The role of online brand community engagement on positive or negative self-expression word-of-mouth

The aim of this research is to explore how online engagement influences positive versus negative consumption-focused self-expression word-of-mouth. Three brands are selected to be the focus of the study (Starbucks, Apple, and McDonald’s), and six online brand communities of these brands participated. The online questionnaire was distributed to those communities and filled in by 600 members of the respective love and hate brand communities. Affection (passion/ aversion) is the motivational factor that leads community members of both types (love and hate) to be active in commenting and writing reviews about the product/ brand. In hate brand communities, connecting negatively influences consumptionfocused self-expression word-of-mouth, whilst brand influence has a positive effect on it. Subjects: Internet/Digital Marketing/e-Marketing; Marketing Communications; Relationship Marketing


PUBLIC INTEREST STATEMENT
Consumers are more and more active in using social media and brand communities to expose their messages. Thus, the process of being engaged in self-expression word-of-mouth gives insights about how brand managers could encourage such comments to be used to cocreate products. The emotional bond-love or aversion-is the most effective factor for consumers be engaged in brand communities. Thus, a brand closer to the consumer, gradually revealing the novelties, responding to questions and comments-always caring-lead consumers more active in co-creation processes. Negative strong feelings and the need to influence others to cause damage to the brand can be used by brand managers to understand the key negative factors of consumer-brand relationships. In this situation, brand should be able to show that they care and want to recover what did wrong. Kumar, 2017). Loureiro (2015) also calls for more research on negative engagement and the effects of hating a brand on its reputation and on managerial implications. Therefore, the importance of this research lies in innovatively comparing the effect of being negatively and positively engaged with online brand communities on the positive/negative consumption-focused selfexpression word-of-mouth (CSWOM). The value is to give insights about which dimensions of negative and positive engagement have greater impact on CSWOM.

Introduction
CSWOM refers to a way of communication about one's consumption activities to expressing one's self -concept and attracting attention to oneself (Saenger, Thomas, & Johnson, 2013). This study focuses on how a positive versus negative state of engagement influences the consumers' motivation to spread word-of-mouth about consumption activities (products and brands) in order to self-express. In this vein, CSWOM emerges as a more appropriate concept than the traditional one to analyze the role of positive versus negative valence of online brand community engagement on the way a consumer self-express about their consumption activities (products and brands) in brand communities.
Following this introduction, next section provides a theoretical foundation pertaining to a review of previous research related to consumer engagement conceptualizations and dimensionality and consumption-focused self-expression word-of-mouth as well as the proposed hypotheses. The next sections describe the research methodology and report on the testing of the hypotheses. Finally, we present the discussion, implications, limitations, and suggestions for future research.

Consumer engagement conceptualization and dimensionality
Previous research has suggested that consumer engagement could result in brand usage intent (Hollebeek, Glynn, & Brodie, 2014), or experiencing firm performance (Kumar et al., 2010;Kumar & Pansari, 2016) or even interactive outcomes such as word-of-mouth, loyalty and consumer retention through consumers' value co-creation (Verhoef, Reinartz, & Krafft, 2010). Brand usage intent refers to the intention to use and prefer the online brand that the consumer is engaged with instead of any other brand. When consumers are engaged with some brand, she/he will be more willing to keep using that brand and could also be more dedicated to express her/himself through the brand by using word-of-mouth.
Regarding experiencing firm performance, Kumar and Pansari (2016, p. 502) claim that if a consumer is engaged with the firm, the consumer will make purchases from the firm and will also provide referrals contributing to the firm's profitability. Hence, if the consumer is willing to repeat the purchase and provide referrals, it is expected that she/he is also likely to be willing to express her/himself using word-of-mouth. Thus, consumers engaged with a firm/brand feel somehow connected and identified with the firm/brand (Steward, Narus, & Roehm, 2018). This process of engagement leads consumers to be more proactive in expressing themselves through word-ofmouth (Beckers et al., 2018;Verhoef et al., 2010).
Consumer engagement transcends concepts of involvement or participation (Gagné, 2003) since it constitutes an active and interactive customer connection with a given object, while involvement and participation fail to reflect the interactive and value co-creation experiences (Brodie, Hollebeek, Jurić, & Ilić, 2011). Further characterizing this interaction, Hollebeek et al. (2014, p. 154) conceptualize consumer-brand engagement as a "consumer's positively valenced brandrelated cognitive, emotional and behavioural activity during or related to focal consumer/brand interactions." Three dimensions emerge from the study of Hollebeek et al. (2014, p. 22) in terms of cognitive processing (processing and elaboration in a particular consumer-brand interaction), affection (positive brand-related affect in a particular consumer-brand interaction), and activation (energy, effort and time spent on a brand in a particular consumer/brand interaction). Yet, the manifestation of particular cognitive, emotional, and behavioral dimensions depends extensively on the engagement actors-engagement subjects/objects-and contexts (Brodie et al., 2011;Li et al., 2018), including media contexts (new online media in contrast to traditional advertising media) (Calder, Malthouse, & Schaedel, 2009;Loureiro, Gorgus, & Kaufmann, 2017). Actually, the scale developed by  is not particularly dedicated to online brand communities. In this respect, Baldus, Voorhees, and Calantone (2015) made a first attempt to conceptualize online brand community engagement as the compelling, intrinsic motivation to continue interacting with an online brand community. To establish the baseline dimensions, consisting of 11 dimensions the authors pursued the grounded theory approach. Consecutively, they employed Churchill's (1979) paradigm to refine the scale which is used in the current study: Brand influence, Brand passion, Connecting, Helping, Like-minded discussion, Rewards (hedonic), Rewards (utilitarian), Seeking assistance, Self-expression, Up-to-date information, and Validation (e.g., Albert, Merunka, & Valette-Florence, 2008;Algesheimer et al., 2005;Batra, Ahuvia, & Bagozzi, 2012;Dholakia et al., 2004).
However, engagement could also have a dark side potentially leading to codestruction of brand value, or impoverishment of value by consumers and providers (Dolan, Conduit, & Fahy, 2016;. Such damaging behavior can be caused by consumers' perceived reputation of the brand, self-confidence, product involvement, proximity to others, and attitudes to the business in general as well as perceived worthiness of complaining (Lau & Ng, 2001). Yet, the action of participating in codestruction of brand value can also be reflected by individuals who are highly motivated in damaging an individual's perception of a specific brand, product or, even, business. These individuals-consumers or non-consumers of a specific brand-do not only share negative feelings and messages toward a specific brand, but also become engaged in doing so. This aspect of consumer behavior creates new research avenues in terms of developing new concepts in the marketing literature (Juric et al., 2016).
According to Hollebeek and Chen (2014) consumers, in a state of positive-valenced consumer engagement, display high commitment toward an object (often, a brand). These consumers are regarded as brand devotees, with a strong connection to the specific brand and a high level of engagement in co-creating value. However, in the same condition of high commitment, albeit characterized by negative-valenced consumer engagement, individuals become brand adversaries, also connected, in a sense, with the focal object, but with the intention to destroy or damage the brand. This is supporting the theory that positive and negative brand engagement are two opposite sides of the same construct. The regularity and severity of negative engagement behaviors have been found to progressively develop through constant incongruence between consumers' expectations and a brand's actual behavior (Chylinski & Chu, 2010). Consumers driven by revenge and the desire to feel less anxious act in diverse ways: while some use negative, even malicious and vindictive word-of-mouth Grégoire, Salle and Tripp (2015) to decrease the frustration caused by dissatisfying experiences, others even go beyond this state and are proactive in attempting to damage the brand (Juric et al., 2016). In addition, the nature of the relationship between consumers and brand is relevant when predicting the intensity of the interactive actions (Hollebeek, Cadogan, & Fastoso, 2018;Juric et al., 2016). Some consumers may forgive brands easier, for previous unethical behavior or unsatisfactory service, when the risk of harm is not high. Other consumers become gradually displeased as the level of harm tends to rise (Riza & Sh., 2015). Loyal consumers expressing strong and deep attachment toward a brand may feel deceived and respond more intensely than others, socalled transactional consumers, who are not deeply connected with the brand (Grégoire & Fischer, 2008). Nonetheless, past research does not explain why consumers react with changing levels of engagement, or why they manifest negative engagement behaviors (Juric et al., 2016), given a large amount of possible events and contexts that motivate such attitude and actions toward the brand. The knowledge about the consequences of brand engagement is scarce, particularly the effects of negative engagement on the way consumers self-express their opinions and recommendations.
Cognitive dissonance theory is addressed in the literature as a way to explain why consumers express themselves negatively about a firm/brand/service after a negative experience has occurred. This theory is rooted in the notion that consumers may go through cognitive dissonance when brands are not able to deliver what consumers are expecting, or when a brand's general business causes concerns. When such events occur, consumers engage in NWOM as a way of decreasing the levels of cognitive dissonance. In doing so, consumers are able to persuade others about their choice of action (Balaji, Khong, & Chong, 2016).
In the current study, we follow Hollebeek's and Chen's (2014) suggestion to consider positive and negative engagement as two opposite sides of the same construct and adopt the scale proposed by Baldus et al. (2015, p. 979) for online brand community engagement and adapt it to negative engagement. The 11 dimensions may be grouped in six core facets (see Table 1): (i) Information. This facet deals with Up-to-date information and Validation: it represents the information that circulates between members and is exposed in the online community, as well as the validation of the information by peers.
(ii) Social identification. By Self-expression and Like-minded experience a member identifies herself/himself with the brand and other members and regards others as similar to her/ himself.
(iii) Connection and influence. Here the members feel to be connected to some good thing bigger than themselves and perceive a certain degree of active influence on the brand. The sentences used for these two dimensions were slightly modified to be adapted to the context of Love and Hate brand communities.
(iv) Interact with members. Helping and Seeking assistance are two dimensions that evidence the interaction among members. The former motive relates to helping other members by sharing knowledge, experience or even time while the latter is aimed at receiving assistance from the community.
(v) Affection. This facet called Brand passion (Baldus et al., 2015) means an ardent positive affection a community member has for the brand. In the current study, a further facet representing Brand aversion, reflecting an ardent negative affection a community member has for the brand, is added. The intention hereby is to express the same intensity of feeling but in a negative way. Therefore, the same sentences of the scale developed by Baldus et al. (2015) for brand passion are used but changing words like "passion" to "aversion." (vi) Rewards. For the first time, hedonic (fun and enjoyment) and utilitarian (monetary rewards, prizes) rewards are also considered in this research. This is considered appropriate due to the evolution of the internet from an informational to a more transactional medium.

Consumption-focused self-expression word-of-mouth
Word-of-mouth (WOM) is a widely known concept employed to express recommendation to friends and family and advocate in favor of a brand or firm (e.g., Mazzarol, Sweeney, & Soutar, 2007;Zeithaml, Berry, & Parasuraman, 1996). Contributing to the complexity of the WOM concept, WOM communication toward a firm can be provided by a single individual once or several times or by various individuals once or at different times (Mazzarol et al., 2007). Unexpectedly, the incidence of positive word of mouth (PWOM) is significantly higher (three times) compared to negative word of mouth (NWOM) (East, Uncles, Romaniuk, & Hand, 2007). This is particularly prevalent at category level, while at brand level the frequency of NWOM may still exceed that of PWOM. PWOM is regarded to be caused by cognitive factors, while NWOM by emotional factors (Sweeney, Soutar, & Mazzarol, 2008).
Several conditions have been identified as facilitators rather than as motivators of WOM. One condition consists in the intention to promote a brand or firm due to positive perceptions developed over time. A different condition that acts as a facilitator is the closeness in the relationship between receiver and giver of WOM. This reflects the notion that WOM is mostly used between family and friends, and the theory of homophily (Mazzarol et al., 2007).
The advancement of electronic WOM (eWOM) has opened avenues for new forms of interactive marketing communication, instead of the limited traditional one-way communication between brands and consumers via mass communication channels (Kim, Wang, Maslowska, & Malthouse, 2016;Rialti et al., 2017). However, due to the similarity of traditional online and offline WOM concepts, characteristics that the literature has identified as important in traditional offline contexts can also be perceived as being relevant in an online environment (Dessart, Veloutsou, & Morgan-Thomas, 2015;Hennig-Thurau, Gwinner, Walsh, & Gremler, 2004). Yet, some researchers challenged this view suggesting that determinants and characteristics of offline WOM may lack appropriateness when trying to describe eWOM and its effects on consumer behavior (Fu, Ju, & Hsu, 2015). Fu et al. (2015) claim that online contexts have several events and aspects that are exclusive to such environments. This view is supported by Hennig-Thurau et al. (2004, p. 39) describing eWOM as "any positive or negative statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet". In particular, it differs from traditional word-of-mouth in three aspects: first, eWOM is presented in the written form, and it can be read; second, it exists in public online forums or websites, available for any user, consumer or brand to observe; third, once online, eWOM is electronically stored and can be used in the future (Kim et al., 2016). Summarizing, eWOM is a rapid and informal way of exchanging thoughts and experiences with other geographically dispersed consumers, concerning goods, brands, or services (Fu et al., 2015;Kim et al., 2016).
Indicating the usefulness of eWOM for individuals desiring to influence others, Sun, Youn, Wu, and Kuntaraporn (2006) hold that eWOM messaging is an activity of "online opinion leadership" in which people-mainly opinion leaders-affect the behavior, thoughts and attitudes of friends or others. In fact, consumers tend to trust more in opinions (positive and negative) found online than ads in mass media (Chu & Choi, 2011;Fu et al., 2015;Kim et al., 2016;Loureiro et al., 2017;Rialti et al., 2017). However, distinct from other previously discussed motivations to share WOM, is the consumption-focused self-expression motive: "the desire to communicate about one's consumption activities for the purpose of expressing one's self-concept and attracting attention to oneself" (Krishnamurthy & Kucuk, 2009, p. 960). Consumers applying consumption-focused self-expression word-of-mouth (CSWOM) do not intend to affect other consumers' purchase intentions or assist them in making consumption decisions (Kokkoris & Kühnen, 2015). CSWOM's sole purpose is to inform others of that consumer's consumption behavior, allowing herself/himself to be heard and to manifest her/his true self. In addition, those involved in spreading eWOM for self-expression do not seek to be perceived as experts or being more innovative when compared to others (Saenger et al., 2013).
The individuals engaged in CSWOM may also discuss that specific brand, to which they are loyal, without having any intent to influence that brand and its success. The action of commenting on liked brands works only as a way of communicating their self, therefore, meeting an emotional need rather than the desiring to influence the brand (Saenger et al., 2013). The degree to which individuals are motivated to engage in CSWOM is also dependent on personal characteristics (Pagani, Hofacker, & Goldsmith, 2011;Thorbjørnsen, Pedersen, & Nysveen, 2007). Since consumption-focused self-expression WOM has no desire to promote the brand, this type of word-of-mouth communication is regarded as one of the most influential and credible types of consumer-generated communications with brands (Saenger et al., 2013).
Consumers seem to participate in expressing negative WOM for several motives: (1) to look for advice on how to resolve the issue in question; (2) to prevent other consumers from undergoing the same experiences; (3) to react against the brand, its products or services; and (4) to express their frustration and anger by using NWOM as a way of decreasing cognitive dissonance (Kim et al., 2016). In addition, consumers may engage in negative WOM in order to draw attention to the problem, which caused their dissatisfaction, with the goal of finding a solution, or as a method for anxiety reduction through venting negative emotions. However, the fact that consumers who experienced a negative event, that caused frustration or disappointment toward the brand, share their feelings online indicates that negative eWOM can be a relevant factor of response behavior (Verhagen, Nauta, & Feldberg, 2013). Underpinned by the cognitive dissonance theory, this study proposes that the fact that a consumer is negatively engaged with an online brand community may influence her/his self-expression word-of-mouth. Brand engagement is linked to consumers' use of brands to enhance self-expression of identity (Bergkvist & Bech-Larsen, 2010). Brand community integration impacts brand identification which, in turn, has an effect on consumer loyalty and WOM communications (Millán & Díaz, 2014). Whilst this identification increases community engagement, negative effects such as normative community pressure and reactance are also reported (Algesheimer et al., 2005; Hollebeek & Chen, 2014).

Hypotheses development
Past research has suggested that the result of customer engagement could be the brand usage intent , or firm performance (Kumar et al., 2010;Kumar & Pansari, 2016) or even interactive outcomes such as word-of-mouth, loyalty, and consumer retention through customers' value co-creation (Verhoef et al., 2010). Brand usage intent means the intention to use and prefer the online brand that consumer is engaged instead of another brand. When consumers are engaged with some brand, he/she will be more willingness to keep use that brand and could also be more dedicated to express themselves through the brand by using word-of-mouth.
Regarding for performance, Kumar and Pansari (2016, p. 502) claim that if a customer is engaged with the firm, the customer will make purchases from the firm and will also provide referrals, which contribute to the firm profurability. We suggest that if the customer is willing to repeat the purchase and provide referrals, it is expected that she/he is also likely to be willing to express her/himself using word-of-mouth. Thus, customers engaged with a firm/brand feel somehow connected and identified with the firm/brand. This process of engagement leads customers to be more proactive in expressing themselves through word-of-mouth (Verhoef et al., 2010). They will talk about the products/brands they use to others (to expose their inner self) and get the attention of others, that is, self-expression word-of-mouth (Saenger et al., 2013). So, the influence of a member in the community (Algesheimer et al., 2005), the ardent affection for the brand (Albert et al., 2008), the feel of being connected to something good and talk with similar people to express opinions and impressions, share and receive up-to-date information, have fun, gain utilitarian rewards and have the opinions validated (Baldus et al., 2015) are all dimensions of being engaged to a brand community. Yet, what dimensions of online engagement could be more effective on CSWOM? We argue that the 11 dimensions of online engagement could enhance CSWOM at different levels of intensity. Therefore, we formulate the following hypotheses: H1: The dimensions brand influence (H 1a ) brand passion (H 1b ), connecting (H 1c ), helping (H 1d ) likeminded discussion (H 1e ), hedonic rewards (H 1f ), utilitarian rewards (H 1g ), seeking assistance (H 1h ), self-expression (H 1i ), up-to-date information (H 1j ), and validation (H 1l ) of online community engagement are related to members' consumption-focused self-expression WOM in love brand communities.
Hollebeek and Chen (2014) highlight that when in positive-valenced consumer engagement, consumers display high commitment toward an object (often, a brand), they are viewed as brand devotees, with a strong connection to the specific brand and engaged in co-creating value. Yet, the opposite could also happen through consumers' unfavorable brand-related feelonhgs, thoughts and behaviors during the engagement process. In their conceptual model, Hollebeek and Chen (2014) suggest that consumers to be negatively engaged with a brand and be equal proactively and passionately involved (feeling aversion) in a negative engagement as others could be in a positive engagement.
Cognitive dissonance theory is often addressed in the literature as a way to explain the reason by which consumers express themselves negatively about a firm/brand/service after a negative experience occurred. This theory is rooted in the notion that consumers may go through cognitive dissonance when brands are not able to deliver an offering what consumers are expecting, or when a brand's general business causes concerns. When such events occur, consumers engage in NWOM as a way of decreasing the levels of cognitive dissonance. In doing so, consumers are able to persuade others about their choice of action (Balaji et al., 2016).
Underpinned by this theory, this study proposes that the fact that a consumer is negatively engaged with an online brand community may influence his/her self-expression word-of-mouth. Brand engagement was linked to consumers' use of brands to enhance self-expression of identity (Bergkvist & Bech-Larsen, 2010). Brand community integration impacts brand identification which, in turn, has an effect on consumer loyalty and WOM communications. (Millán & Díaz, 2014). Yet, not all effects of identification with brand communities may be positive. Whilst this identification increases community engagement, negative effects such as normative community pressure and reactance were also reported (Algesheimer et al., 2005;Hollebeek & Chen, 2014). Thus, it is expected that members of a brand community could be engaged in a negative way to a certain brand. The dimensions of such negative engagement could be similar of those associated to positive engagement, except the passion for the brand, which we adapt to aversion for a brand. Consistent with the argumentation for the positive engagement, the dimensions of negative engagement could enhance CSWOM at different levels of intensity. Therefore, we propose the following hypotheses: H2: The dimensions brand influence (H 2a ) brand aversion (H 2b ), connecting (H 2c ), helping (H 2d ) likeminded discussion (H 2e ), hedonic rewards (H 2f ), utilitarian rewards (H 2g ), seeking assistance (H 2h ), self-expression (H 2i ), up-to-date information (H 2j ), and validation (H 2l ) of online community engagement are related to members' consumption-focused self-expression WOM in hate brand communities.

Sample and procedure
Data were collected in 6 online brand communities, grouped in pairs with each pair belonging to a diverse brand and comprising respectively both valences: positive (love or fan) online brand community and negative (hate or antibrand) community. Hate or anti-brand communities are online spaces that focus negative attention on a specific targeted brand (Bailey, 2004;Krishnamurthy & Kucuk, 2009). Accordingly, a love brand community is considered an online space that focuses positive attention on a specific targeted brand. The three major brands were investigated as follows: Starbucks, Apple, and McDonald's, each brand with a love and a hate brand community. These brands were systematically ranked as top global brands (Interbrand) or top performers on annual growth (Interbrand, 2017) and attracted mixed publicity. The selection of those brands meet the following criteria, by order of importance: (1) the brand must have an official or/unofficial online brand community that mostly displays positive/negative word-of-mouth (online community); (2) each brand should have at least one love (fan) brand community and one hate (antibrand) online community (completely independent from one another) different from the others as well as the social website in which it is posted; (3) the brand's online communities should reveal activity (i.e., posts, comments, reviews); and (4), if possible, the total amount of brands in the study should represent no less than two industries.
In the current study, it seemed a promising route to search for brand communities by following the procedure of Krishnamurthy and Kucuk (2009). Thus, queries were used to identify love and hate brand communities, using positive and negative terms and the name of the brand such as: "anti," "hate," "sux," "bad," "watch," "blow," "sucks," "love," "super," "great," "good," "my," "adoration," and "fan." The raison d'etre is to find brand communities devoted to positive/negative and favorable/unfavorable comments about the brand, depending if it is a love or a hate brand community. Thus, the researchers visit each online community to analyse its characteristics and uniqueness.
The selected communities were contacted by researchers and we asked permission to the coordinator of each community to spread the questionnaire anonymously. We also explained the aim of the research and the need to get a sample representative of the members of the community in terms of age and gender.
A total of 600 fully completed and usable questionnaires (after excluding those with missing values, inconsistent responses or extreme multivariate outliers) were collected, 300 from love brand communities (fan communities) and another 300 from hate brand communities (anti-brand communities) of the same three brands (split equally for the three brand communities).

Measures
The questionnaire was prepared in English. The dimensions for online brand community engagement were adapted from Baldus et al. (2015) with the support of marketing researchers and brand consultants. Given its unique characteristics and applicability, the same scale was used to assess the responses of negative engagement in online communities. Specific modifications were made that allowed for context and valence adaptation while retaining the essence and reasoning of the original item possibly being lost in the transformation (from positive to negative). In order to measure consumption-focused self-expression WOM, six items adapted from Saenger et al. (2013) were employed ("I like to talk about products/brands, so people can get to know me better"; "I like the attention I get when I talk to people about product/brands"; "I talk to people about my consumption/anti-consumption activities to let them know more about me"; "I like to communicate my consumption/anti-consumption activities to people who are interested in knowing about me"; "I like the idea that people want to learn more about me through my consumption/anticonsumption activities"; "I like it when people pay attention to what I say about my consumption activities").
The questionnaire was pretested with 10 consumers, and only a few adjustments needed to be made in order to improve the flow of the survey and clarify any misunderstanding. After that, the questionnaire was distributed to the brand communities contacted by the authors (authors asked permission to collect data and explained the purpose of the research). All scale items were evaluated using a Likert-type, ranging from 1 (one) to 7 (seven) were: 1-Strongly disagree; 2-Mostly disagree; 3-Somewhat disagree; 4-Neither agree nor disagree; 5-Somewhat agree; 6-Mostly agree; 7-Strongly agree. The questionnaire also contained socio-demographic variables and those to characterize the participants (number of hours, on average, spent on the internet per week; number of posts, on average, per week; on a scale from 0 (I hate t) to 10 (I love it), what do you feel about the brand x?).

Results
With regards to the sample of participants from the love brand communities, 40% are female and 60% male. Most of the participants are between 21 and 30 years old (69.71%) (M = 26.41, SD = 5.66). The average number of hours using the Internet per week is 37.03 (SD = 17.11). The average number of posts per week and per participant is 2.53 (SD = 4.62). Having asked how participants felt about the brand on a scale from 0 (I hate it) to 10 (I love it), the average value is 7.43 (SD = 1.61).
Considering the participants of the hate brand communities, 22.70% are female, and 77.30% are male. Most participants are between 21 and 30 years old (59.00%) (M = 29.35, SD = 7.66). The average number of hours using the Internet per week is 34.24 (SD = 11.27). The average number of posts per week and per participant is 2.53 (SD = 4.62). Having asked how participants felt about the brand on a scale from 0 (I hate it) to 10 (I love it), the average value is 0.91 (SD = 1.10).
In both types of brand communities, participants have several different nationalities, such as the United States, the United Kingdom, Canada, Australia, South Africa, India, Belgium, Philippines, Argentine, Portugal, and Spain. The proportion of participants of both genders in the two samples is representative of the number of community members.
The assumptions for multiple linear regression (such as normality, multicollinearity, autocorrelation) were tested followed by the estimation using SPSS23). The hierarchical multiple regression is selected because this allows for a fixed order of entry of variables in order to control for the effects of covariates or to test the effects of certain predictors independent of the influence of others. All participants in both types of communities use or have already experienced the brands on which they comment. Table 2 shows the descriptive statistics of the variables as well as the convergent validity and reliability values. The values of AVE (Average Variance Extracted) are above 0.5, indicating that most of the variance of each indicator (item) is explained by its own construct (Kleijnen, Ruyter, & Wetzels, 2007). All Cronbach's alpha values are above the recommended threshold of 0.7 (Nunnally, 1978).
In addition, a confirmatory factorial analysis (CFA) (see Table 3) of the engagement scales using Lisrel 8.80 was conducted. The CFA of the scale for love online brand engagement offers an excellent fit to the data (χ2 = 1410, df = 725; NNFI = 0.95; CFI = 0.92; RMSEA = 0.07). The same good fit is recognized for hate online brand engagement (χ2 = 2062, df = 764; NNFI = 0.90; CFI = 0.91; RMSEA = 0.08). Hierarchical regression analysis was used to test the hypotheses. Hierarchical regression analysis is a way to show if some dimensions of online brand engagement explain a statistically significant amount of variance of self-expression word-of-mouth after accounting for all other dimensions. Thus, several regression models were built by adding variables to a previous model at each step. The underlying interest is to determine whether newly added dimensions show a significant improvement in R 2 . This framework is considered more adequate for the purpose of the current study because it facilitates understanding on what are the most statistically significant dimensions in influencing self-expression word-of-mouth. Furthermore, amongst the not statistically significant dimensions, those who are still significant when not considering the most significant could be identified. Table 4 reveals that Helping, Seeking assistance, and Self-expression are not statistically significant in explaining online community engagement in love brand communities (for those, who are a brand fans). Brand passion, Rewards (hedonic), Rewards (utilitarian) and Validation account for extra 12.6% of the variance in CSWOM. In fact, Brand passion (β = 0.53, p < 0.001) and Validation (β = 0.22, p < 0.001) are the most significant predictors of CSWOM in love communities, and so H 1b , H 1f , H 1g and H 1l are supported. The hypotheses H 1a , H 1c , H 1e , and H 1j are supported only when the dimensions of brand passion, hedonic rewards, utilitarian rewards, and validation are not presented. The other hypotheses are not supported. Table 5 shows that Validation is not statistically significant in explaining online community engagement in hate brand communities (for those, who are in antibrand communities). Brand influence, Helping, Connecting, Rewards (hedonic), and Rewards (utilitarian) account for extra 29.2% of the variance in CSWOM. Actually, Brand influence (β = 0.364, p < 0.001), Brand aversion (β = 0.447, p < 0.001) and Rewards (hedonic) (β = 0.243, p < 0.001) are the most significant predictors of CSWOM in hate communities, and so H 2a , H 2b , H 2c , H 2f , and H 2g are supported. The hypotheses H 2d , H 2e , H 2h, H 2i , and H 2j are supported only when the dimensions of brand passion, brand aversion, connecting, hedonic rewards, and utilitarian rewards are not presented. The hypothesis H 2l is not supported.

Conclusions
In sum, the valence love/hate does not mean that the same drivers contribute to the CSWOM process. Lovers are more motivated by emotions, passion, and validation, and haters want to spread their negative knowledge about a brand but also contribute to influence the brand to change the behavior or improve the features of the product and get fun in making those comments. Those members that really have an aversion toward the brand tend to be more active than those that merely do not like or are neutral about the brand.

Discussion
The results of the current study show the strength of dimensionalities of engagement on CSWOM for love versus hate online brand communities. Both hypotheses are partially supported. For those who participate in the communities of love for Starbucks, Apple, and McDonald's, Brand passion, Rewards (hedonic), Rewards (utilitarian), and Validation are the most significant predictors to generate CSWOM. The process to develop a passion for the brand and the community not only engages fans with the community but also leads them to be brand captives by developing strong ties with the brand. The ardent affection a community member has for the brand (Baldus et al., 2015) is, indeed, a strong engagement mechanism to enhance the continuity of the consumer-brand relationship. Consistently, Batra et al. (2012) and Albert et al. (2008) have already stressed the strength of the passionate desire, the harmony between consumers and their loved brands as a core factor to love a brand, reflecting higher-arousal and hotter aspects of brand love.
Community members do not use the comments and posts in the online love brand community to get enjoyment and entertainment, and they also do not expect to receive deals, incentives, merchandise, prizes. They want that their opinions and ideas are favorably validated by other members of the community. Validation or the favorable evaluation of the opinions, ideas, and interests by other members of the community reinforce the passion and the positive emotions between a brand and the consumers through the relationships established among the community members. From the findings, a core triad crystallizes that contributes to being continuously engaged in a love brand community. Yet, in the study of Baldus et al. (2015) brand passion, utilitarian and validation were not drivers of participation intention. This study is more aligned with other more conceptual past research (e.g., Algesheimer et al., 2005;Hennig-Thurau et al., 2004) suggesting passion for a brand and the validation of their opinions by others to be motivational for proactive participation. When the dimensions of brand passion, rewards and validation are not considered, brand influence, connecting, and like-minded discussion emerge as significant influencers of CSWOM. Therefore, participants in love brand communities want to talk with people similar to themselves, influence the brand and consider the connection a very positive process. Regarding hate brand communities, the significant relationships between hedonic rewards and CSWOM and utilitarian rewards and CSWOM are remarkable. Hedonic rewards is a dimension related to positive emotions (e.g., fun, enjoyment, entertainment, friendly environment, and social status) gathered through participation in the community and utilitarian rewards associated to incentives, merchandise and prizes are crucial to motivate members to express themselves through word-of-mouth. Even so, when comparing hedonic rewards and utilitarian rewards, the latter seem to be less effective on CSWOM than the former. This confirms the study results of Baldus et al. (2015) that community members are more concerned with fun and entertainment that come from posting comments on brand communities than getting money or other utilitarian rewards from the comments about their consumption. Similarly, Mousavi, Roper, and Keeling (2017) highlighted the importance of cultivating affective commitment in order to develop brand commitment and make consumers more immune to negative information.
Brand aversion is another dimension that encourages members to participate. Although further studies are needed to consolidate this finding, the current study reveals, for the first time, that rewards and passion/aversion are important engagement dimensions creating in members the desire to express themselves about brands (positively or negatively) (Prentice & Loureiro, 2018). Interestingly, the dimension validation is not so important as it is in the case of love brand communities. Members of hate brands communities want to influence the brands and are not so interested in getting their opinions about the brand validated by other members. Rather, brand influence or the degree to which a community member wants to influence the brand is very important for the haters; they like to influence, be provocative and change brand behavior. This dimension is reinforced by brand aversion meaning the opposite of being in love with a brand. Members of this kind are disgusted with the brand or even hate it and perceive hedonic rewards reflecting the entertainment and fun participants have when posting comments and talking about the brand. In other words, those who hate the brand, want to influence it and feel emotional rewards hating the brand, tend to be more active in writing comments about it.
When members of hate brand communities increase the strength of their connection with the brand, they tend to be less engaged in talking and writing comments about the brand and their consumption activities. Similar to Connecting, Like-minded discussion is another dimension negatively related with CSWOM. This means that those who enjoy talking with others similar in the brand community do not tend to talk about their consumption activities. Although the strength of Seeking Assistance exerts less influence than Connection and Like-minded discussion on CSWOM, this dimension has a more negative effect as the other two. Thus, in hate brand communities, members are not proactive in receiving help from fellow community members. Rather, they prefer to express their negative thoughts about the brand and and get some up-to-date information provided by other members.
The topic of social influence of brand communities has been discussed in other studies (e.g., Algesheimer et al., 2005;Dholakia et al., 2004;Prentice & Loureiro, 2018) but, as far as we know, this research is the first attempt to significantly associate this dimension with participating in hate brand communities. Effectively, it could be elicited that members of hate brand communities desire to express themselves in a negative way using the online platform and view that online site a forum where they can express their negative opinion and socialize with other members as suggested by Wallace, Buil, and de Chernatony (2014).
Helping emerges as an important dimension only when the core dimensions (brand influence, brand aversion, rewards, and connecting) are not considered. This means that a community member would like to help fellow community members by sharing knowledge, experience, or time. Therefore, helping is not the priority among those who like to comment on brands and their experience of consumption.
In sum, affection (love/passion) is the main motivational factor that leads community members of both types to become active in commenting and writing reviews about the brands. Regarding brand love communities, information (validations and up-to-date information), as well as connection and brand influence are other important motivational factors. On the other hand, in hate brand communities, Connecting negatively influences on CSWOM, but brand influence has a positive effect on CSWOM.
Although more studies on this topic are called for, it seems that utilitarian rewards are not a mandatory dimension of engagement. Managers of online communities should provide opportunities for sharing positive experiences and create positive engaging experiences (e.g., Innocent big knit campaign to raise money for Age UK). Vice versa, managers of brand communities should not be concerned with preparing and attributing utilitarian rewards (e.g., monetary rewards, time savings, deals or incentives, merchandise, and prizes).
In hate brand communities, members feel rewarded when posting negative opinions about the brand. Moreover, they tend to enjoy such action. They have a negative image and feelings about the brand, and so they want to influence the brand through their negative comments and believe that they are helping others in doing so. Brand managers are advised to be aware and follow the posts and comments made by members of hate communities in order to improve their products. Eventually hate brand community member might become more pro-active in the co-creation process than members involved in love brand communities.

Managerial implications
Overall, it is recommended that brand managers should be aware that both, positive and negative emotions flourish in the online communities. Emotions in online communities need to be carefully managed, as annoyance with fan pages may generate negative WOM among online brand community members (Hutter, Hautz, Dennhardt, & Füller, 2013).
In a positive way, brand community members want to express their ideas or opinions and socialize with others, get information, and influence the brand. A member of a love brand community feels that she/he is involved with a group committed to some good thing bigger than themselves (the brand). In a negative way, members want to expose their negative thoughts and feelings about the brand to others, they enjoy to do that and they think they're helping others doing that.
Thus, brand managers should analyze the information gathered from both types of communities in order to raise the brand, but, at the same time, be able to not become influenced by some reviews that purely express emotion without contributing to improve the products behind the brand. Furthermore, it is recommended that brand managers should understand the situations/ experiences causing a negative feeling about the brand. Managers should understand how long the passion could go; if a passionate member could become a hating one and what the tipping point could be.
The emotional connection and the openness to expose opinions are key aspects to involve customers with the brand. Thus, it is suggested that brand managers should involve members of both types of communities in their co-creation process of new products/services.

Theoretical implications
Regarding theoretical implications, the findings extend the previous studies on online engagement (e.g., Algesheimer et al., 2005;Baldus et al., 2015;Hollebeek & Chen, 2014) by taking a first step in testing the effect of dimensions of an engagement scale with two valences on self-expression word-of-mouth. Another contribution is to enhance understanding on which dimensions of online engagement, proposed by Baldus et al. (2015) adapted to the current study, are more effective on love and which ones are more relevant for hate brand communities.
In this vein, similarities and differences of engagement in both types of communities were analyzed. The value of this research is to give insights about which dimensions of negative and positive engagement have greater impact on CSWOM.

Limitations and further research
The anonymous questionnaire was created based on previous studies and prepared in order to avoid bias. Even so, the study has limitations, and other studies should be conducted in future to validate the findings. For instance, the love and hate brand communities aggregate members of different countries, and so some differences could emerge depending on the national cultures. The levels of cultural dimensions such as collectivism or uncertainty avoidance (more data are needed to understand this issue) could necessitate a differentiation of the findings.
Other variables are suggested to be analyzed together with online engagement and self-expression such as word-of-mouth. For instance, the perception of country-of-origin of the brand could positively or negatively influence brand image and the willingness to participate in love/hate brand communities. A further suggestion is to design a study investigating the tipping point changing a brand lover to a brand hater.
Future studies should increase understanding on how negative reviews and comments posted in hate brand communities can influence the purchase behavior of consumers over time, especially, what expressions written in hate brand communities could cause greater damage in the brand image.