Electronic word-of-mouth on social networking sites: Cue validity and cue utilization perspectives

Abstract

This study explores how people evaluate and adopt electronic word-of-mouth (eWOM) information on social networking sites (SNSs). Applying the logic of the Brunswikian lens model, we develop a framework suggesting that mechanisms of cue utilization (consistency-checking and knowledge-based validation) validate the cues of identity claims (argument strength and perceived expertise) and behavioural residues (graphics, external links, recommendation ratings, and framing of wall postings) in order to judge the perceived credibility of eWOM messages and, consequently, to adopt the suggestion of eWOM messages on SNSs. The research model was empirically evaluated based on web survey results from 445 Facebook users. Results indicate that the cues of identity claims and behavioural residues contribute to perceived credibility of eWOM through two mechanisms of cue utilization. Implications for theory and practice are provided.

Keywords

Behavioral residues cue validity cue utilization electronic word-of-mouth social networking sites

Yu-Hui Fang is an associate professor in the Department of Accounting at Tamkang University, Taiwan (ROC). She received a PhD degree in information management from National Central University and a Master’s degree in Accounting from the University of Houston. Her research interests include electronic commerce, virtual communities, service science, and knowledge management. Her research has appeared in Human Systems Management, Information Systems Journal, International Journal of Electronic Commerce, Journal of the Association for Information Systems, Computers in Human Behavior, Service Business, Pacific Asia Journal of the Association for Information Systems, Journal of E-Business, Internet Research, Online Information Review, and others.

Chia-Ying Li is an Associate Professor in the Department of Business Administration at the National TaiChung University of Science and Technology, Taiwan. She received her Ph.D. degree in Business Administration from National Cheng Kung University, Taiwan. Her research interests are in electronic commerce, strategic marketing management and knowledge management, etc. She has published articles in Human Systems Management, Service Business, Journal of Service Theory and Practice, Asia Pacific Management Review, Computers in Human Behavior, International Journal of Information Management, Journal of Information Science, Expert Systems with Applications, International Journal of Human Resource Management, Industrial Management and Data System, and elsewhere.

1 Introduction

Social networking sites (SNSs) have become one of Web 2.0’s fastest growing services. Nielsen 1 claims that two-thirds of the world’s Internet population visit SNSs or blogging sites, such that time spent on these types of sites is growing at over three times the rate of overall Internet growth. SNS users can interact with others and share information with ease and speed; as a result, this promotes multidirectional communication among users. Social media provides a new platform for social interaction as well as electronic word-of-mouth (eWOM) communication among SNS users. More specifically, social networking technologies help individuals conquer most information asymmetries and thus achieve high levels of market transparency by creating content, such as posting about their experiences and spreading eWOM on SNSs. Given that the social networking phenomenon is dramatically changing the way individuals behave and, consequently, create new opportunities for eWOM communication, the main research question is: How do individuals evaluate and adopt eWOM on SNSs?

eWOM communication has received extensive attention from both academics and practitioners across two major streams. Some have focused on marketers’ efforts to enhance promotions and the adoption of new products through social contagion or communication processes between consumers in target markets [18]. Others have investigated users’ pass-on behaviours, such as their sharing of motivations or inhibitions [10]. Both research streams are derived from the perspective of contribution of information. However, relatively little attention has been devoted to the information evaluation process based on the eWOM receiver’s perspective [8]. The information evaluation process starts with receivers’ judgment of the credibility of eWOM [38] and ends with eWOM adoption. If the potential of SNSs to facilitate the adoption of eWOM is to be realized, we need to identify factors that help users assess the credibility of eWOM and predict successive adoption of eWOM based on receivers’ subjective perceptions. This study aims to fill the knowledge gap by identifying an appropriate theoretical framework, evaluating it empirically, and suggesting opportunities for further work within this research stream.

Brunswik [5] proposed the Brunswikian lens model, used to manipulate environmental cues, as a lens through which observers can make inferences about a target. The model has proven to be valuable in terms of clarifying the process by which individuals utilize cues to form perceptions and make decisions [32]. Therefore, we apply the Brunswikian lens model to scrutinize the process receivers use to assess the credibility of eWOM on SNSs. Brunswik [5] conceptualized the link between an observable cue and an observer’s judgment of that cue as cue utilization. Starting with Hovland et al.’s [24] three facets that influence information assessment (i.e., source, information, and receiver), we analyze observable cues on SNSs in terms of their source and information from the receivers’ perspective, and divide observable cues into two parts: identity claims and behavioural residues [20]. Identity claims are reflected in symbolic statements made by targets (i.e., profile owners) for their own benefit and are intended to reinforce their views, whereas behavioural residues are physical traces of activities conducted in the environment. Research has extended the categories of identity claims and behavioural residues, and the Brunswikian lens, from physical space to static personal web space [36]. Even though personal websites have been portrayed as “highly controlled context[s]” wherein self-expression consists almost exclusively of identity claims, it remains true that “no real-world environment can be completely free of behavioural residue” [36] (p. 124). Extending this logic to examine the assessment of eWOM on SNSs, this leads to our first investigative question on identifying the determinants of identity claims and behavioural residues in the context of SNSs.

Additionally, prior literature on learning has suggested two epistemic strategies— consistency checking and knowledge-based validation— that reflect specific types of cognitive learning strategies directed at validating knowledge claims raised in informational texts [28]. In order to better understand how cue utilization functions in the SNS context, we propose these epistemic strategies are used in the process of cue utilization on SNSs, and may play a role in eWOM evaluation on SNSs. As such, our second investigative question is concerned with how these strategies of cue utilization mediate the relationships between identity claims, behavioural residues, and the perceived credibility of eWOM onSNSs.

Drawing from the Brunswikian lens model and extending epistemic strategies from the learning literature, we explore how identity claims and behavioural residues influence the credibility of eWOM through mediation of cue utilization on SNSs. This study makes several contributions in this field. First, it extends the Brunswikian lens model from the physical world into virtual space, which has been ignored in prior studies. This application also provides a systematic means for analysing eWOM assessments on SNSs. Second, beyond argument strength and perceived expertise, this study identifies behavioural residues as additional key determinants when judging eWOM on SNSs from the receivers’ perspective, which is an issue that has not received much attention thus far. This identification expected to help profile owners manage not only their eWOM messages, but also their behavioural residues to ultimately facilitate receivers’ eWOM adoption. Third, drawing potential mediators from the learning literature, this study attempts to offer preliminary theories on how people evaluate and adopt eWOM on SNSs, and empirically verifies the theoretical model.

2 Theoretical background and hypotheses

2.1 Electronic word-of-mouth on SNSs

eWOM communication refers to any positive or negative statement made by potential, actual or former customers about a product or a company, which is made available to a wider audience via the Internet [22]. Indeed, the disruptive nature of eWOM could become an essential tool for disseminating the truth about fundamental changes taking place in our globalized world, overcoming the traditional media’s conservatism and complicity in protecting the status quo. An eWOM message can be more credible than any other commercial Internet information [25] and widely distributed on SNSs. Given that an SNS such as Facebook “has a richer, more intimate hoard of information about its citizens than any nation has ever had,” 2 the influence of eWOM information on SNSs is powerful and cannot be ignored. However, eWOM has no value unless receivers of SNSs actually apply the information communicated through eWOM and consider it in their purchase decision-making process (i.e., eWOM adoption). eWOM adoption points to the changes in the behaviour of eWOM receivers as a result of accepting what the informant advocates [8]. eWOM adoption can be an effective measure of the persuasiveness of eWOM and the most meaningful consequence of eWOM communication for practitioners. eWOM adoption serves as the dependent variable in this study.

SNSs refer to “web-based services that allow individuals to (1) construct a public or semi-public profile within a bounded system, (2) articulate a list of other users with whom they share a connection and (3) view and traverse their list of connections and those made by others within the system” [6] (p. 211). As such, eWOM on SNSs differs significantly from other web-based eWOM in terms of the visible display of (1) connections (“contacts”, “fans” or “friends”) who are also users of SNSs and (2) information, opinions, reviews, or ideas posted by the profile owner on their profile page [7]. eWOM on profile pages of SNSs are not anonymous, as they are linked to the profile owner and can be observed by the profile owner’s connections if the profile is open to the public. The mutual and decontextualized nature of the interpersonal links between connections and posts on profile pages seems to mitigate the tendency of profile owners to misrepresent information or display a self-idealized virtual identity. Since profile owners may not selectively present a part of their network or posts to a subset of connections, the observable display acts as a sign of credibility of the profile owner’s identity and the information on the profile page [13]. The credibility of eWOM on SNSs, therefore, is likely to rely not only on the source and the information, but also on inferences that receivers make about the profile owners’ activities (i.e., posting activity and posts from connections). The perceived credibility of eWOM reflects the extent to which one perceives online recommendations or reviews on SNSs as believable, true, or factual [8]. As stated by Wathen and Burkell [38], an early key phase in the information persuasion process is the receiver’s judgment of the information’s credibility, which determines how much a receiver learns from and adopts the received information. In the context of eWOM communication, a receiver who believes that an eWOM review is credible tends to learn from it and use it. In contrast, a receiver who perceives an eWOM message as not credible is unlikely to risk acting on the review. Cheung et al.’s [8] work has empirically supported the link between information credibility and adoption. Therefore,

H1: The perceived credibility of eWOM is positively related to eWOM adoption on SNSs.

2.2 Brunswikian lens model

The Brunswikian lens model illustrates that environmental cues function as a kind of lens through which observers draw inferences about the underlying constructs of a target. Based on this approach, a target produces behaviour and creates an artefact that reflects the target’s characteristics. For example, an organized desk could serve as the lens through which an observer perceives the target’s high level of conscientiousness. Brunswik [5] discussed the utility of various cues in terms of cue validity, cue utilization, and functional achievement. Cue validity (or encoding), meaning “good information,” refers to the link between the observable cue and the target’s actual level of the underlying characteristic. Encoding is the objective and visible manifestation of the quality of a sender and the information that is presumed to be some aspect of information communication. Cue utilization (or decoding), a meaning system, refers to the link between the observable cue (e.g., the organized desk) and an observer’s judgment (e.g., conscientiousness). Decoding is the use of observable cues by an observer to infer aspects of information communication in the sender. Functional achievement is defined as the co-occurrence of both cue validity and cue utilization. When functional achievement occurs, observers’ judgments should converge with the underlying construct being observed and result in observer accuracy. The basic assumption is that people exist in an uncertain environment, and judgments about the environment are therefore made on the basis of diverse cues [5]. Judgments of a criterion are made on the basis of cues with different ecological validities. Cues differ in terms of how they are utilized by observers and the inferences drawn by those observers. Drawing on this logic, the Brunswikian lens model has been used to investigate the following factors: individuals’ perceptions of their physical environments, individuals’ predictions of criteria that are related to cues, interpersonal communications, and the verbal and nonverbal communication process [5, 32]. A main advantage of the lens model lies in recognizing that human judgment and the criterion being predicted can be thought of as two functions of cues available within the decisionenvironment.

To date, the Brunswikian lens model has primarily been utilized for investigations in the context of physical surroundings, but has rarely been applied to digital space. Given that the web space on SNSs (e.g., Facebook users’ online profiles and posts on their “wall”) also contains various environmental cues, increasing attention has been devoted to such observable cues in terms of explaining the information process on SNSs [36, 37]. In turn, this study proposes extending use of the lens model to investigate SNSs. First, focus is given to evaluating the credibility of eWOM on SNSs from observers’ perspectives, disregarding the target’s actual level of underlying characteristics. Observers’ assessments have inherent value: their views on a sender’s qualities and the information itself, whether correct or not, help to explain their subsequent attitudes and actions [19]. Second, SNS users can adapt to the medium, utilizing language and other cues to overcome the relative shortage of cue systems to achieve their communication purpose [32]. We, therefore, adopt Gosling et al.’s [20] taxonomy of identity claims and behavioural residues linking individuals to the environments they inhabit as a proxy for cue validity. Third, most previous research on eWOM communication has focused on the antecedents of eWOM credibility, but ignored the decoding process of receivers. This study incorporates the lens model to fill this gap. We take the extra step of considering two epistemic strategies— consistency-checking and knowledge-based validation— as mechanisms of cue utilization that help clarify how meaning systems function on SNSs while evaluating the credibility ofeWOM.

2.3 Cue utilization: Consistency checking and knowledge-based validation on SNSs

As noted above, consistency-checking and knowledge-based validation are viewed in this study as two components of cue utilization that help toexplore appraisals of eWOM on SNSs. These epistemic strategies reflect specific types of cognitive learning strategies that are intended to validate knowledge claims raised in informational texts. According to the learning literature [28], one can benefit fromepistemic strategies directly and indirectly. As a direct benefit, utilizing these strategies can keep one from uncritically encoding fabricated or incorrect information. As an indirect benefit, utilizing these strategies can initiate domain-specific knowledge and an active search for relationships among information scattered across the message. These activities may be assumed to result in a well-organized and tightly integrated knowledge representation. Accordingly, these epistemic strategies are useful in judging the credibility of eWOM on SNSs.

Consistency-checking refers to the extent to which an information receiver can actively monitor and confirm the consistency of eWOM presented on SNSs by searching for additional information, such as whether the eWOM is consistent with other online reviews available. In the case of SNSs, forming perceived credibility toward received eWOM allows observers to mine static repositories of a target’s past interactions, assess profiles on personal profile web pages [37], and “google” relevant information online (i.e., consistency-checking). It is not difficult to obtain reviews and opinions from different experienced users associated with diverse online sources to check for consistency between the reviews. Observers who engage in higher levels of consistency-checking learn specific domain knowledge better and increase their ability to avoid uncritically encoding false information [28]. If the current review is inconsistent with most other opinions of the same product, the receiver will recognize the contradiction and determine the current eWOM review to be less credible [35].Therefore,

H2: Consistency checking is positively related to the perceived credibility of eWOM on SNSs.

Knowledge-based validation refers to the extent to which an information receiver can use his/her prior knowledge and experience to judge the plausibility of eWOM presented in SNS spaces. Receivers can validate a specific eWOM review by detecting the level of agreement between the received eWOM and their past experiences or knowledge regarding the reviewed product. If an eWOM review contradicts prior knowledge or experiences, the receiver may refuse to believe the review and discount its validity. Specifically, when they perceive the current review as congruous with their prior knowledge or experiences, they have more confidence in the received information and generate higher levels of perceived credibility toward that eWOMreview [8].

H3: Knowledge-based validation is positively related to the perceived credibility of eWOM on SNSs.

2.4 Cue validity— Identity claims and behavioural residues on SNSs

According to Hovland et al. [24], three key facets (the source, the information, and the receiver) affect the evaluation of information. Available in the context of SNSs, for the information facet, there are verbal cues (information cues related to linguistic expression, including textual descriptions, numbers and symbols) and nonverbal cues (information cues related to nonlinguistic objects and events, including pictures, videos, etc.) [26]. Nonverbal cues account for 10.3 times as much variance as verbal cues in interpersonal perception [2]. Accordingly, in addition to the frequently mentioned information factor of argument strength, we incorporate one verbal cue (recommendation rating) and two nonverbal cues (graphics and external links) as additional information factors due to their accessibility on SNSs. Likewise, for the source facet, beyond perceived expertise, we propose the framing of wall postings as an additional source factor due to its traceability on SNSs. Furthermore, we classify the abovementioned factors using Gosling et al.’s [20] taxonomy— identity claims and behavioural residues— as indicators of cue validity in helping to comprehend how the source and content of eWOM are judged on SNSs. The rationale for considering them as a proxy for cue validity is that observers employ various features on SNSs to assess the credibility of a profile owner (source) and his/her content. These features include things that the profile owner deliberately displays (identity claims) and things that he/she unintentionally displays (the material left by friends in the profile owner’s electronic social network; behavioural residues) [37]. Taken together, identity claims and behavioural residues can represent cue validity (good information), particularly in the context of SNSs.

2.4.1 Identity claims

Identity claims denote profile owners’ deliberate and intentional displays in digital space. For example, identity claims include information in one’s online profile that is intentionally displayed and under the complete control of the profile owner. Given that argument strength and perceived expertise have been consistently shown to be the most common determinants of information credibility, this study considers argument strength as an information-facet and perceived expertise as a source-facet of identity claims on SNSs. Both facets must be taken into account in order to comprehensively appraise the cue validity of eWOM. Argument strength reflects the extent to which received information can persuade a person to believe something [8]. In our study, it is used to indicate the persuasive strength of received eWOM. Previous research has identified the direct impact of argument strength on receivers’ attitudes, specifically in online surroundings [8]. If the received eWOM is perceived to be persuasive, the receiver will generate a positive attitude toward the eWOM and consider it credible [41]. Perceived expertise refers to the extent to which a source is believed to be capable of making valid assertions [24]. Expertise reflects the knowledge that an individual has about a domain. Given that experts’ eWOM is more credible and reliable than that of non-experts, eWOM receivers are likely to agree more with experts, adjust theirattitudes in accordance with expert opinions, and generate the perception of credibility toward eWOM messages from experts [31].

According to the Brunswikian lens model, observers judge and make sense of the available cues of identity claims (one part of cue validity) so as to infer the perceived credibility of eWOM (cue utilization). On SNSs, “widespread adoption of Internet search engines has made searching an easy and frequently used discovery method for online information seekers” [41] (p. 78). The Internet enables observers to quickly and cheaply access abundant information that is relevant to them. To comprehensively evaluate a specific eWOM review and reduce the risk of receiving misleading information on SNSs, observers try to confirm consistency between the eWOM they receive and what they learn from an online search. For example, after consistency-checking, observers who confirm the greater persuasiveness of the eWOM (argument strength) and that the eWOM is posted by a profile owner with high expertise (perceived expertise) tend to perceive it as having higher credibility [8]. Accordingly, we predict that argument strength and perceived expertise are positively related to consistency checking:

H4: Argument strength is positively related to consistency checking on SNSs.

H5: Perceived expertise is positively related to consistency checking on SNSs.

2.4.2 Behavioural residues

Behavioural residues are the physical traces of activities conducted on SNSs, as previously noted. Behavioural residues denote things that profile owners unintentionally leave in their digital space, such as the observable display of comments, under their eWOM message, by connections (friends and contacts), which may reflect agreement or disagreement with it. As Walther et al. [37] noted, message postings reveal profile owners’ behaviour on SNSs, and 55% of wall postings reflect the profile owner alone or the poster and the profile owner together. Given that behavioural residues can provide additional evidence to support the assertion of a specific eWOM message, these residues function as a supplement to the content of eWOM in a comprehensive evaluation of its credibility. We propose three factors— graphics, external links, and recommendation ratings— as the information facet of behavioral residues in this study, and the framing of wall postings as a source facet of behavioural residueson SNSs.

Graphics, related to traditional graphic design, have been identified as one vital facet of information presentation [27]. In general, graphics play a role in receivers’ eWOM judgment processing by providing them with additional cues. Especially when credibility enhancing cues are presented, receivers are likely to think about credibility overtly. In the case of eWOM communication on SNSs, graphics provide frames of symbolic representations that inform and attract receivers to take a closer look at a specific eWOM message. Zhang et al. [42] confirm the useful role of graphics in promoting further actions and attention by users toward a message. In particular, attractive and clear graphics embedded in an eWOM message compel observers to begin consistency-checking (confirming, through online searches, if that message is consistent with messages provided by others) and knowledge-based validation (verifying its accuracy and credibility based on their prior experience). Therefore,

H6: Graphics are positively related to consistency checking on SNSs.

H7: Graphics are positively related to knowledge-based validation on SNSs.

External links are represented in the user interface and electronic commerce literature as an important element in offering frames of symbolic representations that create an impression on online audiences [39, 42]. As Hlynka and Welsh [23] note, the quantity, functionality and relevance of hyperlinks are vital to the quality of an effective webpage. Further, such links to and from other websites can serve as referrals, entailing third party endorsement and enhancing the site’s credibility [16]. Following the same logic, we propose that external links may play a role in the process of assessing perceived credibility of eWOM messages. Additionally, recommendation ratings reflect the overall ratings of an eWOM message given by other users of SNSs [8]. In this way, SNSs enable users to assess an eWOM message and express their perceptions of it in terms of props (e.g., the number of “likes” clicked) and thecomments/responses to it that they leave for other users to read. It has been acknowledged that such recommendation ratings can considerably affect how observers perceive the credibility of information [8]. If most users of SNSs present a positive rating of an eWOM message, this suggests that most users accept and believe the content of the message.

According to the Brunswikian lens model, receivers validate their knowledge and make inferences about the credibility of eWOM based on behavioural residues. External links and recommendation ratings serve as environmental manifestations and have been confirmed as recognition elements that affect the credibility of websites [16]. Inasmuch as the above arguments can be applied to the process of evaluating eWOM on SNSs, external links and recommendation ratings contribute to judgment of the credibility of eWOM through knowledge-based validation. Specifically, both serve as partial cues to validity (good information) and provide a useful basis for receivers to apply their inferred strategy (i.e., knowledge-based validation). Therefore, both external links and recommendation ratings are proposed to be positively related to knowledge-based validation. Hence,

H8: External links are positively related to knowledge-based validation on SNSs.

H9: Recommendation ratings are positively related to knowledge-based validation on SNSs.

Framing of wall postings reflects the overall rating of a source’s eWOM message, given by friends/connections. Recent research has revealed that friends’ wall postings influence judgments of a profile owner’s credibility. In general, the more positively framed wall postings a profile owner receives, the more trustworthy the profile owner is perceived by observers [34]. Walther et al. [37] noted that the framing of messages generated by the profile owner’s friends significantly impacts observers’ ratings of the profile owner’s credibility. Similar to the reputation function in other computer-mediated communication [8] and other ratings on SNSs, for eWOM communication on SNSs, the framing of wall postings of a profile owner, as rated by other readers, may serve as prominent cues regarding the source’s credibility. As such, behavioural residues left by the profile owner’s friends (rather than explicit identity claims created by the profile owner) are used by observers in the impression formation process [34]. According to the Brunswikian lens model, observers attribute credibility to the profile owner based on the things they observe in the profile owner’s space. Specifically, observers associate non-behavioural cues that reside in the environment (the framing of wall postings in this case) to infer whether such cues agree with or contradict their prior judgment of the profile owner’s credibility (knowledge-based validation). Accordingly, we propose:

H10: The framing of wall postings is positively related to knowledge-based validation on SNSs.

2.5 Consistency-checking and knowledge-based validation as mediators

As discussed earlier, and consistent with the logic behind the Brunswikian lens model, we also posit that consistency-checking is likely to mediate the relationship between two factors of identity claims (argument strength and perceived expertise) and one factor of behavioural residue (graphics) in developing perceived credibility for eWOM messages. Alternatively, knowledge-based validation is likely to mediate four factors of behavioural residue and the perceived credibility of eWOM.

H11: Consistency-checking mediates the positive relationship between the three antecedent variables— argument strength, perceived expertise, and graphics— and perceived credibility of eWOM.

H12: Knowledge-based validation mediates the positive relationship between the four antecedent variables— graphics, external links, recommendation ratings, and framing of wall postings— and perceived credibility of eWOM.

Overall, the research model is shown in Fig. 1.

3 Methodology

3.1 Measures

Measurement items were adapted from existing measures to fit the context of SNSs and are listed, coupled with their sources, in Appendix A. All items were responded to on seven-point Likert scales with the anchors strongly disagree (1) and strongly agree (7). Three steps were taken in order to make the survey suitable to users of SNSs. First, a 90 minute semi-structured, open-ended, face-to-face interview was conducted with three professors familiar with SNSs and the issues surrounding eWOM. These experts offered suggestions regarding our measurement items. Second, a focus group of 57 graduate students who frequently use SNSs was formed to validate the insights gathered in previous steps and the measures that were formulated. This discussion lasted about two hours. Third, another three professors in this stream of research were invited to evaluate the survey questionnaire in terms of logical consistency, contextual relevance, ease of understanding, and the sequence of items. Based on their suggestions, minor modifications were applied to the explanations and examples to eliminate ambiguity.

3.2 Survey administration

The research hypotheses were tested using data collected from Facebook users. Facebook was chosen because it has become the world’s most popular SNS and provides more opportunities than other SNSs for sharing eWOM messages between users. At the end of March 2012, there were more than 125 billion objects with which people interacted (pages, groups, events and community pages) on Facebook.² A web-based survey was employed to target Facebook users. We advertised our web survey on popular websites (e.g., bulletin board systems and Facebook) in Taiwan. We used purposive sampling to invite registered users who received eWOM from Facebook within the past month to participate in this survey; sample elements were selected because they were believed to be representative of the population of interest and were expected to serve the research purpose of our study [11]. Accordingly, in the first question of our survey, we required respondents to indicate whether they had read any eWOM reviews on friends’ Facebook walls in the past month. Then, we asked respondents to recall their most impressive experience related to eWOM over the last month and answer our survey. The first page of the questionnaire explained the study’spurpose, the length of the questionnaire, and the incentive, alongside assurance of participant confidentiality. Sixteen randomly selected respondents were offered a cash incentive of $20USD.

3.3 Sample profile

Initially, 632 respondents completed the survey. The screen question described earlier was applied to exclude those respondents that had not read any eWOM on their friends’ postings to ensure respondent eligibility. The survey yielded a total of 445 valid responses with no missing data for data analysis. The questionnaire also captured respondents’ demographic data such as gender, age, education, and typical activities on Facebook. Among the 445 respondents, all were current Facebook members, 48.2% were males with a mean age of 28.16 years (SD = 8.39) and a mean member tenure of 26.09 months (SD = 12.03). Among the total, 80.9% of our respondents had a university level education or above, and 59.8% spent more than one hour per day on Facebook.

3.4 Data analysis

We used a two-step approach for data analysis as recommended by Anderson and Gerbing [1]. The first step involves analysis of the measurement model, while the second step tests structural relationships among the latent constructs. This two-step approach establishes the reliability and validity of the measures before assessing the model’s structural relationships. SmartPLS 2.0 M3 [29] was used to evaluate both the measurement model and structural model, as partial least squares (PLS) have minimal restrictions in terms of measurement scales, sample size and residual distribution [9].

4 Results

4.1 Measurement model

The measurement model was evaluated for reliability, convergent validity and discriminant validity. As shown in Appendix A, the composite reliability values all exceeded the 0.7 threshold: the lowest was 0.93, indicating adequate reliability [17]. The convergent validity of all scales was established in two ways [17]: (1) all indicator loadings ranged from 0.79 to 0.95, well in excess of the 0.7 threshold, and were statistically significant at the p < 0.001 level; and (2) the average variance extracted (AVE) by each construct ranged from 0.74 to 0.89, exceeding the 0.50 threshold (Appendix A). Adequate discriminant validity was confirmed because the square root of the AVE of a construct was larger than the correlation between construct and other constructs in our model (Table 1) [17]. Based on the above, we concluded that the scales exhibited sufficient construct validity.

Multicollinearity denotes the degree to which an independent variable varies with other independent variables. Given that excessively high multicollinearity would challenge the statistical assumption that the independent variables are truly independent of one another, two additional tests were applied to evaluate multicollinearity— variance inflation factors (VIF) and condition numbers. We conducted a regression analysis with eWOM adoption as the dependent variable and the other nine variables as independent. The VIF ranged from 1.60 to 2.88, well below the recommendedthreshold of 3–5 [21]. The condition number equals the square root of the largest eigenvalue divided by the smallest eigenvalue. The condition number for the current study is 4.62 (√(5.41/0.25)). Belsley et al. [4] advised that condition numbers less than 10 suggest a weak multicollinearity problem. Consequently, we determined that no significant multicollinearity problem existed in our data.

In an effort to prevent common method bias, we used the procedural remedies related to questionnaire design suggested by Podsakoff et al. (2003), such as improving scale items, protecting respondent anonymity, and reducing evaluation apprehension, among others. We also followed Saraf et al.’s [30] recommendation to utilize the single-method-factor approach in PLS after data collection. To assess method variance, a latent-method factor was added to the structural model. We converted each indicator into a single-indicator construct. All major constructs and the method factor become second-order constructs. For each single-indicator construct, we investigated the coefficients of the two incoming paths based on their substantive construct and the method factor, which are equivalent to the observed indicator’s loadings on its substantive construct and the method factor, and can be used to assess common method bias. Overall, common method bias was deemed unlikely to be a serious concern because our results met two criteria: (1) 33 of the 41 method factor loadings were insignificant and (2) the indicators’ substantive variances were substantially greater than their method variances.

4.2 Structural model

In PLS analysis, examination of structural paths and R-square scores of endogenous variables help to assess the explanatory power of a structural model. Table 2 shows the results of the structural path analysis. Paths exhibiting a p-value less than 0.05 are considered significant. Overall, consistency checking and knowledge-based validation together explained 35% of the variance for eWOM credibility. H1 to H10 were supported and the research model accounted for 48% of the variance for eWOM adoption (see Table 2). This offers empirical evidence for the validity and explanatory ability of our model.

The mediating effects of consistency checking and knowledge-based validation (H11 and H12) were assessed using the bootstrapping approach suggested by Zhao et al. [43]; bootstrapping offers a strongermeasure of accuracy for confidence intervals and does not require a normality assumption. Recent approaches to mediation indicate that the mediating relationship is the “total effect” of the sum of direct and indirect effects including the mediator, and that mediation should be established solely by the presence of an indirect effect [43]. As shown in Table 3, we found positive and significant indirect effects of argument strength, perceived expertise, and graphics on eWOM credibility, mediated by consistency checking (B = 0.11 to 0.14, with a 95% confidence interval excluding zero (0.07 to 0.21) at 5000 resamples), supporting H11. Also, we found significant indirect effects of four factors of behavioral residues on eWOM credibility, mediated by knowledge-based validation (B = 0.23 to 0.27), with a 95% confidence interval excluding zero (0.15 to 0.35), supporting H12. According to Zhao et al.’s [43] terminology, given that the direct effects (c path) were all significant, all mediating effects under H11 and H12 should be viewed as complementary mediation, which is similar to Baron and Kenny’s [3] partialmediation.

Prior studies have primarily focused on the influence of argument strength and perceived expertise (identify claims) on eWOM credibility. To demonstrate the contribution of our proposed behavioral residues on eWOM credibility, we also evaluated a model with only identity claims to determine eWOM credibility (R² = 22%) as well as a full model with both identity claims and behavioral residues to determine eWOM credibility (R² = 35%). Then, we compared the R² of both models in order to assess the incremental validity of adding behavioral residues while controlling for identity claims. The difference between the R² values (13%) was subjected to a significance test through the computation of F values and its effect size. The R² difference is statistically significant with an effect size f² of 0.2, which lies between a medium and large effect size. 3 Our results show that the behavioral residues significantly increased R² to eWOM credibility, implying significantly higher explanatory power of the full model as compared to the model with only identity claims. We also calculated the geometric mean of the average communality and average R² (for endogenous constructs), and obtained a goodness-of-fit (GoF) of 0.30 for the model and 0.57 for the full model, where GoF values of 0.5 are seen as quite good [33]. The results indicate a better fit of the structural model with the full model than with the alternative one.

5 Discussion and implications

This study draws upon the Brunswikian lens model in order to advance our understanding of the evaluation process for eWOM adoption in SNSs. It has been acknowledged that people utilize information provided to them online to make inferences about others. However, little is known about the types of information used to make such judgments [37]. As such, this study focuses on both the components of cue validity and strategies of cue utilization to fill the knowledge gap. Specifically, we analyze cue validity in greater detail, identifying argument strength and perceived expertise based on identity claims, graphics, external links, recommendation ratings and framing of wall postings as behavioural residues in the context of SNSs. The findings indicate that even though identity claims play a role in eWOM assessment on SNSs, behavioural residues dominate the evaluation process. As evident in the results section, the full model with both identity claims and behavioural residues has good explanatory power in terms of eWOM credibility and a better fit index than the alternative model.

In addition, the results illustrate how consistency checking and knowledge-based validation mediate cue validity and the credibility of eWOM on SNSs. Specifically, Table 2 reveals that all relationships were statistically significant. Interestingly, three direct effects relationships stand out relative to the others: (1) between graphics and consistency checking (β= 0.29), (2) between recommendation ratings and knowledge-based validation (β= 0.31), and (3) between framing of wall postings and knowledge-based validation (β= 0.25). These results are likely due to the unique nature of SNSs compared with other online and offline contexts relating to the availability of behavioural residues on SNSs. For example, in the context of SNSs, recommendation ratings toward an eWOM review represent the extent to which other users of SNSs agree with that review. The framing of wall postings can be considered a proxy for the reputation of a source on SNSs, given that it can be used to infer source credibility: users make inferences about an informant and his/her message based on comments left by his/her friends. Although such comments are not posted by the informant, users believe them to be sanctioned by the informant and use these cues to judge the credibility of the informant and the message in a matter consistent with the Brunswikian lens model [37]. This is similar to a Chinese proverb claiming that “it must be true if all men say it.” Overall, the inclusion of behavioural residues in this study provides a reasonable starting point for further exploration of the potential factors associated with behavioural residues on SNSs.

5.1 Theoretical implications

The empirical evidence in this study provides support for our proposed model linking cue validity with two strategies of cue utilization to influence the perceived credibility of eWOM reviews and ultimately eWOM adoption. The majority of previous studies have suggested that argument strength and source credibility are the crucial determinants of eWOM credibility [8]. Our findings support this claim. However, we also demonstrate that other factors associated with behavioural residues of SNSs, such as graphics, external links, recommendation ratings and framing of wall postings, play an indirect role in affecting the credibility of eWOM. This is similar to Yu et al.’s [40] observation that “photo”, “link” and “status” messages are popular ways of posting media in the context of SNSs. The findings imply that SNS users take into account traceable behavioural residues that are generated from informants and their interaction with other connections on SNSs while judging the perceived credibility of eWOM. The relationships emphasize two critical issues for researchers: (1) more attention needs to be paid to the analysis of other potential factors of behavioural residues of SNSs in judgments based on eWOM reviews, and (2) such judgments of eWOM reviews need to be studied in greater detail by exploring other possible strategies of cue utilization that impact eWOM adoption.

Our findings suggest that the perceived credibility of eWOM is formed through the strategies of consistency-checking and knowledge-based validation (cue utilization). Extending these two epistemic strategies from the literature to judgments of eWOM credibility in a social networking environment adds value and provides different perspectives, which thus far have been overlooked in SNS research. SNS users who confirmed that an eWOM message was complete and consistent with other reviews generated higher perceived credibility toward that eWOM message: here, consistency-checking helped prevent them from accepting incorrect information. Likewise, SNS users who used their specific knowledge and experience to validate received eWOM messages developed higher perceived credibility toward those eWOM messages (knowledge-based validation). Furthermore, the findings imply that reorganizing users’ thoughts and knowledge toward a message (knowledge-based validation; β= 0.44) is a relatively more prominent feature associated with judging the credibility of eWOM than is consistency-checking (β= 0.21). This finding agrees with the argument that one’s subjective perception is most relevant in terms of predicting one’s response and futurebehaviour [14].

5.2 Practical implications

From a practical perspective, this study suggests that identity claims and behavioural residues affect a user’s judgment of an eWOM message through consistency-checking and knowledge-based validation. Thus, eWOM message informants and SNS designers should take our results into consideration in order to enhance user perceptions of credibility. For example, eWOM informants can look to provide strong, high-quality, and persuasive reviews to convince people of their credibility. Furthermore, to facilitate the initiation of users’ cue utilization, informants need to supplement their messages with clear graphics and useful links to support users’ in their judgment-making process, since messages with obscure photos and broken hyperlinks reduce users’ willingness to evaluate and believe such messages. For SNS designers, it may be worthwhile to design multiple means (e.g., the use of richer media) and technological tools for displaying eWOM reviews, so as to promote better expressiveness and increase the vividness and usefulness of that review. Alternatively, given that recommendation ratings and the framing of wall postings cannot easily be manipulated by informants, but may be more compelling than the things informants say about themselves and the content of their posting [16, 37], the former represent unbiased cues for user judgment. As such, the information collected from the tracking of interaction between a profile owner and his/her friends can be beneficial from the perspective of SNS users. But at the same time, this information can also generate opportunities for unethical abuse by invading the profile owner’s privacy. Designers need to develop appropriate mechanisms to ensure there is a balance between freedom of access to referable information on a profile owner’s wall and protection of the owner from the potential misuse of his/her status messages.

The results show that consistency-checking and knowledge-based validation affect the perceived credibility of eWOM, which ultimately affects eWOM adoption. Although SNS designers cannot control users’ consistency-checking and knowledge-based validation, they can incorporate mechanisms that shorten the time and effort required to undertake this checking and validation, and thereby enhance the effectiveness of eWOM assessment. For instance, designers can create ways for users, in their analysis of message content, to automatically search and compare functionality by means of graphical presentation of summary statistics on the majority of reviews; this will increase user convenience while comparing results. Such approaches can save users both time and effort in terms of retrieving related information from different sources, thereby making the process of eWOM judgment and adoptioneasier.

5.3 Limitations and future research directions

This study has several limitations. First, the results may have been impacted by selection bias given that we intentionally limited its scope, addressing only the process of adopting eWOM for currently-registered users on SNSs. Non-registered individuals might have different concerns about the influence of cue validity, cue utilization, credibility of eWOM, and eWOM adoption. Therefore, future research can explore the influential drivers of eWOM from the perspective of non-registered individuals. Second, we did not include any moderators in our model. Future research is encouraged to investigate other possible moderators (e.g., situational factors) that turn simple, key effects into more insightful conditional relationships [15]. Third, and finally, as the data are cross-sectional, all of the statistically supported relationships can only be viewed as tentative and associational. Future research could test the proposed model longitudinally to verify causal interrelationships among these constructs within SNSs.

Footnotes

, accessed 18 April 2014.

, accessed November 18, 2013.

Cohen [] suggested that the effect size f ² of 0.02, 0.15, and 0.35 are termed small, medium, and large, respectively.

Acknowledgments

This research was supported by the National Science Council (NSC) of Taiwan under grant number NSC 101-2410-H-032-008.

References

Anderson

J.C.

and Gerbing

D.W.

, Structural equation modeling in practice: A review and recommended two-step approach, Psychology Bulletin 3 (1988), 411–423.

Argyle

, Alkema

and Gilmour

, The communication of friendly and hostile attitudes by verbal and non-verbal signals., European Journal of Social Psychology 1(3) (1971), 385–402.

Baron

R.M.

and Kenny

D.A.

, The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations, Journal of Personality and Social Psychology 6 (1986), 1173–1182.

Belsley

D.A.

, Kuh

and Welsch

R.E.

Regression Diagnostics: Identifying Influential Data and Sources of Collinearity, John Wiley and Sons, New York 2004.

Brunswik

Perception and the Representative Design of Psychological Experiments, University of California Press, Berkeley, 1956.

Boyd

D.M.

and Ellison

N.B.

, Social network sites: Definition, history, and scholarship, Journal of Computer-Mediated Communication 13 (2008), 210–230.

Chatterjee

, Drivers of new product recommending and referral behaviour on social network sites, International Journal of Advertising 30(1) (2011), 77–102.

Cheung

M.Y.

, Luo

, Sia

C.L.

and Chen

, Credibility of electronic word-of-mouth: Informational and normative determinants of on-line consumer recommendations, International Journal of Electronic Commerce 13(4) (2009), 9–38.

Chin

W.W.

and Newsted

P.R.

Structural equation modeling analysis with small samples using partial least squares, Sage Publications, Thousand Oaks, 1999.

10.

Chu

S.C.

, Viral advertising in social media: Participation in Facebook groups and responses among college-aged users, Journal of Interactive Advertising 12(1) (2011), 30–43.

11.

Churchill

Marketing Research: Methodological Foundations, Dryden Press, Fort Worth, 1991.

12.

Cohen

Statistical Power Analysis for the Behavioral Sciences, Hillsdale, Erlbaum, 1988.

13.

Donath

and Boyd

, Public displays of connection, BT Technology Journal 22 (2004), 71–82.

14.

Fang

Y.H.

and Li

C.Y.

, Perceived deception in online auctions: Antecedents and consequences, Human Systems Management 33(4) (2014), 181–197.

15.

Featherman

and Fuller

Applying TAM to e-services adoption: The moderating role of perceived risk, Proceedings of the 36th Hawaii International Conference on System Sciences (HICSS-36), Hawaii, USA, 2003.

16.

Fogg

B.J.

Persuasive Technology: Using Computers to Change What We Think and Do, Morgan Kaufmann, San Francisco, 2003.

17.

Fornell

and Larcker

D.F.

, Evaluating structural equation models with unobservable and measurement error, Journal of Marketing Research 18 (1981), 39–50.

18.

Gallaugher

and Ransbotham

, Social media and customer dialog management at Starbucks, MIS Quarterly Executive 9(4) (2010), 197–212.

19.

Gifford

, The role of nonverbal communication in interpersonal relations, In: Horowitz

L.M.

and Strack

, eds., Handbook of Interpersonal Psychology: Theory, Research, Assessment, and Therapeutic Interventions, Wiley, New York, 2011.

20.

Gosling

S.D.

, Ko

S.J.

, Mannarelli

and Morris

M.E.

, A room with a cue: Personality judgments based on offices and bedrooms, Journal of Personality and Social Psychology 82 (2002), 379–398.

21.

Hair

J.F.

, Black

W.C.

, Babin

B.J.

and Anderson

R.E.

, Multivariate Data Analysis. Prentice Hall, Upper Saddle River, 2009.

22.

Hennig-Thurau

, Gwinner

K.P.

, Walsh

and Gremler

D.D.

, Electronic word-of-couth via consumer-opinion platforms: What motivates consumers to articulate themselves on the internet, Journal of Interactive Marketing 18(1) (2004), 38–52.

23.

Hlynka

and Welsh

, What makes an effective home page? A critical analysis, (1996), Retrieved from, http://www.umanitoba.ca/faculties/education/cmns/aect.html

24.

Hovland

C.I.

, Janis

I.L.

and Kelley

H.H.

, Communications and Persuasion: Psychological Studies in Opinion Change, Yale University Press, New Haven, 1953.

25.

Huabl

and Murray

K.B.

, Double agent, MIT Sloan Management Review 43(3) (2006), 8–12.

26.

M.X.

, Tan

C.H.

, Teo

H.H.

and Wei

K.K.

, Effects of product learning aids on the breadth and depth of recall, Decision Support Systems 53(4) (2012), 793–801.

27.

McClure

C.R.

, Wyman

S.K.

, Beachboard

J.B.

and Eschenfelder

K.R.

, Developing system-based and user-based criteria for assessing federal Websites, Journal of the American Society for Information Science 34 (1997), 78–88.

28.

Richter

and Schmid

, Epistemological beliefs and epistemic strategies in self-regulated learning, Metacognition Learning 5(1) (2010), 47–65.

29.

Ringle

C.M.

, Wende

and Will

, SmartPLS 2.0 (M3) Beta, Hamburg, Germany, 2005.

30.

Saraf

, Langdon

C.S.

and Gosain

, IS application capabilities and relational value in interfirm partnerships, Information Systems Research 18(3) (2007), 320–339.

31.

Shen

Y.C.

, Huang

C.Y.

, Chu

C.H.

and Liao

H.C.

, Virtual community loyalty: An Interpersonal- Interaction Perspective, International Journal of Electronic Commerce 15(1) (2010), 49–74.

32.

Tang

, Understanding the electronic word-of-mouth communication process: Communication effectiveness and analytic tools, Doctoral Dissertation, University of Arizona, Tucson, AZ, 2010.

33.

Tenenhaus

, Vinzi

V.E.

, Chatelin

Y.M.

and Lauro

, PLS path modeling, Computational Statistics and Data Analysis 48(1) (2005), 159–205.

34.

Tong

S.T.

, Van Der Heide

, Langwell

and Walther

J.B.

, Too much of a good thing? The relationship between number of friends and interpersonal impressions on Facebook, Journal of Computer-Mediated Communication 13 (2008), 531–549.

35.

Vandenbosch

and Higgins

, Information acquisition and mental models: An investigation into the relationship between behavior and learning, Information Systems Research 7(2) (1996), 198–214.

36.

Vazire

and Gosling

S.D.

, e-Perceptions: Personality impressions based on personal Websites, Journal of Personality and Social Psychology 87 (2004), 123–132.

37.

Walther

J.B.

, Van Der Heide

, Kim

S.Y.

, Westerman

and Tong

S.T.

, The role of friends’ appearance and behavior on eEvaluations of individuals on Facebook: Are we known by the company we keep, Human Communication Research 34 (2008), 28–49.

38.

Wathen

C.N.

and Burkell

, Believe it or not: Factors influencing credibility on the Web, Journal of the American Society for Information Science and Technology 53(2) (2002), 134–144.

39.

Winter

S.J.

, Saunders

and Hart

, Electronic window dressing: Impression management with Websites, European Journal of Information Systems 12 (2003), 309–322.

40.

, Chen

and Kwok

, Toward predicting popularity of social marketing messages, in: Proceedings of the 2011 International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction, Springer-Verlag, Berlin, Heidelberg, 2011.

41.

Zhang

and Watts

S.A.

, Capitalizing on content: Information adoption in two online communities, Journal of the Association for Information Systems 9(2) (2008), 73–94.

42.

Zhang

, Keeling

and Pavur

, Information quality of Commercial website home pages: An explorative analysis, in: Proceedings of the 21st International Conference on Information Systems, Brisbane, Australia, 2000.

43.

Zhao

, Lynch

J.G.

, and

Jr.

and Chen

, Reconsidering Baron and Kenny: Myths and truths about mediation analysis, Journal of Consumer Research 37(2) (2010), 197–206.