T
HEI
NFLUENCE OFW
ORD OFM
OUTH IN THED
IGITALA
GE:
HOW DIFFERENT TYPES OF WORD OF MOUTH INFLUENCE THE CONSUMER DECISION JOURNEY
Peter P. Faasse
Universiteit van Amsterdam
Author Note
Peter P. Faasse, Department of Social Psychology, University of Amsterdam.
This research was supported by Havas Media for marketing research purposes.
Correspondence concerning this article should be directed to Peter P. Faasse,
Delistraat 40 3hoog, Amsterdam, 1094CX. E-Mail: peter.faasse@gmail.com or Michael L.
W. Vliek, Department of Social Psychology, University of Amsterdam, Weesperplein 4,
Amsterdam, 1018 WZ. E-Mail: M.L.W.Vliek@uva.nl or Marloes Rijnders, Havas Media,
Burgemeester A.Colijnweg 2, Amstelveen, 1182 AL. E-Mail:
Table of Contents:
Abstract……….3
Introduction………..4
Word of Mouth………..………..4
Offline WOM, eWOM, social WOM usage………...6
Offline WOM influence vs eWOM influence………...6
Offline WOM and eWOM influence vs social influence………...7
Involvement………...………..8
The Consumer Decision Journey………...9
Current Study………...12 Focusgroups Method………..13 Focusgroups participants..………...13 Focusgroup materials………...13 Focusgroup procedure………....14 Focusgroup results……….…14
Quantitative study Method………...14
Participants……….14
Design………...15
Procedure and Materials……..………...…...…15
Involvement………..16
Consumer Decision Journey……….…16
Influence Questionnaire………17
Offline information source list……….17
Online information source list………..17
Social Media information source lists……….….18
Manipulation Checks………..…..18
Exploratory………...………..……...19
Results………..………...21
Factor and Reliability Analyses……..………...21
Normality and Manipulation checks………...23
Descriptives for usage of offline, online, and social media information…………..………...25
Mixed measures Ancova for offline factors………..25
Mixed measures Ancova for all offline factors, eWOM and social WOM ………..28
Discussion….………...………...35
WOM usage………...35
Influence of Offline WOM………..…………..36
Influence of eWOM………...36
Influence of social WOM………..37
Involvement………...38
Consumer Decision Journey………..38
Limitations and future studies………...….39
Conclusions and implications………...……….40
Abstract
This study looked at how influential different types of word of mouth, or WOM, are on the
decision process consumers go through when buying different types of products. Special
attention was paid to new forms of WOM such as electronic WOM and WOM via social
media. 428 Participants were divided over three between conditions for level of product
involvement (low, medium, high), six within condition for influence factor (offline WOM,
low information influences, advertising, personal experience, eWOM, social WOM) and four
within conditions for stage of the CDJ (awareness, initial consideration, active evaluation,
moment of purchase). Results showed that offline WOM was more influential than
advertising, eWOM, social WOM and low information influences during the consumer
decision journey. Offline WOM was more influential for medium and high involvement
products than for low involvement products. It is recommended marketers and public health
organizations do not overestimate the influence of social WOM and eWOM on consumer
behaviour and focus on creating offline WOM to influence consumers.
Key Words: Word of Mouth, WOM, Consumer Decision Journey, CDJ, Product Involvement, Involvement, Consumer Behaviour, social media, Social WOM, Electric WOM, eWOM, digitalisation
The Influence of Word of Mouth in the Digital Age
In the Netherlands, consumers spent around 68 million euro’s every three months on
consumer goods, in the first two quarters of 2014. This shows the importance of
understanding consumer behaviour for marketers and social scientists. New digital media are
creating new ways for consumers to share product information. These news forms of word of
mouth are also new ways for marketers to influence consumers. Marketers are trying to use
these new possibilities, because traditional advertising are becoming less influential(Rosen,
2000). New possibilities and activities via digital media have helped increase word of mouth
marketing spending, from 76 million dollar in 2001, to an expected three billion dollar in
2013 (“Facebook Statistics”, 2009). But how influential is word of mouth exactly and what is
the influence of new types of word of mouth via digital media?
Introduction
This study will look at the influence of word of mouth or WOM and at the influence of
new types of WOM on consumer behaviour in the digital age. Their influence will be set
against other influence factors on consumer behaviour for comparison. This study will also
look at the influence of different types of WOM for different levels of product involvement.
Involvement has always been strongly related to the way information is processed (Dholakia
& Bagozzi, 2003; Laurent & Kapfener, 1985) and is therefore an important factor to consider.
Finally, this study will look at how the moment during the purchase process in which WOM
is received, effects its influence. This has been unexplored in the current literature, as far as
the researcher can tell. This study will help create new insights in the influence of WOM in
the digital age.
Word of Mouth
Word of mouth has been a topic of interest to psychologists and communication
through WOM conversations. Recent studies by Keller (2007) and Trusov, Bucklin, and
Pauwels (2009) showed that WOM is one of the most important influences on consumer
behaviour and is becoming more and more influential. Crotts (1999) argued that after prior
knowledge, information from friends or family is most important on purchase decisions. More
so than advertising influences. Harrison-Walker (2001) defines word of mouth as follows;
“WOM is informal communication between a perceived non-commercial communicator and one or more receivers regarding a brand, a product, an organization or a service”.
When studying the influence of WOM it is important to understand how it works.
Bughin, Doogan and Vetvik (2010) describe three important factors determining the value of
a WOM message. Firstly there is the sender. Aspects of the sender like familiarity (Brown &
Reingen, 1987; Steffes & Burgee, 2009), expertise (Bansal & Voyer, 2000; Lim & Chung,
2013; Ohanian, 1991; Woodside & Davenport, 1974) and similarity (Cialdini, 2009) are
positively related to the perceived importance of WOM. Secondly there is message content. ,
For example, a message with strong arguments is more influential, for people who are
motivated to process these arguments elaborately (Cacioppo & Petty, 1984). Research by
East, Hammond and Lomax (2008) showed that positive WOM messages were more
influential on purchase behavior than negative messages. So message content, such as quality
of arguments and sentiment help determine WOM influence. The third factor is the place of
communication or context. For example, if a conversation is in public, one might the suspect
the sender to make his message more socially acceptable than when in private. Another
example is that it might be easier for people to lie online than in real life. This shows the
relevance of context in which WOM is received. Besides these three factors it is important to
also consider the receiver as a factor. Pre-existing knowledge and motivation to process
interpreted. So the influence of a WOM message depends on the sender, the content, the
context and the receiver (Bughin et al., 2010).
Digitalization has created new ways of sharing word of mouth. Two important new
types are electronic WOM and social WOM. Electronic WOM or eWOM is word of mouth on
the internet like blogs, reviews and forums. Social WOM is via social media and can be
considered part of eWOM. However because social media have become such a large and
extraordinary part of people’s lives, this study will consider social WOM as a separate type of
WOM. For example, Facebook alone has over 1.31 billion monthly users who spend an
average time of 15 hours and 33 minutes per month on Facebook. To interpret the influence
of these new types of WOM compared to offline WOM, it is important to see how many
people use and how influential people consider these types of WOM.
Offline WOM, eWOM, and social WOM usage. In current western society most
people are online, for example in 2014, 96.5% of the Dutch were online. Social media usage
is also very common, with 90% of the Dutch population using social media (Hendriksma,
2014). However this does not mean WOM via these media types is very frequent. Keller and
Fay (2012) showed that most WOM communication happens offline. They measured this by
making participants keep a journal of all the brand interactions and conversations they had.
This showed that 90% of all WOM conversations are offline conversations. The first
hypothesis of this study therefore is;
Hypothesis 1: offline WOM is used more often than eWOM and social WOM
There are no expectations that eWOM and social WOM differ in usage during the purchase
process. Besides usage it is also important to consider the influence of different types of
WOM.
Offline WOM influence vs eWOM influence. Offline WOM is the traditional form
sender and one or more receivers and is mostly between family, friends or acquaintances.
Keller (2007) showed that most WOM conversations happen offline. Herr, Hardes, and Kim
(1993) demonstrated that face-to-face given information concerning a product was more
influential than a printed format containing the same information. Cialdini (2009) offered that
familiarity with the source makes information more influential. EWOM is also found to be
influential (Liu, 2006). Chevalier and Mayzlin (2006) showed that positive reviews about
books on a webshop significantly increase the sale of those books. EWOM differs from
offline WOM in it can travel quicker, can have a much larger audience and have multiple
contributors (Lee, 2000). EWOM is also often written and therefore available longer (Kiecker
& Cowles, 2002). These factors help eWOM messages get a large reach, however this large
reach also makes eWOM messages less personal. With eWOM the sender is often is
unfamiliar to the receiver making it less influential (Cialdini, 2009). With offline WOM the
sender is often more familiar making the message more influential. Also eWOM, can be
created by hired brand agents making it less credible. For example, in 2013 a yoghurt
company hired several companies to spread positive reviews online (Rushe, 2013). This
makes eWOM appear less trustworthy. These arguments show that the sender factor (Bughin
et al., 2010) of eWOM is in dispute, making it likely that eWOM is less influential than
offline WOM.
Hypothesis 2: eWOM is less influential than offline WOM.
Offline WOM and eWOM influence vs Social WOM influence. Research has
shown that many people on social media often do not interact with WOM message content
(Kimmel & Kitchen, 2014). For example, data provided by Facebook shows that most large
brandpages can have millions of fans, but on average only 0.45% of these fans is active on
these brandpages (Keller, 2012). Social WOM senders are often familiar than eWOM senders.
and is constantly being refreshed and updated (Leskovec, 2011). This makes them less
informative and they are more difficult to process within this blur of information. Social
WOM is also less familiar and not face to face compared to offline WOM. These arguments,
lead to the third hypothesis.
Hypothesis 3: Social WOM is less influential than eWOM and offline WOM. As mentioned earlier, other influences such as advertising and prior knowledge or
experience also influence consumer behavior. To gain a full understanding these factors will
also be measured in this study, to help interpret the relative influence of WOM. Crotts (1999)
argues that most consumers will primarily trust and use their prior knowledge after this offline
WOM is most important followed by advertising. This study exploratory research how
influential eWOM and social WOM are compared to these factors and possible other
influences on consumer behavior.
When assessing the influence of WOM on a purchasing process, it is important to take
into account what product people are purchasing. This is because the way people process and
interpret information like WOM is related to the level of involvement they have with the
product they are purchasing (Petty, Cacioppo, & Schumann, 1983; Laurent & Kapfener,
1985).
Involvement
Mitchell (1979) defined consumer involvement as “an internal state variable that
indicates the amount of arousal, interest or drive evoked by a product class” and this arousal,
interest or drive can determine the level of influence by factors like word of mouth.
Consumer involvement can be divided into two types (Laurent & Kapfener, 1985;
Richins & Bloch, 1986,). The first type is enduring involvement which is ongoing
involvement with a product category and is independent of specific purchase situations. The
specific situation (Bloch & Richins, 1983). This can for example be caused by the costs of a
purchase, a first time purchase, or the purchase of a scarce product. Situational involvement
often dies out shortly after the purchase and is often associated with perceived risk and
perceived importance (Dholakia, 2001; Laurent & Kapfener, 1985). This study will look at
the influence of situational involvement on WOM.
Involvement is a motivational construct that influences how consumers search and
process information (Dholakia & Bagozzi, 2003; Laurent & Kapfener, 1985). Cacioppo and
Petty (1984) showed that when making a decision, highly motivated people do a lot of
research and look specifically for credible and objective information, while unmotivated
people will not look for information and are less critical of information. This is in line with
the MODE-Model by Fazio (1990). This model states that when there is both motivation and
cognitive capacity, people use a deliberate processing mode in which objective information is
actively sought to make a good decision. So with high involvement consumers are motivated
to look for information (Richins, Bloch & McQuarrie, 1992; Cacioppo & Petty, 1984) WOM
is considered as credible and objective information (Steffes & Burgee, 2009). This means that
for higher levels of involvement, people will look for word of mouth making it more
influential. This study does not expect this effect to differ for different forms of WOM
because all are informative. This leads to the following hypothesis.
Hypothesis 4: Influence of offline WOM, eWOM and social WOM is positively related to level of product involvement.
The Consumer Decision Journey
During a purchasing process consumers can go through different stages, in which they
can become aware of products, think about what they already know, look for extra product
information or advice, and make a final decision (Crotts, 1999; Court, Elzinga, Mulder, &
purchasing process. To study the effect of WOM in different stages, this study differentiates
between four stages in a purchase process. These stages are derived from the consumer
decision journey or CDJ (Court et al., 2009).
The CDJ is a model for consumer behaviour developed by Court et al. (2009), by
examining the purchase decisions of over 20.000 consumers. The model states that consumers
are constantly becoming aware of brands through factors like advertising, Word-of-Mouth,
and experience. This is called the awareness stage and is the first stage of the CDJ. If there is
a desire to purchase a product consumers go to the next stage in which an initial set of brands
is considered for purchase. This is called the initial consideration stage. After this consumers
start to gather information about brands and compare different brands, called the active
evaluation stage. Finally consumers will make a final consideration and purchase a product. This stage is called the moment of purchase. Consumers also indicate a post-purchase
experience in which information is gathered from the purchase to inform a next decision
journey (Figure 1). However this last stage is not considered in this study, because it is not
Figure 1. Visual representation of the Consumer Decision Journey proposed by Court et al. (2009).
The decision process in the CDJ has many similarities with rational decision models
from social psychology such as the constraints model by Janis (1989) in which good decision
makers first consider all possible options, similar to the initial consideration stage. Then they
canvas a wide range of alternative actions and search for and incorporate new information,
similar to the active evaluation stage. Finally they reconsider the pros and cons of all the
options and make a decision, similar to the moment of purchase. The stages in the CDJ are
also consistent with other decision models from psychological research (Darley, Blankson, &
Luethge, 2010; Engel, Blackwell & Miniard, 1986; Klein & Calderwood, 1991). These
models all show that when making a decision, people consider their options (initial
consideration), evaluate different possibilities (active evaluation) and make a decision
followed by action (moment of purchase). Considering this, the CDJ is a sufficient model to
study the influence of WOM in different stage of a purchasing process. Its wide usage in
In the awareness stage consumers can become aware of brands via several sources.
WOM does not hold any exceptional influence during this stage compared to other influences.
During the initial consideration stage however WOM has been shown to be very influential on
consideration processes (East et al., 2008). Bughin et al. (2010) confirmed this notion, by
showing that WOM is one of the most important influences during the initial consideration
stage for mobile phones. Court et al. (2009) showed that during the active evaluation,
consumers gather information and compare different brands. Because WOM is considered to
be credible and trustworthy information (Steffes, & Burgee, 2009) it is expected to also be
very influential in this stage. At the moment of purchase a final consideration is made
followed by the actual purchase. In this stage, consumers are relatively close to a purchase,
both in time, psychological distance and sometimes actual distance. According to construal
level theory (Liberman, Trope, & Waslak, 2007), when someone experiences low
psychological distance, this leads to the use of low level construals for processing information
which is related to higher perceived risk. High perceived risk leads to more information
search like WOM (Dholakia, 2000; Nickerson, 1998; Schuette, & Fazio, 1995). This would
make WOM more influential at this stage, at least compared to the awareness stage. So WOM
is expected to be less influential in the awareness stage and more influential in the other
stages. Because all types of WOM are considered information, this study expects this to go
for all types of WOM. This leads to the last hypothesis of this study.
Hypothesis 5: Offline WOM, eWOM and social WOM will be less influential for the awareness stage than for initial consideration stage, the active evaluation stage, and the moment of purchase.
Current study
In the current study, two focusgroups were held to gain insights in the purchasing
method section these focusgroups and its results will be discussed first. Following this the
methodology of the quantitative study will be discussed. To measure the influence factors
offline WOM, eWOM and social WOM and other factors like advertisement and prior
experience (Crotts, 1999), influence of different information sources was measured.
Information sources were for example ‘conversations with family’, ‘TV commercials’,
‘Internetreviews’, or ‘Brandlikes on Facebook’. These were grouped to form the proposed
influence factors. Three different levels of product involvement were created (high, medium
and low) by letting participants imagine buying products of different cost, a car, smartphone
and bag of potato chips (Bloch, 1981; Dholakia, 2001; Hupfer & Gardner, 1971; Laurent &
Kapfener, 1985). The different stages of the CDJ were recreated through several scenarios.
Through these manipulations this study measured the usage and influence of different types of
WOM and other important influence factors on consumer behaviour, for different levels of
product involvement and in different stages of the consumer decision journey.
By researching the influence of different types of WOM under different circumstances
this study aims to gain insights on the influence of new types of WOM on consumer behavior.
In the following two sections, focusgroups and a quantitative study are described that were
done to show how much these different types of WOM are used and how influential they are.
Focusgroups Method
To explore how the CDJ is experienced and which information sources consumers use,
two focusgroups were organized prior to the quantitative study. This helped create the
manipulations, the used questionnaire and the manipulationcheck questions for the
quantitative study.
Focusgroups Participants
The first focusgroup contained ten participants (5 men, 5 women) selected by
having bought a new car. The second focusgroup contained eight participants (4 men, 4
women). These participants were recruited via the personal network of the researcher.
Requirement for participation was having bought a new smartphone in the past two years.
Participants received €40, - either in cash or by money transfer, after participation.
Focusgroups Materials
The focusgroups were held in a laboratory room of the University of Amsterdam,
containing audio and video recording equipment. The room further contained, one large table,
13 chairs and a flipchart.
Focusgroups Procedure
Upon arrival participants were invited to sit around the table with the researcher, a
moderator and a secretary present. The moderator would interview the group about their
purchase behavior for different products assisted by the researcher. The flipcharts were used
to structure their answers. The focusgroups lasted two hours and there was no break. Both
sessions were transcribed and recorded. Anonymity was guaranteed and transcriptions and
recordings were only accessible to the researcher.
Focusgroups Results
The focusgroups offered many insights concerning the experienced stages during the
CDJ. It showed that for low involvement products the different stages are less pronounced.
Results also indicated that when there is a strong brand preference consumers do not go
through the different stages of the CDJ but just buy their preferred brand. The main result
from both focusgroups were information sources participants reported used during the CDJ,
which were used to construct the questionnaire in the quantitative study.
Quantitative Study Method Participants
For the main quantitative study, a total of 450 (M age = 44.29, SD age = 13, 203 men,
225 women, Range 21-65) people filled in the questionnaires. 22 Questionnaires were
discarded due to missing data or straight-lining. Straight-lining happens when a participant
gives the same answer, for a subset of questions. This can be due to a lack of motivation, high
task difficulty or low performance ability and it can undermine data quality (Cole,
McCormick, & Gonyea, 2012). Because there are little clear selection criteria on when to
discard questionnaires because of straight-lining, this study only discarded a questionnaire if
participants answered all multiple choice questions the same. Participants were selected by
SSI sampling agency from their database, based on their purchase history. Participants were
distributed over three conditions for level of involvement, low, medium and high
involvement. This was done based on purchase history. Participants who had ever bought a
new car (high cost) were assigned to the high involvement condition (n = 142). If not, they
were asked if they had bought a new smartphone (medium cost) in the last two years. If so,
they were assigned to the medium involvement condition (n = 140). If not then participants
were asked if they had bought a bag of potato chips (low cost) in the past three months. If so,
they were assigned to the low involvement condition (n = 146). This allocation was done to
ensure participants were able to imagine the purchase process for the products in their
condition. If participants had not bought any of these products they were not considered for
the study. Before the start of the study, informed consent was gathered by asking people to
tick an ‘I agree’ or ‘I disagree’ box after having the terms read of their participation.
Design
Factor analyses for the measured information source items resulted in six different
influence factors (offline WOM, eWOM, social WOM, low information influences,
advertising and personal experience, see results page 24-26). The following mixed design was
influence factor as within subjects variable (offline WOM, eWOM, social WOM, low
information influences, advertising and personal experience) * stage of the CDJ as within
subjects variable (awareness, initial consideration, active evaluation, moment of purchase).
Procedure and Materials
Participants were invited via email to fill in the questionnaire. This could be done at
home on a desktop computer or a laptop, via Qualtrics. The questionnaire was newly
developed for this study and started with informed consent and demographic questions
concerning age, gender and educational level.
Level of involvement. After this participants were asked to answer a maximum of
three questions concerning their purchase history, as described in the participant section.
These questions would indicate to which between subjects condition for level of involvement
participants would be assigned. The particular products used to represent different levels of
involvement based on cost, were taken from by previous studies (Court et al., 2009; Richins,
& Bloch, 1986; Traylor, & Joseph, 1984).
The consumer decision journey. After this, participants in all conditions would read
the first scenario. Scenarios in different conditions for level of involvement were almost
identical except for the type of product. The first scenario recreated the awareness stage of the
CDJ. In this scenario participants were asked to write down all the brands they knew
concerning the product in their condition. This would make them think about all brands they
were aware of and thus recreate the awareness stage. Participants then filled in a
questionnaire measuring influence of different information sources, called the influence
questionnaire. After this participants would read the second scenario which recreated the
initial consideration stage. Participants were asked to write down all brands they would consider buying for the product in their condition. After this participants would fill in the
scenario participants were asked to imagine that they had to narrow down their list of possible
brands by researching, comparing and evaluating the brands they were considering. This was
followed by the influence questionnaire. Finally, in the last scenario for the moment of
purchase, participants were asked to imagine they were about to make a final decision
between brands and then make the actual purchase. After this the influence questionnaire was
filled in. This way influence was measured for all four stages of the CDJ, in all three
conditions for involvement.
The influence questionnaire. The influence questionnaire measured the influence
factors using three kinds of information source lists. One list with offline sources, a list with
online sources, and five lists with social media sources. After a scenario was read, a statement
was presented. For the awareness stage the statement was ‘I knew these brands through…’.
For the initial consideration stage it was ‘When I narrow down to brands I would consider
buying, I am influenced by …’. For the active evaluation stage it was ‘When considering and
researching different brands, I am influenced by…’. For the moment of purchase it was
‘When making my final decision and purchasing a product, I am influenced by…’. After each
statement participants would indicate how much they agreed to this for the first list of offline
information sources (Figure 2). This was done one a 7-point Likert scale one which one
indicated totally disagrees and seven indicated totally agrees.
Figure 2. Example of the influence statement and the used Likert scale, after the first scenario.
Offline information source list. Offline information source items that fit the definition for WOM by Harrison-Walker (2001) were grouped into factor, ‘offline WOM’. These were;
advertisings, radio advertising, magazine advertising, newspaper advertising, billboard
advertising and in store advertising were expected to form the factor advertisement. Personal
experience and product testing were expected to for the factor personal experience. For
consumer magazines, brands people encounter on the streets, brands in movies, advertising in
cinemas, and consumer programs there were no prior expectations about possible factors.
There was also a general internet item added.
Online information source list. If participants scored the general internet item with a three or higher, it was assumed that they found the internet at least a little influential during
this stage of the CDJ. They were then asked to fill in a second list, containing specific online
information source items, called the online list. If they scored a one or a two for internet
influence, they would go to the next scenario. This was done because it was expected that if
people did not find internet influential it would lead to frustration and boredom if they had to
answer specific internet items. This could influence their performance in the remainder of the
study (Ratcliff, 1993), lead to higher drop-out rates or straight-lining. Items on the online list
that fit the WOM definition by Harrison-Walker (2001) were expected to create the influence
factor eWOM. These were; forums, blogs, reviews on webshops, reviews on reviewsites and
comparison websites. Other items that were not considered eWOM were search engines,
popup commercials, banner advertising, brandsites, webshops and email-advertising, there
were no expectations for possible factors for these items. Social media was also added as an
item.
Social media information source lists. Similar to the general internet item, if participants scored a three or higher on the Likert scale for the general social media item, it was assumed
they found social media at least a little influential in this stage of the CDJ. If they scored a one
or a two on the social media item they would go to the next scenario. If they scored a three or
ten different social media that participants could indicate using. These were Facebook,
YouTube, Twitter, LinkedIn, Instagram, Google+, Pinterest, Tumblr, Snapchat, Vine or the
option ‘something else’. If participants indicated using Facebook, YouTube, Twitter,
Google+, LinkedIn or Instagram then they would have to indicate what the influence of
specific information sources of these social media was (table 1). This was not done for the
other social media because they were not commonly used within the population (Oosterveer,
2013). Social media items, that fit the WOM definition by Harrison-Walker (2001), were
grouped to create the influence factor social WOM (table 1). There were no expectations
about other factors. When participants finished the social lists they went to the next scenario.
For a schematic overview of the influence questionnaire see figure 3.
Manipulation checks. After the final influence questionnaire was filled in, participants
were asked two manipulationcheck questions. If they found the stages of the CDJ realistic and
if they were able to really imagine themselves going through the stages during the study. Both
questions were answered on a Likert scale from one (very unrealistic, very hard to imagine)
until seven (very realistic, very easy to imagine). Finally there was a question about whether
or not they had a previous brandpreference for the product in their condition, and which brand
this was. After this they were thanked for their participation and instructed that for more
information concerning the study or their results they could contact the researcher via email.
Participants were not rewarded for their participation. For a schematic visualization of the
complete questionnaire see figure 4.
Exploratory
This study looked at interactions between the three independent variables exploratory.
It also looked exploratory at the other influences on consumer behavior and their interaction
showed that previous brand preference influenced the consumer decision journey it was added
as a covariables besides the manipulationcheck.
Table 1. WOM and non WOM information sources used for the five main social media.
Facebook Twitter YouTube LinkedIn Instagram
-posts about brands by friends -shares of brands by friends -shared brandcontent by friends
-private messages about brands
-invites by friends to like a brand
-tagged brands by friends
-Noncommercial tweets about brands -noncommercial retweets about brands -personal messages -trends on Twitter -Comments -Brands in noncommercial videos -number of views -thumbs up/likes on YouTube -video blogs -video reviews -posts by connections -pictures by friends -video’s by friends -pictures friends liked -videos friends liked
Non WOM information sources - ads on your Facebook
Timeline
- posts by brands you’ve liked
- brand Facebook groups - ads in the right-side ad bar
- brandpages on Facebook
- apps, games or contests from brands - Advertisement tweets - Twitterpages from brands - Tweets from brands you follow
- Brand pages - recommendations by YouTube - banner ads on YouTube - pre-roll ad-videos - Posts by brands - ads on LinkedIn - brandpages - pictures by brands - videos by brands
Results
428 Questionnaires were considered for analysis. First, factor analyses were done for
different information sources. This was done to identify the different influence factors on
consumer behavior.
Factor Analyses and Reliability Analyses
A principal component analysis (PCA) was done on the 22 items of the offline list. An
orthogonal rotation was used (varimax). Kaiser-Meyer-Olkin measure verified the sampling
adequacy for the analysis, KMO = .94 (‘superb’ according to Field, 2009), all KMO values
for individual items were > .82 which is above the limit of .5 (Field, 2009). Bartlett’s test of
sufficiently large for PCA. Four components showed eigenvalues over Kaiser’s criterion of 1
and explained 76.57% of all variance, combined. The scree plot showed a slight inflexion
after 4 components. So these were retained for final analysis.
The first component consisted of information sources that people do not encounter
very often, such as brands in consumer programs or that are not very informative such as
brand placement in movies. This component was named ‘low information factor’ (6 items; α
= .938). The second largest component contained all offline WOM information sources and
also the information source ‘talking to a salesmen’. This was not be considered WOM, and
was deleted from this component. This created a higher level of reliability (5 items; α = .942). The third component consisted of only advertisement information sources and the item
‘brands you see on the streets’. This item could be considered as advertising, so the factor was
retained and called 'Advertising', (5 items; α = .926). The fourth component consisted of two information sources, ‘previous product experience’ and ‘product testing’. This component
was called ‘personal experience’ (2 items; α = .676).
Another factor analysis was done for the 13 online items. A principal component analysis (PCA) was done and orthogonal rotation was used (varimax). Kaiser-Meyer-Olkin
measure verified the sampling adequacy for analysis, KMO = .93 (‘superb’ according to
Field, 2009). Bartlett’s test of sphericity showed χ2 (66) = 2532.54 p < .001, indicating that the correlations between items were sufficiently large for PCA. Two components combined
showed eigenvalues over Kaiser’s criterion of 1 and explained 68.77% of variance. The two
components could be considered ‘often used online information sources’ and ‘less used online
information sources’. Proposed eWOM information sources were divided over these two
components. To still be able to compare eWOM with other influences it was decided to retain
the original items of the eWOM factor. The eWOM factor was highly reliable (5 items; α =
A factor analysis was done over all 38 social media items. This gave a ‘non positive
definite matrix’ indicating there were negative eigenvalues. To counter this it was decided to
limit the amount of items (Field, 2009). This was done by eliminating items from the three
least used social media, Instagram, LinkedIn, and Google+ (D. Oosterveer, 2014). Factor
analysis was done for the remaining items for Facebook, Twitter and YouTube. First a
principal component analysis (PCA) was done. An orthogonal rotation was used (varimax).
Kaiser-Meyer-Olkin measure did not verify the sampling adequacy for the analysis, KMO =
.47. After inspection three items were removed that scored extremely low on the anti-image
correlation matrix, ‘private twitter messages’, ‘pre-roll advertising on YouTube’ and
‘advertisement posts on Facebook’. After this, a Kaiser-Meyer-Olkin measure verified the
sampling adequacy for the analysis, KMO = .61 (‘mediocre’ according to Field, 2009).
Bartlett’s test of sphericity was χ2 (378) = 1349.092 p < .001, this indicated that the correlations between items were sufficiently large for PCA. Seven components combined
showed eigenvalues over Kaiser’s criterion of 1 and explained 88.35% of variance.
Component one contained a total of six social WOM information sources. This component
was considered the best representative factor for social WOM (6 items; α = .956). The
remaining factors were not considered useful for comparison. All factor analyses led to a total
of six factors (Table 2).
Normality and Manipulation Checks
A Kolmogorov-Smirnov test was done to check if influence for the factors offline
WOM, eWOM and social WOM was normally distributed. Results were significant, offline
WOM, D(428) = .08, p < .05, social WOM, D(428) = .43, p < .001, low information factor,
D(428) = .07, p < .001, advertising factor, D(428) = .06, p < .01, and personal experience
D(428) = .09, p < .001 indicating they were not normally distributed. The test was
large samples (N > 30) will have a normal distribution with an equal mean to the population
(Field, 2009) despite a significant Kolmogorov-Smirnov test. Therefore normality was
assumed for all factors.
A one sample t-test was done for the manipulation check for the stages of the CDJ,
which was run to see if participants found the stages of the CDJ realistic. Scores for the
manipulationcheck (M = 3.62, SD = .46) were significantly lower than the median (ME = 4),
t(427) = -16.80, p < .01 (two-tailed), r = .63. Indicating that participants found the different stages of the CDJ not realistic. Manipulationcheck was added as a covariate to control for the
fact that some participants might not have been able to imagine going through the stages of
the CDJ. To counteract multi-collinearity and create more meaningful results (Field, 2009) the
variable manipulationcheck was centered.
It was also checked if participants had a brand preference for the brand in their
condition. Descriptives showed 63.7% of participants had a brand prevalence for chips, 72.9%
control for the effect of previous brand preference, this variable was recoded so that ‘no’ was
0 and ‘yes’ was 1 and then it was added as a covariate in the mixed measures analyses.
Descriptives for Offline, Online, and Social Media Usage
Due to the setup of the research it was not possible to perform chi-square tests or
log-linear analysis between the amount of people using offline items, internet items and social
media items, because not all participants had to answer all questions. Descriptives were used
to see if participants differed in usage of offline, internet and social media information during
purchasing processes. It was assumed that all participants used offline information sources
during a purchase decision. Descriptives showed that internet (n = 296) and social media
(n=155) were found less influential by participants. This is lower than lower than the use of
offline information sources (n=428). This supports the idea that internet and social media are
used less often during purchase decisions than offline information.
Mixed Measures Ancova for Offline Factors
Offline information factors were compared using a mixed measures Ancova design.
The four offline factors, offline WOM, low information factor, advertising and personal
experience and the four stages of the CDJ were measured within subjects and the three levels
of involvement between subjects. The manipulationcheck for the CDJ and previous
brandpreference were added as covariates. Because of the large sample size only significant
results with partial eta squared above .02 (Cohen, J. 1988) are reported. The CDJ
manipulationcheck interacted significantly with the amount of influence participants reported,
F (423, 1) = 28.91, p < .001, ηp2 = .64. When looking at a scatterplot it showed that when
participants found the CDJ less realistic they also reported less influence in general. However
this did not differ between factors. Analyses showed Mauchly’s test of sphericity was
violated, degrees of freedom were corrected using Huyn-Feldt (ε = .801). There was a main
WOM was significantly less influential than personal experience F(1, 423) = 24.76, p < .001,
ηp2= .06. Bonferroni post hoc tests revealed that means for all offline factors were
significantly different making personal experience most influential, followed by offline
WOM, advertising and finally low information influences (Table 3).
There was a weak main effect for stage of the CDJ, F(3, 1269) = 9.30, p < .001, ηp2=
.02. Planned contrasts revealed that effect sizes for individual differences between stages were
too small to be considered. Looking at means per stage there is a drop in influence for offline
factors over all stages, with for the awareness stage (M =4.00, SD = .05), initial consideration
stage (M =3.70, SD = .05), active evaluation stage (M = 3.69, SD = .05) and moment of
purchase (M = 3.56, SD = .06).
There was a significant interaction effect between the offline factors and level of
involvement F(6, 1269) = 28.06, p < .001, ηp2 = .12. A barchart of the interaction effect in
figure 5 shows the offline factors differ in influence over the different levels of involvement.
For low involvement products, personal experience is most influential followed by advertising
and offline WOM and finally low information influences. For medium involvement products,
offline WOM is more influential than for low involvement products. In this condition
personal experience is a less influential and similar to offline WOM. Advertising and finally
high involvement products personal experience is more influential than offline WOM
followed by advertising and finally low information influences. Most important is that offline
WOM is more influential for medium and high involvement than for low involvement (Table
4).
*The Capitol letters below each bar indicate if the means of an offline factor differ significantly for different levels of involvement, different letters indicate significantly different means, p < .05.
** The case letters below each bar indicate if the means within a level of involvement differ significantly between the different offline factors, different letters mean significantly different means, p < .05.
1 2 3 4 5 6 7
Offline WOM Low Information Advertisement Personal Experience
Offline Influence per level of Involvment
There was a significant three way interaction between offline factors, level of
involvement and stage of the CDJ F(18, 3807) = 8.90, p < .001, ηp2 = .04. When looking at
plots for interpretation (figure 6) , influence from offline factors was relatively stable over the
different stages, for low involvement products. For medium involvement however there was a
drop in influence from advertising between the awareness and initial consideration stage and a
small rise between these stages for personal experience. For high involvement products
influence by advertising and low information dropped between the awareness and the initial
consideration stage and between the active evaluation stage and the moment of purchase. For
personal experience, influence rose between the awareness and initial consideration stage and
between the active evaluation stage and the moment of purchase.
Exploratory this study looked at interaction effects for age and gender with offline
factors for different stages of the CDJ. Results showed no significant interaction effects with
age and gender.
Mixed Measures Ancova for offline factors, eWOM and social WOM
Analyses for offline factors, eWOM and social WOM could not be performed because
there were only 51 participants who filled in all information source lists in all different stages
of the CDJ. These participants were not evenly divided over all three levels of involvement
(table 5).
Table 5. n-values for filled-in information lists in all stages of the CDJ, for the different levels of involvement
This
crea
ted a very low power of analysis. Therefore analysis was run with only users of eWOM and
the offline factors.
Low involvement Medium Involvement High Involvement Total
Offline List 146 140 142 428
Online List 27 68 49 144
Analyses showed Mauchly’s test of sphericity was violated, degrees of freedom were
corrected using Huyn-Feldt (ε = .76). The CDJ manipulationcheck significantly interacted
with the amount influence participants reported, F(1, 139) =11.09, p =.001, ηp2 = .08. There
was a main effect for type of factor, F(4, 556) = 2.80, p = 0.39, ηp2 = .02. Post Hoc tests
revealed that eWOM (M = 3.93, SD = .09) was significantly less influential than offline
WOM (M = 4.59, SD = .08), advertising (M = 4.43, SD = .08), and personal experience (M =
5.14, SD = .09), p < .001.
Analyses also showed an interaction effect between influence factors and level of
involvement F(8, 556) = 4.55, p =.001, ηp2 = .06. Contrasts revealed that eWOM was
significantly more influential in the high involvement than in the low involvement condition.
In the low involvement condition personal experience was significantly more influential than
all other factors. Followed by advertisement which was significantly more influential than
eWOM and low information influences. Offline WOM did not differ significantly from
advertisement, low information influences and eWOM. For medium involvement products
personal experience was significantly most influential. After that came offline WOM which
was significantly more influential than advertisement, eWOM and low information
influences. EWOM and advertisement did not differ significantly from each other but were
both more influential than low information influences.For high involvement products
personal experience was most influential.After that offline WOM and advertisement were
most influential but they did not differ significantly from each other. They were both
significantly more influential than low information influences and eWOM (Figure 7, table 5
Figure 7: Offline influence factors and eWOM per level of involvement with error bars. 1 2 3 4 5 6 7
Offline WOM eWOM Low Information Advertisement Personal Experience
Offline Influence and eWOM influence per level of
Involvement
Results also showed a significant three-way interaction between influence factors,
level of involvement and stage of the CDJ F(17.89, 1243.39) = 3.27, p >.001, ηp2 = .05. When
looking at plots for interpretation (figure 8) it seems that offline WOM and low information
influences become more influential in later stages for low involvement products but remain
more stable over allstages for medium and high involvement products. EWOM seems to
remain stable over different stages for different levels of involvement for people who use
internet as an information source.
To compare the influence of social WOM to eWOM and the offline factors, it was
decided to not compare the different stages of the CDJ, but do all analyses per stage
separately. However, samples for the low involvement condition were still considered too
small. So, it was decided not to include the low involvement condition. Remaining sample
sizes were considered large enough and created sufficient power for analyses (Table 7).
Table 7. n-values participants who filled in all information source lists for the awareness stage (awa.), initial consideration stage (ini.), active evaluation stage (act.) and moment of purchase (mom.) for each level of involvement.
The centered manipulationcheck for the CDJ and brandpreference were entered as
covariates. The different factors did not differ significantly in the awareness stage or the
initial consideration stage. For influence in the active evaluation stage, Mauchly’s test of
sphericity was violated, χ2 (14) = 78.89, p <.00. Degrees of freedom were corrected using
Greenhouse-Geisser (ε = .73). Covariates were not significant. Results showed a main effect
for type of influence during the active evaluation stage, F(5, 365) = 3.21, p = .016, ηp2 = .04.
Post hoc tests using Bonferroni correction showed that offline WOM was most influential
together with personal experiencefollowed by eWOM, social WOM, advertising who were
equally influential and finally the low information factor was least influential in the active
evaluation stage (Table 4).
Low Involvement Medium Involvement High involvement Awa. Ini. Act. Mom. Awa. Ini. Act. Mom. Awa. Ini. Act. Mom.
For influence at the moment of purchase, Mauchly’s test of sphericity was violated χ2 (14) = 58.37, p <.00. Degrees of freedom were corrected using Greenhouse-Geisser (ε = .71). Covariates were not significant. A main effect was found for type of influence, F(5,300) =
3.42, p =.013, ηp2 = .05. Results from post hoc using Bonferroni correction showed that
personal experience was most influential followed by offline WOM. Offline WOM was more
influential than social WOM, eWOM and advertising influences but these three did not
significantly differ among each other. Low information influences were significantly less
Discussion
Results showed internet and social media are used less often than offline information
sources. This supports the first hypothesis. Offline WOM was more influential than eWOM
for participants who used both offline information and internet. This supported the second
hypothesis. During the active evaluation stage and the moment of purchase, for medium and
high involvement products, social WOM was less influential than offline WOM and equally
as influential as eWOM. This partially supports the third hypothesis. Offline WOM was found
to be more influential for high and medium involvement products than for low involvement
products. For low involvement products eWOM was less influential than for high
involvement products. This does not support the fourth hypothesis that involvement is
positively related to WOM influence. There was no evidence found that supported hypothesis
five.
WOM usage
Descriptives showed, internet was used by 69%, and social media by 36%, of
participants during a purchase decision. It must be noted that internet and social media usage
was particularly low for low involvement products. People are not motivated to look for
information for these products (Cacioppo & Petty, 1984), but rely mainly rely on their own
experience. However if information is presented people are inclined to use it after all.
Considering 90% of all WOM is offline and it is extremely frequent and easy to encounter
(Engel, Kegerreis & Blackwell, 1969; Keller, 2007) it is likely to be used more often than
eWOM or social WOM that are only found online. Internet and social media were used more
often for medium and high involvement products because consumers are more motivated to
look for information for these products. However, they are also more motivated to look for
influential, making it likely that offline WOM is used more often than eWOM or social
WOM. However, more definitive research is needed to confirm this notion.
Offline WOM influence
Offline WOM was the most influential factor on consumer decision making after
personal experience, more so than advertising, low information influences, eWOM and social
WOM. (Keller & Fay 2012; Crotts 1999). Especially when people are highly involved, they
will look for more information, such as WOM. Offline WOM was shown relatively more
influential for higher levels of involvement. Advertising became relatively less influential for
higher levels of involvement, showing that motivated people look for more trustworthy
information. Advertising was found to be just as influential as offline WOM for low
information influences. For these products, people are probably not motivated to look for
quality of information and just use all information that is presented or that they already have
(prior experience). So, offline WOM is the most influential external factor on consumer
behavior when people are motivated to look for trustworthy and credible information.
The influence of eWOM
Analyses showed that eWOM was less influential than offline WOM and even less
influential than advertising. This shows that familiarity (Cialdini, 2009) is an important factor,
concerning credibility of WOM. Also the fact that this type of WOM is not face-to-face might
add to the lack of influence (Kiecker & Cowles, 2003). Knowing and seeing the person who is
giving you advice appears essential for WOM influence. This is logical because it makes it
easier to trust and assess the quality of WOM (Rocco, 1998). Although eWOM is easy to find
and often comes from previous users or even experts, the unfamiliarity of the source makes
people wary. This causes eWOM to be less trusted and therefore less influential than offline
However the fact that advertisement is more influential than eWOM is strange.
Although the source is unknown with eWOM one would expect it to be more trustworthy and
influential than with advertising. An explanation can be found in the low involvement
condition, where advertisement was particularly more influential than eWOM. For low
involvement products, people are not motivated to look for information such as eWOM,
however advertisement is often forcefully presented in the form of commercials, ads and
billboards and influences people even when they are not looking for information. For high
involvement products, advertising was also significantly more influential than eWOM. This
might be caused by the fact that cars are very high cost. This could cause people to be
influenced more by bargains and special offers, to save money. These special offers are often
promoted via advertisement. Future studies should try to further specify for which products
eWOM is most influential. For example eWOM might be very influential for medium
involvement products that are sold online or for service products such as travel and
accommodation bookings.
Influence of Social WOM
Social WOM was equally as influential as other factors in the early stages of the CDJ.
However social WOM became less influential than offline WOM and previous experience in
later stages. Social WOM and also eWOM were equally as influential as advertising in the
final stages of the study, for people who indicated using social media in these stages. Because
the apparent failure of the manipulation of the CDJ these results are hard to interpret.
However social WOM seems a potentially influential factor, since active users of social media
deemed it equally influential to factors such as eWOM, advertising and in some cases offline
WOM. The amount of time consumers spend on social media and the familiarity (although
behaviour via social WOM. However more research is needed to discover and show how
social WOM can be best used to influence more people.
Level of Involvement
Although results did not fully support the fourth hypothesis, for low involvement
WOM influence is lower than for medium and high involvement products. The lack of a
difference in between medium and high involvement products might be due to the fact that for
some people a smartphone is a relatively high cost purchase. Participants in the smartphone
condition probably had never bought a car, because then they would have been assigned to the
high involvement condition. This makes it more likely a smartphone was a high cost purchase
for these participants. This can explain why offline WOM was not more influential for cars
than for smartphones. Considering this and looking at literature (Dholakia, 2001); Richins,
Bloch & McQuarrie, 1992) it is possible that there is still is positive relation between level of
involvement and WOM influence. Future studies should re-examine this possible relationship.
The Consumer Decision Journey
There were no interaction effects found between stage of the CDJ and types of WOM
and the manipulation of the CDJ failed. This could be because for low involvement products
an elaborate and rational decision process is not realistic. People are not likely to actually go
through all the different stages when buying a bag of chips. A simple one-way Anova for the
manipulationcheck for different levels of involvement showed that people found the CDJ
significantly less realistic for low involvement products (M = 3.43, SD = .50) than for
medium (M = 3.72, SD = .42) and high involvement products (M = 3.73, SD = .39), F(2, 425)
= 22.19, p < .001. It is more likely they will decide based on habit (Van Eelen, 2014).
However also for medium and high involvement products the manipulation failed. This could
“I don’t even consider other brands, I know how my IPhone works and it satisfies all my needs so when my old phone breaks, I just buy a new IPhone.”
and simply buy the same brand as they always have. Descriptives showed that 67.1% of the
participants already had a strong brandpreference in for the product in their condition. During
focusgroups held for this study this notion also came forward (figure 9). Because of this it
remains unclear if WOM differs in influence in different stages of a purchase decision.
Figure 9. Translated quote from an anonymous focusgroup participant.
This study showed that offline WOM is more influential during the consumer decision
processes than eWOM and social WOM. It also shows that WOM is less influential for low
involvement products than for medium and high involvement products.
Limitations and future research
This study contains several problems and limitations. It was assumed that if consumers
indicated the general internet or social media information source as ‘not influential at all’ or
as ‘not influential’, they did not use these media types. However this is arbitrary. People feel
they are not influenced by media types in order to preserve a positive and objective self-image
(Pronin, Ross, & Gilovich, 2004), but use them nonetheless. Also, the routings used in this
study made it difficult to analyse the usage factors and it was impossible to run chi-squared
tests or log-linear analyses. Also the routing caused small samples, making it necessary to
exclude low involvement participants, when comparing influence factors. The questionnaire
itself was newly developed, which causes concerns about its validity. It is advisable that this
questionnaire is compared to other influence measurement tools to ensure its results are valid
Factor analysis over the online list resulted in one high and one low influence factor.
The eWOM factor was created from items from both factors. So within the eWOM factor
there were differences between high influence and low influence items. Low influence items
could have dragged down the level of influence. For future studies it is interesting to consider
focussing on eWOM items that are relatively common and influential and use these for
comparison. This would prevent rare forms of eWOM that are rarely used and not very
influential to negatively impact the data. A similar approach might also be useful for the
factor social WOM.
For future studies interested in researching different stages of a purchase decision
process it is advisable to not measure low involvement products, because rational models do
not apply to these products. Also participants with strong brand preferences, for the used
product should not be included, because they are likely to skip the decision process altogether.
A final concern is on the funding of this study by Havas Media. Due to the interests of
Havas Media in this study and because the researcher used office space at Havas Media
Netherlands, choices concerning the set up and methodology in this study could have been
biased. Although funding and support by commercial organizations offer great opportunities
for researchers to do large and elaborate studies it also heightens the chance of bias, due to a
conflict of interest between commercial and scientific goals. It is advisable for future studies
to practice caution when accepting commercial funding and make very clear arrangements
concerning the methodology and the reporting of results prior to a study.
Conclusions and implications
Although many people use internet and social media, it is likely more people use
offline WOM. This study showed that offline WOM is more influential than eWOM and
products. When speaking about word of mouth it must be made clear which type of WOM is
meant because eWOM and social WOM differ significantly from offline WOM. This is of
interest to marketing, where many brands and media agencies are investing more and more
money into eWOM and social WOM. It is also interesting governmental groups and public
health organizations who want to influence public opinion and lifestyle. Results show that
their focus should be more on creating offline WOM instead of eWOM and social WOM.
Antecedents of positive offline WOM are mainly brand commitment and satisfaction
(Harrison-Walker, 2001; Brown, Barry, Dacin, & Gunst, 2005). However Keller (2007) also
suggests that traditional advertising is a strong driver of brand related conversations and can
create offline WOM. Spreading eWOM and social WOM can also lead to offline WOM.
Research by Van Eelen (2014) showed that online engagement is correlated to offline
engagement, meaning that people who are fans of a brand online are likely to also be fans
offline. So if people are stimulated to create online WOM they will also do this offline. This
is in line with consistency bias (Cialdini, 2009), which states that people want to be perceived
as consistent. So advertising, eWOM and social WOM can still be used to influence
consumers by creating more offline WOM.
The most important thing to take from this study is that offline WOM is the most
influential external factor concerning consumer behaviour, and eWOM and social WOM are
less influential. Future studies need to look at the relationship between advertising, eWOM
and social WOM on the creation of offline WOM. For now, it is clear that although eWOM
and social WOM can influence consumers, it is offline WOM that really influences consumer
References
Bansal, H. S., & Voyer, P. A. (2000). Word-of-mouth processes within a services purchase
decision context. Journal of service research, 3(2), 166-177.
Bloch, P. H. (1981), "An Exploration into the Scaling of Consumers' Involvement with a
Product Class," Advances in Consumer Research, Association for Consumer Research
8, 61-65.
Bloch, P. H., & Richins, M. L. (1983). A theoretical model for the study of product
importance perceptions. Journal of Marketing, 47(3).
Brown, T. J., Barry, T. E., Dacin, P. A., & Gunst, R. F. (2005). Spreading the word:
Investigating antecedents of consumers’ positive word-of-mouth intentions and
behaviours in a retailing context. Journal of the Academy of Marketing Science, 33(2),
123-138.
Brown, J. J., & Reingen, P. H. (1987). Social ties and word-of-mouth referral
behavior. Journal of Consumer research.
Bughin, J., Doogan, J., & Vetvik, O. J. (2010). A new way to measure word-of-mouth
marketing. McKinsey Quarterly, 2, 113-116.
Chevalier, J. A., & Mayzlin, D. (2006). The effect of word of mouth on sales: Online book
reviews. Journal of marketing research, 43(3), 345-354.
Cacioppo, J. T., & Petty, R. E. (1984). The elaboration likelihood model of
persuasion. Advances in Consumer Research, 11(1).
Cialdini, R.B., (2009) Invloed: De zes geheimen van het overtuigen (5th ed.) Den Haag: SDU
uitgevers.
Cole, J. S., McCormick, A. C., & Gonyea, R. M. (2012, April). Respondent use of straight-lining as a response strategy in education survey research: Prevalence and implications.