The influence of word-of-mouth on attitudinal ambivalence during the higher education decision-making process

Abstract

This study investigates the influence of word-of-mouth (WOM) on consumers' attitudinal ambivalence in the context of higher education decision-making. Construal level theory (CLT) is combined with attitudinal ambivalence literature to generate hypotheses about how different types of WOM (i.e., praise and activity) received during the decision-making process reduce attitudinal ambivalence. The subsequent consequences of attitudinal ambivalence for decision-making are also studied. A two-wave survey of applicants to international higher education programs is used to test the hypotheses. This study contributes to the ambivalence literature by showing that different types of WOM information reduce attitudinal ambivalence depending on the temporal closeness of a choice and the consumption of a service. The findings have implications for the management of attitudinal ambivalence and WOM throughout the consumer decision-making process and consequently for assisting consumers in making choices.

Full text

1
THE INFLUENCE OF WORD-OF-MOUTH ON ATTITUDINAL AMBIVALENCE
DURING THE HIGHER EDUCATION DECISION-MAKING PROCESS
Abstract
This study investigates the influence of word-of-mouth (WOM) on consumers’ attitudinal
ambivalence in the context of higher education decision-making. Construal level theory
(CLT) is combined with attitudinal ambivalence literature to generate hypotheses about how
different types of WOM (i.e., praise and activity) received during the decision-making process
reduce attitudinal ambivalence. The subsequent consequences of attitudinal ambivalence for
decision-making are also studied. A two-wave survey of applicants to international higher
education programs is used to test the hypotheses. This study contributes to the ambivalence
literature by showing that different types of WOM information reduce attitudinal ambivalence
depending on the temporal closeness of a choice and the consumption of a service. The
findings have implications for the management of attitudinal ambivalence and WOM
throughout the consumer decision-making process and consequently for assisting consumers
in making choices.
Keywords: Consumer; Ambivalence; Attitude; Word-of-mouth; Decision-making process;
Choice

2
THE INFLUENCE OF WORD-OF-MOUTH ON ATTITUDINAL AMBIVALENCE
DURING THE HIGHER EDUCATION DECISION-MAKING PROCESS
1 INTRODUCTION
Each year, many young adults have to decide whether to pursue their master’s degree studies
at a particular higher education (HE) institution. In their decision-making process, these HE
applicants have to evaluate an institution based on a number of service attributes (Cubillo-
Pinilla et al., 2009; Joseph & Joseph, 1998; Soutar & Turner, 2002). For example, they may
consider whether an institution is reputable and highly ranked, whether it is located in a safe
country, the kinds of job prospects they could expect after graduating from the institution, or
whether the institution has an active student life. In many cases, some attributes are likely to
be evaluated negatively while others are evaluated positively, which forces the applicants to
make difficult trade-offs between the attributes (Soutar &Turner, 2002). For example, an HE
applicant might think that an institution offers high-quality education, but the costs of
studying there would be extremely high. Having both positive and negative evaluations of the
same consumption object is called attitudinal ambivalence (Priester & Petty, 1996), which is
likely to be particularly pronounced in high-involvement contexts (Puccinelli et al., 2009)
such as HE (Cubillo-Pinilla et al., 2009).
Attitudinal ambivalence has multiple implications for consumer behavior. Ambivalent
attitudes are less durable and impactful (Tormala & DeSensi, 2008), and worse predictors of
behavior than univalent (i.e., predominantly positive or negative) attitudes are (Glasman &
Albarracín, 2006). Furthermore, attitudinal ambivalence negatively influences behavioral
intentions (Costarelli & Colloca, 2004) and actual consumption (Berndsen & van der Pligt,

3
2004), and can result in residual doubt (Jewell et al., 2002) and reduced satisfaction (Olsen et
al., 2005) after choices are made. In addition, attitudinal ambivalence is a prominent concept
in consumer behavior because it is rare for a given object not to have some attributes that are
evaluated positively and others that are evaluated negatively (Fazio, 2007). Thus, attitudinal
ambivalence is problematic for marketers, and its reduction should be an important marketing
goal in HE institutions worldwide because of the increasing competition for students
(Durvasula et al., 2011). Based on the existing literature (Olsen et al., 2005), when applicants
have less pre-choice attitudinal ambivalence, they are likely to be more satisfied with their
chosen institution. Satisfaction, in turn, can be expected to increase their likelihood of
recommending the institution and decrease their likelihood of changing to another institution
or quitting their studies, based on previous research (Selnes, 1993).
Ambivalence is also problematic for HE applicants because choice-making (in this study, the
choice of whether to being studies in an institution), is challenging when one’s evaluations
towards the attitude object (HE institution) are conflicting (van Harreveld et al., 2009). One
way of reducing attitudinal ambivalence is by utilizing additional information, which may
enable the applicant to become more favorable or unfavorable towards the institution (Hodson
et al., 2001). However, the existing research does not show whether some types of
information are more useful in this regard than others in the multi-stage decision-making
process. This gap is important because consumers may use different types of information in
different stages of the decision-making process (Bettman & Park, 1980). Hence, the present
study uses construal level theory (CLT) to contribute to the attitudinal ambivalence research.
According to CLT, when the final choice of an object (in this case the HE institution)
becomes temporally closer, its evaluation becomes increasingly concrete and detailed (Trope

4
& Liberman, 2010). Thus, different types of information with varying levels of abstractness
might be beneficial at different stages of the HE decision-making process.
In the present study, this proposition is examined by analyzing WOM information because
WOM, which refers to the “informal communications between consumers concerning the
ownership, usage or characteristics of particular goods, services and/or their sellers” (De
Matos & Rossi, 2008, p. 578) is an important information source for consumers especially in
the service context (Bansal & Voyer, 2000; Mangold et al., 1999; Murray, 1991; Sweeney et
al., 2008). Services are perceived as more challenging than goods to evaluate prior to
purchase because they are intangible, heterogeneous, and perishable (Murray, 1992). WOM
can be used to simplify this complexity and reduce perceived risk (Berger, 2014).
Furthermore, consumers are likely to rely on WOM in making important decisions because it
is perceived as a trustworthy source of information (Berger, 2014). Hence, WOM plays a
significant role in shaping consumers’ attitudes (Brown & Reingen, 1987; Martin & Lueg,
2013) and service evaluations (Lim & Chung, 2011; Mangold et al., 1999; Murray, 1991;
Sweeney et al., 2008). Especially in the HE services context, WOM is a key source of
information (Chapman, 1981; Johnston, 2010), and HE applicants are persuaded by the
comments and advice of their friends and family members (Chapman, 1981; Mazzarol &
Soutar, 2002). Different WOM types have been identified in the existing literature, such as
content richness and negative WOM (Sweeney et al., 2012; Sweeney et al., 2014). In this
study, however, we focus on WOM praise and WOM activity because they vary in their level
of abstractness, which enables advancing ambivalence literature through CLT.
We therefore pose the following research question: How do different types of WOM
information with varying abstractness influence attitudinal ambivalence during two stages of

5
the HE decision-making process (search and choice)? Hence, the present study contributes to
the attitudinal ambivalence literature by demonstrating the reduction of attitudinal
ambivalence at different time points. This understanding is important because decision-
making often involves a multi-stage process (e.g., Puccinelli et al., 2009). Investigations of
only one point of the process yield only a partial understanding of consumption-related
phenomena, such as attitudinal ambivalence. However, by studying two types of information,
this study contributes to the attitudinal ambivalence literature by demonstrating that in terms
of attitudinal ambivalence reduction, information about different levels of abstraction is
required depending on the consumer’s temporal distance from the final choice. From a
managerial perspective, our results show which types of WOM should be promoted at
different stages of the decision-making process to assist consumers in attitudinal ambivalence
reduction, especially with regard to the marketing of HE services. Furthermore, we discuss
potential ways of promoting these different types of WOM. This paper will begin by
proposing a conceptual model and developing hypotheses. The model is then tested using
regression analysis, and the results are discussed from both theoretical and managerial
perspectives.
2 THEORETICAL BACKGROUND
2.1 Attitudinal ambivalence
An attitude refers to “general and enduring favorable or unfavorable feelings about, evaluative
categorizations of, and action predispositions toward stimuli” (Cacioppo & Berntson, 1994, p.
401). Traditionally, attitudes have been conceptualized as unidimensional, that is, as either
positive or negative (Jewell et al., 2002). However, attitudes toward an object can also consist

6
of both positive and negative components, meaning that they are ambivalent (Thompson et
al., 1995). Ambivalence is a property of attitudes (Jewell et al., 2002), meaning that it
operates similarly to an adjective. Hence, attitudes can be described as univalent or
ambivalent. In addition, a summary evaluation such as an attitude consists of components that
can involve emotions, beliefs, or previous behavioral experiences with the object (Fazio
1995). When there are both positive and negative components in the evaluative structure, the
attitude is ambivalent (Krosnick & Petty, 1995). Services are often evaluated based on several
attributes that can be perceived positively or negatively. For example, an HE applicant may
hold both negative beliefs (e.g., high costs of studying in the institution or inactive student
life) and positive beliefs (e.g., the institution is highly ranked and located in a country that has
a good academic reputation) about an HE institution. These conflicting beliefs then constitute
an ambivalent summary evaluation of the HE institution.
According to cognitive dissonance theory, people are inherently motivated to resolve
psychological conflicts (Festinger, 19641) mainly because they are uncomfortable (van
Harreveld et al., 2012). In the case of ambivalence, the uncomfortable feeling arises when
both positivity and negativity are simultaneously accessible, leading to the awareness of
attitudinal ambivalence (Newby-Clark et al., 2002). This case may apply particularly when
there is a need to commit to a choice, leading to uncertainty-induced physiological arousal
(van Harreveld et al., 2012). In addition, when the decision-making process involves a choice
(in the present case, whether to start studies in a particular HE institution or decline the offer),
an ambivalent consumer may anticipate regret for making the wrong choice. For example, the
1 Dissonance differs from ambivalence because it occurs after committing to a particular choice or behavior that
is in conflict with a person’s attitude, whereas ambivalence occurs before a person has committed to one option
(van Harreveld et al., 2009). Furthermore, dissonance involves a conflict between attitude and behavior, whereas
ambivalence involves a conflict between two attitudinal components, such as beliefs about an attitude object
(van Harreveld et al., 2009). Hence, dissonant individuals try to feel good about a choice that they have already
made, whereas ambivalent individuals have not yet committed to a choice and try to make the best possible
decision (van Harreveld et al., 2009).

7
HE applicant may anticipate the thought, “I should not have started my studies in this
institution”. Furthermore, attitudinal ambivalence involves both positive and negative
evaluations of the important aspects of an object (Jewell et al., 2002). Hence, HE applicants
are likely to be motivated to reduce attitudinal ambivalence during decision-making.
The existing literature has studied various coping mechanisms that may help people to resolve
attitudinal ambivalence and therefore make a favorable or unfavorable choice. When regret
about making a bad decision is anticipated and a decision is important, an accuracy
motivation is likely to prevail, leading to effortful problem-focused coping based on the need
to make the best possible decision (van Harreveld et al., 2012). Thus, people are likely to
increase their decision-making efforts in order to increase their confidence in making the
decision (van Harreveld et al., 2009). We expect this to be the case in HE decision-making
because of the importance and investment-like nature of the HE decision. One effortful way to
reduce attitudinal ambivalence is to acquire additional information (Hodson et al., 2001; Jonas
et al., 1997). Because WOM is a key source of information for HE applicants (Chapman,
1981; Johnston, 2010), we propose that it has an influential role in this process.
2.2 Word-of-mouth
WOM has been studied from both the sender and the receiver perspectives, and this study
focuses on the latter. Prior WOM research from the receiver’s perspective has focused on
source characteristics including expertise and tie strength (e.g., Bansal & Voyer, 2000; Gilly
et al., 1998; Voyer & Ranaweera, 2015; Wangenheim & Bayón, 2007); message
characteristics including valence and richness of the message (Laczniak et al., 2001; Sweeney
et al., 2012; Sweeney et al., 2014); and situational characteristics including involvement

8
(Voyer & Ranaweera, 2015; Wangenheim & Bayón, 2007). These have been shown to
influence the effectiveness of WOM. The receivers’ felt need for information and advice is
one of the most frequent triggers of WOM in service decision-making (Mangold et al. 1999),
which implies that receivers tend to actively seek WOM (Berger, 2014). Hence, WOM can be
perceived as discussion-like because it can include casual conversations as well as in-depth
information sharing by the receiver of WOM. This implies that the receiver of WOM can be
both active and passive. Consumers who actively engage in seeking and obtaining WOM
information are more affected by WOM (Bansal & Voyer, 2000), especially during high-risk
decisions (Fang, Lin, Liu & Lin, 2011). Receivers rely on WOM information because it
reduces decision-making risk, simplifies complexity, and increases their confidence (Berger,
2014).
In this study, WOM is conceptualized as consisting of two dimensions: WOM praise and
WOM activity. WOM praise refers to the level of favorableness or valence, whereas activity
refers to the amount and detail of WOM information (Harrison-Walker, 2001). This
conceptualization was developed specifically for the services context (Harrison-Walker,
2001), and WOM praise and WOM activity represent different levels of information
abstractness, which enables answering the research question. WOM activity is a detailed and
concrete form of information. WOM praise, in contrast, consists of more abstract, coherent,
and unambiguous information because it includes simple cues of general favorability rather
than specific details. Furthermore, in positive WOM, abstract language leads to the receiver’s
perception that the sender has a favorable attitude toward a consumption object (Schellekens
et al., 2010), which implies a relationship between WOM praise and abstractness. In terms of
the WOM source, this study focuses on strong-tie sources, that is, WOM senders who know
the receiver personally (Duhan et al., 1997). Specifically, friends and family were selected as

9
the WOM sources in this study. Strong-tie sources are more readily available and perceived as
more influential than weak ties, such as distant acquaintances (Brown & Reingen, 1987).
Accordingly, HE applicants are persuaded by their comments and advice (Chapman, 1981;
Mazzarol & Soutar, 2002), and they are the most influential sources of information in
university decision-making (Johnston 2010).
2.3 The influence of WOM on attitudinal ambivalence
This study focuses on the influence of different types of WOM information (praise and
activity) on attitudinal ambivalence in two different stages (search and choice) of the HE
decision-making process. The search stage involves seeking information about the attributes
of each HE institution, which ends with the application decision (Chapman, 1986). The choice
stage involves choosing whether to accept a place in one of the (normally only one or a few)
institutions that offered a place to the applicant (Chapman, 1986). We apply construal level
theory (CLT) to theorize the types of WOM information that are the most relevant in each
stage of the decision-making process. CLT assumes that as an individual mentally moves
further away from the “here and now” (i.e., psychological distance), he or she will build
increasingly higher-level mental construals about the distant object (Trope & Liberman,
2010). High-level construals are abstract, coherent, simple, unambiguous, and superordinate
mental representations of issues, involving only their central features (Trope & Liberman,
2010). Hence, when the actual consumption of an object is in the distant future, it will be
evaluated based on its intrinsic desirability (Kim et al., 2009). Furthermore, when temporal
distance increases, the pros of issues become more salient than the cons because the former
are higher-level construals (Herzog et al., 2007).2 Hence, by representing the general
2 Cons are subordinate to pros because the importance of pros does not depend on the existence of cons, whereas
cons are only important when pros exist (Trope et al., 2007). For example, if beginning studies in a HE
institution does not have any pros, an applicant would not be very interested in its cons either; instead, the

10
favorableness of WOM, WOM praise is expected to have a central role in reducing attitudinal
ambivalence toward a HE institution in the search stage. Because this study examines the
attitudinal ambivalence of applicants toward an institution, it is likely that their attitudes will
tend to lean towards positivity even though they involve ambivalence. Hence, WOM praise
should decrease (rather than increase) attitudinal ambivalence by further strengthening
positivity in the attitudes. Consequently, it is expected that general evaluations of
favorableness informed by WOM praise reduce attitudinal ambivalence at the search stage3.
However, WOM activity is expected to play a less important role in the search stage due to its
detailed and concrete nature.
H1: WOM praise received in the search stage reduces attitudinal ambivalence in the search
stage to a greater extent than WOM activity does.
The subsequent choice stage involves the resolution of within-alternative conflicts between
the remaining option(s) before the formulation of a summary evaluation, which serves as the
basis for “go/no-go” decisions about the HE institution (Luce et al., 2003). Because the final
choice as well as the actual beginning of studies are now temporally closer to the consumer,
according to CLT, consumers build lower-level construals that serve to preserve detailed
information about an object for immediate use (Trope & Liberman, 2010). Lower-level
construals are concrete and detailed (Trope & Liberman, 2010), and hence better informed by
WOM activity than WOM praise. Thus, the influence of WOM activity on reducing
attitudinal ambivalence is likely to be more pronounced than that of WOM praise at the
choice stage. Therefore the following hypothesis is stated:
applicant simply would not consider the institution. However, an applicant would be interested in an institutions’
pros whether there were cons or not (Trope et al., 2007; example adapted to the HE context).
3 Applicants may encounter negative WOM in the search stage, which could increase ambivalence because it
counters the pre-existing attitude, which we assume leans towards positivity. However, because this study
focuses on ambivalence reduction, the influence of negative WOM is not considered.

11
H2: WOM activity received in the choice stage reduces attitudinal ambivalence in the choice
stage to a greater extent than WOM praise does.
2.4 The role of attitudinal ambivalence in the decision-making process
Attitudinal ambivalence is likely to have different consequences for decision-making in
different stages. In the search stage, the applicant does not have to commit to the choice and
can apply to several institutions that have pros and cons, therefore “remaining on the fence”
concerning the final choice (van Harreveld et al., 2009). In the HE domain, consumers apply
to HE institutions that are “at least minimally acceptable on all major dimensions” (Chapman,
1986, p. 248), but for the consumer, no option may be perfect (Priester et al., 2007). Because
the applicant does not have to make the final choice at the search stage and because there is
no “perfect option,” it is expected that a certain amount of attitudinal ambivalence is not
resolved in the search stage, which increases attitudinal ambivalence in the choice stage.
Thus, the third hypothesis states the following:
H3: Attitudinal ambivalence in the search stage has a positive influence on attitudinal
ambivalence in the choice stage.
Finally, the existing literature suggests that ambivalent attitudes are less predictive of actual
behavior than univalent attitudes are (Glasman & Albarracín, 2006). Hence, if an applicant’s
attitude toward a HE institution leans toward positivity, as we assume in this case, increasing
attitudinal ambivalence turns the net attitude into a more negative direction, and therefore
leads to decreased intention to begin studies in the HE institution. In a similar vein, attitudinal

12
ambivalence has been found to have a negative influence on behavioral intentions (Costarelli
& Colloca, 2004), as well as actual consumption (Berndsen & van der Pligt, 2004). Therefore,
it is expected that:
H4: Attitudinal ambivalence in the choice stage has a negative influence on the final choice
(i.e., decreases the likelihood of accepting a place in the HE institution).
Thus far, we have stated hypotheses concerning the antecedents and consequences of
attitudinal ambivalence in two stages of the HE decision-making process. The hypotheses are
synthesized in Figure 1. In the following section, we introduce the methodology used in the
analyses, which consist of the three regression models (A-C) shown in Figure 1.
----- Insert Figure 1 about here -----
3 METHOD
3.1 Participants and data collection procedure
The sample consists of applicants to international master’s degree programs at four Finnish
HE institutions. The applicants applied to at least one of these four Finnish HE institutions,
and they could also apply to other HE institutions in Finland or in other countries. An email
with a link to an online questionnaire using Qualtrics software was sent to the applicants. The
first data collection (T1) was conducted in the spring of 2012 shortly after the search stage but
before the applicants had received any acceptance information from the HE institutions. The
second data collection phase (T2) was undertaken shortly after the applicants had received

13
acceptance letters from one or more universities, had made their final choice, but had not yet
begun their studies. The sample in the search stage (T1) was (N = 1,718, response rate =
43.8%). Of these, 1,110 respondents participated in a panel through which the second point
(T2) of the data collection was conducted in August 2012. The second data collection period
yielded 481 responses (response rate = 43.3%). Of the respondents, 213 were selected for
subsequent analyses because they had been accepted by a Finnish institution to which they
had applied and therefore they actually made a choice of whether to begin studies in that
institution or not. Of these, 39% had been accepted only by the Finnish institution to which
they had applied, 30.5% had been also accepted by other institutions (in Finland or other
countries) but would start their master level studies at one of the four studied Finnish
institutions, and 30.5% had been accepted by one of the four Finnish institutions but chose to
begin their studies at another institution. The mean age of the respondents was 24 years, and
53% were male. The respondents were from a range of countries, but the majority were from
Asian, Middle-Eastern, and European countries. A subsample of these respondents was
extracted for each analysis.
3.2 Measures
The attitudinal ambivalence measure was composed of choice criteria that were originally
measured by 23 attributes known to influence HE applicants’ choices (e.g., Joseph & Joseph,
1998; Mazzarol & Soutar, 2002). The respondents were asked, “To what extent do you agree
or disagree that the following factors are associated with university X,” and they indicated
their responses on a seven-point Likert scale of 1 = strongly disagree to 7 = strongly agree.
Because attitudinal ambivalence involves both positive and negative evaluations of the

14
important aspects of an object (Jewell et al., 2002)4, the 10 most important criteria were
chosen to calculate the degree of attitudinal ambivalence toward the institution. The
importance of each criterion was determined by asking each respondent to rate the importance
of the choice criteria on a seven-point Likert scale from 1 = not at all important to 7 = very
important. Attitudinal ambivalence scores were calculated using the Griffin formula, which is
one of the most commonly used attitudinal ambivalence formulas:
(P+N)/ 2−
|
P−N
|
, where
P is the positivity score, and N is the negativity score (Thompson et al., 1995; see Appendix A
for details). WOM praise and WOM activity measures were adapted from Harrison-Walker
(2001). The WOM praise measure included two items, and the WOM activity measure
included three items measured on a five-point Likert scale from 1 = strongly disagree to 5 =
strongly agree. Both strong-tie sources (family and friends) were measured separately by
asking the respondents to respond to the WOM questions first in terms of friends and next in
terms of family members. In case they did not receive WOM from one of these sources (e.g.,
the family source), they responded only once to the WOM questions in terms of the source
from which they received WOM (e.g., friends). Choice was operationalized as whether an
applicant chose to begin studies at a Finnish HE institution to which he/she had applied and
been accepted by and was therefore a binary variable (yes/no). Ambivalence was measured
toward the same university.
4 Typical choice criteria identified in previous research include tuition fees and other costs, reasonable entry
requirements, academic reputation of the university and country, good career prospects, campus atmosphere,
friends’ choice of university, and family opinion. Among the most important criteria in the evaluation of an
institution, also known as “pull” factors, are reputation, career prospects, and academic value (e.g., Mazzarol &
Soutar, 2002, Soutar & Turner, 2002). However, cost-related criteria relating to tuition fees, which is a
significant evaluation criterion (e.g., Soutar & Turner, 2002), were excluded from the study because only two of
the participating universities charged tuition fees. All other major choice criteria based on prior research in the
HE literature are included in our attitude ambivalence construct. Thus, we are confident that our choice to focus
on the 10 most important criteria based on self-reports provides a valid description of the complexity of HE
decision making.

15
In addition, age, gender, number of sources from which WOM was received, perceived social
pressure to begin studies at the HE institution, and nationality were included as control
variables. The number of WOM sources refers to whether the respondent had received WOM
praise from both family and friends (number of sources = 2) or only one of these sources
(number of sources = 1). In T1, six respondents did not respond to both WOM praise and
WOM activity items for all of the sources from which they reported receiving WOM. For
example, they might have reported receiving WOM from friends, but answered only to the
questions concerning WOM praise from friends, and left the questions concerning WOM
activity from friends unanswered. Because the vast majority of respondents responded to both
WOM praise and WOM activity items, having separate control variables for the number of
sources from which WOM praise and WOM activity were received would have resulted in
multicollinearity problems in the regression analyses, which were used to test the hypotheses.
Additionally, it cannot be known with certainty whether the respondents left a part of the
questions unanswered because they did not receive WOM praise/activity or because of some
other reason. Hence, only two control variables were eventually created: number of WOM
sources in T1 and number of WOM sources in T2. In this process, it was necessary to exclude
those respondents who had not answered all questions (T1: 6 respondents; T2: 3 respondents).
Additionally, the nationality of the respondents was controlled for by using dummy variables.
The groups used in the dummy coding were European countries (reference group), African
countries, Asian countries, and American countries (including North and South America).
Finally, an item that captured the perceived social pressure to begin studies at a particular HE
institution was included as a control variable. This item was “I feel social pressure to start my
masters’ degree studies at university X”, and it was measured on a seven-point scale from 1 =
strongly disagree to 7 = strongly agree. All scales were measured at both T1 and T2 time
points, except the final choice and perceived social pressure, which were measured at T2. The

16
measures are presented in detail in Appendix B, including the items, scale types, and scale
points.
4 RESULTS
4.1 Validity and reliability of the measures
The items used to measure WOM praise and WOM activity were verified using confirmatory
factor analysis (CFA) (LISREL 8.80). Before conducting the CFA, the mean of each
measurement item for the friends and family source of WOM was calculated. The
measurement model showed a good fit (Table 1). The standardized loadings of the items
ranged between .58 and .85. All the composite reliabilities were acceptable, exceeding the
recommended level of .60 (Bagozzi & Yi, 1988). The average variances extracted (AVE)
were also satisfactory, exceeding .50 (Hair et al., 1998). The CFA showed a good fit: 2
adjusted by the degrees of freedom was .942, which is well below the recommended level of
3.0 (Iacobucci, 2009). Furthermore, discriminant validity was assessed by the Fornell-Larcker
criterion (Fornell & Larcker, 1981), and the results confirmed that the discriminant validity
was acceptable.
----- Insert Table 1 about here -----
There were moderate to high correlations among the independent variables, which raised
concerns about multicollinearity (Table 2). However, the tolerance values ranged
between .561 and .987, and the variance inflation factor (VIF) values ranged between 1.013
and 1.783 in all analyses, which indicated that multicollinearity was not an issue (Hair et al.,
1998). Finally, prior to the regression analyses, the WOM variables were mean-centered to

17
account for individual differences in the use of scales, which was a potential concern because
of the multicultural sample. All analyses were conducted using SPSS 23, except for the
bootstrapping procedure used to test Model A, which was conducted using SPSS 24 including
the bootstrap function.
----- Insert Table 2 about here -----
4.2 Model A: Explaining attitudinal ambivalence at the search stage
In order to test Hypothesis 1, a hierarchical multiple regression was performed (Table 3). In
Model 1, the control variables (gender, age, dummies for nationality, and number of WOM
sources5) were introduced. This model was not statistically significant (F[6,152] = 1.077, p
= .379), and the control variables did not have a significant influence on T1 attitudinal
ambivalence, except the dummy variable indicating that the respondent was from an African
country. In Model 2, WOM praise and WOM activity at T1 were introduced. This model was
statistically significant (F[8,150] = 5.431, p<.001). Both WOM praise (β = -.290, p<.001) and
WOM activity (β = -.231, p<.01) had a negative influence on T1 attitudinal ambivalence. The
model predicted 22.5% of the variance in T1 attitudinal ambivalence, implying that adding
the WOM variables considerably increased the model’s ability to predict T1 attitudinal
ambivalence.
----- Insert Table 3 about here -----
5 Alternative analyses were also run, in which WOM sources were controlled for using dummy variables. The
dummy variables were: has received WOM only from friends (T1 and T2) and has received WOM only from
family (T1 and T2). The group which had received WOM from both family and friends was used as a reference
group in both time points. The results remained similar to those reported in the manuscript for Model A and
Model B.

18
While the results seem to suggest that WOM praise reduces ambivalence to a greater extent
than WOM activity, it was necessary to test whether the difference in their standardized beta
weights was statistically significantly different, i.e., whether WOM praise reduces T1
ambivalence to a greater extent than WOM activity. Therefore, their corresponding 95%
confidence intervals were estimated using a bias corrected bootstrap (1,000 re-samples). If the
confidence intervals overlap by less than 50%, the beta weights can be considered statistically
different from each other at the level of p<.05 (Cumming, 2009). The amount of overlap was
determined by a calculation (Table 4), in which half of the average of the overlapping
confidence intervals was calculated (.076) and added to the WOM activity beta weight lower
bound estimate (-.349), which yielded -.276. Because the WOM praise upper bound estimate
of -.114 exceeds the value of -.276, the confidence intervals overlap by more than 50%. Thus,
the difference between the WOM activity and WOM praise standardized beta weights (∆.59)
was not considered statistically significant. Therefore, H1 is not supported because although
WOM praise appears to reduce T1 ambivalence to a greater extent than WOM activity, the
difference is not statistically significant.
----- Insert Table 4 about here -----
4.3 Model B: Explaining attitudinal ambivalence at the choice stage
In order to test hypotheses H2 and H3, another hierarchical multiple regression was
performed (Table 5). In Model 1, the control variables (gender, age, dummies for nationality,
number of WOM sources, and perceived social pressure) were introduced. This model was
statistically significant (F[7,154] = 3.128, p<.01) and explained 12.4% of the variance in T2
attitudinal ambivalence. Having an Asian nationality (β = -176, p<.1) or American nationality

19
(β = -155, p<.1) were marginally significant, and having an African nationality (β = -.233,
p<.05) and the number of WOM sources (β =-.178, p<.05) were significant predictors of T2
attitudinal ambivalence. In Model 2, attitudinal ambivalence from T1 was introduced. This
model was statistically significant (F[8,153] = 13.005, p<.001). Attitudinal ambivalence at T1
had a positive influence on attitudinal ambivalence at T2 (β = .548; p<.001). In addition, the
number of WOM sources had a marginally significant influence on T2 attitudinal
ambivalence (β = -.121; p<.1). Introducing T1 attitudinal ambivalence into the model
accounted for the largest change in R2, indicating that it was an important predictor of T2
attitudinal ambivalence. In Model 3, WOM praise and WOM activity from T2 were
introduced. This model was statistically significant (F[10,151] = 16.035, p<.001) and
explained 51.5% of the variance in T2 attitudinal ambivalence. T1 attitudinal ambivalence
had a positive influence (β = .467; p<.001), and WOM activity had a negative influence (β =
-.392, p<.001) on T2 attitudinal ambivalence. No other relationships were significant.
Therefore, both Hypothesis 2 and Hypothesis 3 were supported.
----- Insert Table 5 about here -----
4.4 Model C: Explaining choice
To test Hypothesis 4, a hierarchical logistic regression was conducted in which choice was the
dependent binary variable (Table 6). In Model 1, the control variables (gender, age, dummies
for nationality, and perceived social pressure) were introduced. A test of the full model
against a constant-only model showed marginal significance, indicating that the predictors
marginally distinguished between those who chose to begin studies in the HE institution and
those who did not (χ2 = 10.848, p<.1 with d.f. = 6). Nagelkerke’s R2 was .085, indicating a

20
relatively weak relationship between prediction and grouping. Overall prediction success was
70.7% (negative choice = 5.7%, positive choice = 99.2%). The Wald criterion demonstrated
that only age made a significant contribution to the prediction (p<.01). In Model 2, T2
attitudinal ambivalence was introduced. A test of the full model showed significance,
indicating that the predictors distinguished between positive and negative choice (χ2 = 17.341,
p<.05 with d.f. = 7). Nagelkerke’s R2 was .134, indicating a relatively low yet improved
relationship between prediction and grouping compared to Model 1. Overall prediction
success was 71.3% (negative choice = 18.9%, positive choice = 94.2%). These results showed
that the prediction success of negative choices improved after T2 attitudinal ambivalence was
introduced. The Wald criterion implied that attitudinal ambivalence (p<.05) and age (p<.05)
were significant contributors to prediction. The odds ratio indicated that when attitudinal
ambivalence increased by one unit, the odds that a HE applicant would choose to begin
studies at a HE institution were reduced to a third (Exp[B] = .368). Therefore, Hypothesis 4
was supported.
----- Insert Table 6 about here -----
5 DISCUSSION
The results of this study provided empirical evidence that consumers’ attitudinal ambivalence
can be reduced by WOM and that abstract, positive information (i.e., WOM praise) helps to
reduce attitudinal ambivalence early in the decision-making process, whereas its importance
is reduced as the final choice becomes temporally closer. In addition, the reduction of
attitudinal ambivalence requires detailed and concrete information (i.e., WOM activity)
throughout the HE decision-making process. Finally, attitudinal ambivalence carries from one

21
decision-making stage to another, and any remaining attitudinal ambivalence has a negative
influence on the final choice, which indicates the importance of reducing attitudinal
ambivalence from a marketing perspective.
5.1 Theoretical contributions
This paper provides an improved understanding of the role of ambivalence in the decision-
making process. The existing studies provide insight into the role of attitudinal ambivalence
in decision-making at the face of choice (e.g., Pang et al., 2017; Sparks et al., 2001; van
Harreveld et al., 2009), and many studies point to the conclusion that ambivalence has
negative consequences to choice-making (e.g., Berndsen & van der Pligt, 2004; Costarelli &
Colloca, 2004; Penz & Hogg, 2011; Sparks et al. 2001). Our results are in line with this
existing body of literature because it was found that ambivalence at the choice stage has a
negative influence on choice-making. In addition, however, this paper challenges the
prevailing dominance of choice-stage ambivalence in the existing studies by demonstrating
that attitudinal ambivalence has a role also in other stages, and its development in the
previous stages might shape choice-stage ambivalence. This paper extends a previous study,
which also takes different decision-making stages into account (Jewell et al., 2002) by
showing that different types of information influence attitudinal ambivalence in different
stages. In addition, the finding that attitudinal ambivalence in the choice stage originates
partly in the search stage supports prior suggestions about the dynamic nature of attitudinal
ambivalence (Jewell et al., 2002; Schneider et al., 2015). This influence was relatively strong,
which further indicated the importance of attitudinal ambivalence reduction by implying that
unresolved attitudinal ambivalence in an earlier stage could increase attitudinal ambivalence
in the choice stage. These findings add to the existing literature of attitudinal ambivalence in

22
decision-making by turning attention towards attitudinal ambivalence during the decision-
making process, and reveal that even though attitudinal ambivalence may be most “burning”
at the face of choice, its role cannot be ignored in the earlier stages of the decision-making
process.
Additionally, this study contributes to prior research with regard to the factors that can reduce
attitudinal ambivalence (e.g., Hodson et al., 2001; Hänze, 2001; Jonas et al., 1997; Maio et
al., 1996; van Harreveld et al., 2009). The existing research indicates that individuals use
high-effort strategies to reduce ambivalence given that they have sufficient motivation and
ability to do so (van Harreveld et al., 2009). Accordingly, ambivalence can be reduced
through additional information (Hodson et al., 2001; Jonas et al., 1997), and information can
help people develop positive or negative attitudes towards an object (Maio et al., 1996). This
paper extends these findings by demonstrating that all information is not the same in this
regard, and that the relative importance of different types of information varies across
different stages of the decision-making process. Consequently, this paper extends the existing
research, which has found that consumers use different types of information in different
stages of the decision-making process (Bettman & Park, 1980) by showing that this is indeed
the case also in terms of attitudinal ambivalence reduction. CLT was brought into attitudinal
ambivalence literature as a useful framework for theorizing about attitudinal ambivalence
reduction when the temporal closeness to the final choice changed. The finding that WOM
praise had a role in the search stage is in line with the predictions based on CLT. Contrary to
the predictions of CLT, however, the concrete and detailed WOM activity reduced attitudinal
ambivalence in both stages of the HE decision-making process. This result might have been
due to the nature of the HE decision-making process. By the end of the search stage, the
applicants made an application decision. Although it was not the final choice, they were likely

23
to be motivated to apply to institutions that were satisfactory and therefore had a high
accuracy motivation. Hence, these results indicated that WOM activity might reduce
attitudinal ambivalence in the search stage of the HE decision-making process. Therefore, the
findings highlight the importance of studying attitudinal ambivalence reduction in the context
of various decision-making situations.
Finally, while this paper mainly aims to contribute to attitudinal ambivalence research, the
findings also contribute to the WOM literature as they bring insight into what happens after
WOM is received, and how it affects marketing-relevant outcomes (Martin, 2014; Martin &
Lueg, 2013; Yang et al., 2012). The existing WOM research suggests that consumers’
utilization of WOM influences their attitudes, which further influences their purchase
intentions (Martin & Lueg, 2013). We extended these findings by demonstrating similar
effects when the attitude was ambivalent, using actual choices instead of purchase intentions.
Studying actual, “real-life” choice behavior instead of intention-based measures is rare also in
ambivalence studies (Roster & Richins, 2009), and therefore this paper makes a
methodological contribution to both WOM and ambivalence research.
5.2 Managerial implications
The findings demonstrate the importance of considering attitudinal ambivalence in the
marketing of HE services. Because the findings imply that the consequences of attitudinal
ambivalence are negative, its reduction is an important marketing goal. In this regard, the
findings imply that ambivalence can be reduced through the use of WOM information with
varying degrees of abstractness, and the types of WOM required in the search stage and the
choice stage differed. Therefore, the first step by HE marketers should be to recognize the

24
applicants’ decision-making stage. Because the search stage ends when an applicant makes an
application decision (Chapman, 1986), marketing efforts directed at the search stage should
be conducted prior to a HE institution’s deadline for applications. In the choice stage,
however, efforts should be made after the applicants have been notified of their acceptance
but before they make their final choice of whether to begin studies at the institution by which
they have been accepted. In terms of concrete marketing tools, WOM can be influenced
through indirect and direct marketing efforts (Lang & Hyde, 2013). Indirect efforts include
general advertising, such as testimonials, teaser campaigns, and celebrity endorsements,
whereas direct efforts include incentivized WOM and the direct targeting of influencers (Lang
& Hyde, 2013).
In promoting WOM activity, marketers could provide much detailed information to both
friends and family through indirect efforts, such as creating shareable content on the
institution’s website. They can also use direct efforts, such as email campaigns in which the
receivers are encouraged to forward the message to their friends or family members, and
providing informative event marketing for HE applicants and their families (e.g., campus
visits). The families and friends of the applicants were chosen as the focus of this study
because they are both strong-tie sources (Duhan et al., 1997). In the context of HE decision-
making, they are also central sources of WOM (Chapman, 1981; Mazzarol & Soutar, 2002).
Importantly, however, the media usage behavior of strong-tie sources may differ. For
example, because friends are generally younger than family members are, they might be more
responsive to social media campaigns, whereas traditional marketing communications, such
as mass-media advertising, might be more effective in targeting family members (Deloitte,
2015).

25
Furthermore, because the findings indicated that WOM praise received in the search stage
reduces attitudinal ambivalence, it is important to create a positive impression among strong-
tie sources of WOM, perhaps by using indirect efforts such as ad campaigns that have
emotional content or generate conversation. Furthermore, managing the post-purchase WOM
of current customers effectively promotes WOM (Lang & Hyde, 2013), and it is a relatively
manageable way to approach WOM (Buttle, 1998). The alumni and current students of
universities are good examples of post-purchase WOM influencers who could be asked to
share their positive experiences with their friends and family members in order to promote
WOM praise. Finally, the receiver has been shown to trigger WOM discussions by starting
conversations, which happens when receivers gain information by seeking advice or solving
problems (Berger, 2014). Therefore, marketers could target HE applicants through both direct
and indirect marketing to promote both WOM praise and WOM activity.
5.3 Limitations and suggestions for further research
The results of the present study provide the first step toward understanding attitudinal
ambivalence in different stages of the decision-making process. An additional explanation
supporting our results regarding the central role of WOM activity in decision-making is based
on multi-process theories of influence, such as the elaboration likelihood model and the
heuristic systematic model, which assume that information processing operates through two
separate routes (Petty, 1994). The first route relates to high levels of elaboration, and the
receiver focuses on the central merits of issue-relevant information in information processing.
The second route requires less elaboration, and simple cues of information or heuristics can
affect attitudes without extensive thought (Petty et al., 2013). When consumers analyze
information deeply, they form strong attitudes that are durable and resistant to change (Petty

26
et al., 2013). Thus, our results indicate that WOM activity may enable the cognitive
processing of information to a greater extent than WOM praise, and therefore has the
strongest influence on decision-making in the choice stage. Hence, future research is
encouraged to use an information processing perspective to understand changes in attitudinal
ambivalence during the decision-making process. Future research could also address
additional dimensions of WOM, such as negative WOM (Sweeney et al., 2014), cognitive
content, and content richness (Sweeney et al., 2012). Additionally, future studies should
compare the findings between strong-tie and weak-tie sources, as well as investigate other
sources of information and their interactive influence on attitudinal ambivalence, including
TV and the Internet.
Because skeptical attitudes toward advertising (Chang, 2011) and the simultaneous trust and
distrust of an online seller (Moody et al., 2014) can increase attitudinal ambivalence, different
contexts would enable more comprehensive modeling of the antecedents of attitudinal
ambivalence. In terms of personality factors, need for cognition (NFC) and personal fear of
invalidity (PFI) have an influence on attitudinal ambivalence (Thompson & Zanna, 1995), and
tolerance for ambiguity and NFC have an influence on coping with ambivalence (Nowlis et
al., 2002). By taking into account these personality factors, future research would capture a
larger share of attitudinal ambivalence. Moreover, HE decision-making is different from
many other decision-making processes, and although the current findings may be
generalizable to high-involvement, investment-type decision-making, WOM might have
different influences on attitudinal ambivalence in different types of decision-making contexts.
Importantly, the HE decision-making process involves two decisions (i.e., the decision to
apply and the choice of beginning studies in an institution by which the applicant had been
accepted). Because applying to the Finnish HE institutions (towards which ambivalence was

27
studied) is free of charge, there were many applicants, but only a relatively small part of the
applicants became accepted by the institutions. In this study, we were interested in the
influence of WOM on attitudinal ambivalence and subsequent choice towards only one
institution per applicant. Only the part of the applicants who were accepted to that particular
HE institution made a choice about whether to begin studies at the institution, and therefore
only they were included into the sample of this study. However, in future studies it would be
important to study whether those applicants who were rejected received WOM about other
institutions and how it influenced their subsequent ambivalence and decision-making towards
alternative HE institutions. We encourage future research to measure WOM and ambivalence
towards multiple HE institutions to investigate these broader decision-making processes
between multiple options. Finally, although the present study focused on the cognitive
component of attitudes, an interesting future study could investigate ambivalence in the
affective attitude component or take an intercomponent (i.e., affective-cognitive) approach to
attitudinal ambivalence (van Harreveld et al., 2009).

28
REFERENCES
Bagozzi, R. P., & Yi, Y. (1988). On the evaluation of structural equation models. Journal of
the Academy of Marketing Science, 16(1), 74–94.
Bansal, H. S., & Voyer, P. A. (2000). Word-of-mouth processes within a service purchase
decision context. Journal of Service Research, 3(2), 166–177.
Berger, J. (2014). Word of mouth and impersonal communication: A review and directions
for future research. Journal of Consumer Psychology, 24(4), 586–607.
Berndsen, M., & van der Pligt, J. (2004). Ambivalence towards meat. Appetite, 42(1), 71–78.
Bettman, J. R., & Park, C. W. (1980). Effects of prior knowledge and experience and phase of
the choice process on consumer decision processes: A protocol analysis. Journal of
Consumer Research, 7(3), 234–248.
Brown, J. J., & Reingen, P. H. (1987). Social ties and word-of-mouth referral behaviour.
Journal of Consumer Behaviour, 14(3), 350–362.
Buttle, F. A. (1998). Word of mouth: Understanding and managing referral marketing.
Journal of Strategic Marketing, 6(3), 241–254.
Cacioppo, J. T., & Berntson, G. G. (1994). Relationship between attitudes and evaluative
space: A critical review, with emphasis on the separability of positive and negative
substrates. Psychological Bulletin, 115(3), 401–423.
Chang, C. (2011). Feeling ambivalent about going green: Implications for green advertising
processing. Journal of Advertising, 40(4), 19–31.
Chapman, D. W. (1981). A model of student college choice. Journal of Higher Education,
52(5), 490–505.

29
Chapman, R.G. (1986). Toward a theory of college selection: A model of college search and
choice behaviour. In R.J. Lutz (Ed.), Advances in Consumer Research (Vol. 13, pp. 246–
250). Provo: Association for Consumer Research.
Costarelli, S., & Colloca, P. (2004). The effects of attitudinal ambivalence on pro-
environmental behavioural intentions. Journal of Environmental Psychology, 24(3), 279–
288.
Cubillo-Pinilla, J. M., Zuniga, J., Losantos, I. S., & Sanchez, J. (2009). Factors influencing
international student’s evaluations of higher education programs. Journal of American
Academy of Business, 15(1), 270–278.
Cumming, G. (2009). Inference by eye: Reading the overlap of independent confidence
intervals. Statistics in Medicine, 28(2), 205-220.
De Matos, C. A., & Rossi, C. A. V. (2008). Word-of-mouth communications in marketing: A
meta-analytic review of the antecedents and moderators. Journal of the Academy of
Marketing Science, 36(4), 578–596.
Deloitte. (2015). Digital democracy survey, executive summary. Retrieved from
https://www2.deloitte.com/us/en/pages/technology-media-and-telecommunications/
articles/digital-democracy-survey-generational-media-consumption-trends.html/ Accessed
19.2.17.
Duhan, D. F., Johnson, S. D., Wilcox, J. B., & Harrell, G. D. (1997). Influences on consumer
use of word-of-mouth recommendation sources. Journal of the Academy of Marketing
Science, 25(4), 283–295.
Durvasula, S., Lysonski, S., & Madhavi, A. D. (2011). Beyond service attributes: Do personal
values matter? Journal of Services Marketing, 25(1), 33–46.
Fang, C. H., Lin, T. M., Liu, F., & Lin, Y. H. (2011). Product type and word of mouth: a
dyadic perspective. Journal of Research in Interactive Marketing, 5(2/3), 189–202.

30
Fazio, R. H. (1995). Attitudes as object-evaluation associations: Determinants, consequences,
and correlates of attitude accessibility. In R. E. Petty & J. A. Krosnick (Eds.), Attitude
strength: Antecedents and consequences (pp. 247-282). Mahwah: Erlbaum.
Fazio, R. H. (2007). Attitudes as object–evaluation associations of varying strength. Social
Cognition, 25(5), 603–637.
Festinger, L. (1964). Conflict, decision, and dissonance. Stanford, CA: Stanford University
Press.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable
variables and measurement error. Journal of Marketing Research, 18(1), 39–50.
Gilly, M. C., Graham, J. L., Wolfinbarger, M. F., & Yale, L. J. (1998). A dyadic study of
interpersonal information search. Journal of the Academy of Marketing Science, 26(2), 83–
100.
Glasman, L. R., & Albarracín, D. (2006). Forming attitudes that predict future behaviour: A
meta-analysis of the attitude-behaviour relation. Psychological Bulletin, 132(5), 778–822.
Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1998). Multivariate data
analysis. (5th ed.). Upper Saddle River, New Jersey: Prentice Hall.
Harrison-Walker, L. J. (2001). The measurement of word-of-mouth communication and an
investigation of service quality and consumer commitment as potential antecedents.
Journal of Service Research, 4(1), 60–75.
Herzog, S. M., Hansen, J., & Wänke, M. (2007). Temporal distance and ease of retrieval.
Journal of Experimental Social Psychology, 43(3), 483–488.
Hodson, G., Maio, G. R., & Esses, V. M. (2001). The role of attitudinal ambivalence in
susceptibility to consensus information. Basic and Applied Social Psychology, 23(3), 197–
205.

31
Hänze, M. (2001). Ambivalence, conflict, and decision making: Attitudes and feelings in
Germany towards NATO’s military intervention in the Kosovo war. European Journal of
Social Psychology, 31(6), 693–706.
Iacobucci, D. (2009). Everything you always wanted to know about SEM (structural
equations modeling) but were afraid to ask. Journal of Consumer Psychology, 19(4), 673–
680.
Jewell, R. D., Coupey, E., & Jones, M. T. (2002). Catch a tiger by his toe: Ambivalence in
decision making in the 2000 presidential election. In S. M. Broniarczyk, & K. Nakamoto
(Eds.), Advances in consumer research (Vol. 29, pp. 333–338). Valdosta: Association for
Consumer Research.
Johnston, T. C. (2010). Who and what influences choice of university? Student and university
perceptions. American Journal of Business Education, 3(10), 15–24.
Jonas, K., Diehl, M., & Brömer, P. (1997). Effects of attitudinal ambivalence on information
processing and attitude-intention consistency. Journal of Experimental Social Psychology,
33(2), 190–210.
Joseph, M., & Joseph, B. (1998). Identifying needs of potential students in tertiary education
for strategy development. Quality Assurance in Education, 6(2), 90–96.
Kim, Y. J., Park, J., & Wyer, R. S. (2009). Effects of temporal distance and memory on
consumer judgments. Journal of Consumer Research, 36(4), 634–645.
Krosnick, J. A., & Petty, R. E. (1995). Attitude strength: An overview. In R. E. Petty & J. A.
Krosnick (Eds.), Attitude strength: Antecedents and consequences (pp. 1-24). Mahwah:
Erlbaum.
Lang, B., & Hyde, K. F. (2013). Word of mouth: What we know and what we have yet to
learn. Journal of Consumer Satisfaction, Dissatisfaction & Complaining Behavior, 26, 3–
18.

32
Laczniak, R. N., DeCarlo, T. E., & Ramaswami, S. N. (2001). Consumers’ responses to
negative word-of-mouth communication: An attribution theory perspective. Journal of
Consumer Psychology, 11(1), 57–73.
Lim, B. C., & Chung, C. M. (2011). The impact of word-of-mouth communication on
attribute evaluation. Journal of Business Research, 64(1), 18–23.
Luce, M. F., Jia, J., & Fischer, G. W. (2003). How much do you like it? Within-alternative
conflict and subjective confidence in consumer judgments. Journal of Consumer Research,
30(3), 464–472.
Maio, G. R., Bell, D. W., & Esses, V. M. (1996). Ambivalence and persuasion: The
processing of messages about immigrant groups. Journal of Experimental Social
Psychology, 32(6), 513-536.
Mangold, W. G., Miller, F., & Brockway, G. R. (1999). Word-of-mouth communication in
the service marketplace. Journal of Services Marketing, 13(1), 73−89.
Martin, W.C. (2014). Independent versus Incentivized Word-of-Mouth: Effects on Listeners.
Academy of Marketing Studies Journal, 18(1), 261-271.
Martin, W. C., & Lueg, J. E. (2013). Modeling word-of-mouth usage. Journal of Business
Research, 66(7), 801–808.
Mazzarol, T., & Soutar, G. N. (2002). “Push-pull” factors influencing international student
destination choice. International Journal of Educational Management, 16(2), 82–90.
Moody, G. D., Galletta, D. F., & Lowry, P. B. (2014). When trust and distrust collide online:
The engenderment and role of consumer ambivalence in online consumer behavior.
Electronic Commerce Research and Applications, 13(4), 266–282.
Murray, K. B. (1991). A test of services marketing theory: Consumer information acquisition
activities. Journal of Marketing, 55(1), 10–25.

33
Murray, K. B. (1992). Health care service decision influences: An exploratory investigation of
search and nonsearch criteria for professionals and patients. Marketing Health Services,
12(1), 24–38.
Newby-Clark, I. R., McGregor, I., & Zanna, M. P. (2002). Thinking and caring about
cognitive inconsistency: When and for whom does attitudinal ambivalence feel
uncomfortable? Journal of Personality and Social Psychology, 82(2), 157–166.
Nowlis, S. M., Kahn, B. E., & Dhar, R. (2002). Coping with ambivalence: The effect of
removing a neutral option on consumer attitude and preference judgments. Journal of
Consumer Research, 29(3), 319–334.
Olsen, S. O., Wilcox, J., & Olsson, U. (2005). Consequences of ambivalence on satisfaction
and loyalty. Psychology and Marketing, 22(3), 247–269.
Pang, J., Keh, H. T., Li, X., & Maheswaran, D. (2017). “Every coin has two sides”: The
effects of dialectical thinking and attitudinal ambivalence on psychological discomfort and
consumer choice. Journal of Consumer Psychology, 27(2), 218-230.
Penz, E. & Hogg, M.K. (2011). The role of mixed emotions in consumer behaviour:
Investigating ambivalence in consumers’ experiences of approach-avoidance conflicts in
online and offline settings. European Journal of Marketing, 45(1/2), 104-132.
Petty R. E. (1994). Two routes to persuasion: State of the art. In P. Eelen, M. Pichelle, G.
d'Ydewalle, & P. Bertelson (Eds.), 25th International Congress of Psychology: State of the
art lectures (pp. 229–247). Erlbaum: Hillsdale, NJ.
Petty, R. E., Wheeler, S. C., & Tormala, Z. T. (2013). Persuasion and attitude change. In I. B.
Weiner & M. J. Lerner (Eds.), Comprehensive handbook of psychology: Vol. 5, personality
and social psychology (2nd ed., pp. 269–390), John Wiley & Sons: New York.

34
Priester, J. R., & Petty, R. E. (1996). The gradual threshold model of ambivalence: Relating
the positive and negative bases of attitudes to subjective ambivalence. Journal of
Personality and Social Psychology, 71(3), 431–449.
Priester, J. R., Petty, R. E., & Park, K. (2007). Whence univalent ambivalence? From the
anticipation of conflicting reactions. Journal of Consumer Research, 34(June), 11–21.
Puccinelli, N. M., Goodstein, R. C., Grewal, D., Price, R., Raghubir, P., & Stewart, D. (2009).
Customer experience management in retailing: Understanding the buying process. Journal
of Retailing, 85(1), 15–30.
Roster, C.A. & Richins, M.L. (2009). Ambivalence and attitudes in consumer replacement
decisions. Journal of Consumer Psychology 19(1), 48-61.
Schellekens, G. A., Verlegh, P. W., & Smidts, A. (2010). Language abstraction in word of
mouth. Journal of Consumer Research, 37(2), 207–223.
Schneider, I. K., van Harreveld, F., Rotteveel, M., Topolinski, S., van der Pligt, J., Schwarz,
N., & Koole, S. L. (2015). The path of ambivalence: Tracing the pull of opposing
evaluations using mouse trajectories. Frontiers in Psychology, 6, 1–12.
Selnes, F. (1993). An examination of the effect of product performance on brand reputation,
satisfaction and loyalty. European Journal of Marketing, 27(9), 19–35.
Soutar, G. N., & Turner, J. P. (2002). Students’ preferences for university: A conjoint
analysis. International Journal of Education Management, 16(1), 40–45.
Sparks, P., Conner, M., James, R., Shepherd, R., & Povey, R. (2001). Ambivalence about
health‐related behaviours: An exploration in the domain of food choice. British Journal of
Health Psychology, 6(1), 53-68.
Sweeney, J. C., Soutar, G. N. and Mazzarol, T. (2008). Factors influencing word of mouth
effectiveness: Receiver perspectives. European Journal of Marketing, 42(3/4), 344–364.

35
Sweeney, J. C., Soutar, G. N., & Mazzarol, T. (2012). Word of mouth: Measuring the power
of individual messages. European Journal of Marketing, 46(1/2), 237–257.
Sweeney, J. C., Soutar, G. N., & Mazzarol, T. (2014). Factors enhancing word-of-mouth
influence: Positive and negative service-related messages. European Journal of Marketing,
48(1/2), 336–359.
Thompson, M. M., & Zanna, M. P. (1995). The conflicted individual: Personality‐based and
domain specific antecedents of ambivalent social attitudes. Journal of Personality, 63(2),
259–288.
Thompson, M. M., Zanna, M. P., & Griffin, D. W. (1995). Let's not be indifferent about
(attitudinal) ambivalence. In R.E. Petty & J.A. Krosnick (Eds.), Attitude strength:
Antecedents and consequences (pp. 361–386). Mahwah: Erlbaum.
Tormala, Z. L., & DeSensi, V. L. (2008). The perceived informational basis of attitudes:
implications for subjective ambivalence. Personality and Social Psychology Bulletin,
34(2), 275–287.
Trope, Y., Liberman, N., & Wakslak, C. (2007). Construal levels and psychological distance:
Effects on representation, prediction, evaluation, and behavior. Journal of Consumer
Psychology, 17(2), 83–95.
Trope, Y., & Liberman, N. (2010). Construal-level theory of psychological distance.
Psychological Review, 117(2), 440–463.
Van Harreveld, F., Schneider, I. K., Nohlen, H., & van der Pligt, J. (2012). The dynamics of
ambivalence: Evaluative conflict in attitudes and decision making. In B. Gawronski & F.
Strack (Eds.), Cognitive consistency: A fundamental principle in social cognition (pp. 267–
284). New York, Guilford Press.

36
Van Harreveld, F., van der Pligt, J., & de Liver, Y. N. (2009). The agony of ambivalence and
ways to resolve it: Introducing the MAID model. Personality and Social Psychology
Review, 13(1), 45–61.
Voyer, P. A., & Ranaweera, C. (2015). The impact of word of mouth on service purchase
decisions: examining risk and the interaction of tie strength and involvement. Journal of
Service Theory and Practice, 25(5), 636–656.
Wangenheim, F. V., & Bayón, T. (2007). The chain from customer satisfaction via word-of-
mouth referrals to new customer acquisition. Journal of the Academy of Marketing
Science, 35(2), 233–249.
Yang, S., Hu, M., Winer, R.S. Assael, H. and Chen, X. (2012). An empirical study of word-
of-mouth generation and consumption. Marketing Science, 31(6), 952-963.

37
Table 1. Confirmatory Factor Analysis
p-value χ2(d.f.) RMSEA CFI/GFI NNFI
.544 22.600(24) .000 1.000/.971 1.002
1 2 3 4
1. WOM praise T1 1.000
2. WOM activity T1 .664 1.000
3. WOM praise T2 .554 .482 1.000
4. WOM activity T2 .249 .508 .744 1.000
Mean 3.619 3.510 3.449 3.449
S.D. .794 .761 .805 .873
AVE .602 .509 .571 .665
CR .750 .754 .856 .723

38
Variable 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.
1. Gender 1.000
2. Age -.227*** 1.000
3. African nationality -.192** .441*** 1.000
4. Asian nationality -.047 -.227*** -.353*** 1.000
5. American nationality -.064 .096 -.051 -.353*** 1.000
6. # WOM channels T1 .010 -.051 .082 .038 -.069 1.000
7. # WOM channels T2 .041 .032 .058 -.030 -.006 .056 1.000
8. Social pressure -.042 .041 .057 .000 -.069 .119 .151* 1.000
9. WOM activity T1 .051 .039 .223** .026 -.129 .150 .038 -.045 1.000
10. WOM activity T2 .068 .046 .185* .043 -.088 .225** .215** -.048 .480*** 1.000
11. WOM praise T1 .036 .064 .077 .018 .010 .153* .-006 -.053 .476*** .229** 1.000
12. WOM praise T2 -.007 .114 .171* -.164* .124 .125 .110 .059 .351*** .570*** .477*** 1.000
13. Ambivalence T1 -.049 -.034 -.112 -.045 -.040 -.193** -.004 -.011 -.379*** -.320*** -.419*** -.318*** 1.000
14. Ambivalence T2 -.028 -.120 -.165 -.045 -.074 -.183* -.140 .075 -.355*** -.537*** -.295*** -.371*** .613*** 1.000
15. Choice .041 .173* .051 -.006 .002 .081 .003 -.013 .229** .323** .025 .222** .012 -.219** 1.000
Note: Significance †= <.1; *= <.05; **= <.01; ***= <0.001
Table 2. Correlations

39
Table 3. Model A: Results
Dependent variable:
T1 ambivalence
N = 158 B(S.E.) Beta t-value B(S.E.) Beta t-value
Independent
variable Model 1 Model 2
Gender -.039(.067) -.050 -.590 .001(.061) .001 .013
Age .007(.011) .059 .632 .007(.010) .061 .722
Africa (dummy) -.412(.168) -.244* -2.460 -.211(.157) -.125 -1.344
America (dummy) -.121(.169) -.064 -.715 -.111(.154) -.058 -.720
Asia (dummy) -.065(.085) -.075 -.768 -.008(.077) -.009 -.102
# of WOM channels -.033(.069) -.038 -.470 .028(.064) .003 .446
WOM praise T1 -.141(.040) -.290*** -3.522
WOM activity T1 -.117(.043) -.231** -2.730
R2.041 .225
Sig. F change .379 .000
Note: Significance †= <.1; *= <.05; **= <.01; ***= <0.001

40
Table 4. Calculation used to determine the statistical significance of the difference between
the standardized beta weights of WOM activity and WOM praise.
95% Confidence intervals Lower bound Point
Upper
bound
WOM activity T1 -.349 -.231 -.068
WOM praise T1 -.393 -.290 -.114
Calculation
WOM activity T1: Difference between point and lower bound .118
WOM praise T1: Difference between point and upper bound .176
Average of the two differences between point and lower bound .147
Average divided by two .0735
WOM activity lower bound+average divided by two (above) -.2755
Conclusion -.114>-.276; difference not significant

41
Table 5. Model B: Results
Dependent variable:
T2 ambivalence
N = 161 B(S.E.) Beta t-value B(S.E.) Beta t-value B(S.E.) Beta t-value
Independent variable Model 1 Model 2 Model 3
Gender -.117(.072) -.132 -.1.609 -.055(.060) -.062 -.908 -.037(.055) -.042 -.679
Age -.019(.012) -.135 -.1.505 -.016(.010) -.112 -.1511 -.015(.009) -.111 -1.652
Asia (dummy) -.172(.090) -.176† -.1.914 -.083(.075) -.085 -1.108 -.032(.069) -.033 -.462
America (dummy) -.287(.164) -.155† -.1.811 -.145(.137) -.076 -1.061 -.175(.125) -.091 -1.399
Africa (dummy) -.503(.198) -.233* -2.537 -.242(.167) -.112 -1.452 -.067(.155) -.031 -.436
# of WOM channels -.228(.097) -.178* -2.338 -.156(.081) -.121† -1.922 -.044(.077) -.034 -.569
Social pressure .015(.020) .059 .770 .041(.016) .055 .870 .006(.015) .025 .426
Ambivalence T1 .552(.065) .548*** 8.488 .470(.063) .467*** 7.471
WOM activity T2 -.199(.038) -.392*** -.5.181
WOM praise T2 .022(.041) .040 .535
R2.124 .405 .515
Sig. F change .004 .000 .000
Note: Significance †=<.1; *=<.05; **=<.01; ***=<0.001

42
Table 6. Model C: Results
Dependent variable: Choice
N = 174 B(S.E.) Wald
Exp(B
) B(S.E.) Wald
Exp(B
)
Independent variable Model 1 Model 2
Constant -.4.920(2.045) 5.788* .007 -5.261(2.060) 6.520* .005
Gender .579(.372) 2.430 1.785 .462(.378) 1.493 1.588
Age .224(.081) 7.583** 1.251 .206(.081) 6.448* 1.229
Asia (dummy) .302(.442) .467 1.353 .174(.451) .149 1.190
America (dummy) -.240(837) .082 .787 -.473(.859) .303 .632
Africa (dummy) .215(1.252) .030 1.240 -.346(1.279) .073 .707
Social pressure -.003(.100) .001 .997 .041(.103) .019 1.015
Ambivalence T2 -.999(.400) 6.229* .368
χ2 statistic (Model) 10.848 (d.f. = 6); p = .093 17.341 (d.f. = 7); p = .015
χ2 statistic (Step) 10.848 (d.f. = 6); p =.093 6.493 (d.f. = 1); p = .011
Nagelkerke R2.085 .134
% of correctly classified
observations: neg. choice 5.7 18.9
% of correctly classified
observations: pos. choice 99.2 94.2
% of correctly classified
observations: total 70.7 71.3
Note: Significance †=<.1; *= <.05; **= <.01; ***= <0.001

43
Figure 1. Theoretical framework. The relationships shown with dashed lines are assumed non-significant. Age, gender, nationality, and number
of WOM channels are included in each model (A-C) as control variables. Additionally, perceived social pressure is used as a control variable in
Models B and C.

44
Appendix A. Calculation of positivity, negativity, and attitudinal ambivalence
The chosen criteria scales were first recoded using a scale ranging from -3 to +3. Next, the
scale was split at the middle point. Because all the original choice criteria were labeled
positively, the values from -3 to -1 reflected negativity, and the values from 1 to 3 reflected
the positivity of each choice criterion. In order to calculate the attitudinal ambivalence score
using the choice criteria, the positivity and negativity values of each criterion were
aggregated. Because there were different numbers of positive and negative responses for
many of the items, the positivity score was calculated by dividing the sum of positive
responses to all items by the total number of the items (i.e., 10), and the same was done for
the negative responses. Table 1 below shows two examples of the calculation of positivity and
negativity scores of hypothetical participants.
Table 1. Examples of the Calculation of the Positivity and Negativity Scores
Choice criteria R1, original
response
R1, recoded
response
R2, original
response
R2, recoded
response
(1) The degrees offered have
academic value
1 -3 5 1
2) The degrees offer good job
prospects
3 -1 6 2
(3) The program fulfills my
academic needs
6 2 6 2
(4) A clean and safe study
environment
5 1 2 -2
(5) Country’s high academic
reputation
4 0 4 0
(6) Institution is well-known for
its reputation
7 3 3 -1
(7) Reasonable living costs
(accommodation, food,
traveling, etc.)
6 2 7 3
(8) A high level of security in
the host country
5 1 7 3
(9) University’s high-ranking
position
2 -2 5 1
(10) An active student life 3 -1 6 2
Sum of negative responses -7 -3
Sum of positive responses 9 14
Negativity score -0.7 -0.3
Positivity score 0.9 1.4
Note: R1=Respondent 1; R2=Respondent 2
Next, the positivity and negativity scores were combined into an overall attitudinal
ambivalence score according to the following formula:
(P+N)/ 2−
|
P−N
|
, where P is the
positivity score and N is the negativity score (Thompson et al., 1995). Because both positivity
and negativity components should be inserted into the equation with a positive sign, the
negativity scores were first multiplied by -1. Thus, the values for attitudinal ambivalence
could range from -1.5 (i.e., univalent positive or negative attitude) to 3 (when both positivity
and negativity scores are 3). In the example above, the attitudinal ambivalence score is
(0.9+0.7)/2-│0.9-0.7│= 0.6 for Respondent 1, and (1.4+0.3)/2-│1.4-0.3│= -0.25 for
Respondent 2, indicating that Respondent 1 was more ambivalent.

45
Appendix B. Measures
Items Scale points Cronbach’s
α
Source
WOM activity
(1) The university has been frequently
mentioned in discussions
Five-point Likert scale:
1=‘strongly disagree’;
5=‘strongly agree’
Harrison-
Walker (2001)
(2) I have had more discussions about the
university than about other universities
(3) The discussions about the university tend
to be in great detail
WOM praise
(1) In the discussions, the university was
praised
Five-point Likert scale,
1=‘strongly disagree’;
5=‘strongly agree’
T2: α = .710
Harrison-
Walker (2001)
(2) The discussions have only been on good
things about the university
Choice criteria (Criteria used for attitudinal ambivalence calculation, i.e., the most important criteria)
(1) The degrees offered have academic value Seven-point Likert scale:
1=’strongly disagree’;
7=’strongly agree’
N.A. Joseph &
Joseph (1998);
Mazzarol &
Soutar (2002)
(2) The degrees offer good job prospects
(3) The program fulfills my academic needs
(4) A clean and safe study environment
(5) Country’s high academic reputation
(6) Institution is well-known for its reputation
(7) Reasonable living costs (accommodation,
food, traveling, etc.)
(8) A high level of security in the host country
(9) University’s high-ranking position
(10) An active student life
Other choice criteria originally measured
(11) A low level of racial discrimination in the
host country
N.A. Joseph &
Joseph (1998);
Mazzarol &
Soutar (2002)
(12) University’s cost of education/tuition fees
is/are reasonable
(13) The possibility of studying for a doctoral
degree
(14) Reasonable entry/admission requirements
(15) The university is centrally located
(16) A positive city image
(17) The possibility to work during one’s
studies
(18) The availability of advice and help with
organizing everyday life in the host country
(19) The availability of financial
help/scholarships
(20) Recommendations from alumni or
current students
(21) My friends are applying to the same
university
(22) I know someone who has studied or is
currently studying at the university
(23) I know someone who has studied or is
currently studying in the country
Importance of choice criteria
The same items were used as in the choice
criteria measure
Seven-point Likert scale:
1=’not at all important’;
7=’very important’
N.A. Joseph &
Joseph (1998);
Mazzarol &
Soutar (2002)

46