BISE – RESEARCH PAPER

In eWOM We Trust

A Framework of Factors that Determine the eWOM Credibility

Electronic word-of-mouth (eWOM) as an instrument of marketing communicationinfluences many purchasing decisions. The paper identifies major determinants ofcredibility from a consumer’s point of view. Drawing on dual process theory and sourcemodels, hypotheses are derived and tested. The paper provides evidence that expertise,trustworthiness, and aggregate rating have a positive impact on online recommendationcredibility. The study also demonstrates that involvement could moderate theserelationships.

DOI 10.1007/s12599-013-0261-9

The Author

Dr. Bettina Lis, AssistantProfessor (�)Media ManagementJohannes Gutenberg-University55128 MainzGermanylis@uni-mainz.de

Received: 2012-07-12Accepted: 2013-01-26Accepted after two revisions byProf. Dr. Spann.

Published online: 2013-05-08

This article is also available in Ger-man in print and via http://www.wirtschaftsinformatik.de: Lis B (2013)In eWOM We Trust. Ein Modell zurErklärung der Glaubwürdigkeit voneWOM. WIRTSCHAFTSINFORMATIK.doi: 10.1007/s11576-013-0360-8.

© Springer Fachmedien Wiesbaden2013

1 Introduction

In purchase decisions, consumers in-creasingly look for information on prod-ucts and services in the internet (Lee et al.2008). Results of representative studiesshow that one third actively uses online-recommendations for information onproducts and services (Heckathorne2010; Nielsen et al. 2010). People, whobased their purchase decisions on ad-

vertising and professional advice a fewyears ago, now more and more refer torecommendations of online users (Leeet al. 2008). The exchange of provider in-formation between consumers via inter-net is called electronic Word-of-Mouth(eWOM) (Hennig-Thurau et al. 2004).An eWOM recommendation is charac-terized by a positive, neutral, or negativeprovider-relevant piece of informationpublished in the internet by a consumer(Rafaeli and Raban 2005). As informa-tion thus spreads exponentially and atlow cost, eWOM communication is animportant factor for businesses.

Whereas traditional Word-of-Mouth(WOM) shows a direct connection be-tween sender and receiver with a signif-icant tie strength, eWOM is character-ized by indirect and mostly public com-munication with normally no social con-nection between the sender of a mes-sage and the receiver (Godes and May-zlin 2004; Hennig-Thurau et al. 2004).Thus each consumer may issue and re-ceive recommendations worldwide at anytime. The assessment of a recommenda-tion by the reader however proves dif-ficult (Smith et al. 2007). Prior stud-ies show that credibility is especially im-portant for the final valuation of elec-tronic consumer recommendations: thehigher the credibility of an online rec-ommendation, the more likely it is thatthe receiver follows the sender’s productrecommendation (Wathen and Burkell2002). Whereas the positive effects ofcredibility on eWOM adoption couldbe sufficiently confirmed (Cheung et al.2009), little is known about the deter-minants of eWOM credibility (McKnight

and Kacmar 2006). Thus in many stud-ies credibility is given as an explanationfor the effects of eWOM communication(e.g., Cheung et al. 2008), but credibilityitself seldom is the object of research.

Studies on eWOM have up until nowhad their focus mainly on the success ofa product (Chevalier and Mayzlin 2006;de Bruyn and Lilien 2008), the motivesfor publishing eWOM (Hennig-Thurauet al. 2004; Walsh and Mitchell 2010), andan optimal seeding strategy (Berger andMilkman 2012; Hinz et al. 2011). An earlyconceptual study on eWOM credibilityis given by Wathen and Burkell (2002).They indicate the multi-dimensionalityof credibility in the context of eWOM andgive a theoretical model as a basis for fur-ther empirical research. E.g., Brown et al.(2007) emphasize credibility, besides tiestrength and homophily, as the key vari-able for the assessment of eWOM com-munication. Concrete identification ofcredibility determinants is not provided,however. Another study of Mackiewicz(2008) explains credibility, due to theanonymity in the eWOM context, solelyby the linguistic quality of consumerrecommendations, whereas O’Reilly andMarx (2011) analyze the credibility ratingmainly against the background of techni-cal aspects. A further contribution comesfrom Cheung et al. (2009). In their analy-sis they give two determinants of eWOMcredibility: the strength of the recom-mendation and its value in the context(of the product) (Cheung et al. 2009).As their study is however confined to theChinese area, the authors suggest to ex-pand the study to individualistic cultures“(. . . ) to permit cross-cultural compar-ison of the relative Silbentrennung im-

Business & Information Systems Engineering 3|2013 129

BISE – RESEARCH PAPER

pacts of normative influence on eWOMacceptance” (Cheung et al. 2009, p. 31).This is where the present study comes in.Significant determinants of eWOM cred-ibility are to be defined from the viewof the consumer. Here credibility in dis-tinction to Brown et al. (2007) is notexamined as a whole, but its particulardeterminants are defined in detail. Thefocus is less on linguistic (Mackiewicz2008) and technical aspects (O’Reillyand Marx 2011) than on determinantsderived from the source models. Thusthe source-credibility model by Hovlandet al. (1953) and the source-attractivenessmodel by McGuire (1985) explain thebasic credibility determinants and theirmode of function in the classic WOMcontext. The present study extends ex-isting knowledge by examining whetherthe classic (informational) determinantsof the source models can be transmittedto the online context. Furthermore it canbe inferred, in line with the dual processtheory of Deutsch and Gerrard (1955),that not only informational factors havean influence on credibility, but also nor-mative ones, which are also to be ana-lyzed in the present study (Cheung et al.2009).

In sum, the present study amends exist-ing literature in the following points. Re-search on eWOM credibility so far showsno integrated empirical or theoretical sat-uration. Considering this, the objectiveof this contribution is to supplement thepresent discussion on eWOM with thetopic of eWOM credibility, which has sofar not been sufficiently researched. Onthe one hand, the focus is on regardingcredibility as a central construct. On theother hand, conceptual and operationaldetails of the determinants of eWOMcredibility are described. In this expan-sion of existing literature the aim is notto gain an isolated view of singular fac-tors (as linguistic and technical aspects).Instead, its degree of novelty lies in the in-tegrated view of various credibility deter-minants in a comprising research model,so far not analyzed in this context. Thusthis study aims at showing eWOM cred-ibility as a theoretical construct, consist-ing of two dimensions, and at putting thetwo dimensions into operation by corre-sponding determinants. In addition thisstudy contributes to the need of researchin the question of the influence of norma-tive factors on eWOM credibility, as men-tioned by Cheung et al. (2009). Whereasstudies so far only focus on informa-tional determinants of eWOM credibil-ity (e.g., Mackiewicz 2008), this paper

amends previous analyses by includingnormative determinants in the study.

The subject of this research are onlinerecommendations, as they are a widelyspread form of eWOM accepted by theusers (Hennig-Thurau et al. 2004). Theyrepresent so-called “consumer-created”information found in various contexts: inopinion platforms (as epinion.com), fo-rums, blogs, or as integrated part of anonline shop (as amazon.com) (Chen andXie 2008). They are characterized by in-formal, interpersonal, normally not com-mercially oriented, product-related com-munication of an unspecified number ofpersons (Chen and Xie 2008).

Drawing on dual process theory andsource models relevant factors of the per-ceived eWOM credibility will be identi-fied and their relevance for influencingcredibility will be examined in the follow-ing. Starting from theoretical considera-tions, hypotheses on the effects of par-ticular determinants will be deduced andvalidated by structural equation model-ing based on a sample of 643 subjects.The results show that the sender’s exper-tise and trustworthiness as well as the ag-gregated rating are significant factors inthe perceived credibility of online rec-ommendations. Additionally it becomesclear that involvement moderates thestrength of this relation.

2 Theoretical Framework andHypotheses

2.1 Source Credibility

Central significance for credibility re-search has to be granted to the conceptof source and context oriented rating ofcredibility. The interest here concentrateson the characteristics of a communica-tion source as perceived by the receiver(Hovland et al. 1953). The central idea ofit being an attribution and not an inher-ent quality of texts is mirrored in quite afew definitions. For example, Tseng andFogg (1999) see credibility as a “perceivedquality (. . . ) it does not reside in an ob-ject, a person, or a piece of information(. . . )” (Tseng and Fogg 1999, p. 40).

A suitable approach to define partic-ular determinants of credibility is of-fered by source models. They deter-mine the conditions under which thesender or the source of a message ap-pears credible (McCracken 1989). Thesource-credibility model of Hovland et al.

(1953) assumes that information origi-nating from a credible source influencesattitudes, opinions, and conduct of thereceiver. The credibility of a message ishere determined by two dimensions: theexpertise given to the sender owing to hisspecial competence, and his trustworthi-ness, describing the objectivity and sin-cerity of the sender (Hovland et al. 1953).Credibility according to Hovland et al.(1953) thus is a function with the di-mensions “expertise” and “trustworthi-ness”. The only relevant factor for the rat-ing of these parameters is the subjectiveperception of the receiver.

Following the source-attractivenessmodel of McGuire (1985), the credibilityof a message additionally depends uponthe attractiveness of the communicator.Central quality of source-attractivenessis the similarity or social homophily re-spectively between sender and receiver(von Wangenheim and Bayón 2004). Inconsequence, the model assumes that asource is attractive for the sender andthus credible if it closely resembles him(McGuire 1985). The three-componentsmodel of Ohanian (1990, 1991) finallycombines the two models of Hovlandet al. (1953) and McGuire (1985) anddeclares the factors expertise, trustwor-thiness and homophily to be the essen-tial determinants of message credibility(Ohanian 1990).

The source models are criticized on thegrounds of their assumption that just thesource itself is decisive for the effective-ness of a message. Influence of third par-ties is not taken into account. In line withthe dual process theory of Deutsch andGerrard (1955), however, it can be as-sumed that not only informational in-fluences have a bearing on credibility,but that there are also normative ones(Deutsch and Gerrard 1955). Whereasinformational factors of influence referto the information and arguments ex-changed during the discussion, norma-tive factors refer to the efforts of mem-bers of the group to remain conform tothe other members of the group and tobe rated positively (Deutsch and Gerrard1955). Normative influence thus exists assoon as the sender has access to the opin-ions and views of others (Kaplan andMiller 1987).

The dual process theory of Deutschand Gerrard (1955) can be positionedin social psychology and so far has beenmainly used when examining the credi-bility of information in physical scenarios

130 Business & Information Systems Engineering 3|2013

BISE – RESEARCH PAPER

(Cohen and Golden 1972; von Wangen-heim and Bayón 2004). It can be classifiedto belong to the block of dual process the-ories in company with the elaboration-likelihood model (Petty and Cacioppo1986) and the heuristic-systematic model(Eagly and Chaiken 1993). These eachpostulate two kinds of information pro-cessing which differ in the extent to whichindividuals associate with the argumentsof a message. The sender is describedhere as a peripheral stimulus. The the-ory of Deutsch and Gerrard (1955) ap-pears to be significant for this study, asbeside the classical informational dimen-sions of credibility also normative factorsare taken into account. It also lends itselfas a superior concept to the classificationof the particular credibility dimensionsdescribed in the following.

2.2 Informational Determinants

2.2.1 Source Expertise

Following the source-credibility model,the perceived expertise of the sender isa significant determinant of credibility(Hovland and Weiss 1951). This exper-tise can be defined “as the extent to whicha person is perceived to possess knowl-edge, skills or experience and thereby isconsidered to provide accurate informa-tion” (Ohanian 1990, p. 44). It refers tothe knowledge of a sender on a prod-uct or a service. A receiver will probablyturn directly to a sender whom he consid-ers knowledgeable and experienced (Yaleand Gilly 1995). He presumes that thesender has substantial and especially use-ful information due to his high expertise(Bansal and Voyer 2000).

Empirical studies demonstrate that in-formation provided by experts has a greatimpact on the receiver. Bone (1995), Yaleand Gilly (1995), as well as Gilly et al.(1998), for example, established for clas-sical WOM that in the rating of prod-ucts the influence on the receiver in-creases when the WOM originates from asender with high expertise. In the contextof services, Bansal and Voyer (2000) andalso von Wangenheim and Bayón (2004)show a positive correlation between theexpertise of the sender and its impact onthe receiver. Following the tenor of thesource-credibility model, it can be pre-sumed that the sender’s expertise alsoforms a relevant factor in the rating ofcredibility in online recommendations(Bansal and Voyer 2000). Thus receiverschoose senders of high expertise as they

expect them to provide highly qualifiedinformation (McCracken 1989). A senderwith high expertise appears more credi-ble, since the receiver has little cause todoubt the correctness of this informationdue to the knowledge and the compe-tence of the communicator (Kroeber-Rieland Weinberg 2003, p. 504c.). This pre-sumption is supported by the theoreti-cal study of Wathen and Burkell (2002).Additionally, experts often possess morepower of persuasion. Due to their exten-sive knowledge and experience, expertsshould be better able to convince otherconsumers and thus appear more credi-ble (von Wangenheim and Bayón 2004).So we find:

H1: The higher the reviewer’s level of ex-pertise, the more his or her online recom-mendations will be perceived as credible.

2.2.2 Source Trustworthiness

Along with expertise, the source-credibility model names perceived trust-worthiness of the sender as another de-terminant of credibility (Hovland et al.1953). The credibility of informationcoming from a trustworthy source isdoubted less by the receivers than oneof an origin considered not trustworthy(Sparkman and Locander 1980). Herecredibility and trustworthiness are rela-tional constructs and therefore requireat least two actors. Whereas credibil-ity describes a comprehensive relationalprocess, trustworthiness refers to certainaspects within this relation (Hovlandet al. 1953). A reviewer and thus his rec-ommendation is acknowledged as trust-worthy if the statement is judged valid,honest, and to the point (Hovland andWeiss 1951). The issue therefore is thedegree of objectivity and sincerity thesender is granted. The construct trust-worthiness here is closely related to theidea of trust (McKnight and Chervany2002). Whereas trustworthiness relates tothe cognitive-affective component (trust-ing beliefs, Jones 1996), trust refers to theaspect of behavior in the form of will-ingness or intention to rely on a differentperson (trusting intentions, Büttner andGöritz 2008).

With regard to the source-credibilitymodel it can be assumed that the re-viewer’s trustworthiness plays a role inassessing eWOM credibility. A trustwor-thy reviewer showing a high degree of ob-jectivity and sincerity appears more cred-ible, as the receiver has no cause to ques-tion the validity of the given informa-tion. Therefore it is more likely that the

receiver will rely on the transmitted in-formation in the case of high trustwor-thiness and find it more influential andcredible than in the case of low trust-worthiness (Huang and Chen 2006). Em-pirical studies in the context of classi-cal WOM support this assumption. E.g.,Wilson and Sherrell (1993) confirm thispositive effect on the change of atti-tude when the source is rated as trust-worthy. As, however, consumers in thecontext of eWOM normally cannot di-rectly judge whether a recommendationis trustworthy or not, they use indirectmethods such as evaluating the consis-tence of the arguments or the objectivityof the contents. Therefore:

H2: The higher the reviewer’s level oftrustworthiness, the more his or her on-line consumer recommendations will beperceived as credible.

2.2.3 Social Homophily

Besides expertise and trustworthiness afurther factor appears important forthe assessment of credibility: social ho-mophily (Miller and Hoppe 1973). Socialhomophily or similarity between senderand receiver emerges as a central com-ponent from the source-attractivenessmodel (McGuire 1985). The constructdescribes the similarity of two individualsconcerning particular attributes (Rogers1983). Social homophily can be differ-entiated according to demographic (age,gender, education, occupation) and/orperceived attributes (values, preferences)(Lazarsfeld and Merton 1964; Gilly et al.1998). Concerning the emergence of ho-mophily there are significant differencesbetween an 3 online and offline con-text. Due to reduced information the rat-ing in an online context basically resultsfrom the contents of the website. Ac-cording to Gilly et al. (1998), the demo-graphic determinants such as gender orsocio-economic status are of less impor-tance than the perceived attributes suchas similar values (Blanton 2001) or pref-erences (Brewer and Webber 1994). Inthe text of reviews consumers look forvalues and experiences matching theirown character and ideas. If a recommen-dation contains such information and thereader senses similar values and prefer-ences, this leads to an increased perceivedhomophily (Blanton 2001).

According to the source-attractivenessmodel it can be assumed that social ho-mophily is also significant for the credi-bility rating of online recommendations.Receiving and viewing a viral message

Business & Information Systems Engineering 3|2013 131

BISE – RESEARCH PAPER

causes an emotional reaction on the re-cipient’s side (Gilly et al. 1998). Therecipient is motivated to validate theseemotions by social comparison. A senderwith a perceived high affinity is morelikely to give reason for the accuracy ofhis emotions and thus is rated as morecredible (Gibbons and Gerrard 1991).Empirical studies in the field of classi-cal WOM support this assumption. Theyshow that a greater homophily betweensender and receiver has a positive ef-fect on the sender’s influence (Gilly et al.1998). Homophile sources are more fre-quently used in a consumer decision (vonWangenheim and Bayón 2004). The so-called “like-me” principle is a funda-mental concept of human communica-tion. Thus individuals tend to interactwith other individuals who are similar tothem (Laumann 1966, p. 29). Thereforea transmission of ideas and informationhappens more often between individualsof high homophily (Rogers 1983). Thusfor eWOM follows:

H3: The higher the perceived homophilybetween the reviewer and the reader, thehigher the perceived credibility of theonline consumer recommendations willbe.

2.3 Normative Determinants

2.3.1 Aggregated RecommendationRating

In line with the dual process theory, onemay assume that not only informationalinfluences have bearings on the credibil-ity 3 of online recommendations, but thatalso normative ones exist (Deutsch andGerrard 1955). Normative influence is al-ways given when the sender has access tothe views and opinions of different peo-ple (Kaplan and Miller 1987). In the on-line context, recommendation rating isa possibility to represent the opinion ofothers. Online forums allow users to eval-uate the contents of recommendationsaccording to quality, utility, and so on.Recommendation rating is the productof a multitude of singular ratings (Qiuand Li 2010). When reading for examplea negatively rated recommendation, therecipient is more likely to question thistext and doubt its credibility (Qiu and Li2010).

In this context, group pressure in linewith the conformity thesis of Asch (1951)plays a decisive role. If a person approvesof the group opinion, this reduces the

feeling of dissonance. Groups perform anormative influence, which means thereis little dispute of contents but rather apoorly reflected takeover. When individ-uals in a certain situation have no ac-cess to complete information, others maybecome information sources. Here thegroup opinion receives more credibilitythan an individual opinion (Asch 1951).Conformity is achieved by following themajority’s opinion in order to avoid per-sonal insecurity (Asch 1951). FollowingAsch (1951) it can be assumed that therecipient of a recommendation believesit more credible when the contents arerated positive by other users.

H4: The more positive the rating of therecommendation, the higher the perceivedcredibility of the online consumer recom-mendation will be.

2.4 Moderating Effect of Involvement

Besides the shown direct effects, fur-ther relational structures can exist whichmoderate the force of effectual relationsin dependence on various conditions. Inliterature the moderating role of involve-ment is of central importance (Petty andCacioppo 1986). E.g., the elaboration-likelihood model (ELM) indicates thatthe motivation for information engineer-ing is essentially determined by involve-ment (Petty and Cacioppo 1986). Gen-erally involvement can be defined as apersonally perceived relevance of an ob-ject based on the own needs, values,and interests of a person (Zaichkowsky1985). Involvement depends on the in-ner participation or respectively the men-tal engagement of a consumer regard-ing an object (Celsi and Olson 1988).Thus consumers with low involvementalso experience low need for informa-tion, whereas consumers of high in-volvement extensively look for informa-tion which provides added informationalvalue (Zaichkowsky 1985).

From this, two central findings canbe deduced for this study. The simpleheuristic equalization “positive sender at-tributes = higher credibility” does nothold beyond contexts. Rather, the rel-evance of particular sender attributesvaries in dependence on involvement(de Bruyn and Lilien 2008). If one fol-lows the argumentation of ELM above,highly involved receivers are specially af-fected by the communicator’s character-istics which possess an added informa-tional value. According to de Bruyn and

Lilien (2008) it can be assumed that theattributes of the sender in eWOM com-munication normally are processed withhigh involvement. Consequently, a differ-entiated view on the influence of the par-ticular determinants and on their effectson eWOM credibility is required.

As shown, the sender’s expertise is adecisive factor for influencing a receiverin the eWOM context. Principally, thestrong influence of a sender with highexpertise results from the availability ofquantitatively and qualitatively high-endinformation. As highly involved receiversaccording to ELM mainly look for infor-mation with large added informationalvalue, and senders with high expertise areinfluential due to the quality of their in-formation, it can be assumed that highlyinvolved receivers are more strongly in-fluenced than those with low involve-ment. This argumentation is supportedby von Wangenheim and Bayón (2004)as well as Petty et al. (1983). Petty et al.(1983) demonstrate that both purchaseintention and attitude of highly involvedconsumers in regard to a product are in-fluenced more strongly by the expertiseof the sender. For the context of eWOMwe can assume:

H5a: Involvement strengthens the rela-tionship between expertise and the per-ceived credibility of online consumer rec-ommendations.

In connection with the trustworthi-ness it can however be assumed thatqualitative aspects of information havea lower priority (Leonard-Barton 1985).When assessing trustworthiness, partic-ipants prefer for example the objectiv-ity of the statements (Ohanian 1990).Senders of high trustworthiness seem toexert their influence less by the qualityof their information but mainly due toits perceived reliability. Following the ar-gumentation of ELM it can be assumedthat highly involved receivers are less in-fluenced by the sender’s trustworthinessthan those with low involvement.

H5b: Involvement weakens the relation-ship between trustworthiness and theperceived credibility of online consumerrecommendations.

In addition, it can be argued accordingto von Wangenheim and Bayón (2004)that the positive relation between so-cial homophily and credibility is strongerwith highly involved receivers. The rea-son for this is that even though source

132 Business & Information Systems Engineering 3|2013

BISE – RESEARCH PAPER

Fig. 1 Research model

credibility has priority for the highly in-volved receiver, the sender should alsobe homophile to the receiver. Follow-ing the source-attractiveness model it canbe assumed that the recommendation ofa sender with homophile attitudes andvalues is more relevant as he is seenas a valid source for the satisfaction ofone’s own needs (Ohanian 1990). A ho-mophile sender will be assumed to havesimilar preferences and to convey infor-mation with an emotional or informa-tional added value to the recipient. Priceand Feick (1984) argue similarly whenthey say that communication betweenhomophile senders is especially effectiveand simple due to similar values.

H5c: Involvement strengthens the rela-tionship between homophily and the per-ceived credibility of online consumer rec-ommendations.

Finally Lord et al. (2001) show thatin purchase situations characterized by ahigh level of involvement, the main in-fluence is informational and not norma-tive. Highly involved individuals are gen-erally less susceptible to third parties inregard to their opinions and views (Sherifand Hovland 1965). For the most part,individuals of high involvement have def-inite and strongly anchored positions andopinions. In the context of this study itcan be assumed that high involvementweakens the relation between recommen-dation rating (as a normative measure ofinfluence) and credibility.

H5d: Involvement weakens the relation-ship between recommendation rating andthe perceived credibility of online con-sumer recommendations.

2.5 Effect on Perceived eWOMCredibility on eWOM Adoption

A successful eWOM communication pro-cess is concluded by the eWOM adop-tion, that is, the acceptance of the recom-mendation of the relevant review (Suss-man and Siegal 2003). Several studiesshow that a basic requirement for theadoption of eWOM is its credibility (e.g.,McKnight and Kacmar 2006). Thus re-cipients adopt a recommendation froma reliable source more readily than onewhich they estimate as unreliable (Bansaland Voyer 2000). Petty et al. (2002) con-firm this by varying the credibility of asender in an experiment: When they de-scribed the source as credible, the recip-ients for the most part did not doubtit and adopted the information imme-diately. In addition, several studies (e.g.,Petty et al. 1983; Ohanian 1990) allow theconclusion that credible sources lead toa more positive attitude and acceptanceof the described object on the part of therecipient than less credible ones. Directeffects of credibility could be shown forpurchase intention (Hu et al. 2008) andalso for information adoption (Cheunget al. 2009). ELM in particular serves asa theoretical support due to the informa-tion processing postulated (Bhattacherjeeand Sanford 2006). Thus ELM was usedas a theoretical explanation for the in-formation adoption of their test subjectsby Sussman and Siegal (2003). It can beassumed:

H6: The higher the perceived credibilityof online consumer recommendations, themore likely they are to be adopted.

Figure 1 presents the research frame-work. It includes all of the informationaland normative determinants that explainthe consumer’s process of forming con-sumers’ perceived credibility of onlinerecommendations.

3 Methodology

3.1 Sample and Data Collection

To verify the described hypotheses, anonline survey was used. A first pretestamong 18 students of social and eco-nomic studies aged 20 to 28 resultedin no objections. The main study tookplace from March to June 2011 with aninterrogation of 2000 users of a lead-ing online consumer discussion forumwhich remains undisclosed for reasonsof discretion. It is an online communitywhose main contents are user-oriented,site-related ratings on a local basis. Usersrate businesses, locations, and services(e.g., hotels, restaurants, or fitness stu-dios). The entries include personal as-sessments and recommendations fromusers for users. The provider offers a sys-tem which indicates the “online repu-tation” of the users and the possibilityto rate the recommendations of othermembers.

All in all, 634 test subjects participatedin the survey, which equals a 34 % rateof return. The relation between gendersis nearly balanced: 45 % men, 55 %women. As to occupations, the threestrongest groups were clerks with 52 %,students 38 %, and pensioners 10 %.As for formal education, the “Abitur”

Business & Information Systems Engineering 3|2013 133

BISE – RESEARCH PAPER

Table 1 Constructmeasurement items Construct Source Cronbach’s α Instruments

Expertise (EXP) Ohanian (1991) 0.856 The reviewer is an expert

The reviewer is experienced

The reviewer is knowledgeable

The reviewer is qualified

The reviewer is skilled

Trustworthiness(TRUST)

Ohanian (1991) 0.901 The reviewer is undependable

The reviewer is honest

The reviewer is reliable

The reviewer is sincere

The reviewer is trustworthy

Homophily(HOMO)

McCroskey et al.(1974);

0.912 The reviewer is very similar to me/is verydifferent from me

McCroskey andYoung (1981)

The reviewer thinks a lot like me/doesn’t thinklike me at all

AggregatedRecommendationRating (AGG)

Cheung et al. (2009) 0.879 Based on the review rating, the review wasfound to be favorable by other audiences

Based on the review rating, the review ishighly rated by other audiences

Based on the review rating, the review is good

Perceived eWOMCredibility(CRED)

Cheung et al. (2009) 0.904 I think the review is factual

I think the review is accurate

I think the review is credible

eWOM ReviewAdoption(ADOP)

Cheung et al. (2009) 0.898 To what extent do you agree with the review?

Information from the review contributed tomy knowledge of the product discussed

The review made it easier for me to make mypurchase decision

The review has enhanced my effectiveness inmaking a purchase decision

The review motivated me to take purchasingaction

Involvement(INV)

Zaichkowsky (1985) 0.911 I am interested in online recommendations

I always wanted to know more about onlinerecommendations, so I appreciate if friendsgive me some explanations

Online recommendations are a hobby of mine

Online recommendations are important to me

(university entrance examination) repre-sented 40 %, the “Fachabitur” (techni-cal diploma) 28 %, and the “Realschu-labschluss” (certificate of secondary edu-cation) 22 %. The average age was 34.5years. Nearly all of the interviewees hadan experience with the internet of threeyears or more. 84.4 % use the internet formore than one hour per day.

3.2 Measures

Variables were measured through seven-point Likert scales, ranging from stronglyagree (1) to strongly disagree (7). In allcases, items were extracted from previ-ous research. First, expertise and trust-

worthiness were measured with the itemsproposed by Ohanian (1991). To mea-sure social homophily, the approachesof McCroskey et al. (1974) and Mc-Croskey and Young (1981) respectivelywere employed. The measurement of therecommendation rating was shaped ac-cording to the scale of Cheung et al.(2009). Measuring eWOM credibility isbased on the inventory of Cheung et al.(2009). To determine the involvement thescale of Zaichkowsky (1985) was used.In the context of this study the on-line recommendations refer to productsand services respectively. Involvement inthis context concerns the descriptionsof products and services in these rec-

ommendations. Finally, eWOM adoptionwas assessed through five items used byCheung et al. (2009). It proved to be es-pecially suited to these consumer recom-mendations. Table 1 provides a completelist of the measurement items.

3.3 Common Method Bias

As the dependent and independent vari-ables were taken for the same per-son, potentially the problem of Com-mon Method Bias may arise (Pod-sakoff and Organ 1986). Therefore thedata raised have to be checked ac-cordingly. With reference to Podsakoffet al. (2003), Harman’s single-factor test

134 Business & Information Systems Engineering 3|2013

BISE – RESEARCH PAPER

Table 2 Results ofdiscriminate validityanalysis

EXP TRUST HOMO AGG CRED ADOP INV

EXP 0.683

TRUST 0.338 0.670

HOMO 0.220 0.391 0.731

AGG 0.146 0.494 0.130 0.770

CRED 0.375 0.384 0.183 0.338 0.790

ADOP 0.145 0.124 0.171 0.128 0.301 0.782

INV 0.153 0.138 0.139 0.143 0.131 0.257 0.832

Table 3 Results ofhypotheses testing(∗ p < 0.1; ∗∗ p < 0.05;∗∗∗ p < 0.01)

Hypothesesbasic model

Standardizedestimates

t-value

H1 (EXP→ CRED) 0.449∗∗∗ 5.523

H2 (TRUST → CRED) 0.479∗∗∗ 6.224

H3 (HOMO→ CRED) 0.021 0.756

H4 (AGG→ CRED) 0.338∗∗∗ 3.729

H6 (CRED→ ADOP) 0.750∗∗∗ 8.126

was executed both with an explorative(EFA) and a confirmatory factor anal-ysis (CFA). The EFA-approach of theunrotated principal-components analy-sis shows seven factors with an intrin-sic value of larger than 1 (Kaiser crite-rion). The factor with the highest intrin-sic value covers only 26.5 % of the wholevariance. The danger of common methodbias can thus be considered low. In ad-dition, CFA was used. Here the qualityof the one-factor solution of CFA wascompared to the quality of the measure-ment model used (Podsakoff et al. 2003).The results show that the quality of theone-factor model is significantly lower(χ2/d.f . = 6.65; RMSEA = 0.225; CFI =0.70; GFI = 0.61; NFI = 0.59). There-fore a substantial common method biascannot be assumed.

4 Results

4.1 Validity of Measures

The structural model shown in Fig. 1was estimated using the maximum like-lihood algorithm with AMOS version 19.The use of structural equation models re-quires not only the specification of theapplication, the estimation of the param-eters, but also the listing of suitable cri-teria to rate the quality of the specifiedmodel

The validity of the model was as-sessed using traditional methods. Ta-ble 2 presents the correlations among theframework’s variables. Overall, the mea-surement scales show sufficient reliabil-

ity and validity; more specifically, for allconstructs the composite reliability ex-ceeds the threshold value of 0.6 (Bagozziand Yi 1988). All coefficient alpha val-ues exceed the threshold value of 0.7 rec-ommended by Nunnally (1978). All thefactor loadings are significant (p < 0.01),which Bagozzi and Yi (1988) suggest asa criterion of convergent validity. Fur-thermore, item reliabilities are above therecommended value of 0.4 (Bagozzi andYi 1988). The discriminate validity ofthe construct measures was assessed onthe basis of Fornell and Larcker’s (1981)criterion which indicates that discrimi-nate validity is supported if the averagevariance extracted exceeds the squaredcorrelations between all pairs of con-structs. Table 2 indicates that each ex-plained variance estimate on the diag-onal is greater than the correspondingsquared inter-factor correlation estimatebelow the diagonal. All constructs ful-filled this requirement, which suggeststhat their degree of discriminate validityis sufficient.

The measures of overall fit meet con-ventional standards, suggesting that themodel fits the data well: χ2/d.f . =3.806, root mean square error of approx-imation [RMSEA] = 0.066, standardizedroot mean square residual [SRMR] =0.036, normed fit index [NFI] = 0.983,goodness of fit index [GFI] = 0.90, andcomparative fit index [CFI] = 0.926.

As a further rating criterion, the ratioof chi-square test size and number of de-grees of freedom was brought in. Here,too, the model shows an acceptable fit on

the whole (χ2/d.f . = 3.806) (Carminesand McIver 1981).

4.2 Results on the Level of the StructuralModel

4.2.1 Results of the Basic Model

Table 3 shows the standardized estimatesof the model tested.

The results confirm strong positive re-lationships between expertise and credi-bility γ = 0.45, p < 0.01), trustworthi-ness and credibility (γ = 0.48, p < 0.01),and recommendation rating and credi-bility (γ = 0.34, p < 0.01); therefore, H1,H2, and H4 are supported. Homophily,however, did not have a significant effecton credibility γ = 0.02, not significant[n.s.]); therefore, H3 is not supported.Finally, credibility is a strong predictorof adoption (γ = 0.75, p < 0.01), whichsupports H6.

4.2.2 Interaction Effects of Involvement

To test H5a to H5d, which refers to themoderating role of involvement, hierar-chical multiple regression analyses – alsocalled moderated regression – was used(Aiken and West 1991, pp. 49). Here con-nections influenced by the existence ofone or several additional predictor vari-ables of an independent predictor anda dependent criterion are examined. Forthis study, we focused on the questionwhether force and direction of the rela-tion between predictor variables (exper-tise, trustworthiness, social homophily,

Business & Information Systems Engineering 3|2013 135

BISE – RESEARCH PAPER

Table 4 Results of hierarchical regres-sion analysis: Moderating effect of in-volvement on expertise-credibility rela-tionship (∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p <

0.01)

Predictors Model 1β

Model 2β

Expertise (EXP) 0.375∗∗∗ 0.391∗∗∗

Involvement (INV) 0.031 0.014

EXP × INV 0.253∗∗∗

R2 0.09∗∗∗ 0.10∗∗∗

�R2 0.01∗∗∗

Table 5 Results of hierarchical re-gression analysis: moderating effectof involvement on trustworthiness-credibility relationship (∗ p < 0.1;∗∗ p < 0.05; ∗∗∗ p < 0.01)

Predictors Model 1β

Model 2β

Trustworthiness(TRUST)

0.384∗∗∗ 0.395∗∗∗

Involvement(INV)

−0.094 −0.051

Trust × INV 0.239∗∗

R2 0.08∗∗∗ 0.13∗∗∗

�R2 0.05∗∗∗

and recommendation rating) and crite-rion variables (credibility) vary depend-ing on the characteristics of the mod-erator variables (involvement). To testthis relationship each predictor variablewas initially centered (converted into de-viation score form) to minimize multi-collinearity, and interaction terms wereformed as the product of the centeredpredictors (Aiken and West 1991, pp. 49).The results of the moderated regressionanalysis are presented in Tables 4, 5, 6and 7.

As shown in Table 4, expertise and in-volvement were able to explain 9 % ofthe variance in eWOM credibility (p <

0.001). The addition of the interactionterm produces a significant incrementin the amount of variance explained incredibility (�R2 = 0.01; p < 0.01) in-dicating that involvement moderates theexpertise-credibility relationship.

The moderating effect of involvementon the trustworthiness-credibility rela-tionship was also examined using a hi-erarchical regression analysis. As shownin Table 5, trustworthiness and involve-ment were able to explain 8 % of

Table 6 Results of hierarchical regres-sion analysis: moderating effect of in-volvement on homophily-credibility re-lationship (∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p <

0.01)

Predictors Model 1β

Model 2β

Homophily (HOMO) 0.183 0.187

Involvement (INV) 0.021 0.075

HOMO × INV 0.212∗∗

R2 0.09∗∗∗ 0.13∗∗∗

�R2 0.04∗∗∗

Table 7 Results of hierarchical regres-sion analysis: moderating effect of in-volvement on recommendation rating–credibility relationship (∗ p < 0.1; ∗∗ p <

0.05; ∗∗∗ p < 0.01)

Predictors Model 1β

Model 2β

Aggregatedrating (AGG)

0.338∗∗∗ 0.396∗∗∗

Involvement(INV)

−0.066 −0.064

AGG × INV −0.273∗∗

R2 0.10∗∗∗ 0.15∗∗∗

�R2 0.05∗∗∗

the variance (p < 0.001). The additionof the interaction term again producesa significant increment in the amountof variance explained in eWOM cred-ibility (�R2 = 0.05; p < 0.01), indi-cating that involvement moderates thetrustworthiness-credibility relationship.

As can be seen in Table 6, the interac-tion of homophily and involvement hasa significant effect on credibility (β =0.212, p < 0.05). Homophily and in-volvement were able to explain 9 % ofthe variance in eWOM credibility (p <

0.001). The addition of the interactionterm produces a significant incrementin the amount of variance explained incredibility (�R2 = 0.04; p < 0.01) in-dicating that involvement moderates thehomophily-credibility relationship.

Finally, Table 7 shows aggregated rec-ommendation rating and involvementwere able to explain 10 % of the variancein eWOM credibility (p < 0.001). Theaddition of the interaction term producesa significant increment in the amount ofvariance explained in credibility (�R2 =0.05; p < 0.01), indicating that involve-

ment also moderates the recommenda-tion rating-credibility relationship.

The nature of the significant inter-action was explored using simple slopeanalyses (Aiken and West 1991; Fitzsi-mons 2008; Hennig-Thurau et al. 2012).The significant interactions were probedusing the techniques outlined by Aikenand West (1991). In this procedure, theeffects of parenting variables on out-come variables are estimated at 1 stan-dard deviation below the means (low)and 1 standard deviation above the mean(high). Figure 2 presents the slope plotsof the interactions.

Figure 2 shows the relationships be-tween the four predictors and eWOMcredibility as a function of involvement.The first slope plot supports the theo-retical argument: expertise led to highercredibility when involvement was high(b = 3.275; t = 6.47; p < 0.001) thanwhen it was low (b = 1.66; t = 4.63; p <

0.001). The spotlight analysis shows thatthis increase of value in the high involve-ment constellation is significant at p <

0.001. This supports H5a.With respect to trustworthiness, the

opposite can be observed: high involve-ment weakens the positive effects oftrustworthiness on eWOM credibility. Ananalysis of the simple slopes demon-strates that with both strong and weakinvolvement an increase of higher trust-worthiness is associated with higher cred-ibility. This increase, however, is less sig-nificant with high involvement (b = 1.08;t = 3.22; p < 0.001) than with weakinvolvement (b = 2.96; t = 5.25; p <

0.001). H5b thus is confirmed.The figure depicting the interaction be-

tween homophily and involvement showsthat the relation between homophily andcredibility increases with high values ofthe moderator. The analysis of “simpleslopes” results in significant increases forcredibility in dependence of homophilyfor both low involvement (b = 1.11; t =3.57; p < 0.001) and high involvement(b = 3.29; t = 6.48; p < 0.001). H5c thuscan be regarded as confirmed.

With reference to aggregated recom-mendation rating, a significant inter-action of involvement can likewise befound. The computation of the simpleslopes shows that with both strong andlow involvement an increase of the aggre-gated rating is associated with more cred-ibility. This increase, however, is less pro-nounced with strong involvement (b =1.96; t = 4.78; p < 0.001) than with lowinvolvement (b = 3.35; t = 6.57; p <

0.001). H5d can be confirmed.

136 Business & Information Systems Engineering 3|2013

BISE – RESEARCH PAPER

Fig. 2 Interactions between the predictors and involvement

5 Discussion

5.1 Summary

The empirical findings show that bothinformational and also normative deter-minants influence the credibility of on-line recommendations. Thus the two in-formational determinants expertise andtrustworthiness are significant factors ofinfluence for eWOM credibility. The re-sults of the source models (Hovland andWeiss 1951) can therefore be transferredto the eWOM environment: receivers ofonline recommendations for their ratingsmostly rely on the expertise and the trust-worthiness of the sender. In contrast, theinformational factor of social homophilycould not be confirmed. The results showthat congruency between the attributesof the participants in an online environ-ment is less relevant than could be provedin studies of the offline world (von Wan-genheim and Bayón 2004). This can cer-tainly be explained by the fact that im-pressions of homophily cannot be gen-erated easily in the environment of on-

line recommendations due to the lack ofinformation and personal knowledge.

In extension to previous studies, whichfocused mostly on the effects of infor-mational determinants, this study showsthat normative factors significantly influ-ence the credibility rating of online rec-ommendations. Thus reviewers of onlinerecommendations make use of the ag-gregated rating as an indicator for cred-ibility, besides expertise and trustworthi-ness. This result allows new views indi-cating that normative influences are sig-nificant in the eWOM environment aswell. Thus aggregated recommendationrating is based on the appraisals of therecommendations by other consumers.Are they rated positively, their credibilityrises. This can be explained by the dual-process theory and the conformity the-sis of Asch (1951), ascribing normativesocial influences to a need for conformity.

In addition this study confirmed a pos-itive relation between eWOM credibil-ity and final eWOM adoption. Empiri-cal findings show that credibility has apositive effect on the final eWOM adop-

tion and the connected purchase inten-tion. The results of the moderated regres-sion show a larger influence of expertiseand homophily on highly involved re-ceivers than on those less involved. In linewith ELM this could be explained withdiffering needs for information of highlyand less involved people (Petty and Ca-cioppo 1986). Highly involved receiversshow a strong need for information andfor the most part refer to expertise whenassessing credibility. In addition, they ac-tively look for a high degree of homophilybetween sender and receiver.

5.2 Implications

This study presents an analysis of themost important factors for credibilityrating of online recommendations. Fromthe results, important implications canbe derived for businesses, providers ofonline-forums, and online retailers. Anunderstanding of the determinants ofeWOM credibility carries implicationsfor the identification of credible recom-mendations. The appraisal and under-

Business & Information Systems Engineering 3|2013 137

BISE – RESEARCH PAPER

standing of the factors may help busi-nesses learn how customers rate the cred-ibility of eWOM. If, for example, busi-nesses use online recommendations toobserve the market, the identified fac-tors can add to an effective processingof the information found. The moni-toring of recommendations is a usefulmethod to identify consumer problemsand desires, and indicates product im-provements. Businesses can utilize iden-tified factors as starting points to cate-gorize generated information and eval-uate it from a consumer’s perspective.This knowledge is necessary to confirmthe importance of eWOM credibility andadoption as targets to be achieved. Thestudy shows that publishing credible rec-ommendations can be an effective way toraise eWOM adoption and thus the pur-chase probability. Decisive here are thefactors expertise and trustworthiness ofthe sender as well as the aggregated rec-ommendation rating. Businesses shouldenforce the observation of these factorswhen judging consumer recommenda-tions. In addition to previous studies ineWOM environment, the results showthat the normative influence in ratingcredibility has also to be considered. Forbusinesses this means that also in theonline context opinions of third partiesare relevant. They might use, for ex-ample, aggregated recommendations toidentify reviewers who are rated espe-cially positively by others. These could beexplicitly chosen to, e.g., test products.Also for providers of online consumerforums the findings from this study areof great benefit. The factors which all inall lead to higher credibility are pointedout. Providers should be more aware ofthese factors to raise the credibility oftheir eWOM forums. One possibility re-garding expertise, e.g., would be to installbonus systems to commit users who reg-ularly provide high quality recommen-dations. Furthermore, the implementa-tion of a rating system for the improve-ment of trustworthiness may be advanta-geous. Even though providers of forumsare not in a position to exercise con-trol on the contents of normative infor-mation, they might consider making thenormative aspects of a recommendationstand out. They could develop and pub-lish aggregated rating systems to make re-lations with normative information eas-ier. A further option might be a newdefinition of the rating system to makemulti-dimensional ratings possible. Con-trary to a sole offering of a general rat-

ing, the recommendation could be val-ued according to different criteria. Tak-ing into account the results of the moder-ating influence of involvement, providersshould introduce different rating systemsfor high and low involvement products.For low involvement products they canlay the focus on short recommendationsand a rating using just a few criteria onlysuch as notes or star icons. With high in-volvement products the focus should beon the reviewer’s expertise, i.e., especiallyqualified reviewers should be labeled.

The results also have important impli-cations for online retailers. Whereas on-line forums usually favor a “consumerprotective” approach, consumer recom-mendations have a quasi direct relevanceto the turnover of online retailers. For re-tailers it should be all the more impor-tant to provide credible reviews. The re-sults demonstrate that credible reviewsincrease eWOM adoption and thus theprobability of purchase. For online retail-ers the analysis shows a direct connectionbetween credible reviews and turnover. Afurther gain is to be found in the provi-sion of credible reviews as the orderingof “wrong” articles can be avoided. Thismakes it possible to save costs, for exam-ple by rendering logistics to return prod-ucts redundant. For the online retailerhimself credible recommendations canhave a positive effect. High-quality rec-ommendations may allow him to standout from competitors and to generate thecustomers’ trust. As for aggregated rec-ommendation rating, online retailers canidentify reviewers who were judged espe-cially positively and install them as inde-pendent testimonial reviewers. To distin-guish these from their use for the busi-nesses, a confirmed independence couldgenerate a specific value for customers.If the retailer succeeds in establishing afixed group of trustworthy reviewers, hecan enhance his business model by meansof a qualitative component which ex-ceeds the easily substituted pure distribu-tion function and thus generates a com-petitive advantage in a hotly contestedmarket.

All in all, the results demonstrate thatas much information on the review-ers as possible should be gathered. Asa minimum, a function should be in-stalled to evaluate and/or comment onthe recommendations, so that customerscan exchange opinions of the qualityof the recommendations. In addition,rubrics such as “background” or “occu-pation” of the reviewer would be use-ful. Alternatively, short guidelines should

be included which appear when writ-ing a review and show which informa-tion a “good” recommendation shouldcontain. Also, more incentive systemsshould be created to insure a large num-ber of credible consumer recommenda-tions. In this context bonuses for review-ers such as vouchers, priority purchases,or free delivery could be offered. Addi-tionally, a box “reviewer of the month”could reward reviewers by granting themattention.

5.3 Limitations and Future Research

Although this study has determined therelevant factors of the influence on thecredibility of online recommendations, itis just a first step and ought to be de-veloped further in different aspects. Sug-gestions for further research result fromthe various restrictions of this study. Thesample, for example, was limited to anonline discussion forum for consumers.Therefore I advise caution in generaliz-ing the results of this study. Which fac-tors are decisive for a receiver when rat-ing credibility also depends on the kindof information available. However, re-sults should be applicable to other on-line consumer forums as well. Futureresearch should also broaden the cur-rent approach and integrate additionalvariables into the context of direct andmoderating effects of eWOM. Thus fu-ture studies could integrate dimensionsreferring to text (e.g., choice of words,comprehensibility, and design) into theresearch design and analyze contents.

References

Aiken LS, West SG (1991) Multiple regression:testing and interpreting interactions. SagePublication, Newbury Park

Asch SE (1951) Opinions and social pressure.Scientific American 193:31–35

Bagozzi RP, Yi Y (1988) On the evaluation ofstructural equation models. Journal of theAcademy of Marketing Science 16(1):74–94

Bansal HS, Voyer PA (2000) Word-of-mouthprocesses within a services purchase deci-sion context. Journal of Service Research3(2):166–177

Berger J, Milkman KL (2012) What makesonline content viral. Journal of MarketingResearch 49(2):192–205

Bhattacherjee A, Sanford C (2006) Influenceprocesses for information technology ac-ceptance: an elaboration likelihood model.MIS Quarterly 30(4):805–825

Blanton H (2001) Evaluating the self in thecontext of another: the three-selves modelof social comparison assimilation and con-trast. In: Moskowitz GB (ed) Cognitive socialpsychology: the Princeton symposium onthe legacy and future of social cognition.Erlbaum, Mahwah, pp 75–87

138 Business & Information Systems Engineering 3|2013

BISE – RESEARCH PAPER

Bone PF (1995) Word of mouth effectson short-term and long-term productjudgements. Journal of Business Research32(3):213–223

Brewer MB, Webber JG (1994) Self-evaluationeffects of interpersonal versus intergroupsocial comparison. Journal of Personalityand Social Psychology 66(2):268–275

Brown J, Broderick AJ, Lee N (2007) Word ofmouth communication within online com-munities: conceptualizing the online socialnetwork. Journal of Interactive Marketing21(3):2–20

Büttner OB, Göritz AS (2008) Perceived trust-worthiness of online shops. Journal of Con-sumer Behaviour 7(1):35–50

Carmines EG, McIver JP (1981) Analysingmodels with unobserved variables: analy-sis of covariance structures. In: BohmstedtGW, Borgatta EF (eds) Social measurement.Sage, Thousand Oaks, pp 65–115

Celsi RL, Olson JC (1988) The role of involve-ment in attention and comprehension pro-cesses. The Journal of Consumer Research15(2):210–224

Chen Y, Xie J (2008) Online consumer review:word-of-mouth as a new element of mar-keting communication mix. ManagementScience 54(2):477–491

Cheung CMK, Lee MKO, Rabjohn N (2008)The impact of electronic word-of-mouth.Internet Research 18(3):229–247

Cheung M, Luo C, Sia CL, Chen H (2009) Cred-ibility of electronic word-of-mouth: infor-mational and normative determinants ofonline consumer recommendations. Inter-national Journal of Electronic Commerce13(4):9–38

Chevalier JA, Mayzlin D (2006) The effectof word of mouth on sales: online bookreviews. Journal of Marketing Research43(3):345–354

Cohen JB, Golden E (1972) Informational so-cial influence and product evaluation. Jour-nal of Applied Psychology 56(1):54–59

de Bruyn A, Lilien G (2008) A multi-stagemodel of word-of-mouth influencethrough viral marketing. InternationalJournal of Research in Marketing25(3):151–163

Deutsch M, Gerrard HB (1955) A study of nor-mative and informational social influenceupon individual judgment. Journal of Ab-normal and Social Psychology 51(3):629–636

Eagly AH, Chaiken S (1993) The psychology ofattitudes. Harcourt Brace Jovanovich, SanDiego

Fitzsimons GJ (2008) Death to dichotomizing.Journal of Consumer Research 35(1):5–8

Fornell C, Larcker DF (1981) Evaluating struc-tural equation models with unobservablevariables and measurement error. Journalof Marketing Research 18(1):39–50

Gibbons FX, Gerrard M (1991) Downwardcomparison and coping with threat. In: SulsJ, Wills TA (eds) Social comparison: con-temporary theory and research. LawrenceErlbaum, Hillsdale, pp 317–345

Gilly MC, Graham JL, Wolfinbarger MF (1998)A dyadic study of interpersonal informa-tion search. Academy of Marketing Science26(2):83–100

Godes D, Mayzlin D (2004) Using online con-versations to study word of mouth commu-nication. Marketing Science 23(4):545–560

Heckathorne W (2010) Speak now or foreverhold your tweets. Two in five say they aimto influence others when they express theirpreferences online (Harris Interactive Poll).http://www.harrisinteractive.com/Insights/HarrisVault.aspx. Accessed 2012-03-12

Hennig-Thurau T, Gwinner K, Walsh G, Grem-ler D (2004) Electronic word-of-mouth viaconsumer-opinion platforms: what moti-vates consumers to articulate themselveson the internet? Journal of Interactive Mar-keting 18(1):38–52

Hennig-Thurau T, Marchand A, Marx P (2012)Can automated group recommender sys-tems help consumers make better choices?Journal of Marketing 76(5):89–109

Hinz O, Skiera B, Barrot C, Becker JU (2011)Seeding strategies for viral marketing: anempirical comparison. Journal of Market-ing 75(6):55–71

Hovland C, Weiss W (1951) The influenceof source credibility on communicationeffectiveness. Public Opinion Quarterly15(4):635–650

Hovland CI, Janis IL, Kelley HH (1953) Com-munication and persuasion. Yale UniversityPress, New Haven

Hu N, Liu L, Zhang J (2008) Do online re-views affect product sales? The role of re-viewer characteristics and temporal effects.Information Technology and Management9(3):201–214

Huang JH, Chen YF (2006) Herding in onlineproduct choice. Psychology & Marketing23(5):413–428

Jones K (1996) Trust as an affective attitude.International Journal of Ethics 107(1):4–25

Kaplan MF, Miller CE (1987) Group deci-sion making and normative versus informa-tional influence: effects of type of issue andassigned decision rule. Journal of Personal-ity and Social Psychology 53(2):306–313

Kroeber-Riel W, Weinberg P (2003) Kon-sumentenverhalten, 8th edn. Vahlen,München

Laumann EO (1966) Prestige and associa-tion in an urban community, 2nd edn.Bobbs-Merrill, Indianapolis

Lazarsfeld PF, Merton RK (1964) Friendship associal process. A substantive and method-ological analysis. In: Berger M, Abel T, PageCH (eds) Freedom and control in modernsociety. Octagon, New York, pp 18–66

Lee J, Park D-H, Han I (2008) The effectof negative online consumer reviews onproduct attitude: an information process-ing view. Electronic Commerce Researchand Applications 7(3):341–352

Leonard-Barton D (1985) Experts as negativeopinion leaders in the diffusion of a tech-nological innovation. Journal of ConsumerResearch 11(4):914–926

Lord KR, Lee M-S, Choong P (2001) Differ-ences in normative and informational so-cial influence. Advances in Consumer Re-search 28(1):280–285

Mackiewicz J (2008) Reviewer motivations,bias, and credibility in online reviews. In:Kelsey S, Amant K (eds) Handbook of re-search on computer mediated communica-tion. IGI Global, Hershey, pp 252–266

McCracken G (1989) Who is the celebrityendorser? Cultural foundations of the en-dorsement process. Journal of ConsumerResearch 16(3):310–321

McCroskey J, Hamilton P, Weiner A (1974)The effect of interaction behavior onsource credibility, homophily, and interper-sonal attraction. Human CommunicationResearch 1(1):42–52

McCroskey J, Young T (1981) Ethos andcredibility: the construct and its measure-ment after three decades. The Central StateSpeech Journal 32(1):24–34

McGuire W (1985) Attitudes and attitudechange. In: Gardner L, Elliott A (eds) Hand-book of social psychology, New York, vol 2,pp 233–346

AbstractBettina Lis

In eWOM We Trust

A Framework of Factors thatDetermine the eWOM Credibility

Electronic word-of-mouth (eWOM) isan important factor in marketing com-munication. As more people use eWOMto assist them in making purchase deci-sions, the process by which they evalu-ate the credibility of these online rec-ommendations becomes increasinglyrelevant. Although previous studieshave recognized that credibility is oneof the most important antecedents ofeWOM adoption, little is known aboutthe drivers of this credibility. Thus, thispaper examines factors that influencethe perceived credibility of consumeronline recommendations. Drawing ondual process theory and source models,hypotheses were derived and testedwith structural equation modeling on abasis of 643 consumers. Generally, thepaper provides evidence that expertise,trustworthiness, and aggregate ratingare the most significant factors of theperceived eWOM credibility. The studyalso demonstrates that involvementcould moderate these relationships.

Keywords: Electronic word-of-mouth,Perceived credibility, Viral marketing,Dual process theory, Online consumerrecommendations

Business & Information Systems Engineering 3|2013 139

BISE – RESEARCH PAPER

McKnight DH, Chervany NL (2002) What trustmeans in e-commerce customer relation-ships: an interdisciplinary conceptual ty-pology. International Journal of ElectronicCommerce 6(2):35–59

McKnight DH, Kacmar C (2006) Factors ofinformation credibility for an Internet ad-vice site. In: Proc. 39th Hawaii internationalconference on system sciences, Hawaii, pp1–10

Miller DT, Hoppe RA (1973) The effect of re-gional similarity-dissimilarity on commu-nicator credibility. Language and Speech16(3):211–217

Nielsen, MTV, VW (2010) Me public. Aglobal study on social media youth:executive summary. http://www.viacombrandsolutions.de/media/6_research/studien_pdfs/. Accessed 2012-03-12

Nunnally JC (1978) Psychometric theory.McGraw-Hill, New York

O’Reilly K, Marx S (2011) How young, techni-cal consumers assess online WOM credibil-ity. Qualitative Market Research: An Inter-national Journal 14(4):330–359

Ohanian R (1990) Construction and valida-tion of a scale to measure celebrity en-dorsers’ perceived expertise, trustworthi-ness, and attractiveness. Journal of Adver-tising 19(3):39–52

Ohanian R (1991) The impact of celebrityspokespersons ’perceived image on con-sumers’ intention to purchase. Journal ofAdvertising Research 31(1):46–54

Petty RE, Cacioppo J, Schumann D (1983) Cen-tral and peripheral routes to advertisingeffectiveness: the moderating role of in-volvement. Journal of Consumer Research10(10):135–146

Petty RE, Cacioppo JT (1986) Elaboration like-lihood model. In: Berkowitz L (ed) Ad-vances in experimental social psychology,San Diego, pp 123–205

Petty RE, Priester J, Brinol P (2002) Mass mediaattitude change: implications of the elab-oration likelihood model of persuasion. In:Bryant J, Zillmann D (eds) Media effects: ad-vances in theory and research. LawrenceErlbaum, Mahwah, pp 155–198

Podsakoff N, Organ D (1986) Self-reportsin organizational research: problemsand prospects. Journal of Management12(4):531–544

Podsakoff PM, MacKenzie SB, Lee JY, Pod-sakoff NP (2003) Common method biasin behavioral research: a critical review ofthe literature and recommended remedies.Journal of Applied Psychology 88(5):879–903

Price LL, Feick LF (1984) The role of inter-personal sources in external search: an in-formational perspective. Advances in Con-sumer Research 11:250–255

Qiu L, Li D (2010) Effects of aggregate rat-ing on eWOM acceptance: an attributiontheory perspective. In: Proc PACIS 2010

Rafaeli S, Raban DR (2005) Information shar-ing online: a research challenge. Interna-tional Journal of Knowledge and Learning1(2):62–79

Rogers EM (1983) Diffusion of innovations.Free Press, New York

Sherif M, Hovland CI (1965) Social judgement:assimilation and contrast effects in com-munication and attitude change, 2nd edn.Yale University Press, New Haven

Smith T, Coyle JR, Lightfoot E, Scott A (2007)Reconsidering models of influence: the re-lationship between consumer social net-works and word-of-mouth effectiveness.

Journal of Advertising Research 47(4):387–397

Sparkman R, Locander W (1980) Attributiontheory and advertising effectiveness. Jour-nal of Consumer Research 7(7):219–224

Sussman SW, Siegal WS (2003) Informationalinfluence in organizations: an integratedapproach to knowledge adoption. Infor-mation Systems Research 14(1):47–65

Tseng S, Fogg BJ (1999) Credibility and com-puting technology. Communications of theACM 42(5):39–44

von Wangenheim F, Bayón T (2004) Theeffect of word-of-mouth on servicesswitching: measurement and moderatingvariables. European Journal of Marketing38(9/10):1173–1185

Walsh G, Mitchell VW (2010) The effect ofconsumer confusion proneness on wordof mouth, trust, and customer satisfaction.European Journal of Marketing 44(6):838–859

Wathen CN, Burkell J (2002) Believe it or not:factors influencing credibility on the web.Journal of the American Society for Infor-mation Science and Technology 53(2):134–144

Wilson EJ, Sherrell DL (1993) Source effects incommunication and persuasion research: ameta-analysis of effect size. Journal of theAcademy of Marketing Science 22(2):101–122

Yale LJ, Gilly MC (1995) Dyadic perceptionsin personal information search. Journal ofBusiness Research 32(3):225–237

Zaichkowsky JL (1985) Measuring the in-volvement construct. The Journal of Con-sumer Research 12(3):341–352

140 Business & Information Systems Engineering 3|2013