Word-of-Mouth Communications in Marketing A Meta-A

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/227196796

Word-of-Mouth Communications in Marketing: A Meta-Analytic Review of


the Antecedents and Moderators

Article in Journal of the Academy of Marketing Science · December 2008


DOI: 10.1007/s11747-008-0121-1

CITATIONS READS

994 7,222

2 authors:

Celso Augusto de Matos Carlos Alberto Vargas Rossi


Universidade NOVA de Lisboa Universidade Federal do Rio Grande do Sul
67 PUBLICATIONS 2,554 CITATIONS 43 PUBLICATIONS 2,449 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cross-Cultural Consumer Behavior View project

Consumer Debt - Endividamento de Consumidores View project

All content following this page was uploaded by Carlos Alberto Vargas Rossi on 17 June 2014.

The user has requested enhancement of the downloaded file.


J. of the Acad. Mark. Sci. (2008) 36:578–596
DOI 10.1007/s11747-008-0121-1

ORIGINAL EMPIRICAL RESEARCH

Word-of-mouth communications in marketing: a meta-analytic


review of the antecedents and moderators
Celso Augusto de Matos & Carlos Alberto Vargas Rossi

Received: 27 February 2008 / Accepted: 8 September 2008 / Published online: 23 September 2008
# Academy of Marketing Science 2008

Abstract Although word-of-mouth (WOM) activity has Interpersonal influence has received great attention in social
been studied as an outcome variable of other constructs psychology and there is ample evidence for the relevance of
such as satisfaction, less attention has been given to the interpersonal communications when individuals make
antecedents and moderators of WOM when considering choices in different contexts, including those of consump-
WOM as a central construct. Hence, we propose a model of tion (Katz and Lazarsfeld 1955). This effect is also
WOM antecedents and moderators using a meta-analytic investigated in consumer behavior, and its models consider
review. The results show that all antecedents have signif- this interpersonal influence when information is sought as
icant effects on WOM activity, with customer commitment well as given (Engel et al. 1995). Empirical studies show that
showing the strongest effect. The following hypotheses are customers are even more likely to rely on these interpersonal
also supported: (1) WOM valence is a significant moder- communications, known as “word of mouth” (WOM), in the
ator, (2) cross-sectional studies show a stronger influence of service context because of the intangibility and experiential
satisfaction and loyalty on WOM activity than longitudinal nature of services (Murray 1991; Zeithaml et al. 1993).
studies, and (3) studies of WOM behavior show a weaker In the marketing context, WOM communications are
link between loyalty and WOM activity than studies of defined as “informal communications directed at other
WOM intentions. In addition, we show that satisfaction has a consumers about the ownership, usage, or characteristics
stronger relationship with positive WOM than loyalty, of particular goods and services and/or their sellers”
whereas (dis)loyalty has a stronger relationship with negative (Westbrook 1987, p. 261). Arndt (1967) defined word of
WOM than does (dis)satisfaction. We discuss this finding mouth as “oral, person-to-person communication between a
based on the different natures of positive and negative WOM. perceived non-commercial communicator and a receiver
concerning a brand, a product, or a service offered for sale”
Keywords Word-of-mouth activity . (p. 190). These definitions are consistent with recent studies
Word-of-mouth valence . Antecedents . Moderators . about WOM (Gruen et al. 2006; Harrison-Walker 2001;
Meta-analysis Wangenheim 2005; Wangenheim and Bayón 2007).
There have been a great number of studies about WOM
This article is based on the first author’s dissertation. since the frequently cited Arndt (1967) study, although
C. A. de Matos (*)
many of these investigations are focused on other con-
Marketing Department, School of Management, structs, such as satisfaction, and consider WOM merely as
Federal University of Rio Grande do Sul (PPGA-EA-UFRGS), one of the behavioral consequences (Mittal et al. 1999;
P.O. Box 532, 93.220-970, Sapucaia do Sul, Oliver 1980; Swan and Oliver 1989). A recent investigation
Rio Grande do Sul, Brazil
e-mail: celsomatos@yahoo.com.br
by Mazzarol et al. (2007, p. 1478) has emphasized exactly
this point in saying, “Little research, however, has
C. A. V. Rossi addressed antecedents of WOM when considering WOM
Marketing Department, School of Management, as a focal construct.” Although the positive effects of the
Federal University of Rio Grande do Sul (PPGA-EA-UFRGS),
Rua Washington Luiz, 855, Centro, 90.010-460,
antecedents of WOM, such as satisfaction, loyalty, quality,
Porto Alegre/RS, Brazil commitment, trust, and perceived value, are well estab-
e-mail: cavrossi@ea.ufrgs.br lished in the literature, there is still a lack of integration for
J. of the Acad. Mark. Sci. (2008) 36:578–596 579

WOM Approach Studies’ Characteristics


the bivariate relationships involving WOM. Hence, we - WOM valence (positive, - Survey or experiment
negative or mixed) - Cross-sectional or longitudinal
conduct a systematic review of these antecedents using a - WOM incidence - Student or non-student
meta-analysis in order to provide a quantitative integration (intention or behavior) - Products or services

of the previous studies after adjusting for measurement and


sampling error, to make comparisons about the strength of Antecedents
each antecedent in influencing WOM, and to identify Satisfaction
studies’ characteristics as possible moderators.
Using the systematic and quantitative meta-analytic
review, this study makes several contributions to the field. Loyalty

Primarily, it is the first meta-analytic effort to assess WOM


as a focal construct and to investigate its antecedents and Quality
moderators. Additionally, it identifies the different method-
Word-of-Mouth
ological approaches and the main antecedents of WOM in Activity
empirical studies. It evaluates the moderating effects of Commitment

studies’ characteristics and different WOM approaches such


as WOM valence and WOM incidence. Moreover, it finds that Trust
satisfaction has a stronger relationship with positive WOM
than loyalty, while (dis)loyalty has a stronger relationship with
negative WOM than does (dis)satisfaction. Lastly, it identifies Perceived
Value
research questions worthy of future investigations into WOM.
Our review is structured as follows. First, we present a
conceptual framework with the proposed hypotheses to Figure 1 Conceptual framework of the antecedents and moderators
guide the meta-analysis. Second, we describe the meta- of word-of-mouth activity.
analytic procedures, including the search process, database
development, effect sizes computation, and integration.
Third, we present a quantitative summary of the adjusted analyses. The theoretical rationale for the relationships
mean effect sizes for the pair-wise relationships between presented in Fig. 1 is discussed next.
WOM and its antecedents. Fourth, we present the moder-
ating analysis and the test of the different roles of Antecedents of WOM
satisfaction and loyalty for positive and negative WOM
activity. Finally, we discuss the main findings. Satisfaction

The dominant model for conceptualizing and measuring


Conceptual framework customer satisfaction has been the expectancy disconfirma-
tion theory. This view holds that customers evaluate a
We developed the conceptual framework shown in Fig. 1 product or service performance and compare their evalua-
based on the literature review of WOM. This meta-analytic tion with their expectations prior to purchase or consump-
framework describes the relationships between WOM and tion (Oliver 1980). In this approach to satisfaction as a
the most investigated antecedents. Although the positive post-choice evaluative judgment regarding a specific pur-
effects of the WOM antecedents depicted in Fig. 1 are well chase selection, satisfaction is understood by its transaction-
established in the literature, a systematic review of these specific component (Oliver 1981). Another approach sees
antecedents using a meta-analysis enables us to make satisfaction as the customers’ evaluations of multiple
comparisons about the strength of each in influencing experiences with the same product or service provider over
WOM. time (Bolton and Drew 1991), and given that this cumula-
Among the potential moderators, we focus on WOM tive construct incorporates previous experiences, the cumu-
valence, WOM incidence, and the studies’ characteristics. It lative satisfaction construct will contain an element of
is common for meta-analyses to investigate the variability customer attitude (Westbrook and Oliver 1991). Subsequent
of effect sizes across studies’ characteristics (Assmus et al. studies, however, have demonstrated that, in addition to the
1984; Pan and Zinkhan 2006). It has been suggested that cognitive view, customer satisfaction also contains emo-
other variables, such as switching costs and customer tional components (Liljander and Strandvik 1997; Oliver
experience, could also intervene between WOM and its and Westbrook 1993; Straus and Neuhaus 1997). A recent
correlates, but the limited number of studies testing these study (Martin et al. 2008) agrees with Zeelenberg and
variables as moderators precluded their assessment in our Pieters (2004) in suggesting that emotionally based satis-
580 J. of the Acad. Mark. Sci. (2008) 36:578–596

faction is a stronger predictor of future behavioral inten- proach is commonly used to include repatronage intentions
tions than traditional cognitive measures. and WOM intentions as items of a unidimensional loyalty
The level of customer satisfaction has an influence on construct. Two empirical studies conducted by Söderlund
two purchase behaviors, namely, repurchase intentions and (2006) support the argument that repurchase intentions and
WOM (Bearden and Teel 1983; Maxham and Netemeyer WOM should be considered as separate constructs and the
2002a, b; Oliver 1980; Ranaweera and Prabhu 2003; Richins “cocktail approach” should be avoided. There are also other
1983). Specifically, the likelihood of customers spreading authors modeling repurchase intentions and WOM as
WOM will depend on their satisfaction level for at least two independent constructs, including Blodgett et al. (1993),
reasons. First, the extent to which the product or service Gruen et al. (2006), Jones and Reynolds (2006), Maxham
performance exceeds the customer’s expectations might (2001), and Maxham and Netemeyer (2002a, b).
motivate him or her to tell others about his or her positive Dick and Basu (1994) called attention to the fact that
experience. In the context of service recovery, for instance, there were few studies investigating the influence of loyalty
the salience and recency of the experience might explain on WOM. Yet, even after a decade, there have been very
why satisfaction with the recovery prompts customers to tell few studies which have empirically tested the unidirectional
family and friends about their positive experience (Maxham effect of loyalty on WOM (Carpenter and Fairhurst 2005;
and Netemeyer 2002b). Second, to the extent that the Gounaris and Stathakopoulos 2004; Reynolds and Arnold
customer’s expectations are not fulfilled, possibly creating a 2000; and Sichtmann 2007). It is more common to measure
customer regret experience, this customer will engage in the bi-directional association (correlation) between these
WOM behavior as a form of “venting” his or her negative two constructs when considering them as behavioral
emotions, such as anger and frustration, reducing anxiety, outcomes, especially of customer satisfaction (Arnett et al.
warning others, and/or seeking retaliation (Anderson 1998; 2003; Gremler and Gwinner 2000; Jones and Reynolds
Oliver 1997; Richins 1984; Sweeney et al. 2005). 2006; Macintosh 2007; Maxham and Netemeyer 2002a, b,
Indeed, there are a number of studies supporting the 2003; Price and Arnould 1999; Reynolds and Beatty 1999;
significant effect of satisfaction on WOM (Brown et al. 2005; Söderlund 1998). Thus, because of the availability of the
Heckman and Guskey 1998; Heitmann et al. 2007; Hennig- correlations between loyalty and WOM, we could test the
Thurau et al. 2002; Mittal et al. 1999; Price and Arnould effects of loyalty on WOM in our meta-analysis.
1999; Söderlund 2006; Swan and Oliver 1989; Wangenheim Loyalty is hypothesized as an antecedent of WOM
and Bayón 2007). Thus, our conceptual framework proposes: because to the extent customers are more loyal to a given
provider, they are also more likely to (1) give positive
H1a: There is a significant positive effect of satisfac-
recommendations of the company to the individuals in their
tion on WOM activity.
reference group (friends and relatives), (2) have greater
motivation for processing new information about the
Loyalty company, and (3) have stronger resistance to being
persuaded by contrary information (Dick and Basu 1994,
Loyalty is defined in the marketing context as “an intention p. 107). Moreover, in a disloyalty situation, in which
to perform a diverse set of behaviors that signal a customers switch providers, they are also likely to spread
motivation to maintain a relationship with the focal firm, negative WOM about the provider in order to reduce their
including allocating a higher share of the category wallet to cognitive dissonance (Wangenheim 2005). In other words,
the specific service provider, engaging in positive word of they try to convince themselves about their decision by
mouth (WOM), and repeat purchasing” (Sirdeshmukh et al. convincing others, which is one of the strategies often used
2002, p. 20). Note that this conceptualization considers for reducing post-decision dissonance. As mentioned
positive WOM as a component of loyalty. This approach is before, customers might also engage in negative WOW
common in a great number of studies in the marketing for other reasons such as to release negative emotions, to
literature (Bloemer et al. 1999; Jones and Sasser 1995; warn others, and/or to retaliate (Richins 1984; Sweeney et
Jones and Taylor 2007; Lam et al. 2004; Zeithaml et al. al. 2005). Based on this rationale, we propose:
1996), and we found that it was employed in 40 studies
H1b: There is a significant positive effect of loyalty on
reviewed in our meta-analysis. These studies could not,
WOM activity.
however, be included because they did not present specific
results for the WOM construct.
However, a recent study has questioned this “cocktail Quality
approach” in which loyalty is measured by an aggregated
mix of items that form different components of loyalty Our framework focuses on quality of services, rather than
(Söderlund 2006). Söderlund (2006) states that this ap- that of products, because the available studies do so. Only
J. of the Acad. Mark. Sci. (2008) 36:578–596 581

one of the observations investigated a product, namely an 1994). In the multidimensional approach, commitment is
e-CRM software. We show in the “Results” section that this composed of “affective” (positive emotional attachment),
observation produced findings similar to those studies “continuance” (perceived costs associated with leaving the
investigating services. organization), and “normative” (perceived moral obligation
Different models have been proposed to measure and toward the organization) commitment. Some authors
evaluate the determinants of service quality. Parasuraman et consider two dimensions, namely “affective” and “high-
al. (1985, 1988) developed the SERVQUAL model in sacrifice (calculative) commitment” (Fullerton 2003; Harrison-
which quality is evaluated from the eyes of the customers Walker 2001; Jones et al. 2007). While the former is related
and defined as the discrepancy between customers’ expect- to the customer identification with, and involvement in, a
ations or desires and their perceptions. In other words, particular organization, the latter refers to the customer’s
quality is understood as the ability to meet or exceed sense of being “locked in” to the service provider, due to
customers’ expectations. The instrument proposed by these constraints like loss of benefits and costs for switching
authors (SERVQUAL) suggests that service quality is provider. On the other hand, the unidimensional approach
composed of five dimensions—tangibles, reliability, re- measures commitment as an overall evaluation of the
sponsiveness, assurance, and empathy. customers’ engagement with the organization.
Customers’ perceptions of service quality have an Higher commitment customers (relationship-oriented
important relationship with their behavioral responses, customers) are those who also present higher identification
especially loyalty and WOM. For WOM, when evaluations with the company and hold feelings of attachment to
of service quality are high, the customer’s behavioral maintaining valued relationships, reflecting the affective
intentions in terms of recommendations are favorable, dimension of the construct (Harrison-Walker 2001). More-
strengthening the relationship between customers and the over, customers are more likely to increase their commit-
company (Parasuraman et al. 1988; Zeithaml et al. 1996). ment with companies that recognize and reward their status
On the other hand, when customers perceive service of special customer (Lacey et al. 2007). Therefore, these
performance as inferior, they are likely to manifest customers are likely to provide favorable WOM about the
complaining behavior, including private responses (nega- company as a need to reinforce their decision to enter the
tive WOM) and/or defection (Zeithaml et al. 1996). Hence, relationship as a good one. Even when experiencing lower
customers recommend the company to others when they levels of satisfaction, these customers are likely to endorse
perceive high service quality and spread negative WOM the company in order to keep cognitive consistency and
when they perceive low service quality. justify their favorable attitude and strong identification with
Empirical studies have demonstrated that service quality the company (Brown et al. 2005).
is a relevant predictor of WOM (Bloemer et al. 1999; Indeed, Brown et al. (2005) have demonstrated in their
Boulding et al. 1993; Harrison-Walker 2001; Zeithaml et al. longitudinal study that for higher-commitment customers,
1996). A positive relationship presented in these studies positive WOM behavior is less dependent on the satisfac-
demonstrates that the higher (lower) the perceived quality, tion level. The reason is that high-commitment customers
the higher (lower) the WOM activity of the customers. talk positively about the company regardless of their
Based on this discussion, we propose: satisfaction level, whereas low-commitment customers will
provide favorable recommendations to the extent that they
H1c: There is a significant positive effect of quality on
are satisfied. This finding is in agreement with other studies
WOM activity.
which establish that commitment has a positive relationship
with WOM (Hennig-Thurau et al. 2002; Lacey et al. 2007).
Commitment Based on the above discussion, we expect that customers
with higher commitment would have a greater likelihood of
Commitment can be defined as “an enduring desire to spreading positive WOM because their WOM behavior is
maintain a valued relationship” (Moorman et al. 1992, stimulated by either (1) a high satisfaction condition (when
p. 316). This definition is in agreement with Dwyer et al. satisfaction stimulates WOM) or (2) a low satisfaction state
(1987) conceptualization of buyer-seller relationships and is (when a cognitive consistency mechanism prompts the
also consistent with Morgan and Hunt’s (1994, p. 23) customer to endorse the company in order to reduce
definition of commitment as “an exchange partner believing cognitive dissonance). Hence, we propose:
that an ongoing relationship with another is so important as
H1d: There is a significant positive effect of commit-
to warrant maximum efforts at maintaining it.” Commit-
ment on WOM activity.
ment is measured in the marketing literature either as a
multidimensional construct (Gruen, Summers, and Acito We noticed in our review that some studies related only
2000) or a unidimensional construct (Morgan and Hunt the dimensions of commitment (affective and continuance)
582 J. of the Acad. Mark. Sci. (2008) 36:578–596

with WOM without providing a relationship between of what is received and what is given” (Zeithaml 1988,
overall commitment and WOM (Fullerton 2003; Harrison- p. 14), or in other words, a trade-off between benefits or
Walker 2001). Since the meta-analysis had to combine gets (quality, convenience, volume, etc.) and costs or gives
similar studies, we included only those studies that related (money, time, efforts, etc.).
overall commitment with WOM. Previous studies have investigated the influence of service
quality on customers’ behavioral intentions (Boulding et al.
Trust 1993; Zeithaml 1988). Based on this literature, Hartline and
Jones (1996) proposed that perceived value also has an
Trust refers to “a willingness to rely on an exchange partner influence on customers’ behavioral intentions, especially on
in whom one has confidence” (Moorman et al. 1993, p. 82). WOM. One explanation is that customers who perceive that
For Morgan and Hunt (1994, p. 23), trust exists “when one they receive relatively high value tend to become more
party has confidence in an exchange partner’s reliability and committed to the organization and seek to recommend
integrity.” Thus, confidence and reliability are two important others of the reference group to become loyal to the same
factors for conceptualizing trust in the marketing context. Re- organization (McKee et al. 2006).
search in this field has demonstrated that customers’ trust— Also, perceived value might have an influence on WOM
either in the overall organization or in the employees—is because it is a more tangible signal in the service encounter
significantly influenced by customers’ satisfaction (Kau and since it includes price in the “give” component, and price
Loh 2006; Morgan and Hunt 1994; Ranaweera and Prabhu can be considered a more extrinsic and tangible attribute
2003; Singh and Sirdeshmukh 2000). These findings show when compared to other cues used to infer service quality
that the higher (lower) the customer’s satisfaction with the by customers, such as competence and responsiveness of
organization, the higher (lower) his or her trust in it. employees (Hartline and Jones 1996).
Trust also has an important effect on behavioral constructs, Consistent with the above rationale, perceived value has
especially on the customer’s propensity to leave or stay with been hypothesized as a predictor or a correlate of WOM in
the same service provider (Garbarino and Johnson 1999; a number of studies (Durvasula et al. 2004; Gruen et al.
Morgan and Hunt 1994; Singh and Sirdeshmukh 2000). 2006; Hartline and Jones 1996; Keiningham et al. 2007;
Indeed, empirical findings have shown that higher levels of McKee et al. 2006). Given that the majority of these studies
trust are associated with a greater tendency to offer favorable are in the services context—only one of the ten observa-
WOM (Garbarino and Johnson 1999; Gremler et al. 2001; tions was related to a product context—our investigation as
Ranaweera and Prabhu 2003). This is based on the rationale well focuses on the service context.
that customers mostly provide recommendations to other Based on the above rationale, we propose:
individuals of their reference group, such as a friend or a
H1f: There is a significant positive effect of perceived
relative, and, thus, a customer will be more likely to endorse
value on WOM activity.
a provider that he or she has previous experience with and
confidence in (Gremler, Gwinner, and Brown 2001). But even
when customers are offering advice to others, no matter Moderators
who the receiver is, there is a risk of being wrong and a WOM
giver would not like to be wrong (Mazzarol et al. 2007). The different effects found across studies could be due to
Another possible reason for the influence of trust on different approaches to WOM. For example, there are
WOM is an indirect effect through satisfaction. Trust studies measuring positive WOM, others measuring nega-
creates benefits for customers such as lower anxiety, tive WOM, and yet others using both negative and positive
uncertainty, and vulnerability about the transaction. These items in the WOM construct. There are studies that work
benefits influence satisfaction, which in turn affects WOM, with intentions of WOM and others measuring WOM
especially in a service context that is relatively more informed behavior. Moreover, studies’ characteristics in
complex (Garbarino and Johnson 1999; Hennig-Thurau et terms of methodological designs could also have an
al. 2002). Based on this discussion, we propose: influence on the results. These effects are discussed next.
H1e: There is a significant positive effect of trust on
WOM valence
WOM activity.
In an effort to develop a scale to measure WOM, Harrison-
Perceived value Walker (2001) conceptualized WOM as being composed by
the factors of frequency, number of contacts, detail of the
Customer-perceived value is the “consumer’s overall shared information, and praise (valence). Harrison-Walker’s
assessment of the utility of a product based on perceptions (2001) study eventually showed two dimensions of WOM
J. of the Acad. Mark. Sci. (2008) 36:578–596 583

communications. The first was “WOM activity,” which is that customers in the extremes of high or low satisfaction
included aspects of how often the WOM communication are more likely to spread WOM. Indeed, as discussed
takes place, the number of people told, and the quantity of earlier, research has shown that highly satisfied customers
information provided by the sender. The second was have a desire to tell others about their positive experience
“WOM praise,” reflecting the valence of the WOM (Brown et al. 2005; Söderlund 2006; Swan and Oliver
communication (positive, negative, or neutral). This dis- 1989; Wangenheim and Bayón 2007). Thus, we expect that
tinction between WOM activity and WOM valence is a customers spreading positive WOM are those customers
relevant contribution by Harrison-Walker (2001) to the who have high satisfaction and, hence, when WOM
WOM literature, and she proposes that researchers should assumes a positive valence, there will be a direct relation-
measure not only the WOM activity but also the WOM ship between satisfaction and WOM (higher satisfaction is
valence, which is in agreement with earlier investigations associated with higher positive WOM).
about WOM (Halstead 2002; Halstead and Dröge 1991; On the other hand, we also expect that customers
Swan and Oliver 1989) as well as with more recent research spreading negative WOM are those with the lowest
(Heitmann et al. 2007). satisfaction levels, such as those experiencing a service
However, most of the studies in the WOM literature failure followed by an unsatisfactory recovery. The reason
measure WOM as a single unidimensional construct, is that disappointed customers (those with unsatisfied
addressing WOM valence only indirectly by measuring the expectations) often experience negative emotions such as
likelihood of positive or negative recommendations. In some regret, frustration, and anger (Bonifield and Cole 2007;
instances, studies work with a neutral position, mixing Sweeney et al. 2005). As a consequence, they use negative
positive and negative measures among the construct. WOM as a way of “venting” their negative emotions and
In our theoretical framework, we follow Harrison- achieving a feeling of relief after commenting on the
Walker’s (2001) conceptualization of WOM by establishing incident (Sundaram, Mitra, and Webster 1998; Sweeney et
WOM activity as the outcome variable and WOM valence al. 2005; Wetzer et al. 2007). Therefore, we expect that
as a moderator (see Fig. 1). The reason for doing this is that customers spreading negative WOM are those with lower
the focus of the reviewed WOM literature is on the aspects satisfaction and, thus, when WOM assumes a negative
of WOM activity, including frequency, number of people valence, there will be an inverse relationship between
told, and amount of shared information (WOM valence is satisfaction and WOM (lower satisfaction is associated with
considered only indirectly, as discussed before). In taking higher negative WOM).
this position, we recognize that WOM is not a unidimen- Based on the above discussion, we propose:
sional construct, and that we will include under WOM
H2a: Under the condition of positive (negative) WOM
activity studies that measure frequency, relative number of
valence, there is a direct (inverse) relationship between
people told or relative quantity of information shared, since
satisfaction and WOM activity.
these were the facets of WOM activity supported by
Harrison-Walker’s (2001) study. WOM valence will be The same justification is extended to the relationship
addressed as a moderator by categorizing WOM as between loyalty and WOM. As we discussed in H1b,
positive, negative, or mixed. In the test of this moderation, customers who are more loyal to a company are more likely
we evaluate whether the effect of the antecedent on WOM to spread positive recommendations about this company,
activity differs with the categories of positive, negative, and either as a cognitive consistency mechanism (Wangenheim
mixed WOM valence. 2005) or as part of a self-enhancement motivation, in
This is a common approach for a test of moderators in a which customers provide WOM to praise what they have a
meta-analysis (Lipsey and Wilson 2001). In other words, if strong relationship/identification with (Brown et al. 2005;
the effect of a given predictor, such as satisfaction, in the Sundaram et al. 1998). Thus, we expect that customers
outcome variable (WOM activity) is contingent on a third spreading positive WOM are those who have higher
variable (WOM valence as positive, negative, or neutral), loyalty, which suggests a direct relationship between
then the predictor-outcome relationship should be analyzed loyalty and WOM activity for positive WOM valence.
for each condition of the third variable (moderator). In contrast, we expect that customers spreading negative
WOM valence has been receiving more attention in WOM are those with lower loyalty or in a disloyalty
recent investigations about WOM, and there is evidence situation, such as those in a condition of complaint with no
that extremely positive and extremely negative WOM are recovery or dissatisfactory recovery (Voorhees et al. 2006),
the most common examples of WOM (Chevalier and or those customers who have switched providers and who
Mayzlin 2006; Duan et al. 2008; Mazzarol et al. 2007; spread negative WOM about the former provider as a
Sweeney et al. 2005; Zeelenberg and Pieters 2004). The means of reducing cognitive dissonance (Wangenheim
rationale, as shown in Oliver (1997) and Anderson (1998), 2005). Therefore, if customers spreading negative WOM
584 J. of the Acad. Mark. Sci. (2008) 36:578–596

are those with low or no loyalty, we can propose that when patterns in a meta-analysis: research context, model specifi-
WOM assumes a negative valence, an inverse relationship cation, measurement methods, and estimation procedure. But
is expected between loyalty and WOM (lower loyalty is since our unit of analysis was bivariate correlations, we
associated with higher negative WOM). Based on this looked for systematic differences in the study characteristics.
rationale, we propose: This is a common procedure in meta-analyses using
correlations (Henard and Szymanski 2001; Pan and Zinkhan
H2b: Under the condition of positive (negative) WOM
2006). We propose four characteristics as potential moder-
valence, there is a direct (inverse) relationship between
ators in our framework: method (survey or experiment),
loyalty and WOM activity.
design (cross-sectional or longitudinal), subject (student or
non-student), and context (product or service).
WOM incidence
Experimental versus survey method The use of experimen-
The relationship between an antecedent and WOM can also tal or non-experimental investigation might contribute to
vary across studies investigating WOM intention and explaining the variability in the effect sizes. Given that in
WOM informed behavior. Brown et al. (2005), for example, an experimental context the researcher is able to randomly
report a correlation of .74 between satisfaction and WOM assign subjects to the different conditions and control the
when measuring WOM intentions and .35 when measuring covariates that might influence the results, there is a greater
WOM informed behavior. Studies measuring WOM inten- chance of less error variance in the denominator of the
tions might present higher correlations because of the effect size and, consequently, there are larger effect sizes
attitudinal nature of the measures. In other words, respond- (Crosno and Dahlstrom 2008; Geyskens et al. 1998; Pan
ents usually try to keep their responses coherent and those and Zinkhan 2006). Based on this rationale, we propose:
manifesting high (low) satisfaction will be more likely to
H4a: The relationship between satisfaction and WOM
provide high (low) intentions of behavior. But studies that
activity is stronger (weaker) for experimental (non-
combine different measures (satisfaction as an evaluation
experimental) studies.
but WOM as a performed behavior) are more likely to
H4b: The relationship between loyalty and WOM
capture the “inconsistencies” of customer behavior. For
activity is stronger (weaker) for experimental (non-
instance, some customers are satisfied, have intentions to
experimental) studies.
provide favorable recommendations, but do not convert this
plan into practice because of a lack of opportunity. This is
in agreement with previous meta-analyses showing signif- Cross-sectional versus longitudinal design The strength of
icant differences between measures of intention (intentions the relationships between satisfaction, loyalty, and WOM
vs. estimates of performance) for the link between attitude might vary over time, just like consumers’ perceptions
and behavior (Sheppard et al. 1988). In this review, studies may change over time (Maxham and Netemeyer 2002a).
measuring intentions provided stronger correlations, while Hence, a study that asks the customer to recall the
those measuring estimates of behavior provided weaker consumption experience or approaches the participant just
correlation. Based on this rationale, we propose: after the consumption experience is likely to produce
associations (between satisfaction time 1 and WOM time 1,
H3a: The relationship between satisfaction and WOM
for instance) that are stronger when compared to a study
activity is stronger (weaker) when the study measures
that uses measures from different time frames (between
intentions (reported behavior) of WOM activity.
satisfaction time 1 and WOM time 2). The rationale is that
H3b: The relationship between loyalty and WOM
measuring predictors and criterion variables at the same
activity is stronger (weaker) when the study measures
point in time may produce artifactual covariance indepen-
intentions (reported behavior) of WOM activity.
dent of the content of the constructs themselves (Podsakoff
et al. 2003). Moreover, another common method bias is the
Study characteristics respondents’ tendency to maintain consistency in their
responses. Hence, if respondents are highly satisfied they
Differences across studies can also be explained by will manifest a high propensity toward positive WOM as
variations in studies’ characteristics in terms of survey or well as manifested WOM behavior. But if satisfaction is
experiment approach, cross-sectional or longitudinal meas- measured in, say, time 1 and WOM in time 2, it is likely
ures, student or non-student subjects, scenario or non- that there will be greater heterogeneity in responses.
scenario-based research, and the product or service context. Therefore, this greater heterogeneity is more likely to
Assmus et al. (1984) suggest that four categories of increase error variance and produce lower effect sizes (Fern
characteristics might contribute to identifying systematic and Monroe 1996).
J. of the Acad. Mark. Sci. (2008) 36:578–596 585

Even when the longitudinal study uses measures that are H6b: The relationship between loyalty and WOM
taken from the same point in time, it is likely that activity is stronger (weaker) for studies using a student
intercorrelations will be different when compared to the (non-student) sample.
ones from the first attempt, just because participants have
provided answers before and some time has passed. For
Product versus service context Services are known as being
instance, Brown et al.’s (2005) longitudinal study presented
intangible, perishable, and requiring higher involvement
lower correlations between satisfaction, commitment, and
between customers and sellers during the production and
WOM in their second survey. The possible reason might be
consumption process (Parasuraman et al. 1985). As
a dissipation effect because of the time that had elapsed
suggested by Palmatier et al. (2006), this close interaction
since the consumption experience.
between customers and sellers might make relationships
Based on this discussion, we propose:
more critical for services. This might also be the case for
H5a: The relationship between satisfaction and WOM WOM behavior and its relationship with customer satisfac-
activity is stronger (weaker) for cross-sectional (longi- tion and loyalty, since the process of spreading WOM
tudinal) studies. involves relationships between customers and sellers and
H5b: The relationship between loyalty and WOM activity customers with other customers. Szymanski and Henard
is stronger (weaker) for cross-sectional (longitudinal) (2001), for instance, have found a stronger correlation
studies. between satisfaction and repeat purchasing when services
rather than products are investigated. Likewise, Grinstein
(2008) as well has found a stronger effect for the link
Student versus non-student subjects Sample homogeneity between market orientation and innovation consequences
might have an influence on the magnitude of the effect for service firms rather than for manufacturing ones. Based
sizes. Student samples are likely to be more homogeneous, on this discussion, we propose:
while non-student samples are usually heterogeneous.
Given that a homogeneous sample tends to give responses H7a: The relationship between satisfaction and WOM
that vary less across the scale values, there might be a activity is stronger (weaker) for studies in the service
problem of restricted ranges and attenuated effect sizes (product) context.
(Fern and Monroe 1996). On the other hand, a non-student H7b: The relationship between loyalty and WOM
sample, as a more heterogeneous sample, is more likely to activity is stronger (weaker) for studies in the service
increase error variance, which would also produce weaker (product) context.
effect sizes (Fern and Monroe 1996). However, it is
difficult to predict which of these sources of error is bigger
than the other. Thus, based exclusively on these influences, The different roles of satisfaction and loyalty for favorable
we might not be sure about which type of sample would and unfavorable WOM
have a weaker or stronger mean effect.
Considering the particular context of WOM activity and In the previous section, we proposed WOM valence as a
the influence of satisfaction and loyalty as antecedents, we moderator, suggesting that for positive WOM the effect of
could argue that students have greater dependence on their the antecedents would be a direct one, whereas for negative
satisfaction and loyalty levels in order to provide recommen- WOM the effect would be an inverse one. But we also have
dations. The reason is that students are more susceptible to reason to expect that the antecedent with the strongest
reference group influences (Park and Lessig 1977), with effect on WOM differs for positive and negative WOM.
greater chance that they actively search for information from Specifically, research by Sweeney et al. (2005) has
opinion leaders. In consequence, students will be more likely empirically demonstrated that positive WOM is more
to provide WOM when this recommendation is driven by cognitive, more carefully considered, and more associated
satisfaction and loyalty, given that they have less-solidified with service quality-related comments. Conversely, nega-
cognitive structures than more mature individuals (Park and tive WOM is more emotionally based and is passed on
Lessig 1977). Given this influence in the WOM context, there more immediately. Thus, negative WOM might be more
is a potential consequence of a bias toward stronger effects of related to behavioral intentions (has a conative aspect) and
satisfaction and loyalty in a student sample when compared positive WOM could have a stronger evaluative (attitudi-
to a non-student sample. Based on this rationale, we propose: nal) component, driven by careful rational evaluation.
Considering that satisfaction is often approached by its
H6a: The relationship between satisfaction and WOM cognitive mechanisms and it often has an attitudinal
activity is stronger (weaker) for studies using a student component (Oliver 1980; Westbrook and Oliver 1991),
(non-student) sample. especially when authors measure cumulative satisfaction, it
586 J. of the Acad. Mark. Sci. (2008) 36:578–596

is reasonable to expect that satisfaction (instead of loyalty) The remaining 115 studies (242–127) could not be
will have a stronger relationship with positive WOM, since included for several reasons: (1) results were originated
this type of WOM also depends on cognitive evaluations from logit/probit models and no correlations were presented
and is carefully considered before being spread. (17 studies); (2) WOM was cited/measured in the context of
On the other hand, negative WOM is passed on more consumer complaining behavior, but it was not related to
immediately and to a greater number of people. Because of any of the variables of the proposed model (17 studies); (3)
the emotional nature of negative WOM (Schoefer and results were reported only in terms of number of recom-
Diamantopoulos 2008; Sweeney et al. 2005) and its greater mendations (11 studies); (4) results for WOM were
correspondence with intentions and behavior, we could expect integrated with customer lifetime value and no correlation
that (dis)loyalty, instead of (dis)satisfaction, has a stronger was presented (five studies); and (5) WOM was measured
relationship with negative WOM. This view is consistent but it was not related to any of the constructs of the
with Samson’s (2006) argument that brand loyalty should proposed model (65 studies). The inclusion rate of 53%
have a greater association with negative WOM because is comparable to other meta-analyses in marketing by
negative experiences with a brand usually have a greater Szymanski and Henard (2001; 59%) and Pan and Zinkhan
impact on repurchase intentions than do positive experiences. (2006; 56%).
Based on this discussion, we propose: We evaluated each study for measures of the relationship
between WOM, its antecedents, and the moderators. Two
H8: Satisfaction has a stronger relationship with
criteria were used for inclusion: (1) correlational studies
positive WOM than loyalty, while (dis)loyalty has a
had to present the correlation coefficient (r) or the
stronger relationship with negative WOM than does
standardized regression coefficient between WOM and
(dis)satisfaction.
any antecedent of the theoretical model and (2) studies of
group contrasts had to present the related statistic (t test,
F-ratios with one df in the numerator) for the relationship
Method between WOM and any antecedent. We could convert these
statistics to r coefficients using common guidelines in meta-
Search process and coding of studies analysis (see Lipsey and Wilson 2001, p. 198; Peterson and
Brown 2005).
We employed a literature search in different scientific We selected the correlation coefficient r as the primary
databases in order to identify the studies relevant to our effect size metric because it is easier to interpret and is a
research. First, we searched for “word of mouth” and scale-free measure, a common approach for meta-analytical
“recommendation” in abstracts and keywords in each of the review in the marketing literature (DelVecchio et al. 2006;
28 marketing-related journals that are included in the Web Eisend 2006, 2004; Franke and Park 2006; Janiszewski et
of Science Journal Citation Reports impact factor (JCR al. 2003; Palmatier et al. 2006; Pan and Zinkhan 2006). In
2006). This journal-by-journal search used Ebsco, Proquest, following Peterson and Brown’s (2005) recommendations
and Elsevier Science Direct scientific search engines. After that a meta-analysis should include the maximum number
that, we conducted a search on Proquest, Ebsco, and of effect sizes to make the results more generalizable, we
Emerald using the phrase “word of mouth” to look in also incorporated those studies that provided only the
abstracts and keywords. Then we searched for articles and standardized regression (beta) coefficients. We used the
working papers on the internet, using Google Scholar. formula suggested by Peterson and Brown (2005) to
Finally, we searched for dissertations and theses in Proquest estimate the correlations from the beta coefficients. The
and sent e-mails to the researchers in the domain asking for formula used is r ¼ :98b þ :05l, where λ is a variable that
their published and unpublished works. After completing equals 1 when β is non-negative and 0 when β is negative.
our search process in November 2007, we obtained a total We found that from the 162 observations in the data set,
of 591 studies done in 1955–2007. Of these 591 studies, 105 (65%) provided the Pearson correlation coefficients, 19
there was a body of 349 that included theoretical papers, (12%) supplied a t or F statistic of a group contrast, and 38
qualitative investigations, quantitative studies that merely (23%) were completed with the correlation derived from the
cited “word of mouth,” and studies using WOM items beta coefficient. Since the studies we reviewed had a broad
inside a loyalty construct (the “cocktail approach”), which array of research designs, using r as the common effect size
left 242 papers. From this amount (242 quantitative studies enabled us to integrate the various research designs. Before
measuring WOM), we selected 127 (53%) studies, which pooling the samples, we tested whether studies providing
produced 162 independent samples, 348 effect sizes, and a correlations would be different from those in which
cumulated N of 64,364 subjects. A list of the studies used in correlations were derived from beta coefficients but no
our empirical meta-analysis is available on request. significant difference was found for the link between
J. of the Acad. Mark. Sci. (2008) 36:578–596 587

satisfaction and WOM (mean r=.39; mean converted critical effect size (see Lipsey and Wilson 2001, p. 166;
r=.22; F=2.57; p<.112) and between loyalty and WOM Hunter and Schmidt 2004, p. 501). We used a level of .05
(mean r=.31; mean converted r=.43; F=.41; p<.526). as “just significant,” similar to Grewal et al. (1997).
All identified studies were then examined in terms of the Homogeneity of the effect size distribution was tested by
following relevant variables: authors, year, journal, method the Q statistic, distributed as a chi-square (see Lipsey and
(survey vs. experiment), design (cross-sectional vs. longitu- Wilson 2001, p. 115). If the null hypothesis of homogeneity
dinal), subjects (students vs. non-students), context (product is rejected, differences in effect sizes may be attributed to
vs. service), WOM valence (positive, negative, or mixed), factors other than sampling error alone, like moderating
WOM communication (sent, received, or both), WOM variables related to studies’ characteristics.
incidence (intention vs. behavior), scenarios (used vs. non- In each relationship in which the Q statistic was
used), total sample size, number of items used to measure significant (heterogeneity evidence), a moderator analysis
each variable, reliability of each variable, and effect sizes. was performed, considering the studies’ characteristics that
From our 162 observations, 76 (47%) used the term were coded based on information provided in the articles.
“repurchase intentions,” 33 (20%) used “loyalty,” two (1%) These moderators included method (survey vs. experiment),
measured real behavior, and 51 (32%) did not measure this design (cross-sectional vs. longitudinal), subjects (students
construct. Because most of these studies measured inten- vs. non-students), context (products vs. services), WOM
tions of loyalty, we considered repurchase intentions and valence (positive vs. negative), and WOM incidence
loyalty intentions together, naming the construct “loyalty.” (intention vs. behavior).
When studies measured the constructs using single items
Procedures for effect size integration (141 cases, or 32% of the 442 measurement situations) or
when reliability values were unavailable (28, or 9% of the
The developed database integrated the effect sizes accord- total of 301 situations using two or more items in the
ing to common guidelines in the meta-analysis literature measured variables), the reliabilities were estimated using
(Lipsey and Wilson 2001). Since the true relationship the Spearman–Brown procedure suggested by Hunter and
between variables is mainly influenced by sampling and Schmidt (2004, p. 311), which has been used in other meta-
measurement error, correlations were first weighed by the analyses in marketing (Franke and Park 2006; Grewal et al.
inverse variance and then by the inverse variance corrected 1997). For instance, of the 162 observations providing
for measurement error (cf. Lipsey and Wilson 2001, p. 110). effect sizes for WOM, 58 (36%) were based on single items
We first converted the original correlations to Fisher’s Z and had an estimated reliability of .711; 9 (6%) observa-
statistic before applying the variance weight and reliability tions were based on two items and had an estimated
adjustment. After computing the mean values and confi- reliability of .831; three (2%) observations were based on
dence intervals for Fisher’s Z values, we converted them three items and had an estimated reliability of .88; and two
back to r using the appropriate formulas (cf. Lipsey and (1%) observations were based on four items and had an
Wilson 2001, p. 63–64). We adjusted the effect sizes for estimated reliability of .908. This same procedure was
unreliability by applying the reliability weight directly to applied to the other constructs of the model. The advantage
the inverse variance using w’= w ×(rxx) × (ryy). In this of the Spearman–Brown procedure is that the estimated
approach, the effect size is corrected both for sampling reliability takes into account the number of items used to
and measurement error (Lipsey and Wilson 2001, p. 110). measure the construct, so that constructs with a greater
In view of this approach, we present our effect size number of items have higher estimated reliabilities.
integration in three stages, based first on observed
correlations, then on correlations weighed by sampling
error, and finally on correlations weighed by both sampling Results
and measurement error.
A confidence interval is presented for each effect size Descriptive statistics
and it is significant when it does not include zero. When the
mean effect size is significant, a fail safe N is calculated, After reviewing the sample size of each of the 162
estimating the number of non-significant and unavailable observations in the data set, we identified three outliers.
studies that would be necessary to bring the cumulative While the sample size of the remaining studies varied from
effect size to a non-significant value (known as the “file 40 to 1,405, the outliers had sample sizes of 6,201, 3,206,
drawer problem;” Rosenthal 1979). This statistic is an and 1,986 respondents. The distortion of these outliers
indication of the robustness of the results. The formula used would be greater for the link between perceived value and
is k×(r−rc)/rc, where k is the number of studies, r is the WOM, which would be .48 with the inclusion of the
mean effect size, and rc is the “just significant” level or outliers and .59 after the exclusion. This difference was
588 J. of the Acad. Mark. Sci. (2008) 36:578–596

caused by the observed effect size of .35 for the link Examining the confidence intervals of the antecedents of
between perceived value and WOM in the outlier with a WOM, we notice that the links satisfaction–WOM and
greater sample size (6,201). All analyses of effect size loyalty–WOM have the lowest dispersion around the mean
integration were performed after the outliers were excluded. effect. The file drawer N was also higher (greater than 510)
Because one of the three outliers came from a study that for these links, indicating that a higher number of studies
provided more than one observation in the data set, the with non-significant effect size would be needed in order to
number of observations, after excluding the outliers, reduce the mean effect size to a level of “just significant.”
reduced from 162 to 159, but the number of studies from In other words, this result is an indication that the mean
127 to 125. Thus, the analyses were based on 159 effect sizes of satisfaction–WOM and loyalty–WOM are
observations from 125 studies. more unlikely to be null, as the greater the file drawer
Table 1 presents the results for the integration of effect number, the greater our confidence that the mean effect size
sizes of the antecedents of WOM. As we show in Table 1, is not null. For example, to bring the significant link
the correlates of WOM with greater frequency in the satisfaction–WOM down to the level of .05 (used as the
literature are satisfaction and loyalty, with 113 and 75 “just significant” level), it would be necessary to find 844
observations respectively. Following Lipsey and Wilson’s studies with null-results to be included in our analysis. Even
(2001) suggestion for analyzing the magnitude of effect the lower file drawer N of 91 studies for the antecedent trust
sizes (r<.10 as small; r=.25 as medium, and r>.40 as large suggests that it is unlikely there is this number of studies
effect size), we can see in Table 1 that the antecedents of finding null results for the effects of trust on WOM.
WOM have medium-to-large mean effects for the integrated But a highly heterogeneous subset of effect sizes in the
effect size (corrected for sampling and measurement error), links satisfaction–WOM and loyalty–WOM—as indicated
with the relationships varying from .39 (loyalty–WOM) to by the Q statistic (see Table 1) and the greater ranges (see
.66 (commitment–WOM). All the proposed relationships columns of minimum and maximum)—also indicates that
were found to be significant, irrespective of the number of moderating variables might contribute to explaining the
observations. Thus, hypotheses H1a, H1b, H1c, H1d, H1e, variance in these effect sizes. The analysis of moderators is
and H1f were supported. presented next.
We note that only one of the 15 observations in the
relationship between quality and WOM dealt with the Moderating effects
quality of a product (an e-CRM software). This observation
presented a correlation of .53 and a sample of 143 in the In order to test if the studies’ characteristics could
dataset. Table 1 shows that the simple average r was .50 contribute to explaining the variability in the effect sizes,
while the adjusted r was .53, demonstrating that the only we conducted an analysis of moderation using regression
study about products presented a relationship for the link analysis, with effect sizes as dependent variables (correla-
quality-WOM very similar to that of studies investigating tions corrected for reliability) and moderators as indepen-
services. dent dummy-coded variables, a common approach for

Table 1 Descriptive statistics for effect size integration of the antecedents of WOM

Antecedents ka Ob Nc Min. Max. Simple Sample- Sample- Sig. LCI UCI Q testd Sig File
average weighed weighed drawer
r adjusted reliability N
average r adjusted r

Satisfaction 89 113 34,828 −.76 .95 .44 .42 .42 .000 .41 .43 9,945.65 .000 844
Loyalty 62 75 25,872 −.83 .93 .38 .37 .39 .000 .38 .40 5,811.34 .000 510
Quality 9 15 2,797 −.31 .74 .50 .53 .53 .000 .50 .56 225.65 .000 144
Commitment 8 12 5,391 .21 .78 .54 .60 .66 .000 .64 .68 245.38 .000 146
Trust 9 11 4,659 .18 .76 .44 .45 .46 .000 .44 .49 137.36 .000 91
Perceived 10 10 5,096 .11 .97 .66 .56 .59 .000 .57 .61 716.87 .000 108
Value

File Drawer number was attenuated at .05


LCI lower confidence interval, UCI upper confidence interval, Sig. significance, Min. minimum, Max. maximum, WOM word-of-mouth
a
Number of studies
b
Number of observations
c
Combined N
d
Test of homogeneity
J. of the Acad. Mark. Sci. (2008) 36:578–596 589

testing moderators (Eisend 2006; Szymanski and Henard We see from the findings in Table 2 that the regression
2001). An advantage of the regression approach is that it models are relatively free of collinearity, since the maxi-
takes into account the variability in the other moderators as mum variance inflation factor (VIF) was below the
well. This procedure is not recommended if there are few recommended threshold of 10 (Hair et al. 1998). The
effect sizes, namely, a ratio of less than five observations models for the satisfaction–WOM and loyalty–WOM
per independent variable, because results might be biased relationships presented a high proportion (about 77%) of
by low statistical power and capitalization on sampling variance explained by the predictors. These two models
error (Hunter and Schmidt 2004, p. 70). Because the links were statistically significant at .001 (see Model F and p
satisfaction–WOM and loyalty–WOM are the ones with the level).
greater number of observations (90 and 66, respectively), For the satisfaction–WOM relationship, we see from
we conducted a separate regression for each of them. The Table 2 that two moderators were significant predictors.
results are presented in Table 2. The first significant moderator was the WOM valence
Given that WOM valence had three categories (negative, (negative vs. positive). The significant positive coefficient
positive, and mixed), one alternative was to use two (β=.90, p<.001) means that studies measuring positive
dummies in the regression model, namely (1) WOM (negative) WOM presented a direct (inverse) effect of
positive (1 for positive WOM and 0 otherwise) and (2) satisfaction on WOM, supporting H2a. The second signif-
WOM negative (1 for negative WOM and 0 otherwise). icant moderator was study design, supporting H5a. The
However, this procedure would bias the results because negative coefficient (β=−.10, p<.071) reveals that the
mixed WOM was more similar to positive WOM (both effect of satisfaction on WOM was stronger for cross-
presented a positive mean for satisfaction–WOM). Thus, sectional studies (.42) than for longitudinal studies (.38). A
we decided to drop the “mixed WOM” category and use total of 90 observations presented completed information
one dummy considering only negative vs. positive WOM. for all predictors and the dependent variable.

Table 2 Regression results for the moderator analysis

Hypotheses Moderator Variable Antecedent of WOM

Satisfaction Loyalty

H2a, H2b WOM valence (0 = negative; 1 = positive)


Beta .90*** .88***
Number of correlations: negative, positive 19, 71 12, 54
Mean effect size: negative WOM; positive WOM −.51; .62 −.58; .54
H3a, H3b WOM incidence (0 = intention; 1 = behavior)
Beta −.05 −.13**
Number of correlations: intention, behavior 84, 6 57, 9
Mean effect size: WOM intention; WOM behavior – .42; .18
H4a, H4b Method (0 = experiment; 1 = survey)
Beta .01 −.05
Number of correlations: experiment, survey 11, 79 2, 64
H5a, H5b Design (0 = cross-sectional; 1 = longitudinal)
Beta −.10* −.14**
Number of correlations: cross-sectional, longitudinal 81, 9 58, 8
Mean effect size: WOM cross-sectional; WOM longitudinal .42; .38 .39; .29
H6a, H7b Subjects (0 = students; 1 = non-students)
Beta .01 −.03
Number of correlations: students, non-students 20, 70 10, 56
H7a, H7b Context (0 = product; 1 = service)
Beta −.05 .01
Number of correlations: product, service 11, 79 8, 58
R2 adjusted (n listwise) .784 (90) .768 (66)
Model F (p level) 54.79 (.000) 36.8 (.000)
Maximum variance inflation factor 1.81 1.24

Standardized coefficients are presented; mean effect sizes are displayed for each category of the significant moderators
*p<.10, **p<.05, ***p<.01
590 J. of the Acad. Mark. Sci. (2008) 36:578–596

The link loyalty–WOM presented a similar pattern when while (dis)loyalty has a stronger relationship with negative
considering that (1) studies measuring positive (vs. nega- WOM than does (dis)satisfaction. The segmented results
tive) WOM presented a direct (inverse) mean effect of are presented in Table 3.
loyalty on WOM (β=.88, p<.001) and (2) longitudinal We can see in Table 3 that for positive WOM, the effects
studies presented a lower mean effect size (.29) than cross- of both antecedents are positive, although satisfaction has a
sectional investigations (.39), with a significant regression stronger effect (r=.62) when compared to loyalty (r=.54),
coefficient (β=−.14, p<.031). These findings support H2b with a significant difference (z=6.82, p<.001). Conversely,
and H5b. Furthermore, WOM incidence was significant for for negative WOM conditions, the effects of loyalty on
loyalty–WOM (β=−.13, p<.037), with the negative coef- WOM are stronger (r = −.58) when compared to the
ficient indicating that the association between loyalty and influence of satisfaction (r= −.51), with a significant
WOM was stronger (.42) for the studies measuring difference (z=2.76, p<.003). These findings support H8.
intention of WOM and weaker (.18) for those measuring a An additional finding, which is in agreement with the
reported behavior of WOM, supporting H3b. However, the previous findings, is that the condition of mixed WOM
WOM incidence moderator was not significant in the produced a positive mean effect size for both satisfaction
satisfaction–WOM relationship, not supporting H3a. and loyalty, and the effect of satisfaction was stronger
We can also see from Table 2 that the influence of (r=.42) when compared to loyalty (r=.30), with a significant
satisfaction and loyalty on WOM was not different between difference (z=3.30, p<.001).
studies (1) using experimental versus survey method (no Even if we applied a Bonferroni correction and lowered
support for H4), (2) researching students or non-student the alpha level of .05 by the number of comparisons, which
subjects (no support for H6), and (3) investigating the would set the alpha for significance at .017 (.05/3=.017),
product or the service context (no support for H7). the findings would still support the difference between
It is interesting to note that for both the satisfaction– satisfaction–WOM and loyalty–WOM for both positive
WOM and loyalty–WOM relationships there were strong WOM (p<.001) and negative WOM (p<.003). We discuss
differences between conditions of positive and negative these findings next.
WOM. These analyses supported the relevant role of WOM
valence as a moderator. Next we test whether a change in
the WOM valence changes the antecedent that is most Discussion and conclusion
strongly associated with WOM activity (H8).
The meta-analysis we report in this study provides a
The different roles of satisfaction and loyalty quantitative integration of the main constructs that are
associated with WOM and a summary of the relationships
From the total of 159 observations, 101 (63%) dealt with between these constructs and WOM. Although there have
positive WOM, 31 (20%) with mixed WOM, and 27 (17%) been a great number of studies investigating WOM, there is
with negative WOM. We proposed in H8 that satisfaction still a lack of integration for the bivariate relationships
has a stronger relationship with positive WOM than loyalty, involving WOM. Thus, the present meta-analysis seeks to

Table 3 Relative strength of satisfaction and loyalty for different conditions of WOM valence

WOM Antecedent ka Ob Nc Min. Max. Simple Sample- Sample- Sig. LCI UCI Q Testd Sig File
valence average weighed weighed drawer
r adjusted reliability N
average r adjusted r

Positive Sat 54 71 22,208 −.11 .95 .62 .61 .62 .000 .61 .63 2,332.72 .000 816
Loy 44 54 20,523 −.01 .93 .57 .52 .54 .000 .53 .55 1,942.33 .000 532
Negative Sat 17 19 5,900 −.76 −.14 −.45 −.49 −.51 .000 −.53 −.48 272.98 .000 173
Loy 10 12 3,676 −.83 .14 −.59 −.55 −.58 .000 −.60 −.55 312.38 .000 126
Mixed Sat 18 23 6,720 −.36 .85 .42 .41 .42 .000 .39 .45 907.98 .000 170
Loy 8 9 1,673 −.46 .77 .33 .25 .30 .000 .24 .35 221.83 .000 45

LCI lower confidence interval, UCI upper confidence interval, Sig. significance, Min. minimum, Max. maximum, Sat satisfaction, Loy loyalty
a
Number of studies
b
Number of observations
c
Combined N
d
Test of homogeneity
J. of the Acad. Mark. Sci. (2008) 36:578–596 591

fill this gap by presenting a conceptual framework of the and non-experimental studies (H4), student-based versus
antecedents and moderators of WOM and empirically non-student-based research (H6), and the service versus the
investigating the hypotheses of this model. Our contribu- product context (H7). Although the differences between
tions are derived from this meta-analytic review. these conditions have been proposed in previous meta-
First, the direct-effect analysis reveals that commitment analyses (Crosno and Dahlstrom 2008; Geyskens et al.
is the main correlate of WOM (r=.66), followed by 1998; Pan and Zinkhan 2006; Szymanski and Henard
perceived value (.59), quality (.53), trust (.46), satisfaction 2001), the findings from this literature have not provided
(.42), and loyalty (.39). Most of these mean effects can be manifest differences. Our findings show that when control-
interpreted as a large effect (greater than .40, Lipsey and ling for the influence of the other methodological moder-
Wilson 2001). These findings support the relevant role of ators, the effects of satisfaction and loyalty on WOM tend
the commitment variable in the customer’s propensity to to be similar whether the researcher conducts an experiment
spread recommendations (positive or negative) about the or a survey, uses students or not in the sample, and
provider, in agreement with Brown et al. (2005), Fullerton investigates services or products. What does have an effect
(2003), and Lacey et al. (2007). is the decision to investigate negative or positive WOM,
Second, findings from the multiple regression support measure WOM intention or behavior, and use a cross-
the moderating role of WOM valence (H2) because for sectional or a longitudinal design.
positive WOM valence, a direct relationship is found Sixth, our findings supported the hypothesis that
between the antecedents (satisfaction and loyalty) and satisfaction has a stronger relationship with positive
WOM activity. On the other hand, for negative WOM WOM than loyalty, whereas (dis)loyalty has a stronger
valence, an inverse association was found. This finding relationship with negative WOM than does (dis)satisfaction
supports Szymanski and Henard’s (2001) finding of a link (H8). This effect can be explained by the differences
between satisfaction and negative WOM and extends their between behavioral intentions and attitudes. In other words,
results to the association between loyalty and WOM. given that (1) positive WOM is more strongly related to
Third, WOM incidence (intention or behavior) was a cognitive evaluations and it is carefully considered before
significant moderator in the loyalty–WOM relationship being spread (Sweeney et al. 2005), and that (2) satisfaction
(H3b). As expected, evidence from the analysis shows that is often interpreted by its cognitive mechanisms and it often
the influence of loyalty on WOM is significantly lower for has an attitudinal component (Oliver 1980; Westbrook and
studies measuring reported behavior of WOM activity Oliver 1991), we could expect satisfaction to have a
when compared to those measuring intentions of WOM stronger relationship with WOM activity for positive
activity. Brown et al.’s (2005) study found a similar result WOM valence.
when measuring the influence of satisfaction on WOM On the other hand, because negative WOM is more
activity. This difference in the level of influence can be emotionally based, is passed on more immediately, and
explained by the attitudinal nature of the WOM intentions therefore has a greater association with an intentional/
measure. So, it is more likely that researchers measuring behavioral component (Schoefer and Diamantopoulos
WOM as the intention to recommend will find a stronger 2008; Sweeney et al. 2005), we can predict that (dis)loyalty
relationship between WOM and other measures, such as is more strongly associated with WOM activity for negative
loyalty intentions, because respondents often try to main- WOM valence than is (dis)satisfaction. This finding is
tain a cognitive consistency when answering questionnaire consistent with the notion that negative WOM is usually a
items (Podsakoff et al. 2003). behavior of customers who have switched providers (a
Fourth, the moderator analysis also revealed that the disloyalty behavior), and then spread negative evaluations
studies’ design (cross-sectional vs. longitudinal) also had a of the former provider as a means to reduce cognitive
significant influence in both satisfaction–WOM and loyalty– dissonance, to vent negative emotions, to warn others, and/
WOM associations (H5). As hypothesized, cross-sectional or to seek retaliation (Richins 1984; Sweeney et al. 2005;
studies presented stronger mean effects than longitudinal Wangenheim 2005).
ones. A possible explanation is that the influence of This support for H8 is consistent with a recent study
satisfaction and loyalty on WOM may dissipate over time which shows that emotionally based satisfaction is a
and if variables are measured just after the consumption stronger predictor of future behavioral intentions than
experience, they might capture a stronger association between traditional cognitive measures (Martin et al. 2008). It is
the variables. Another possibility is that the cross-sectional also consistent with the proposition that consumers are
studies could suffer from common-method bias, which would more likely to engage in WOM when they experience
increase the estimated correlations (Podsakoff et al. 2003). significant emotional experiences (Dick and Basu 1994;
Fifth, there was no difference in the effects of satisfac- Söderlund and Rosengren 2007). Our meta-analytic review
tion and loyalty on WOM when comparing experimental empirically confirms that more emotionally based WOM
592 J. of the Acad. Mark. Sci. (2008) 36:578–596

presents a stronger association with customers’ behavioral and customer experience, could not be included in our
intentions. analysis, as is the case in other meta-analyses (see for example
As the main contribution for managers, the present meta- Kirca et al. 2005). Moreover, our moderating analysis could
analysis shows that satisfaction is more associated with be performed only for the antecedents of WOM with the
positive WOM than loyalty, whereas (dis)loyalty is more greatest number of observations (satisfaction and loyalty).
associated with negative WOM than is (dis)satisfaction. Another possible limitation is the fact that the reviewed
This finding reinforces the notion that satisfied customers studies might have methodological weaknesses other than
are not necessarily loyal (Matos et al. 2007; Reichheld the sample size and the measurement error taken into
1994) and it expands this knowledge by empirically account in our meta-analysis. Considering, for example,
demonstrating that negative and positive WOM are differ- that studies come from different sources, such as disserta-
ently associated with key marketing variables (satisfaction tions and papers published in journals of different quality,
and loyalty). While negative WOM has a strong behavioral decisions related to design, data collection, and analyses
component, positive WOM is more evaluative and attitudi- could have an influence on the obtained results. These
nal-based. An implication of this finding is that managers extraneous sources of error might also contribute to the
might “reap” the benefits of positive WOM in the middle or variability in the effect sizes across studies. Unfortunately,
long term as customer satisfaction is “sown.” On the other these factors are difficult, if not impossible, to control in the
hand, the damages of negative WOM can be experienced in context of a meta-analysis.
the short term if customers are dissatisfied and switch Future research should further develop the WOM
providers. literature by integrating the perspectives of the WOM
The study’s support for the significant role of customer recipient and WOM giver, as conducted by Gilly et al.
commitment as an important predictor of WOM activity is (1998). An extension to this approach could include those
another of its contributions to managerial practice. After (moderating) variables that are relevant to communication
combining studies and controlling for the effects of theory (Hovland et al. 1953) but which have been omitted
sampling and measurement error, the strongest effect of in WOM research, such as source credibility, source
commitment (.66) on WOM indicates that the most attractiveness, message congruence, message repetition,
important way for managers to increase customers’ likeli- situational involvement of the recipient, and risk perceived
hood of favorable recommendations could be to increase by the recipient. These variables should be relevant in
customers’ commitment to the organization. Fullerton understanding how the received WOM influences custom-
(2003) has demonstrated that the positive effect of ers’ propensity to pass the information to others.
commitment on retention is more pronounced when Another aspect worthy of future investigation is the
customers are committed through shared values and affective mechanisms related to the situations in which
identification rather than dependence and switching costs. customers provide either positive or negative WOM, since
Thus, managers could increase customers’ commitment by commitment presented a significant association with WOM,
improving the identification and the values shared between and commitment usually reflects a positive affect towards
customers and the company as an incentive for customers’ the organization. Correspondingly, future investigations
advocacy. about WOM and its relationship with satisfaction and
loyalty could be extended to include specific emotions,
Limitations and future research directions such as anger, regret, frustration, and disappointment, in
order to understand the likely emotional and behavioral
Several limitations that are common to meta-analytic aspect of negative WOM when compared to the more
reviews in the marketing literature are also present here. cognitive positive WOM.
For instance, we could not include studies that measured Moreover, longitudinal research could examine whether
WOM as a component of a broader loyalty or behavioral the influence of satisfaction and loyalty on WOM dissipates
intentions construct because of the “cocktail approach” over time or if the stronger relationships found for the
used in measurement (Söderlund 2006). We were also not cross-sectional studies are related to the predictor and
able to include the type of WOM (sent vs. received) as a outcome variables being measured at the same point in time
moderator in the regression because the few cases of (common-method bias). Considering that WOM researchers
received WOM were from studies that did not have are often investigating customers’ relationships with pro-
complete information for the other moderators (the variable viders, collecting data at more than one point in time might
became a constant in the listwise procedure and had to be help in producing a more realistic picture of the relation-
excluded). Also, our analysis is limited to moderators used ships between customers and companies.
in existing studies, and other variables that have been In addition, future research should also test whether the
suggested as moderators for WOM, such as switching costs practice of measuring WOM as an intention of recommen-
J. of the Acad. Mark. Sci. (2008) 36:578–596 593

dation (instead of a manifested real behavior) is leading behavior and repatronage intentions. Journal of Retailing, 69(4),
399–428.
researchers to believe in a stronger effect when in fact the
Bloemer, J., de Ruyter, K., & Wetzels, M. (1999). Linking perceived
effect is smaller. Indeed, most consumer researchers grapple service quality and service loyalty: a multi-dimensional perspec-
with the question of whether to use intentions or behavior tive. European Journal of Marketing, 33(11/12), 1082–1106.
in measuring behavioral constructs like WOM. Our review Bolton, R. N., & Drew, J. H. (1991). A longitudinal analysis of the
impact of service changes on customer attitudes. Journal of
suggests a large and significant difference (.42 versus .18)
Marketing, 55(1), 1–10.
between these approaches in the WOM context. Hence, Bonifield, C., & Cole, C. (2007). Affective responses to service
researchers should be aware of the serious limitation of failure: anger, regret, and retaliatory versus conciliatory
measuring only intentions and make efforts to measure real responses. Marketing Letters, 18(1), 85–99.
Boulding, W., Kalra, A., Staelin, R., & Zeithaml, V. A. (1993). A
behavior, or even use both intention and behavior measures dynamic process model of service quality: from expectations to
in order to assess the differences between them. behavioural intentions. Journal of Marketing Research, 30(1), 7–
Furthermore, researchers must ask whether the effects 27.
suggested above are influenced by the context of online Brown, T. J., Barry, T. E., Dacin, P. A., & Gunst, R. F. (2005).
Spreading the word: investigating antecedents of consumers’
WOM. Although previous studies have investigated elec-
positive word-of-mouth intentions and behaviors in a retailing
tronic WOM (Gruen et al. 2006; Hennig-Thurau et al. context. Journal of the Academy of Marketing Science, 33(2),
2004; Stewart and Pavlou 2002), more research is needed to 123–138.
investigate the issues presented here in the online WOM Carpenter, J. M., & Fairhurst, A. (2005). Consumer shopping value,
satisfaction, and loyalty for retail apparel brands. Journal of
context. It is possible that the differential effect of
Fashion Marketing and Management, 9(3), 256–269.
satisfaction and loyalty on negative and positive WOM is Chevalier, J., & Mayzlin, D. (2006). The effect of word of mouth on
even more pronounced in an online environment, given the sales: online book reviews. Journal of Marketing Research, 43
great power of dissatisfied customers to use the Internet to (3), 345–354.
Crosno, J. L., & Dahlstrom, R. (2008). A meta-analytic review of
vent negative emotions, to warn other customers, or to seek opportunism in exchange relationships. Journal of the Academy
retaliation. of Marketing Science, 36(2), 191–201.
In summary, we provide insight into the antecedents of DelVecchio, D., Henard, D. H., & Freling, T. H. (2006). The effect of
WOM activity, the approaches and studies’ characteristics sales promotion on post-promotion brand preference: a meta-
analysis. Journal of Retailing, 82(3), 203–213.
that moderate these relationships, and how the influence of
Dick, A. S., & Basu, K. (1994). Customer loyalty: toward an
satisfaction is more pronounced for positive WOM and integrated conceptual framework. Journal of the Academy of
(dis)loyalty for negative WOM. We hope these findings Marketing Science, 22(2), 99–113.
provide renewed insights for WOM researchers and Duan, W., Gu, B., & Whinston, A. B. (2008). The dynamics of online
word-of-mouth and product sales – an empirical investigation of
advance the theory of WOM communications in marketing.
the movie industry. Journal of Retailing, 84(2), 233–242.
Durvasula, S., Lysonski, S., Mehta, S. C., & Tang, B. P. (2004).
Acknowledgments We thank the Brazilian Funding Council for Forging relationships with services: the antecedents that have an
Research (CNPq and CAPES) and the Graduate School of Manage- impact on behavioral outcomes in the life insurance industry.
ment (PPGA-EA-UFRGS) for their financial support. We are also Journal of Financial Services Marketing, 8(4), 314–326.
thankful to the editor, David W. Stewart, and the three anonymous Dwyer, F. R., Schurr, P., & Oh, S. (1987). Developing buyer-seller
JAMS reviewers for their insightful comments. relationships. Journal of Marketing, 51, 11–27 April.
Eisend, M. (2004). Is it still worth to be credible? A meta-analysis
of temporal patterns of source credibility effects in marketing.
In B. Kahn, & M. F. Luce (Eds.), Advances in consumer
research, vol. 31 (pp. 352–357). Urbana: Association of
References
Consumer Research.
Eisend, M. (2006). Two-sided advertising: a meta-analysis. Interna-
Anderson, E. W. (1998). Customer satisfaction and word-of-mouth. tional Journal of Research in Marketing, 23(2), 187–198.
Journal of Service Research, 1(1), 1–14. Engel, J. F., Blackwell, R. D., & Miniard, P. W. (1995). Consumer
Arndt, J. (1967). Word-of-mouth advertising and informal vommuni- behavior (8th ed.). Orlando: Dryden.
cation. In D. Cox (Ed.), Risk taking and information handling in Fern, E. F., & Monroe, K. B. (1996). Effect-size estimates: issues and
consumer behaviour. Boston: Harvard University. problems in interpretation. Journal of Consumer Research, 23,
Arnett, D. B., German, S. D., & Hunt, S. D. (2003). The identity 89–105 September.
salience model of relationship marketing successes: the case of Franke, G. R., & Park, J.-E. (2006). Salesperson adaptive selling
nonprofit marketing. Journal of Marketing, 67, 89–105 April. behavior and customer orientation: a meta-analysis. Journal of
Assmus, G., Farley, J. U., & Lehmann, D. R. (1984). How advertising Marketing Research, 43(4), 693–702.
affects sales: meta-analysis of econometric results. Journal of Fullerton, G. (2003). When does commitment lead to loyalty? Journal
Marketing Research, 21(1), 65–74. of Service Research, 5(4), 333–344.
Bearden, W. O., & Teel, J. E. (1983). Selected determinants of Garbarino, E., & Johnson, M. S. (1999). The different roles of
consumer satisfaction and complaint reports. Journal of Market- satisfaction, trust, and commitment in customer relationships.
ing Research, 20(1), 21–28. Journal of Marketing, 63(2), 70–87.
Blodgett, J. G., Granbois, D. H., & Walters, R. G. (1993). The effects Geyskens, I., Steenkamp, J.-B. E. M., & Kumar, N. (1998). General-
of perceived justice on complainants’ negative word-of-mouth izations about trust in marketing channel relationships using
594 J. of the Acad. Mark. Sci. (2008) 36:578–596

meta-analysis. International Journal of Research in Marketing, Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis:
15(3), 223–249. correcting error and bias in research findings (2nd ed.). Beverly
Gilly, M. C., Graham, J. L., Wolfinbarger, M. F., & Yale, L. J. (1998). Hills: Sage.
A dyadic study of interpersonal information search. Journal of Janiszewski, C., Noel, H., & Sawyer, A. G. (2003). A meta-analysis of
the Academy of Marketing Science, 26(2), 83–100. the spacing effect in verbal learning: implications for research on
Gounaris, S., & Stathakopoulos, V. (2004). Antecedents and con- advertising repetition and consumer memory. Journal of Con-
sequences of brand loyalty: an empirical study. Journal of Brand sumer Researcher, 30(1), 138–149.
Management, 11(4), 283–306. Jones, M. A., & Reynolds, K. E. (2006). The role of retailer interest
Gremler, D. D., & Gwinner, K. P. (2000). Customer-employee rapport on shopping behavior. Journal of Retailing, 82(2), 115–126.
in service relationships. Journal of Service Research, 3(1), 82– Jones, M. A., Reynolds, K. E., Mothersbaugh, D. L., & Beatty, S. E.
104. (2007). The positive and negative effects of switching costs on
Gremler, D. D., Gwinner, K. P., & Brown, S. W. (2001). Generating relational outcomes. Journal of Service Research, 9(4), 335–355.
positive word-of-mouth communication through customer- Jones, T. O., & Sasser, W. E., Jr. (1995). Why satisfied customers
employee relationships. International Journal of Service Industry defect. Harvard Business Review, 73, 88–99 November-December.
Management, 12(1), 44–59. Jones, T., & Taylor, S. F. (2007). The conceptual domain of service
Grewal, D., Kavanoor, S., Fern, E. F., Costley, C., & Barnes, J. loyalty: how many dimensions? Journal of Services Marketing,
(1997). Comparative versus noncomparative advertising: a meta- 21(1), 36–51.
analysis. Journal of Marketing, 61(4), 1–15. Katz, E., & Lazarsfeld, P. F. (1955). Personal influence. Glencoe:
Grinstein, A. (2008). The effect of market orientation and its Free.
components on innovation consequences: a meta-analysis. Kau, A.-K., & Loh, E. W.-Y. (2006). The effects of service recovery
Journal of the Academy of Marketing Science, 36(2), 166–173. on consumer satisfaction: a comparison between complainants
Gruen, T. W., Osmonbekov, T., & Czaplewski, A. J. (2006). eWOM: and non-complainants. Journal of Services Marketing, 20(2),
the impact of customer-to-customer online know-how exchange 101–111.
on customer value and loyalty. Journal of Business Research, 59 Keiningham, T. L., Cooil, B., Aksoy, L., Andreassen, T. W., &
(4), 449–456. Weiner, J. (2007). The value of different customer satisfaction
Gruen, T. W., Summers, J. O., & Acito, F. (2000). Relationship and loyalty metrics in predicting customer retention, recommen-
marketing activities, commitment, and membership behaviors in dation, and share-of-wallet. Managing Service Quality, 17(4),
professional associations. Journal of Marketing, 64(3), 34–49. 361–384.
Hair, J. F., Jr., Anderson, R. E., Tatham, R. L., & Black, W. C. (1998). Kirca, A. H., Jayachandran, S., & Bearden, W. O. (2005). Market
Multivariate data analysis. Upper Saddle River: Prentice Hall. orientation: a meta-analytic review and assessment of its
Halstead, D. (2002). Negative word of mouth: Substitute for or antecedents and impact on performance. Journal of Marketing,
supplement to consumer complaints? Journal of Consumer 69, 24–41 April.
Satisfaction, Dissatisfaction and Complaining Behavior, 15(1), Lacey, R., Suh, J., & Morgan, R. M. (2007). Differential effects of
1–12. preferential treatment levels on relational outcomes. Journal of
Halstead, D., & Dröge, C. (1991). Consumer attitudes toward Service Research, 9(3), 241–256.
complaining and the prediction of multiple complaint responses. Lam, S. Y., Shankar, V., Krishna Erramilli, M., & Murthy, B. (2004).
Advances in Consumer Research, 18, 210–216. Customer value, satisfaction, loyalty, and switching costs: an
Harrison-Walker, L. J. (2001). The measurement of word-of-mouth illustration from a business-to-business service context. Journal
communication and an investigation of service quality and Academy of Marketing Science, 32(3), 293–311.
customer commitment as potential antecedents. Journal of Liljander, V., & Strandvik, T. (1997). Emotions in service satisfaction.
Service Research, 4(1), 60–75. International Journal of Service Industry Management, 8(2),
Hartline, M. D., & Jones, K. C. (1996). Employee performance cues 148–169.
in a hotel service environment: influence on perceived service Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis.
quality, value, and word-of-mouth intentions. Journal of Business Beverly Hills: Sage.
Research, 35(3), 207–215. Macintosh, G. (2007). Customer orientation, relationship quality, and
Heckman, R., & Guskey, A. (1998). The relationship between alumni relational benefits to the firm. Journal of Services Marketing, 21
and university: toward a theory of discretionary collaborative (3), 150–159.
behavior. Journal of Marketing Theory and Practice, 6(2), 97– Martin, D., O’Neill, M., Hubbard, S., & Palmer, A. (2008). The role
112. of emotion in explaining consumer satisfaction and future
Heitmann, M., Lehmann, D. R., & Herrmann, A. (2007). Choice goal behavioural intention. Journal of Services Marketing, 22(3),
attainment and decision and consumption satisfaction. Journal of 224–236.
Marketing Research, 44, 234–250 May. Matos, C. A., Henrique, J. L., & Rossi, C. A. V. (2007). Service
Henard, D. H., & Szymanski, D. M. (2001). Why some new products recovery paradox: a meta-analysis. Journal of Service Research,
are more successful than others. Journal of Marketing Research, 10(1), 60–77.
38(3), 362–375. Maxham, J. G., III (2001). Service recovery’s influence on consumer
Hennig-Thurau, T., Gwinner, K. P., & Gremler, D. D. (2002). satisfaction, positive word-of-mouth, and purchase intentions.
Understanding relationship marketing outcomes: An integration Journal of Business Research, 54(1), 11–24.
of relational benefits and relationship quality. Journal of Service Maxham, J. G., III, & Netemeyer, R. G. (2002a). A longitudinal study
Research, 4(3), 230–247. of complaining customers’ evaluations of multiple service
Hennig-Thurau, T., Gwinner, K. P., Walsh, G., & Gremler, D. D. failures and recovery efforts. Journal of Marketing, 66(4), 57–71.
(2004). Electronic word-of-mouth via consumer-opinion plat- Maxham, J. G., III, & Netemeyer, R. G. (2002b). Modeling customer
forms: what motivates consumers to articulate themselves on the perceptions of complaint handling over time: the effects of
internet? Journal of Interactive Marketing, 18(1), 38–52. perceived justice on satisfaction and intent. Journal of Retailing,
Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication 78(4), 239–252.
and persuasion: psychological studies of opinion change. New Maxham, J. G., III, & Netemeyer, R. G. (2003). Firms reap what they
Haven: Yale University Press. sow: the effects of shared values and perceived organizational
J. of the Acad. Mark. Sci. (2008) 36:578–596 595

justice on customers’ evaluations of complaint handling. Journal Reynolds, K. E., & Beatty, S. E. (1999). Customer benefits and
of Marketing, 67, 46–62 January. company consequences of customer-salesperson relationships in
Mazzarol, T., Sweeney, J. C., & Soutar, G. N. (2007). Conceptualizing retailing. Journal of Retailing, 75(1), 11–32.
word-of-mouth activity, triggers and conditions: an exploratory Richins, M. L. (1983). Negative word-of-mouth by dissatisfied
study. European Journal of Marketing, 41(11/12), 1475–1494. consumers: a pilot study. Journal of Marketing, 47(1), 68–
McKee, D., Simmers, C. S., & Licata, J. (2006). Customer self- 78.
efficacy and response to service. Journal of Service Research, 8 Richins, M. L. (1984). Word-of-mouth communications as negative
(3), 207–220. information. In T. C. Kinnear (Ed.), Advances in consumer
Mittal, V., Kumar, P., & Tsiros, M. (1999). Attribute-level performance, research (vol. 11, (pp. 697–702)). Ann Arbor: Association for
satisfaction, and behavioral intentions over time: a consumption- Consumer Research.
system approach. Journal of Marketing, 63(2), 88–101. Rosenthal, R. (1979). The ‘file drawer problem’ and tolerance for null
Moorman, C., Deshpandé, R., & Zaltman, G. (1993). Factors affecting results. Psychological Bulletin, 86, 638–641.
trust in market research relationships. Journal of Marketing, 57 Samson, A. (2006). Understanding the buzz that matters: negative vs
(1), 81–101. positive word of mouth. International Journal of Market
Moorman, C., Zaltman, G., & Deshpandé, R. (1992). Relationships Research, 48(6), 647–657.
between providers and users of market research: the dynamics of Schoefer, K., & Diamantopoulos, A. (2008). The role of emotions in
trust within and between organizations. Journal of Marketing translating perceptions of (in)justice into postcomplaint behav-
Research, 29(3), 314–339. ioral responses. Journal of Service Research, Available at:
Morgan, R., & Hunt, S. (1994). The commitment-trust theory of <http://jsr.sagepub.com/cgi/rapidpdf/1094670508319091v1>.
marketing relationships. Journal of Marketing, 58(3), 20–38. Sheppard, B. H., Hartwick, J., & Warshaw, P. R. (1988). The theory of
Murray, K. B. (1991). A test of services marketing theory: consumer reasoned action: a meta-analysis of past research with recom-
information acquisition activities. Journal of Marketing, 55, 10– mendation for modifications and future research. Journal of
25 January. Consumer Research, 15(3), 325–343.
Oliver, R. L. (1980). A cognitive model of the antecedents and Sichtmann, C. (2007). An analysis of antecedents and consequences
consequences of satisfaction decisions. Journal of Marketing of trust in a corporate brand. European Journal of Marketing, 41
Research, 17(4), 460–469. (9/10), 999–1015.
Oliver, R. L. (1981). Measurement and evaluation of satisfaction Singh, J., & Sirdeshmukh, D. (2000). Agency and trust mechanisms in
processes in retail settings. Journal of Retailing, 57(3), 25–48. consumer satisfaction and loyalty judgments. Journal of the
Oliver, R. L. (1997). Satisfaction: a behavioral perspective on the Academy of Marketing Science, 28(1), 150–167.
consumer. New York: McGraw-Hill. Sirdeshmukh, D., Singh, J., & Sabol, B. (2002). Consumer trust,
Oliver, R. L., & Westbrook, R. A. (1993). Cognitive, affective, and value, and loyalty in relational exchanges. Journal of Marketing,
attribute bases of the satisfaction response. Journal of Consumer 66(1), 15–37.
Research, 20(3), 418–430. Söderlund, M. (1998). Customer satisfaction and its consequences on
Palmatier, R. W., Dant, R. P., Grewal, D., & Evans, K. R. (2006). customer behaviour revisited. International Journal of Service
Factors influencing the effectiveness of relationship marketing: a Industry Management, 9(2), 169–188.
meta-analysis. Journal of Marketing, 70(4), 136–153. Söderlund, M. (2006). Measuring customer loyalty with multi-item
Pan, Y., & Zinkhan, G. M. (2006). Determinants of retail patronage: a scales—a case for caution. International Journal of Service
meta-analytical perspective. Journal of Retailing, 82(3), 229–243. Industry Management, 17(1), 76–98.
Parasuraman, A., Zeithaml, V., & Berry, L. L. (1985). A conceptual Söderlund, M., & Rosengren, S. (2007). Receiving word-of-mouth
model of service quality and its implications for future research. from the service customer: An emotion-based effectiveness
Journal of Marketing, 49(4), 41–50. assessment. Journal of Retailing and Consumer Services, 14(1),
Parasuraman, A., Zeithaml, V., & Berry, L. L. (1988). SERVQUAL: a 123–136.
multiple-item scale for measuring consumer perception of service Stewart, D. W., & Pavlou, P. A. (2002). From consumer response to
quality. Journal of Retailing, 64(1), 12–41. active consumer: measuring the effectiveness of interactive
Park, C. W., & Lessig, V. P. (1977). Students and housewives: media. Journal of the Academy of Marketing Science, 30(4),
differences in susceptibility to reference group influence. Journal 376–396.
of Consumer Research, 4, 102–110 September. Straus, B., & Neuhaus, P. (1997). The qualitative satisfaction model.
Peterson, R. A., & Brown, S. P. (2005). On the use of beta coefficients International Journal of Service Industry Management, 8(3),
in meta-analysis. Journal of Applied Psychology, 90(1), 175–181. 236–249.
Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. Sundaram, D. S., Mitra, K., & Webster, C. (1998). Word-of-mouth
(2003). Common method biases in behavioral research: a critical communications: a motivational analysis. Advances in Consumer
review of the literature and recommended remedies. Journal of Research, 25, 527–531.
Applied Psychology, 88(5), 879–903. Swan, J. E., & Oliver, R. E. (1989). Postpurchase communications by
Price, L., & Arnould, E. J. (1999). Commercial friendships: service consumers. Journal of Retailing, 65(4), 516–533.
provider-client relationships in context. Journal of Marketing, 63, Sweeney, J. C., Soutar, G. N., & Mazzarol, T. (2005). The difference
38–56 October. between positive and negative word-of-mouth—emotion as a
Ranaweera, C., & Prabhu, J. (2003). On the relative importance of differentiator. In Proceedings of the ANZMAC 2005 Conference:
customer satisfaction and trust as determinants of customer Broadening the Boundaries (pp. 331–337). Perth, Australia:
retention and positive word of mouth. Journal of Targeting, University of Western Australia.
Measurement and Analysis of Marketing, 12(1), 82–90. Szymanski, D. M., & Henard, D. H. (2001). Customer satisfaction: a
Reichheld, F. F. (1994). Loyalty and the renaissance of marketing. meta-analysis of the empirical evidence. Journal of the Academy
Marketing Management, 2(4), 10–21. of Marketing Science, 29(1), 16–35.
Reynolds, K. E., & Arnold, M. J. (2000). Customer loyalty to the Voorhees, C. M., Brady, M. K., & Horowitz, D. M. (2006). A voice
salesperson and the store: examining relationship customers in an from the silent masses: an exploratory and comparative analysis
upscale retail context. The Journal of Personal Selling & Sales of noncomplainers. Journal of the Academy of Marketing
Management, 20(2), 89–98. Science, 34(4), 514–527.
596 J. of the Acad. Mark. Sci. (2008) 36:578–596

Wangenheim, F. v. (2005). Postswitching negative word of mouth. word-of-mouth communication. Psychology & Marketing, 24(8),
Journal of Service Research, 8(1), 67–78. 661–680.
Wangenheim, F. v., & Bayón, T. (2007). The chain from customer Zeelenberg, M., & Pieters, R. (2004). Beyond valence in customer
satisfaction via word-of-mouth referrals to new customer acqui- dissatisfaction: a review and new findings on behavioral
sition. Journal of the Academy of Marketing Science, 35(2), 233– responses to regret and disappointment in failed services. Journal
249. of Business Research, 57(4), 445–455.
Westbrook, R. A. (1987). Product/consumption based affective Zeithaml, V. A. (1988). Consumer perceptions of price, quality, and
responses and postpurchase processes. Journal of Marketing value: a means-end model and synthesis of evidence. Journal of
Research, 24(3), 258–270. Marketing, 52(3), 2–22.
Westbrook, R. A., & Oliver, R. L. (1991). The dimensionality of Zeithaml, V. A., Berry, L. L., & Parasuraman, A. (1993). The nature
consumption emotion patterns and consumer satisfaction. Jour- and determinants of customer expectations of service. Journal of
nal of Consumer Research, 18(1), 84–91. the Academy of Marketing Science, 21(1), 1–12.
Wetzer, I. M., Zeelenberg, M., & Pieters, R. (2007). ‘Never eat in that Zeithaml, V. A., Berry, L. L., & Parasuraman, A. (1996). The behavioral
restaurant, I Did!’: exploring why people engage in negative consequences of service quality. Journal of Marketing, 60(2), 31–47.

View publication stats

You might also like