Content uploaded by Angela Sasse
Author content
All content in this area was uploaded by Angela Sasse on Nov 11, 2014
Content may be subject to copyright.
Unpublished Manuscript - Please do not circulate
1
Too Close for Comfort: A Study of the Effectiveness and
Acceptability of Rich-Media Personalized Advertising
Miguel Malheiros, Charlene Jennett, Snehalee Patel, Sacha Brostoff, M. Angela Sasse
University College London
Gower Street
London WC1E 6BT, UK
{m.malheiros, c.jennett, s.brostoff, a.sasse}@cs.ucl.ac.uk, snehalee.patel.10.ucl.ac.uk
ABSTRACT
Online display advertising is predicted to make $25.27
billion this year. Advertisers believe targeted and
personalized ads to be more effective, but many users are
concerned about their privacy. We conducted a study where
30 participants completed a simulated holiday booking task;
each page showing ads with different degrees of
personalization. Participants fixated twice as long when ads
contained their photo. Participants reported being more
likely to notice ads with their photo, holiday destination,
and name, but also increasing levels of discomfort with
increasing personalization. We conclude that greater
personalization in ad content may achieve higher levels of
attention, but that the most personalized ads are also the
least acceptable. The noticeability benefit in using
someone‟s photo to make them look at an ad may be offset
by the privacy cost. As more personal data becomes
available to advertisers, it becomes important that these
trade-offs are considered.
Author Keywords
Targeted advertising, personalization, privacy.
ACM Classification Keywords
H.1.2 User/Machine Systems: Human Factors.
General Terms
Human Factors
INTRODUCTION
Display advertising (banner ads and pop-ups) accounts for
approximately one third of the total online advertising
market and is predicted to reach $25.27 billion this year,
with a 36% growth to $34.4 billion in 2013 [20]. Many
users are desensitized to traditional display advertising and
actively avoid looking at online banner ads [10]. Over time,
response rates to banner ads have fallen dramatically [14].
Techniques used by advertisers to overcome this problem
include targeted advertising and personalization. Targeted
advertising refers to the practice where ads are matched to
the user‟s interest. The more relevant the ad is to the user,
the more attractive it is. Personalization refers to the
inclusion of information in the ad content that identifies or
characterizes the recipient. It is sometimes used alongside
targeting to further increase the appeal of an ad. These
techniques have been found to achieve higher click-through
rates [32] and in turn more sales. However, they also create
ads which have the potential to be more invasive to users,
intruding on their privacy [30]. Yet there exists scope for
even greater personalization of advertisements. What will
happen to internet users‟ perceptions of privacy should
these more powerful techniques for personalization be
deployed? Will increasingly personalized ads lead to
increased revenues for advertisers and their clients, or
might it lead to a still greater experience of privacy
invasion, and rejection of products, services and sites
hosting the ads?
We report a study that explored participants‟ responses to
ads with varying degrees of personalization toward the
individual recipient, including a newer type that
incorporates personally identifying information (PII) about
the viewer within each ad (i.e. the participant‟s name and
photograph). We first present background on users‟
perceptions of targeted advertising and personalization. We
then describe the study where participants interacted with
web pages with increasingly targeted and personalized ads.
Their attention towards the ads was measured using eye-
tracking while their perceptions were collected with a
questionnaires and interviews. The results show that greater
personalization in ad content may achieve higher levels of
attention, with participants spending almost twice as much
time looking at an ad containing a photo of themselves than
at a standard picture ad. However, increasing
personalization also increased discomfort, with 80%
disagreeing they were comfortable with their photos being
used in the ads. We conclude that advertisers should strive
to identify high-value data items that can be used to achieve
„sweet spot‟ personalization that results in noticeable,
interesting ads that are also comfortable for the user, and
avoid data items that may increase the noticeability of their
ads at the expense of users‟ comfort.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise,
or republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee.
CHI 2012,
May 5-10, 2012, Austin, TX, USA.
Copyright 2012 ACM xxx-x-xxxx-xxxx-x/xx/xx...$10.00.
Unpublished Manuscript - Please do not circulate
2
BACKGROUND
Display Advertising
Targeted ads are mainly plain content text ads – such as
Google‟s AdSense, which generates $6 billion in revenue
[11]. The two most common forms of targeting are
contextual and behavioral. Contextual advertising (CA)
describes ads delivered based on an automated matching of
keywords from the content a user is currently viewing with
keywords for an advertisement. CA aims to complement the
website content and relies on information collected in real-
time. Behavioral advertising (BA) describes ads which are
delivered based on information collected about a user‟s web
browsing behavior over time, such as websites visited,
topics viewed and search engine queries. This data is used
to profile users into interest categories (e.g. „golf
enthusiast‟) and relevant ads are served. Examples of BA ad
networks include Google‟s Double Click, Yahoo! Network,
AOL Advertising and Scientific Media.
Past research suggests that BA can improve the click-
through rate (CTR) of an ad by as much as 670% [32]; and
the action-through rates (ATR; percentage of ads resulting
in sale) are more than double those of standard advertising,
6.8% and 2.8% respectively [4]. It is likely that targeted ads
are more effective because they are more relevant to users.
A strong correlation was found between users liking an ad
and its perceived relevance, those who dislike advertising
being the least likely to see any relevance in what they see
[15].
There is also evidence to suggest that BA is more effective
than CA. Studies conducted by advertising agencies found
that the same ads received 17% more fixations in unrelated-
content sites than related-content sites [19]; and the CTR
was more than 100% higher for ads in unrelated-content
sites, and the ATR was 19% higher, compared to related-
content sites [25]. Such results could be due to the „surprise
effect‟: when a user looking for a product finds an ad on an
unrelated site, s/he might react to the unexpected event by
engaging with the ad [19, 25]. Another explanation is that
contextual ads could suffer from the „clamor effect‟: when
too many adverts for the same product try and catch the
user‟s attention, the user might avoid looking at any of
them and instead choose to stay focused on the editorial
content [19, 25].
However, being served more relevant adverts does not
necessarily mean that users will perceive targeted
advertising positively – as can be seen in Table 1, studies
exploring the perceptions of users have had mixed results.
Reasons for disliking targeted advertising include
perceived privacy costs. Users dislike the idea of being
followed, describing BA as „invasive‟ [15, 23]. This has
Researchers
Year
N
Population
Survey Method
Findings
Internet
Advertising
Bureau and
Olswang [15]
2009
1,004
UK
Online
23% found the concept of BA appealing and 20%
found it unappealing. When asked whether they
would prefer BA as opposed to non-targeted ads, 27%
opted for BA while 17% preferred non-targeted ads.
Turow et al.
[30]
2009
1,000
US
Phone
66% did not want ads tailored to their interests,
compared to 32% yes and 2% maybe.
McDonald and
Cranor [23]
2009
2010
14
314
US
US
In-depth
interviews.
Online
Only 21% wanted the benefits of relevant advertising.
40% said that they would be more careful online if
they knew that advertisers were collecting data; 15%
said that they would stop using sites with BA.
Hastak &
Culnan [13]
2010
2,064
US
Online
46% were uncomfortable with BA, 31% were neutral
and 22% were comfortable.
Office of Fair
Trading [26]
2010
1,320
UK
Not Reported
40% held neutral views about BA, 28% disliked it and
24% welcomed it. 57% said that the practice of BA
would make no difference to their internet use, 5%
that they would limit their internet use, and 1% that
they would stop using the internet altogether.
TrustE [1]
2011
1,004
US
Not Reported
54% did not like BA and 37% had experienced a time
when they had felt uncomfortable with a targeted
online ad.
Table 1. Surveys investigating targeted advertising
Unpublished Manuscript - Please do not circulate
3
been termed the „creepiness factor‟, a sense that someone
has been „snooping‟ into a part of your life that should
remain private [16]. Other perceived privacy costs
identified in the literature include:
Cookies being installed on the user‟s computer [26];
The storage of personal data without the user‟s
knowledge [15];
PII being attached to the user‟s Internet browsing [1];
Being labeled by advertisers in ways the user considers
unfair [30];
Potential embarrassment to the user if using a shared
computer [26];
Other companies having access to the user‟s data [15];
Data collected being used for purposes other than
advertising [26].
CA raises fewer objections than BA [26]; because no
tracking is involved, there are fewer risks associated with
data storage or data sharing.
The benefits of targeted ads include:
Free access to ad-funded content [2, 26];
A reduction in irrelevant ads [26];
A reduction in the cost of good services [26];
Decreased search times [26].
The Internet Advertising Bureau suggest that the benefit of
ad-funded Internet services to the user outweighs the
privacy costs: they found that users were only prepared to
pay one-sixth of the total surplus gained to avoid
advertising and personal information-usage nuisance [2].
Users might argue, however, that it is not fair for
advertisers to expect them to make such a trade-off.
McDonald and Cranor [23] found that 69% believe privacy
is a right, 61% think it is „extortion‟ to pay to keep their
data private, and only 11% say they would pay to avoid ads.
Factors that could help alleviate users‟ privacy concerns
include transparency and control. Research findings suggest
that users feel more comfortable with BA in situations
where they are actively told when targeted ads are being
shown [13, 26]. Users are also more comfortable after
finding out PII is not stored [1, 15] and that they have the
option to opt-out [13, 15, 26].
One limitation of the existing research, however, is that the
majority of the user studies tend to be surveys that ask users
to reflect on the practice of targeted advertising. Arguably,
how a person feels about the practice in the abstract might
be different to how they feel when they are actually
presented with a concrete instance of a targeted ad in a
browsing situation.
Rich Media
Rich media - such as images, video and pop-ups – are
increasingly being used in display advertising. By making
the ad highly visible relative to the website content, the ad
is made harder for the user to ignore. Pop-ups have been
found to be more memorable than standard banner ads [9].
However, such advertising can also be experienced as
disruptive because it diverts the user from their online
goals. When an ad is considered disruptive, negative
attitudes can develop, affecting brand perception and
leading to „ad avoidance‟ [22]. The more important the
task, the more disruptive the interruption is likely to be
perceived.
With the growth of targeted ads, it is possible that
advertisers will try to combine targeting with high
visibility. Only one study has investigated users‟ possible
response to this approach. Goldfarb and Tucker [11]
conducted a large-scale field experiment on 2,892 web
advertising campaigns, comparing CA campaigns, rich
media campaigns, and campaigns that did both. They
conclude that users‟ purchase intent increased when CA and
rich media were used as separate strategies; but when these
strategies were combined, users‟ purchase intent decreased.
They suggest that users may tolerate CA more than other
ads because they potentially provide useful information;
however, when such ads are made highly visible, this has a
negative effect because it increases the user‟s awareness of
being targeted and their perceptions of being manipulated
by advertisers.
Personalization
Personalization is said to increase the appeal of an ad,
because the user is more likely to assume that there is a
match between his/her self and the product [3]. However,
highly personalized messages can also have negative
effects, depending on the degree to which the personal
information used in the message uniquely identifies or
characterizes the recipient. This is referred to as
„personalization reactance‟ - when the user feels
constrained in the sense of being too identifiable or
observable by the firm. White et al. [31] suggest that three
factors influence personalization reactance: the level of
personalization, whether or not justification for
personalization is present, and the perceived utility of the
service. In their study, they used highly personalized email
ads that addressed the customer by their name, state of
residence and movie preferences. They found that when the
perceived utility of the service was low, participants
experienced personalization reactance in response to highly
personalized messages that were not justified, resulting in
Unpublished Manuscript - Please do not circulate
4
lower click-through intentions. By contrast, when the
perceived utility of the service was high, the justification of
personalization was less important because highly
personalized messages were less likely to elicit reactance.
Only one research study has investigated the effects of
targeted display advertising and personalization. Tucker et
al. [29] conducted a randomized field experiment where
they compared the click-through rates of two different
Facebook ad formats, before and after the introduction of
improved privacy controls. In the targeted and personalized
ad format, the ad explicitly mentioned the user‟s
undergraduate institution, or the name of a celebrity the
user was a „fan‟ of, e.g. „As a [undergraduate institution
name] graduate, you know that strong women matter...‟ In
the targeted and non-personalized format, the message
referred to a broader user characteristic, e.g. „You know that
strong women matter...‟ They found that after Facebook‟s
introduction of improved privacy controls, users were twice
as likely to click the targeted personalized ads. As a result,
they suggest that if sites are successful at reassuring
consumers that they are in control of their privacy,
personalization of online ads can be used to generate higher
CTR.
Compared to email personalization, relatively low levels of
personalization are currently used in targeted display ads. In
particular, PII has not yet been used to personalize targeted
display ads. PII has been formally defined as „information
that can be used to distinguish or trace an individual‟s
identity‟; examples include a person‟s name and
photographic images [21]. Research studies suggest that PII
can make a message more noticeable. For example, in
psychology, the famous „cocktail party effect‟ describes
how a person can hear his/her own name being said
amongst many voices in a crowded room [6, 24]. More
recently, it has been suggested that people have prioritized
processing for their own name and their own face [28], and
that people have difficulty disengaging their attention from
self-referential stimuli [7, 8].
The majority of advertisers involved in BA claim not to
keep people‟s real names in their databases and often cite
this layer of anonymity as a reason why BA should not be
considered intrusive [27]. However, it is reported that some
companies, such as Rapleaf, do keep PII [27]. Also there is
evidence to suggest that advertisers have access to PII, even
if they are not using it: several studies have found that there
is „information leakage‟ from online social networks to
third-party advertisers, which can include PII and sexual-
orientation [12, 17, 18].
Another relevant finding is that it is a common belief
amongst Internet users that advertisers have access to PII. A
recent study found that over 30% of users believed that sites
they are registered with (e.g. Facebook, Google, Microsoft
Live, Yahoo) share PII with advertisers without their
consent; and more than half of users (52%) believe that
their PII are attached to tracking activity [1]. Following on
from this, we question how would users respond if
advertisers were open about having access to PII, and PII
was used to personalize advertising content?
Research Aims
In summary, it is evident that targeted advertising is an
effective strategy for advertisers, resulting in ads that are
better noticed by users and clicked more often; but at the
same time, users have a number of privacy concerns
regarding how data is collected and used. In this paper we
ask what will happen to internet users‟ perceptions of
privacy should more powerful techniques for
personalization be deployed? For example, how would
users react to ads that use PII? The aim of the current
research is to explore how users respond to targeted ads that
use rich media and PII. In particular, we wanted to explore
their reaction to the following types of ad:
Untargeted rich media ads;
Targeted rich media ads;
Personalized rich media ads, using PII of first name and
photo.
User studies investigating people‟s opinions of targeted ads
have tended to be survey-based, asking participants to rate
their level of agreement with various statements. We argue
however, that how a person feels about the practice of
targeting might be different to how they feel when
presented with targeted ads in an actual browsing situation.
To explore people‟s responses to our ad types, we designed
a study where participants were given the task of booking a
holiday. As the participant went through the booking
process, they were exposed to ads that became increasingly
personal – on the first page they were presented with
standard ads, on the second page they were presented with
ads that targeted them based on their holiday booking input,
and on the final page they were presented with personalized
ads that used their name and photo in the ad content. In
particular, we wanted to understand the following research
questions:
RQ1. Which ads did participants notice most / least?
RQ2. Which ads did participants find the most
comfortable / uncomfortable?
RQ3. Which ads were participants most / least likely to
take an interest in?
Unpublished Manuscript - Please do not circulate
5
METHOD
Participants
There were 30 participants (15 male, 15 female). Their ages
ranged from 19 to 55 years (mean age = 28 years, SD =
10.1). 22 were university students and 8 were university
staff, recruited from an opportunity sample.
Stimuli
A travel website („Flyaway‟) was created using HTML,
CSS and JavaScript. The website was split into three pages,
each page containing four banner adverts (top left, top right,
bottom left, bottom right). The adverts were all the same
size (221 by 336 pixels) and consisted of text and rich
media. See Figure 1 for examples of the adverts.
Figure 1. Examples of ‘Flyaway’ ads. Top left: a general
holiday ad. Top right: a holiday ad based on holiday selection
‘Dubai’. Bottom left: an ad based on the age selection ‘18-34’.
Bottom right: an anti-aging cream ad using the participant’s
first name and modified photo
Page 1 allowed the participant to select their journey
information (destination, journey type, departure date and
time, return date and time) and a series of additional
questions to „qualify for our exclusive offers‟ (relationship
status, do you own a car, do you have travel insurance, age
group). The ads on this page were general ads about
holidays and flights. See Figure 1, top left ad, for an
example.
Page 2 allowed the participant to select the number of
tickets and to enter their name, address and payment details.
The ads on this page were targeted using the holiday
destination the participant chose on Page 1 (e.g. local
hotels, restaurants) and their answers to the additional
questions on Page 1 (e.g. dating website, car loan, travel
insurance). See Figure 1, top right, for an example.
Page 3 confirmed the booking and informed the participants
that their booking reference would be emailed to them
shortly. The ads on this page were targeted using the age
range the participant chose on Page 1 (e.g. clubbing, life
cover), addressed the participant by their first name, and
used the participant‟s photo to show them what they could
look like with / without a particular product (e.g. hair salon,
anti-wrinkle cream). See Figure 1, bottom left and right ads,
for examples. The participant‟s photo was obtained from
the university database when the participant signed up for
the study and was modified using Photoshop. The
modifications were changing the hair-style in one version,
and artificially aging the appearance of the individual by 40
years in another version.
Apparatus
The website was displayed on a Dell desktop computer
using Internet Explorer 7. Eye movements were measured
with a Tobii X50 eye tracker and Tobii Studio 2.0.4
software. Total fixation duration (TFD) was collected in
order to gauge noticing of the stimuli ads (RQ1), with
longer durations indicating ads that had been noticed more.
The post-task interview was recorded using an audio
recorder.
Materials
A post-task questionnaire was created that consisted of 13
statements, which participants had to rate how on a 5-point
Likert scale, indicating their level of agreement. Q1 was a
general statement, where participants rated their awareness
of the website‟s ads. The 12 questions that followed then
focused on four of the targeted ads: holiday destination,
age, name and photo. Participants were asked to rate each
of the ad types for how likely they were to notice the ad
(RQ1), how comfortable they felt with the ad (RQ2) and
how likely they were to take an interest in the ad (RQ3).
Procedure
The experimenters applied for permission to conduct the
study through the university‟s ethical review process. They
sought permission to use the participant‟s university ID
photo (from a publicly accessible page) and to display
modified versions of it to the participant during the study.
The study took place in a usability lab and took
approximately 30 minutes per participant. It was advertised
as an experiment to investigate „Perceptions of a Travel
Unpublished Manuscript - Please do not circulate
6
Website.‟ Participants signed a consent form detailing the
procedure of the experiment, what equipment would be
used, informed that the data would be held in accordance
with local data protection law, and of their right to
withdraw from the experiment at any time without
consequence. However, participants were not told that the
focus of the study would be the website‟s adverts, and that
their photo would be used for a subset of the ads.
Participants were asked to book a flight to a destination of
their choice and to „talk aloud‟ about their thoughts of the
website. While they did the task, eye tracking and video
recording were used to record their reactions. Once the task
was completed, the researcher reviewed the Tobii screen
recording with the participant and this time asked
participants to specifically talk about what they thought of
the ads on each page.
Next participants were asked to complete a post-study
questionnaire, which asked them to rate the ads with regard
to how noticeable, comfortable and likely to elicit interest
they were. They then took part in an interview exploring
their perceptions of targeted and personalized advertising in
the context of their prior experience.
At the end of the study participants were fully debriefed and
informed that the photos of themselves would not be
published, and that all data relating to them from the
experiment could be destroyed at their request. All
participants were paid £5 for their time.
RESULTS
Attention
Eye tracking data was analyzed using Tobii Studio 2.0.4
Software. The four ads on each page were defined as areas
of interest (AOI). Aggregating the data for the four AOIs,
descriptive statistics for TFD were then calculated for each
page. (Note that 5 participants were excluded from the
sample due to poor data quality.)
Page
Total Fixation Duration (s)
Mean
SD
1
4.6
3.8
2
4.7
5.4
3
9.5
6.3
Table 2. Descriptive statistics for total fixation duration (n=25)
As can be seen in Table 2, the ads on Page 3 received twice
as much attention (mean TFD = 9.5 seconds) as the ads on
Page 1 and Page 2. A repeated measures one-way ANOVA
was revealed that there was a significant effect for Total
Fixation Duration (TFD), F (2, 48) = 10.16, p<.001.
Bonferroni-corrected pairwise comparisons (sig. level =
.016) revealed that the TFD for Page 3 was significantly
higher than Page 1 (p=.009) and Page 2 (p<.001).
Two ads on Page 3 were compared, to test the effect of
displaying an ad with the participant‟s photo while
controlling for potential differences between pages. An ad
which used the participant‟s age and a standard picture
(top-left) was compared against an anti-ageing cream ad
using the participant‟s photo (top-right). The ad with the
participant‟s photo was looked at for 5.8 seconds longer
than the standard picture ad (mean TFDs = 13.0 seconds
and 7.2 seconds respectively). A repeated measures t-test
revealed that this difference was statistically significant, t
(24) = 3.2, p=.003.
Questionnaire Results
The questionnaire responses for all participants (n=30) were
analyzed using SPSS. Four questions were negated (Q6,
Q7, Q9, Q11), so that for all items 1 = strongly disagree, 2
= disagree, 3 = neutral, 4 = agree and 5 = strongly agree.
Likely to Notice
The majority of participants agreed that they were more
likely to notice ads that use their photo (97%), holiday
destination (77%) and name (57%); see Table 3.
I am more likely to notice
adverts that use my…
+ ve
0
- ve
Holiday destination (Q2)
23
(77%)
5
(17%)
2
(7%)
Age (Q5)
7
(27%)
13
(43%)
9
(30%)
Name (Q8)
17
(57%)
6
(20%)
7
(23%)
Photo (Q11)
29
(97%)
0
(0%)
1 (3%)
Table 3. Frequencies for Noticing. +ve = Strongly Agree or
Agree, 0 = Neutral, - ve = Disagree or Strongly Disagree
(n=30)
Descriptive statistics revealed that participants rated adverts
using their photo highest for being noticeable (M=4.6),
followed by adverts using their holiday destination
(M=3.8), their name (M=3.5) and age (M=3.0). A repeated-
measures one-way ANOVA revealed that there was a
significant effect for Noticing, F (3, 87) = 16.0, p<.001.
Bonferonni-corrected pairwise comparisons (sig. level
=.008) revealed that photo was rated significantly more
noticeable than holiday destination (p=.005), age (p<.001)
and name (p<.001). Holiday destination was rated
significantly more noticeable than age (p<.001).
Feeling Comfortable
87% of participants agreed that they would feel comfortable
with their holiday destination being used in ads and more
Unpublished Manuscript - Please do not circulate
7
than two-thirds of participants disagreed that they would
feel comfortable with their photo (80%) or name (66%)
being used in ads; see Table 4.
I feel comfortable with
adverts that use my…
+ ve
0
- ve
Holiday destination (Q3)
26
(87%)
3
(10%)
1
(3%)
Age (Q6)
7
(23%)
13
(43%)
10
(33%)
Name (Q9)
7
(23%)
4
(13%)
19
(66%)
Photo (Q12)
3
(10%)
3
(10%)
24
(80%)
Table 4. Frequencies for Comfort. +ve = Strongly Agree or
Agree, 0 = Neutral, - ve = Disagree or Strongly Disagree
(n=30)
Descriptive statistics revealed that participants rated adverts
using their holiday destination (M=4.0) as most
comfortable, followed by adverts using their age (M=2.9),
their name (M=2.3) and photo (M=1.7). A repeated
measures one-way ANOVA was conducted to test for
significance. To compensate for violations of the sphericity
assumption (Mauchley‟s W(df=5) = .65, p=.037), the
significance levels were adjusted according to the lower-
bound procedure. There was a significant effect for
Comfort, F (1, 30) = 26.7, p<.001. Bonferonni-corrected
pairwise comparisons (sig. level = .008) revealed that
holiday was rated significantly more comfortable than age
(p<.001), name p<.001) and photo (p<.001). Age was rated
as significantly more comfortable than photo (p=.001).
Taking An Interest
77% of participants agreed that they would be more likely
to take an interest in ads that used their holiday destination
and over half of participants disagreed that they would be
more likely to take an interest in ads that used their photo
(67%) or name (57%); see Table 5.
I’m more likely to take
an interest in adverts
that use my…
+ ve
0
- ve
Holiday destination (Q4)
23
(77%)
6
(20%)
1
(3%)
Age (Q7)
7
(30%)
16
(53%)
5
(17%)
Name (Q10)
5
(17%)
8
(27%)
17
(57%)
Photo (Q13)
10
(23%)
0
(0%)
20
(67%)
Table 5. Descriptive statistics for Interest. +ve = Strongly
Agree or Agree, 0 = Neutral, - ve = Disagree or Strongly
Disagree (n=30)
Descriptive statistics revealed that participants rated adverts
using their holiday destination (M=3.9) highest for interest
followed by adverts using their age (M=3.1), their name
(M=2.4) and photo (M=2.4). A repeated-measures one-way
ANOVA was conducted to test for significance. To
compensate for violations of the sphericity assumption
(Mauchley‟s W(df=5) = .47, p<.001), the significance levels
were adjusted according to the lower-bound procedure.
There was a significant effect for Interest, F (1, 30) = 13.7,
p<.001. Bonferonni-corrected pairwise comparisons (sig.
level = .008) revealed that holiday was rated significantly
more comfortable than age (p<.001), name (p<.001) and
photo (p<.001).
Interviews
Interview transcripts were analyzed using thematic analysis,
a “method for identifying, analyzing and reporting patterns
(themes) within data” [5]. This type of analysis involves
coding (tagging) interesting sections of the transcript in a
consistent way and subsequently grouping those codes into
themes. Themes help explain what the data means and
relate it to the research questions [5]. Five major themes are
described in this section: (1) relevance of the ads; (2)
perception of own photo; (3) how advertisers obtain
personal data; (4) the extent to which advertisers access and
use personal data; and (5) other people seeing ads with
another person‟s data.
Relevance
The majority of participants (23) identified „relevance‟ of
the ads as an important factor in how they perceived
targeted and personalized advertising. A relevant ad was
described as an ad which was related to the individual‟s
interests, activity on the website, or topic of the website. In
the context of our study that meant ads related to holidays
were considered by these participants as more relevant.
Relevance was associated to a more positive perception of
the ads. P5 said “I mean I think that they are more
attractive if they have things that are relevant to me […]”.
Own photo
More than half the participants (19) expressed negative
reactions to seeing ads with their own photo. When
referring to the ads in the study that manipulated their photo
into looking older and having different haircuts these
participants used adjectives such as: “disturbing” (P2,
P25), “strange” (P4), “weird” (P12, P13), “freaky” (P13),
“creepy” (P14), or “terrible” (P17).
When asked how they would react to ads that used their
own photo these participants said they would feel
uncomfortable. P1 said “[…] the face is a very important
thing and identifying yourself is important but umm it‟s
Unpublished Manuscript - Please do not circulate
8
creepy yeah and it might turn me off and it might turn
several people off a lot.”
In order to gauge the strength of feeling and judge possible
consequences for ad-hosts of using this level of
personalization, participants were asked about how they
would feel if a site that is frequently used and relied upon,
such as Facebook, started using their photo in ads targeted
at them. 5 participants said they would quit the site. 6 said
they would continue using it even though they would feel
uncomfortable about the use of their photo. 7 said they
would be comfortable.
17 participants mentioned that ads with their own photo
would be more noticeable. P27 said “[…] well in terms of
advertising it might work well if you use someone‟s picture
because you immediately notice that.”
How did they get my data?
For 18 participants, how advertisers had obtained their data
and where it had come from influenced how they perceived
targeted ads. One specific issue was data from one website
or company being used to show ads on another website.
P10 said “I don‟t understand how they know what you‟ve
been looking at on another website.”
Understanding how the ad had been created had a
comforting effect. P18 said “Yeah, I would prefer targeted
adverts as long as I knew how they got the fact that they‟re
targeted. As long as, yeah, I was aware of, it was just you
know that i could see that I looked at it before and they
were just advertising something, and that was it, then I‟d be
more comfortable and happy with that […]”
In our study, the photos of participants were obtained from
the university pages. Knowing this made participants more
comfortable with its use in ads. P5 said “Yeah I would be
surprised and a bit umm not comfortable with it, I mean the
fact that I know that it is a university, that it is my university
picture and that I am at university, then it doesn‟t make me
uncomfortable [...]” Not realizing where the photo came
from made participants uncomfortable. P14 said “I think
that‟s weird, because I‟m like „Where did they get that
picture?‟”
Access to / use of personal data
For 17 participants the extent of personal data that
advertisers had access to, and used for creating ads had an
effect on their perceptions. For example, P7 said “I don‟t
want anything specifically focused on me because then
again it presumes that my life is pretty open but for instance
if you‟re digging into my life it‟s none of your business.
Consent to use personal data in targeted advertisement was
mentioned by 5 participants. Using individual‟s personal
data without consent in order to create ads was perceived
negatively. P4 said “[…] I don‟t think I would want my
image being used for something without my knowledge, I
mean if they like approached people and asked to use it
then that would be different but I wouldn‟t want it used
without my knowledge.”
Other people seeing ads with my data
9 participants were concerned about the potential for other
people to see ads with their data because of errors in the
targeting, or people sharing computers. For example, if
personalized ads started to make use of personal photos, the
wrong photo could be displayed to the wrong person. P19
said “Well they have to be rather accurate to know which
… I mean there may be … are so many, many names, have
the same name so they may get the wrong picture from a
person with the same name.” Computers storing an
individual‟s web browsing behavior could also introduce
problems if they are shared. P11 said “The computer or the
website will have the memory of my searching. The next
time my friend or somebody else uses my computer they can
see what I bought. If I just, I only buy the cream or
moisturizer, those kind of things, that‟s okay. But if it‟s very
private I don‟t want them to be able to see that.”
DISCUSSION
The goal of the study was to explore participants‟
perceptions of rich-media targeted and personalized
advertising. We investigated how participants perceive
targeted and personalized ads that use increasingly personal
data with regards to noticeability, interest, and comfort.
Questionnaire results indicate that depending on the data
item used to create an ad it can become significantly more
or less noticeable. Ads which use the participant‟s photo,
name, or holiday destination are more likely to be noticed.
Ads that used their photo were perceived by participants as
being significantly more noticeable than ads that used their
age, name, or holiday destination. An individual‟s photo is
not commonly displayed without her/his knowledge as part
of an ad in commercial websites, so it is possible that they
were considered highly noticeable due to a „surprise effect‟
[19, 25]. An additional explanation is that individuals are
slower to disengage their attention when looking at a photo
of themselves [7, 8], so it‟s possible that they looked at ads
with their photo for longer periods of time and more times
than the other ads. This possibility is supported by the mean
TFD results. Ads on Page 3 were looked at for significantly
longer than ads on Pages 1 and 2. Also, when comparing
the ads at the top of Page 3, the ad with the participant‟s
photo was looked at for significantly longer than the ad
with the standard picture.
The level of interest participants had in the different types
of ads was significantly influenced by the type of data item
used. Questionnaire results revealed that they were more
likely to take an interest in ads that use their holiday
destination, and less likely to take an interest in ads that use
their name and photo. The use of age had no effect on
interest. Ads that used holiday destination were considered
Unpublished Manuscript - Please do not circulate
9
significantly more likely to raise interest than ads that used
age, name, or photo. This can be attributed to the fact that
the task participants were asked to complete was
intrinsically related to holidays; thus ads with holiday
destination may have been seen as more relevant in the
context than the other ads. This explanation is supported by
the interviews which show that the majority of participants
identified „relevance‟ of an ad as having a positive
influence in how they perceived it. The link between
relevance of an ad and whether people like it has also been
suggested in past research [15]. This supports the
conclusion that in order to make users interested in their ads
advertisers should make an effort to make ads relevant for
the context users are engaging with.
The type of data item used in the ads has a significant effect
on how comfortable participants were with it. Participants
reported being comfortable with ads using their holiday
destination, neutral about ads using their age, and
uncomfortable with ads using their name or photo. Ads that
used holiday destination were rated significantly more
comfortable than ads that used the other three types of data.
Ads that used photo were rated significantly less
comfortable than ads that used age and holiday destination,
with the majority of participants saying they felt
uncomfortable with the use of their photos in ads. Again,
relevance of the ad may be used to explain these results:
previous research has shown that individuals are more
comfortable with personal data use in ads when it is
perceived as relevant [31]. It is likely that participants
perceived holiday destination as relevant data item in that
context, but not their own photo. Additionally, the
interviews indicated that not knowing how advertisers had
obtained the data used to create targeted ads was
discomforting. It is possible that, while it was clear for
participants that holiday destination had been collected
from the form they were filling in, it was more difficult to
remember the source of their photo. Advertisers should
avoid using personal data that make users feel
uncomfortable about ads. The use of personal photos in
particular may upset users and lead them to reject services,
as indicated by participants‟ answers to the possibility of
Facebook employing this type of advertising. The
interviews suggest that asking users for consent before
using their data in advertising could alleviate their
concerns.
The type of personal data used in creating the targeted ads
has, according to the questionnaire results, a highly
significant effect on noticeability, interest, and comfort.
Items such as holiday destination in our scenario help to
create ads which are both considered noticeable, interesting,
and comfortable – so should be of great value for
advertisers since they will help get the attention of potential
customers, convert that attention into purchases, while not
creating feelings of privacy invasion on the individual.
Identifying these data types in different contexts on the web
should be of great interest to advertisers. At the same time
advertisers should also be careful with data items that can
increase noticeability of ads but which are considered too
sensitive to be used in ads by individuals. There could be a
short term benefit in using someone‟s photo to make them
look at an ad, but if that ad makes the individuals
uncomfortable then the privacy cost may offset the
noticeability benefit. As more personal data becomes
available to advertisers on the web, it becomes more
important that these trade-offs are considered.
CONCLUSIONS
The findings described in this paper suggest that users‟
perceptions of targeted ads using rich-media vary
depending on the type of data used to create the ads, with
comfort decreasing as the level of personalization increases.
Advertisers should strive to identify high-value data items
that can be used to achieve „sweet spot‟ personalization that
results in noticeable, interesting ads that are also
comfortable for the user. At the same time, advertisers
should be wary of using data items that can increase the
noticeability of their ads at the expense of users‟ comfort
since this could be counterproductive for the advertised
brand.
To our knowledge, this is the first study of how
personalized rich media ads are perceived by users and
where different types of personalization were compared
with regards to their impact on user perceptions. Past
studies on targeted advertising have typically been surveys,
whereas we gauged participant‟s live reactions to adverts in
the lab. It was also the first study to investigate users‟
reactions to ads that used their own photo. Targeted
advertising seeks to make ads more relevant to the recipient
and related to their interests. It is becoming increasingly
prevalent and, with advertising companies having access to
new sources of personal data such as social networks, we
believe the trend toward targeting may become a trend
toward personalization. Therefore, by using participants‟
photos in ads for anti-aging cream or makeovers we are
anticipating what the future of display advertising can be.
The main limitation of this study was the size and
composition of the participant sample. It was also
participants‟ first interaction with ads that used their photo.
Further research is needed to determine whether users
habituate to these ads over time, if different users perceive
these ads more positively than others, or if combination
with other types of content changes users‟ perceptions.
Participants being asked to „talk aloud‟ may also have
artificially increased their sensitivity for the ads.
Although this was a first-step study, we can state
confidently that the use of PII in this context is a complex
issue that needs to be handled with care and that imprecise
targeting or personalization could deter potential customers
Unpublished Manuscript - Please do not circulate
10
from engaging with the brand. Data quality and
personalization errors in advertising are topics we would
like to pursue in future research.
REFERENCES
1. 2011 Consumer Research Results: Privacy and Online
Behavioral Advertising, TRUSTe, 2011.
2. Consumers driving the digital update: The economic
value of online advertising-based services for
consumers, IABEurope, London, 2010.
3. Anand, B. and Shachar, R. Targeted advertising as a
signal. Quantitative Marketing and Economics, 7 3
(2009), 237-266.
4. Beales, H. The Value of Behavioral Targeting, Network
Advertising Initiative, Washington, D.C., 2010.
5. Braun, V. and Clarke, V. Using thematic analysis in
psychology. Qualitative Research in Psychology, 3
(2006), 77-101.
6. Cherry, E. Some Experiments on the Recognition of
Speech, with One and with Two Ears. J. Acoust. Soc.
Am., 25 5 (1953), 975.
7. Devue, C. and Brédart, S. Attention to self-referential
stimuli: Can I ignore my own face? Acta Psychologica,
128 2 (2008), 290-297.
8. Devue, C., Van der Stigchel, S., Brédart, S. and
Theeuwes, J. You do not find your own face faster; you
just look at it longer. Cognition, 111 1 (2009), 114-122.
9. Diao, F. and Sundar, S.S. Orienting Response and
Memory for Web Advertisements. Communication
Research, 31 5 (2004), 537-567.
10. Drèze, X. and Hussherr, F.-X. Internet advertising: Is
anybody watching? Journal of Interactive Marketing, 17
4 (2003), 8-23.
11. Goldfarb, A. and Tucker, C. Online Display
Advertising: Targeting and Obtrusiveness. Marketing
Science (2011).
12. Guha, S., Cheng, B. and Francis, P., Challenges in
Measuring Online Advertising Systems. in IMC '10:
Proceedings of the 10th ACM SIGCOMM conference on
Internet measurement conference, ACM, (2010).
13. Hastak, M. and Culnan, M.J. Future of Privacy Forum
Online Behavioral Advertising “Icon” Study Summary
of Key Results, 2010.
14. Hollis, N. Ten Years of Learning on How Online
Advertising Builds Brands. Journal of Advertising
Research, 45 02 (2005), 255-268.
15. Kean, A. and Dautlich, M. A guide to online
behavioural advertising, Internet Advertising Bureau,
London, 2009.
16. Knowledge@Wharton. Privacy on the web: Is it a losing
battle? Knowledge@Wharton Philadelphia, PA, 2008.
17. Korolova, A., Privacy Violations Using Microtargeted
Ads: A Case Study. in Data Mining Workshops
(ICDMW), 2010 IEEE International Conference on,
(2010) 474-482.
18. Krishnamurthy, B. and Wills, C., On the leakage of
personally identifiable information via online social
networks. in WOSN '09: Proceedings of the 2nd ACM
workshop on Online social networks, ACM, (2009) 7-
12.
19. Marcus, N. Behavioral Targeting Has More Visibility
iMEDIA CONNECTION, 2006.
20. MarketingCharts. Global web ad spend to rise 31% in 2
yrs, Watershed Publishing, 2011.
21. McCallister, E., Grance, T. and Scarfon, K. Guide to
Protecting the Confidentiality of Personally Identifiable
Information (PII), NIST, Gaithersburg, MD 2010.
22. McCoy, S., Everard, A., Polak, P. and Galletta, D. The
effects of online advertising. Commun. ACM, 50 3
(2007), 84-88.
23. McDonald, A.M. and Cranor, L.F. Beliefs and
behaviors: Internet users‟ understanding of behavioural
advertising Telecommunications Policy Research
Conference, 2010.
24. Moray, N. Attention in dichotic listening: Affective cues
and the influence of instructions. Quarterly Journal of
Experimental Psychology, 11 1 (1959), 56-60.
25. Newcomb, K. Behavioral ads convert better out of
context Clickz, New York, 2006.
26. OFT. Online targeting of advertising and prices: A
market study, Office of Fair Trading, London, 2010.
27. Steel, E. A web pioneer profiles users by name Wall
Street Journal, New York, NY, 2010.
28. Tacikowski, P. and Nowicka, A. Allocation of attention
to self-name and self-face: An ERP study. Biological
Psychology, 84 2 (2010), 318-324.
29. Tucker, C. Social Networks, Personalized Advertising,
and Privacy Controls, NET Institute, 2010.
30. Turow, J., King, J., Hoofnagle, C.J., Bleakley, A. and
Hennessy, M. Americans Reject Tailored Advertising
and Three Activities that Enable It. SSRN (2009).
31. White, T., Zahay, D., Thorbjørnsen, H. and Shavitt, S.
Getting too personal: Reactance to highly personalized
email solicitations. Marketing Letters, 19 1 (2008), 39-
50.
32. Yan, J., Liu, N., Wang, G., Zhang, W., Jiang, Y. and
Chen, Z. How much can behavioral targeting help
online advertising? Proceedings of the 18th
Unpublished Manuscript - Please do not circulate
11
international conference on World wide web, ACM,
Madrid, Spain, 2009, 261-270.
