You are on page 1of 20

Advertising Effectiveness and Annoyance Dynamics

Examining the Advertising Effectiveness and


Annoyance Dynamics
Completed Research Paper

Anindya Ghose Param Vir Singh


New York University Carnegie Mellon University
Stern School of Business Tepper School of Business
44 West 4th Street 5000 Forbes Avenue
New York, NY 10012 Pittsburgh, PA 15213
aghose@stern.nyu.edu psidhu@cmu.edu

Vilma Todri
Emory University
Goizueta Business School
1300 Clifton Road
Atlanta, GA 30309
vtodri@emory.edu

Abstract
Digital advertisers often harness technology-enabled advertising scheduling strategies, such as
ad repetition at the consumer level, to improve advertising effectiveness. However, such strategies
can elicit annoyance in consumers, who increasingly deploy ad-blocking technologies. Our study
captures this trade-off between effective and annoying display advertising. We propose a hidden
Markov model (HMM) that allows us to investigate both the enduring impact of display
advertising on consumers' purchase decisions and the potential of persistent display advertising
to stimulate annoyance in consumers. Additionally, we study the structural dynamics of these
display advertising effects by allowing the corresponding effects to be contingent on the latent
state of the funnel path in which consumers reside. Our findings demonstrate a tension exists
between generating interest and triggering annoyance in consumers; whereas display
advertising has an enduring impact on transitioning consumers further down in the funnel path,
display-advertising exposures beyond a frequency threshold can have an adverse effect by
increasing the probability that consumers will be annoyed. Investigating the dynamics of these
annoyance effects, we reveal that consumers who reside in different stages of the funnel path
exhibit considerably different tolerance for annoyance stimulation. Our paper contributes to the
scant but important literature of annoyance effects in digital advertising and has significant
managerial implications for the online advertising ecosystem.

Keywords: adverse effects, technology-enabled strategies, consumer annoyance, dynamic


models, Markov models, online advertising

Introduction
Online display advertising started as a limited experiment in the 1990s (Ryan and Graham 2014) and has
grown into an industry worth $32.17 billion (eMarketer 2016b). To put the size of the digital-advertising
market into perspective, in 2016, digital display-ad spending eclipsed search-ad spending in the United
States for the first time and accounted for the largest share of digital-ad spending. One factor that has
greatly contributed to this remarkable growth and dominance of display advertising is the digital

Thirty Eighth International Conference on Information Systems, South Korea 2017 1


Advertising Effectiveness and Annoyance Dynamics

transformation of the advertising industry fueled by the advancement of information technology systems.
Admittedly, information technology has revolutionized the way firms deliver advertising messages and
connect with prospective customers online (Adamopoulos and Todri 2014; Todri and Adamopoulos 2014).
For instance, thanks to various technological innovations, advertisers are nowadays able to track
consumers’ digital footprints at a fine-grained level and purchase advertising inventory on a per-impression
consumer basis via programmatic instantaneous auctions (Chen et al. 2011).
As a result, firms can effectively harness technology-enabled ad-scheduling strategies and exert great
control over the frequency of display-advertising exposures at the consumer level. Advertisers’ refined
ability to control the frequency of display-advertising exposures has important implications for the online
advertising ecosystem. On the one hand, the ability to control the media scheduling and engage in increased
ad repetition at the consumer level has the potential to improve the effectiveness of display advertising.
That is, advertisers often choose to increase the frequency of individual-level display-advertising exposures
because ad repetition can more effectively influence consumers’ learning (Cacioppo and Petty 1979; Naik et
al. 1998) and positively affect key communication dimensions such as recall, attitude, and purchase
intention (Belch 1982; Campbell and Keller 2003; Sawyer 1981). In addition, as websites become more
cluttered with content, advertisers choose to increase the frequency of individual-level display-advertising
exposures in order to draw consumers’ attention and prolong engagement with the advertisements. On the
other hand, however, advertisers’ technology-enabled ability to engage in increased ad repetition at the
consumer level can lead to excessive display-advertising exposures, which might trigger annoyance in
consumers and possibly undermine advertisers’ goals (Aaker and Bruzzone 1985). Specifically, persistent
display-advertising exposures might elicit annoyance in consumers as advertising intervenes with the users’
main goal of navigating the Internet or completing a task online (Goldstein et al. 2014).
As a result of the perceived goal impediment, annoyed consumers cope with the negative effects of
annoyance by actively minimizing the duration of their actual exposure to the advertisements (Goldstein et
al. 2014; Wilbur 2015) to eliminate the source of nuisance (Elliott and Speck 1998). Hence, annoyed
consumers’ actions would entail moving to another part of the webpage to avoid being directly exposed to
display ads that elicit annoyance or, alternatively, clicking away from webpages that contain the annoying
ads. Fundamentally, these annoyed consumers’ reactions reduce the time the consumers are exposed to the
display ads (Cho 2004) (i.e., reducing the in-view time exposure). In this paper, we utilize such key
consumer behaviors in order to better understand consumers' annoyance elicitation and we allow the
corresponding consumer behavior theory to inform and guide the structure of the empirical model.
The substantial issue of annoyance has gained significant traction in the digital advertising ecosystem due
to the growing body of anecdotal evidence that digital advertising can be a source of significant annoyance
for consumers. In particular, the remarkable popularity of ad-blocking software1 (e.g., over 500 million
downloads of Adblock Plus as of January 2016 (Williams 2016)) indicates consumers increasingly seek to
avoid exposure to online advertisements. In fact, consumers cite annoying advertisements as the main
reason they adopt such software (eMarketer 2016a). This mainstream adoption of ad-blocking software has
detrimental effects both for advertisers, because ad blockers eliminate an effective and powerful channel of
communication with the consumers, as well as for publishers, because ad blockers disrupt their main
business revenue model (Scott 2016). Hence, it is important to understand what triggers consumers’
annoyance with digital advertising, and use such knowledge to accordingly adapt the promotional and
publishing strategies.
Towards this direction, we examine the trade-off between effective and annoying display advertising. We
propose a hidden Markov model (HMM) that allows us to simultaneously investigate both the enduring
impact of display advertising on consumers’ purchase decisions and the potential of display advertising to
stimulate annoyance in consumers. Our paper makes a significant contribution to the relevant literature on
advertising effectiveness and consumer annoyance in technology-mediated environments. First, this paper
conducts an empirical analysis to understand whether and when technology-enabled frequent display-
advertising exposures can elicit consumers’ annoyance. This contribution is of paramount importance as,
to the best of our knowledge, this is the first paper to examine whether display advertising scheduling (i.e.,

1Ad blocking or ad filtering is a type of software that can remove or alter advertising content from a
webpage.

Thirty Eighth International Conference on Information Systems, South Korea 2017 2


Advertising Effectiveness and Annoyance Dynamics

the frequency of firm-initiated advertising exposures) can trigger consumers’ annoyance and, thus, the
paper advances our understanding of the antecedents of consumer annoyance. Second, this paper further
contributes to the literature on consumer annoyance as it also examines the structural dynamics of the
effective and annoying display-advertising effects by allowing the corresponding effects to be contingent
on the latent state of the funnel path in which consumers reside. Hence, the paper not only allows
consumers to have heterogeneous responses to advertising but also captures heterogeneous responses in
consumers’ annoyance elicitation. This is an important contribution as drawing on existing theories it is not
clear whether a consumer who has moved further down in the purchasing funnel path will be more or less
easily annoyed with frequent display advertising exposures compared to a consumer who resides in the
early stages of the funnel path. For instance, on the one hand, consumers who reside in the later stages of
the funnel path might be less easily annoyed because they have developed more favorable attitudes toward
the brand but, on the other hand, consumers who reside in the early stages of the funnel path are less
informed about the brand and, thus, they might be the ones who are less easily annoyed with frequent
display advertising exposures because such interactions can convey valuable information to them.
Therefore, this is a question that requires empirical investigation and, hence, our paper also contributes to
the literature by studying such structural dynamics of consumer annoyance.
Our findings demonstrate a tension exists between generating interest and triggering annoyance in
consumers via display advertising. Although display advertising can have an enduring impact on
transitioning consumers further down in the funnel path and increasing the likelihood of a purchase,
display-advertising exposures beyond a certain frequency threshold substantially increase the probability
that consumers will be annoyed. Investigating the dynamics of these annoyance effects, we reveal that
consumers who reside in different stages of the funnel path exhibit different thresholds for annoyance
stimulation. In particular, we find that consumers who reside in the early stage of the funnel path can be
annoyed by as few as three display-advertising exposures within the same day, whereas consumers who
reside in a later engaged stage of the funnel path would need at least seven display-advertising exposures
within the same day to reach a substantial probability of annoyance effects. Our findings have significant
managerial implications both for advertisers, who would like to capitalize on the enhanced opportunities
that digital advertising unlocks without stimulating annoyance in consumers, and the rest of the digital
advertising ecosystem.

Literature Review
In this section, we discuss the related literature and how this paper builds upon and extends various streams
of related research. First, our paper is closely related to the literature examining the effectiveness of
advertising and the impact of ad repetition on enhancing advertising effectiveness. The literature on
advertising effectiveness postulates that the main reason advertising is effective is that it conveys relevant
information to consumers (Nelson 1970; Nelson 1974) in a direct or indirect way. Specifically, consumers
are often not informed about the existence of products or their respective attributes due to the imperfect
and asymmetrical information structure of the markets (Spence 1976; Stiglitz 1989). Therefore, firms seek
to eliminate this type of ignorance by investing in advertising aimed at influencing consumers’ learning
(Animesh et al. 2010; Bagwell 2007; Feng and Xie 2012; Stigler 1961). Studying factors that can enhance
advertising effectiveness, prior literature has demonstrated ad repetition can improve consumers’ learning
(Cacioppo and Petty 1979; Heflin and Haygood 1985; Naik et al. 1998) and can affect communication
dimensions such as recall, positive attitude, and purchase intention (Anand and Shachar 2009; Campbell
and Keller 2003; Ghose and Yang 2009; Sawyer 1981), rendering the increased frequency of advertising
exposures an effective digital-media scheduling strategy for firms. Besides, digital advertisers today need
to increasingly rely on ad repetition in order to enhance the effectiveness of their marketing campaigns,
because consumers experience increased information overload (Adamopoulos 2016; Tan et al. 2015) as,
for instance, websites have become more cluttered with content (Anderson and De Palma 2012). Similarly,
the proliferation of product offerings is increasing (Brynjolfsson et al. 2003), leading to even greater
consumer uncertainty.
Our paper also contributes to a growing body of literature related to annoyance and irritation in advertising.
Focusing on traditional means of advertising, several studies have identified aspects of advertising that can
stimulate negative feelings in consumers and trigger annoyance. For instance, a few papers have shown that
irritation can occur as a function of the advertising content and the creative execution of the TV commercial

Thirty Eighth International Conference on Information Systems, South Korea 2017 3


Advertising Effectiveness and Annoyance Dynamics

(Aaker and Bruzzone 1985). In particular, content that overdramatizes a situation, “puts down” a person
with respect to intelligence or sophistication, and, in general, exaggerates can stimulate annoyance in
consumers (Aaker and Bruzzone 1985; Bauer and Greyser 1968). Furthermore, advertisements that
excessively stimulate consumers’ senses can also elicit feelings of annoyance. For example, consumers
might feel overwhelmed when they are exposed to advertisements that are too noisy or strident
(Pokrywczynski and Crowley 1993). Moving beyond the traditional means of advertising, the concept of
annoyance has received scant attention in the context of online media. In this stream of literature, Yun Yoo
and Kim (2005), focusing on the presentation of advertising content, find that rapidly animated
advertisements that excessively stimulate consumers’ senses can lead to more negative consumer attitudes.
Similarly, Goldstein et al. (2014) also focus on the content of a display advertisement as a source of
annoyance in digital media. They find that animated display advertisements incur an increased cognitive
cost to the consumers as well as an economic cost to the publishers. An important question, however, which
currently remains unanswered, is the extent to which other advertising strategies, beyond the
advertisement content itself, can stimulate annoyance in consumers. Hence, this paper fills this gap in the
advertising and annoyance literature and addresses this question by evaluating whether technology-
enabled digital-media scheduling strategies can stimulate annoyance in consumers.
Another stream of literature that is closely related to our paper is that of structural dynamics. The related
literature suggests marketing activities, such as advertising, work in a series of discrete stages and transition
consumers over time between these discrete stages, which are often regarded as latent. Capturing these
consumer behavior dynamics is crucial, because research has shown that consumers’ responses to ads vary
as they progress to different stages in the decision process (e.g., (Abhishek et al. 2012; Netzer et al. 2008)).
These discrete-stage models of buying behaviors, also known as stages of a conversion “funnel path”,
typically capture the effect of advertising in increasing consumers’ awareness and then raising interest in
the product and the brand (Abhishek et al. 2012; Barry 1987; Tellis 2003). Besides, allowing for structural
marketing dynamics of advertising effectiveness, we can also examine the dynamics of annoyance effects.
Hence, our paper contributes to the literature of annoyance in advertising by also examining whether
annoyance effects are contingent on the latent state of the conversion funnel path in which consumers
reside. For instance, we examine whether the users who exhibit higher interest in the product (i.e., users
who reside in the later latent stages of the funnel path) have a higher or lower tolerance for annoyance
elicitation.

Empirical Data
Our data set was collected in collaboration with a major US online media analytics and optimization plat-
form2. The media analytics company handles the entire online campaign of an online retailer that sells
experience and credence goods (i.e., goods whose utility impact is difficult or impossible for the consumer
to ascertain) and runs paid search as well as display-advertising campaigns across various websites. We
have a random sample of 5,000 real-world consumers who were exposed to online advertisements that were
run by the specific online retailer during a period of six months in 2014. During this period, the online
company did not engage in online or offline advertising activities apart from those we observe in the data
set. The data set contains firm-initiated advertising exposures and user-initiated actions, such as direct
visits to the advertiser’s website (Chen and Hitt 2002) and organic search-engine clicks. Hence, an
observation in the data set corresponds to a display-advertising impression, a search click (paid or organic),
a direct visit to the advertiser’s website, and an actual purchase, if any. During the aforementioned period
of six months, the advertiser placed the ads to a set of popular websites. A unique characteristic of this data
set is the access to granular information regarding the viewability of display-advertising exposures to
consumers3. For instance, we know when an impression was visible on a consumer’s screen area and hence
whether it was deemed viewable. Tracking information regarding the viewability of impressions reveals a
remarkably high percentage of the display-advertising impressions (on average 55%) are never rendered

2 Due to the nature of the data-sharing agreement, we are unable to reveal the name of the firm.
3Information on viewability is collected through (1) a geometric method that typically involves comparing
the position of the four corners of the ad relative to the host webpage and then comparing the four corners
of the browser’s viewport relative to the host webpage and (2) browser-optimization monitoring techniques.

Thirty Eighth International Conference on Information Systems, South Korea 2017 4


Advertising Effectiveness and Annoyance Dynamics

viewable; similar statistics have been reported from various ad media analytics platforms (Ghose and Todri
2016). Hence, to accurately estimate the impact of display advertising in generating interest and stimulating
annoyance, we filter out all the non-viewable impressions for the estimation of the model. Thus, we are able
to alleviate biases that could have emerged due to this measurement error, because such non-viewable
impressions are essentially not eligible to have an impact on consumers’ transition across the funnel path
and their buying behavior.
We construct the consumer funnel paths by connecting the touchpoints (i.e., consumer-firm interactions,
such as advertising exposures or customer-initiated search actions) based on the anonymous unique
customer identifiers. Subsequently, we construct all the variables at the time-period level (i.e., a day) for
each individual. The individual-level data have been collected through cookies (i.e., digital code stored on
user’s web browser) and advanced tracking techniques, such as fingerprinting (i.e., a technique that allows
advertisers to collect information regarding the characteristics of the computer, such as installed plug-ins
and software, in order to form a unique signature similar to random friction ridges on a human finger). If a
funnel path is successful, we do observe the corresponding conversion, which means the consumer made
an actual purchase on the advertiser’s website. Furthermore, we distinguish between brand- and non-
brand-related search queries based on the campaign that was triggered to serve the search advertisement.
In particular, the corresponding campaigns have refined targeting criteria that prevent a non-brand search-
advertising campaign from serving an advertisement to a brand-related search query. Finally, we also have
information on the duration of the exposure to the display advertisements. Table 1 provides the descriptive
statistics of the aforementioned variables at the observational level.

Table 1. Descriptive Statistics of Variables.

Variable Mean St. Deviation


Display Ads 0.11853 1.2077
Brand Search Clicks 0.00121 0.0524
Non-brand Search Clicks 0.00039 0.0212
Organic Search Clicks 0.00133 0.0517
Direct Visits 0.00419 0.0516
Online Purchase 0.00071 0.0267
In-View Exposure Display Ads 0.28207 1.6134

Modeling Consumer Behavior Dynamics


We propose a hidden Markov model (HMM) in order to study the dynamics and the trade-off between
effective and annoying digital-display advertising. An HMM is a type of stochastic signal model in which
the system being modeled is assumed to be a Markov process with unobserved hidden states (Rabiner
1989). In particular, an HMM is a doubly stochastic process with an underlying stochastic process that is
not directly observable, and can only be observed through another set of stochastic processes that produce
the sequence of observed signals (Rabiner and Juang 1986). Each latent state has a probability distribution
over the possible output signals; therefore, the sequence of the signals provides information about the
sequence of the states. HMMs have been employed to study temporal pattern-recognition problems, such
as speech recognition, and they remain the dominant machine-learning technique for speech applications.
More recently, HMMs have been employed to study various aspects of user behavior, involving longitudinal
individual-level data (e.g., (Abhishek et al. 2012; Kokkodis and Lappas 2016; Netzer et al. 2008; Sahoo et
al. 2012; Singh et al. 2011)). HMMs can also be considered a generalization of mixture models, which are
already popular in the fields of marketing and information systems.
HMMs are a powerful modeling technique, the use of which provides a number of significant advantages
for our research question and empirical setting. First, it allows us to capture the latent nature of consumers’
state of annoyance and transition across the funnel path by utilizing the signals consumers emit throughout
the funnel path. For instance, although annoyance is a latent state in the funnel path, consumers who have
been annoyed engage in distinct behaviors that signal their annoyance (Goldstein et al. 2014; Wilbur 2015;
Woltman Elpers et al. 2003) and these behaviors are utilized by the HMM to infer annoyance of consumers.

Thirty Eighth International Conference on Information Systems, South Korea 2017 5


Advertising Effectiveness and Annoyance Dynamics

Second, the proposed technique allows the theory drawn from the relevant streams of literature to inform
the structure of the HMM while estimating the effect of time-varying variables from the data. Third, the
proposed technique enables us to study the structural dynamics of the effective and annoying advertising
effects by allowing the magnitude of these effects to be contingent on the latent state in which consumers
reside.
For estimating the proposed HMM, we consider a random set of consumers who have interacted with the
focal brand over time. We construct the consumer funnel paths by connecting the advertising exposures
based on the anonymous customer identifiers. Each funnel path is initiated the first time the advertiser
success- fully targets a customer with display advertising (i.e., display impression is rendered viewable) or
the first time the customer engages in an active search behavior (e.g., direct visit, search engine click)
(Abhishek et al. 2012). If a funnel path is successful, we observe the corresponding conversion, which means
the consumer made an actual purchase on the advertiser’s website. Each funnel path is divided into time
periods. For each period (i.e., day), we observe the display advertisements that might affect consumers’
search and purchase behaviors as well as the actual search clicks, advertising response, and purchasing
behaviors, if any. For any given period denoted as 𝑡, a consumer 𝑖 resides in one of the latent states of the
funnel path. To capture the annoying effects of advertising, if any, we augment the classic funnel-path states
(i.e., Awareness, Interest, Purchase 4 ) with the additional unobserved state of annoyance. Hence, the
proposed HMM consists of three hidden (unobserved) states s ∈ {Annoyance, Awareness, Interest} and an
observed state, the Purchase state. Consumers can transition to the Annoyance state from anywhere across
the funnel path before they actually make an online purchase, if any. If consumers reach the Annoyance
state, they can probabilistically transition back to the Awareness or Interest state at any time. The possible
state transitions across the three latent states of the proposed HMM are graphically depicted in Figure 1.

Figure 1. Graphical representation of probabilistic transitions of consumers across the HMM states.

4The designated names of the funnel path states effectively capture the distinct consumer behaviors across
the various states that the fitted HMM model recovers as indicated in the section Empirical Results. In
particular, consumers who reside in the Awareness state are aware of the brand as they have engaged in
active search process or they have seen display advertisements that were rendered viewable. Moreover,
consumers in the Internet state are indeed observed to express higher interest in the brand through various
channels (e.g., direct visits, paid searches, etc.) while consumers in the Annoyance state are the ones who
do not engage in such actions and, more importantly, they also actively minimize their exposures to the
display ads, which constitutes a typical behavior of annoyed consumers (Goldstein et al. 2014; Wilbur 2015;
Woltman Elpers et al. 2003)

Thirty Eighth International Conference on Information Systems, South Korea 2017 6


Advertising Effectiveness and Annoyance Dynamics

More specifically, in the Awareness state reside consumers who have successfully been targeted by
advertisements and are merely aware of the existence of the product but are still away from making a
purchase. At any time period, consumers who reside in the Awareness state might progress further down
in the funnel path (i.e., forward transition) and reach the Interest state. In the first stage of the funnel path,
consumers reside in the Awareness state. In the Interest state, consumers possess more favorable attitudes
toward the advertiser and actively express their interest in the focal brand and product by explicitly
engaging in active search behaviors (i.e., direct website visits, organic search clicks, paid search clicks, etc.).
At the same time, we also allow backward transitions along the funnel path (i.e., consumers’ transition from
the Interest state to the Awareness state) in order to capture a possible loss of consumers’ interest for the
focal product and brand. At any time period 𝑡, consumers can also transition to the Annoyance state if
persistent display-advertising exposures in the same time period have triggered their annoyance.
Consumers who reside in the Annoyance state are distinguished from consumers who reside in the
Awareness or Interest state as annoyed consumers cope with the negative effects of annoyance by actively
minimizing the duration of their actual exposure to the display advertisements (Goldstein et al. 2014;
Wilbur 2015; Woltman Elpers et al. 2003) in order to eliminate the source of noise or nuisance (Elliott and
Speck 1998). If consumers reach the Annoyance state, they can transition back to the Awareness or Interest
state at any time. The transition of consumers across the latent states (i.e., Annoyance, Awareness, Interest)
is probabilistically determined and is affected by the advertising exposures consumers encounter in their
funnel path to purchase. In the proposed HMM, the transition across the latent states is characterized by a
Markov process. This stochastic process is then transformed in the observed search, advertising response,
and buying behaviors (i.e., emission variables). Furthermore, every latent state is characterized by its own
“stickiness” capturing the varying levels of difficulty pertaining to consumers’ transitions from the state in
which they reside to another state. Finally, the state of Purchase is an observed state in which the customer
expresses ultimate interest in the product by completing an online purchase, if any, successfully exiting the
current funnel path and entering a new one since the focal products can be purchased repeatedly.

Hidden Markov Model’s Main Components

Consider, for consumer 𝑖 , a fixed state sequence 𝑆 𝑖 = 𝑆&' 𝑆&( 𝑆&) … 𝑆&+ , where 𝑆&' is the initial state for
consumer 𝑖 and 𝑆&, ∈ {1 = 𝐴𝑛𝑛𝑜𝑦𝑎𝑛𝑐𝑒, 2 = 𝐴𝑤𝑎𝑟𝑒𝑛𝑒𝑠𝑠, 3 = 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡, 4 = 𝑃𝑢𝑟𝑐ℎ𝑎𝑠𝑒} and an observed
outcome sequence 𝑂 𝑖 = 𝑂&' 𝑂&( 𝑂&) … 𝑂&+ . Given the set of latent states and the sequence of observed
outcomes, the HMM compromises three elements: the initial state distribution 𝜋 , the state transition
probability distribution 𝑄, and the observed outcome probability vector 𝐴. The probability that we observe
the outcome sequence 𝑂 𝑖 given the state sequence 𝑆 𝑖 and the parameter set 𝜆 = (𝜋, 𝑄, 𝐴) is
+

𝑃 𝑂 𝑖 𝜆, 𝑆 𝑖 = 𝑃(𝑂&, |𝜆, 𝑆&, ).


,K'

Then, we obtain
𝑃 𝑂 𝑖 𝜆, 𝑆 𝑖 = 𝑎 𝑂&' 𝑆&' 𝑎 𝑂&( 𝑆&( … . 𝑎 𝑂&+ 𝑆&+ ,
where 𝑎 𝑂&, 𝑆&, is the probability of observing outcome 𝑂&, given that consumer 𝑖 is in state 𝑆&, at time 𝑡.
Note that 𝑎 𝑂&' 𝑆&' is an element of the outcome probability vector 𝐴(𝑖, 𝑡).
The probability of state sequence 𝑆 𝑖 is given by
𝑃 𝑆 𝑖 𝜆 = 𝜋 𝑖 𝑞 𝑆&' , 𝑆&( … 𝑞 𝑆&, , 𝑆&,M' … 𝑞 𝑆&+N' , 𝑆&+ ,
where 𝜋 𝑖 is the initial probability that consumer 𝑖 is in state 𝑆&' in period 𝑡 = 1 and 𝑞 𝑆&, , 𝑆&,M' is the
probability that consumer 𝑖 is in state 𝑆&,M' in period 𝑡 + 1 given that she was in state 𝑆&, in period 𝑡. Note
that 𝑞 𝑆&, , 𝑆&,M' is an element of the state transition matrix 𝑄 𝑖, 𝑡 → 𝑡 + 1 for consumer 𝑖. The probability
that 𝑂 𝑖 and 𝑆 𝑖 occur simultaneously is
𝑃 𝑂 𝑖 , 𝑆 𝑖 𝜆 = 𝑃 𝑂 𝑖 𝜆, 𝑆 𝑖 𝑃 𝑆 𝑖 𝜆 .
Therefore, the probability that the observed outcome sequence 𝑂(𝑖) given the model parameter set 𝜆 is the
likelihood of observing this sequence and is obtained by summing the aforementioned equation over all
possible values of state sequence 𝑆(𝑖):

Thirty Eighth International Conference on Information Systems, South Korea 2017 7


Advertising Effectiveness and Annoyance Dynamics

𝐿 𝑂 𝑖 =𝑃 𝑂 𝑖 𝜆 = 𝑃 𝑂 𝑖 𝜆, 𝑆 𝑖 𝑃 𝑆 𝑖 𝜆 .
∀ S(&)

The individual likelihood can be written in more compact matric notation as following:
𝐿 𝑂 𝑖 = 𝜋 𝑖 𝛬 𝑖, 1 𝑄 𝑖, 1,2 𝛬 𝑖, 2 𝑄 𝑖, 2,3 … . 𝑄 𝑖, 𝑇-1, 𝑇 𝛬 𝑖, 𝑇 𝟏' , (1)
where the matrix 𝛬 𝑖, 𝑡 = 𝑑𝑖𝑎𝑔(𝑎 𝑂&, 𝑆&, = 1 , 𝑎 𝑂&, 𝑆&, = 2 , 𝑎(𝑂&, |𝑆&, = 3)) and 𝟏 is a 𝑛 𝑥 1 vector of ones.
As discussed, two main components of an HMM are the transition matrix and the state-dependent choices
(emission variables). In the proposed model, the transition matrix is governed by a sequence of first order
Markovian transitions 𝑄(𝑖, 𝑡, 𝑡 + 1). The transition of consumers across the latent states of the HMM are
probabilistically determined and the advertising exposures consumers encounter during the last time
period are allowed to affect these transitions. At each time period 𝑡, consumers may transition in another
state or continue residing in the same state. The probability that a consumer 𝑖 who resides in state 𝑗 in time
period 𝑡 transitions to state 𝑘 in time period 𝑡 + 1 is denoted as 𝑞&, 𝑗, 𝑘 = 𝑃 𝑆&,M' = 𝑘 𝑆&, = 𝑗 ). Section
“State Transition Matrix” discusses in detail the state transition matrix for the proposed HMM.
Furthermore, another important component of an HMM is the state-dependent choices of information-
gathering, advertising response and purchase likelihood behaviors. These state dependent choices pertain
to the probability of the consumer exhibiting these behaviors at time period 𝑡 and are contingent on the
state the consumer resides during that period. We denote the probability of purchase (i.e., conversion) as
𝑃(𝐶&, = 1|𝑆&, = 𝑗), where 𝑆&, is the state of customer 𝑖 at time period 𝑡 in a Markov process and 𝐶&, denotes
the decision of the consumer 𝑖 to make an online conversion at time period 𝑡. Similarly, we denote the
probabilities that the consumer will search a specific number of times through different channels and the
probability that the consumer will view the display advertisement for 𝑣 units of time as described in Table
2. Section “State Dependent Consumer Behavior” discusses in detail the state-dependent choice
probabilities for the proposed HMM.

Table 2. Definition of state-dependent consumer choices.

Notation Description of state-dependent consumer actions


Probability that consumer 𝑖 will make an online conversion (i.e., actual
𝑃(𝐶&, = 1|𝑆&, = 𝑗)
purchase) at time period 𝑡 given that she resides in state 𝑗.
Probability that consumer 𝑖 will directly visit the advertiser’s website d times at
𝑃(𝐷&, = 𝑑&, |𝑆&, = 𝑗)
time period 𝑡 given that she resides in state 𝑗.
Probability that consumer 𝑖 will click on advertiser’s organic search results 𝑜
𝑃(𝑅&, = 𝑟&, |𝑆&, = 𝑗)
times at time period 𝑡 given that she resides in state 𝑗.
Probability that consumer 𝑖 will click on advertiser’s brand-related paid search
𝑃(𝐵&, = 𝑏&, |𝑆&, = 𝑗)
results 𝑏 times at time period 𝑡 given that she resides in state 𝑗.
Probability that consumer 𝑖 will click on advertiser’s generic paid search results
𝑃(𝐺&, = 𝑔&, |𝑆&, = 𝑗)
𝑔 times at time period 𝑡 given that she resides in state 𝑗.
Probability that consumer 𝑖 will view the display advertisement on average 𝑣
𝑃(𝑉&, = 𝑣&, |𝑆&, = 𝑗)
amount of time at time period 𝑡 given that she resides in state 𝑗.

Hence, the information-gathering, advertising response and purchase likelihood behavior of a consumer
corresponds to a multivariate outcome denoted as 𝑂&, = {𝐶&, , 𝐷&, , 𝑅&, , 𝐵&, , 𝐺&, , 𝑉&, }.

State Transition Matrix


The transition between states is modeled as a random walk allowing consumers to probabilistically move
from one state to another. The transition further down in the funnel path (e.g., moving from the Awareness
state to the Interest state) captures the effectiveness of advertising (i.e., the enduring impact of advertising)
towards increasing consumers’ propensity to make an online purchase while the backward transition in the

Thirty Eighth International Conference on Information Systems, South Korea 2017 8


Advertising Effectiveness and Annoyance Dynamics

funnel path (e.g., moving from the Interest state to the Awareness state) captures the loss of interest for the
focal product and brand. At the same time, the probabilistic transition of customers to the Annoyance state
captures the annoying effects of display advertising. Thus, the random walk transition matrix for states s ∈
{1=Annoyance, 2=Awareness, 3=Interest, 4= Purchase} is defined as follows:
𝑞𝑖𝑡 (1,1) 𝑞𝑖𝑡 (1,2) 𝑞𝑖𝑡 (1,3) 𝑞𝑖𝑡 (1,4)
𝑞𝑖𝑡 (2,1) 𝑞𝑖𝑡 (2,2) 𝑞𝑖𝑡 2,3 𝑞𝑖𝑡 (2,4)
𝑄 𝑖, 𝑡, 𝑡 + 1 = , (2)
𝑞𝑖𝑡 (3,1) 𝑞𝑖𝑡 (3,2) 𝑞𝑖𝑡 (3,3) 𝑞𝑖𝑡 (3,4)
0 1 0 0
g
where 𝑞&, 𝑗, 𝑘 = 𝑞&, 𝑆&, = 𝑗, 𝑆&,M' = 𝑘 = 𝑃 𝑆&,M' = 𝑘 𝑆&, = 𝑗 ) and for each state 𝑗 , hK' 𝑞&, 𝑗, 𝑘 = 1 and
0 ≤ 𝑞&, 𝑗, 𝑘 ≤ 1 ∀ 𝑗, 𝑘 ∈ {1, 2, 3, 4}.
The propensity for transition from sate 𝑗 to state 𝑘 for consumer 𝑖 during time period 𝑡 (𝑃𝑆𝑇&,jh ) is affected
by advertising exposures such as display advertising and search advertising as well as customer initiated
actions that can have a positive towards increasing consumers’ propensity of making a purchase (i.e., direct
visits). Hence,
l o p q
𝑃𝑆𝑇&,jh = 𝛽jh 𝐷𝑖𝑠𝑝𝑙𝑎𝑦&, + 𝛽jh 𝐵𝑟𝑎𝑛𝑑𝑆𝑒𝑎𝑟𝑐ℎ&, + 𝛽jh 𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ&, + 𝛽jh 𝐷𝑖𝑟𝑒𝑐𝑡𝑉𝑖𝑠𝑖𝑡&,
r s
+ 𝛽jh 𝑂𝑟𝑔𝑎𝑛𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ&, + 𝛽jh ; 𝑗 , 𝑘 = 2,3 . (3)
Here, 𝐷𝑖𝑠𝑝𝑙𝑎𝑦&, captures the frequency of display advertising exposures for consumer 𝑖 in period t (i.e.,
viewable display advertising impressions) and BrandSearch•€ captures the frequency of search engine paid
clicks for brand-related keywords for consumer 𝑖 in period 𝑡 while GenericSearch•€ captures the frequency
of search engine paid clicks for generic-related keywords (i.e., keywords that do not contain the advertisers’
l o p
brand name) for consumer 𝑖 in time period 𝑡. The parameter vectors 𝛽jh , 𝛽jh and 𝛽jh capture the effect of
various types of digital advertising (i.e., display advertising, branded search clicks, and generic search clicks
respectively) towards transitioning customers from state 𝑗 to state 𝑘. Additionally, 𝐷𝑖𝑟𝑒𝑐𝑡𝑉𝑖𝑠𝑖𝑡&, captures
the frequency of direct website visits to the advertiser’s website for individual 𝑖 at time period 𝑡 while the
q
corresponding parameter vector 𝛽jh measures the impact of these direct visits towards transitioning
customers from state 𝑗 to state 𝑘. In a similar vein, 𝑂𝑟𝑔𝑎𝑛𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ&, captures the frequency of search engine
r
organic clicks (i.e., non-paid clicks) for consumer 𝑖 in period 𝑡 and the parameter vector 𝛽jh captures the
s
effect of such actions towards consumers’ transition from state 𝑗 to state 𝑘. Finally, 𝛽jh is the constant term
that captures the baseline probability of consumers’ transition from state 𝑗 to state 𝑘. Overall, the transition
probability from state 𝑗 to state 𝑘 is determined as follows:
exp 𝑃𝑆𝑇𝑖𝑡𝑗𝑘
𝑞𝑖𝑡 𝑗, 𝑘 = ; 𝑗, 𝑘 = 2,3 , 𝑗 ≠ 𝑘.
1 + exp 𝑃𝑆𝑇𝑖𝑡𝑗𝑘
Regarding the consumers’ transition to the Annoyance state, we allow firm-initiated marketing interactions
to operate as a mechanism of annoyance. In particular, the probability of transition to the Annoyance state
is a function of the frequency of exposures of viewable display advertisements at the consumer level during
time period 𝑡. Hence, the transition probability to the Annoyance state (state 𝑠=1) is defined as following:
l 0
exp (𝛽𝑗1 𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑖𝑡 + 𝛽𝑗1 )
𝑞𝑖𝑡 (𝑗, 𝑘 = 1) = 𝐴 0 ; 𝑗 = 2,3 .
1 + exp (𝛽𝑗1 𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑖𝑡 + 𝛽𝑗1 )
Finally, consumers who reside in the Annoyance state can probabilistically move back to the states of the
funnel path as follows:
0
exp (𝛽1𝑘 )
𝑞𝑖𝑡 (𝑗 = 1, 𝑘) = 0 ; 𝑘 = 2,3 .
1 + exp ( 𝑘 𝛽1𝑘 )

Thirty Eighth International Conference on Information Systems, South Korea 2017 9


Advertising Effectiveness and Annoyance Dynamics

State-dependent Consumer Behavior and Likelihood of Purchase


Consumers’ search behavior, their responsiveness to display-advertising exposures, and their likelihood of
conducting a purchase are contingent on the current state of the funnel path in which consumers reside. As
consumers move further down in the funnel path, they are more likely to actively engage in information-
gathering behaviors. For every consumer, we observe the multivariate outcome of information-gathering,
advertising-response, and purchase-likelihood behavior denoted as 𝑂&, = {𝐶&, , 𝐷&, , 𝑅&, , 𝐵&, , 𝐺&, , 𝑉&, }.
We employ a binary logit model in order to model the consumers’ decision to make a purchase and we allow
the purchase decision to be conditional on the state of the augmented funnel path the consumers resides.
Hence, the probability of an online conversion is:
exp (𝛾j )
𝑃 𝐶&, = 1 𝑆&, = 𝑗 = ; 𝑗 = 1,2,3 .
1 + exp (𝛾j )
As shown in the aforementioned equation, the consumers are allowed to have a different propensity to
purchase depending on the state 𝑗 they are located since the constant term is a state-specific coefficient.
We model both the consumers’ decision to gather information across various channels, as they progress
through the funnel path, and the consumers’ responsiveness to the firm-initiated display advertisements as
state-dependent behaviors. The following dependent variables that capture the information-seeking
behavior of users through different information channels are drawn from Poisson distributions with
respective state-specific rate parameter vector 𝝀 = {𝜆qˆ , 𝜆rˆ , 𝜆oˆ , 𝜆pˆ } . Respectively, the variable that
captures the average duration of exposure of the consumer to the display advertisement(s) is drawn from
an Exponential distribution with state-specific parameter 𝜇Šˆ . Hence:

• -‘Œ
‹ŒŽ• • ˆ
ˆ
𝑃 𝐷&, = 𝑑&, 𝑆&, = 𝑗 = ,
’(“Ž• M')

• -‘”
‹”Ž• • ˆ
ˆ
𝑃 𝑅&, = 𝑟&, 𝑆&, = 𝑗 = ,
’(–Ž• M')

˜ -‘—
‹—Ž• • ˆ ; 𝑗 = {1,2,3}.
ˆ
𝑃 𝐵&, = 𝑏&, 𝑆&, = 𝑗 = ,
’ ™Ž• M'

› -‘š
‹šŽ• • ˆ
ˆ
𝑃 𝐺&, = 𝑔&, 𝑆&, = 𝑗 = ,
’(œŽ• M')

𝑃 𝑉&, = 𝑣&, 𝑆&, = 𝑗 = 𝜇Šˆ exp 𝜇Šˆ 𝑣&, .

Identification
To ensure the model is fully identifiable with the data set, we conduct the following simulation. Specifically,
we simulate a data set of display-advertising exposures and consumers’ search as well as purchase actions
with specific parameter values so that the generated data set resembles the actual observed data set. Then,
we estimated the proposed model with the simulated data set and found we are able to recover the true
parameters. This relieves a potential concern about the empirical identification of the model.
Furthermore, to ensure identification of the states, the consumer-action probabilities are non-decreasing
for the latent states, similar to all the papers that have employed HMM to study consumer behavior (e.g.,
(Abhishek et al. 2012; Netzer et al. 2008)). The following paragraphs illustrate the identifiability constraints
and discuss in detail the set of consumer behavior theories that inform and guide these constraints in our
empirical setting.

Thirty Eighth International Conference on Information Systems, South Korea 2017 10


Advertising Effectiveness and Annoyance Dynamics

As far as consumers’ active search behavior is concerned, consumers move throughout the discrete stages
of the funnel path and more actively engage in a deliberation process along the conversion funnel path. That
is, as consumers move forward in the funnel path, they are more likely to conduct more research about the
product and the brand (Bettman et al. 1998; Browne et al. 2007; Howard and Sheth 1969; Lancaster and
Withey 2006; Tam and Ho 2006) in order to reduce their utility uncertainty associated with the purchase
under consideration. This consumer uncertainty pertaining to the surplus associated with the alternatives
is especially high for experience and credence or experience goods (Chen et al. 2004; Nelson 1974). The un-
certainty governing the purchase decision pertains, for instance, to the existence of certain product features
the importance of such features, and the knowledge regarding which alternatives have which features
(Brynjolfsson and Smith 2000; Urbany et al. 1989). As users move further down in the funnel path, they
engage more in information-seeking activities aimed at resolving such uncertainties. At the same time,
users who reside in the Annoyance state exhibit the lowest probability of engaging in information-
acquisition activities, because annoyance triggered by persistent advertisements negatively affects
consumers’ brand evaluation, persuasion, and ultimately interest in the advertised product (Fennis and
Bakker 2001; Morimoto and Chang 2006). Therefore, the restriction 𝜆' ≤ 𝜆( ≤ 𝜆) is implemented as
follows:
𝑠
𝜆𝑠 = 𝜆1 + exp (𝜆𝑠' ) ; 𝑠 = 2,3 .
𝑠′=2
Similarly, we restrict the probability of making an online purchase to be non-decreasing for the states. This
restriction 𝛾' ≤ 𝛾( ≤ 𝛾) is implemented as follows:
𝑠
𝛾𝑠 = 𝛾1 + exp (𝛾𝑠' ) ; 𝑠 = 2,3 .
𝑠′=2
As far as the state of annoyance is concerned, persistent advertising exposure can trigger annoyance in
consumers because advertising intervenes with the users’ main goal of completing a task online or simply
navigating the Internet (Bauer and Greyser 1968; Burke et al. 2005; Goldstein et al. 2014; Jenkins et al.
2016; Morimoto and Chang 2006). As a result of the perceived goal impediment, consumers who have been
frequently exposed to persistent advertising exposures cope with the negative effects of annoyance by
actively avoiding advertisements and minimizing the duration of their actual exposure to them (Goldstein
et al. 2014; Wilbur 2015; Woltman Elpers et al. 2003) to eliminate the source of noise or nuisance (Elliott
and Speck 1998). For instance, in the context of traditional TV advertising, annoyed consumers exhibit
similar behaviors. In particular, they often leave the room or switch TV channels to actively and physically
avoid annoying advertising (Abernethy 1991). In a similar vein, in the context of digital advertising, annoyed
consumers would either move to another part of the webpage to avoid being directly exposed to display ads
that elicit annoyance or click away from webpages that contain the annoying ads. Fundamentally, these
annoyed consumers’ reactions result in reducing the time consumers are exposed to the display
advertisements (i.e., reducing the in-view time exposure for a viewable display impression). Therefore, the
restriction pertaining to the average duration of the user’s exposure to display advertisements 𝜇' ≤ 𝜇( ≤
𝜇) is as follows:
𝑠
𝜇𝑠 = 𝜇1 + exp (𝜇𝑠' ) ; 𝑠 = 2,3 .
𝑠′=2
We use maximum likelihood estimation (MLE) to estimate the parameters of the HMM: we use the joint
likelihood function to estimate the transition matrix parameters and the state-dependent consumer-action
parameters. We use the sequential BFGS Newton-Raphson algorithm (LeSage 2005) to maximize the
likelihood given equation (1). We estimate the proposed HMM with the consumers initially residing in the
Awareness state for computational convenience purposes and parsimony of model parameters.
Nonetheless, we also verify the stability of the results by conducting the analysis with various randomly
selected parameter starting values and initial state distributions as discussed in the robustness-check
section.

Thirty Eighth International Conference on Information Systems, South Korea 2017 11


Advertising Effectiveness and Annoyance Dynamics

Empirical Results
In this section, we discuss the estimated parameters of the proposed HMM. In particular, Tables 4 and 3
show the parameters of the emission and transition variables of the proposed HMM model, respectively.
Based on the results presented in Table 3, we notice the parameters that capture the effect of advertising
exposures on transitioning the consumers across the various states are statistically significant. This finding
indicates advertising has an impact on transitioning consumers across the funnel path and, hence, increases
their propensity to make a purchase. In particular, display-advertising exposures increase consumers’
propensity to transition from the Awareness state to the Interest state, as revealed by the state-specific
l
transition coefficient of (𝛽() = 1.8755, p<0.01). By contrast, display-advertising exposures have a
statistically significant negative effect on the transition of the consumers from the Interest state to the
l
Awareness state, with a coefficient of (𝛽)( =-0.7162, p<0.1). Moreover, the findings reveal the coefficients
that capture the effect of display-advertising exposures on transitioning the consumers in the Annoyance
state are positive and statistically significant. Specifically, display-advertising exposures can have a
significant effect on transitioning consumers to the Annoyance state when they reside in the Awareness
l l
state (𝛽(' =6.7033, p<0.01) as well as when they reside in the Interest state (𝛽)' =2.2499, p<0.01).
Overall, these findings reveal a fundamental tension in digital display advertising between generating
interest and stimulating annoyance in consumers; that is, display advertising assists in transitioning
consumers towards the late stages of the funnel path and, at the same time, display-advertising exposures
beyond a certain frequency threshold within the same day can stimulate annoyance and transition
consumers to the Annoyance state. Besides, the results also highlight that technology-enabled advertising
scheduling (i.e., the frequency of individual-level exposures) can indeed trigger consumers’ annoyance. We
further elaborate on the important trade-off between effective and annoying advertising in the following
paragraphs when we discuss the estimated transition probabilities across the states of the funnel path as a
function of display-advertising exposures. Furthermore, our results effectively demonstrate consumers’
heterogeneous responses with respect to annoyance elicitation. The results unveil that the magnitude of the
annoyance effect is different across the stages of the funnel path in which consumers reside, revealing
interesting structural dynamics. Moreover, as previously discussed, drawing on existing theories it is not
clear whether a consumer who has moved further down in the purchasing funnel path will be more or less
easily annoyed with frequent display advertising exposures compared to a consumer who resides in the
early stages of the funnel path. For instance, one could postulate that consumers who reside in the later
stages of the funnel path might be less easily annoyed because they have developed more favorable attitudes
toward the brand. On the other hand, one could also postulate that consumers who reside in the early stages
of the funnel path are less informed about the brand and, thus, they might be the ones who are less easily
annoyed with frequent display advertising exposures because such interactions can convey valuable
information to them. Interestingly, examining the structural dynamics of annoyance effects, our findings
reveal that consumers who reside in the later stages of the funnel path exhibit higher tolerance for
annoyance stimulation. We further elaborate on this unexpected finding in the following paragraphs.
Moreover, the results presented in Table 4, allow us to gain insights into how consumers behave across the
latent states of the funnel path. Based on these results, we notice that as consumers move further down in
the funnel path (i.e., forward transition in the funnel path), they are substantially more likely to make an
online purchase. In particular, the state-specific constant term of the respective purchase probabilities is
(𝛾' = -5.4032, p<0.01) for the Annoyance state, (𝛾( = -5.2812, p<0.01) for the Awareness state and (𝛾) = -
4.0945, p<0.05) for the Interest state.

Table 3. Hidden Markov Model (HMM) Results.

Variables Coefficients St. Error


q 0.4044 *** 0.0607
Direct Visit (Awareness, Interest) (𝛽() )
Variables
Transitio

q -0.6440 * 0.0738
Direct Visit (Interest, Awareness) (𝛽)( )
n

ž 1.3578 *** 0.1321


Organic Search Click (Awareness, Interest) (𝛽() )
ž -0.0609 0.1005
Organic Search Click (Interest, Awareness) (𝛽)( )

Thirty Eighth International Conference on Information Systems, South Korea 2017 12


Advertising Effectiveness and Annoyance Dynamics

o -0.0683 0.2580
Brand Search Click (Awareness, Interest) (𝛽() )
o -1.4208 *** 0.1097
Brand Search Click (Interest, Awareness) (𝛽)( )
p -2.1860 *** 0.1448
Non Brand Search Click (Awareness, Interest) (𝛽() )
p -1.0870 *** 0.1124
Non Brand Search Click (Interest, Awareness) (𝛽)( )
l 1.8755 *** 0.3450
Display Ads (Awareness, Interest) (𝛽() )
l -0.7162 *** 0.2195
Display Ads (Interest, Awareness) (𝛽)( )
l 6.7033 *** 0.0201
Display Ads (Awareness, Annoyance) (𝛽(' )
l 2.2499 *** 0.1848
Display Ads (Interest, Annoyance) (𝛽)' )
Log-Likelihood -352066.96
BIC Value -352220.27

Focusing on the consumer information-gathering actions, we observe that there exist differences in the
coefficients of these state-dependent actions across the three latent states. In other words, consumers tend
to exhibit different information-gathering behaviors when they reside in different stages of the funnel path.
In particular, consumers who reside in the Interest state are more likely to engage in active search behaviors
having a coefficient of direct visit of (𝜆qŸ = 4.4899, p<0.01), a coefficient of organic (non-paid) search click
of (𝜆rŸ = 2.8282, p<0.05), a coefficient of brand paid search click of (𝜆oŸ =1.1289, p<0.1) and lastly a
coefficient of generic paid search click (𝜆pŸ = 5.9042, p<0.01). However, consumers who reside in the
Awareness state, an earlier stage in the funnel path, rarely or only occasionally engage in such active search
behaviors; the coefficient of organic (non-paid) search click of (𝜆r = 0.0719, p<0.01), for brand paid search
click (𝜆o = 0.1010, p<0.01), for direct visit is (𝜆q = 1.9047, p<0.01) and lastly for non-brand paid search
clicks (𝜆p = 0.0492, p<0.01). At the same time, consumers who reside in the Annoyance state rarely
engage in such information gathering behaviors. In particular, for consumers who reside in the Annoyance
state, the coefficient of direct visit is (𝜆qŸ = 0.0031, p<0.01), for organic (non-paid) search click is (𝜆r¡ =
0.0037, p<0.01), for brand paid search click is (𝜆o¡ = 0.0031, p<0.01), and for non-brand paid search
clicks is (𝜆p¡ = 0.0033, p<0.01). Besides, another important distinction between the latent states is the
variation in the tendency of the consumers to control their exposures to firm-initiated display
advertisements. We notice that consumers who reside in the Annoyance state seek to minimize their
exposure to display advertisements, as captured by the coefficient of in-view exposure (𝜇Š¡ = 6.6200,
p<0.01), in contrast to the users who reside in the Awareness state (𝜇Š = 6.8324, p<0.01) or Interest state
(𝜇ŠŸ = 8.8004 , p<0.01). Overall, the estimates of Table 4 indicate that the likelihood of the user to engage
in information-gathering actions and the responsiveness of the users to the advertising exposures increases
as consumers move further down in the funnel path.
An important aspect of the proposed HMM model is the ability to examine the effect of time-varying
variables, such as display-advertising exposures, on the transition of the consumers across the latent states.
To gain a better understanding of the estimated transition coefficients, we estimate the impact of the
frequency of the display-advertising on the transition probabilities across the latent states. Figures 2a and
2b depict the transition probabilities for consumers who reside in the Awareness and Interest state,
respectively.

Thirty Eighth International Conference on Information Systems, South Korea 2017 13


Advertising Effectiveness and Annoyance Dynamics

(a) Awareness State (b) Interest State

Figure 2. Transition probabilities as a function of display-advertising exposures for consumers who reside in
(a) Awareness State and (b) Interest State.

Evaluating the impact of display advertising on the transition probabilities across the states, we notice that
exposing consumers to a display advertisement increases the likelihood that the consumer will move further
down the funnel path and exhibit a higher propensity to make an online purchase. In particular, a successful
exposure to one display advertisement (i.e., a viewable display-advertising impression) has a positive
impact on the likelihood that a consumer who resides in the Awareness state will move to the Interest state
(from 37.38% to 79.57%). As the frequency of display advertising increases from one to two exposures in
the same time period (i.e., day), the corresponding transition probability further increases without
increasing the probability of annoyance elicitation. However, when the frequency of display advertising
increases to three exposures within the same day, the probability of transitioning from the Awareness state
to the Interest state decreases to 50.26%. This decrease happens because, as Figure 2a shows, with three
display- advertising exposures, annoyance effects can be triggered for consumers who reside in the
Awareness state as the transition probability to the Annoyance state increases to 49.43%. The annoyance
effects become substantial for consumers who reside in the Awareness state when they are exposed to four
display-advertising exposures in the same period, because the probability of transition to Annoyance state
greatly increases, as Figure 2a shows. In other words, we notice that digital advertisers indeed face trade-
offs in their decisions regarding the optimal frequency of display-advertising exposures at the individual
level. On the one hand, display advertising increases the likelihood of transition to the Interest state,
improving consumers’ learning. On the other hand, display advertising, when it is over-utilized beyond a
certain threshold, can also increase the likelihood of transition to the Annoyance state.
By contrast, the marginal impact of display-adverting exposures and the trade-offs of display-advertising
exposures for consumers who already reside in the Interest state are substantially different. As Figure 2b
shows, a successful (i.e., viewable) display-advertising exposure increases to 63.63% the probability that a
consumer who resides in the Interest state will continue residing in the same state. We notice the marginal
impact of one display-advertising exposure is higher when consumers reside in the early stages of the funnel
path (i.e., higher for consumers who reside in the Awareness state rather than consumers who reside in the
Interest state). However, increasing the probability that consumers will continue residing in the Interest
state is still important because the Interest state is less “sticky” than the Awareness state. The probability
that the consumer will continue residing in the Interest state without any additional advertisements during
this time period is 55.61%. As Figure 2b shows, one display-advertising exposure increases the probability
that consumers will continue residing in the Interest state, whereas two display-advertising exposures in
the same day continue increasing the aforementioned probability without triggering any annoyance effects
to consumers. Similarly, three advertising exposures in the same day effectively increase to 77.31% the
probability of staying in the Interest state. We also notice that even with four or five display-advertising
exposures in the same time period, the upward trend continues. This unexpected pattern reveals that

Thirty Eighth International Conference on Information Systems, South Korea 2017 14


Advertising Effectiveness and Annoyance Dynamics

consumers in the Interest state, who have developed more favorable attitudes towards the brand, exhibit
higher tolerance for annoyance stimulation compared to consumers who reside in the Awareness state.
However, after this point, the advertisers should start considering the trade-offs between effective and
annoying advertising, because the annoyance effects start to kick in with six display-advertising exposures
within the same period. In particular, the probability of transition to the Annoyance state becomes
increasingly significant with six or more advertising exposures: six exposures increases the probability to
15.53%, seven increases it to 64.21%.

Figure 3. Probability of a consumer residing in a specific state


as a function of the number of display-advertising exposures.

Furthermore, Figure 3 depicts the probability of a consumer residing in a specific state as a function of the
number of display-advertising exposures. We notice the overall probability a consumer resides in the
Annoyance state becomes larger than the probability a consumer resides in the Awareness or Interest state
as the frequency of the advertising exposures within the same day increases beyond six display
advertisements. However, once consumers transition to the Annoyance state, as a result of persistent
display-advertising exposures within the same day, they are likely to stay in the Annoyance state because it
is a fairly “sticky” state. For instance, Figure 4a illustrates the probability to continue residing in the
Annoyance state after certain time periods once consumers have moved to the Annoyance state. Finally,
our model unveils that when consumers probabilistically transition from the Annoyance state, they are
more likely to move to the Awareness state rather than the Interest state.
Besides, Figure 4b depicts the overall probability of a consumer making a purchase as a function of the
number of display-advertising exposures. As Figure 4b illustrates, increasing the frequency of the
advertising exposures can increase consumers’ likelihood of making a purchase, but exceeding a threshold
of display-advertising exposures in the same time period can undermine advertisers’ goals.

Robustness Checks

To enhance the robustness of our analysis and evaluate the sensitivity of results to alternative specifications,
we run our analysis with alternative initial state distributions. For an HMM with a transition matrix that is
a function of time-varying covariates, the initial state distribution can also be specified as the stationary
distribution of the transition matrix by solving the equation 𝝅 = 𝝅 𝑄 under the constraint that g£K' 𝜋£ = 1
where 𝑄 is the transition matrix with the parameter estimates and all the covariates are set to the mean
value of the corresponding variables across all the individuals and time periods. We find that the estimated
transition probabilities are positive confirming that the transition matrix is aperiodic and irreducible
ensuring the existence and uniqueness of the stationary distribution. Besides, the estimated HMM with the

Thirty Eighth International Conference on Information Systems, South Korea 2017 15


Advertising Effectiveness and Annoyance Dynamics

initial state distribution fixed at 80% in Awareness state, 20% in Interest state and 0% in Annoyance state
yields very similar results.

(a) (b)

Figure 4. (a) Probability a consumer continues residing in the Annoyance state in the following time periods
given that she has already reached the state of Annoyance in time period t. and (b) Probability a consumer
makes a purchase as a function of the number of display-advertising exposures.

Additionally, we check the robustness of our results to different period-length specifications beyond the one
employed in section State Transition Matrix (i.e., day). In particular, we estimated the HMM model with
the alternative periods of half a day and two days, and the results are similar. Moreover, we also checked
the sensitivity of the HMM model in different constructions of the explanatory variables (i.e., presence in
current and/or previous time period), and our results remain similar.

Conclusions, Managerial Implications, and Limitations


The main contribution of this study is the investigation of annoyance effects in the context of display
advertising. The proposed framework and empirical model capture the trade-offs in digital advertising by
allowing persistent advertising exposures to not only have an enduring impact on consumers’ purchase
decisions, but also to potentially trigger annoyance in consumers. An important aspect of our paper’s
contribution is that it allows technology-enabled advertising strategies, beyond the typical advertisement
content, to stimulate annoyance in consumers. Essentially, this study extends the current literature on
annoyance effects by investigating whether excessive display-advertising exposures can trigger annoyance
in consumers. Our findings reveal an interesting tension between generating interest and triggering
annoyance in consumers. A second important contribution of this paper is the investigation of the structural
dynamics of effective and annoyance effects in the context of digital advertising, by allowing the
corresponding effects to be contingent on the latent state of the funnel path in which consumers reside.
Investigating the dynamics of the annoyance effects, we reveal that consumers who reside in different stages
of the funnel path indeed exhibit different thresholds toward annoyance stimulation. Specifically, we find
the threshold of annoyance in display-advertising exposures is at least two times higher for consumers who
reside in the Interest state than for those who reside in the Awareness state.
Our paper’s findings reveal important managerial implications for firms that employ digital advertising.
The ability of the proposed HMM to incorporate time-varying covariates in the transition probability matrix
allows advertisers to investigate the impact of marketing activities on consumers’ purchase decisions. First,
our findings reveal the interesting trade-off between effective and annoying display-advertising exposures,
and thus highlight the need for digital advertisers to adapt their online display-advertising scheduling, while
taking into consideration the probability of triggering annoyance and its detrimental effects on consumers’
likelihood of purchase. Such schedule adaptations are of paramount importance given the revealed

Thirty Eighth International Conference on Information Systems, South Korea 2017 16


Advertising Effectiveness and Annoyance Dynamics

dynamics of advertising effectiveness and annoyance effects. That is, advertisers must take into
consideration the annoyance effects because display advertising is more impactful at the early stages of the
funnel but the threshold for annoyance elicitation is lower at these stages of the funnel path. Depending on
which stage of the funnel path consumers reside in, advertisers can effectively adapt their media scheduling
of advertising exposures in order to stimulate interest without triggering annoyance. Besides, employing
the proposed model, researchers and practitioners can segment the customers into latent states and
examine the effect of various time-varying marketing stimuli on transitioning the consumers further down
in the funnel path. This segmentation provides an important tool to advertisers to predict consumers’
subsequent purchase behavior and devise effective marketing-media scheduling strategies to maximize the
probability that the consumers will move forward in the funnel path and increase their probability of
making an online purchase.
Furthermore, our findings that tension exists in display advertising between stimulating interest and
generating annoyance have significant managerial implications for publishers and the advertising network
affiliates. For instance, the findings reveal the need for publishers and advertising networks to facilitate the
tracking of advertising exposures at the consumer level and provide the appropriate mechanisms that would
allow advertisers to enforce a limit on the frequency of advertising exposures. Besides, given the importance
of the revealed structural dynamics in the advertising effectiveness and annoyance effects, our findings also
indicate it would be strategically meaningful for the publishers and advertising networks to track subtle
signals of consumer browsing behavior at a granular level. Such tracking would appropriate inferences
regarding the potential magnitude of the advertising effectiveness and threshold for annoyance elicitation.

Conclusions

In this paper, we use a novel digital-advertising data set to estimate a hidden Markov model (HMM) to
study the effective and annoying display-advertising effects and their corresponding dynamics. We propose
an HMM that offers deep insights into the drivers of the dynamics of digital-advertising exposures. Our
findings demonstrate display advertising has an enduring impact on transitioning consumers further down
in the funnel path. Studying the dynamics of the display-advertising effectiveness, we find these effects are
contingent on the latent state in which consumers reside; a successful display-advertising exposure has a
higher impact for consumers who reside in the early stages of the funnel path compared to consumers who
already reside in the Interest state. Nonetheless, the impact of a display-advertising exposure for consumers
who reside in the Interest state is still significant and non-negligible. The augmentation of the funnel path
with an Annoyance state allows us to study whether and when persistent display-advertising exposures
operate as a mechanism of annoyance for consumers. Interestingly, we find that trade-offs exist in digital
advertising, encapsulated in the tension between generating interest and triggering annoyance in
consumers. In particular, we find display-advertising exposures beyond a certain threshold in the same
period can substantially increase the probability of annoying the consumers. Finally, investigating the
dynamics of these annoyance effects, we reveal that consumers who reside in different stages of the funnel
path exhibit different thresholds toward exhibiting annoyance effects. This finding is an important aspect
that the proposed HMM allows advertisers to effectively capture. Specifically, we find the threshold of
annoyance in display-advertising exposures is much higher for consumers who reside in the Interest state
compared to the Awareness state. In addition, studying the annoyance effects has important managerial
implications for firms, as our estimations reveal the Annoyance state is a substantially “sticky” state calling
attention to the lingering negative effects of frequent display-advertising exposures in the same time period.

Limitations and Future Research Directions

Our paper has several limitations that arise primarily from the restrictions of the data set. For instance, in
the present study, we do not consider the effect of organic word-of-mouth interactions (Adamopoulos et al.
2015; Adamopoulos and Todri 2015) to which some consumers could possibly be exposed. Furthermore,
future research could investigate the potential impact of other marketing factors, beyond the frequency of
advertising exposures, in triggering annoyance in consumers. In the same vein, future research can also
examine how the annoyance effects triggered by persistent advertising exposures vary for different product
categories and different brands. Notwithstanding these limitations, we hope our paper paves the way for
more research in this increasingly important and emerging stream of work of annoyance in the context of
digital advertising.

Thirty Eighth International Conference on Information Systems, South Korea 2017 17


Advertising Effectiveness and Annoyance Dynamics

References
Aaker, D.A., and Bruzzone, D.E. 1985. "Causes of Irritation in Advertising," The journal of marketing
(49:2), pp 47-57.
Abernethy, A.M. 1991. "Physical and Mechanical Avoidance of Television Commercials: An Exploratory
Study of Zipping, Zapping and Leaving," Proceedings of the American Academy of Advertising: New
York: The American Academy of Advertising, pp. 223-231.
Abhishek, V., Fader, P., and Hosanagar, K. 2012. "The Long Road to Online Conversion: A Model of Multi-
Channel Attribution," Working paper, Carnegie Mellon University.
Adamopoulos, P. 2016. "Unexpectedness and Non-Obviousness in Recommendation Technologies and
Their Impact on Consumer Decision Making." Stern School of Business, New York.
Adamopoulos, P., Ghose, A., and Todri, V. 2015. "Estimating the Impact of User Personality Traits on Word-
of-Mouth: Text-Mining Microblogging Platforms." Available at SSRN:
https://ssrn.com/abstract=2679199.
Adamopoulos, P., and Todri, V. 2014. "Social Media Analytics: The Effectiveness of Promotional Events on
Brand User Base in Social Media," Proceedings of the 35th International Conference on Information
Systems, ICIS: AIS, p. 10.
Adamopoulos, P., and Todri, V. 2015. "The Effectiveness of Marketing Strategies in Social Media: Evidence
from Promotional Events," Proceedings of the 21th ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining: ACM, pp. 1641-1650.
Anand, B.N., and Shachar, R. 2009. "Targeted Advertising as a Signal," QME (7:3), pp 237-266.
Anderson, S.P., and De Palma, A. 2012. "Competition for Attention in the Information (Overload) Age," The
RAND Journal of Economics (43:1), pp 1-25.
Animesh, A., Ramachandran, V., and Viswanathan, S. 2010. "Research Note—Quality Uncertainty and the
Performance of Online Sponsored Search Markets: An Empirical Investigation," Information Systems
Research (21:1), pp 190-201.
Bagwell, K. 2007. "The Economic Analysis of Advertising," Handbook of industrial organization (3), pp
1701-1844.
Barry, T.E. 1987. "The Development of the Hierarchy of Effects: An Historical Perspective," Current issues
and Research in Advertising (10:1-2), pp 251-295.
Bauer, R.A., and Greyser, S.A. 1968. Advertising in America, the Consumer View. Division of Research:
Graduate School of Business Administration, Harvard University.
Belch, G.E. 1982. "The Effects of Television Commercial Repetition on Cognitive Response and Message
Acceptance," Journal of Consumer Research (9:1), pp 56-65.
Bettman, J.R., Luce, M.F., and Payne, J.W. 1998. "Constructive Consumer Choice Processes," Journal of
consumer research (25:3), pp 187-217.
Browne, G.J., Pitts, M.G., and Wetherbe, J.C. 2007. "Cognitive Stopping Rules for Terminating Information
Search in Online Tasks," MIS quarterly (39:1), pp 89-104.
Brynjolfsson, E., Hu, Y., and Smith, M.D. 2003. "Consumer Surplus in the Digital Economy: Estimating the
Value of Increased Product Variety at Online Booksellers," Management Science (49:11), pp 1580-1596.
Brynjolfsson, E., and Smith, M.D. 2000. "Frictionless Commerce? A Comparison of Internet and
Conventional Retailers," Management science (46:4), pp 563-585.
Burke, M., Hornof, A., Nilsen, E., and Gorman, N. 2005. "High-Cost Banner Blindness: Ads Increase
Perceived Workload, Hinder Visual Search, and Are Forgotten," ACM Transactions on Computer-
Human Interaction (TOCHI) (12:4), pp 423-445.
Cacioppo, J.T., and Petty, R.E. 1979. "Effects of Message Repetition and Position on Cognitive Response,
Recall, and Persuasion," Journal of personality and Social Psychology (37:1), p 97.
Campbell, M.C., and Keller, K.L. 2003. "Brand Familiarity and Advertising Repetition Effects," journal of
Consumer Research (30:2), pp 292-304.
Chen, P.-Y., and Hitt, L.M. 2002. "Measuring Switching Costs and the Determinants of Customer Retention
in Internet-Enabled Businesses: A Study of the Online Brokerage Industry," Information systems
research (13:3), pp 255-274.
Chen, P.-Y., Wu, S.-y., and Yoon, J. 2004. "The Impact of Online Recommendations and Consumer
Feedback on Sales," in: Proceedings of the 25th International Conference on Information Systems,
ICIS. AIS.

Thirty Eighth International Conference on Information Systems, South Korea 2017 18


Advertising Effectiveness and Annoyance Dynamics

Chen, Y., Berkhin, P., Anderson, B., and Devanur, N.R. 2011. "Real-Time Bidding Algorithms for
Performance-Based Display Ad Allocation," Proceedings of the 17th ACM SIGKDD international
conference on Knowledge discovery and data mining: ACM, pp. 1307-1315.
Cho, C.-H. 2004. "Why Do People Avoid Advertising on the Internet?," Journal of advertising (33:4), pp
89-97.
Elliott, M.T., and Speck, P.S. 1998. "Consumer Perceptions of Advertising Clutter and Its Impact across
Various Media," Journal of Advertising Research (38:1), pp 29-30.
eMarketer. 2016a. "Reasons for Using Ad Blockers According to Ad Blocking Users Worldwide, Q2 2016 (%
of Respondents) ".
eMarketer. 2016b. "Us Digital Display Ad Spending to Surpass Search Ad Spending in 2016,"
http://www.emarketer.com/.
Feng, J., and Xie, J. 2012. "Research Note—Performance-Based Advertising: Advertising as Signals of
Product Quality," Information Systems Research (23:3-part-2), pp 1030-1041.
Fennis, B.M., and Bakker, A.B. 2001. "“Stay Tuned—We Will Be Back Right after These Messages”: Need to
Evaluate Moderates the Transfer of Irritation in Advertising," Journal of Advertising (30:3), pp 15-25.
Ghose, A., and Todri, V. 2016. "Toward a Digital Attribution Model: Measuring the Impact of Display
Advertising on Online Consumer Behavior," MIS Quarterly (40:4), pp 889-910.
Ghose, A., and Yang, S. 2009. "An Empirical Analysis of Search Engine Advertising: Sponsored Search in
Electronic Markets," Management Science (55:10), pp 1605-1622.
Goldstein, D.G., Suri, S., McAfee, R.P., Ekstrand-Abueg, M., and Diaz, F. 2014. "The Economic and
Cognitive Costs of Annoying Display Advertisements," Journal of Marketing Research).
Heflin, D.T., and Haygood, R.C. 1985. "Effects of Scheduling on Retention of Advertising Messages,"
Journal of Advertising (14:2), pp 41-64.
Howard, J.A., and Sheth, J.N. 1969. The Theory of Buyer Behavior. Wiley New York.
Jenkins, J.L., Anderson, B.B., Vance, A., Kirwan, C.B., and Eargle, D. 2016. "More Harm Than Good? How
Messages That Interrupt Can Make Us Vulnerable," Information Systems Research (27:4), pp 880-
896.
Kokkodis, M., and Lappas, T. 2016. "Realizing the Activation Potential of Online Communities," in:
Proceedings of the 37th International Conference on Information Systems, AIS (ed.).
Lancaster, G., and Withey, F. 2006. Marketing Fundamentals. Routledge.
LeSage, J.P. 2005. "Econometrics Toolbox for Matlab," Internet site: http://www.spatial-
econometrics.com/ (Accessed April 28, 2005)).
Morimoto, M., and Chang, S. 2006. "Consumers’ Attitudes toward Unsolicited Commercial E-Mail and
Postal Direct Mail Marketing Methods: Intrusiveness, Perceived Loss of Control, and Irritation,"
Journal of Interactive Advertising (7:1), pp 1-11.
Naik, P.A., Mantrala, M.K., and Sawyer, A.G. 1998. "Planning Media Schedules in the Presence of Dynamic
Advertising Quality," Marketing Science (17:3), pp 214-235.
Nelson, P. 1970. "Information and Consumer Behavior," The Journal of Political Economy), pp 311-329.
Nelson, P. 1974. "Advertising as Information," The Journal of Political Economy), pp 729-754.
Netzer, O., Lattin, J.M., and Srinivasan, V. 2008. "A Hidden Markov Model of Customer Relationship
Dynamics," Marketing Science (27:2), pp 185-204.
Pokrywczynski, J., and Crowley, J.H. 1993. "The Influence of Irritating Commercials on Radio Listening
Habits," Journal of Radio Studies (2:1), pp 51-68.
Rabiner, L. 1989. "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition,"
Proceedings of the IEEE (77:2), pp 257-286.
Rabiner, L., and Juang, B.-H. 1986. "An Introduction to Hidden Markov Models," ASSP Magazine, IEEE
(3:1), pp 4-16.
Ryan, K.M., and Graham, R.S. 2014. "Digital Display Advertising," in: Taking Down Goliath. Springer, pp.
85-100.
Sahoo, N., Singh, P.V., and Mukhopadhyay, T. 2012. "A Hidden Markov Model for Collaborative Filtering,"
MIS Quarterly (36:4), pp 1329-1356.
Sawyer, A.G. 1981. Repetition, Cognitive Responses, and Persuasion.
Scott, M. 2016. "Rise of Ad-Blocking Software Threatens Online Revenue," in: New York Times.
Singh, P.V., Tan, Y., and Youn, N. 2011. "A Hidden Markov Model of Developer Learning Dynamics in Open
Source Software Projects," Information Systems Research (22:4), pp 790-807.
Spence, M. 1976. "Informational Aspects of Market Structure: An Introduction," The Quarterly Journal of
Economics (90:4), pp 591-597.

Thirty Eighth International Conference on Information Systems, South Korea 2017 19


Advertising Effectiveness and Annoyance Dynamics

Stigler, G.J. 1961. "The Economics of Information," The journal of political economy (69:3), pp 213-225.
Stiglitz, J.E. 1989. "Imperfect Information in the Product Market," Handbook of industrial organization
(1), pp 769-847.
Tam, K.Y., and Ho, S.Y. 2006. "Understanding the Impact of Web Personalization on User Information
Processing and Decision Outcomes," MIS quarterly (30:4), pp 865-890.
Tan, B., Yi, C., and Chan, H.C. 2015. "Research Note—Deliberation without Attention: The Latent Benefits
of Distracting Website Features for Online Purchase Decisions," Information Systems Research (26:2),
pp 437-455.
Tellis, G.J. 2003. Effective Advertising: Understanding When, How, and Why Advertising Works. Sage
Publications.
Todri, V., and Adamopoulos, P. 2014. "Social Commerce: An Empirical Examination of the Antecedents
and Consequences of Commerce in Social Network Platforms," in: Proceedings of the 35th
International Conference on Information Systems. AIS.
Urbany, J.E., Dickson, P.R., and Wilkie, W.L. 1989. "Buyer Uncertainty and Information Search," Journal
of consumer research (16:2), pp 208-215.
Wilbur, K.C. 2015. Recent Developments in Mass Media: Digitization and Multitasking. Elsevier.
Williams, B. 2016. "Adblock Plus and (a Little) More," in: Decadblock: Adblock Plus turns 10.
https://adblockplus.org/ Adblock Plus.
Woltman Elpers, J.L., Wedel, M., and Pieters, R.G. 2003. "Why Do Consumers Stop Viewing Television
Commercials? Two Experiments on the Influence of Moment-to-Moment Entertainment and
Information Value," Journal of Marketing Research (40:4), pp 437-453.
Yun Yoo, C., and Kim, K. 2005. "Processing of Animation in Online Banner Advertising: The Roles of
Cognitive and Emotional Responses," Journal of Interactive Marketing (19:4), pp 18-34.

Thirty Eighth International Conference on Information Systems, South Korea 2017 20

You might also like