Nguyễn Thị Thanh Thuý

1632300205
MKTG 462- Homework 3
Cite the article:
Tran, T. P. (2017). Personalized ads on Facebook: An effective marketing tool for online marketers. Journal of Retailing
and Consumer Services, 39, 230–242.doi:10.1016/j.jretconser.2017.06.010
The main purpose of the article is to examine the effects of perceived personalised ads on Facebook on customer attitude
and behaviral intention through ad avoidance, ad skepticism, and ad credibility
IV: perceived personalised on Facebook. Three dimensions are ad avoidance, ad skepticism, and ad credibility
DV: ad attitude
The second purpose is to use two set of data including validation data set for measurement model analysis and the
estimation data set for structural model analysis through online survey
Finally, Develop suitable market segments based on personal responses of Facebook personalized advertisements.
perceived ad ad ad avoidance
personalization skepticism credibility
To create a The is a refers to the efforts that consumers engage in in order to avoid being exposed
personalised ad disbelief customer's to commercial content.
matching with of assessment The extent of ad advoice is affected by
customer customers of the + demo-graphic features such as income and age
interests on towards authenticity +communication difficulty
Facebook ad and +media factors like general perception of media, or time exposed to a
based on their message believability medium
previous of an + perceived ad clutter, prior annoying experiences, and perceive goal
activities advertiseme hindrance
nt's content.
Result: -Ad sekpticism does not singificant effect on ad avoidance and ad attitude . The empirical findings indicate that
personalised advertising on Facebook and social networking sites have impacted customers' perceptions of ads significantly.
Perceived personalisation encourages positive reactions, such as ad credibility and attitude, and decreases negative reactions,
such as ad skepticism. To make the advertising individualized, this study used the viewer's previous internet activity.
-Three target segments consist of Ad haters (a high level of resistance to ads), Ad Lovers (a high level of favoritism), and Ad
Accommodators (stay in the middle) => e-retailers can adjust marketing tactics to improve the efficient and effective
advertising budget and promote the positive customer attitude.
What I learn from the article and my opinions
The finding illustrates that perceived personalization promotes the positive responses and reduces the negative respinses
towards ad. Therefore, perceived personalization is the underlying psychological mechanism of message effectiveness. I
wonder that whether a customer perceive a non-personalised message as personalized or to perceive a personalized message
as non-personalized? If it is possible, the measurement of perceived personlised process may not be correct, because the
starting point of a personalization process is individual preferences/interests, but not everyone knows what they really want,
they may be inclined to construct interest by situations => how people perceive preferences can vary based on how they are
elicited => what a marketer runs ads on social media as Facebook based on data collected from previous customers activities
on websites may not nescessary and message from that personalized process may not create favorable outcomes
In brief, from the article, I have two questions:
Firstly, Is it conceivable for someone to interpret a non-personalized message as personal, or a personalized message as
non-personalized?
Secondly, what elements in a message may be used most effectively for personalization?
 -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services

journal homepage: www.elsevier.com/locate/jretconser

Personalized ads on Facebook: An effective marketing tool for online 0$5.

marketers
Trang P. Tran
College of Business, East Carolina University, 3rd Floor Bate Building, Mail Stop 503, Greenville, NC 27858, USA

A R T I C L E I N F O A B S T R A C T

Keywords: A growing number of researchers have examined the effects of personalized advertising in traditional media,
Personalization however, little has been known about personalized advertising on Facebook. The primary objectives of this
Cluster analysis research are threefold: (1) Develop a comprehensive model that captures the effects of perceived personalized
PLS-SEM ads on Facebook on customer attitudinal and behavioral reactions (ad credibility, ad avoidance, ad skepticism,
Social media
ad attitude, and behavioral intention) to the ad; (2) Test hypothesized relationships using two data sets collected
through an online survey; and (3) Develop appropriate customer segments based on personal views of perso-
nalized ads on Facebook. The paper reveals that eleven out of thirteen hypotheses are supported and that three
market segments are identified including Ad Lovers, Ad Accommodators, and Ad Haters. The paper concludes
with conclusion and discussions highlighting managerial and research implications.

1. Introduction Facebook to deliver more relevant, more personalized experiences,”

said Jay Henderson, a director of strategy for IBM commerce (Finley,
Advertising strategies have been changed drastically as a result of 2015). These “personalized experiences” are associated with a number
development of social media. Online advertisers have used social media of benefits, for instance, reducing customer's resistance against ad (Baek
(i.e. Facebook) to connect customers with companies, developing new and Morimoto, 2012), enhancing ad credibility (Xu, 2006) and im-
opportunities for customers know about their brands and products proving brand awareness (Johns and Perrott, 2008).
(Comscore Media, 2009). To draw customers’ attention, online mar- Although researchers have attempted to investigate the impact of
keters have employed social networking sites to customize ads more personalized ads in traditional media (Baek and Morimoto, 2012; Yu
appealing to customers. That is reasonable to understand why more and Cude, 2009), website (Awad and Krishnan, 2006; Bleier and
companies are employing social media as marketing tools and why Eisenbeiss, 2015; Ho and Bodoff, 2014), or mobile (Kim and Han, 2014;
social networking sites like Facebook are preferred to non-virtual media Xu, 2006), little has been made to examine the effects of personalized
in creating ads (Waters et al., 2011). ads on Facebook (Keyzer et al., 2015; Tucker, 2014). Given the fact that
Statistical figures show that Facebook is an ideal platform for ad- Facebook has grown to be the most popular social media and that
vertisement. Facebook has 1.39 billion active users visiting the website personalization has been increasingly utilized as an advertising strategy
per month and 92% of social marketing companies have chosen that makes an ad more relevant to the users, the impact of personalized
Facebook as a marketing tool, that help Facebook reach the level of advertising on Facebook is worth investigating (Taylor et al., 2011).
revenue of 12.7 billion in 2014 (Rudolph, 2015). Additionally, many The current research is developed to fill the research gap. The primary
companies or service providers, either small or large, choose to create objectives of this research are threefold: (1) Develop a comprehensive
online ad on Facebook because it is less expensive than other media. model that captures the effects of perceived personalized ads on Face-
Online advertising has moved to a new phase after IBM teamed up book on customer attitudinal and behavioral reactions (ad credibility,
with Facebook on May 6, 2015 in attempt to create better advertisings ad avoidance, ad skepticism, ad attitude, and behavioral intention) to
on the world's largest social network by incorporating Facebook's tar- the ad; (2) Test hypothesized relationships using two data sets collected
geting technology into IBM's services for marketers (Finley, 2015). For through an online survey; and (3) Develop appropriate customer seg-
example, a retailer can develop an ad of a product that a customer ments based on personal views of personalized ads on Facebook.
previously views on its websites and place it to the customer's Facebook The research is organized as follows. First, related literature on
feed. Through this cooperation, both companies hope that the adver- personalized advertising is elaborated which is then followed by hy-
tising campaigns more effective. “These new capabilities will allow potheses development. Methodology is then presented with formative

E-mail address: [email protected].

http://dx.doi.org/10.1016/j.jretconser.2017.06.010
Received 16 March 2017; Received in revised form 12 June 2017; Accepted 20 June 2017
‹(OVHYLHU/WG$OOULJKWVUHVHUYHG
T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

and reflective measurement model and structural model, mediation advertising media. Early research indicates that personalized mailing
tests, and cluster analysis. The paper ends with discussions and con- does not generate better response rate (Weilbacher and Walsh, 1952).
clusion. More recent research shows that customers feel unpleasant when re-
ceiving an advertisement that is not targeted specifically to them
2. Theoretical background (Pavlou and Stewart, 2000). Customers are only interested in receiving
messages that are related to purchase decision or desired responses.
2.1. Perceived personalization Personalized advertising through mobile phones does not have a posi-
tive effect either if the message is sent to customers without permission
The concept of personalization emerged as early as late nineteenth (Tsang et al., 2004) because they feel their privacy is possibly threa-
century (Ross, 1992) and usually referred to segmentation and tar- tened. Therefore, personalized advertising makes customers unhappy,
geting and profiling while other researchers use this term in the context leading to negative responses and feeling to advertising, such as ad-
of mass customization, or one-on-one marketing (Petrison et al., 1997). vertising avoidance, or skepticism.
Broader meaning of personalization has been employed in practice that
includes, but is not limited to, tailoring the product, tailoring content of 3. Hypotheses development
message, or location diagnosis (Wind and Rangaswamy, 2001).
Personalization can be understood to be an activity of developing 3.1. The effects of perceived personalization
individualized communication to a particular customer which is tai-
lored based on the customer's implied or stated interests (Roberts and Drawing on the foundation of perceived utility (i.e., benefits or re-
Zahay, 2012). Imhoff et al. (2001) conceptualize personalization as a wards) of targeted advertising (Weilbacher and Walsh, 1952), perso-
firm's capability of identifying and treating customers on the individual nalized advertising is a strategic approach to optimizing advertising
basis through customized individual messages (i.e., special incentives, messages through matching with customer characteristics and interests.
target banner ads). Alternatively, personalization is perceived as a A personalized advertisement well-tailored to a customers’ need pro-
firm's practice of utilizing customers’ information and delivering tar- vides useful information, and therefore affects the way customers re-
geted solutions to those customers (Peppers and Rogers, 1998). spond when they are exposed to the ad – the response that is measured
Although several definitions are proposed in seminal literature, the by ad avoidance, ad skepticism, ad attitude and ad credibility.
core value of personalization is in common – personalization is the Ad avoidance refer to activities in which users are involved to
process (Vesanen and Raulas, 2006). Kotler et al. (2001) view perso- prevent their exposure to the advertisement content (Speck and Elliott,
nalization as a process of targeting, segmentation, and positioning. 1997a). In Facebook, ad avoidance can be measured in three levels:
Peppers and Rogers (1997) consider personalization as a way to obtain affection (i.e., I hate the ad), cognition (i.e., I ignore the ad), and be-
personal information of customers and tailor favorable solutions. Per- havior (i.e., I install AdBlock to block the ad). Ad skepticism, on the
sonalization is part of a company's customer relationship management other hand, is perceived as a customer's predisposition to disbelieve the
in which the company collects customer data through an interactive messages conveyed by an ad (Obermiller and Spangenberg, 1998). Ad
process between customers and the company and the process of col- skepticism plays a role in developing a customer's tendency to resist the
lecting data is referred to as learning relationship (Peppers et al., 1999). advertisement and to be suspicious about persuasive stimuli (Knowles
Personalization can be applied in the online setting and is integrated and Linn, 2004). Both ad avoidance and ad skepticism capture negative
in the system to help website interfaces more appealing. Data mining is responses that viewers have about an ad. However, since personalized
used to collect data on the website which then is analyzed to make the ads are created on the basis of the viewers’ prior interaction on Internet,
web more personalized (Pierrakos et al., 2003). Adomavicius and the viewers will have a lower degree of ad avoidance and ad skepticism
Tuzhilin (2005) interpret personalized as a Understanding-Delivering- when being exposed to those ads than when being exposed to general
Measuring cycle. This concept is consistent with Murthi and Sarkar's ads (Obermiller and Spangenberg, 1998; Pasadeos, 1990).
(2003) 3 phase conceptualization of personalization: (1) learn what On the other hand, ad credibility captures a positive view of an ad.
customers need, (2) match what customers need to what is offered, and If an ad is perceived to be credible, it is truthful and believable
(3) evaluate the two previous phases. (MacKenzie and Lutz, 1989). As explained earlier, a personalized ad is
Personalization is also viewed as an integrated marketing commu- tailored according to a viewer's interactive activity, the viewer would
nication that is derived from two way communication system, ability to trust the ad more. Stated differently, personalized ads improve ad
control responses of each party in communication process, database credibility (Kim and Han, 2014) and enhance customer's positive ad
application and personalized communication relationship (Peltier et al., attitude (Xu, 2006) (see Fig. 1). Formally,
2003). An integration of the four components plays a role in synthe-
sizing key elements necessary for personalized marketing and the whole H1. Ad avoidance is negatively affected by perceived personalization
process of personalization. Drawing on the concepts of personalization, on Facebook.
this research defines personalized advertising on Facebook as “the H2. Ad skepticism is negatively affected by perceived personalization
process of advertising in which a retailer develops a customized ad of a on Facebook.
product or service on Facebook based on prior customer activities on the
Internet.” H3. Ad attitude is positively affected by perceived personalization on
Both consumers and marketers are beneficiaries of personalized Facebook.
advertising. From a customer's perspective, personalized advertising H4. Ad credibility is positively affected by perceived personalization on
helps customers concentrate on the product that they want as a result of Facebook.
matching between customers’ preferences and the product advertised
(Bleier and Eisenbeiss, 2015). So the customers do not waste time
searching through a large assortment of products as they do in a regular 3.2. The effects of ad skepticism
searching (Srinivasan et al., 2002). Compared to traditional marketing,
personalized advertising is a cost-saving solution. This advertising is Ad skepticism is a customer's predisposition to disbelieve the mes-
sent to selected customers through highly customized commercial sages conveyed by an ad (Obermiller and Spangenberg, 1998). There-
messages (Kim et al., 2001). fore, people are prone to be wary and distrustful when being exposed to
Besides the advantages provided by personalized advertising, ne- a proposal, a message or ad advertisement. A key characteristic of ad
gative responses to this type of advertising is also reported across skepticism is its stability and its profound impact on customers’


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Fig. 1. Conceptual Model. Note: PER: Perceived

Personalized Ads, CRE: Ad Credibility, AVOID: Ad
Avoid, SKEP: Ad Skepticism, ATTD: Ad Attitude,
INT: Behavioral Intention.

responses to advertisement. Growing empirical evidence has shown 3.3. The effects of ad credibility
that customer skepticism is a driving factor of reactance responses to
persuasive messages in an advertisement (Obermiller and Spangenberg, According to advertising literature, ad credibility is perceived as the
1998). ‘‘extent to which the consumer perceives claims made about the brand
Simonson (2005) emphasize that no matter which media the com- in the ad to be truthful and believable’’ (MacKenzie and Lutz, 1989)
panies use to advertise, advertising has the same purpose: to persuade while Pavlou and Stewart (2000) define ad credibility as "predictability
customers or to sell products. Customers become skeptical after they and fulfillment of implicit and explicit requirements of an agreement".
realize the companies use marketing tactics trying to manipulate them. In other words, ad credibility is a customer's evaluation of truthfulness
Similarly, Knowles and Linn (2004) claim that a customer's distrust of a and believability of the content of an ad. Credibility is perceived as one
message derived from skepticism likely results in a high level of re- of the most critical components of a persuasive message that changes
sistance. So the customer becomes guarded and wary when he is ex- the outcomes of the message (Perloff, 1993; Choi and Rifon, 2002). Any
posed to a personalized message that makes him to be engaged in a message that is not credible is often avoided or ignored by the viewers.
biased processing. In a recent study investigating teenagers’ use of so- Advertisers who believe how ad credibility plays a role in changing an
cial networking sites, Kelly et al. (2010) find out that four key ante- attitude toward a brand want to make sure that the ad sent out to the
cedents affecting teenagers’ tendency to avoid ads: (1) whether the market is credible. If an ad is more credible, readers are more likely to
users expect any negative experience, (2) whether the ad is relevant and be convinced that the claims in the ad are truthful (Cotte et al., 2005).
(3) how much skeptical the users are toward the ad message and (4) The role of ad credibility in improving message persuasiveness is
how much skeptical the users are toward the ad medium. In the per- stressed by the hierarchy of effect models (Thorson et al., 1992).
sonalized advertising context on Facebook, a customer's reaction to an The effects of ad credibility on ad outcomes (i.e., attitude toward a
ad is not different. The customer understands an ad customized based brand, or purchase intention) is well documented (Deshpandé and
on his or her preferences as an attempt to manipulate, hence he be- Stayman, 1994; Goldberg and Hartwick, 1990; Kavanoor et al., 1997;
comes more skeptical. Ohanian, 1990). Choi, Marina and Rifon (2002) investigate the drivers
When a customer is skeptical, he is less interested in the ad, and less and outcomes of online ad credibility using structural equation mod-
reliant on the ad, and therefore he is more inclined to avoid the ad- elling. The paper concludes that credibility perceptions enhance ad
vertisement, such as, zapping or zipping through television commer- attitude and brand attitude that result in purchase intention. The results
cials (Speck and Elliott, 1997a). In light of persuasion knowledge, show that although website credibility does not have direct impact on
skeptical customers interpret advertising customized to individual in- brand attitude and purchase intention, advertiser credibility does – the
terests or lifestyles as an advertiser's effort to manipulate and persuade finding that is in line with the previous research conducted in tradi-
customer. As such, the skeptical customers generally do not like the ad tional media (Goldsmith et al., 2000). Formally,
and find a way to avoid it. They do so because customer's skepticism
H7. Ad avoidance is negatively affected by ad credibility on Facebook.
toward personalized ads is associated with general distrust of ad tactics
designed by marketers that encompass customer names and personal H8. Ad skepticism is negatively affected by ad credibility on Facebook.
preferences. Therefore,
H9. Ad attitude is positively affected by ad credibility on Facebook.
H5. Ad avoidance is positively affected by ad skepticism on Facebook.
H6. Ad attitude is negatively affected by ad skepticism on Facebook. 3.4. The effect of ad avoidance

Ad avoidance is a key research area attracting attentions of


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

marketers and advertisers because it is perceived as one of the biggest and ad attitude
obstacles for marketers. Ad avoidance can be understood as activities in
H12. Ad credibility and ad avoidance mediate the relationship between
which users are involved to prevent their exposure to the advertisement
perceived personalized ads on Facebook and ad attitude
content (Speck and Elliott, 1997a). Prior research investigates drivers
and effects of ad avoidance in such traditional media as radio, televi- H13. Ad credibility and ad skepticism mediate the relationship between
sion, newspapers and magazine, and suggests two ways that viewers perceived personalized ads on Facebook and ad attitude
could use to avoid a television commercial. The first is flipping that is
similar to the way of flipping pages of newspapers or magazine, and the
second is zapping – the same way that people are zapping radio ads
5. Methodology
(Heeter and Greenberg, 1985).
Key factors affecting the extent of ad avoidance include demo-
5.1. Pretest
graphic features (i.e., income and age), communication difficulty (i.e.,
search challenge), or media factors (i.e., general perception of media, or
The objective of the pre-test was to see whether or not participants
time exposed to a medium). More recent literature has identified in-
perceived an online ad to be personalized. Before the survey was ad-
ternet-related factors influencing viewers’ tendency to avoid online
ministered, participants were given instructions of the survey to ensure
advertisements that include, but are not limited to, perceived ad clutter,
that only those who were qualified (i.e., having a Facebook account)
prior annoying experiences, and perceive goal hindrance (Cho and
could proceed to the questionnaire. Next, the definition of personalized
Cheon, 2004). Ad avoidance is made up by three components: affection,
advertising was provided. Then, participants were asked to report
cognition and behavior (Cho and Cheon, 2004). Affection component of
whether they saw a personalized advertisement on their Facebook ac-
ad avoidance refers to one's personal emotional feeling about the ad.
count (“Yes” or “No”). Participants who answered “No” to this question
Cognition component of ad avoidance means beliefs that a person has
were excluded from the analyses. Those who answered “Yes” were
about the object. And behavior component of ad avoidance represent
asked to provide a brand name that they saw on Facebook and describe
actions to avoid the ad. When a viewer is inclined to avoid or be
briefly about the situation where they were involved in an activity
skeptical against a personalized ad, the view does not have a favorable
online before they saw that ad. They were then asked to write what they
attitude toward the ad. Therefore,
thought about the ad that they just saw. Finally, participants indicated
H10. Ad attitude is negatively affected by ad avoidance on Facebook. the extent to which they were thinking about the ad on Facebook
(1=Personalized, and 7=General). This scale was used as a manip-
ulation check. Two hundred and eighty responses that fulfilled re-
3.5. The effects of ad attitude
quirements were collected. One sample t-test created to compare the
mean of the sample and the midpoint (4) between personalized and
The impact of personalized ads on customers’ attitudes toward ads
general advertisement confirmed that participants thought the ad they
has been explored in the mobile and website context (Xu, 2006). Em-
encountered on Facebook was personalized (M = 3.12, SD = 2.17, t
pirical studies show that personalized ads could improve the effec-
(df=279) = −6.81, p < 0.001). Therefore, our results show that the
tiveness of internet advertising (Kim et al., 2001). Through persona-
manipulation check for perceived personalization worked.
lized ads, customers are able to find more appropriate ads that are
specifically tailored to their preferences or interests. Key advantages
associated with personalized ads include favorable attitudes toward
5.2. Main study
online ads and increased likelihood to purchase (Chakraborty et al.,
2003; Chen, 1999). In a social networking site like Facebook, it is an-
The primary objectives of the main study was to test hypothesized
ticipated that personalization would evoke positive attitudes toward the
relationships illustrated in the conceptual model. The beginning of this
ads which, in turn, lead to purchase the products or items being ad-
study was similar to the pretest. Then respondents were assigned to one
vertised. That means, customers will be more likely to plan to act on the
of two scenarios with each being associated with a personalized ad-
informational claims they see on the ad, such as, checking on the
vertisement of one particular product (laptop versus shoes). A preamble
website, or viewing the product advertised, or making order online.
of laptop scenario was written as follows: “Imagine you were searching
Therefore
online to find a new laptop that meets your expectation in terms of brand,
H11. Purchase intention is positively affected by ad attitude on screen size, color, price and configuration. One day later when you opened
Facebook. your Facebook account, you recognized there was an advertisement of the
laptop's brand that you had searched.”
Pictures of the ads, copied directly from Facebook website, were
4. Mediation effects
displayed on the right hand side of the preamble. This was followed by
the questionnaire, demographic questions, and validation questions
As earlier discussed, a personalized ad is tailored according to a
(that are created to make sure the respondents read and answer the
viewer's geographic location, preferences or prior interactive activities,
questionnaire carefully). Out of the sample of 711 responses collected
the viewers believe that ad to be more trustworthy and more credible.
from Amazon Mechanical Turk (MTurk) were 613 completed observa-
Stated differently, personalized ads improve ad credibility (Kim and
tions that met requirements. Demographical information presented in
Han, 2014; MacKenzie and Lutz, 1989). Credibility plays a role in
Table 1 showed that more than half of the respondents were within 20
changing customer perception about the ad. A credible message is more
and 30 years old (54.01%), there were more female than males (female:
persuasive (Perloff, 1993; Choi and Rifon, 2002). Therefore, any mes-
55.32%, male: 44.68%), students accounted only for about one third of
sage that is not credible is often avoided or ignored by the viewers
the total sample, most of them spent from 1 to 3 h on Facebook a day
(Heeter and Greenberg, 1985; Speck and Elliott, 1997b; Thorson et al.,
(48.61%).
1992). Similarly, when an ad is not trustworthy, the viewers are highly
To avoid the use of the same set of data to validate scales and to test
skeptical and doubtful of the message in the ad (Obermiller and
hypotheses, the data set is randomly divided into two data sets: the first
Spangenberg, 1998). Therefore it is anticipated that ad credibility and
data set (called validation data set) to be used for measurement model
ad avoidance serve as mediators between perceived personalized ads
analysis (n = 206), and the second data set (called estimation data set)
and ad attitude. In a similar logic, it is expected that ad credibility and
for structural model analysis (n = 407).
ad skepticism serve as mediators between perceived personalized ads


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Table 1 rejected (or formative). Bonferroni-adjusted confidence intervals are

Demographic Characteristics (n = 613). used to test multiple hypotheses. If zero lies in the range of a confidence
interval, the tetrad vanishes since it is not significant from zero. The
Variables Percent
results, reported in Table 2, reveal three reflective constructs including
Age perceived personalization, ad attitude, and behavioral intention, and
19 or younger 2.45 three formative constructs including ad avoidance, ad skepticism, and
20–30 54.01
ad credibility.
31–40 26.68
41–50 10.15
More than 50 6.71 7. Formative measurement model
Gender
Male 44.68 In this measurement model, indicators of each formative construct,
Female 55.32
serving as possible independent drivers of the latent construct, should
Students vs. non-students
Students 32.24 not correlate with each other. This can be checked by multicollinearity
Non-students 67.76 test or by VIF (Diamantopoulos and Winklhofer, 2001). Maximum VIF
Time of using Facebook per day values for formative constructs (ad skepticism, ad credibility, and ad
Less than 1 h 26.68
avoidance) is 4.758 which is lower than the threshold of 5 (see Table 3).
1–3 h 48.61
3–5 h 17.84 Therefore it is confirmed that the problem of multicollinearity does not
5–7 h 4.58 exist.
More than 7 h 2.29 The next step is to test relative contribution (or significance) of each
Income indicator to the corresponding construct. Testing of relative contribu-
Less than $20,000 32.41
tion can be done by measuring whether the relationship is significant
$20,000-$39,999 28.48
$40,000-$59,999 16.69 through bootstrapping method using 5000 subsamples. The indicator
$60,000-$79,999 10.15 weights and significant levels are used for that purpose (see Table 3). In
$80,000-$99,000 4.91 this step, all indicators that have significant level (p < ; 0.05) are re-
More than $100,000 7.36
tained, while indicators that are not significant should be tested in the
Living place
USA 63.50
following phase (testing of absolute contribution).
India 33.72 Then testing of absolute contribution is applied for indicators with
Other 2.78 insignificant weights. According to Hair et al. (2013a, 2013b), only if a
formative indicator has insignificant weight and low loading, it can be
eliminated. Other nonsignificant formative indicators remain in the
5.2.1. Measurements measurement model due to significant correlation between those in-
All measurements were adapted from existing literature: ad avoid- dicators and latent variables (p < ; 0.05).
ance (Cho and Cheon, 2004; Speck and Elliott, 1997b), ad skepticism For example, a formative item of ad avoidance (avoid 4 “I have tried
(Obermiller and Spangenberg, 1998), perceived personalization to set up my account to avoid personalized advertising from Facebook”) has
(Srinivasan et al., 2002), ad credibility (Tsang et al., 2004), and be- insignificant weight (0.090, p > ; 0.05), but its loading on the latent
havioral intention and ad attitude (Taylor et al., 2011). The measure- construct is significant (the result is not reported in the paper). That
ments used 7-Likert scales with 1 being “Strongly disagree” and 7 means this item has insignificant relative contribution, but significant
“Strongly agree”. absolute contribution, therefore this item is retained. After the same
procedure is applied for all formative items, every item is retained in
6. Analytical results the measurement model.

6.1. PLS approach 7.1. Reflective measurement model

Structural equation analysis (SEM) is a popular statistical method Reflective constructs in the measurement model is assessed by re-
for multivariable data analysis. There are two types of SEM: covariance- liability and validity (Chin and Newsted, 1999; Hulland, 1999). Relia-
based (CB-SEM) and variance-based (partial least squares SEM or PLS- bility is evaluated using three criteria: factor loadings, composite re-
SEM) (Ringle et al., 2015). PLS-SEM is used to estimate path relation- liability and internal consistency reliability. The results revealed
ships on the basis of available data with an objective of minimizing the satisfactory factor loadings of all reflective items on their corresponding
error terms of endogenous variable. This method can work with a small latent constructs (with the minimum of 0.795), high reliability
sample size in a complex model. It is not bound by the normal data 0.932–0.963) and high internal consistency reliability 0.908–0.943) in
distribution assumption (Chin and Newsted, 1999; Haenlein and all latent constructs (see Table 3).
Kaplan, 2004). Therefore, PLS-SEM is selected for the current study. Two types of validity that is measured in PLS-SEM: convergent and
Two types of constructs are usually considered in PLS path analysis: discriminant. The former is represented by average variance explained
reflective and formative. To decide whether a construct is formatively (AVE). The AVE values of all latent variables are greater than 0.7,
or reflectively measured, a confirmatory tetrad analysis (CTA-PLS) was therefore convergent validity was established (Hair et al., 2010). The
used (Gudergan et al., 2008). Pursuing guidelines from literature latter is tested by comparing the AVE value of one construct and
(Coltman et al., 2008), an ex post test was run on the measurement squared Pearson's correlation coefficients between that construct and
model with CTA-PLS. The procedure first identifies the indicators of others (Fornell and Larcker, 1981). Discriminant validity was con-
measurement model, calculates the discrepancies between a products of firmed since the AVEs was greater than the squared correlation coef-
different pairs of covariance (or tetrad), and stops until it calculates the ficients (see Table 3). Since the requirements for testing reliability and
discrepancies for every pair of indicators. If there is no discrepancy, a validity are met, we move to structural model.
tetrad vanishes.
One assumption in the reflective measurement model is all model 7.2. Structural model
implied nonredundant tetrads vanish (Gudergan et al., 2008). However,
if at least one tetrad does not vanish, the null hypothesis that the re- The structural model is assessed using the estimation data set. The
flective direction of relationships exists in these measurement models is results of the test are discussed in the following section. First,


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Table 2
CTA-PLS Results.

Latent Valuables Model-Implied Residual Value Bootstrap CIa

Nonredundant t-Value
Vanishing Tetrad

Ad Skepticism (formative) τskep, 1234 0.210 2.190 [−0.195;1.114]

τskep, 1243 0.274 0.102 [−0.877;0.831]
τskep, 1235 0.214 3.453 [0.065;1.395]b
τskep, 1352 0.206 3.138 [−1.278;0.004]
τskep, 1362 0.241 3.032 [−1.476;0.028]
τskep, 1237 0.219 2.267 [−0.188;1.174]
τskep, 1238 0.235 3.342 [0.049;1.516]b
τskep, 1293 0.315 0.183 [−0.925;1.040]
τskep, 1254 0.262 1.271 [−0.489;1.146]
τskep, 1274 0.280 2.167 [−0.271;1.471]
τskep, 1284 0.308 2.198 [−0.285;1.635]
τskep, 1492 0.235 0.522 [−0.614;0.849]
τskep, 1285 0.315 2.039 [−0.348;1.617]
τskep, 1276 0.288 2.555 [−0.172;1.625]
τskep, 1268 0.278 2.392 [−0.204;1.525]
τskep, 1682 0.223 0.276 [−0.756;0.630]
τskep, 1782 0.168 0.446 [−0.597;0.450]
τskep, 1456 0.236 0.322 [−0.660;0.808]
τskep, 1475 0.173 0.629 [−0.436;0.644]
τskep, 1479 0.197 1.054 [−0.410;0.818]
τskep, 1498 0.250 0.163 [−0.822;0.735]
τskep, 1587 0.267 1.936 [−0.314;1.351]
τskep, 1579 0.243 1.269 [−0.453;1.062]
τskep, 1697 0.222 0.470 [−0.588;0.793]
τskep, 1896 0.297 0.905 [−0.660;1.189]
τskep, 1234 0.192 0.282 [−0.656;0.542]
τskep, 2684 0.248 1.205 [−0.469;1.079]
Personalized Ads (reflective) τper, 1234 0.424 0.823 [−1.441;0.742]
τper, 1243 0.270 1.380 [−0.327;1.062]
τper, 1235 0.410 0.356 [−1.213;0.899]
τper, 1352 0.389 0.972 [−0.617;1.386]
τper, 1345 0.406 1.869 [−0.286;1.806]
Ad Credibility (formative) τcred, 1237 0.233 1.642 [−0.900;0.143]b
τcred, 1273 0.243 2.864 [−1.233;−0.142]
Ad Avoidance (formative) τavoid, 1234 0.390 2.524 [−0.032;1.980]
τavoid, 1234 0.531 0.842 [−0.933;1.802]
τavoid, 1234 0.438 2.593 [−0.007;2.251]
τavoid, 1234 0.323 2.241 [−1.551;0.111]
τavoid, 1234 0.613 4.723 [1.280;4.442]b
Ad Attitude (reflective) τattd, 1234 0.296 2.202 [−1.312;0.017]
τattd, 1234 0.404 0.479 [−1.096;0.717]
Behavioral Intention (reflective) τint, 1235 0.196 0.626 [−0.561;0.318]
τint, 1253 0.355 0.627 [−0.571;1.019]

Note: (a) Adjustment of the 5% bias corrected bootstrap (two-tailed) confidence interval (CI) limits uses the Bonferroni method to account for multiple testing issues; (b) Mode-implied
nonredundant tetrads that do not vanish.

collinearity is assessed. All VIF values are below the threshold value of effect size of SKEP is zero (0.012). In explaining SKEP, the effect size of
5 (with maximum value of 4.535) indicating that collinearity does not CRE is large while that of PER is medium. For AVOID, all effect sizes is
reach critical level, and therefore it is not an issue for the estimation of zero while that of CRE is small.
the PLS path model. Second, the determination coefficients (R2) re- Then, Stone-Geisser's Q2 values (Geisser, 1975; Stone, 1974) is
present the model's predictive accuracy which is measured by the employed to measure the model's predictive relevance. In particular,
amount of explained variance of the endogenous constructs in the when the model's predictive relevance is established, the model pre-
structural model. As a rule of thumb, there are three values of R2: 75%, cisely predicts the data points of indicators in reflective measurement
50%, or 25% representing substantial, moderate, or weak level, re- models of endogenous constructs. In the structural model, Q2 values
spectively (Hair et al., 2013a, 2013b; Henseler et al., 2009). The R2 lager than zero for a certain reflective endogenous latent variable show
values of all endogenous constructs are greater than 75% except those that the path model's predictive relevance for this construct. Using the
of CRED (59.1%) and AVOID (31.3%). As such, the model's predictive blinding procedure for assessing the predictive relevance of the path
accuracy on ATTD, and SKEP substantial while that on CRE is medium model, the current research indicates that all Q2 values for ATTD,
and that on AVOID is weak. AVOID, INT, and SKEP are larger than zero, therefore, the path model's
Third, the f2 effect size is employed to analyze the relevance of predictive relevance for those constructs is established. As f2 was used
constructs in explaining selected endogenous latent constructs. The to measure effect size for R2, q2 was used to compute effect size for Q2.
method previously used by existing research has been used in this re- As a rule of thumb, q2 values of 0.02, 0.15, and 0.35 respectively in-
search to test the effect size (Cohen, 1992; Schroer and Hertel, 2009). dicate that an exogenous construct has small, medium, or large pre-
Effect size could be small, medium, and large depending on whether f2 dictive relevance for a certain endogenous construct (see Table 4).
values are 0.02, 0.15, and 0.35, respectively. The results in Table 4 Finally, a bootstrapping resampling procedure is implemented to
show that in explaining ATTD, the effect size of CRE is large (0.435), test whether path coefficients estimated in the structural model are
while those of AVOID (0.070) and PER (0.042) are small while the significant. To do so, we select 5000 bootstrapping samples and assess


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Table 3
Measurement Model Evaluation.

Items VIF CR AVE α AVE > Corr2 Loadings/ Mean SD t-value

Weights

Ad avoidance (formative)
1. I intentionally ignore any personalized advertising on Facebook. 2.263 n.a. n.a. n.a. n.a. 0.248 0.241 0.134 1.847
2. I hate any personalized advertising on Facebook. 3.068 0.478 0.448 0.161 2.975
3. It would be better if there were no personalized advertising on Facebook. 2.097 0.443 0.460 0.140 3.164
4. I have tried to set up my account to avoid personalized advertising from 2.859 0.120 0.118 0.159 0.756
Facebook.
5. I have sought assistance to prevent me from receiving personalized advertising 2.399 −0.612 −0.614 0.149 4.097
from Facebook.

Ad skepticism (formative)
1. I can depend on getting the truth in most personalized advertising on Facebook. 2.831 n.a. n.a. n.a. n.a. 0.071 0.069 0.080 0.895
(R)
2. Personalized advertising's aim is to inform the consumer. (R) 2.897 −0.201 −0.199 0.097 2.070
3. I believe personalized advertising on Facebook is informative. (R) 3.922 0.364 0.365 0.099 3.679
4. Personalized advertising on Facebook is generally truthful. (R) 4.213 0.090 0.086 0.093 0.969
5. Personalized advertising on Facebook is a reliable source of information about 4.160 0.019 0.023 0.090 0.208
the quality and performance of products. (R)
6. Personalized advertising on Facebook is truth well told. (R) 4.246 0.215 0.210 0.109 1.977
7. In general, personalized advertising on Facebook presents a true picture of the 3.257 −0.093 −0.092 0.087 1.076
product being advertised. (R)
8. I feel I have been accurately informed after viewing most personalized 4.758 0.348 0.341 0.107 3.261
advertising on Facebook. (R)
9. Most personalized advertising on Facebook provides consumers with essential 2.868 0.269 0.271 0.076 3.522
information. (R)

Perceived Personalization (reflective) 0.932 0.733 0.908 0.733 > 0.653

1. This personalized advertising on Facebook makes purchase 0.795 0.792 0.038 21.077
recommendations that match my needs.
2. I think that this personalized advertising on Facebook enables me to order 0.887 0.887 0.018 49.055
products that are tailor-made for me.
3. Overall, this personalized advertising on Facebook is tailored to my 0.873 0.870 0.029 30.490
situation.
4. This personalized advertising on Facebook makes me feel that I am a 0.820 0.819 0.027 30.184
unique customer.
5. I believe that this personalized advertising on Facebook is customized to 0.900 0.900 0.016 54.586
my needs.

Ad Credibility (formative) n.a. n.a. n.a. n.a.

1. I use personalized advertising on Facebook as a reference for purchasing 2.349 0.382 0.386 0.080 4.783
2. I trust personalized advertising on Facebook 3.594 0.414 0.411 0.078 5.319
3. The content provided by personalized advertising on Facebook is credible 3.447 0.296 0.292 0.084 3.532

Ad Attitude (reflective) 0.957 0.882 0.933 0.882 > 0.716

1. I like the idea of using personalized advertising on Facebook 0.939 0.939 0.011 88.708
2. Using personalized advertising on Facebook is a good idea 0.947 0.947 0.010 92.275
3. Using personalized advertising on Facebook is a wise idea 0.932 0.931 0.014 66.981

Behavioral Intention (reflective) 0.963 0.898 0.943 0.898 > 0.741

1. I will use personalized advertising on Facebook to consume whenever I have a 0.929 0.929 0.013 70.832
chance
2. I intend to use personalized advertising on Facebook for shopping after I receive 0.955 0.955 0.007 134.618
it
3. I expect to use personalized advertising on Facebook to purchase after receiving it 0.958 0.957 0.007 142.617

Note: The seven point response cues for each item are 1 = "strongly disagree" to 7 = "strongly agree"; (R) indicates reversed coded item.

standard error estimates using t-tests (Chin, 1998; Sellin and Keeves, of dependent variables, which are reported in Table 5. Path coefficients
1997; Hulland, 1999). Then the inner model (structural model) eva- and their significant values were used to test the hypotheses. The results
luation of PLS explains the path estimates of the independent variables show that perceived personalized ads do not have significant impact on

Table 4
Effect Sizes.

R2 Q2 ATTD AVOID SKEP

Path f2 Effect Size q2 Effect Size Path f2 Effect Size q2 Effect Size Path f2 Effect Size q2 Effect Size
Coefficients Coefficients Coefficients

ATTD 0.811 0.738

AVOID 0.313 0.105 −0.139 0.070 0.042
CRE 0.591 0.513 0.600 0.435 0.294 −0.463 0.076 0.019 −0.599 0.665 0.247
PER 0.154 0.042 0.027 −0.102 0.005 0.001 −0.335 0.208 0.094
SKEP 0.779 0.583 −0.101 0.012 0.000 0.017 0.000 0.000


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Table 5
Hypotheses Testing.

Hypotheses Coefficient Sample Mean Standard Error t-Values p-Values

PER - > AVOID H1: not supported −0.102 −0.090 0.076 1.344 0.179
PER - > SKEP H2: supported −0.335 −0.332 0.049 6.794 0.000
PER - > ATTD H3: supported 0.154 0.152 0.047 3.241 0.001
PER - > CRE H4: supported 0.769 0.770 0.026 29.984 0.000
SKEP - > AVOID H5: not supported 0.017 0.047 0.098 0.169 0.866
SKEP - > ATTD H6: not supported −0.101 −0.109 0.058 1.748 0.081
CRE - > AVOID H7: supported −0.463 −0.454 0.088 5.274 0.000
CRE - > SKEP H8: supported −0.599 −0.603 0.048 12.552 0.000
CRE - > ATTD H9: supported 0.600 0.593 0.054 11.147 0.000
AVOID - > ATTD H10: supported −0.139 −0.140 0.030 4.583 0.000
ATTD - > INT H11: supported 0.876 0.876 0.014 60.873 0.000
PER - > CRE - > AVOID - > ATTD H12: supported –
PER - > CRE - > SKEP - > ATTD H13: supported –

ad avoidance (β = −0.102, p > ; 0.05) (that means H1 is not sup- The third indirect effect, labeled "Ind3" in the output, is the indirect
ported), but have negative significant impact on ad skepticism (β = effect of personalized ads on Facebook on ad attitude through ad
−0.335, p < ; 0.05) (or H2 is supported), positive significant impact on avoidance (PER - > AVOID - > ATTD). Estimated as the product of the
ad attitude (β = 0.154, p < ; 0.05) (or H3 is supported), positive sig- effect of personalized ad on ad avoidance and the effect of ad avoidance
nificant impact on ad credibility (β = 0.769, p < ; 0.05) (or H4 is on ad attitude, this indirect effect (a2b2) is 0.012. However, since zero
supported) lies in the range of CI (−0.011 to 0.040), this effect is not significant.
However, different from what is expected, ad skepticism do not The total indirect effect in the serial multiple mediator model, de-
have significant impact on ad avoidance (β = 0.017, p > ; 0.05) (or H5 fined as the sum of all the specific indirect effects, is 0.672 which is
is not supported), and ad attitude (β = −0.101, p > ; 0.05) (or H6 is significant because CI does include zero 0.600–0.768). Therefore the
not supported). On the other hands, ad credibility has significant im- results lend evidence that ad credibility and ad avoidance serve as
pact on ad avoidance (β = −0.463, p < ; 0.05) (or H7 is supported), ad mediators of the effects of personalized ads on ad attitude.
skepticism (β = −0.599, p < ; 0.05) (or H8 is supported), and ad at- The contrast option of PROCESS calculated all positive pairwise
titude (β = 0.600, p < ; 0.05) (or H9 is supported). Ad avoidance has comparisons between specific indirect effects. These comparisons are
significant impact on ad attitude (β = −0.139, p < ; 0.05) (or H10 is found in the PROCESS output as indicated as (C1), (C2), and (C3). As
supported). Finally, ad attitude has positive significant impact on be- can be seen, the effect of personalized ad on ad attitude through ad
havioral intention (β =0.876, p < ; 0.05) (or H11 is supported) credibility (Ind1) is statistically different from the effect of personalized
ad on ad attitude through ad credibility and ad avoidance (Ind2) since
the CI does not include zero 0.517–0.698).
8. Mediation tests
This concludes that personalized ad has a larger impact on ad atti-
tude through ad credibility than it does through its effect on ad cred-
8.1. Ad credibility and avoidance: mediating the link between personalized
ibility, which influences ad avoidance. A similar observation is seen in
ads and ad attitude
(C2) that tests whether personalized ad has a larger effect on ad attitude
through ad credibility than it does through its effect on ad avoidance
Mediation effect was tested using PROCESS by (Hayes, 2013) with
which influences ad attitude. The results lend evidence of support of
application of the serial multiple mediator model. The model consists of
this statement. However, the confidence intervals for the last contrast
four indirect effects as products of regression coefficients connecting
(C3) include zero, meaning this indirect effect is not statistically dif-
between PER and ATTD. The procedure was implemented with 95%
ferent.
bias-corrected bootstraps confidence intervals based on 10,000 boot-
strap samples. The results were reported in Table 6.
The first indirect effect, labeled "Ind1" in the output, is the specific 8.2. Ad credibility and skepticism: mediating the link between personalized
indirect effect of personalized ads on Facebook on ad attitude through ads and ad attitude
ad credibility (PER - > CRE - > ATTD), estimated as a1b1 = 0.630.
Since both upper and lower limit of the confidence interval (CI) are A similar procedure was applied to tested mediation effects in the
above zero 0.556–0.724), this indirect effect can be interpreted as relationship PER - CRED - SKEP – ATTD. Among three indirect effects,
significantly positive. That means, those who perceived a Facebook ad only is the first indirect effect (PER - > CRED - > ATTD) labeled “Ind1”
to be personalized believed that the ad was more credible, and this significant while the other two effects are not. Total effect is 0.675 and
increased credibility enhanced the customer's positive attitude toward significant because the CI does not include zero 0.580–0.778). Overall,
the ad. the total indirect effect in the serial multiple mediator model is sig-
The second indirect effect, labeled "Ind2" in the output, is the spe- nificantly different from zero. Therefore the results lend support that ad
cific indirect effect of personalized ads on Facebook on ad attitude credibility and ad skepticism serve as mediators of the effects of per-
through ad credibility and ad avoidance in serial, with ad credibility sonalized ads on ad attitude (see Table 7).
affecting ad avoidance, which in turn impact ad attitude (PER - > CRE
- > AVOID - > ATTD). Estimated as a1d21b1 = 0.030, the effect is 9. Cluster analysis
significant because zero does not belong to CI 0.011–0.056). Relative to
those who viewed a Facebook ad, those who perceived a Facebook ad to After the model was tested using the estimation sample, two sam-
be more personalized believed that the ad was more credible, and this ples were combined to identify the market segments on the basis of four
increased credibility (as a1 is positive), which in turn reduced the exogenous variables in the model that included ad avoidance, ad
customer's tendency to avoid the ad (as d21 is negative), and the de- skepticism, personalized ad, and ad credibility. A two-step cluster
creased ad avoidance enhanced positive attitude toward ad (because b2 analysis was employed to carry out the task and mean scores from all
is negative). respondents of the combined sample served as the input for the analysis


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Table 6
Mediation Testing - Part A.

Mediation Test: PER - CRE - AVOID - ATTD

Consequence

M1 (CRED) M2 (AVOID) Y (ATTD)

Antecendent Coeff. SE p Coeff. SE p Coeff. SE p

X (PER) a1 0.861 0.036 0.000 a2 −0.075 0.070 0.281 c′ 0.228 0.040 0.000
M1 (CRED) – – – d21 −0.211 0.062 0.001 b1 0.732 0.036 0.000
M2 (AVOID) – – – – – – b2 −0.164 0.029 0.000
constant iM1 0.028 0.150 0.853 iM2 5.152 0.188 0.000 iY 0.861 0.183 0.000
R2 = 0.588 R2 = 0.099 R2 = 0.810
F(1,405) = 578.913, p < ; 0.001 F(2,404) = 22.290, p < ; 0.001 F(3,403) = 570.832, p < ; 0.001

Total, Direct, Indirect Effects

Total effect of X on Y
Effect SE t p LLCI ULCI
0.899 0.038 23.504 0.000 0.824 0.975
Direct effect of X on Y
Effect SE t p LLCI ULCI
0.228 0.040 5.664 0.000 0.149 0.307
Indirect effect(s) of X on Y
Effect Boot SE BootLLCI BootULCI
Total: 0.672 0.042 0.600 0.768
Ind1 0.630 0.043 0.556 0.724
Ind2 0.030 0.012 0.011 0.056
Ind3 0.012 0.013 −0.011 0.040
(C1) 0.600 0.045 0.517 0.698
(C2) 0.617 0.046 0.535 0.716
(C3) 0.017 0.021 −0.020 0.062
Indirect effect key
Ind1 per - > cred - > attd
Ind2 per - > cred - > avoid - > attd
Ind3 per - > avoid - > attd
Specific indirect effect contrast definitions
(C1) Ind1 minus Ind2
(C2) Ind1 minus Ind3
(C3) Ind2 minus Ind3

(Punj and Stewart, 1983). Ad Accommodators (39.8%). This segment is relatively large as op-
In the first step, Ward's hierarchical clustering with square posed to two other segments. In terms of characteristics, they are
Euclidean distance was implemented as a preliminary solution to find a standing in the middle between Ad Haters and Ad Lovers in all variables
number of clusters. This procedure was usually used to develop being examined. Particularly, they do not hate the ads as much as do Ad
homogenous clusters with relatively similar size (Hair et al., 2010). Haters, and do not love ads as much as do Ad Lovers.
Following the procedure, coefficients in the Agglomeration Schedule
were used as criteria to determine a cluster solution. The results showed 10. Cluster profiling
that the coefficients increased substantially from a 3 cluster solution to
a 2 cluster solution. This indicated that the 3 cluster solution was The first validation test was performed with respect to two depen-
considered appropriate. dent variables of the model (ad attitude, and behavior intention). The
The second step was completed through non-hierarchical clustering results revealed that the three clusters differ significantly in these two
procedure (K mean) with the solution of 3 clusters that was determined variables. Particularly, Chi-square test showed that three clusters were
in the first step. Three clusters were named Ad Haters (or Ad Hating significantly different in ad attitude (χ2(2) = 811.89, p < ; 0.01) and in
Segment), Ad Accommodators (or Ad Accommodating Segment), and behavioral intention (χ2(2) = 674.90, p < ; 0.01) with Ad Lover having
Ad Lovers (or Ad Loving Segment). For the sake of brevity, Ad Haters, the highest scores and Ad Haters having the lowest scores.
Ad Accommodators, and Ad Lovers are used. The results were showed There was no significant difference among three clusters for age
in Table 8. Chi-square tests confirmed that those variables were sig- (χ2(2) = 1.35, p > ; 0.05), but significant difference in time on Facebook
nificantly different across three clusters (see Table 8). Pairwise com- (χ2(2) = 12.66, p < ; 0.01), and income (χ2(2) = 10.45, p < ; 0.01).
parison with Bonferroni post hoc procedure each pair of three clusters Pairwise comparison indicated that time spent on Facebook, however,
were significantly different. was not significant different between Ad Haters and Accommodators,
Ad Haters (29.5%). Consumers in this segment demonstrate a high neither was income between Ad Haters and Accommodators.
level of resistance to ads (including ad avoidance and skepticism) and Within cluster analysis revealed that the differences among three
do not think ads to provide much value as do customers in other two clusters could be seen in gender. Among Ad Haters, female customers
segments perceive (exemplified though low score in both personalized were more than twice male customers (Female: 70.72%, Male: 29.28%)
ads and ad credibility). while there were more male than female in the Ad Lovers segment
Ad Lovers (30.7%). Unlike Ad Haters, consumers in this segment (Female: 42.25%, Male: 57.75%). Breakdown between students and
demonstrate a high level of favoritism toward the ad (including per- non-students was evidenced among three segments. Non-students ac-
sonalized ad and ad credibility) and a low level of resistance against ads counted for majority numbers across three clusters.
(indicated by low scores in both ad avoidance, and ad skepticism). Put all together, it can be seen that compared to Ad Haters, Ad


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Table 7
Mediation Testing - Part B.

Mediation Test: PER - CRE - SKEP - ATTD

Consequence

M1 (CRED) M2 (SKEP) Y (ATTD)

Antecendent Coeff. SE p Coeff. SE p Coeff. SE p

X (PER) a1 0.861 0.036 0.000 a2 −0.345 0.035 0.000 c′ 0.225 0.047 0.000
M1 (CRED) – – – d21 −0.495 0.031 0.000 b1 0.744 0.048 0.000
M2 (SKEP) – – – – – – b2 −0.044 0.060 0.458
constant iM1 0.028 0.150 0.853 iM2 7.396 0.094 0.000 iY 0.344 0.456 0.451
R2 = 0.588 R2 = 0.782 R2 = 0.794
F(1,405) = 578.913, p < ; 0.001 F(2,404) = 725.893, p < ; 0.001 F(3,403) = 518.759, p < ; 0.001

Total, Direct, and Indirect Effects

Total effect of X on Y
Effect SE t p LLCI ULCI
0.899 0.038 23.504 0.000 0.824 0.975
Direct effect of X on Y
Effect SE t p LLCI ULCI
0.225 0.047 4.830 0.000 0.133 0.316
Indirect effect(s) of X on Y effect(s)
Effect Boot SE BootLLCI BootULCI
Total: 0.675 0.051 0.580 0.778
Ind1 0.641 0.050 0.543 0.732
Ind2 0.019 0.031 −0.037 0.082
Ind3 0.015 0.025 −0.031 0.064
(C1) 0.622 0.071 0.482 0.755
(C2) 0.625 0.066 0.494 0.751
(C3) 0.004 0.009 −0.006 0.036
Indirect effect key
Ind1 per - > cred - > attd
Ind2 per - > cred - > skep - > attd
Ind3 per - > skep - > attd
Specific indirect effect contrast definitions
(C1) Ind1 minus Ind2
(C2) Ind1 minus Ind3
(C3) Ind2 minus Ind3

Table 8
Cluster Profile.

Numeric Variables

Ad Haters (n=181, 29.5%) Ad Accommodators (n=244, 39.8%) Ad Lovers (n=188, 30.7%) Test Statistic

M SD M SD M SD F (2, 608) Sig.

Ad avoidance 5.12 1.09 3.45 1.15 3.90 1.55 92.46 0.00

Ad skepticism 5.84 0.83 4.16 0.69 2.59 0.72 879.04 0.00
Personalized ad 2.41 1.11 4.19 0.79 5.52 0.70 587.62 0.00
Ad credibility 1.62 0.66 3.63 0.84 5.53 0.75 1204.82 0.00
Ad attitude 1.69 0.89 3.93 1.01 5.58 0.85 811.89 0.00
Behavioral intention 1.49 0.71 3.33 1.15 5.26 0.98 674.90 0.00
Age 2.65 1.02 2.71 0.91 2.56 0.89 1.35 0.26
Time on Facebook 1.89 0.85 2.00 0.90 2.34 0.93 12.66 0.00
Income 2.82 1.55 2.52 1.56 2.12 1.31 10.45 0.00

Categorical Variables

Ad Haters (n=181, 29.5%) Ad Accommodators (n=244, 39.8%) Ad Lovers (n=188, 30.7%)

Number % d % Number %

Gender
Male 53 29.28% 112 46.09% 108 57.75%
Female 128 70.72% 131 53.91% 79 42.25%
Students vs. Non-Students
Students 62 34.25% 74 30.45% 61 32.62%
Non-Students 119 65.75% 169 69.55% 126 67.38%

Note: pairwise comparison tests showed significant differences among three clusters at the .05 significant level (Bonferroni post hoc test) in all variables except age, students vs. non-
students, time on Facebook (Ad Haters and Accommodators) and income (Ad Haters and Accommodators)


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Lovers have more positive attitude toward the ad, that lead to higher Facebook, an ad can be made personalized using different criteria, such
behavioral intention after seeing a personalized ad (i.e, go for shopping as gender, location, prior purchase, or interaction. The current study
or make an order). Another key difference among three segments is has employed one criterion – the viewer's prior interaction on the In-
those belonging to Ad Lovers have spent significant more time on ternet – to make the ads personalized. This type of perceived perso-
Facebook than Ad Haters, but their income is much less than that of Ad nalization helps improve customer perception of ad as long as the ad is
Haters. With respect to gender, there were many more female in Ad perceived to be personalized. This finding is in line with other adver-
Accommodators and Ad Lovers than Ad Haters, while there was the tising literature (Dijkstra, 2005; Webb et al., 2007).
highest number of female in Ad Accommodators, followed by that in Ad Several multiple moderator mechanism has been tested in this re-
Haters, and lowest in Ad Lovers. search. The results confirm that the effects of perceived personalization
can be almost attributed to the role of mediators (ad credibility and ad
11. Discussion avoidance in the first mediation test or ad credibility and ad skepticism
in the second mediation test). In particular, in the first mediation test,
The paper has developed a comprehensive model that captures the since a personalized ad is viewed as more credible, the viewers are less
relationship between personalized ads on Facebook on customer's re- likely to avoid the ad and ultimately have more positive attitude toward
sponses to the ads. The model was tested through two phases using the ad. Likewise, in the second mediation test, because the viewers see a
different data sets (validation and estimation). The results collected personalized ad to be more credible, they are less skeptical against the
through MTurk revealed that all hypothesized relationships are sup- ad and ultimately have more positive attitude toward the ad.
ported except the effects of personalized ads on ad avoidance, ad The current research provides insights consistent to previous re-
skepticism on ad avoidance, and ad skepticism on ad attitude. In ad- search in terms of the influence of personalization (Baek and Morimoto,
dition, mediation tests implemented using PROCESS (Hayes, 2013) 2012; Keyzer et al., 2015). In two separate experiments about Facebook
confirm two mediation effects including (1) ad credibility and ad ads customized based on viewer's gender, Keyzer et al. (2015) confirm
avoidance mediate the relationship between personalized ads and ad that perceived personalization enhances customer responses toward
attitude and (2) ad credibility and ad skepticism mediate the relation- Facebook ads. The authors further explain that this effect is fully at-
ship between personalized ads and ad attitude. Cluster analysis per- tributed to perceived relevance as a mediator. That means if a message
formed on the combined data sets shows that respondents are grouped is more relevant, the message is perceived more positive. The current
in three different market segments whose size is about one third of the research, however, employ other moderators and tested the mediation
total. Those segments include Ad Haters, Ad Accommodators and Ad effects using serial multiple mediation analysis. And the results indicate
Lovers and each segment demonstrates their distinct perception about that other variables like ad credibility, ad skepticism, or ad avoidance
personalized ads on Facebook. could be a mediator, beside perceived relevance as mentioned in Keyzer
The current research is conducted after the announcement about the et al. (2015).
partnership between Facebook and IBM which has moved advertising The findings of this research provide further explanation to see-
technology to a next level where online retailers are able to place a mingly contradictory results found in previous research arguing that
personalized ad on the user's Facebook after their previous interactions personalization does not have positive significant impact on customer
on the Internet. The empirical findings illustrate that personalized ads response (Maslowska et al., 2011). Inconsistences may exist due to the
on Facebook have changed dramatically the ways customers perceive fact that those studies use actual personalization and manipulate it as 0
about the ads. Specifically, perceived personalization enhances positive versus 1. Nevertheless, perceived and actual personalization is not al-
responses (i.e., improving ad credibility and ad attitude) and reduces ways the same (Kramer et al., 2007). Drawing on the concept of per-
negative responses toward an ad (i.e., lowering ad skepticism). ceived personalization, the current paper asserts that perceived perso-
Although personalized ads do not have direct impact on ad avoidance, nalization induces positive customer attitude through altering
but they affects ad avoidance through other variables (i.e., ad cred- mediation variables (i.e., enhancing ad credibility, reducing ad avoid-
ibility). ance, or lowering ad skepticism). That results sheds light into the fact
The role of ad credibility cannot be underestimated in the context of that when future researchers consider personalized advertising, they
personalized advertising on Facebook. The findings demonstrate that should incorporate perceived personalization in addition to actual
the effects that ad credibility has on ad avoidance, ad skepticism and ad personalization.
attitude are all significant. That means that like in traditional media,
trustworthiness and believability of an ad on Facebook plays a role in 13. Managerial implications
helping customers eliminate their skepticism about the ad, reduce ad
avoidance, and develop favorable attitude toward the ad. This research is the first of the kind implemented in response to the
Unlike existing advertising literature (Baek and Morimoto, 2012), application of new advertising technology as part of integrated efforts
ad skepticism does not play a role in affecting ad avoidance and ad between IBM and Facebook to see whether personalization in adver-
attitude. These insignificant relationships could be attributed to (1) the tising changes customer's perception about the ads on Facebook.
nature of personalized advertising on Facebook in which even though Therefore, the findings of this research also provide implications for
customers are skeptical toward the content of the ad, they are not likely advertisers and marketers. An ability of online retailers to integrate
to avoid or ignore it and their attitude toward the ad is probably not their ads into a Facebook's user's account has added a new dimension of
affected; and (2) the bias that is derived from the pre-selected ads advertising that is based on customer's interactivity on the Internet.
(laptop or shoes) which are more related to the young group of Face- This new technology would enable e-retailers to accomplish their goals
book users. of advertising by improving positive effects (i.e., ad credibility) while
removing negative effects associated with the ad (i.e., ad avoidance,
12. Theoretical implications and ad skepticism). Ultimately ad attitude is significant enhanced
through the direct or indirect effects of personalized advertising, which
The findings of the research offer a number of theoretical implica- then leads to a higher behavioral intention toward the ad (Ganesh et al.,
tions. In consistence with the previous research focusing on traditional 2010; Senecal and Nantel, 2004).
media (Baek and Morimoto, 2012; Pavlou and Stewart, 2000; Tam and Adopting technology, online retailers are able to move from push
Ho, 2005), the results of the current research lends evidence that per- advertising to trust-based advertising that is associated with a number
ceived personalization plays a role in enhancing customer perception of of advantages. Personalized ads are developed on the basis of new
and customer response to an ad on social networking sites. Through technology. For e-retailers who consider choosing a platform to run an


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

effective advertising campaign, Facebook would definitely be con- 2010.01.003.

Geisser, S., 1975. The predictive sample reuse method with applications. J. Am. Stat.
sidered because the retailers could post personalized ads on the users’ Assoc. 70 (350), 320–328. http://dx.doi.org/10.1080/01621459.1975.10479865.
Facebook based on their previous online interaction (Aguirre et al., Goldberg, M.E., Hartwick, J., 1990. The effects of advertiser reputation and extremity of
2015). Three market segments including Ad Lovers, Ad Accom- advertising claim on advertising effectiveness. J. Consum. Res. 17 (2), 172–179.
http://dx.doi.org/10.1086/208547.
modators, and Ad Haters provide invaluable data for e-retailers that Goldsmith, R.E., Lafferty, B.A., Newell, S.J., 2000. The impact of corporate credibility and
help them tailor appropriate advertising strategies to targeted market celebrity credibility on consumer reaction to advertisements and brands. J. Advert.
segments, potentially leading to improved efficiency and effectiveness 29 (3), 43–54. http://dx.doi.org/10.1080/00913367.2000.10673616.
Gudergan, S.P., Ringle, C.M., Wende, S., Will, A., 2008. Confirmatory tetrad analysis in
of advertising budget and enhanced customer attitude toward the ad. PLS path modeling. J. Bus. Res. 61 (12), 1238–1249. http://dx.doi.org/10.1016/j.
jbusres.2008.01.012.
14. Limitations and future research Haenlein, M., Kaplan, A.M., 2004. A beginner's guide to partial least SquaresAnalysis.
Underst. Stat. 3 (4), 283–297. http://dx.doi.org/10.1207/s15328031us0304_4.
Hair, J.F., Black, W.C., Babin, B.J., Anderson, R.E., 2010. Multivariate Data Analysis.
There are some limitations associated with the current study that Retrieved from: 〈http://works.bepress.com/joe_hair/2〉.
open avenue for future research. First, the results may be biased based Hair, J.F., Hult, G.T.M., Ringle, C., Sarstedt, M., 2013a. A Primer on Partial Least Squares
on the selection of two product categories. Further research could Structural Equation Modeling (PLS-SEM). SAGE.
Hair, J.F., Hult, G.T.M., Sarstedt, M., 2013b. A Primer on Partial Least Squares Structural
employ more variable of product categories and brands to improve Equation Modeling, 1st edition. SAGE Publications, Inc, Los Angeles.
external validity. Second, the data were collected from online custo- Hayes, A.F., 2013. Introduction to Mediation, Moderation, and Conditional Process
mers using questionnaire, another research method (i.e., experimental Analysis: A Regression-Based Approach. Guilford Press.
Heeter, C., Greenberg, B.S., 1985. Profiling the zappers. J. Advert. Res. 25 (2), 15–19.
design) could be used to see whether the results hold the same that Henseler, J., Ringle, C.M., Sinkovics, R.R., 2009. The use of partial least squares path
would help improve generalizability of the study. Third, the study is modeling in international marketing. In: New Challenges to International Marketing.
limited to applying ad attitude and behavior intention as dependent Emerald Group Publishing Limited, Vol. 20, pp. 277–319. Retrieved from 〈http://
www.emeraldinsight.com/doi/abs/10.1108/S1474-7979(2009)0000020014〉.
variables, however, there are several other variables like brand equity, Ho, S.Y., Bodoff, D., 2014. The effects of web personalization on user attitude and be-
brand engagement that are highly relevant to managerial practice havior: an integration of the elaboration likelihood model and consumer search
should be considered in the future research. Finally, this research fo- theory. MIS Q. 38 (2) (497-A10).
Hulland, J., 1999. Use of partial least squares (PLS) in strategic management research: a
cuses only on personalized ads based on customer's prior views or
review of four recent studies. Strateg. Manag. J. 20 (2), 195–204. http://dx.doi.org/
searching. Future research should take into account how online mar- 10.1002/(SICI)1097-0266(199902)20:2<195::AID-SMJ13>3.0.CO;2-7.
keters incorporate customers’ activities pertaining to a brand such as Johns, R., Perrott, B., 2008. The impact of internet banking on business‐customer re-
lationships (are you being self‐served?). Int. J. Bank Mark. 26 (7), 465–482. http://
“likes” or “share” to create personalized advertising on Facebook and
dx.doi.org/10.1108/02652320810913846.
see how these ads change customer's perceptions about the ads. Kavanoor, S., Grewal, D., Blodgett, J., 1997. Ads promoting OTC medications: the effect
of ad format and credibility on beliefs, attitudes, and purchase intentions. J. Bus. Res.
References 40 (3), 219–227. http://dx.doi.org/10.1016/S0148-2963(96)00238-X.
Kelly, L., Kerr, G., Drennan, J., 2010. Avoidance of advertising in social networking sites.
J. Interact. Advert. 10 (2), 16–27. http://dx.doi.org/10.1080/15252019.2010.
Adomavicius, G., Tuzhilin, A., 2005. Personalization technologies: a process-oriented 10722167.
perspective. Commun. ACM 48 (10), 83–90. http://dx.doi.org/10.1145/1089107. Keyzer, F.D., Dens, N., Pelsmacker, P.D., 2015. Is this for me? How Consumers Respond to
1089109. Personalized Advertising on Social Network Sites. J. Interact. Advert. 15 (2),
Aguirre, E., Mahr, D., Grewal, D., de Ruyter, K., Wetzels, M., 2015. Unraveling the per- 124–134. http://dx.doi.org/10.1080/15252019.2015.1082450.
sonalization paradox: the effect of information collection and trust-building strategies Kim, J.W., Lee, B.H., Shaw, M.J., Chang, H.L., Nelson, M., 2001. Application of decision-
on online advertisement effectiveness. J. Retail. 91 (1), 34–49. http://dx.doi.org/10. tree induction techniques to personalized advertisements on internet storefronts. Int.
1016/j.jretai.2014.09.005. J. Electron. Commer. 5 (3), 45–62. http://dx.doi.org/10.1080/10864415.2001.
Awad, N.F., Krishnan, M.S., 2006. The personalization privacy paradox: an empirical 11044215.
evaluation of information transparency and the willingness to be profiled online for Kim, Y.J., Han, J., 2014. Why smartphone advertising attracts customers: a model of Web
personalization. MIS Q. 30 (1), 13–28. advertising, flow, and personalization. Comput. Hum. Behav. 33, 256–269. http://dx.
Baek, T.H., Morimoto, M., 2012. Stay away from me. J. Advert. 41 (1), 59–76. http://dx. doi.org/10.1016/j.chb.2014.01.015.
doi.org/10.2753/JOA0091-3367410105. Knowles, E.S., Linn, J.A., 2004. Resistance and Persuasion. Psychology Press.
Bleier, A., Eisenbeiss, M., 2015. The importance of trust for personalized online adver- Kramer, T., Spolter-Weisfeld, S., Thakkar, M., 2007. The Effect of Cultural Orientation on
tising. J. Retail. 91 (3), 390–409. http://dx.doi.org/10.1016/j.jretai.2015.04.001. Consumer Responses to Personalization. Mark. Sci. 26 (2), 246–258. http://dx.doi.
Chakraborty, G., Lala, V., Warren, D., 2003. What do customers consider important in org/10.1287/mksc.1060.0223.
B2B websites? J. Advert. Res. 43 (01), 50–61. MacKenzie, S.B., Lutz, R.J., 1989. An Empirical Examination of the Structural
Chen, Q., 1999. Attitude toward the site. J. Advert. Res. 39 (5), 27–37. Antecedents of Attitude toward the Ad in an Advertising Pretesting Context. J. Mark.
Cho, C.-H., Cheon, H.J., 2004. Why do people avoid advertising on the internet? J. 53 (2), 48–65. http://dx.doi.org/10.2307/1251413.
Advert. 33 (4), 89–97. http://dx.doi.org/10.1080/00913367.2004.10639175. Maslowska, E., Smit, E., van den Putte, B., 2011. Is personalized communication superior?
Choi, S.M., Rifon, N.J., 2002. Antecedents and consequences of web advertising cred- The effectiveness of personalization and the role of consumers’ characteristics. In:
ibility. J. Interact. Advert. 3 (1), 12–24. http://dx.doi.org/10.1080/15252019.2002. ACR Asia-Pacific Advances, AP-09. Retrieved from 〈http://acrwebsite.org/volumes/
10722064. 1009100/volumes/ap09/AP-09〉.
Cohen, J., 1992. A power primer. Psychol. Bull. 112 (1), 155–159. http://dx.doi.org/10. Murthi, B.P.S., Sarkar, S., 2003. The role of the management sciences in research on
1037/0033-2909.112.1.155. personalization. Manag. Sci. 49 (10), 1344–1362. http://dx.doi.org/10.1287/mnsc.
Coltman, T., Devinney, T.M., Midgley, D.F., Venaik, S., 2008. Formative versus reflective 49.10.1344.17313.
measurement models: two applications of formative measurement. J. Bus. Res. 61 Obermiller, C., Spangenberg, E.R., 1998. Development of a scale to measure consumer
(12), 1250–1262. http://dx.doi.org/10.1016/j.jbusres.2008.01.013. skepticism toward advertising. J. Consum. Psychol. 7 (2), 159–186. http://dx.doi.
Cotte, J., Coulter, R.A., Moore, M., 2005. Enhancing or disrupting guilt: the role of ad org/10.1207/s15327663jcp0702_03.
credibility and perceived manipulative intent. J. Bus. Res. 58 (3), 361–368. http://dx. Ohanian, R., 1990. Construction and validation of a scale to measure celebrity endorsers'
doi.org/10.1016/S0148-2963(03)00102-4. perceived expertise, trustworthiness, and attractiveness. J. Advert. 19 (3), 39–52.
Deshpandé, R., Stayman, D.M., 1994. A tale of two cities: distinctiveness theory and http://dx.doi.org/10.1080/00913367.1990.10673191.
advertising effectiveness. J. Mark. Res. 31 (1), 57–64. http://dx.doi.org/10.2307/ Pasadeos, Y., 1990. Perceived Informativeness of and irritation with local advertising. J.
3151946. Mass Commun. Q. 67 (1), 35–39. http://dx.doi.org/10.1177/107769909006700107.
Diamantopoulos, A., Winklhofer, H.M., 2001. Index construction with formative in- Pavlou, P.A., Stewart, D.W., 2000. Measuring the Effects and Effectiveness of Interactive
dicators: an alternative to scale development. J. Mark. Res. 38 (2), 269–277. http:// Advertising. J. Interact. Advert. 1 (1), 61–77. http://dx.doi.org/10.1080/15252019.
dx.doi.org/10.1509/jmkr.38.2.269.18845. 2000.10722044.
Dijkstra, A., 2005. Working mechanisms of computer-tailored health education: evidence Peltier, J.W., Schibrowsky, J.A., Schultz, D.E., 2003. Interactive integrated marketing
from smoking cessation. Health Educ. Res. 20 (5), 527–539. http://dx.doi.org/10. communication: combining the power of IMC, the new media and database mar-
1093/her/cyh014. keting. Int. J. Advert. 22 (1), 93–115. http://dx.doi.org/10.1080/02650487.2003.
Fornell, C., Larcker, D.F., 1981. Evaluating structural equation models with unobservable 11072841.
variables and measurement error. J. Mark. Res. 18 (1), 39–50. http://dx.doi.org/10. Perloff, R.M., 1993. The Dynamics of Persuasion, Communication Textbook Series.
2307/3151312. General Communication Theory and Methodology.
Ganesh, J., Reynolds, K.E., Luckett, M., Pomirleanu, N., 2010. Online shopper motiva- Petrison, L.A., Blattberg, R.C., Wang, P., 1997. Database marketing: past, present, and
tions, and e-store attributes: an examination of online patronage behavior and future. J. Interact. Mark. 11 (4), 109–125. http://dx.doi.org/10.1002/(SICI)1522-
shopper typologies. J. Retail. 86 (1), 106–115. http://dx.doi.org/10.1016/j.jretai. 7138(199723)11:4<109::AID-DIR12>3.0.CO;2-G.


T.P. Tran -RXUQDORI5HWDLOLQJDQG&RQVXPHU6HUYLFHV²

Pierrakos, D., Paliouras, G., Papatheodorou, C., Spyropoulos, C.D., 2003. Web usage Taylor, D.G., Lewin, J.E., Strutton, D., 2011. Friends, Fans, and Followers: do Ads Work
mining as a tool for personalization: a survey. Use. Model. Use.-Adapt. Interact. 13 on Social Networks? J. Advert. Res. 51 (1), 258–275. http://dx.doi.org/10.2501/
(4), 311–372. http://dx.doi.org/10.1023/A:1026238916441. JAR-51-1-258-275.
Punj, G., Stewart, D.W., 1983. Cluster analysis in marketing research: review and Thorson, E., Chi, A., Leavitt, C., 1992. Attention, memory, Attitude, and Conation: a test
Suggestions for Application. J. Mark. Res. 20 (2), 134–148. http://dx.doi.org/10. of the advertising hierarchy. ACR North Am. Adv., NA-19.. 〈http://acrwebsite.org/
2307/3151680. volumes/7323/volumes/v19/NA-19〉.
Roberts, M., Zahay, D., 2012. Internet Marketing: Integrating Online and Offline Tsang, M.M., Ho, S.-C., Liang, T.-P., 2004. Consumer attitudes toward mobile advertising:
Strategies. Cengage Learning. an empirical study. Int. J. Electron. Commer. 8 (3), 65–78. http://dx.doi.org/10.
Schroer, J., Hertel, G., 2009. Voluntary engagement in an open web-based encyclopedia: 1080/10864415.2004.11044301.
wikipedians and why they do it. Media Psychol. 12 (1), 96–120. http://dx.doi.org/ Tucker, C.E., 2014. Social networks, personalized advertising, and privacy controls. J.
10.1080/15213260802669466. Mark. Res. 51 (5), 546–562. http://dx.doi.org/10.1509/jmr.10.0355.
Senecal, S., Nantel, J., 2004. The influence of online product recommendations on con- Vesanen, J., Raulas, M., 2006. Building bridges for personalization: a process model for
sumers' online choices. J. Retail. 80 (2), 159–169. http://dx.doi.org/10.1016/j.jretai. marketing. J. Interact. Mark. 20 (1), 5–20. http://dx.doi.org/10.1002/dir.20052.
2004.04.001. Waters, R.D., Canfield, R.R., Foster, J.M., Hardy, E.E., 2011. Applying the dialogic theory
Simonson, I., 2005. Determinants of Customers' Responses to Customized Offers: con- to social networking sites: examining how university health centers convey health
ceptual Framework and Research Propositions. J. Mark. 69 (1), 32–45. http://dx.doi. messages on Facebook. J. Soc. Mark. 1. pp. 211–227. http://dx.doi.org/10.1108/
org/10.1509/jmkg.69.1.32.55512. 20426761111170713.
Speck, P.S., Elliott, M.T., 1997a. Predictors of Advertising Avoidance in Print and Webb, M.S., Hendricks, P.S., Brandon, T.H., 2007. Expectancy priming of smoking ces-
Broadcast Media. J. Advert. 26 (3), 61–76. http://dx.doi.org/10.1080/00913367. sation messages enhances the placebo effect of tailored interventions. Health Psychol.
1997.10673529. 26 (5), 598–609. http://dx.doi.org/10.1037/0278-6133.26.5.598.
Speck, P.S., Elliott, M.T., 1997b. Predictors of Advertising Avoidance in Print and Weilbacher, W.M., Walsh, H.R., 1952. Mail questionnaires and the personalized letter of
Broadcast Media. J. Advert. 26 (3), 61–76. http://dx.doi.org/10.1080/00913367. transmittal. J. Mark. 16 (3), 331–336. http://dx.doi.org/10.2307/1247547.
1997.10673529. Wind, J., Rangaswamy, A., 2001. Customerization: the next revolution in mass custo-
Srinivasan, S.S., Anderson, R., Ponnavolu, K., 2002. Customer loyalty in e-commerce: an mization. J. Interact. Mark. 15 (1), 13–32. http://dx.doi.org/10.1002/1520-
exploration of its antecedents and consequences. J. Retail. 78 (1), 41–50. http://dx. 6653(200124)15:1<13::AID-DIR1001>3.0.CO;2-#.
doi.org/10.1016/S0022-4359(01)00065-3. Xu, D.J., 2006. The influence of personalization in affecting consumer attitude toward
Stone, M., 1974. Cross-Validatory choice and assessment of statistical predictions. Journal mobile advertising in China. J. Comput. Inf. Syst. 47 (2), 9–19.
of the Royal statistical Society. Ser. B (Methodol.) 36 (2), 111–147. Yu, J. (Hyunjae), Cude, B.J., 2009. Possible disparities in consumers' perceptions toward
Tam, K.Y., Ho, S.Y., 2005. Web Personalization as a Persuasion Strategy: an Elaboration personalized advertising caused by cultural differences: U.S. and Korea. J. Int.
Likelihood Model Perspective. Inf. Syst. Res. 16 (3), 271–291. http://dx.doi.org/10. Consum. Mark. 21 (4), 251–269. http://dx.doi.org/10.1080/08961530802282166.
1287/isre.1050.0058.