Drivers of in‐group and out‐of‐group electronic word‐of‐mouth (eWOM)

Purpose – The purpose of this study is to address a recent call for additional research on electronic word‐of‐mouth (eWOM). In response to this call, this study draws on the social network paradigm and the uses and gratification theory (UGT) to propose and empirically test a conceptual framework of key drivers of two types of eWOM, namely in‐group and out‐of‐group.Design/methodology/approach – The proposed model, which examines the impact of usage motivations on eWOM in‐group and eWOM out‐of‐group, is tested in a sample of 302 internet users in Portugal.Findings – Results from the survey show that the different drivers (i.e. mood‐enhancement, escapism, experiential learning and social interaction) vary in terms of their impact on the two different types of eWOM. Surprisingly, while results show a positive relationship between experiential learning and eWOM out‐of‐group, no relationship is found between experiential learning and eWOM in‐group.Research limitations/implications – This is the first study inve...


Introduction
[While] word of mouth has always been the most effective form of communication, [nowadays] there is a lost generation of marketeers. . . who do not understand the web and social networks (Simon Clift, Unilever Head of Marketing, Financial Times, April 6, 2010).
Social networks are a defining feature of today's electronic landscape (Bruyn and Lilien, 2008). Within these social networks, it is common for individuals to provide and receive information and informal advice on products and services. This is usually referred to as electronic word-of-mouth (eWOM), which is conceptualised as "any positive or negative statement made by . . . [an individual] . . . which is made available to a multitude of people and institutions via Internet" (Hennig-Thurau et al., 2004, p. 39).
In contrast, word-of-mouth (WOM), the precursor to eWOM, may be defined as person-to-person, oral communication between a receiver and a sender (Lee and Youn, 2009). In this communication, the source is perceived as a non-commercial message that relates to a brand, product or service (Alon and Brunel, 2006;Arndt, 1967). WOM has been recognised as a key force in the marketplace as it influences overall consumers' attitudes, beliefs and behaviour patterns (Bansal and Voyer, 2000;; see Sweeney et al., 2011;Mazzarol et al., 2007), and specifically consumers' product judgements (Bone, 1995;Summers, 1972) and purchase decisions (Lampert and Rosenberg, 1975;Lau and Ng, 2001).
While most traditional WOM occurs among individuals who know and trust each other (Gupta and Harris, 2010), the internet facilitates not only communication with family, friends, and co-workers but also unknown people (Kavanaugh et al., 2005). Indeed, most eWOM occurs with individuals who are strangers (Gupta and Harris, 2010). Given the dissimilar tie strengths among individuals, two different types of eWOM develop, namely eWOM in-group (eWOM with close friends or family), and eWOM out-of-group (eWOM with individuals beyond a person's social, familial and collegial circles) (see Brown and Reingen, 1987;Matsumoto, 2000). This study aims to investigate these two types of eWOM.
Given the "ease of eWOM generation and dissemination" (Gupta andHarris, 2010, p. 1042) and its impact on consumer buying behaviour , researchers have been calling for more research into eWOM for a number of years (Gupta and Harris, 2010;Valck, 2006;Zhang et al., 2010). Thus far, scholars have examined a wide range of eWOM issues, including the value of eWOM to organisations (e.g. Liu, 2006), its links with purchase decisions and purchase intentions (e.g. , its ability to persuade consumers (e.g. Zhang et al., 2010), its antecedents (e.g. Jayawardhena and Wright, 2009;Gruen et al., 2006;Mazzarol et al., 2007;Sweeney et al., 2008), and its consequences (e.g. Park and Lee, 2008;Huang et al., 2011;Wangenheim and Bayón, 2004). Despite the considerable volume of studies on eWOM, it is important to acknowledge that eWOM still remains a very under-researched area (Zhang et al., 2010). Specifically, what drives individuals to engage in different types of eWOM characterised by diverse tie strengths remains underexplored.
Accordingly, this study's objective is to address this gap in the eWOM literature by investigating the impact of usage motivations on eWOM In-goup and out-of-group. This distinction is important because information circulated through weak ties is more novel than information that flows through strong ties (Granovetter, 2005; see Weenig EJM 47,7 and Midden, 1991), and, therefore, the impact of usage motivations on eWOM might differ for in-group and out-of-group. Although some studies distinguish between in-group and out-of-group for the traditional WOM (see Brown and Reingen, 1987;Granovetter, 1973), to the best of our knowledge, no study has investigated the determinants of these two types of eWOM.
From a managerial perspective, understanding the drivers of eWOM in-group and out-of-group can help the company as a whole benefit from consumers' generated eWOM and marketing managers, in particular, in implementing strategic decisions on website design and product positioning aligned with our results.
This study draws on the social network paradigm and the uses and gratification theory (UGT) to propose a conceptual framework of the motivational drivers of eWOM in-group and out-of-group. In the next section, the theoretical background that underpins the relationships in this study is presented, and the research hypotheses are developed. In the following sections, the research methodology is discussed followed by the analysis and the results. A discussion of the results and their implications for academics and practitioners is presented. The paper concludes with the study's limitations and future research directions.

Model development and hypotheses
The conceptual framework postulates that motivations to use the Internet are positively related to eWOM. Figure 1 shows the conceptual framework.

Social network paradigm
It is our contention that the social network paradigm provides a strong theoretical basis for explaining eWOM. A social network can be defined as a social structure representation in which people are points, connected by lines that represent relationships (Granovetter, 1976). This paradigm assumes these ties link "social actors" (Freeman, 2004, p. 3) in a network formed by one or more "nodes" of individuals in social networks or using websites (Wellman, 2008). Information is exchanged among people who have interpersonal ties that differ in strength. The ties' strength results Drivers of electronic wordof-mouth from a "combination of the amount of time, the emotional intensity, the intimacy [. . .] and the reciprocal services which characterise the tie" (Granovetter, 1973(Granovetter, , p. 1361. Depending on the strength of the ties, these can be classified as weak or strong ties. Weak ties, also called secondary ties, are those established with people with whom one rarely has contact with; strong or primary ties are those connections with family members, close friends and colleagues (Granovetter, 1973;see Brown and Reingen, 1987). Therefore, the social network paradigm is important in an eWOM context, since weak ties tend to connect members of different groups, and therefore out-of-group communication emerges. On the other hand, strong ties tend to be established in specific groups in which in-group communication takes place (Matsumoto, 2000;Granovetter, 1973). Both strong and weak ties are important to promote eWOM because, in combination, they allow widespread information diffusion from one tightly knit group to a bigger, cohesive social segment (Brown and Reingen, 1987;Granovetter, 1973).

Uses and gratification theory: internet usage drivers
Much of the research on internet usage (e.g. Cuillier and Piotrowski, 2009;Grant, 2005) suggests that internet usage is driven by different drivers. An underlying theory that supports this notion is the uses and gratification theory (UGT) (Blumler and Katz, 1974). Employing the UGT in an internet context is not new. In fact, from its early days, researchers have applied UGT to explain internet usage (Morris and Ogan, 1996;Newhagen and Rafaeli, 1996;Charney and Greenberg, 2001;Flanagin and Metzger, 2001). The UGT builds upon three basic principles (Blumler, 1979): first, individuals are goal directed in their behaviour; second, they are active media users; and third, these active users are aware of their needs and select media to gratify them. Scholars have long recognised the importance of individual differences in determining behaviours. Furthermore, it has been shown that individual desires influenced by personality affects how a person seeks gratification (Conway and Rubin, 1991). An individual's values, beliefs, needs, and motives affect his or her behaviours, such as media usage and selection, in order to satisfy a set of psychological needs. As such, the use of a medium such as the internet is aligned with the three principles of the UGT.
We rely on both the UGT and the social network paradigm in our conceptual framework's hypotheses development.

Mood enhancement and escapism
Moods are attached to all human activities, and influence a wide range of cognitive processes and explicit behaviours (Bagozzi et al., 1999;Cohen and Andrade, 2004;Schwarz, 1998). In fact, researchers focused on the question of how moods influence behaviour in shopping, information search, and selecting preferential channels, for a considerable period of time (Puccinelli et al., 2009). Evidence suggests that mood enhancement is based on the pleasure-seeking principle, according to which individuals are thought to constantly search for feel good activities to attain a good mood (Cohen and Andrade, 2004). Indeed, mood enhancement has been found to be one of the strongest motivations for internet usage, especially among young people (Grant, 2005).

EJM 47,7
Escapism, on the other hand, is "a classic motivation associated with most types of media" and particularly with the internet amongst young people (Grant, 2005, p. 612). Escapism has been defined as a state of psychological immersion and absorption (Mathwick and Rigdon, 2004) in which people escape from their everyday concerns and responsibilities for a period of time. Several internet activities are suited to escapism, including surfing the news, weblogs, social networking sites, participating in forums and chat room discussions, as well as spontaneous and constant e-mailing (Charney and Greenberg, 2001;Grant, 2005).
It is possible to identify motives that reflect such needs and personal goals. Using UGT, previous research identified escapism (Abelman and Atkin, 1997) as a motive for using that media. Given that mood enhancement is one of the strongest motivations for internet usage, and because it encourages individuals to think in a broader, more abstract fashion (Labroo and Patrick, 2009) thus facilitating an individual's immersion and absorption, we postulate that: H1. The internet's use for mood enhancement is positively related to the internet's use for escapism.

Mood enhancement and experiential learning
Experiential learning is related to becoming familiar with a certain subject through some type of exposure (Braunsberger and Munch, 1998). Muthukrishnan and Kardes (2001) postulate that individuals often feel that they are learning from experiences when these experiences are enjoyable. Furthermore, research indicates that visual elements, such as pictures (McQuarrie and Mick, 2003), colours (Gorn et al., 2004;Mandel and Johnson, 2002) and aesthetic designs (Veryzer and Hutchinson, 1998) greatly influence information search and elaborative processing (Loken, 2006). Hence, if the use of the internet involves websites that contain features such as those described above (pictures, aesthetic design, etc.), the individual's mood might be enhanced and therefore (enjoyable) learning might take place. Grant (2005, p. 611) argues that while mood enhancement is "a more powerful motivator in absolute terms [. . .], information searching for learning purposes", i.e. experiential learning, "may ultimately be the internet's real point of difference." From a theoretical perspective, we observed that UGT postulates that an individual uses the internet not only because he/she is goal directed but because they also seek to gratify their needs. Based on the UGT framework, Abelman and Atkin (1997) observed the information seeking behaviour of internet users. It is plausible that internet users find that while experiential learning takes place, a pleasurable experience is also occurring. This is because, "the primary use of computer-mediated forms of communication and the Web involves entertainment" (Eighmey and McCord, 1998, p. 189). Additionally, gratification (such as mood enhancement) can be sought in an electronic communication medium, such as the internet, through informational learning and socialisation ( James et al., 1995). Finally, research has demonstrated that a strong correlation exists between moods and learning (Bagozzi et al., 1999). Given that individuals can experience experiential learning through the use of the internet and they also find the use of the internet a pleasurable experience, it is conceivable that: H2. The internet's use for mood enhancement is positively related to the internet's use for experiential learning.

Drivers of electronic wordof-mouth
Escapism and social interaction. Internet activities motivated by escapism are generally associated with positive social outcomes , namely social connectivity. This is because online connectivity offers new opportunities to individuals for social interaction by enabling them to interact with large numbers of others. If not for the internet, such interactions and resulting relationships would have been unlikely, if not impossible, to emerge (Bargh and Mckenna, 2004). The internet offers different forms of social interaction. On one hand, it enables one-to-one relationships with a high level of privacy and personalisation (Kang, 2000). For instance, the internet supports long-distance relationships -across regional boundaries and the globe (Wellman et al., 2001) -and facilitates nearly cost-free, continual communication among family members, friends, colleagues and acquaintances, long-lost friends and co-workers who are physically distant. Such social interaction is supported by software programmes such as Skype which allow individuals to communicate across the world, not only with text but with real-time voices and images, thus resembling actual, in-person communication. On the other hand, individuals can send and receive a great deal of information via social networks, e-mails and blogs, across socially integrated online communities (Lee and Zaichkowsky, 2006), and consequently achieve escapism.
Research on internet developments, such as Second Life, also confirms the importance of social interactions (Chesney et al., 2009). Nevertheless, the strength and quality of online relationships can vary. Some researchers argue that they are very similar to those developed in person (Parks and Floyd, 1996) while others indicate that online relationships are less valuable than offline ones, with their benefits dependent on whether they supplement or substitute offline social relationships . What is not disputed is that the internet allows users to escape reality. This escapism does not threaten social life and in fact allows users to enlarge their social networks (DiMaggio et al., 2001;Howard et al., 2001) and is aligned with the principles of the social network paradigm. Overall, online tools may promote escapism and will probably expand social contacts (Wellman et al., 2001). Thus, we propose the following hypothesis: H3. The internet's use for escapism is positively related to the internet's use for social interaction.
Social interaction and eWOM. WOM in virtual communities is a key marketing issue, because within these groups information can reach millions of individuals (Brown et al., 2007). Community is defined as a set of interlinked relationships that meets members' needs (Kalyanam and McIntyre, 2002). Virtual communities can resemble traditional primary reference groups, such as friends and family members (Jepsen, 2006), as well as secondary reference groups, such as colleagues and co-workers. Virtual community members consider those communities as "places" for contact with people who share their interests (Maignan and Lukas, 1997;Wellman and Gulia, 1999). These virtual communities offer many opportunities for developing friendships and nurturing close relationships, as a consequence of shared interests, values and beliefs (McKenna et al., 2002). Membership and participation in a relevant virtual group may indeed become a central part of an individuals' social life (Bargh and Mckenna, 2004). The fact that virtual community members tend to engage in substantial WOM exchanges (Alon et al., 2002) justifies eWOM's importance from a marketing perspective. Based on the social EJM 47,7 network paradigm, following Brown and Reingen (1987) and Matsumoto (2000), we can observe that eWOM in-group occurs in groups characterised by close relationships or strong ties, such as family and close friends; while eWOM out-of-group generally occurs between people with weaker ties, such as in social networking groups aimed at reaching the mass public. Since eWOM is a social phenomenon that occurs in group settings (see Alon and Brunel, 2006;Brown and Reingen, 1987), the more consumers interact in a group, the more likely they will be to use eWOM to reflect their knowledge and enhance their reputation as experts about specific products (see Wojnicki, 2006). Hence, it can be postulated that: H4a. The internet's use for social interaction is positively related to eWOM in-group.
H4b. The internet's use for social interaction is positively related to eWOM out-of-group.
Experiential learning and eWOM. E-communication enables people to share information and opinions with others more easily than ever before . The internet has extended consumers' options for gathering assumedly unbiased product information from their peers . Furthermore, the internet provides consumers the opportunity to offer their unique consumption-related advice by engaging in eWOM in message boards, internet forums, chat rooms and social networking sites. In particular, internet forums give consumers the opportunity and ability to share experiences, opinions and knowledge with other consumers (Bickart and Schindler, 2002). When consumers generate information based on their personal experiences, this information tends to exert more impact on others' attitudes and holds more credibility than if it were generated by advertising companies and corporate marketing departments (Walsh et al., 2009;Bickart and Schindler, 2002;Kempf and Smith, 1998). Moreover, eWOM's credibility is justified by the fact that other "consumers are perceived to have no vested interest in the product and no intentions to manipulate the reader" (Bickart and Schindler, 2002, p. 428). Hence, consumers find the information exchanged on internet social networks more relevant and trustworthy, as the information reflects product consumption in real-world settings by other consumers and is free from marketeers' interests (Bickart and Schindler, 2002;Jepsen, 2006). As Granovetter (1973) noted in his expounding of the social network paradigm, this information exchange may depend on a combination of the amount of time, the emotional intensity, and the intimacy of the networks. Based on the UGT framework, earlier it was argued that internet users use the internet medium for experiential learning as this was likely to be positively related to mood enhancement. Therefore, consumers who become familiar with a service or product through experiential learning are therefore likely to engage in eWOM about that experience with other consumers as it is a positive experience. Hence, we expect that: H5a. The internet's use for experiential learning is positively related to eWOM in-group.
H5b. The internet's use for experiential learning is positively related to eWOM out-of-group.

Research method
To test the hypotheses, we conducted a survey of internet users in Portugal. We used a convenience sample of internet users. The individuals in the sampling frame were university undergraduate students from one faculty within a university who were invited to participate in the study through an e-mail. In the subsequent lectures students were made aware of the importance of this study. Three hundred and ten e-mails were sent to students and 302 students agreed to participate. This study's questionnaire was initially developed in English and then translated into Portuguese. To avoid translation errors, the questionnaire was back-translated into English by a different researcher (see Douglas and Craig, 1989). The questionnaire was then given to a pre-test sample of thirty young adults who use the internet regularly before being distributed to the 302 respondents.
The respondents' ages ranged from 18 to 35 years old, 25 per cent of the students were 21 years old or younger, 50 per cent of the students were 22 or 23 years old, and the remaining students were 24 years old or older. Most students were female (58.9 per cent). With regard to the internet usage behaviours, 33.1 per cent of our respondents use the internet on a daily basis for up to 29 minutes, 27.5 per cent use it from 30 to 59 minutes, 22.8 per cent use it from 1 h to 1 h 59 m and the remaining (16.6 per cent) use it for more than 2 hours daily. These results are in line with the fact that an estimated 97.3 per cent of Portuguese young adults use the internet on a regular basis (Marktest, 2009).
Measures for the constructs were adapted from existing studies (Grant, 2005;Lam and Mizerski, 2005). The six constructs were mood enhancement, escapism, experiential learning, social interaction, eWOM in-group and eWOM out-of-group. Respondents were asked to assess all the items, using a seven-point Likert scale, ranging from "1 -strongly disagree" to "7 -strongly agree". A complete listing of the questionnaire items can be found in Table I. All scales' internal reliability (Cronbach, 1951) is significant: an average of 0.81 (see Lages et al., 2008). Although all constructs present Cronbach alphas above the recommended value of 0.70 (Nunnally, 1978), the construct "Social Interaction" presents a a of 0.67, which may be considered questionable (Cronbach and Shavelson, 2004). We have decided to include this construct because this value is near the recommended level of 0.70 considering that this construct comprises only two variables. Other studies in many contexts present a values between 0.60 and 0.70 (see Lages and Lages, 2005;Ntoumanis, 2001).

Measurement analysis
To assess the measures' validity, the items were subjected to a confirmatory factor analysis (CFA), using LISREL 8.72 (Jöreskog and Sorbom, 1996). In this model, each item is restricted to load on its pre-specified factor. Despite the fact that the chi-square for this model is significant (x 2 ¼ 648:43, df ¼ 174, p , 0:001), fit indices reveal an acceptable fit: the comparative fit index (CFI) is 0.93, the incremental fit index (IFI) is 0.93 and the Tucker-Lewis fit index (TLI) is 0.92. Since fit indices can be improved by allowing more terms to be freely estimated, we also assessed the root mean square error of approximation (RMSEA), which assesses fit and assigns a penalty for lack of parsimoniousity (Holbert and Stephenson, 2002). The RMSEA of this measurement model is .095, which indicates a satisfactory fit to the population (Chen et al., 2008). We EJM 47,7 also assessed the standardised root mean square residual (RSMR), which has a value of 0.069 and thus indicates a good fit (Hu and Bentler, 1999).
All six constructs have acceptable levels of composite reliability, namely 0.7 or higher (Bagozzi, 1980). Also Fornell and Larcker's (1981) variance extracted values are above the recommended level of 0.50 for all six constructs (see Table II Notes: a ¼ Internal reliability (Cronbach, 1951); r ¼ composite reliability (Bagozzi, 1980) Sources: a Grant (2005); b Lam and Mizerski (2005)  Drivers of electronic wordof-mouth of their intercorrelation) being less than the average variance explained in the items by the construct (Fornell and Larcker, 1981). The correlations among all constructs and the average variance extracted for each construct are presented in Table II. Convergent validity is evidenced by each item's large and significant standardised loadings on its intended construct, with an average loading size of 0.76 (see Table I). Hence, none of the correlations in the final model were sufficiently high to jeopardise discriminant validity (Anderson and Gerbing, 1988).

Structural model estimation
In line with recent research (Cinite et al., 2009;Walsh et al., 2009), we estimated the structural equation model (see Figure 2), using the maximum likelihood (ML) estimation procedure in LISREL 8.72. The model contains six constructs, which correspond to 21 observable variables (see Table I). Where covariance based structural equation modelling is employed, Nunnally (1978) suggests an ad hoc rule of thumb that requires ten observations per indicator.  (Fornell and Larcker, 1981); ME ¼ Mood enhancement, EL ¼ Experiential learning, Esc ¼ Escapism, SI ¼ Social interaction, eWOMIG ¼ e-word-of-mouth in-group, eWOMOG ¼ e-word-of-mouth out-of-group reveal that the final model reproduces the population covariance structure, and that the observed and predicted covariance matrices have an acceptable discrepancy between them. Because the reduced chi-squared statistic (x 2 =df ¼ 3:88) is more than the recommended threshold of 3 (Hair et al., 2006), we proceeded to examine the Mardia's coefficient and found that its value is superior to 3, which suggests that the data might not be normally distributed. When faced with such a distribution, Satorra and Bentler (2001) argue that it may be more appropriate to correct the test statistic rather than to use different estimation methods. The SB chi-square statistic (which incorporates a scaling correction for the chi-square statistic when distributional assumptions are violated), corrected for non-normality is calculated at 406.33 (x 2 =df ¼ 2:34). Since this study comprises a large sample size -of 200 or more -, the "detrimental effects of nonnormality" are reduced and may even be negligible (Hair et al., 2006, p. 80). Also Tabachnick and Fidell (2001) highlight that for large samples, variables with statistically significant kurtosis do not usually have a big impact in the analysis. Table III contains the estimation of direct, indirect and total effects for the structural model.
Our results indicate that mood enhancement has a highly positive direct impact on escapism, as well as on experiential learning, which provides support for H1 and H2. Mood enhancement explains 29 per cent of escapism's variance and 32 per cent of experiential learning's variance (see Figure 2). H3 was also confirmed, as escapism has a highly positive impact on social interaction. The percentage of variance in social interaction, explained by its antecedents, is 49 per cent. Surprisingly, we found that while experiential learning has a non-significant impact on eWOM in-group, it has a highly positive impact on eWOM out-of-group. Finally, we proposed that social interaction has a positive impact on eWOM in-group (H5a) and on eWOM out-of-group (H5b). Our results therefore support both H5a and H5b. Overall, the variance in eWOM in-group and eWOM out-of-group, explained by their respective antecedents, is 67 per cent and 51 per cent, respectively.
With the use of path models, we estimated not only the direct, but also indirect and total effects among latent variables (Bollen, 1989). Table III shows that all five indirect effects are highly significant and positive. Mood enhancement has a positive indirect effect on eWOM in-group (0.35, p , 0:01), and eWOM out-of-group (0.37, p , 0:01). The total and indirect effect of escapism on eWOM in-group is highly significant and positive (0.56, p , 0:01); likewise, the indirect effect of escapism on eWOM out-of-group is positive (0.43, p , 0:01).

Discussion
EWOM is an important tool for all organisations, as it influences consumer behaviour and attitudes towards products, brands and the organisation itself. WOM, and in particular eWOM, has an impact on customer loyalty intentions (Gruen et al., 2006), influences sales (Chevalier and Mayzlin, 2006) and ultimately the firm's revenue (Liu, 2006). Despite its importance and a considerable amount of research on eWOM, there have been recent calls for additional research on the topic (Gupta and Harris, 2010;Zhang et al., 2010 The current study is therefore an attempt to advance our understanding of eWOM, and in particular the drivers of different types of eWOM. In the next sections, the theoretical and practical implications of the research are discussed.

Theoretical implications
This paper provides a number of theoretically grounded contributions to eWOM literature. First, this study offers insights into eWOM dynamics. In particular, our results demonstrate that when internet users aim to enhance their mood, namely through entertainment, amusement, excitement and relaxation, they enter a state of psychological immersion and absorption, which takes them away from their everyday worries and responsibilities, setting the ground for social interaction. Simultaneously, when using the internet to enhance their mood, individuals tend to become more familiar with certain goods and services by gathering information from other peer consumers and thus experiencing learning.
Second, in examining the influence of experiential learning on eWOM in-group and out-of-group, the results demonstrate that experiential learning is not related to eWOM in-group, but it does have a positive relationship with eWOM out-of-group. This differential effect of experiential learning on eWOM out-of-group and eWOM in-group is anchored in the premise that the information circulated through weak ties is more novel than information that flows through strong ties (Granovetter, 2005;see Weenig and Midden, 1991) as strong-tie individuals tend to validate their common knowledge when sharing information (see Phillips et al., 2004). The underlying reason is that an individual's in-group members move in the same circles and therefore a substantial overlap of information already exists among them. On the other hand, an individual's out-of-group members have contact with people whom the individual does not know. As such, more novel information is generated (Granovetter, 2005(Granovetter, , 1983) and more experiential learning may occur and subsequently be shared with out-of-group members. Additionally, weak-tie sources are more numerous and varied than strong-tie sources strengthening the argument that information gathered in weak-tie groups is richer and more meaningful to information seekers (Duhan et al., 1997). Thus, while group members with strong ties tend to validate their common knowledge when sharing information, unique knowledge is received from individuals with whom one has weak ties (see Phillips et al., 2004). Another possible explanation for our results is that when individuals use the internet for experiential learning, they engage more in eWOM out-of-group, given that they spend less or no (face-to-face) time with their out-of-group members (Granovetter, 2005) in comparison to their in-group members. Hence, in line with socio-psychological studies (e.g. Weenig and Midden, 1991), this study's results support Granovetter's (1983Granovetter's ( , 1973) "strength-of-weak ties" hypothesis.
Finally, we also respond to a call in the literature for additional research on the mood enhancement construct (Davis, 2009), by illustrating its central role in driving other internet usage motivations and ultimately eWOM. Mood enhancement has been found to be positively related to escapism, which in turn is positively related to social interaction. This study also found that social interaction among users will ultimately influence eWOM in-group and eWOM out-of-group.

Managerial implications
In line with Kozinets et al.'s (2010) and Ha and Perks' (2005)work, our results suggest that when individuals use the internet, they are likely to engage in eWOM. Thus, in their marketing efforts, companies should design their websites to generate entertainment and amusement, while providing information about their products which appeals to consumers. Companies can capitalise on the internet by coordinating web designers and marketeers' tasks to provide an ingenious and appealing website design anchored in rich content, such as videos, aimed at lifting consumers' moods. For example, Blendtec (a seller of powerful blenders, mainly to private households), created a web page containing a video where an iPhone was thrown into a blender soon after the launch of the iPhone. The light-hearted video was a resounding success with 6.9 million views, and dedicated social media pages with discussions on the virtues of Blendtec products, which resulted in sales growth of 700 per cent. It is also apparent that, if websites facilitate social interaction, they will benefit from consumers engaging in eWOM with both in-group and out-of-group members. Our results confirm findings of online environment research that asserts that consumers come together to interact socially Jepsen, 2006). As a result, discussion participants share product information and gain general information about the company itself. For example, visitors to the website www.clubpenguin.com/puffle/ can play online games, interact with fellow visitors, engage in eWOM about the site and "puffles", and ultimately buy "puffles" in a retail store. Companies should therefore strive to provide opportunities for social interactions on their website and, at the very least, provide links to Facebook and other social websites, which ultimately promote eWOM. Dominos Pizza, for example, showed a 10 per cent increase in sales in 2010, following a Facebook recipe campaign which encouraged users to start an eWOM campaign (Ohngren, 2012). Another example is Babylicious, a company that relies solely on eWOM for promotion. Marketing managers should also consider whether their product lends itself to promotion via eWOM, and if the product is responsive to eWOM promotion, managers should facilitate it.
We found a differential influence of experiential learning on eWOM in-group and out-of-group, specifically that experiential learning is important in eWOM out-of-group. This signifies that individuals are prepared to devote their time and energy to start conversing with others provided they feel that they are learning and it is enjoyable. "My Starbucks idea" (http://mystarbucksidea.force.com/ideaHome) is perhaps an illustration of this. The site allows users to submit suggestions to be voted on by Starbucks consumers, and the most popular suggestions are highlighted and reviewed. In effect, Starbucks have managed to get individuals to create content, and harness the resulting eWOM by adding a feature called "Ideas in Action" blog that gives updates to users on the status of changes suggested.
Consumers do regard eWOM as reliable information sources, far more so than advertising and marketing messages (Walsh et al., 2009;Bickart and Schindler, 2002;Kempf and Smith, 1998). Although companies might be advised to make their websites entertaining and informative, and to provide opportunities for social interaction (for example by creating discussion boards about specific brands and products), such provision can also lead to adverse eWOM, particularly from dissatisfied consumers. This is the main reason why some companies, such as Ryanair, still do not provide this service. However, in an environment in which dissatisfied consumers are free and able EJM 47,7 to create their own forums, discussion boards, and so on, it might be more prudent to cater to their needs and offer these on the company's own website, rather than having them setting up their own information channels. If a firm provides customers with an appropriate forum or discussion board on their website, the firm will benefit from gaining up-to-date information and feedback on consumer dissatisfaction and, as such, will be able to monitor and address the consumer's concerns promptly.
In summary, organisations need to develop websites that are simultaneously entertaining and informative, and that provide social interaction opportunities in order to generate eWOM. Given that online communities are open to everyone, the firm may decide to monitor the information exchanged in the most important communities (e.g. Facebook, MySpace, Twitter and LinkedIn). The firm can then act upon the eWOM information, whether it is positive or negative, and regard it as a great opportunity to receive product feedback, and also to reach their consumers in a more subtle way. For example, the company might post reply messages on online communities to help "spread" their message. In addition, businesses may also apply content-management practices to eWOM content and use it to their advantage. Owing to the interest in social networks and their potential marketing effect, many organisations around the world have an extremely strong financial incentive to understand and facilitate information exchange among individuals who engage in eWOM.

Research limitations and future research directions
Despite this study's theoretical and practical contributions, we acknowledge its limitations. The first limitation is that the questionnaire might have created common method variance, which might in turn have inflated the relationships among the constructs. This could be a threat if the respondents were aware of the conceptual framework of interest. However, respondents were not informed of the purpose of the study, and all of the constructs' items were separated and mixed, making it difficult for respondents to detect which items measured which factors.
A second limitation relates to the convenience sample characteristics, which limit the generalisability of the results. In particular, the sample comprises young adults, who are University students, in Portugal. Future studies with larger samples could allow for a comparison between young, middle-aged and older internet users. This research was conducted in a country in which the internet usage rate among young adults is extremely high (97.3 per cent). Future studies could replicate our study across a different sample and in diverse cultural contexts, characterised by various levels of internet access and usage. It may be that the internet usage motivations will differ, as well as their impact on both eWOM in-group and out-of-group.
Another potential limitation stems from the fact that we used two items to reflect the social interaction construct. It would be desirable if future studies would use at least three items to measure this construct. Another key issue to be explored in future research is the consequences of these two types of eWOM -in-group and out-of-group -and their relative impact on the firm's performance. Additionally, there may be moderator relationships that have not been taken into account in this model. Nevertheless, given that the proposed hypotheses are new, from a theoretical viewpoint, it is important to first understand the direct relationships and then, in a later study, once these relationships are well-established, to explore the role of possible Drivers of electronic wordof-mouth moderator variables. Suggestions for further research include considering age, gender and education level as moderators of the relationships between social interaction and e-WOM in-group and out-of-group and between experiential learning and e-WOM in-group and out-of-group. Finally, future studies can investigate the antecedents of both eWOM in-group and out-of-group by focusing on the volume of eWOM generated for each type.