066 0.164 0.214 0.264 -0.144 -0.043 -0.039 -0.017 -0.341 0.819 0.799 0.654 0.651 0.653 0.067 0.098 0.110 0.047

-0.019 0.118 -0.101 0.155 -0.011 0.077 0.051 0.036 0.045 0.078 0.052 0.083 0.156 0.014 0.071 -0.046 -0.027 -0.104 0.081 0.091 0.063 0.023 0.126 0.856 0.782 0.825 0.766

44

Table A3. Discriminant Validity through the Square Root of AVE H-IND H-IND

V-IND

H-COLL

V-COLL

AMT

DEPTH

HONESTY

INTENT

VALENCE

TRUST

RECI

PRIV

V-IND

0.452 (0.672) .236

H-COLL

.137

0.493 (0.702) .044

V-COLL

.148

.168

0.498 (0.706) .419

AMT

-.055

.099*

.109*

0.622 (0.789) -.022

DEPTH

-.028

.091*

.105*

.028

0.617 (0.786) .671

HONESTY

.220

-.037

.244

.159

.163

0.672 (0.820) .307

INTEG

.192

-.023

.205

.125

.097*

.104*

0.564 (0.751) .521

VALENCE

-.020

.125

.206

.091*

.203

.179

.100*

0.623 (0.789) .136

TRUST

.121

.033

.338

.150

.207

.216

.396

.295

0.648 (0.805) .170

RECIPROCITY

.035

.046

.153

.153

.266

.310

.176

.122

.086*

0.770 (0.875) .171

PRIVACY

.114

.097*

.070

.196

-.344

-.351

-.196

-.136

-.028

-.211

0.659 (0.812) -.190

SOCIAL

-.044

.175

.062

.066

.289

.287

-.036

-.056

.123

.113

.283

0.581 (0.762) -.160

ANONYMITY

-.033

.020

-.001

.048

.116

.130

-.035

-.049

.110*

.053

.128

.000

SOCIAL

0.560 (0.748) .119

ANON

0.664 (0.815)

45

TESTING FOR COMMON METHODS BIAS Because the self-reported data was collected using online surveys with similar-in-appearance scales, we tested for common methods bias to establish that it was not a likely factor in our data collection. To do so, we used two approaches. The first approach—which is increasingly in dispute—was to conduct Harman‘s single-factor test (Podsakoff et al., 2003). This test required that we run an exploratory, unrotated factor analysis on all of the firstorder constructs. The aim of the test is to see if a single factor emerges that explains the majority of the variance in the model. If so, then common-method bias likely exists on a significant level. The result of our factor analysis produced 55 distinct factors, the largest of which only accounted for 13.9% of the variance of the model. The second approach, which is more accepted, was simply to examine a correlation matrix of the constructs (see measurement model statistics in Table A4) and to determine if any of the correlations were above 0.90; if so, this qualifies as strong evidence that common methods bias exists (Pavlou et al., 2007). In this analysis, we collapsed the self-disclosure subconstructs to one primary construct, the two individualism subconstructs to individualism, and the two collectivism subconstructs to collectivism, as these were the ultimate units of analysis. In no case did our correlations reach this threshold. Given that our data passed both tests of common method bias, we conclude there is little reason to believe the data exhibit negative effects from common methods bias. These correlations and descriptive statistics are shown in Table A4.

Table A4. Descriptive Statistics (n=528) Construct

(1)

(2)

(3)

(4)

(5)

(6)

µ

SD

Individualism (1)

4.84

0.77

Collectivism (2)

5.35

0.84

Self-disclosure (3)

4.15

0.79

.116

.237

Trust (4)

5.05

1.15

.096*

.296

.411

Reciprocity (5)

4.1

1.33

.054

.184

.312

.171

Privacy risk (6)

4.72

1.18

.137

.152

-.347

-.211

-.190

Social influence (7)

3.32

1.15

.091*

.075

.187

.113

.283

-.160

Anonymity (8)

3.27

1.38

-.003

.026

.082

.053

.128

.000

(7)

.183

.119

ESTABLISHING CROSS-CULTURAL EQUIVALENCE Because we had participants similarly represented from two distinct cultures (France and the U.K.), it was important to establish cross-cultural equivalence and a lack of bias in cross-cultural comparisons; otherwise, we could not be sure that any differences could not truly be attributed to cultural differences rather than to measurement artefacts (Karahanna et al., 2002). To establish equivalence we addressed cross-cultural issues that can introduce (1) construct bias, (2) method bias, and item (3) bias. Construct bias. To prevent construct bias (where a construct means different things in different cultures) we primarily used constructs that had been carefully constructed and validated in multiple studies. To further empirically lack of construct bias, we divided the data by national culture (U.K. and France) and performed separate factorial validity checks on each of the two portions of data. The idea here is that if the factorial validity of each portion is different, then the different cultures may have different understandings of the constructs. Our results showed that virtually the same factors were produced; thus, we can infer that our methods, construct selection, and measurement items reasonably prevented construct bias. Method bias. To prevent method bias (bias in instrument scores because of flaws in the instrument or its administration) we were careful to keep the instructions very simple and clear, to provide for the same conditions in administering the instrument, to focus on respondents who were familiar with social networking, and to establish simple, unambiguous communication with the respondents.

To further maintain cross-cultural equivalence we tried to sample similar types of respondents from each country in terms of basic demographics. We looked at differences in terms of education level, age, internet experience, being a privacy victim, and gender. This is summarized in Table A5. There were no statistical differences between the countries in terms of internet experience, being privacy victims, and gender; however, there were demographic differences in education and age. However, in computing the effect sizes (via Cohen‘s), we determined that there was no meaningful difference in education level and that the difference in age represented a small effect size. Thus, for extra caution, both education level and age were added as covariates to clean up this slight noise in our model (these covariates had virtually no effect on our model).

Table A5. Demographic Differences across Countries Country

Education level

Age

Internet experience

Privacy victim

FR

Mean N Std. Deviation UK Mean N Std. Deviation F-statistic (1,519)

3.40 263 1.26 3.18 266 1.21 5.85

33.62 263 12.56 36.02 266 11.95 4.96

5.87 263 1.80 6.07 266 1.68 0.841

2.27 263 1.59 2.17 266 1.55 2.13

p-value

0.016*

0.026*

0.360 (n/s)

Effect size (Cohen‘s D)

0.07

0.20

n/a

0.145 (n/s) n/a

Meaningful difference?

Negligible Small No No effect effect ***p < 0.001, ** p < 0.01, * p < 0.05

Gender (1=male; 2=female) 1.53 263 5.0 1.56 266 5.0 0.308 0.579 (n/s) n/a No

Item bias. To prevent item bias (measurement artefacts from poor translation, inappropriate terms, and wording complexity) we used as simple and clear of wording as possible. We had two different sets of professional translators translate and back translate all of the items to ensure proper translation into French. Unfortunately, statistical procedures to establish lack of item bias are inconsistent, controversial, and no agreed-upon method has been established (Karahanna et al., 2002). Thus, because item bias also directly influences construct bias and method bias (Karahanna et al., 2002), we infer that our lack of construct bias and method bias (as corrected for by the added covariates) implies a lack of item bias.

47

Table A6.Exploring Differences in Country Samples (Using MANOVA) country

H-IND

V-IND

HCOLL

VCOLL

Ind.

Coll.

Amount

Depth

Honesty

Integrity

Val

SD

FR

Mean SD

5.68 0.90

4.09 1.12

5.14 0.96

5.55 1.14

4.88 0.82

5.31 0.89

2.90 1.29

3.08 1.32

5.07 1.17

5.29 1.11

3.76 1.13

4.07 0.75

UK

Mean

5.46

4.13

5.27

5.55

4.80

5.39

3.24

3.20

4.88

5.06

4.61

4.22

SD

0.90

0.99

0.88

1.03

0.71

0.78

1.25

1.42

1.17

1.15

1.04

0.82

F-statistic p-value

7.59 0.006**

0.10 0.756 (n/s) n/s

1.21 0.272 (n/s) n/s

1.87 0.172 (n/s) n/s

0.98 0.323 (n/s) n/s

3.72 0.054+

5.91 0.015*

81.61 0.000***

4.30 0.039*

FR > UK

3.04 0.082 (n/s) n/s

9.59 0.002**

result

0.34 0.562 (n/s) n/s

FR > UK

FR > UK

UK > FR

UK > FR

country

TRUST

Reciproc

PRIVACY

SOCIAL

ANON

FR

Mean SD

5.07 1.05

4.01 1.32

4.66 1.18

3.40 1.13

3.16 1.39

UK

Mean

5.03

4.19

4.78

3.25

3.38

SD F-statistic p-value result

1.24

1.33

1.18

1.17

1.37

0.32 0.572 (n/s) n/s

3.64 0.057+

1.34 0.247 (n/s) n/s

1.86 0.173 (n/s) n/s

3.65 0.057+

UK > FR

UK > FR

UK > FR

To statistically compare the UK-only and France-only path models, we used the established approach to multigroup comparisons in PLS as demonstrated by Moores and Chang (Moores & Chang, 2006). Accordingly, a pooled standard error term t-test was used to determine the statistical significance of all the path coefficients (βs) by UK and France samples. We used the Smith-Satterthwait test because our data were not normally distributed, and variances of the comparison groups were not equal. These results are summarized in Table A7, which indicates no meaningful difference existed in the path coefficients across the models. This finding allows us to propose one model for both cultures.

Table A7. Multi-group PLS Comparisons of Path Coefficients of the Two Countries Path

France path

France SE

UK path

H1. Social influence to use an online community ' self-disclosure H2. Reciprocity ' self-disclosure H3. Online community trust ' self-disclosure H4. Privacy risk beliefs ' (-) self-disclosure H5. Anonymity ' self-disclosure H6. Collectivism ' self-disclosure H7. Individualism ' (-) self-disclosure

0.232

0.082

0.128

0.187 0.256 -0.268 0.044 0.052 -0.090

0.070 0.133 0.139 0.066 0.118 0.143

0.158 0.182 -0.289 0.081 0.187 0.166

UK SE

df

Effect size

0.088

tSig.? statistic 0.000 0.867 no

0.084 0.092 0.083 0.066 0.078 0.095

0.020 0.026 0.026 0.009 0.020 0.030

negligible negligible negligible negligible negligible negligible

0.265 0.458 0.130 -0.397 -0.957 -1.488

no no no no no no

negligible

ONLINE APPENDIX 1 Clay Posey, Paul Benjamin Lowry, Tom L. Roberts, and Selwyn Ellis (2010). “The CultureInfluenced Online Community Self-Disclosure Model: The Case of Working Professionals in France and the UK Who Use Online Communities,” European Journal of Information Systems (EJIS) , vol. 19(2), pp. 181-195 (doi:10.1057/ejis.2010.15; published online 9 March 2010) This appendix is intended for online distribution to supplement the published article; it is not intended to be published in the actual article.

MEASUREMENT SCALES All scales were measured as 7-point Likert-like scales anchored on “strongly disagree, moderately disagree, slightly disagree, neither disagree nor agree, slightly agree, moderately agree, strongly agree.” Questions were framed in the context of online communities. Construct Subconstruct Items Trust in Community N/A trust1. Overall, the people in my online social network are very trustworthy Adapted to online trust2. Members of my online social network are usually considerate of context from Jarvenpaa one another’s feelings and Leidner (1999) trust3. The people in my online social network are friendly trust4. I can rely on those with whom I disclose personal information to in my online social network Privacy Risk Beliefs N/A prb1. In general, it is risky to give my private information to others Adapted to online online context from Malhotra, prb2. There is a high potential for loss associated with giving my Kim, and Agarwal personal information to others online (2004) prb3. There is too much uncertainty associated with giving my personal information to others online prb4. Providing others my private information online involves many unexpected problems *prb5. I feel safe giving my private information to others online Social Influence to use N/A si1. People who influence my behavior think that I should use an online an online community community si2. People who are important to me think that I should use an online Adapted to online community context from Venkatesh si3. I use an online community because of the proportion of my friends et al. (2003) that use it si4. People I am associated with who use an online community gain important benefits for using it si5. Having a personal profile on an online community is considered a status symbol Reciprocity N/A rec1. When others disclose personal information online, I believe that they expect me to do the same Based on Kankanhalli, rec2. Other online users trust me to return the favor of sharing personal Tan, and Wei (2005) information (Wasko & Faraj, 2005) rec3. I know that other users online disclose information about themselves, so it is only fair to do the same Self-disclosure Amount *amt1. I do not often talk about myself online amt2. I usually talk about myself for fairly long periods at a time Adapted to online *amt3. My conversation online lasts the least time when I am discussing context from (Wheeless, myself 1978; Wheeless & amt4. I often talk about myself online Grotz, 1976) amt5. I often discuss my feelings about myself online Depth dep1. I intimately disclose who I really am, openly and fully in my conversation online dep2. I often disclose intimate, personal things about myself without hesitation online dep3. I feel that I sometimes do not control my self-disclosure of personal or intimate things I tell about myself online dep4. Once I get started, I intimately and fully reveal myself in my selfdisclosures online Honesty hon1. I always feel completely sincere when I reveal my own feelings and experiences online

2

Intent

Valence

Anonymity (Pinsonneault & Heppel, 1998)

n/a

Vertical and horizontal collectivism / individualism (Triandis & Gelfand, 1998)

n/a

hon2. My online self-disclosures are completely accurate reflections of who I really am * hon3. I am not always honest in my self-disclosures online Hon4. My statements online about my own feelings, emotions, and experiences are always accurate self-perceptions Hon5. I am always honest in my self-disclosures online int2. When I express my personal feelings online, I am always aware what I am doing and saying int3. When I reveal my feelings about myself online, I consciously intend to do so int4. When I am self-disclosing online, I am consciously aware of what I am revealing val1. I usually disclose positive things about myself online *val2. I normally reveal “bad” feelings I have about myself online val3. I normally express my “good” feelings about myself online val4. On the whole, my disclosures about myself online are more positive than negative Based on Pinsonneault’s conceptualization of identification Id1. My identity can be distinguished from other users online Id2. Other users online know my real identity Id3. I believe none of the other users online know who I really am Id4. My identity is hidden from other users online 1. I’d rather depend on myself than others. 2. I rely on myself most of the time; I rarely rely on others. 3. I often do “my own thing.” 4. My personal identity, independent of others, is very important to me. 5. It is important that I do my job better than others. 6. Winning is everything. 7. Competition is the law of nature. 8. When another person does better than I do, I get tense and aroused. 9. If a coworker gets a prize, I would feel proud. 10. The well-being of my coworkers is important to me. 11. To me, pleasure is spending time with others. 12. I feel good when I cooperate with others. 13. Parents and children must stay together as much as possible. 14. It is my duty to take care of my family, even when I have to sacrifice what I want. 15. Family members should stick together, no matter what sacrifices are required. 16. It is important to me that I respect the decisions made by my groups. Items 1-8 (individualism); Items 1-4 (horizontal individualism); Items 5-8 (vertical individualism); Items 9-16 (collectivism); Items 9-12 (horizontal collectivism); Items 13-16 (vertical collectivism).

*=reverse-coded item

3

ESTABLISHING FACTORIAL VALIDITY Because our constructs are reflective, establishing factorial validity using the latest techniques with PLS was very straightforward. We performed confirmatory factor analysis (CFA) on the reflective constructs to establish factorial validity, following the procedures as outlined in (Gefen & Straub, 2005; Straub et al., 2004). We first established convergent validity. According to Gefen and Straub (2005), “convergent validity is shown when each of the measurement items loads with a significant t-value on its latent construct” (p. 93). To do so, we generated a bootstrap with 200 resamples. We then examined the t-values of the outer model loadings; all of the outer loadings were significant at the .05 α level. These results indicate strong convergent validity in our model for the constructs (Table A1).

Table A1. Convergent Validity Results of Reflective Constructs Construct

Items

Social influence  soc1  soc2  soc3  soc4  soc5  Reciprocity  rec1  rec2  rec3  Trust  trst1  trst2  trst3  trst4  trst5  Privacy  priv1  priv2  priv3  priv4  priv5  SD‐Amount  sdam1  sdam2  sdam3  sdam4  sdam5  SD‐Valence  sdvl1  sdvl2  sdvl3  sdvl4  SD‐Integrity  sdin1  sdin2 

t-statistic

53.09***  49.23***  22.13***  24.59***  22.19***  33.14***  47.78***  39.97***  79.12***  35.61***  51.47***  35.68***  53.65***  41.97***  22.53***  23.42***  30.81***  24.71***  24.11***  23.92***  13.18***  53.61***  37.24***  38.33***  2.51**  21.62***  28.66***  41.50***  17.80*** 

SD‐Honesty 

SD‐Depth 

Anonymity 

H‐Ind 

V‐Ind 

H‐Coll 

V‐Coll 

sdin3  sdhn1  sdhn2  sdhn3  sdhn4  sdhn5  sddp1  sddp2  sddp3  sddp4  an1  an2  an3  an4  tria1  tria2  tria3  tria4  tria5  tria6  tria7  tria8  tria9  tria10  tria11  tria12  tria13  tria14  tria15  tria16 

32.52***  39.36***  22.20***  15.14***  23.58***  25.74***  35.65***  37.29***  12.93***  45.96***  64.04***  28.86***  42.67***  32.73***  18.98***  29.51***  10.32***  6.97***  29.57***  40.49***  13.43***  17.69***  31.63***  26.71***  11.49***  19.18***  43.92***  29.62***  21.79***  8.80*** 

***p < 0.001, ** p < 0.01

To establish discriminant validity of our indicators, we used two common techniques: (1) correlating the latent variable scores and (2) calculating the average variance extracted (AVE). The first approach requires one to generate correlations of the latent variable scores with all the measurement items. These correlations represent a confirmatory factor analysis where the correlations are the actual loadings. Exact guidelines on this have not yet been established, but the fundamental idea is that “all the loadings of the measurement items on their assigned constructs should be an order of magnitude larger than any other loading” (Gefen & Straub, 2005, p. 93). Using latent variable scores, strong discriminant validity was established for all items with the exception of item 4 of vertical collectivism, item 3 of SD-amount, item 3 of SD-depth, and item 2 of SD-valence (Table A2). Because these items showed potential for overlap on other constructs, these items were removed to improve discriminant validity. The second approach that we used to establish discriminant validity was the AVE test. “Conceptually, the

5

AVE test is equivalent to saying that the correlation of the construct with its measurement items should be larger than its correlation with the other constructs” (Gefen & Straub, 2005, p. 94), which is similar to correlation tests with multi-trait, multi-method matrices. The AVE is calculated through PLS-GRAPH by computing the variances shared by the items of a particular construct. See Table A3 (the AVE square roots are represented as the bold and underlined diagonal elements). Off-diagonal elements in the table represent the correlations between the constructs. To establish discriminant validity, the diagonal elements must be greater than the off-diagonal elements for the same row and column. The AVE analysis showed very strong discriminant validity for all subconstructs and thus further confirmed our choices of items to retain and drop. Finally, we established the reliability of the measures. Reliability refers to the degree to which a scale yields consistent and stable measures over time (Straub, 1989). PLS computes a composite reliability score (similar to Cronbach’s α in that both are measures of internal consistency) as part of its integrated model analysis. Specifically, composite reliability is an index that reflects the impact of error on the measure (Raykov & Grayson, 2003). Each construct in our research model demonstrated high levels of reliability that exceeded the standard thresholds.

6

Table A2. Establishing Discriminant Validity of Reflective Constructs Using Latent Scores

triand1 triand2 triand3 triand4 triand5 triand6 triand7 triand8 triand9 triand10 triand11 triand12 triand13 triand14 triand15 triand16 sdamt1 sdamt2 sdamt3 sdamt4 sdamt5 sddep1 sddep2 sddep3 sddep4 sdhon1 sdhon2 sdhon3 sdhon4 sdhon5 sdint1 sdint2

H-Ind

V-Ind

H-Coll

0.765 0.782 0.633 0.548 0.239 0.182 0.212 0.017 0.102 0.087 0.012 0.164 0.107 0.156 0.081 0.084 -0.044 -0.037 -0.090 -0.020 -0.042 0.044 -0.061 -0.078 -0.037 0.148 0.181 0.136 0.185 0.223 0.222 0.122

0.199 0.134 0.186 0.161 0.743 0.779 0.590 0.670 -0.093 0.085 0.048 0.119 0.109 0.110 0.180 0.020 -0.028 0.137 -0.111 0.137 0.061 0.025 0.085 0.149 0.113 -0.016 -0.003 -0.176 0.016 0.044 -0.026 -0.021

0.037 0.044 0.181 0.181 0.133 0.058 0.141 -0.140 0.744 0.755 0.593 0.775 0.302 0.352 0.306 0.420 0.059 0.047 -0.022 0.070 0.171 0.117 0.071 0.066 0.088 0.212 0.184 0.127 0.233 0.245 0.204 0.150

V-Coll 0.114 0.118 0.151 0.108 0.165 0.151 0.163 -0.043 0.340 0.318 0.364 0.437 0.794 0.762 0.746 0.499 -0.029 0.004 -0.113 -0.002 0.037 0.052 -0.005 0.096 0.103 0.137 0.141 0.080 0.221 0.154 0.139 0.121

Amt

Depth

-0.100 -0.083 -0.049 0.004 0.022 0.060 0.002 0.094 0.078 0.035 0.071 0.048 -0.015 -0.076 -0.024 0.014 0.741 0.737 0.548 0.837 0.787 0.476 0.586 0.366 0.547 0.214 0.138 -0.022 0.187 0.102 0.004 0.149

-0.065 -0.039 -0.024 -0.057 0.044 0.112 -0.009 0.135 0.102 0.063 0.070 0.008 0.034 -0.007 0.069 0.086 0.386 0.602 0.224 0.564 0.626 0.762 0.799 0.586 0.827 0.303 0.187 0.067 0.252 0.169 -0.015 0.138

Honest 0.149 0.180 0.147 0.091 -0.016 0.039 0.015 -0.123 0.193 0.197 0.123 0.200 0.076 0.142 0.126 0.188 0.102 0.139 0.041 0.128 0.163 0.387 0.165 0.012 0.193 0.809 0.800 0.641 0.787 0.816 0.439 0.422

Integ

Val

Trust

0.143 0.131 0.152 0.120 0.020 -0.004 0.038 -0.087 0.195 0.161 0.026 0.134 0.057 0.104 0.085 0.140 0.082 0.029 0.009 0.111 0.103 0.198 0.026 -0.110 0.033 0.390 0.436 0.218 0.451 0.457 0.829 0.692

-0.081 -0.097 0.022 0.048 0.105 0.094 0.116 -0.027 0.125 0.071 0.077 0.201 0.061 0.028 0.113 0.117 0.071 0.034 0.007 0.159 0.165 0.130 0.072 -0.012 0.112 0.052 0.086 -0.004 0.087 0.071 0.127 0.051

0.024 0.057 0.026 0.100 0.021 0.039 0.020 -0.050 0.245 0.209 0.191 0.248 0.106 0.092 0.118 0.205 0.185 0.107 0.045 0.213 0.208 0.265 0.152 0.027 0.136 0.314 0.221 0.169 0.355 0.250 0.236 0.191

Reci -0.048 0.057 0.081 -0.022 0.043 0.006 0.039 0.028 0.151 0.080 0.069 0.090 0.095 0.132 0.141 0.080 0.134 0.226 0.108 0.221 0.315 0.247 0.247 0.179 0.277 0.158 0.147 -0.018 0.169 0.140 0.061 0.098

Privacy 0.093 0.092 0.063 0.030 0.098 0.047 0.064 -0.012 -0.032 0.013 0.095 0.070 0.119 0.154 0.121 0.048 -0.202 -0.174 -0.133 -0.201 -0.263 -0.260 -0.290 -0.103 -0.157 -0.122 -0.145 -0.092 -0.169 -0.160 -0.060 -0.140

Soc-in -0.054 -0.013 0.003 -0.065 0.092 0.109 0.022 0.158 0.044 0.023 0.057 0.012 0.069 0.012 0.075 0.011 0.135 0.265 0.009 0.261 0.260 0.186 0.206 0.278 0.281 0.044 -0.029 -0.126 0.011 -0.013 -0.073 0.037

anon -0.040 -0.042 -0.033 -0.050 -0.027 0.006 0.013 0.027 0.016 -0.029 0.039 -0.041 0.034 0.028 0.071 0.037 0.042 0.117 -0.034 0.079 0.126 0.099 0.112 0.117 0.124 0.007 -0.007 -0.061 0.007 -0.047 -0.029 -0.050

sdint3 sdval1 sdval2 sdval3 sdval4 trust1 trust2 trust3 trust4 trust5 recip1 recip2 recip3 privcy1 privcy2 privcy3 privcy4 privcy5r socinf1 socinf2 socinf3 socinf4 socinf5 anon1 anon2 anon3 anon4

0.177 -0.069 0.008 -0.039 0.068 0.091 0.165 0.095 0.087 0.119 0.073 0.054 -0.024 0.088 0.090 0.155 0.069 0.053 -0.062 -0.039 -0.085 0.056 0.007 -0.016 -0.095 -0.067 0.018

-0.002 0.087 -0.041 0.138 0.090 -0.024 0.045 0.047 0.045 0.018 0.041 0.026 0.037 0.103 0.055 0.062 0.145 0.008 0.129 0.139 0.077 0.141 0.158 0.038 -0.027 0.038 -0.041

0.143 0.165 0.034 0.197 0.166 0.313 0.301 0.302 0.270 0.320 0.107 0.145 0.125 0.073 0.027 0.089 0.120 -0.057 -0.025 0.015 0.073 0.082 0.082 -0.011 -0.023 -0.030 0.049

0.095 0.078 0.026 0.145 0.105 0.160 0.186 0.181 0.111 0.186 0.131 0.116 0.146 0.213 0.216 0.161 0.144 0.011 0.058 0.045 0.014 0.071 0.057 0.058 0.001 0.023 0.072

0.017 0.127 -0.350 0.277 0.064 0.187 0.166 0.202 0.119 0.143 0.117 0.198 0.296 -0.268 -0.157 -0.222 -0.256 -0.359 0.198 0.219 0.169 0.170 0.173 0.045 0.066 0.101 0.086

-0.031 0.087 -0.413 0.255 0.060 0.174 0.160 0.158 0.147 0.157 0.179 0.202 0.383 -0.276 -0.106 -0.201 -0.221 -0.410 0.248 0.243 0.192 0.271 0.233 0.133 0.125 0.160 0.089

0.384 0.036 -0.041 0.134 0.095 0.345 0.392 0.314 0.326 0.320 0.094 0.169 0.178 -0.175 -0.023 -0.055 -0.154 -0.241 -0.082 -0.017 -0.034 0.053 -0.042 -0.021 -0.032 -0.029 0.017

0.825 0.049 0.055 0.131 0.144 0.227 0.288 0.200 0.230 0.204 0.061 0.144 0.099 -0.124 -0.056 -0.010 -0.108 -0.192 -0.066 -0.048 -0.030 -0.024 -0.044 -0.047 -0.098 -0.042 0.053

0.096 0.832 0.253 0.791 0.791 0.141 0.075 0.167 0.076 0.112 0.016 0.114 -0.008 0.070 0.011 0.010 -0.028 -0.066 0.065 0.041 0.029 0.061 0.073 0.091 0.053 0.064 0.128

0.170 0.095 -0.069 0.191 0.129 0.898 0.809 0.836 0.847 0.887 0.023 0.187 0.152 -0.156 -0.066 -0.070 -0.135 -0.269 0.005 0.033 0.124 0.075 0.058 0.050 0.018 0.075 0.031

0.080 0.087 -0.261 0.147 0.026 0.176 0.133 0.165 0.131 0.133 0.799 0.812 0.805 -0.092 -0.059 -0.093 -0.054 -0.288 0.235 0.238 0.122 0.194 0.175 0.138 0.103 0.095 0.114

-0.054 0.014 0.109 -0.054 -0.054 -0.195 -0.166 -0.158 -0.211 -0.155 -0.067 -0.107 -0.262 0.792 0.684 0.751 0.735 0.673 -0.077 -0.148 -0.171 -0.090 -0.044 -0.010 -0.015 -0.062 0.030

-0.047 0.057 -0.210 0.140 0.063 0.070 0.050 0.115 0.065 0.066 0.164 0.214 0.264 -0.144 -0.043 -0.039 -0.017 -0.341 0.819 0.799 0.654 0.651 0.653 0.067 0.098 0.110 0.047

-0.019 0.118 -0.101 0.155 -0.011 0.077 0.051 0.036 0.045 0.078 0.052 0.083 0.156 0.014 0.071 -0.046 -0.027 -0.104 0.081 0.091 0.063 0.023 0.126 0.856 0.782 0.825 0.766

8

Table A3. Discriminant Validity through the Square Root of AVE H-IND H-IND

V-IND

H-COLL

V-COLL

AMT

DEPTH

V-IND

0.452 (0.672) .236

H-COLL

.137

0.493 (0.702) .044

V-COLL

.148

.168

0.498 (0.706) .419

AMT

-.055

.099*

.109*

0.622 (0.789) -.022

DEPTH

-.028

.091*

.105*

.028

0.617 (0.786) .671

HONESTY

.220

-.037

.244

.159

.163

0.672 (0.820) .307

INTEG

.192

-.023

.205

.125

.097*

.104*

VALENCE

-.020

.125

.206

.091

TRUST

.121

.033

.338

RECIPROCITY

.035

.046

PRIVACY

.114

SOCIAL ANONYMITY

*

HONESTY

0.564 (0.751) .521 *

INTENT

0.623 (0.789) .136

VALENCE

TRUST

RECI

PRIV

.203

.179

.100

.150

.207

.216

.396

.295

0.648 (0.805) .170

.153

.153

.266

.310

.176

.122

.086*

0.770 (0.875) .171

.097*

.070

.196

-.344

-.351

-.196

-.136

-.028

-.211

0.659 (0.812) -.190

-.044

.175

.062

.066

.289

.287

-.036

-.056

.123

.113

.283

0.581 (0.762) -.160

-.033

.020

-.001

.048

.116

.130

-.035

-.049

.110*

.053

.128

.000

SOCIAL

0.560 (0.748) .119

ANON

0.664 (0.815)

9

TESTING FOR COMMON METHODS BIAS Because the self-reported data was collected using online surveys with similar-in-appearance scales, we tested for common methods bias to establish that it was not a likely factor in our data collection. To do so, we used two approaches. The first approach—which is increasingly in dispute—was to conduct Harman’s single-factor test (Podsakoff et al., 2003). This test required that we run an exploratory, unrotated factor analysis on all of the firstorder constructs. The aim of the test is to see if a single factor emerges that explains the majority of the variance in the model. If so, then common-method bias likely exists on a significant level. The result of our factor analysis produced 55 distinct factors, the largest of which only accounted for 13.9% of the variance of the model. The second approach, which is more accepted, was simply to examine a correlation matrix of the constructs (see measurement model statistics in Table A4) and to determine if any of the correlations were above 0.90; if so, this qualifies as strong evidence that common methods bias exists (Pavlou et al., 2007). In this analysis, we collapsed the self-disclosure subconstructs to one primary construct, the two individualism subconstructs to individualism, and the two collectivism subconstructs to collectivism, as these were the ultimate units of analysis. In no case did our correlations reach this threshold. Given that our data passed both tests of common method bias, we conclude there is little reason to believe the data exhibit negative effects from common methods bias. These correlations and descriptive statistics are shown in Table A4.

Table A4. Descriptive Statistics (n=528) Construct

(1)

(2)

(3)

(4)

(5)

(6)

µ

SD

Individualism (1)

4.84

0.77

Collectivism (2)

5.35

0.84

Self-disclosure (3)

4.15

0.79

.116

.237

Trust (4)

5.05

1.15

.096*

.296

.411

Reciprocity (5)

4.1

1.33

.054

.184

.312

.171

Privacy risk (6)

4.72

1.18

.137

.152

-.347

-.211

-.190

Social influence (7)

3.32

1.15

.091*

.075

.187

.113

.283

-.160

Anonymity (8)

3.27

1.38

-.003

.026

.082

.053

.128

.000

(7)

.183

.119

ESTABLISHING CROSS-CULTURAL EQUIVALENCE Because we had participants similarly represented from two distinct cultures (France and the U.K.), it was important to establish cross-cultural equivalence and a lack of bias in cross-cultural comparisons; otherwise, we could not be sure that any differences could not truly be attributed to cultural differences rather than to measurement artefacts (Karahanna et al., 2002). To establish equivalence we addressed cross-cultural issues that can introduce (1) construct bias, (2) method bias, and item (3) bias. Construct bias. To prevent construct bias (where a construct means different things in different cultures) we primarily used constructs that had been carefully constructed and validated in multiple studies. To further empirically lack of construct bias, we divided the data by national culture (U.K. and France) and performed separate factorial validity checks on each of the two portions of data. The idea here is that if the factorial validity of each portion is different, then the different cultures may have different understandings of the constructs. Our results showed that virtually the same factors were produced; thus, we can infer that our methods, construct selection, and measurement items reasonably prevented construct bias. Method bias. To prevent method bias (bias in instrument scores because of flaws in the instrument or its administration) we were careful to keep the instructions very simple and clear, to provide for the same conditions in administering the instrument, to focus on respondents who were familiar with social networking, and to establish simple, unambiguous communication with the respondents.

To further maintain cross-cultural equivalence we tried to sample similar types of respondents from each country in terms of basic demographics. We looked at differences in terms of education level, age, internet experience, being a privacy victim, and gender. This is summarized in Table A5. There were no statistical differences between the countries in terms of internet experience, being privacy victims, and gender; however, there were demographic differences in education and age. However, in computing the effect sizes (via Cohen’s), we determined that there was no meaningful difference in education level and that the difference in age represented a small effect size. Thus, for extra caution, both education level and age were added as covariates to clean up this slight noise in our model (these covariates had virtually no effect on our model).

Table A5. Demographic Differences across Countries Country

Education level

Age

Internet experience

Privacy victim

3.40 263 1.26 3.18 266 1.21

33.62 263 12.56 36.02 266 11.95

5.87 263 1.80 6.07 266 1.68

2.27 263 1.59 2.17 266 1.55

Gender (1=male; 2=female) 1.53 263 5.0 1.56 266 5.0

F-statistic (1,519)

5.85

4.96

0.841

2.13

0.308

p-value

0.016*

0.026*

0.360 (n/s)

0.579 (n/s)

Effect size (Cohen’s D)

0.07

0.20

n/a

0.145 (n/s) n/a

Meaningful difference?

Negligible Small No No effect effect ***p < 0.001, ** p < 0.01, * p < 0.05

FR

UK

Mean N Std. Deviation Mean N Std. Deviation

n/a No

Item bias. To prevent item bias (measurement artefacts from poor translation, inappropriate terms, and wording complexity) we used as simple and clear of wording as possible. We had two different sets of professional translators translate and back translate all of the items to ensure proper translation into French. Unfortunately, statistical procedures to establish lack of item bias are inconsistent, controversial, and no agreed-upon method has been established (Karahanna et al., 2002). Thus, because item bias also directly influences construct bias and method bias (Karahanna et al., 2002), we infer that our lack of construct bias and method bias (as corrected for by the added covariates) implies a lack of item bias.

11

Table A6.Exploring Differences in Country Samples (Using MANOVA) country

FR UK

H-IND

UK

HCOLL

VCOLL

Ind.

Coll.

Amount

Depth

Honesty

Integrity

Val

SD

Mean

5.68

4.09

5.14

5.55

4.88

5.31

2.90

3.08

5.07

5.29

3.76

4.07

SD

0.90

1.12

0.96

1.14

0.82

0.89

1.29

1.32

1.17

1.11

1.13

0.75

Mean

5.46

4.13

5.27

5.55

4.80

5.39

3.24

3.20

4.88

5.06

4.61

4.22

SD

0.90

0.99

0.88

1.03

0.71

0.78

1.25

1.42

1.17

1.15

1.04

0.82

F-statistic p-value

7.59 0.006**

0.10 0.756 (n/s) n/s

1.21 0.272 (n/s) n/s

1.87 0.172 (n/s) n/s

0.98 0.323 (n/s) n/s

3.72 0.054+

5.91 0.015*

81.61 0.000***

4.30 0.039*

FR > UK

3.04 0.082 (n/s) n/s

9.59 0.002**

result

0.34 0.562 (n/s) n/s

FR > UK

FR > UK

UK > FR

UK > FR

country

FR

V-IND

TRUST

Reciproc

PRIVACY

SOCIAL

ANON

Mean

5.07

4.01

4.66

3.40

3.16

SD

1.05

1.32

1.18

1.13

1.39

Mean

5.03

4.19

4.78

3.25

3.38

SD

1.24

1.33

1.18

1.17

1.37

0.32 0.572 (n/s) n/s

3.64 0.057+

1.34 0.247 (n/s) n/s

1.86 0.173 (n/s) n/s

3.65 0.057+

F-statistic p-value result

UK > FR

UK > FR

UK > FR

To statistically compare the UK-only and France-only path models, we used the established approach to multigroup comparisons in PLS as demonstrated by Moores and Chang (Moores & Chang, 2006). Accordingly, a pooled standard error term t-test was used to determine the statistical significance of all the path coefficients (βs) by UK and France samples. We used the Smith-Satterthwait test because our data were not normally distributed, and variances of the comparison groups were not equal. These results are summarized in Table A7, which indicates no meaningful difference existed in the path coefficients across the models. This finding allows us to propose one model for both cultures.

Table A7. Multi-group PLS Comparisons of Path Coefficients of the Two Countries Path 

France path

France SE

UK path

H1. Social influence to use an online  community ' self‐disclosure  H2. Reciprocity ' self‐disclosure  H3. Online community trust ' self‐disclosure H4. Privacy risk beliefs ' (‐) self‐disclosure H5. Anonymity ' self‐disclosure  H6. Collectivism ' self‐disclosure  H7. Individualism ' (‐) self‐disclosure 

0.232

0.082

0.128

0.187 0.256 ‐0.268 0.044 0.052 ‐0.090

0.070 0.133 0.139 0.066 0.118 0.143

0.158 0.182 ‐0.289 0.081 0.187 0.166

UK SE

df

Effect size

0.088

t‐ Sig.? statistic 0.000 0.867 no

0.084 0.092 0.083 0.066 0.078 0.095

0.020 0.026 0.026 0.009 0.020 0.030

negligible negligible negligible negligible negligible negligible

0.265 0.458 0.130 ‐0.397 ‐0.957 ‐1.488

no no no no no no

negligible