Predicting Advertising Success Beyond

VINOD VENKATRAMAN, ANGELIKA DIMOKA, PAUL A.
PAVLOU, KHOI VO,

WILLIAM HAMPTON, BRYAN BOLLINGER, HAL E. HERSHFIELD,
MASAKAZU ISHIHARA, and RUSSELL S. WINER*
In the past decade, there has been a tremendous increase in the use of
neurophysiological methods to better understand marketing phenomena
among academics and practitioners. However, the value of these
methods in predicting advertising success remains underresearched.
Using a unique experimental protocol to assess responses to 30-second
television ads, the authors capture many measures of advertising
effectiveness across six commonly used methods (traditional self-reports,
implicit measures, eye tracking, biometrics, electroencephalography, and
functional magnetic resonance imaging). These measures have been
shown to reliably tap into higher-level constructs commonly used in
advertising research: attention, affect, memory, and desirability. Using time-
series data on sales and gross rating points, the authors attempt to relate
individual-level response to television ads in the lab to the ads’
aggregate, market-level elasticities. The authors show that functional
magnetic resonance imaging measures explain the most variance in
advertising elasticities beyond the baseline traditional measures. Notably,
activity in the ventral striatum is the strongest predictor of real-world,
market-level response to advertising. The authors discuss the findings and
their significant implications for theory, research, and practice.
Keywords: advertising elasticities, neuroscience, biometrics, implicit

measures, market response modeling
Online Supplement: http://dx.doi.org/10.1509/jmr.13.0593
Predicting Advertising Success Beyond

Traditional Measures: New Insights from
Neurophysiological Methods and Market
Response Modeling
*Vinod Venkatraman is Assistant Professor of Marketing (e-mail: [email protected]), Angelika Dimoka is Associate Professor of Marketing
(e-mail: [email protected]), Paul A. Pavlou is Milton F. Stauffer Professor of Information Technology and Strategy (e-mail: [email protected]), Khoi Vo
is Senior Research Associate (e-mail: [email protected]), and William Hampton is Research Assistant (e-mail: [email protected]), Center for
Neural Decision Making, Fox School of Business, Temple University. Bryan Bollinger is Assistant Professor of Marketing, Fuqua School of Business, Duke
University (e-mail: [email protected]). Hal E. Hershfield is Assistant Professor of Marketing, Anderson School of Management, University of Cali-
fornia, Los Angeles (e-mail: [email protected]). Masakazu Ishihara is Assistant Professor of Marketing (e-mail: [email protected]),
and Russell S. Winer is William H. Joyce Professor of Marketing (e-mail: [email protected]), Stern School of Business, New York University. The
authors are grateful for funding support from the Advertising Research Foundation (ARF) and the corporate sponsors of this study. They thank Horst Stipp
from ARF and Jim Thompson from Temple University for their support and guidance throughout the project, Dave Walker from IPSOS-ASI for assistance
with the protocol design for the traditional surveys, Jessica Fleck (Stockton College) for assistance with collection of EEG data, and Rene San Martin (Duke
University) for assistance with the EEG analyses.
© 2015, American Marketing Association Journal of Marketing Research

ISSN: 0022-2437 (print), 1547-7193 (electronic) 436 Vol. LII (August 2015), 436–452
Predicting Advertising Success Beyond Traditional Measures 437
Companies spend millions of dollars each year striving neurophysiological measures and market sales. Specifically,
for advertising success. Advertising spending occurs for Berns and Moore (2012) use fMRI data to predict music
pretesting copy alternatives while a campaign is being popularity by measuring the brain activity of 27 adolescents
developed as well as for in-market analyses after the cam- when listening to 15-second song clips. The authors use
paign is launched. Many methods have been developed to three-year data from Nielsen SoundScan to show a signifi-
pretest ads, ranging from self-reported “traditional” mea- cant link between brain activity (ventral striatum) and sales.
sures such as recall, liking, and purchase intent to neuro- A common question practitioners ask is whether neuro-
physiological measures such as functional magnetic reso- physiological methods are “valuable”—that is, do they con-
nance imaging (fMRI). Furthermore, sophisticated statistical tribute anything beyond traditional methods in predicting ad
approaches, referred to as marketing-mix modeling, have success? The current study is an attempt to address this
also been used to evaluate ex post the impact of advertising question. Our objectives are twofold: First, we explore how
spending across multimedia. measures from commonly used neurophysiological methods
Marketing textbooks draw a distinction between rational tap into higher-level constructs commonly used in advertis-
and emotional advertising (Batra, Myers, and Aaker 1996), ing research (attention, affect, memory, and desirability).
wherein the former refers to advertising of factual or con- Specifically, we aim to study the relationships among neu-
scious information and the latter refers to advertising that rophysiological measures as well as their relationship to tra-
targets unconscious and emotional processes. Marketers ditional measures. Second, we aim to explain the variance
have relied on various approaches to measure rational pro- in real-life advertising success (captured using market
cesses for decades (e.g., Lucas and Britt 1963). These mea- response models) using the various neurophysiological
sures include recognition, recall, liking, and persuasion. measures relative to traditional advertising measures.
Marketers have also attempted to measure unconscious The outline of this article is as follows. We first provide a
automatic reactions to advertising. As Stewart (1984) notes, brief literature review on advertising research and introduce
popular methods have included several physiological the four key constructs in advertising research. Second, we
approaches such as pupillary response, heart rate, eye provide an overview of the proposed neurophysiological
movements, voice pitch analysis, and neuroimaging, all of methods and measures and their relationship to the four
which are commonly referred to as neurophysiological advertising constructs. Third, we describe the research
methods. design and experimental protocol and discuss the empirical
The past decade has experienced an explosion of research relationships among all our measures. We then describe
in neuroscience and the use of multiple neurophysiological how we estimated the elasticities for the 30-second televi-
methods to study marketing, consumer behavior, and adver- sion ads, followed by the empirical analysis linking all mea-
tising phenomena, broadly referred to as consumer neuro- sures to ad elasticities. Finally, we discuss the study’s con-
science or neuromarketing (e.g., Ariely and Berns 2010; tributions and implications for researchers and practitioners.
Camerer, Loewenstein, and Prelec 2005; Dimoka 2012;
Smidts et al. 2014; Venkatraman et al. 2012; Yoon et al. BACKGROUND, LITERATURE REVIEW, AND THEORY
2012). A new industry has also been built around neuromar- DEVELOPMENT
keting tools in the past decade.1 This growth is due to a We review the key constructs examined in advertising
combination of technological advances in fMRI, electroen- research by discussing how they have been assessed using
cephalography (EEG), eye tracking, and other neurophysio- traditional self-reported measures. We then introduce the
logical tools with increased accessibility of these methods proposed neurophysiological methods and discuss how they
due to decreased administration costs (Dimoka, Pavlou, and can capture traditional advertising constructs more directly
Davis 2011). and objectively.
Although there has been considerable research in both
academia and industry using neurophysiological measures Constructs in Advertising Research
to better understand consumer responses to advertising Advertising research has long relied on self-reported
(Ohme et al. 2010; Stipp and Woodard 2011), it is important measures (Biel and Bridgwater 1990; Du Plessis 1994;
to examine whether these measures actually translate into Haley and Baldinger 2000; Poels and Dewitte 2006; Smit,
real-life advertising success. The overall goal of this article Van Meurs, and Neijens 2006; Walker and Dubitsky 1994).
is to link traditional and neurophysiological measures to Researchers and practitioners have used traditional methods
actual market responses to advertising in terms of advertis- of copy testing, such as focus groups and surveys, to collect
ing elasticities.2 Although, to our knowledge, no research responses toward ads. These measures are inexpensive,
has directly linked neurophysiological measures to advertis- accessible, quick, and relatively simple to analyze; in addi-
ing elasticities, one study has found a relationship between tion, they offer insight into consumer brand attitudes and
preferences as well as into how different ad executions
1The neuromarketing–advertising practice link received a boost through
moderate these responses. The common traditional mea-
the Advertising Research Foundation’s (ARF’s) “Neuro 1.0” project (Stipp sures can be broadly classified into measures focused on (1)
and Woodard 2011). The goal of the project was to “help the (advertising) ad execution (e.g., liking, excitability, recall) and (2) the
industry learn how best to apply the capabilities of neuromarketing to real product featured in the ad (e.g., attitudes, purchase intent),
marketing issues and decisions” (Stipp and Woodard 2011, p. 5). and both have been used to explain advertising success.
2The impetus for this study came from the follow-up project to Neuro
1.0. The project, called “Neuro 2.0,” was initiated by ARF and underwrit-
The early AIDA (attention, interest, desire, action) model
ten by six corporate sponsors that provided both ads for testing and time- argues that every advertising process begins with capturing
series data for estimating elasticities. attention, followed by information assimilation and compre-
438 JOURNAL OF MARKETING RESEARCH, AUGUST 2015
hension, which leads to desirability, followed by action often attributed to better ad processing, these measures do
(Strong 1925). This and other “hierarchy of effects” models not necessarily distinguish between processing due to
have been the backbone of advertising research for the past encoding (when the ad is presented) and retrieval (when the
few decades (Barry and Howard 1990). Recent research has recognition test is performed).
extended this notion of hierarchy (temporal sequence) by Desirability. In traditional advertising research, desirabil-
shifting focus to a set of core constructs: attention, affect, ity refers to the extent to which people desire the product
memory, and desirability. These core constructs can affect featured in the ad. Marketing managers routinely use mea-
advertising success independently or in combination (Haley sures such as purchase intent as a strong correlate of desir-
and Baldinger 2000; Morwitz, Steckel, and Gupta 2007; ability and subsequent market behavior. To quantify the spe-
Walker and Dubitsky 1994). We begin with a review of cific effects of an ad, researchers measure purchase intent as
these four constructs and discuss how they are assessed a change in the level of desirability for the product pre- and
using traditional measures. post-exposure to the ad. This shift measure for purchase
Attention. Attention is defined as the ability to focus on intent is often weakened by the varying amounts of brand
certain aspects of the environment while ignoring others. equity. Because popular brands tend to have higher premea-
Advertising researchers often refer to attention as the ability sure scores, it is important to account for this bias before
to attract focus to an ad. Common measures of attention are making any judgments about shifts in desirability. Walker
liking, informativeness, excitability, and relevancy of the ad and Dubitsky (1994) propose a method by which change
(Biel and Bridgwater 1990; Brown and Stayman 1992; scores are normalized by using a baseline predicted average
Schlinger 1979; Smit, Van Meurs, and Neijens 2006). Newer result (PAR) score to remove brand-specific effects not
methods, such as eye tracking, provide more direct mea- associated with ad exposure. Yet broader concerns still
sures of attention, as we discuss subsequently. Specifically, remain about the relationship between these intent measures
researchers make a distinction between endogenous (“top- and subsequent purchasing (Morwitz, Steckel, and Gupta
down”) attention, in which specific aspects of the ad are 2007). Consumers are not capable of perfectly predicting
explicitly selected and processed, and exogenous (“bottom- the future, either in terms of how they represent their inten-
up”) attention, in which features of the stimulus attract tions or how these intentions will change over time. The
attention and processing. We argue that such distinctions, strength of the predictability also depends on the context,
while critical, cannot be captured using traditional self- novelty, and specificity of the products concerned (Morwitz
reported measures. and Fitzsimons 2004). Therefore, we contend that the extent
Affect. Emotion refers to a relatively brief episode of of reward-related activation in the brain during the actual ad
coordinated brain, physiological, and behavioral changes provides a better and more direct measure of desirability, as
that facilitate a response to an external or internal event of we describe next.
significance (Davidson, Scherer, and Goldsmith 2009).
Affect, though often used as a synonym for emotion, refers Newer Methods in Advertising Research
to the outward expression of an emotion. We contend that In the past decade, there has been a burgeoning use of
affect, in the context of advertising, can be broadly classi- neurophysiological methods to understand consumer behav-
fied into two dimensions: valence (relative pleasantness/ ior. In this section, we provide a brief introduction to the
unpleasantness) and arousal (physiological and subjective methods used in this study. Other sources are available for
intensity). In the AIDA model, emotions and affect were more detailed reviews of each method (e.g., Dimoka 2012;
considered merely a means for attracting attention. Thus, Huettel, Song, and McCarthy 2008; Potter and Bolls 2012;
they have often been inferred with self-reported measures Shaw 2003; Wedel and Pieters 2008).
such as liking and excitability (Poels and Dewitte 2006; Implicit measures. Despite their popularity, self-reported
Walker and Dubitsky 1994). These measures represent post measures are inherently subjective and incomplete because
hoc introspection about affect experienced from an earlier they only capture conscious, declared opinions (Micu and
stimulus and thus could be distorted by a variety of factors, Plummer 2010). As a result, implicit testing has emerged as
including higher cognitive processes. Neurophysiological an alternative to capture the unconscious nature of con-
methods, in contrast, provide a more direct measure of sumer preferences, attitudes, and information processing
affect, as we discuss subsequently. (Greenwald, McGhee, and Schwartz 1998). The Implicit
Memory. Memory refers to the mechanisms by which Association Task (IAT) is a commonly used measure that
past experiences influence behavior. Therefore, memory is captures the strength of association among concepts and
often associated with encoding (which occurs during the avoids tapping into the consumer’s conscious thought.
past event), consolidation (which occurs during the inter- Specifically, differences in response latencies for brands
vening period), and retrieval (which occurs at a future time). paired with positive and negative words in IAT have been
Retrieval success is often used as a proxy for the depth to used as a measure of emotional valence (Dimofte 2010).
which information was encoded (Mandler 1980). Advertis- Eye tracking. Next to traditional and implicit measures,
ing research, like most memory research, has focused on the eye tracking is perhaps the most accessible method for cap-
retrieval aspects to evaluate the quality of ads. The empha- turing ad response. Eye tracking has a high temporal resolu-
sis has been on two retrieval measures in particular: recall, tion (60–120 Hz) and provides insight into temporal pro-
in which participants generate the target with partial or no cesses. Compared with old camera-based systems (with
cues, and recognition, in which participants distinguish the chin rest and head straps), modern eye trackers use an opti-
targets from novel distractors (Du Plessis 1994; Singh, cal camera to identify the position of the pupil and cornea
Rothschild, and Churchill 1988). Although better memory is using infrared/near-infrared light that evokes corneal reflec-
tion. By tracking participants’ gaze when viewing ads, we eccrine (sweat) glands following exposure to certain stimuli
can capture not only which information was processed but (Potter and Bolls 2012). Skin conductance is frequently
also the order and duration of these processes. Eye tracking used as a tool to measure tonic activity of the SNS. Due to
has been used as a direct measure of attention. For example, the nature of physiological responses, SCR is also preceded
bottom-up factors, such as color and luminance, have a by a small latency (delay). Skin conductance amplitude and
strong effect on initial eye movements (Leven 1991). In response latency provide direct measures of arousal when
addition, the percentage of valid fixations (total amount of watching an ad, unlike self-reported measures, which are
time eyes are focused on the ad) provides an index of over- often based on introspection at a later time. Still, SCR can-
all attention or engagement with the ad (relative to distrac- not reliably indicate emotional valence (Potter and Bolls
tions). The number of fixations and mean dwell times pro- 2012).
vide a measure of the depth to which information within an EEG. Perhaps the most commonly used neuroscience
ad is processed (Venkatraman, Payne, and Huettel 2014). method in advertising research (Wang and Minor 2008),
Longer dwell times and fewer fixations represent more EEG can reveal variations in electrical signals of cortical
detailed processing (Horstmann, Ahlgrimm, and Glockner brain regions as a function of internal or external variables.
2009). Finally, eye tracking can also measure pupil dilation3 These variations are recorded at different frequencies—
(physiological response of the sympathetic nervous system), delta rhythms (<4 Hz), theta rhythms (4–7 Hz), alpha
which provides additional insight into the degree of arousal rhythms (8–12 Hz), and beta rhythms (15–30 Hz)—and cor-
following an external stimulus (Hess and Polt 1960). respond to different physiological phenomena. Electroen-
Biometrics. Biometrics refers to the physiological or cephalography provides high temporal resolution but low
automatic responses to an external stimulus. Biometrics has spatial resolution because it is restricted to measuring only
become increasingly popular in marketing and advertising cortical brain activity. Here, we focus primarily on the alpha
research because they can provide insight into unconscious frequency band, which is inherently inhibitory and thus
processes and affect (Potter and Bolls 2012). Common inversely related to underlying brain activity (Jensen and
physiological responses include heart rate, breathing, and Mazaheri 2010; Shaw 2003). Specifically, we focus on two
skin conductance. measures: occipital alpha activity and frontal asymmetry.
Heart rate, also called pulse, is the speed of the heartbeat The occipital alpha measures the extent of activation/gating
and is typically measured with an electrocardiogram, which
in the visual system and thus provides an index of visual
measures the electrical activity of the heart using external
processing and exogenous attention (Foxe and Snyder 2011;
skin electrodes. Heart rate is controlled by two antagonistic
Jensen and Mazaheri 2010). We predicted that the more
systems: the sympathetic nervous system (SNS) and the
effective ads would have reduced occipital alpha. Similarly,
parasympathetic nervous system (PNS) (Potter and Bolls
the relationship between affect and hemispheric asymme-
2012). The SNS (termed the “fight-or-flight system”) repre-
sents the body’s automatic response to external stimulus. tries in frontal brain activity has a long history in psychol-
Activation of this system increases heart rate, also called ogy and neuroscience (Davidson 2004; Demaree et al. 2005;
heart rate acceleration, which provides an independent mea- Harmon-Jones, Gable, and Peterson 2010). This frontal
sure of arousal (Wang, Lang, and Busemeyer 2011). Con- asymmetry measure (ln[F4] – ln[F3]) argues for greater
versely, the PNS (termed the “rest-and-digest system”) responses in the alpha band frequencies for positive stimuli
refers to a calm and relaxed state that is characterized by in the left hemisphere (F3) and negative stimuli in the right
slower heart rate, or heart rate deceleration. Increased heart hemisphere (Davidson et al. 1990; Tomarken, Davidson,
rate deceleration in response to an ad implies increased abil- and Henriques 1990). However, others have argued that
ity to focus on the ad and thus provides an independent greater activation in the left hemisphere (smaller alpha)
measure of attention (Lang et al. 1999). merely reflects approach motivation, independent of emo-
Breathing frequency, or respiration rate, refers to the tional valence (Harmon-Jones et al. 2006; Sutton and
number of breaths taken within a fixed amount of time, typi- Davidson 1997). In this study, we expect that the more
cally 60 seconds, to yield a breaths-per-minute (BPM) mea- effective ads will be associated with higher values of frontal
sure. Activation of the SNS leads to an increase in respira- asymmetry (approach behavior).
tion rate, which can then be used as a measure of arousal. fMRI. Functional magnetic resonance imaging is a non-
Breathing also influences the SNS/PNS by temporarily invasive method that localizes and tracks changes in blood
blocking PNS influence on heart rate, resulting in increased oxygenation during cognitive tasks (Ogawa et al. 1990). The
heart rate, but subsequent exhalation removes this block and blood oxygenation level–dependent contrast is based on the
decreases heart rate. The undulation in heart rate caused by fact that hemoglobin has different magnetic properties
respiration is called respiratory sinus arrhythmia, which has depending on its oxygenation state. Because neural activity
also been used as a measure of arousal and affective pro- following a specific task utilizes oxygen within specific
cesses (Potter and Bolls 2012). areas of the brain, the brain vasculature responds by increas-
Skin conductance response (SCR), also known as electro- ing the flow of oxygen-rich blood into the region. This leads
dermal response, occurs when the skin transiently becomes to a localized increase in blood oxygenation level–
a better electrical conductor due to increased activity of the dependent signal intensity in that region of the brain, which
is then measured using high-field magnetic resonance scan-
3We do not analyze pupil dilation further in this study, because the ads ners (Huettel, Song, and McCarthy 2008). Accordingly,
were not controlled for luminance and brightness, which are known to fMRI provides an indirect and correlative measure of local
affect pupil dilation. brain activity at high spatial resolution (approximately 1
mm3) and good temporal resolution (approximately 2–5 Table 1

seconds). METHODS, MEASURES, AND CONSTRUCTS
Neural activations can be used as a direct measure of
exogenous and endogenous attention. Exogenous attention Constructs
is measured through activation in the primary visual cortex Measures Attention Affect Memory Desirability
(greater visual processing) and amygdala (arousal). In con-
Traditional
trast, top-down attention depends on goals, internal states,
Liking x x
and expectations (e.g., health cues could help modulate Excitability x x
choices of different food items) and is associated with acti- Familiarity x
vation in the dorsolateral prefrontal cortex (dlPFC). The Recognition x
dlPFC is the “executive” part of the brain that helps process Purchase intent x
contextual information (Hare, Malmaud, and Rangel 2011; IAT
Miller and Cohen 2001). IAT valence x
IAT memory x
The amygdala has been a focus of research on affect and
Eye Tracking
emotions because it is a key part of the limbic system and Fixation count x
connects to subcortical structures that process autonomic Dwell time x
functions (Pessoa and Adolphs 2010; Phelps 2004). Across Pupil size x
several studies, the amygdala has consistently been shown Biometrics
to be involved in various aspects of emotional processing. Heart rate deceleration x
Specifically, the magnitude of its activation is often related Respiratory sinus arrhythmia x
Heart rate acceleration x
to affective intensity and has been described as being Skin conductance x
greater for negative than for positive stimuli (Critchley et al. EEG
2005; Dimoka 2010; Sabatinelli et al. 2005). However, Occipital alpha x
lesion studies have indicated that the amygdala may play a Frontal asymmetry x
more important role in emotional arousal than in valence fMRI
(Glascher and Adolphs 2003). dlPFC x
It is advantageous to use fMRI to measure memory vmPFC x x
Amygdala x
because it provides a direct measure of the strength of Hippocampus x
encoding during the ad. For example, we can explicitly Ventral striatum x
identify brain regions that show greater activation for stim-
uli that were remembered versus forgotten. The hippocam- of these neurophysiological measures can predict advertis-
pus has been shown to be critical for memory across many
ing success.
neuroimaging studies (Zola-Morgan and Squire 1993), and
lesions to this region affect a person’s ability to form new EXPERIMENTAL METHODS
memories and associations (Corkin 1984).
Activation in the ventromedial prefrontal cortex All studies were approved by Temple University’s Insti-
(vmPFC) and ventral striatum are viewed as key measures tutional Review Board. We tested a total of 37 ads in the
of desirability. The vmPFC has consistently been linked to main study. All ads were 30-second television ads drawn
willingness to pay for a wide range of branded products from six companies (described in detail subsequently) and
across different studies (Plassmann, O’Doherty, and Rangel included 15 unique brands. Participants were recruited from
2007; Plassmann, Ramsøy, and Milosavljevic 2012). The a large city in the U.S. Northeast through online and print
ventral striatum is the primary dopaminergic target in the ads. Interested participants were required to fill out an
brain and thus plays an important role in the prediction and online prescreening questionnaire at least two days before
consumption of rewards (Knutson, Delgado, and Phillips participating in the study. In addition to basic demographics
2010; Levy et al. 2011). The ventral striatum also plays an (gender, ethnicity, age, employment status, and income), we
important role in wanting, which refers to motivation or also collected information about television and television-
approach behavior toward rewards. Although wanting is ad-watching habits during the prescreening. Participants
often correlated with liking (hedonic value of reward), the who did not watch television or television ads were
two concepts can be distinguished through the manipulation excluded from the study. To measure the participants’ pre-
of dopamine levels (Berridge 2007). Recent research has disposition to products and brands, we showed them images
shown that activation in the ventral striatum during product
of brands featured in the study and collected information
evaluation is the strongest predictor of subsequent pur-
about their product familiarity, purchase intent, usage intent,
chases (e.g., Berns and Moore 2012; Knutson et al. 2007).
We summarize the various neurophysiological measures and recommendation intent. To minimize biases, we
under the four proposed constructs (attention, affect, mem- included other products from competitors as part of the pre-
ory, and desirability) in Table 1. A more detailed version screening questionnaire. For the main study, we collected
with references is available in Web Appendix A. As Table 1 data from a total of 277 participants across four separate
shows, these constructs can be assessed using measures phases. The experimental protocol was largely identical
from different methodologies. The approach taken in this across phases, except for minor methodology-specific
study is first to compare across these measures along the modifications. Next, we describe each of the four phases
four key advertising constructs and then to determine which briefly. Additional details are available in Web Appendix B.
Phase 1: Traditional and Implicit Measures familiarity, purchase intent, usage intent, and recommenda-
A total of 186 participants (86 women; mean age = 39 tion intent from the baseline measures obtained during pre-
14 years) completed Phase 1. All studies were conducted in screening. We normalized these change scores using a PAR
a laboratory for greater experimental control. A lab assistant measure to remove effects that were brand specific and not
briefed participants and obtained their signed informed con- associated with ad exposure (Walker and Dubitsky 1994).
sent before any data collection. Participants were then After viewing all ads, participants were given a five-
seated in front of a computer with headphones. All stimuli minute break before a surprise recognition test. We identi-
were presented using E-Prime 2.0 (Psychology Software fied two salient moments for each ad based on internal
Tools), and responses were captured using mouse and key- pretesting. We then used one of the screenshots (unbranded
board. Participants were provided $15 as compensation for moment) interspersed with foils (screenshots drawn from
this phase of the study. similar products). Participants were asked to indicate
We summarize the basic protocol in Figure 1, Panels A whether each of the screenshots was old (from ads they had
and B.4 Because some brands were repeated across ads, we seen in the session) or new on a six-point scale that included
split our protocol into two pods. In pod 1, participants confidence measures (Web Appendix B). To calculate the
watched the ads from each of the 15 unique brands. After a hit rate, we converted the responses into a simple binary
five-minute anagram distractor task, they retrieved as many scale.
brands as possible from the ads they had just watched, as a For 80 (41 women) of the 186 participants, we adminis-
free recall measure. The participants then completed pod 2, tered a modified version of the IAT (Greenwald, McGhee,
consisting of the remaining 22 ads. Ads were rotated within and Schwartz 1998) after the recognition test.5 Similar to
each pod across participants. After each ad, participants the original IAT, participants were asked to sort stimuli into
were asked a series of ten self-report questions, which were different categories as quickly as possible. Participants cate-
drawn primarily from “The ARF Copy Research Validity gorized words as either positive (e.g., love) or negative
Project” (Haley and Baldinger 2000) and the “Advertising (e.g., death) and categorized images as representing indoor
Research Foundation Copy Research Workshops” from the or outdoor scenes (Web Appendix B). The images were
early 1990s. They included five measures of the ad (liking, salient, unbranded screenshots drawn from the ads, inter-
excitability, relevance, informativeness, and familiarity) and spersed with foils selected from competitor ads. We used
four measures of products featured in the ad (purchase the difference in response latencies to ad images versus foil
intent, recommendation intent, usage intent, and famil- images as an implicit measure of memory (previously seen
iarity). All questions are listed in Web Appendix B. To images are likely to be retrieved more quickly). We refer to
assess desirability, we measured the change in product this measure as IAT_Memory. The difference in response
4We carried out two pilot studies to address concerns that (1) answering 5Participants also completed a second implicit task called the Affect
self-reported measures immediately after ad could bias memory measures Misattribution Procedure (Murphy and Zajonc 1993) after the IAT. How-
and (2) the large number of ads tested could affect participants’ engage- ever, because the Affect Misattribution Procedure focused only on the con-
ment and recognition scores (for details, see Web Appendix B). stricted set of unique brands, we do not discuss it further.
Figure 1
EXPERIMENTAL PROTOCOL
A: Behavioral Protocola
Ad 1 Question
Ad 2
Question ... Distractor Free
Ad 16
Question ... Recognition
Block Block Task Recall Block
Pod 1 Pod 2
B: fMRI Protocolb
Question Block
Ad 1 + Commercial
Familiarity
Commercial
Liking
Purchase
Intent
+ Ad 2 ...
30 s 4s 5s 5s 5s 8-15 s
aParticipants viewed 37 30-second television spots, separated into two pods. In Pod 1 (15 ads), all ads represented unique brands. In Pod 2, the remaining
ads were presented. Ads were rotated within each pod across participants. A question block followed each ad, in which a series of self-reported measures was
obtained. A recall task was administered for the unique brands in between the two pods, following a distractor task. Surprise recognition tests were adminis-
tered at the end of the study.
bFor the fMRI protocol, the two pods were divided into five runs. Each run was exactly eight minutes long. Each 30-second ad was followed by a 4-second
fixation (a cross presented in the center of the screen) and sequential presentation of three questions. Participants had up to 5 seconds to answer each question.
If a response was recorded before the 5-second limit, the remaining time was filled with a fixation. There was a variable intertrial interval of 8 to 12 seconds
before the next ad. Eight ads were presented in each run.
latency between when each image was paired with a posi- ads with unique brands, participants rested for four minutes
tive or negative word served as an implicit measure of emo- with eyes open and fixated on a cross in the center of the
tional valence toward each ad. We refer to this measure as screen. This rest period acted as a distractor for the subse-
IAT_Valence. We excluded 22 participants who had error quent free recall test that was administered in the scanner
rates greater than two standard deviations from the mean through an intercom. We summarize the timing for the
(Greenwald, McGhee, and Schwartz 1998), resulting in a fMRI protocol in Figure 1, Panels A and B, and provide
total of 58 participants for subsequent analysis. The details about the fMRI sequence in Web Appendix B. After
excluded participants had greater difficulty in reversing the fifth run, a recognition test was administered outside the
responses between blocks, possibly because of fatigue from scanner on a laptop.
the lengthy experimental protocol. We constructed a first-level general linear model (Friston
et al. 1994) for each of the ads for each participant. This
Phase 2: Eye Tracking and Biometrics model consisted of one regressor (30 seconds) for each of the
A total of 29 participants (11 women; mean age = 33 10 ads. All three traditional measures (familiarity, liking, and
years) completed the eye-tracking and biometric studies. purchase intent) were collapsed across ads into three regres-
After a briefing similar to the traditional phase, participants sors per run. Motion parameters were included in both mod-
sat in front of a Tobii T60XL eye tracker and were affixed els as an effect of noninterest. We then constructed a second-
with BIOPAC (MP150) BioNomadix wireless physiology level model for each of the 37 ads as one-sample t-tests in
devices for collecting skin conductance, heart rate, and SPM8 using contrast images from the first-level model for
breathing data. Stimuli were presented using E-prime 2.0. each of the 29 participants (Berns and Moore 2012). Then,
The protocol was similar to Phase 1, with additional breaks we built a third-level model (also a one-sample t-test) using
to mitigate participant fatigue. contrast images from the second level and additional covari-
For eye tracking, information about fixations and gaze ates (from within- or out-of-sample traditional measures).
locations were exported from Tobii and analyzed using in- For analysis with covariates, statistical images had a thresh-
house scripts in MATLAB. Biometric data were pre- old of p < .001, uncorrected. We also preselected an indepen-
processed using the Acqknowledge 4.0 package. For heart dent set of four regions on the basis of their established role
rate data, raw tonic data were analyzed using Acqknowl- in measuring the core constructs, for a region-of-interest
edge’s Heart Rate Variability procedure. Although typical (ROI) analysis (Web Appendix B). For each ROI, the
heart rate analysis focuses on the low-frequency (.04–.15 parameter estimates for each ad were obtained from the sec-
Hz) and high-frequency (.15–.40 Hz) components as mea- ond level using MarsBaR toolbox for SPM (Brett et al.
sures of accelerations and decelerations, they often require 2002). Unless specified otherwise, all brain activations in
events of longer durations for reliable estimations compared this study refer to these preselected ROIs and are not from
with the 30-second ads used here. Therefore, we created a any specific models.
coding system to identify heart rate accelerations and decel-
Phase 4: EEG
erations using the phasic heart rate signal. We coded any
positive shifts from the baseline (measured before the start We obtained high-density EEG data from 29 participants
of each ad) as heart rate acceleration and any negative shifts (15 women; mean age = 25 6 years). We used a 129-channel
from baseline as heart rate deceleration. We analyzed event- HydroCel Geodesic Sensor Net (Electrical Geodesics Inc.)
related SCR data using Acqknowledge’s built-in exploratory with a Cz reference to record EEG data. The protocol was
data analysis. Additional details about the setup and specific almost identical to the fMRI protocol (the liking measure
analyses are available in Web Appendix B. was not obtained due to a coding error). Raw EEG data were
first filtered using a bandpass filter (HP:.01 Hz; LP: 40 Hz)
Phase 3: fMRI and rereferenced to linked mastoids before performing inde-
Thirty-three participants (15 women, mean age = 29 8 pendent component analysis using EEGLAB (Delorme and
years) completed the fMRI protocol. All participants were Makeig 2004). Artifacts (horizontal eye movement, vertical
right-handed, healthy people with normal or corrected-to- eye movement, eye blinks, and general discontinuities) were
normal vision and were free of any hearing problems. All automatically detected and removed using ADJUST 1.1, an
participants provided written consent before participating independent plug-in for EEGLAB (Mognon et al. 2011). We
and received $40 as compensation. Four participants were then extracted alpha activity (8–12 Hz) from 17 channels
excluded from analysis due to excessive movement. for the final analysis (for additional details, see Web Appen-
The fMRI protocol was similar to that of Phase 1, with dix B). Extracted alpha activity was log-transformed and
several minor changes. Unlike the ten self-reported mea- baseline corrected for each channel on a moment-to-
sures in the traditional protocol, fMRI participants were moment basis. We then estimated the aggregate mean for
asked only three questions to keep the overall duration rea- frontal asymmetry (ln[F4] – ln[F3]) and occipital (Oz)
sonable. We selected measures about ad familiarity, liking, across the entire 30 seconds of each commercial.
and purchase intent because these measures showed the RELATIONSHIP AMONG MEASURES OF
most variability in our preliminary analysis. We then ADVERTISING EFFECTIVENESS
divided the 37 ads into five runs with breaks between each
run to reduce fatigue. To keep run lengths consistent (eight Relationship Among Traditional Measures
videos per run), we added one filler ad that was always the We first examined the relationship among the traditional
last video in the second run as well as two others at the end self-reported measures. We restricted this analysis to the
of the fifth run. After the first two runs, which consisted of 186 participants from Phase 1. Web Appendix C summa-
rizes the pairwise rank correlations across ads among the ates, we used the average liking and purchase intent mea-
different measures. We found significant positive correla- sures for each of the ads across participants from all four
tions among the various ad-related measures such as liking, phases as covariates to identify regions in the brain that
familiarity, relevance, and informativeness. We also found tracked these measures.
significant correlations among the various product-related Using the liking measure as a covariate, we found signifi-
measures—namely, changes in purchase intent, usage cant activations in the right amygdala, dlPFC, and vmPFC
intent, recommendation intent, and familiarity with the (Figure 2, top). Historically, liking has been argued to repre-
products featured in the ad. The ad-related measures were sent both cognitive and affective processes. The pattern of
also correlated with the product-related measures. Finally, activations found here is consistent with the presence of
recognition was significantly correlated with excitability these two components: the amygdala represents affective
and liking. processing (Pessoa and Adolphs 2010; Phelps 2004), and
We next categorized the traditional advertising measures the dlPFC represents cognitive processing (Miller and
using factor analysis. Using a Varimax rotation, we found Cohen 2001). To better understand the nature of this inter-
that the 11 measures loaded mainly onto three factors: one action, we used a bootstrapping mediation analysis
factor loaded strongly on all the 6 ad-related measures; the (Preacher and Hayes 2004) to investigate whether the
second factor loaded on all product-related change mea- amygdala activation mediated the effect of liking on the
sures and weakly on liking, excitability, and relevance; and
the third factor loaded strongly on recognition and weakly Figure 2
on excitability and liking (Web Appendix C). Therefore, we NEURAL CORRELATES OF LIKING AND MEMORY FOR
selected liking, change in purchase intent, and recognition TELEVISION ADS
as key traditional measures for further analysis because they
loaded highly onto one of the three different factors and
were most consistent with prior copy testing research.
Relationship Among Traditional Measures Across Samples
A different set of participants completed each of the four
experimental phases, so we analyzed the consistency in the
traditional measures across phases. The four measures col-
lected across all phases were ad familiarity, liking, purchase
intent, and recognition. For each of these measures, we cal-
culated a measure of internal consistency (Cronbach’s
alpha) across phases. We found strong consistency for lik-
ing ( = .916), familiarity ( = .796), change in purchase
intent ( = .739), and recognition ( = .931). This suggests
that the self-reported measures across the various sets of
participants were consistent in this study.
Relationship Between Biometric and Traditional Measures
We then examined the relationship between the biometric
and traditional measures for all participants in Phase 2 (Web
Appendix C). We found that deceleration correlated with
liking (r = .37, p < .05), recognition (r = .34, p < .05), and
change in purchase intent (r = .46, p < .01). These findings
are consistent with deceleration providing an independent
measure of increased attention. There was also a negative
correlation between heart rate acceleration and deceleration
(r = –.52, p < .001). Therefore, we focused on heart rate
deceleration for the remaining analyses. We did not find any
significant correlations between SCRs and any traditional
measures. Finally, among the various eye-tracking variables,
we found that the percentage of valid fixations was signifi-
cantly correlated with liking (r = .38, p < .05). This is again
consistent with the finding that ads that were liked were
associated with increased attention and processing. All other
associations were not statistically significant.
Relationship Between fMRI and Traditional Measures Notes: The top panel shows brain regions that positively correlated with
We aimed to elucidate the neural correlates of the three an out-of-sample measure of average likability across ads. Using a thresh-
old of p < .001, we found significant activation in the right amygdala, right
key traditional measures: liking, purchase intent, and recog- dlPFC, and right vmPFC. In the bottom panel, using subsequent-memory
nition. Unlike typical fMRI analyses that use aggregate analysis, we found significant activation in the bilateral hippocampus when
measures from the sample of fMRI participants as covari- we compared activity for recognized versus unrecognized ads.
dlPFC. Liking had a significant effect on dlPFC activation Next, we explore how the neurophysiological measures
( = .30, p < .01). However, when introducing amygdala predicted ad effectiveness, measured with market response
activation as a mediator, the direct effect of liking on the models. We had market response data for 26 of the 37 ads
dlPFC became insignificant ( = .07, p = .43), while the tested in the study (details highlighted in the following sec-
indirect effect through the amygdala was significant, imply- tion). Therefore, we restricted the number of measures used
ing that the amygdala fully mediated the effect of liking on in the prediction models a priori. We selected 17 measures6
the dlPFC. Therefore, we contend that liking leads to on the basis of empirical findings, relationships to the core
increased arousal and affect, which in turn translates to constructs, and relevance to prior advertising literature.
greater top-down attention and cognitive processing. They included four traditional measures (liking, product
Finally, activation in vmPFC is consistent with Berns and familiarity, change in purchase intent, and recognition),7
Moore (2012), who find activation in vmPFC to covary with two implicit measures (IAT_Valence and IAT_Memory),
likeability ratings of audio songs. two eye-tracking measures (percentage of valid fixations
We next used purchase intent as a covariate to identify and total number of fixations), three biometric measures
brain regions that tracked desirability for the products fea- (heart rate deceleration, SCR amplitude, and BPM), two
tured in the ad. We used only the postmeasures of purchase EEG measures (frontal asymmetry and occipital alpha), and
intent (“How likely are you to purchase the product in the four fMRI areas (vmPFC, dlPFC, amygdala, and ventral
ad you just watched?”) for this analysis because they are striatum). Table 2 summarizes the measures, their means
closest to the fMRI activations. We found activation in the across ads, and correlations between them for these ads.
vmPFC and more rostral region of the anterior cingulate
cortex. Again, activation in the vmPFC is consistent with ADVERTISING ELASTICITY ANALYSIS
other studies that have postulated an important role for this In the second stage of the study, we aim to link the 17
region in estimating willingness to pay for products and for measures across traditional and neurophysiological methods
product valuation (e.g., McClure et al. 2004; Plassmann, to market-level response to advertising. Our study addresses
O’Doherty, and Rangel 2007; Plassmann, Ramsøy, and the following primary question: Which of the measures
Milosavljevic 2012). explains the most variance in market response to advertising
Finally, we identified regions in the brain that tracked beyond the traditional measures that have been used in
recognition. We ran a traditional subsequent-memory analy- theory and practice for many years? To answer this ques-
sis focused on the moments in the ad that corresponded to tion, we developed a two-step process. In Step 1, we esti-
the images used in the recognition test. In a separate model, mate a sales response model by specifying and estimating
we included two additional regressors that were each two the market responses to the television ads on a company-by-
seconds long and classified as “remember” or “forgot” on company basis. In Step 2, given the response parameters
the basis of whether the participants correctly classified the estimated in Step 1, we regress these parameters on differ-
image in the subsequent recognition test. We then used a ent subsets of the variables aggregated over participants
paired t-test at the second level to look for differences in the from the six sets of measures.
brain between these two regressors. Consistent with our pre-
dictions, we found significant activation in the bilateral hip- Step 1: Estimating Advertising Elasticities
pocampus (Figure 2, bottom). In other words, ads that had We acquired sales and gross rating points8 (GRPs) data
higher activations in the hippocampus (stronger encoding) from four of the seven companies in the study as well as
during the initial presentation were more likely to be elasticities estimated directly by one of the other companies.
remembered in a surprise memory test later. These findings We were not able to obtain demand or elasticity data from
are consistent with prior studies that have also shown a two of the companies. Given differences in industry types
strong link between memory-related activation in the hip- and data availability, we sought a measure of response that
pocampus and brand preferences (e.g., McClure et al. would be comparable across the different companies and
2004). product categories. “Advertising elasticity” is the percent-
age change in sales due to a 1% change in the advertising
Summary of Relationships Across Traditional and
measure being utilized (e.g., expenditures, GRPs) and has
Neurophysiological Measures
been used extensively in the literature. Advertising elas-
First, we confirmed the consistency of traditional mea- ticities have several attractive features. First, they are
sures across all four phases. Second, we demonstrated high dimensionless, so they can be estimated independently of
reliability in the self-reported measures between the various the units of analysis. Second, they can be computed for any
samples used in the study, which enabled us to compare and dependent variable using suitable variable transformations.
integrate data across the different methodologies. We then
demonstrated relationships of the neurophysiological mea-
6We also considered excitability and activation in the hippocampus but
sures to the appropriate construct (attention, arousal, mem- subsequently excluded them because they were very highly correlated (r >
ory, and desirability) and to one another. For example, we .8) with liking and amygdala, respectively.
found a strong positive correlation between the SCR ampli- 7Although recall is a popular measure of memory, we did not analyze it
tude and frontal asymmetry measure (r = .39, p < .05) here because it was restricted only to the 15 unique brands.
8Gross rating points are a measure of the size of an audience for an
obtained from EEG, suggesting that ads with higher arousal advertisement. They measure the reach of an ad in terms of the percentage
levels as measured by SCR amplitude were also associated of the target audience exposed multiplied by the frequency with which the
with higher frontal asymmetry (greater approach behavior). exposure occurs.
Table 2
RELATIONSHIPS BETWEEN MODELED VARIABLES
Eye Tracking Biometrics

Traditional Number Percentage HR
IAT EEG fMRI
Recog- of of SCR Decele-
M SD Liking Familiarity PI nition Memory Valence Fixations Fixation Amplitude ration BPM Oz Asymmetry Amyg dlPFC vSTR
Liking 3.69 .22
Familiarity 3.91 .35 .70**
PI –.05 .17 .52** .22
Recognition .83 .14 .19 –.13 .06
IAT memory –105.81 106.94 .05 .21 .13 –.03
IAT valence 33.03 160.90 –.21 –.19 –.10 –.22 –.31
Number of fixations 75.08 6.67 –.26 –.19 –.26 .38 .16 .06
Percentage of fixation .74 .03 .29 .13 .02 .06 .06 –.30 .03
Predicting Advertising Success Beyond Traditional Measures
SCR amplitude 1.48 .70 .12 .02 .25 –.12 .28 –.29 –.30 .12
HR deceleration 11.06 4.33 .37 .21 .19 .31 .06 –.22 –.14 –.11 –.03
BPM 17.35 .70 –.04 –.30 –.15 –.22 –.32 .06 –.07 .31 –.27 –.23
Occipital alpha –.10 .02 –.08 –.08 .18 –.04 .01 –.08 .18 –.12 .09 –.02 –.10
Frontal asymmetry .00 .01 .13 .00 –.04 .30 –.15 .14 –.13 .02 .39* –.02 –.10 –.08
Amyg .44 .18 .44* .27 .08 .29 –.24 –.03 –.14 –.14 –.21 .39* –.08 .07 .05
dlPFC .98 .16 .34 .06 .38 .04 .38 –.12 –.10 –.05 .06 .10 –.01 .02 –.07 .31
vSTR .13 .11 –.03 –.21 .14 .03 .11 .05 -.43* –.09 –.08 .20 .07 .05 .28 .13 .12
vmPFC .28 .19 .28 .57** –.14 –.19 .24 .11 –.23 –.12 –.20 .27 –.22 –.16 .07 .40* .26 .21
*p < .05.
**p < .01.
Notes: = (Post – Pre) – PAR; PI = purchase intent; IAT = Implicit Association Test; SCR = skin conductance rate; HR = hear rate; BPM = breaths per minute; Oz = occipital; Amyg = amygdala; dlPFC =
dorsolateral prefrontal cortex; vSTR = ventral striatum; vmPFC = ventromedial prefrontal cortex.
445
The general model, which is estimated by industry for brand Table 3

i, takes the following form: ESTIMATED AD ELASTICITIES
(1) DVit = Git i + Zit + Xt + it, and

Company Ad Estimate SE
(2) Git = Git – 1 + GRPit, A Ad 1 .11* .07

A Ad 2 .11** .05
where A Ad 3 .16*** .06
A Ad 4 .16*** .05
GRPit = a vector of GRPs for all ad spots at time t for brand i; A Ad 5 .10 .13
Git = a vector of advertising goodwill stock for all ad spots j at B Brand 1, Ad 1 –.01 .02
time t for brand i; B Brand 1, Ad 2 .09* .05
i = a vector of advertising effectiveness for all ad spots j for B Brand 2, Ad 1 .09*** .02
brand i; B Brand 3, Ad 1 .18*** .06
B Competitor Brand 1, Ad 1 .26*** .07
= advertising carryover (1 – is the exponential decay B Competitor Brand 2, Ad 1 .09*** .01
rate); B Competitor Brand 2, Ad 2 .09*** .02
Zit = industry and/or brand-specific variables including media, C Ad 1 –.05 .12
price, penetration, and so on; C Ad 2 –.01 .13
Xt = time-related control variables (e.g., seasonality) at time t; C Ad 3 .13** .05
and C Ad 4 .41 .27
it = unobserved shock to sales for brand i at time t. C Ad 5 –.12* .07
D Brand 1, Ad 1 .33* .17
The parameters of interest are the i, which, given suffi- D Brand 1, Ad 2 .12 .16
cient data, enable us to estimate the effectiveness of the spe- D Brand 2, Ad 1 .17*** .05
D Brand 2, Ad 2 .11 .07
cific ads and then calculate the ad elasticities. Goodwill was E Ad 1 .26 —
entered linearly, which has the advantage that it does not E Ad 2 .54 —
matter if some ad exposures are aggregated and some are E Ad 3 .23 —
not. For several companies, we had GRP data for the spe- E Ad 4 .47 —
E Ad 5 .39 —
cific ad that was tested in the lab, whereas the GRPs of other
ads were aggregated. *p < .10.
If the true dependent variable of interest at the individual **p < .05.
***p < .01.
level is consumer utility and if we have data that enable us
to calculate market share, we could estimate the model
given in Equation 1 using the log-odds ratio as the depen- ket share data between 2010 and 2012 as well as unaided
dent variable, log(Sit) – log(S0t), where S0t is the market brand recall for the focal company and its major competi-
share of the outside option and the interpretation of the esti- tors. For this company, we therefore used equivalent models
mated coefficients would be the effect of the independent for two dependent variables: log-odds and recall. We esti-
variables on consumer utility (assuming a consumer-level mated the two-equation system using seemingly unrelated
Type I extreme value utility shock over which we integrate). regression analysis. By including the additional information
In cases in which we did not have competitor information, the in the recall measure, we could better account for the unob-
utility interpretation was still possible if the market size and served shocks. The set of control variables included a week-
share of the outside option did not vary systematically with of-year fourth-order polynomial, year-specific week-of-year
our control variables. This is because log(Sit) – log(S0t) = second-order polynomials, and brand dummy variables.
log(q/M) – log(S0t) = log(q) – log(M) – log(S0t), and so the Company B. Company B is a consumer products com-
log(M) and log(S0t) terms were absorbed in the regression pany. Data included weekly national GRP and market share
intercept when using log demand as the dependent variable. data between 2010 and 2012 for two related product cate-
A limitation of the analysis is that some of the executions gories. As with Company A, we had competitor demand and
ran for only a short period of time. It was thus impossible to GRP data, so we were able to use the log-odds ratio as the
separate short-term and long-term advertising effects. As a dependent variable. To construct the dependent variable, we
result, we created a cumulative advertising term and, draw- used sales divided by a price index as the demand variable.
ing on some auxiliary analyses, only estimated long-term The control variables included the price index, the everyday
elasticities on the basis of Equation 1, with = .9. In the fol- base price, promotion variables (log of the percentage of
lowing subsections, we provide a brief description of each sales sold with an ad feature with display, feature without
company’s estimates. A total of five firms—two consumer display, display without feature, and promotion only),
product firms, one large and one multinational financial brand-specific week third-order polynomials, and week
services firm, and one large Internet travel services firm— dummy variables. Company B also provided information on
provided data for this study. Because the companies pro- ads from one of its major competitors, which was useful to
vided different data for the estimation of Equation 1 and fully specify the model.
represented different product types and market conditions, we Company C. Company C is a large financial services
estimate five separate demand models with different sets of company. Because we did not have competitor data, we
controls. Table 3 presents all results. For additional company- used the log of the sales data for the dependent variable. As
specific details, see Web Appendix D. with Company A, we had an additional dependent variable
Company A. Company A is a large Internet travel services that would also reflect explained unobserved demand
company. Data included weekly national GRP data and mar- shocks—namely, the web channel click-through rate. We
used log click-through as a second dependent variable in a Finally, we ran the regressions separately using different
seemingly unrelated regression. The control variables sets of measures because we did not have a sufficient num-
included the Dow Jones Industrial Average, a week-of-year ber of observations to include all measures simultaneously
fourth-order polynomial, and a week third-order polynomial. in a single regression. However, we contend that such an
Company D. Like Company B, Company D is a large analysis is relevant for practitioners, who are highly
consumer products company. Data included weekly sales unlikely to invest in all the methods at once and have all
and advertising GRP between 2009 and 2012 for four adver- these measures.
tising executions. As with Company B, we used sales Our primary goal was to investigate which of the set of
divided by a price index as the dependent variable. Control measures best explains the variation in advertising elas-
variables included the price index and advertising GRP for ticities beyond traditional measures. However, we first ran a
untested advertising executions. set of regressions with each set of variables (traditional, IAT,
Company E. Company E is a multinational financial serv- biometrics, fMRI, and EEG) separately with individual-
ices company. Unlike the other four companies, this com- company dummy variables to control for fixed-effect differ-
pany provided its own advertising elasticity estimates.9 ences among companies. Although not a focus of our analy-
sis, these results (Web Appendix E) show that the traditional
Estimates variables were by far the best predictors of ad elasticities.
The estimates of the advertising elasticities for all tested They produced a 72% improvement in adjusted R2, beyond
ads, which would be used as the dependent variables in the the company dummies.
second-stage regression, appear in Table 3. The only nega- To assess which measures best improved the explanation
tive elasticity estimates are small and not significant; in of the advertising elasticities beyond these traditional mea-
these cases, we replaced the estimate with zero. Three of the sures, we included each set of nontraditional measures with
five ad elasticities from Company A are significant at 10%, the traditional measures and company dummies in separate
as are five of the seven from Company B, one of the five for regressions. We then conducted an F-test and assessed
Company C (due to data limitations), and one of the two for whether each method adds a significant explanatory power
Company D. We do not know the significance of the elas- after controlling for the traditional measures. We present the
ticities provided directly by Company E. The mean of the results in Table 4. We found that when we controlled for tra-
positive, significant elasticities was .14, which is within the ditional measures, only fMRI measures were significant
range found by many studies and meta-analyses of advertis- predictors of ad elasticities (p < .011). Consistent with this
ing effectiveness (e.g., Sethuraman, Tellis, and Briesch result, fMRI measures were the only variables to produce a
2011). positive percentage increase in adjusted R2. Table 4 also
presents the parameter estimates of the relationships between
Step 2: Neurophysiological Predictors of Ad Elasticities the individual measures and the market-level advertising
In Step 1, we recover the long-term effectiveness of elasticities, controlling for brand heterogeneity and the tra-
advertising for brand i and ad j (the ij). In Step 2, we esti- ditional measures. Notably, the only significant result is the
mate the effects of the various multimethod measures on the positive impact of the activation in ventral striatum.
effectiveness of television advertising on sales (i.e., the ad These results suggest that for researchers interested in
elasticities). Let ij be the ad elasticity for brand i and ad j. utilizing one physiological approach beyond the traditional
We transformed the elasticities for the second stage to use self-reported measures, fMRI would be the best candidate.
logs of elasticity because we could control for proportional To test whether some of the other measures explain the
differences in ad effectiveness across industries’ brand same variance as traditional measures, we ran additional
dummy variables. In practice, we use log(.1 + ij) to prevent regressions with each of the sets of measures in isolation,
taking the log of zero. The general form of the model uti- without controlling for traditional measures. We found that
lized is eye tracking and EEG measures were moderate predictors
of ad elasticities (for details, see Web Appendix E). There-
(3) log(.1 + ij) = Wij + Nij + i + ij,
fore, it is likely that eye tracking and EEG measures could
where Wij represents the traditional measures for ad j by potentially explain much of the same variance in ad elas-
brand i (including company-specific effects for purchase ticities as the traditional self-reported measures.
intent), Nij includes the nontraditional measures (including
DISCUSSION
implicit measures, eye tracking, biometrics, EEG, and
fMRI), and i are company fixed effects. In the past decade, a new industry has rapidly grown
We had to make several accommodations to the data and around neuroscience applied to marketing, as marketing
results to estimate Equation 3 because of the limited degrees practitioners increasingly look to neuroscience methods to
of freedom. Importantly, as noted previously, we used the better understand consumer behavior and advertising. Yet
reduced set of variables for each category of measures, and healthy skepticism exists in both academia and practice
in addition, we created aggregated values of the measures about the contribution and value of these methods to mar-
by taking the means across the relevant set of respondents. keting. This is the first study to provide a framework for
how academic research on neuroscience can inform adver-
tising practice. Using a unique experimental protocol, we
9This company did not disclose exact model and method for estimating
their advertising elasticities, which might introduce an inconsistency issue.
obtained multiple measures of advertising effectiveness
Therefore, as a robustness check, we conducted another set of analyses by across the six most commonly used methods (traditional
dropping observations for this company (Web Appendix E). self-reports, implicit measures, eye tracking, biometrics,
Table 4
EFFECTS OF TRADITIONAL AND NEUROPHSYIOLOGICAL MEASURES ON ADVERTISING ELASTICITIES BEYOND
TRADITIONAL MEASURES
Model 1 Model 2 Model 3 Model 4 Model 5

IAT Eye Tracking EEG fMRI Biometrics
Est. SE Est. SE Est. SE Est. SE Est. SE
Company Dummies
Constant 1.90 2.36 4.10 3.38 2.36 2.47 1.74† 1.20 2.31 3.64
Company A –.708** .234 –.706** .257 –.549* .271 –.851*** .138 –.652† .370
Company B –.582** .222 –.604** .236 –.418* .211 –.508*** .099 –.470 .499
Company C –.697† .423 –.893* .437 –.777* .416 –.933** .295 –.670 .582
Company D –.142 .245 –.145 .142 –.297 .230 –.072 .159 –.122 .411
Company E .161 .180 .024 .229 .107 .198 –.152 .157 .107 .229
Traditional Measures
Liking .091 .560 –.363 .572 –.548 .585 –.123 .417 .012 .911
Familiarity –.286 .345 –.205 .308 .121 .263 –.178 .340 –.148 .302
Recognition –1.53*** .401 –.925 .720 –1.04* .506 –1.51*** .359 –1.50† .833
Company A PI 2.70** .842 1.82† .981 2.16* .969 3.09*** .822 2.33 1.52
Company B PI 4.17** 1.74 4.22** 1.78 5.76*** .884 8.29*** 1.42 4.99† 3.18
Company B’s Competitor PI .030 .758 .505 .921 .077 .812 –.022 .571 –.045 1.18
Company C PI –.399 .731 –.641 .899 –.844 1.06 –.984† .655 –.777 1.30
Company D PI –1.57† .986 –1.01* .508 –.001 .674 –1.79** .516 –.824 .683
Company E PI 3.12* 1.58 .374 2.82 .882 1.77 .300 1.51 1.90 2.94
Implicit Measures
IAT memory 8.01e-4 6.42e-4
IAT valence 7.68e-5 3.26e-4
Eye Tracking
Number of fixations –.011 .009
Percentage of fixation –.724 2.08
EEG
Occipital alpha 2.48 1.90
Frontal asymmetry 3.38 5.09
fMRI
Amyg –.164 .253
dlPFC .330 .319
vSTR .869** .239
vmPFC .400 .480
Biometrics
SCR amplitude .017 .078
HR deceleration 2.81e-04 .002
BPM –.052 .121
Adjusted R2 .580 .498 .521 .856 .378
Percentage change in adjusted R2 8.0% –7.3% –3.0% 59.4% –29.6%
F-test p-value .471 .389 .399 .011 .949
†p < .20.
*p < .10.
**p < .05.
***p < .01.
Notes: The percentage changes in adjusted R2 and the F-test p-values are computed against the model with company dummies and traditional measures. The
base category for company dummies is Company B’s competitor. IAT = Implicit Association Test; PI = purchase intent; Amyg = amygdala; dlPFC = dorsolateral
prefrontal cortex; vSTR = ventral striatum; vmPFC = ventromedial prefrontal cortex; SCR = skin conductance rate; HR = heart rate; BPM = breaths per minute.
EEG, and fMRI). Furthermore, we demonstrated the rela- with a variety of traditional and neurophysiological meth-
tive contribution of these measures in predicting advertising ods that allow for direct comparisons of these methods. Pre-
elasticities using independent and objective measures of viously, comparisons across methods were inferred from
real-world advertising success obtained with marketing-mix parallel findings across studies using different stimuli and
modeling. Our findings suggest that neurophysiological protocols, thereby preventing a direct comparison. This
methods can explain significantly greater variance in adver- method integration has important implications for both aca-
tising elasticities than traditional advertising methods alone. demic research and practice. For academics, this study
We discuss the broad implications of these findings next. paves the way for similar efforts in other areas, such as con-
sumer decision making. For practitioners, it provides a
Implications for Theory, Research, and Practice proof of concept for the integration of traditional advertis-
This study makes three important contributions. First, we ing methods with neurophysiological approaches by using a
develop and test a unique multimethodological experimen- common protocol as well as a novel perspective toward cap-
tal protocol that examines the same stimuli (television ads) turing key marketing variables. Integration of evidence
from multiple methodologies will also likely lead to the physiological measures, particularly fMRI, which explained
development of better theories and models in marketing that the most incremental variance in advertising elasticities
are grounded on biological plausibility, which would ulti- beyond traditional measures. Only one other published
mately benefit both academics and practitioners. study, to our knowledge, has shown such a relationship
The second major contribution relates to the examination between neurophysiological measures and market outcomes
of interrelationships among the measures obtained from by using fMRI responses to song clips in the lab to explain
both traditional and neurophysiological methods (biomet- subsequent sales of music albums (Berns and Moore 2012).
rics, EEG, and fMRI) because they correspond to the key The additional predictive power in our study can be traced
constructs associated with advertising success (attention, back to specific neurophysiological processes (activation in
affect, memory, and desirability) (Table 1). In the past, dif- ventral striatum), which tap into a specific construct (desir-
ferences in terminology and language may have isolated ability). The ventral striatum, through its strong dopaminer-
academics and practitioners. We hope this study (e.g., Table gic connections, has been shown to play an important role in
1) helps clarify some of the pertinent advertising constructs reward processing. Specifically, it has been associated with
and how they can be measured differently with multiple the motivation of “wanting” something, rather than just
methods. A first step toward integrating these methods is to “liking” (Knutson et al. 2007). Therefore, the finding that
demonstrate the commonalities and differences among mea- the ventral striatum explains the most incremental variance
sures. We show high reliability across samples for the self- over traditional measures in this study is consistent with its
reported measures (demonstrating robustness of these mea- role in measuring desirability for the products featured in
sures) and largely consistent patterns of correlations both the ad. Given that advertising firms spend millions of dol-
across the four key constructs (attention, affect, memory, lars on advertising, our findings have important implica-
and desirability) and across measures (in support of internal tions for practice in that they help elucidate which particular
validity). For example, we show strong correlations among methods and exact measures better predict real-world adver-
liking, number of fixations, heart rate deceleration, and acti- tising success.
vation in dlPFC, consistent with the higher-level construct
of attention. Strikingly, liking was also correlated signifi- Limitations and Suggestions for Further Research
cantly with excitability and activation in the amygdala, con- The pioneering nature of this study opens up possibilities
sistent with previous intuition that liking measures both for further research. First, due to the small number of ads,
rational cognitive and also affective unconscious compo- we had to restrict the number of measures in our prediction
nents. Our mediation analysis further supports the notion analysis. Although we were able to use multiple measures
that affective processes may regulate the degree of top- across the most common neurophysiological methods, other
down attention. measures (e.g., pupil dilation) and methods (e.g., facial
It is also important to acknowledge that we did not con- electromyography, facial coding) were not part of this study.
firm all expected relationships, possibly because of variabil- The lower p-values and multiple second-level models could
ity in responses across participants and smaller sample sizes also raise concerns about false positives. However, all second-
for some of the methods. For example, skin conductance stage analyses were theoretically motivated and grouped by
measures did not correlate with any other arousal measure. methodologies to assess additional variation explained beyond
This could suggest that skin conductance measures may traditional measures. We find that fMRI explains significant
actually capture different aspects of arousal or merely repre- additional variation (at just over 5%) even when using con-
sent a limitation (as we discuss subsequently). We contend, servative Bonferroni correction for multiple second-level
however, that our findings have implications for theory of models. Still, further research should include additional
how various marketing measures relate to one another and methods, measures, and ads to provide greater degrees of
what we can learn from such relationships. These findings freedom for more comprehensive testing.
also provide valuable insight for advertising theory and Second, we could potentially obtain more precise esti-
measurement about the nature of higher-level constructs mates of ad elasticities by running ads in randomly selected
commonly used in advertising research. geographic markets to increase the variation in the ad GRP
Finally, and perhaps most importantly, this is one of the data. One of the challenges in estimating the ad elasticities
few studies to demonstrate the relationship between labora- in this study was that the ads had already been aired and
tory measures and real-world market outcomes, with obvi- were all part of national campaigns, limiting the variation in
ous implications for practitioners. After obtaining data using GRPs to time-series variation. Directly controlling the var-
a well-controlled experimental protocol in a lab from a rela- iation in ad GRPs would help minimize any biases in esti-
tively small number of participants who viewed television mates of the elasticities resulting from advertising endo-
ads, we effectively explained the real-world advertising geneity. Such endogeneity is not a great concern here
elasticities of these ads. Although not the main goal of the because our ad elasticities are consistent with prior litera-
research, we found that traditional measures explain the ture. Even if there is some bias in our estimates, it should
most variance in advertising elasticities after controlling for not be correlated with either the traditional or neurophysio-
firm differences. This finding gives further support to more logical measures, leaving second-stage results unaffected.
than 50 years of advertising research demonstrating that Third, we limited all analyses in this study to aggregate
measures such as purchase intent are good predictors of data across all 30 seconds of the television ads. However,
advertising success. certain methods, such as biometrics and EEG, may be more
More importantly, we show that the predictions of adver- effective in identifying interesting variations within portions
tising success can be substantially improved with neuro- of the ad (because of their high temporal resolution). These
subtle variations may have been washed out when we Critchley, Hugo D., Pia Rotshtein, Yoko Nagai, John O’Doherty,
aggregated our measurement across the entire ad. Therefore, Christopher J. Mathias, and Raymond J. Dolan (2005), “Activ-
future studies should focus on identifying interesting tem- ity in the Human Brain Predicting Differential Heart Rate
poral components within each ad (e.g., branding moments, Responses to Emotional Facial Expressions,” Neuroimage, 24
(3), 751–62.
final seconds) and relate them to advertising success. It is
Davidson, Richard J. (2004), “What Does the Prefrontal Cortex
very possible that the biometric and EEG measures may be ‘Do’ in Affect? Perspectives on Frontal EEG Asymmetry
more effective for these temporal aspects within an ad than Research,” Biological Psychology, 67 (1/2), 219–33.
measures from fMRI (Ohme et al. 2009). ———, Paul Ekman, Clifford D. Saron, Joseph A. Senulis, and
Finally, relative to traditional methods, neurophysiologi- Wallace V. Friesen (1990), “Approach-Withdrawal and Cerebral
cal methods are typically more expensive and less accessi- Asymmetry: Emotional Expression and Brain Physiology. I,”
ble. Further research could explore the incremental value of Journal of Personality and Social Psychology, 58 (2), 330–41.
each method relative to its cost and accessibility compared ———, Klaus R. Scherer, and H. Hill Goldsmith (2009), Handbook
with traditional methods. of Affective Sciences. New York: Oxford University Press.
Delorme, Arnaud and Scott Makeig (2004), “EEGLAB: An Open
Conclusion Source Toolbox for Analysis of Single-Trial EEG Dynamics
Including Independent Component Analysis,” Journal of Neuro-
A wide variety of methods have been developed to assess
science Methods, 134 (1), 9–21.
advertising effectiveness, ranging from traditional self- Demaree, Heath A., D. Erik Everhart, Eric A. Youngstrom, and
reported measures to eye tracking and neurophysiological David W. Harrison (2005), “Brain Lateralization of Emotional
tools. In this study, we provide insight into the relative con- Processing: Historical Roots and a Future Incorporating ‘Domi-
tribution of each of these methods in the context of television nance,’” Behavioral and Cognitive Neuroscience Reviews, 4 (1),
advertising. Specifically, we collected, analyzed, integrated, 3–20.
and compared the role of several methods and measures in Dimofte, Claudiu V. (2010), “Implicit Measures of Consumer
predicting real-world advertising success. Our findings Cognition: A Review,” Psychology & Marketing, 27 (10), 921–37.
clearly demonstrate the potential of neurophysiological Dimoka, Angelika (2010), “What Does the Brain Tell Us About
measures to complement traditional measures in improving Trust and Distrust? Evidence from a Functional Neuroimaging
the predictive power of advertising success models. In addi- Study,” MIS Quarterly, 34 (2), 373–96.
——— (2012), “How to Conduct a Functional Magnetic Resonance
tion to guiding practitioners toward supplementary mea- (fMRI) Study in Social Science Research,” MIS Quarterly, 36
sures that could enhance their efforts to predict advertising (3), 811–40.
effectiveness, this study demonstrates the potential of neu- ———, Paul A. Pavlou, and Fred D. Davis (2011), “NeuroIS: The
roscience applied to marketing research and practice by Potential of Cognitive Neuroscience for Information Systems
extending existing measures, helping enrich marketing Research,” Information Systems Research, 22 (4), 687–702.
theories, and improving models of marketing success. Du Plessis, Erik (1994), “Recognition Versus Recall,” Journal of
Advertising Research, 34 (3), 75–91.
REFERENCES Foxe, John J. and Adam C. Snyder (2011), “The Role of Alpha-
Ariely, Dan and Gregory S. Berns (2010), “Science and Society Band Brain Oscillations as a Sensory Suppression Mechanism
Neuromarketing: The Hope and Hype of Neuroimaging in Busi- During Selective Attention,” Frontiers in Psychology, 2 (accessed
ness,” Nature Reviews Neuroscience, 11 (4), 284–92. April 6, 2015), [available at http://journal.frontiersin.org/article/
Barry, Thomas E. and Daniel J. Howard (1990), “A Review and 10.3389/fpsyg.2011.00154/full].
Critique of the Hierarchy of Effects in Advertising,” Inter- Friston, Karl J., Andrew P. Holmes, Keith J. Worsley, Jean-Baptiste
national Journal of Advertising, 9 (2), 121–35. P. Poline, Chris D. Frith, and Richard S.J. Frackowiak (1994),
Batra, Rajeev, John G. Myers, and David A. Aaker (1996), Adver- “Statistical Parametric Maps in Functional Imaging: A General
tising Management, 5th ed. Upper Saddle River, NJ: Prentice Linear Approach,” Human Brain Mapping, 2 (4), 189–210.
Hall. Glascher, Jan and Ralph Adolphs (2003), “Processing of the Arousal
Berns, Gregory S. and Sara E. Moore (2012), “A Neural Predictor of Subliminal and Supraliminal Emotional Stimuli by the Human
of Cultural Popularity,” Journal of Consumer Psychology, 22 Amygdala,” Journal of Neuroscience, 23 (32), 10274–82.
(1), 154–60. Greenwald, Anthony G., Debbie E. McGhee, and Jordan L.K.
Berridge, Kent C. (2007), “The Debate Over Dopamine’s Role in Schwartz (1998), “Measuring Individual Differences in Implicit
Reward: The Case for Incentive Salience,” Psychopharmacol- Cognition: The Implicit Association Test,” Journal of Personal-
ogy, 191 (3), 391–431. ity and Social Psychology, 74 (6), 1464–80.
Biel, Alexander L. and Carol A. Bridgwater (1990), “Attributes of Haley, Russell I. and Allan L. Baldinger (2000), “The ARF Copy
Likable Television Commercials,” Journal of Advertising Research Validity Project,” Journal of Advertising Research, 40
Research, 30 (3), 38–44. (6), 114–35.
Brett, Matthew, Jean-Luc Anton, Romain Valabregue, and Jean- Hare, Todd A., Jonathan Malmaud, and Antonio Rangel (2011),
Baptiste Poline (2002), “Region of Interest Analysis Using the “Focusing Attention on the Health Aspects of Foods Changes
MarsBar Toolbox for SPM 99,” Neuroimage, 16 (2), S497. Value Signals in vmPFC and Improves Dietary Choice,” Jour-
Brown, Steven P. and Douglas M. Stayman (1992), “Antecedents nal of Neuroscience, 31 (30), 11077–87.
and Consequences of Attitude Toward the Ad: A Metaanalysis,” Harmon-Jones, Eddie, Philip A. Gable, and Carly K. Peterson
Journal of Consumer Research, 19 (1), 34–51. (2010), “The Role of Asymmetric Frontal Cortical Activity in
Camerer, Colin, George Loewenstein, and Drazen Prelec (2005), Emotion-Related Phenomena: A Review and Update,” Biologi-
“Neuroeconomics: How Neuroscience Can Inform Economics,” cal Psychology, 84 (3), 451–62.
Journal of Economic Literature, 43 (1), 9–64. ———, Lacey Lueck, Meghan Fearn, and Cindy Harmon-Jones
Corkin, Suzanne (1984), “Lasting Consequences of Bilateral (2006), “The Effect of Personal Relevance and Approach-
Medial Temporal Lobectomy: Clinical Course and Experimental Related Action Expectation on Relative Left Frontal Cortical
Findings in H.M.,” Seminars in Neurology, 4 (2), 249–59. Activity,” Psychological Science, 17 (5), 434–40.
Hess, Eckhard H. and James M. Polt (1960), “Pupil Size as National Academy of Sciences in the United States of America,
Related to Interest Value of Visual Stimuli,” Science, 132 87 (24), 9868–72.
(3423), 349–50. Ohme, Rafal, Dorota Reykowska, Dawid Wiener, and Anna
Horstmann, Nina, Andrea Ahlgrimm, and Andreas Glockner Choromanska (2009), “Analysis of Neurophysiological Reac-
(2009), “How Distinct Are Intuition and Deliberation? An Eye- tions to Advertising Stimuli by Means of EEG and Galvanic
Tracking Analysis of Instruction-Induced Decision Models,” Skin Response Measures,” Journal of Neuroscience, Psychol-
Judgment and Decision Making, 4 (5), 335–54. ogy, and Economics, 2 (1), 21–31.
Huettel, Scott A., Allen W. Song, and Gregory McCarthy (2008), ———, ———, ———, and ——— (2010), “Application of Frontal
Functional Magnetic Resonance Imaging, 2nd ed. Sunderland, EEG Asymmetry to Advertising Research,” Journal of Eco-
MA: Sinauer Associates. nomic Psychology, 31 (5), 785–93.
Jensen, Ole and Ali Mazaheri (2010), “Shaping Functional Archi- Pessoa, Luiz and Ralph Adolphs (2010), “Emotion Processing and
tecture by Oscillatory Alpha Activity: Gating by Inhibition,” the Amygdala: From a ‘Low Road’ to ‘Many Roads’ of Evaluat-
Frontiers in Human Neuroscience, 4 (accessed April 6, 2015), ing Biological Significance,” National Review of Neuroscience,
[available at http://journal.frontiersin.org/article/10.3389/fnhum. 11 (11), 773–83.
2010.00186/full]. Phelps, Elizabeth A. (2004), “Human Emotion and Memory: Inter-
Knutson, Brian, Mauricio R. Delgado, and Paul E.M. Phillips actions of the Amygdala and Hippocampal Complex,” Current
(2010), “Representation of Subjective Value in the Striatum,” in Opinion in Neurobiology, 14 (2), 198–202.
Neuroeconomics: Decision Making and the Brain, Paul W. Plassmann, Hilke, John O’Doherty, and Antonio Rangel (2007),
Glimcher, Colin F. Camerer, Ernst Fehr, and Russell A. Pol- “Orbitofrontal Cortex Encodes Willingness to Pay in Everyday
drack, eds. San Diego: Academic Press, 389–406. Economic Transactions,” Journal of Neuroscience, 27 (37),
———, Scott Rick, G. Elliott Wimmer, Drazen Prelec, and George 9984–88.
Loewenstein (2007), “Neural Predictors of Purchases,” Neuron, ———, Thomas Zoëga Ramsøy, and Milica Milosavljevic (2012),
53 (1), 147–56. “Branding the Brain: A Critical Review and Outlook,” Journal
Lang, Annie, Paul Bolls, Robert F. Potter, and Karlynn Kawahara of Consumer Psychology, 22 (1), 18–36.
(1999), “The Effects of Production Pacing and Arousing Con- Poels, Karolien and Siegfried Dewitte (2006), “How to Capture
tent on the Information Processing of Television Messages,” the Heart? Reviewing 20 Years of Emotion Measurement in
Journal of Broadcasting & Electronic Media, 43 (4), 451–75. Advertising,” Journal of Advertising Research, 46 (1), 18–37.
Leven, Wilfried (1991), Blickverhalten von Konsumenten: Grund- Potter, Robert F. and Paul David Bolls (2012), Psychophysiologi-
lagen, Messung und Anwendung in der Werbeforschung. Hei- cal Measurement and Meaning: Cognitive and Emotional Pro-
delberg: Physica-Verlag. cessing of Media. New York: Routledge.
Levy, Irfat, Stephanie C. Lazzaro, Robb B. Rutledge, and Paul W. Preacher, Kristopher J. and Andrew F. Hayes (2004), “SPSS and
Glimcher (2011), “Choice from Non-Choice: Predicting Con- SAS Procedures for Estimating Indirect Effects in Simple Medi-
sumer Preferences from Blood Oxygenation Level–Dependent ation Models,” Behavior Research Methods, Instruments, &
Signals Obtained During Passive Viewing,” Journal of Neuro- Computers, 36 (4), 717–31.
science, 31 (1), 118–25. Sabatinelli, Dean, Margaret M. Bradley, Jeffrey R. Fitzsimmons,
Lucas, Darrell B. and Steuart H. Britt (1963), Measuring Advertis- and Peter J. Lang (2005), “Parallel Amygdala and Inferotempo-
ing Effectiveness. New York: McGraw-Hill. ral Activation Reflect Emotional Intensity and Fear Relevance,”
Mandler, George (1980), “Recognizing: The Judgment of Previous Neuroimage, 24 (4), 1265–70.
Occurrence,” Psychological Review, 87 (3), 252–71. Schlinger, Mary Jane (1979), “A Profile of Responses to Commer-
McClure, Samuel M., Jian Li, Damon Tomlin, Kim S. Cypert, cials,” Journal of Advertising Research, 19 (2), 37–46.
Latane M. Montague, and P. Read Montague (2004), “Neural Sethuraman, Raj, Gerard J. Tellis, and Richard A. Briesch (2011),
Correlates of Behavioral Preference for Culturally Familiar “How Well Does Advertising Work? Generalizations from
Drinks,” Neuron, 44 (2), 379–87. Meta-Analysis of Brand Advertising Elasticities,” Journal of
Micu, Anca Cristina and Joseph T. Plummer (2010), “Measurable Marketing Research, 48 (June), 457–71.
Emotions: How Television Ads Really Work,” Journal of Adver- Shaw, John C. (2003), The Brain’s Alpha Rhythms and the Mind: A
tising Research, 50 (2), 137–53. Review of Classical and Modern Studies of the Alpha Rhythm
Miller, Earl K. and Jonathan D. Cohen (2001), “An Integrative Component of the Electroencephalogram with Commentaries on
Theory of Prefrontal Cortex Function,” Annual Review of Neu- Associated Neuroscience and Neuropsychology, 1st ed. Amster-
roscience, 24, 167–202. dam: Elsevier.
Mognon, Andrea, Jorge Jovicich, Lorenzo Bruzzone, and Marco Singh, Surendra N., Michael L. Rothschild, and Gilbert A.
Buiatti (2011), “ADJUST: An Automatic EEG Artifact Detector Churchill (1988), “Recognition Versus Recall as Measures of
Based on the Joint Use of Spatial and Temporal Features,” Psy- Television Commercial Forgetting,” Journal of Marketing
chophysiology, 48 (2), 229–40. Research, 25 (February), 72–80.
Morwitz, Vicki G. and Gavan J. Fitzsimons (2004), “The Mere- Smidts, Ale, Ming Hsu, Alan G. Sanfey, Maarten A.S. Boksem,
Measurement Effect: Why Does Measuring Intentions Change Richard B. Ebstein, Scott A. Huettel, et al. (2014), “Advancing
Actual Behavior?” Journal of Consumer Psychology, 14 (1/2), Consumer Neuroscience,” Marketing Letters, 25 (3), 1–11.
64–74. Smit, Edith G., Lex Van Meurs, and Peter C. Neijens (2006),
———, Joel H. Steckel, and Alok Gupta (2007), “When Do Pur- “Effects of Advertising Likeability: A 10-Year Perspective,”
chase Intentions Predict Sales?” International Journal of Fore- Journal of Advertising Research, 46 (1), 73–83.
casting, 23 (3), 347–64. Stewart, David W. (1984), “Physiological Measurement of Adver-
Murphy, Sheila T. and Robert B. Zajonc (1993), “Affect, Cogni- tising Effects,” Psychology and Marketing, 1 (1), 43–48.
tion, and Awareness: Affective Priming with Optimal and Sub- Stipp, Horst and Robert P. Woodard (2011), “Uncovering Emotion:
optimal Stimulus Exposures,” Journal of Personality and Social Using Neuromarketing to Increase Ad Effectiveness,” research
Psychology, 64 (5), 723–39. report, Advertising Research Foundation, (accessed April 6,
Ogawa, Sonoko, Toni M. Lee, A.R. Kay, and David W. Tank 2015), [available at https://thearf-org-aux-assets.s3.amazonaws.
(1990), “Brain Magnetic Resonance Imaging with Contrast com/research/ARF_Using_Neuromarketing_to_Increase_Ad_
Dependent on Blood Oxygenation,” Proceedings of the Effectiveness.pdf].
Strong, Edward K., Jr. (1925), “Theories of Selling,” Journal of Walker, David and Tony M. Dubitsky (1994), “Why Liking Mat-
Applied Psychology, 9 (1), 75–86. ters,” Journal of Advertising Research, 34 (3), 9–18.
Sutton, Steven K. and Richard J. Davidson (1997), “Prefrontal Wang, Yong J. and Michael S. Minor (2008), “Validity, Reliability,
Brain Asymmetry: A Biological Substrate of Behavioral and Applicability of Psychophysiological Techniques in Market-
Approach and Inhibition Systems,” Psychological Science, 8 ing Research,” Psychology & Marketing, 25 (2), 197–232.
(3), 204–210. Wang, Zheng, Annie N. Lang, and Jerome R. Busemeyer (2011),
Tomarken, Andrew J., Richard J. Davidson, and Jeffrey B. Hen- “Motivational Processing and Choice Behavior During Televi-
riques (1990), “Resting Frontal Brain Asymmetry Predicts sion Viewing: An Integrative Dynamic Approach,” Journal of
Affective Responses to Films,” Journal of Personality and Communication, 61 (1), 71–93.
Social Psychology, 59 (4), 791–801. Wedel, Michel and Rik Pieters (2008), “A Review of Eye-Tracking
Venkatraman, Vinod, John A. Clithero, Gavan J. Fitzsimons, and Research in Marketing,” in Review of Marketing Research, Vol.
Scott A. Huettel (2012), “New Scanner Data for Brand Mar- 4, Naresh K. Malhotra, ed. Armonk, NY: M.E. Sharpe.
keters: How Neuroscience Can Help Better Understand Differ- Yoon, Carolyn, Richard Gonzalez, Antoine Bechara, Gregory S.
ences in Brand Preferences,” Journal of Consumer Psychology, Berns, Alain A. Dagher, Laurette Dubé, et al. (2012), “Decision
22 (1), 143–53. Neuroscience and Consumer Decision Making,” Marketing Let-
———, John W. Payne, and Scott A. Huettel (2014), “An Overall ters, 23 (2), 473–85.
Probability of Winning Heuristic for Complex Risky Decisions: Zola-Morgan, Stuart and Larry R. Squire (1993), “Neuroanatomy
Choice and Eye Fixation Evidence,” Organizational Behavior of Memory,” Annual Review of Neuroscience, 16, 547–63.
and Human Decision Processes, 125 (2), 73–87.
Predicting Advertising Success Beyond Traditional Measures:
New Insights from Neurophysiological Methods and Market Response Modeling
VINOD VENKATRAMAN, ANGELIKA DIMOKA, PAUL A. PAVLOU, KHOI VO,
WILLIAM HAMPTON, BRYAN BOLLINGER, HAL E. HERSHFIELD, MASAKAZU
ISHIHARA, and RUSSELL S. WINER
WEB APPENDIX
2
WEB APPENDIX A
Constructs in Advertising Research
TABLE A.1: LITERATURE REVIEW ON MEASURES FOR

CORE CONSTRUCTS IN ADVERTISING RESEARCH
Constructs
Measures Attention Affect Memory Desirability
Liking 12, 40 38, 40, 43 12, 40
Traditional
Excitability 16, 31, 38

Familiarity 35
Recognition 7, 37
Purchase Intent 12, 25, 40
IAT Valence 4, 11
IAT
IAT Memory 21
Fixation Count 23, 30, 42
Tracking
Eye
Dwell Time 23, 30, 42

Pupil Size 10, 15
Heart Rate Deceleration 20, 32, 39, 41
Biometrics
Respiratory Sinus Arrhythmia 1, 32

Heart Rate Acceleration 3, 41
Skin Conductance 20, 32, 39, 41
Visual Gating 9, 17, 34
EEG
Frontal Asymmetry 6, 14, 26, 33, 36

Dorsolateral Prefrontal Cortex 13, 18
Ventromedial Prefrontal Cortex 24 8, 22, 29, 30
fMRI
Amygdala 27, 28
Hippocampus 5, 44
Ventral Striatum 2, 19, 30
References for Table A.1
1. Appelhans, Bradley M. and Linda J. Luecken (2006), "Heart rate variability as an index
of regulated emotional responding," Review of General Psychology, 10 (3), 229-40.
2. Berns, Gregory S. and Sara E. Moore (2012), "A neural predictor of cultural popularity,"
Journal of Consumer Psychology, 22 (1), 154-60.
3. Bolls, Paul D., Annie Lang, and Robert F. Potter (2001), "The effects of message valence
and listener arousal on attention, memory, and facial muscular responses to radio
advertisements," Communication Research, 28 (5), 627-51.
4. Brunel, Frederic F., Brian C. Tietje, and Anthony G. Greenwald (2004), "Is the implicit
association test a valid and valuable measure of implicit consumer social cognition?,"
Journal of Consumer Psychology, 14 (4), 385-404.
3
5. Corkin, Suzanne (1984), "Lasting Consequences of Bilateral Medial Temporal

Lobectomy - Clinical Course and Experimental Findings in Hm," Seminars in Neurology,
4 (2), 249-59.
6. Demaree, Heath A., D. Erik Everhart, Eric A. Youngstrom, and David W. Harrison
(2005), "Brain Lateralization of Emotional Processing: Historical Roots and a Future
Incorporating “Dominance”," Behavioral and Cognitive Neuroscience Reviews, 4 (1), 3-
20.
7. Du Plessis, Erik (1994), "Recognition Versus Recall," Journal of Advertising Research,
34 (3), 75-91.
8. Falk, Emily B., Elliot T. Berkman, and Matthew D. Lieberman (2012), "From Neural
Responses to Population Behavior: Neural Focus Group Predicts Population-Level Media
Effects," Psychological Science, 23 (5), 439-45.
9. Foxe, John J. and Adam C. Snyder (2011), "The role of alpha-band brain oscillations as a
sensory suppression mechanism during selective attention," Frontiers in Psychology, 2.
10. Graur, Simona and Greg Siegle (2013), "Pupillary motility: bringing neuroscience to the
psychiatry clinic of the future," Curr Neurol Neurosci Rep, 13 (8), 365.
11. Greenwald, Anthony G., Debbie E. McGhee, and Jordan L. K. Schwartz (1998),
"Measuring individual differences in implicit cognition: The implicit association test,"
Journal of Personality and Social Psychology, 74 (6), 1464-80.
12. Haley, Russell I. and Allan L. Baldinger (2000), "The ARF Copy Research Validity
Project," Journal of Advertising Research, 40 (06), 114-35.
13. Hare, Todd A., Jonathan Malmaud, and Antonio Rangel (2011), "Focusing attention on
the health aspects of foods changes value signals in vmPFC and improves dietary
choice," J Neurosci, 31 (30), 11077-87.
14. Harmon-Jones, Eddie, Philip A. Gable, and Carly K. Peterson (2010), "The role of
asymmetric frontal cortical activity in emotion-related phenomena: a review and update,"
Biol Psychol, 84 (3), 451-62.
15. Hess, Eckhard H. and James M. Polt (1960), "Pupil Size as Related to Interest Value of
Visual Stimuli," Science, 132 (3423), 349-50.
16. Holbrook, Morris B. and Rajeev Batra (1987), "Assessing the Role of Emotions as
Mediators of Consumer Responses to Advertising," Journal of Consumer Research, 14
(3), 404-20.
17. Jensen, Ole and Ali Mazaheri (2010), "Shaping functional architecture by oscillatory
alpha activity: gating by inhibition," Frontiers in Human Neuroscience, 4.
18. Knudsen, Eric I. (2007), "Fundamental components of attention," Annu Rev Neurosci, 30,
57-78.
19. Knutson, Brian, Scott Rick, G. Elliott Wimmer, Drazen Prelec, and George Loewenstein
(2007), "Neural predictors of purchases," Neuron, 53 (1), 147-56.
20. Lang, Annie, Paul Bolls, Robert F. Potter, and Karlynn Kawahara (1999), "The effects of
production pacing and arousing content on the information processing of television
messages," Journal of Broadcasting & Electronic Media, 43 (4), 451-75.
21. Mandler, George (1980), "Recognizing - the Judgment of Previous Occurrence,"
Psychological Review, 87 (3), 252-71.
22. McClure, Samuel M., Jian Li, Damon Tomlin, Kim S. Cypert, Latane M. Montague, and
P. Read Montague (2004), "Neural correlates of behavioral preference for culturally
familiar drinks," Neuron, 44 (2), 379-87.
4
23. Milosavljevic, Milica, Vidhya Navalpakkam, Christof Koch, and Antonio Rangel (2012),
"Relative visual saliency differences induce sizable bias in consumer choice," Journal of
Consumer Psychology, 22 (1), 67-74.
24. Mo, Jue, Yuelu L. Liu, Haiqing Q. Huang, and Mingzou Z. Ding (2013), "Coupling
between visual alpha oscillations and default mode activity," Neuroimage, 68, 112-18.
25. Morwitz, Vicki G., Joel H. Steckel, and Alok Gupta (2007), "When do purchase
intentions predict sales?," International Journal of Forecasting, 23 (3), 347-64.
26. Ohme, Rafal, Dorota Reykowska, Dawid Wiener, and Anna Choromanska (2010),
"Application of frontal EEG asymmetry to advertising research," Journal of Economic
Psychology, 31 (5), 785-93.
27. Pessoa, Luiz and Ralph Adolphs (2010), "Emotion processing and the amygdala: from a
'low road' to 'many roads' of evaluating biological significance," Nat Rev Neurosci, 11
(11), 773-83.
28. Phelps, Elizabeth A. (2004), "Human emotion and memory: interactions of the amygdala
and hippocampal complex," Current Opinion in Neurobiology, 14 (2), 198-202.
29. Plassmann, Hilke, John O'Doherty, and Antonio Rangel (2007), "Orbitofrontal cortex
encodes willingness to pay in everyday economic transactions," J Neurosci, 27 (37),
9984-8.
30. Plassmann, Hilke, Thomas Zoëga Ramsøy, and Milica Milosavljevic (2012), "Branding
the brain: A critical review and outlook," Journal of Consumer Psychology, 22 (1), 18-
36.
31. Poels, Karolien and Siegfried Dewitte (2006), "How to Capture the Heart? Reviewing 20
Years of Emotion Measurement in Advertising," Journal of Advertising Research, 46 (1),
18-37.
32. Potter, Robert F. and Paul David Bolls (2012), Psychophysiological measurement and
meaning : cognitive and emotional processing of media. New York: Routledge.
33. Rothschild, Michael L., Yong J. Hyun, Byron Reeves, Esther Thorson, and Robert
Goldstein (1988), "Hemispherically Lateralized EEG as a Response to Television
Commercials," Journal of Consumer Research, 15 (2), 185-98.
34. Rothschild, Michael L., Esther Thorson, Byron Reeves, Judith E. Hirsch, and Robert
Goldstein (1986), "EEG Activity and the Processing of Television Commercials,"
Communication Research, 13 (2), 182-220.
35. Schlinger, Mary Jane (1979), "A Profile of Responses to Commercials," Journal of
Advertising Research, 19 (2), 37-46.
36. Silberstein, Richard B. and Geoffrey E. Nield (2012), "Measuring Emotion in
Advertising Research: Prefrontal Brain Activity," Pulse, IEEE, 3 (3), 24-27.
37. Singh, Surendra N., Michael L. Rothschild, and Gilbert A. Churchill (1988),
"Recognition Versus Recall as Measures of Television Commercial Forgetting," Journal
of Marketing Research, 25 (1), 72-80.
38. Smit, Edith G., Lex Van Meurs, and Peter C. Neijens (2006), "Effects of Advertising
Likeability: A 10-Year Perspective," Journal of Advertising Research, 46 (1), 73-83.
39. Thorson, Esther and Annie Lang (1992), "The Effects of Television Videographics and
Lecture Familiarity on Adult Cardiac Orienting Responses and Memory,"
Communication Research, 19 (3), 346-69.
40. Walker, David and Tony M. Dubitsky (1994), "Why Liking Matters," Journal of
Advertising Research, 34 (3), 9-18.
5
41. Wang, Zheng, Annie N. Lang, and Jerome R. Busemeyer (2011), "Motivational
Processing and Choice Behavior During Television Viewing: An Integrative Dynamic
Approach," Journal of Communication, 61 (1), 71-93.
42. Wedel, Michel and Rik Pieters (2008), "A Review of Eye-Tracking Research in
Marketing," in Review of Marketing Research Vol. 4, Naresh K. Malhotra, ed. Vol. 4.
Armonk, N.Y. ; London: M.E. Sharpe.
43. Youn, Seounmi, Tao Sun, William D. Wells, and Xinshu S. Zhao (2001), "Commercial
liking and memory: Moderating effects of product categories," Journal of Advertising
Research, 41 (3), 7-13.
44. Zola-Morgan, Stuart and Larry R. Squire (1993), "Neuroanatomy of Memory," Annual
Review of Neuroscience, 16, 547-63.
6
WEB APPENDIX B
Supplemental Methods
Traditional Measures Survey Questions
1. Commercial Familiarity
How familiar are you with this commercial?
5 – Very familiar
4 – Familiar
3 – Neutral
2 – Unfamiliar
1 – Very unfamiliar
2. Liking
Thinking about the commercial you just saw, please indicate which statement best describe
your feelings about the commercial.
5 – I liked it very much
4 – I liked it
3 – I neither liked nor disliked it
2 – I disliked it
1 – I disliked it very much
3. This commercial is: Exciting ---- Boring (7 point scale)
4. This commercial is: Informative ---- Uninformative (7 point scale)
5. This commercial is: Relevant ---- Irrelevant (7 point scale)
6. Recognition
Instructions: You will be shown a series of images. Some of these images are from the
commercials that you saw at the beginning of the study (old) while others are not (new).
Please indicate if the image is OLD or NEW using the following scale:
1. Definitely new
2. Probably new
3. Maybe new
4. Maybe old
5. Probably old
6. Definitely old
Scales 1 - 3 are considered a “no, I do not recognize this image” and scales 4 - 6 are
considered a “yes, I do recognize this image.” To create a binary measure, responses of 1 – 3
7
are considered as a “no” and scored as a 0, and responses of 4 – 6 are considered as a “yes”
and scored as a 1.
7. Product Familiarity
How familiar are you with the product in the commercial?
5 – Very familiar
4 – Familiar
3 – Neutral
2 – Unfamiliar
1 – Very unfamiliar
8. Purchase Intent
How likely are you to purchase this product?
5 – Very unlikely
4 – Unlikely
3 – Undecided
2 – Likely
1 – Very likely
9. Usage Intent
Would you use this product?
5 – Definitely would use product
4 – Probably would use product
3 – Might use product
2 – Probably wouldn't use product
1 – Definitely wouldn't use product
10. Recommendation Intent

Would you recommend this product to someone else?
5 – Very unlikely
4 – Unlikely
3 – Undecided
2 – Likely
1 – Very likely
Protocol Development: Pilot Studies
There were two concerns with this protocol: (a) answering a series of self-reported measures
immediately after each ad could lead to enhanced encoding, potentially biasing the memory
measures, and (b) the large number of ads tested in a single session could affect participant’s
engagement and recognition scores. We carried out a series of pilot studies to address these
concerns. First, we used 6 ads (one from each company) and tested two versions of the protocol
(Figure A1). In Group 1 (G1, N=57), participants were presented ads followed by self-reported
measures at the end of each ad. At the end, they completed a recognition test.. In Group 2 (G2,
N=51), participants only viewed the ads in the first stage, completed a surprise recognition test, and
then watched ads again before the self-reported measures. We did not find any differences in
8
recognition scores between the two groups across all the recognition images (t(106) = -.286, n.s.).
Instead, we found strong rank correlations between the groups for recognition scores from the six
DGV ȡ S.05). Therefore, we decided to use the single-stage version, since it was shorter and
better suited for the neuropsychological methods. To address the second concern, we ran a third
group of participants (G3, N=50) using the same protocol as G1, but the 6 ads were now
interspersed with 26 other ads. The key issue was whether the increase in number of ads would
reduce memory performance. We did not find a significant difference in recognition scores between
the two groups (t = -QV ,QIDFWWKHUHFRJQLWLRQVFRUHVZHUHPDUJLQDOO\EHWWHU ȝG1 = .75 and
ȝG3 = .83) when more ads were included, consistent with the theory that increased variety improves
memory performance. We also found high rank correlations between the two groups for recognition
DFURVVWKHVL[FRPPRQDGV ȡ = .72, p < .1). This provides evidence that the increased number of
ads used in our study did not impact participant engagement or performance.
Figure B.1
TRADITIONAL PROTOCOL PILOT STUDIES
Implicit Association Test
Similar to classic IAT tasks, participants were asked to make judgments on two orthogonal
categories. In our IAT task, participants were asked to judge: 1) whether or not a stimulus word
was positive (e.g. love, peace) or negative (e.g. hate, death); and 2) whether a picture depicts an
outdoor scene or an indoor scene. Judgments were divided into five separate testing blocks.
Participants made judgments via two response keys (left and right). In the first block, participants
classified word stimuli as positive (left) or negative (right). In the second block, participants
classified images as indoor (left) or outdoor (right). In the third block, participants classified both
words and images. Indoor images were paired with positive words (left key) and outdoor images
with negative words (right). In the fourth block, participants classified images as indoor (right) or
outdoor (left) again, but the response mappings were switched from block 2. The fifth block was
similar to the third block except the pairing was reversed. The order of associations was
randomized across participants.
9
Figure B.3
EYE TRACKING AND BIOMETRIC EXPERIMENTAL SETUP
Phase 2: Eye-tracking and Biometrics Setup
Our Biometrics/Eye Tracking behavioral lab is equipped with a Tobii T60XL eye tracking
monitor and a BIOPAC (MP150) BioNomadix wireless physiology devices for collecting skin
conductance (EDA), heart rate (EKG), and breathing data. In order to integrate and synchronize
these different devices, two computers were utilized: one dedicated to stimuli presentation using
E-Prime 2.0 and the other to recording eye tracking data with Tobii Studio 3.2 (Tobii
Technology) and biophysiology data with Acqknowledge 4.0 (Biopac Systems Incorporated). All
time markers were synchronized using E-Prime 2.0. A gigabit Ethernet desktop switch facilitated
data communication between the PC towers and Tobii T60XL. A parallel port connection
allowed communication between E-Prime and Acqknowledge. This setup is summarized in
Figure B.2.
For EKG measurements (Figure B.3, panel A), three disposable wet gel (Ag/AgCl) snap
electrodes were applied to the participant’s torso at three locations: 1) the left rib cage,
superficial to the rib bone to avoid the intercostal muscles and diaphragm (positive lead) 2) the
10
right collarbone, inferior to the clavicle, lateral to the pectoral muscles (negative lead)and 3) the
right shoulder, symmetrical to left shoulder (ground lead). Note that the ground lead can also be
placed on the right rib cage, symmetrical to the left rib cage.
For skin conductance (Figure B.3, panel B), participants were asked to wash and wipe their
second (index) and third (middle) fingers on their left hand, before disposable snap electrodes
with isotonic gel were applied to the distal phalanges of both fingers.
For breathing measures (Figure B.3, panel C), a respiration transducer belt was wrapped around
the chest. To test the set up, participants were asked to take in a deep breath, hold it for a few
seconds, and exhale. This temporary change in state induces increased electrodermal activity and
disturbances in the participant’s respiration pattern, thus confirming both devices were
functioning properly. At least 5 minutes of resting baseline for each physiological measure was
recorded before the start of the experiment. Original image in panel A used with permission from
lifeinthefastlane.com.
Figure B.3
NEUROPHYSIOLOGICAL MEASUREMENT SET-UP
(A) EKG | (B) Skin conductance | (C) Breathing
Eye tracking and Biometric Analysis: Information regarding event-related time markers, pupil
size, event type and gaze duration, and gaze spatial locations were exported from Tobii and
analyzed with MATLAB. Biometric data were pre-processed using Acqknowledge 4.0 package.
For heart rate data, raw tonic data was analyzed using Acqknowledge’s Heart Rate Variability
procedure, with a modified Pan-Tompkins QRS detector used to identify R waves with minimum
beats per minute of 30, maximum beats per minute of 240, and spline resampling frequency of
8Hz. Though typical heart rate analysis focus on the low frequency (.04 – .15 Hz) and high
frequency (.15 – .40 Hz) components as measures of accelerations and decelerations, they often
require events of longer durations for reliable estimations compared to the 30-second
commercials. Therefore, we created a coding system to identify heart rate accelerations and
11
decelerations using the phasic heart rate signal. We coded any positive shifts from baseline
(measured prior to the start of each ad) as heart rate acceleration and any negative shifts from
baseline as heart rate deceleration. We coded shifts of 5 – 9 beats per min with a score of 1, 10-
14 beats per min with a score of 2, and anything greater than 15 with a score of 3. Scores were
summed across the 30 seconds independently for positive and negative shifts, providing a
measure of heart rate acceleration and deceleration, respectively.
For event-related skin conductance response data was analyzed using Acqknowledge’s built in
EDA analysis. Phasic skin conductance activity was derived from the raw data using a 0.05 Hz
high pass filter with baseline window of 5 seconds, activation threshold of 0.02 mS and rejection
level of 10% of the max.
For breathing, raw data were downsampled from 1000 samples per second to 50 samples per
second. A band pass FIR filter (between .05Hz and 1Hz) was applied to the resampled data. The
breaths per minute measure was then derived from these filtered data.
Phase 3: FMRI Acquisition Parameters
We acquired fMRI data on a 3T Siemens Vario scanner using an EPI sequence with parameters:
TR = 2000ms; TE = 30ms; 34 axial slices parallel to the AC-PC plane, with voxel size of
3.75*3.75*3.8mm. High-resolution T1-weighted MPRAGE anatomical images were acquired
and used for normalizing and coregistering individual data. SPM8 (Functional Imaging
Laboratory, UCL, London) was used for all fMRI analyses. Preprocessing included slice-time
correction, motion correction, spatial normalization, and smoothing (with 8 mm Gaussian kernel)
in that order. Region of interests (ROI) were extracted using MarsBaR toolbox for SPM8. Four
main regions were defined: amygdala, ventral striatum, dorsolateral prefrontal cortex, and
ventromedial prefrontal cortex (Figure B.4).
Figure B.4
FMRI REGIONS OF INTEREST
Phase 4: EEG Apparatus & Extracted Channels
High-density EEG data were recorded using a 129-channel HydroCel Geodesic Sensor Net using
a Cz reference (Electrical Geodesics, Inc.). Sensor impedance levels were below 50 K:,
appropriate for use with the Net Amps 300 high-impedance amplifier. Data were sampled at 250
Hz, and filtered using an analog .1 – 100Hz bandpass filter.
12
In order to stabilize the signal for F3, F4, and Oz, we calculated each measure as an average of
the main sites (F3\E24; F4\E124; Oz\E75) and the neighboring sites. Figure B.5 shows the main
electrode sites and their neighbors.
Figure B.5
EEG CHANNEL LOCATION
13
WEB APPENDIX C
Supplemental Results
Table C.1
TRADITIONAL MEASURE CORRELATIONS ACROSS ALL ADS
Familiarity Liking Excitability Informativeness Relevancy Recognition ¨)DPLOLDULW\ ¨3, ¨8,
Familiarity -
Liking .44** -
*
Excitability .40 .88** -
**
Informativeness .44 .28 .28 -
** ** **
Relevancy .59 .62 .54 .59** -
Recognition .10 .22 .23 .06 -.09 -
** * **
¨)DPLOLDULW\ .24 .46 .39 -.07 .47 -.05 -
** ** **
¨3, .09 .50 .50 .24 .55 .05 .76** -
** ** ** **
¨8, .03 .46 .44 .13 .44 .09 .76 .96** -
** ** ** ** **
¨5, .07 .63 .62 .23 .46 .23 .71 .91 .91**
Notes:
1. * p < .05, ** p < .01
2. Familiarity = Commercial Familiarity
¨)DPLOLDULW\ %UDQG3URGXFW)DPLOLDULW\6KLIW 3$5FRUUHFWHG
¨ 3RVW– Pre) – PAR
5. PI = Purchase Intent; UI = Usage Intent; RI = Recommendation Intent
Table C.2
RELATIONSHIP BETWEEN PHASE 1 AND 2 MEASURES ACROSS ALL ADS
Eye-tracking Biometric
SCR HR
# Fixations % Fixation BPM
Amplitude Deceleration
Familiarity .14 .16 -.14 .12 -.16
* *
Liking .04 .38 -.01 .37 .14
* **
Excitability -.02 .33 .06 .52 .05
Informativeness -.01 .21 .01 -.06 .02
*
Relevancy -.02 .34 -.01 .00 -.09
*
Recognition .29 .25 .00 .34 -.27
*
¨)DPLOLDULW\ -.08 -.01 -.22 .39 -.14
**
¨3, -.24 .13 -.06 .46 -.16
¨8, -.16 .31 -.08 .30 -.14
* *
¨5, -.22 .40 -.07 .33 .03
Notes:
1. * p<.05, ** p<.01
2. SCR = Skin conductance response; HR = Heart rate; BPM = Breaths per minute
14
Table C.3
COMPONENT MATRIX
Component
1 2 3
Familiarity .824
Liking .501 .569 .448
Excitability .475 .536 .487
Informativeness .760
Relevancy .442 .809
Recognition .906
¨Familiarity .859
¨3, .948
¨8, .959
¨5, .912
Notes:
1. Extraction Method: Principal Component Analysis.
2. Rotation Method: Varimax with Kaiser Normalization.
3. Rotation converged in 4 iterations.
)DPLOLDULW\ &RPPHUFLDO)DPLOLDULW\¨ Familiarity = Product Familiarity Shift
15
WEB APPENDIX D
Elasticity calculation by company details
Company A details:
In addition to the listed control variables, we also tried including Facebook and Magazine GRPS,
but they provided little explanatory power. Due to the limited number of exposures for the
different ad copy, we aggregated the creative at the campaign level – the Company has multiple,
distinct campaigns with very similar ads within-campaign.
Figure D.1
DEMAND AND EXPOSURE
Table D.1
COMPANY A SUR REGRESSION RESULTS FOR GOODWILL PARAMETERS
Log-odds Unaided recall
Ad campaign Estimate S.E. Estimate S.E.
**
1 1.186 -.72 3.363 -1.296
2 1.119* -.526 1.656 -.947
3 1.537** -.561 .854 -1.011
4 1.749** -.598 -.396 -1.077
5 .156 -.195 .638 -.351
Other 1.259*** 1.243*** .519
R-squared .091 .790
N 1,035
Notes: * p<.10, ** p<.05, *** p<.01
In Figure D.1, we show the log-odds ratio for the company and its competitors and the GRPs for
the company’s ads, over the 200-week period. The mean level of GRPs for the campaigns over
16
the data period vary between 4.7 and 19.8, and the mean levels conditional on exposures that
week vary between 42.8 and 68.0 with a maximum of 159.7 in one week for campaign 2. Each
campaign had non-zero exposures for at least 16 weeks (campaign 1) and no more than 45 weeks
(campaign 4). SUR regression results can be found in Table D.1.
Company B details:
Company B was unique in that they provided ad GRPs for not only their own ads but also for
competitors. In addition, one of their ads was for a brand in a second, related category. Brands 1-
3 belong to the company, and brand 3 is in the second category. Brands 4 and 5 belong to
competitors in the first category. Figures D.2 and D.3 show the log odds ratio and GRPs for the
tested ads in the two categories.
Figure D.2
DEMAND AND EXPOSURE DATA, CATEGORY 1
Figure D.3
DEMAND AND EXPOSURE DATA, CATEGORY 2
17
Over the given period, the average GRPs for the different ads range from .036 to 10.3. If an ad
was shown in a given week, then the average GRPs range from 30.4 to 232.1, and the number of
weeks a specific ad is shown ranges from eight weeks to 101 weeks. The maximum GRPs for
the tested ads in any one week is 479.7. GRPs for all ads are included in the regressions, and the
estimates for the tested ads are shown in Table D.2. In addition, we included the initial goodwill
of all ads at the beginning of the data period (using the .10 decay rate) to control for initial
conditions since two of the ads ran at the very beginning of the period.
Table D.2
COMPANY B REGRESSION RESULTS FOR GOODWILL PARAMETERS
Brand Ad Estimate S.E.

1 1 -.013 .055
1 2 .225 .128
***
2 1 .419 .106
**
3 1 .625 .216
***
4 1 .207 .052
***
5 1 .218 .032
***
5 2 .226 .043
R-squared .997
N 1,957
Notes: * p < .10, ** p < .05, *** p < .01
To maintain consistency with the rest of the analyses, we calculate the elasticities using the
average weekly category goodwill of 418, to avoid the issue of much higher elasticities
calculated for brands with greater GRPs. This also yields the same results in the second stage
(except for the brand intercepts) as if we just used the estimates of the demand equation, which
can be interpreted as utility coefficients. This is because we use a log-linear specification in
which case any multiple of the demand variable (such as the average GRPs) is absorbed in the
brand fixed effects in the second stage regression.
Company C details:
As with Company A, we had an additional dependent variable which would also reflect
explained unobserved demand shocks, namely the web channel click-through rate. We used log
click-through as a second dependent variable in an SUR regression. The DV is new money. We
show the demand over time in Figure D.4, which clearly exhibits considerable seasonality. In
addition, we show the variation in the ads, both the tested ads and the non-tested ads.
It is clear that the company had many other exposures from ads that were not tested. The average
GRPs for the tested ads over the period of the data range from .17 to 2.0, whereas the average
total GRPs are 41.9. The average GRPs for each tested ad conditional on the ad being shown that
week range from 2.1 to 9.0.The number of weeks a specific tested ad is shown ranges from 17
18
weeks to 47 weeks. The maximum GRPs for one of the tested ads in any one week is 29.5, and
the maximum total GRPs in any one week is 157.
Figure D.4
DEMAND AND EXPOSURE DATA
Regression results can be found in Table D.3.
Table D.3:
COMPANY C REGRESSION RESULTS
Log new money Click-through
Ad Estimate S.E. Estimate S.E.
1 -.913 1.982 -9.522 6.437
2 -.215 2.221 8.576 7.215
3 2.154 *
.908 5.516 2.951
4 6.952 4.577 13.249 14.868
5 -2.068 1.193 2.852 3.874
other .008 .222 .310 .723
R-squared .614 .355
N 208
Notes: * p<.10, ** p<.05, *** p<.01
Company D details:
Company D’s data set includes two brands, each with two tested advertising executions. We
conduct a regression separately for each brand. The dependent variable is unit sales, and the
independent variables include advertising GRPs for both tested and untested advertising
executions, and the price index. For Brand 1, we also include a time trend to control for a
19
declining sales trend. Figure D.5 show weekly unit sales and advertising GRPs for tested
advertising executions.
Figure D.5:
UNIT SALES ADVERTISING GRPS
Regression results are reported in Table D.4.
Table D.4:
COMPANY D REGRESSION RESULTS
Brand 1 Brand 2
Estimate S.E. Estimate S.E.
Ad 1 1006.8* 525.3 333.8*** 93.13
Ad 2 359.1 486.6 306.6 205.8
Non-tested ads 192.5*** 64.92 926.9*** 188.1
Price -1.00e+6*** 9.19e+4 -2.43e+5*** 9.10e+4
Time trend -1925.4*** 239.5
Constant 5.19e+6*** 2.93e+5 1.36e+6*** 3.64e+5
R2 .709 .328
N 203 70
Notes: * p<.10, ** p<.05, *** p<.01
Company E details:
This company provided us with their own estimates for the advertising elasticities, and we
therefore have no details for how they were calculated.
20
WEB APPENDIX E
Method with most explanatory power
The first question we wanted to answer was which of the six sets of measures best
explains variation in the advertising elasticities? To answer this, we estimated the second stage
model using each set of measures in isolation as independent variables to see how much
variation in the elasticities they explain. The estimation results are reported in Table E.1. To
assess the statistical significance of each method in explaining variation in the advertising
elasticities, we conducted an F-test and reported the p-value at the bottom of each column.
As seen from the table, none of the models are significant at the conventional p<.05 level.
However, when we relax the threshold given the low degrees of freedom, the most significant set
of predictors, after controlling for brand differences, are the eye-tracking measures (p<.077)
followed by the traditional measures (p<.118) and the EEG measures (p<.175). This is an
interesting finding since we have noted that the traditional measures have been in the ad
researcher’s toolkit for many years. In fact, traditional measures explained the most variance in
elasticities beyond the brand dummies (based on adjusted R2), followed by eye-tracking and then
EEG. It appears that eye tracking and EEG methods may be as compelling when trying to
explain market-level behavior.
Table E.1
EFFECTS OF TRADITIONAL AND NEUROPHSYIOLOGICAL MEASURES ON

ADVERTISING ELASTICITIES
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Company Eye
Traditional IAT EEG fMRI Biometrics
dummies Tracking
2
Adjusted R .313 .537 .343 .450 .338 .181 .262
% change in adjusted
- 71.6% 9.6% 43.8% 11.2% -42.2% -16.3%
R2
F-test p-value - .118 .301 .077 .175 .940 .499
Notes: The percentage changes in adjusted R2 and the F-test p-values are computed against the model with company dummies
only (Model 1).
Robustness Check: Estimation results without Company E
Company E provided their own estimates for the advertising elasticities, but did not disclose the
exact model and the method they used, which might introduce an inconsistency issue. Therefore,
as a robustness check, we conducted another set of estimations by dropping the observations for
Company E. Tables E.2 and E.3 present the results without Company E. Overall, our previous
findings are all supported with this alternative data set. For the second set of results (Table E.3),
we found that, though some of the previously non-significant methods became significant, the
fMRI measures remain to be one of the most significant predictors beyond the traditional
measures.
21
Table E.2
EFFECTS OF TRADITIONAL AND NEUROPHYSIOLOGICAL MEASURES ON
ADVERTISING ELASTICITIES WITHOUT FIRM E
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Eye
Company dummies Traditional IAT EEG fMRI Biometrics
Tracking
Adjusted R2 .119 .650 .140 .328 .231 -.002 .092
% change in adjusted R2 - 446.2% 17.6% 175.6% 94.1% -101.7% -22.7%
F-test p-value - < .001 .419 .077 .074 .789 .368
Notes: The percentage changes in adjusted R2 and the F-test p-values are computed against the model with company dummies
only (Model 1).
22
Table E.3
EFFECTS OF NEUROPHYSIOLOGICAL MEASURES ON ADVERTISING
ELASTICITIES BEYOND TRADITIONAL MEASURES WITHOUT FIRM E
Model 8 Model 9 Model 10 Model 11 Model 12
IAT Eye Tracking EEG fMRI Biometrics
Est. S.E. Est. S.E. Est. S.E. Est. S.E. Est. S.E.
Company dummies
Constant 4.55 3.36 5.31+ 3.25 3.61 3.66 2.49+ 1.31 4.70+ 2.29
Company A -.963*** .233 -.914*** .222 -.769* .342 -.999*** .134 -1.05*** .213
Company B -.866** .274 -.753** .196 -.622* .299 -.537** .104 -.959** .296
Company C -1.29+ .710 -1.00* .467 -.970+ .652 -1.16** .313 -1.01* .433
Company D -.425 .356 -.093 .218 -.322 .434 .056 .150 -.472 .313
Traditional measures
Liking -.693 .990 -.109 .629 -.581 1.34 .400 .375 -.183 .787
Familiarity -.157 .204 -.429+ .228 -.061 .511 -.617+ .305 -.270 .352
Recognition -1.68** .511 -1.56** .516 -1.52 1.16 -2.14*** .371 -2.42** .558
Company A x PI 2.91** .887 2.50** .971 2.91* 1.32 4.36** .997 4.30*** .924
Company B x PI 7.73*** 1.54 5.12** 1.29 6.44*** 1.24 9.94** 1.97 10.71** 2.70
Company B’s Comp x PI .747 1.08 .558 .924 .498 1.13 -.049 .374 .641 1.03
Company C x PI -.778 .938 -.739 .750 -.771 1.18 -1.77+ .841 -.196 .649
Company D x PI -.325 1.12 -1.51** .541 -.471 1.78 -1.87** .575 -.649 .865
Implicit measures
IAT Memory -6.90e-5 6.03e-4
IAT Valence 5.34e-4 3.79e-4
Eye Tracking
# Fixations -.012+ .007
% Fixation -1.55 1.84
EEG
Occipital Alpha 1.33 4.22
Frontal Asymmetry 1.71 8.21
fMRI
Amyg -.159 .231
dlPFC -.231 .539
vSTR .482* .190
vmPFC .992+ .521
Biometrics
SCR Amplitude -.121+ .077

HR Deceleration -.002 .002
BPM -.030 .114
Adjusted R2 .692 .705 .527 .871 .634
% change in adjusted R2 6.5% 8.5% -18.9% 34.0% -2.5%
F-test p-value .402 .144 .952 .034 .317
Notes:
1. + p < .20, * p < .10, ** p < .05, *** p < .01
2. The percentage changes in adjusted R2 and the F-test p-values are computed against the model with company
dummies and traditional measures (Model 2 in Table E1).
Copyright of Journal of Marketing Research (JMR) is the property of American Marketing
Association and its content may not be copied or emailed to multiple sites or posted to a
listserv without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.

Predicting Advertising Success Beyond

Uploaded by

Copyright:

Available Formats

Predicting Advertising Success Beyond

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Predicting Advertising Success Beyond

Uploaded by

Copyright:

Available Formats

VINOD VENKATRAMAN, ANGELIKA DIMOKA, PAUL A.

PAVLOU, KHOI VO,

Keywords: advertising elasticities, neuroscience, biometrics, implicit

Online Supplement: http://dx.doi.org/10.1509/jmr.13.0593

Predicting Advertising Success Beyond

© 2015, American Marketing Association Journal of Marketing Research

mm3) and good temporal resolution (approximately 2–5 Table 1

Eye Tracking Biometrics

The general model, which is estimated by industry for brand Table 3

(1) DVit = Git i + Zit + Xt + it, and

(2) Git = Git – 1 + GRPit, A Ad 1 .11* .07

Model 1 Model 2 Model 3 Model 4 Model 5

New Insights from Neurophysiological Methods and Market Response Modeling

VINOD VENKATRAMAN, ANGELIKA DIMOKA, PAUL A. PAVLOU, KHOI VO,

WILLIAM HAMPTON, BRYAN BOLLINGER, HAL E. HERSHFIELD, MASAKAZU

ISHIHARA, and RUSSELL S. WINER

Constructs in Advertising Research

TABLE A.1: LITERATURE REVIEW ON MEASURES FOR

Excitability 16, 31, 38

Dwell Time 23, 30, 42

Respiratory Sinus Arrhythmia 1, 32

Frontal Asymmetry 6, 14, 26, 33, 36

References for Table A.1

5. Corkin, Suzanne (1984), "Lasting Consequences of Bilateral Medial Temporal

Traditional Measures Survey Questions

3. This commercial is: Exciting ---- Boring (7 point scale)

4. This commercial is: Informative ---- Uninformative (7 point scale)

5. This commercial is: Relevant ---- Irrelevant (7 point scale)

10. Recommendation Intent

Protocol Development: Pilot Studies

Implicit Association Test

Phase 2: Eye-tracking and Biometrics Setup

Phase 3: FMRI Acquisition Parameters

Phase 4: EEG Apparatus & Extracted Channels

Elasticity calculation by company details

Notes: * p<.10, ** p<.05, *** p<.01

Brand Ad Estimate S.E.

Notes: * p < .10, ** p < .05, *** p < .01

Regression results can be found in Table D.3.

Notes: * p<.10, ** p<.05, *** p<.01

Regression results are reported in Table D.4.

Notes: * p<.10, ** p<.05, *** p<.01

Method with most explanatory power

EFFECTS OF TRADITIONAL AND NEUROPHSYIOLOGICAL MEASURES ON

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Robustness Check: Estimation results without Company E

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

SCR Amplitude -.121+ .077

You might also like

(1) DVit = Git i + Zit + Xt + it, and

Notes: * p<.10, p<.05, * p<.01

Notes: * p < .10, p < .05, * p < .01

Notes: * p<.10, p<.05, * p<.01

Notes: * p<.10, p<.05, * p<.01