Professional Documents
Culture Documents
Module 5. Learning Memory Neuromarketing Compendium 2nd Ed p.273 To p.308
Module 5. Learning Memory Neuromarketing Compendium 2nd Ed p.273 To p.308
SELECTED'READINGS'IN''
CONSUMER)NEUROSCIENCE)&)
NEUROMARKETING)
!
2nd$edition$
!
Compiled$by$
Thomas$Zoëga$Ramsøy$$
2014
LEARNING & MEMORY
What%is%a%memory?%What%is%remembering,%and%can%there%be%different%kinds%of%recall?%As%soon%as%
we%think%about%memory,%we%probably%think%about%the%recall%of%vivid%playbacks%of%what%
happened%the%other%day,%who%said%what,%and%how%you%felt.%
Memory%is%a%term%covering%a%whole%plethora%of%different%functions%and%processes.%Reflexes,%
instinctual%behaviors,%habits%and%reading%are%all%in%part%memory%processes,%and%with%different%
kinds%of%access%to%prior%learning.%For%example,%you%probably%do%not%have%a%detailed%movieClike%
playback%of%when%you%actually%learned%to%ride%a%bike,%but%nevertheless%you%can%easily%jump%on%a%
bike%and%ride%it%more%or%less%on%auto%pilot.%Your%conscious%recollection%of%riding%a%bike%is%not%
necessary%for%you%to%use%your%learned%and%automatized%behavior%of%riding%a%bike.
Recall%can%therefore%also%come%in%many%shades.%Think%of%a%brand%name.%The%first%name%that%
came%to%mind%would%be%an%instance%of%“free%recall”.%If%I%write%a%brand%name%like%“Pepsi”%you%will%
recognize%it,%and%“recognition”%is%a%different%memory%process.%If%I%ask%you%to%judge%your%
knowledge%about%the%brand%“BMW”%you%will%access%your%semantic%memory%and%make%a%selfC
evaluated%rating%of%your%own%memory.
Not%all%memory%types%are%relevant%to%consumer%behavior%–%or%are%they?%Maybe%even%the%
memory%types%we%take%most%for%granted%are%also%part%of%our%consumer%choice%process?
ARTICLE IN PRESS
YNIMG-05947; No. of pages: 8; 4C:
NeuroImage xxx (2009) xxx–xxx
NeuroImage
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / y n i m g
Tuning down the emotional brain: An fMRI study of the effects of cognitive load on
the processing of affective images
Lotte F. Van Dillen a,c,⁎, Dirk J. Heslenfeld b, Sander L. Koole a
a
Department of Social Psychology, VU University, Van der Boechorststraat 1, 1081 BT, Amsterdam, The Netherlands
b
Department of Cognitive and Clinical Neuropsychology, VU University, Van der Boechorststraat 1, 1081 BT, Amsterdam, The Netherlands
c
Department of Social and Organizational Psychology, Utrecht University, Heidelberglaan 1, 3584 CS Utrecht, The Netherlands
a r t i c l e i n f o a b s t r a c t
Article history: The present research examines whether cognitive load can modulate the processing of negative emotional
Received 1 August 2008 stimuli, even after negative stimuli have already activated emotional centers of the brain. In a functional
Revised 13 January 2009 magnetic resonance imaging (fMRI) study, participants viewed neutral and negative stimuli that were
Accepted 14 January 2009
followed by an attention-demanding arithmetic task. As expected, exposure to negative stimuli led to
Available online xxxx
increased activation in emotional regions (the amygdalae and the right insula). Subsequently induced task
Keywords:
load led to increased activation in cognitive regions (right dorsolateral frontal cortex, right superior parietal
Cognition cortex). Importantly, task load down-regulated the brain's response to negative stimuli in emotional regions.
Emotion Task load also reduced subjectively experienced negative emotion in response to negative stimuli. Finally,
Interaction coactivation analyses suggest that during task performance, activity in right dorsolateral frontal cortex was
Amygdala related to activity in the amygdalae and the right insula. Together, these findings indicate that cognitive load
Insula is capable of tuning down the emotional brain.
Dorsolateral frontal cortex © 2009 Elsevier Inc. All rights reserved.
IAPS
Negative emotional experiences are an inescapable aspect of Reekum et al., 2007). For example, in one study, negative visual
human life. Although brief episodes of negative emotions can be distracters engaged the amygdalae during participants' judgements
adaptive (Öhman, 2007), negative emotions may become problematic whether two bars were like oriented or not (Pessoa et al., 2002), but
when they persist over time. Indeed, there is mounting evidence that only when the difference in orientation of two bars was easy to judge.
enduring negative emotional states impair both psychological and When the orientation of the two bars was difficult to judge, so that the
physical health (Lyubomirsky et al., 1999; Sapolsky, 1994). It is central task became more attentionally demanding, the amygdalae no
therefore vital for people to develop ways of effectively dealing with longer differentiated between the negative and neutral distracters. In
negative emotion. another study (Erk et al., 2007), amygdalae responses to negative
One important way of dealing with negative emotion consists of scenes were smaller when participants concurrently performed a
minimizing the amount of attentional resources that are devoted to working memory task that was highly rather than moderately
processing negative information (Ochsner and Gross, 2005; Nolen- demanding.
Hoeksema et al., 1993; Van Dillen and Koole, 2007). More specifically, Although past work has made important progress, it remains
performing an attention-demanding task has been found to attenuate unclear precisely how cognitive load influences the emotional brain.
the emotional impact of negative stimuli (Erber and Tesser, 1992; One possibility, which has been proposed in prior work, is that
Erthal et al., 2005; Glynn et al., 2002; Morrow and Nolen-Hoeksema, cognitive load prevents the processing of the emotional impact of
1990; Pessoa et al., 2002; Van Dillen and Koole, 2007; in press). In negative stimuli altogether. From this perspective, cognitive load may
recent years, the neural effects of task load on processing of emotional cause an emotionally relevant stimulus to simply bypass emotional
stimuli have begun to receive more systematic attention. Several circuits. Given that most prior research relied on demanding distractor
studies have demonstrated that activity in emotion processing regions tasks that visually competed with emotional information (Erthal et al.,
of the brain in response to negative emotional stimuli, such as the 2005; Okon-Singer et al., 2007; Pessoa et al., 2002), cognitive load may
amygdalae, depend on the availability of attentional resources for have led participants to overlook some of the emotional information.
processing of these stimuli (Blair et al., 2007; Erk et al., 2007; Okon- In support of this notion, Van Reekum et al. (2007) demonstrated
Singer et al., 2007; Mitchell et al., 2007; Pessoa et al., 2002; Van that individuals spent less time fixating emotion-relevant stimulus
features when they were instructed to reappraise negative visual
⁎ Corresponding author. Fax: +31 30 253 4718. scenes in less emotional terms than when they were instructed to
E-mail address: L.F.vanDillen@uu.nl (L.F. Van Dillen). merely attend to these pictures. This variation in gaze fixation could
1053-8119/$ – see front matter © 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.neuroimage.2009.01.016
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
2 L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx
explain a significant amount of the reduction of activation in than accompanied the display of the emotional stimulus. As a
emotional brain regions during reappraisal compared to passive consequence, any effects of task load on neural responses to the
viewing. Accordingly, attentional deployment strategies (such as an emotional stimuli could not be accounted for by differences in visual
additional task load or attending to non-emotional information) may processing strategies, such as controlling visual attention to emotional
allow people to manage unwanted emotions even before these rather than to non-emotional stimulus features (Van Reekum et al.,
emotions have been fully aroused (Gross, 2001). 2007).
Another possibility, which is highlighted in the present work, is In an experimental study, participants were exposed to neutral or
that task load is capable of down-regulating emotional circuits even negative emotional pictures, after which they performed an arith-
after these circuits have been mobilized. This would imply that metic task that made varying demands on processing resources
emotional brain regions operate quite flexibly, in that they can still be (Ashcraft and Kirk, 2001), and rated their emotional state (Fig. 1).
modulated by contextual demands even when the initial emotional Throughout the experiment, participants' brain responses were
response has unfolded. In line with this, behavioral research suggests monitored using functional magnetic resonance imaging (fMRI). We
that task load can still modulate emotional processing when the task predicted that, relative to exposure to neutral pictures, exposure to
succeeds, rather than accompanies the emotional stimulus (Erber and negative pictures would activate both negative feelings and emotional
Tesser, 1992; Gerin et al., 2006; Glynn et al., 2002; Van Dillen and brain circuits such as the amygdalae and the insulae (Phan et al.,
Koole, 2007). For instance, participants reported less negative feelings 2002). In addition, we predicted that task load would increase
in response to negative pictures when they subsequently tried to solve activation in brain regions implicated in cognitive processing, such
complex rather than simple math equations (Van Dillen and Koole, as the dorsolateral frontal cortex, the superior parietal cortex, and the
2007). Perhaps then, performing a demanding task may similarly dorsal occipital cortex (Duncan and Owen, 2000; Rypma et al., 1999).
down-regulate the unfolding of the emotional brain response to a Most importantly, we hypothesized that cognitive load would
previously displayed emotional stimulus. modulate subjectively reported emotional states and the activation of
The present research was designed to investigate the neural emotional brain circuits. That is, we predicted neural activity in
dynamics by which cognitive load modulates the emotional brain regions implicated in emotional processing to decrease with increases
response, even after this response has already been initiated. in task load. Although the negative pictures should initiate a greater
Importantly, in the present study, the cognitive task followed rather response in emotional brain areas than neutral pictures, we expected
that the unfolding of this response over time would be attenuated
when participants subsequently performed a highly demanding
arithmetic task rather than a moderately demanding arithmetic
task. In short, we predicted that task load would modulate the
unfolding of the emotional brain response, even after this response
was already initiated.
Method
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx 3
ratings between 4.00 and 5.00). Negative pictures included images of Tournoux, 1988). Voxel time series were standardized and corrected
scenes with burn victims, physical assaults, and angry faces. Neutral for serial correlations.
pictures depicted scenes of people in conversation, scenes of nature or FMRI data were first analyzed at each voxel (whole brain) and then
buildings, and neutral faces. The two picture sets were matched as specifically for a number of regions of interest (ROI's). The initial
much as possible on dimensions such as living beings, faces, and random-effects whole-brain analysis served to identify brain regions
complexity. that responded in any way to negative pictures in our task. A multirun/
In each trial, a negative or neutral picture appeared on screen for multisubject GLM design matrix was constructed to model the
4 s. After picture presentation, participants had to perform an relevant brain responses for each run and participant (Friston et al.,
arithmetic task. The complexity of the task was randomly varied. In 1999). The matrix consisted of regressors predicting hemodynamic
half of the trials, the arithmetic task consisted of a more complex responses to each combination of picture valence and task load, as
equation such as ‘2 ⁎ 8 + 12 = 28’. These equations always combined a well as the mood scale and possible arithmetic errors. ROI's were
summation or subtraction with a product or division. In the remaining defined on the basis of the whole-brain activation obtained in
trials, the arithmetic task consisted of a much simpler equation, such response to negative pictures that were followed either by a simple
as ‘7 + 2 = 9’. The latter equations only consisted of either a summation or a complex arithmetic task. We used a “negative-fixation” contrast
or a subtraction. Participants judged whether the equation was correct because we did not want to miss any region responding to negative
by pressing a button with either their left or right index finger. pictures. We also included regressors for arithmetic errors and
Participants had 4 s to make this response. For nine participants, the negative mood reports, as these may activate emotional brain regions
right button represented the correct response and the left button the as well, and may increase the error term if left unexplained. All
incorrect response, while for eight participants this order was regressors were convolved with a standard hemodynamic response
reversed. function, and we analyzed the resulting beta weights at each voxel.
At the end of each trial, participants rated, with a button response, The statistical threshold for this random-effects whole-brain analysis
how unpleasant they felt at that moment (from 1 = not at all, to 8 = very was p = .05 after Bonferroni correction for multiple comparisons.
much). For eight participants, this scale was reversed (ranging from The whole-brain analysis revealed activation clusters in a number
right to left). In between trials, participants were asked to relax. To of regions, such as the bilateral amygdalae, the bilateral inferior
avoid systematic overlap of BOLD responses within and between insulae, the right dorsolateral frontal cortex and the right superior
trials, the interval between the arithmetic task and the mood scale, as parietal cortex (see Table 1 for all ROI's). Peaks of activation in each
well as the interval between the mood scale and the beginning of the region were located, significant voxels surrounding those peaks were
next trial, was set randomly to either 4,000 or 8,000 ms. The duration selected, and their time courses were averaged for a subsequent
of each trial accordingly was 20, 24 or 28 s. The onset of each trial was detailed analysis of the shape of the responses at each ROI.
synchronized to the onset of an fMRI volume. For each ROI and subject, we computed the course of the
Prior to the presentation of the first block, participants were given hemodynamic responses to each valence/load combination by
a block of 16 practice trials to get familiar with the experimental set- means of a deconvolution analysis. A GLM regressor was assigned to
up and the scanner. After the experiment, participants were thanked each of 8 fMRI volumes following the onset of a picture, separately for
for their efforts, debriefed, and paid by the experimenter. each combination of picture valence and task load. This allowed us to
A personal computer controlled presentation of the experimental compute the shape of the hemodynamic response without prior
trials and recorded participants' responses. The experimental trials assumptions, independent for each condition, ROI and subject. These
were presented in E-prime (Psychology Software Tools, Inc., Pitts- deconvolved hemodynamic responses were then broken into time
burgh, USA). Participants responded by pressing fiber-optic buttons frames of 4 s duration (i.e., fMRI volume 1–2, 3–4, 5–6, 7–8), and used
(Lumitouch Photon Control, Burnaby, Canada). in subsequent random-effects analyses of the detailed interactions
between task load and picture valence. The motivation for the 4 s (2
MRI procedure and analysis volumes) time frames was a compromise between temporal specifi-
city and sufficient signal-to-noise. That is, we wanted to analyze the
Brain imaging was performed on a 1.5 T Siemens Sonata scanner
(Siemens Medical Systems, Erlangen, Germany) equipped with a
volume head coil. Functional volumes consisted of 24 near axial slices Table 1
Statistical regions of interest
acquired using an EPI sequence with the following parameters:
repetition time = 2 s, echo time = 50 ms, flip angle = 90°, slice Brain region (Brodmann area, hemisphere) Talairach Volume (ml)
thickness = 4.2 mm, slice gap = 0.84 mm, acquisition matrix = 64 × 64 coordinates (mm)
pixels, in-plane resolution = 3 × 3 mm. Series of 392 volumes were x y z
acquired in each of the four blocks of trials. Images were on-line Dorsolateral frontal cortex (BA 6/44, right) 40 2 30 0.210
motion corrected. After the functional session, a three-dimensional Superior parietal cortex (BA 7, right) 224 − 59 53 0.331
Fasciculus uncinatus (BA 34, left) −28 7 −11 0.353
structural scan was acquired using a T1-weighted MP-RAGE sequence
Dorsal occipital cortex (left) −22 − 79 19 13.979
with the following scanning parameters: repetition time = 2730 ms, Dorsal occipital cortex (right) 23 − 77 18 13.129
echo time = 3.43 ms, inversion time = 1000 ms, flip angle = 7°, 160 Ventral occipital cortex (left) −22 −64 −5 30.286
sagittal slices, slice thickness = 1 mm, acquisition matrix = 256 × 224 Parahippocampal cortex (left) −30 − 25 −14 0.763
pixels, in-plane resolution = 1 × 1 mm. Parahippocampal cortex (right) 34 − 19 −16 0.643
Medial anterior temporal cortex (BA 38, right) 34 −4 −23 0.221
Preprocessing and statistical analyses of the MRI data were
Amygdala (left) − 19 −5 −9 0.211
performed using BrainVoyager 2000 software (Brain Innovation, Amygdala (right) 22 −8 −10 0.450
Maastricht, The Netherlands). The first 2 volumes were discarded in Inferior insula (left) −31 −1 −9 465
order to avoid differences in T1 saturation. Voxel time-series of the Inferior insula (right) 31 −3 −9 0.173
Pulvinar (left) − 18 − 24 0 0.118
remaining volumes were high-pass filtered (0.01 Hz), temporally
Pulvinar (right) 19 − 24 0 0.986
smoothed (2.5 s FWHM Gaussian kernel), and corrected for slice Locus coeruleus 0 − 30 −28 0.244
acquisition times. Finally, volumes were 3D spatially smoothed (6 mm Superior colliculus (left) −6 − 24 0 0.934
FWHM Gaussian kernel). Each functional run was manually co- Superior colliculus (right) 6 − 24 0 0.800
registered to the individual 3D structural scan, re-sampled, and Brain areas responding to negative pictures as revealed by the random-effects whole-
transformed into Talairach space (Goebel et al., 2001; Talairach and brain analysis (p b .05, corrected).
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
4 L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx
time courses of the effects of task load in detail, ideally at each volume, selected regions were then broken into four time frames of two fMRI
but then the number of tests would become large, and the signal-to- volumes each (1–4 s, 5–8 s, 9–12 s, and 13–16 s). These time frames
noise ratio of each test low. Therefore we grouped 2 volumes, which were further tested in subsequent analyses of variance for effects of
allowed us to analyze the time courses of the effects at sufficient picture valence and task load.
temporal detail and with sufficient signal-to-noise. These subsequent analyses revealed that there were no effects of
picture valence or task load during the first time frame following
Results picture onset (t = 1–4 s). Activation in response to the neutral and
negative pictures peaked at time = 6 s after picture onset (see Fig. 2),
Arithmetic performance which corresponds to the delay of the hemodynamic response.
Analyses of variance of the responses during the corresponding time
To investigate whether our manipulation of task load was frame (t = 5–8 s) revealed that, at this point, all regions of interest
successful, we analyzed participants' correct responses and response (except for the locus coeruleus and parahippocampal cortex)
times. Analyses of variance (ANOVA's) revealed that participants responded more strongly to negative pictures than to neutral pictures
performed better on simple arithmetic tasks than on complex (all p b . 05,). Initially, negative pictures thus engaged these brain
arithmetic tasks (M = 97% correct, SD = 1.09 versus M = 84% correct, regions to a greater degree than neutral pictures. The results of these
SD = 5.56; F(1, 16) = 161.59, p b .0001). Participants were also faster on analyses are given in Table 2.
simple than on complex arithmetic equations, F(1, 16) = 858.38, Because the task was presented 4 s after picture onset, effects of
p b .0001 (M = 1929 ms, SD = 160 versus M = 3083 ms, SD = 145). These task load on the response to the picture would be expected at about
effects confirm that the complex arithmetic equations were more 10 s and onwards. In line with this, activation in cognitive
difficult than the simple arithmetic equations. The effects of task processing regions, such as the right dorsolateral frontal cortex,
complexity did not interact with picture valence for neither peaked at time = 10 s, e.g. during the third time frame (t = 9–12 s; see
participants' correct responses (F b 1) nor participants' response Fig. 2). Moreover, if task load indeed modulates the temporal
times (F b 1). unfolding of emotional responses, this may imply an interaction
effect between picture valence and task load on participants' brain
Self-reported negative emotion activity during this time frame. We indeed found statistically
significant interactions between picture valence and task load in
To examine whether task load modulates participants' self- several regions during the third time frame (t = 9–12 s) in right
reported negative emotion, we analyzed the effect of both task load dorsolateral frontal cortex (F(1, 16) = 4.69, p = .046), right superior
(high, low) and picture valence (negative, neutral) on their self- parietal cortex (F(1, 16) = 5.28, p = . 035), the left dorsal occipital
reports, with participants response time differences between complex cortex (F(1,16) = 5.16, p = 0.037), the left amygdala (F(1, 16) = 4.74,
and simple arithmetic trials as a covariate. As expected, this ANCOVA p = 0.045), the right amygdala (F(1, 16) = 4.82, p = .043), and the right
yielded a significant effect of picture valence, F(1, 16) = 30.39, p = .000. inferior insula (F(1, 16) = 7.70, p = .014) (see Table 3 for an overview).
On average, participants reported more negative emotion after trials
with negative pictures than after trials with neutral pictures The effect of task load on brain responses to negative pictures
(respectively, M = 4.55, SD = 1.08 versus M = 3.15, SD = 1.14).
The analysis further yielded the predicted interaction between task Recall that we predicted neural responses in emotion regions to
load and picture valence, F(1, 16) = 16.50, p = .001. To interpret this negative pictures to be attenuated when followed by a complex task
effect, we analyzed the effects of task load separately in each valence (high load) rather than a simple task (low load), while we expected
condition. In line with previous research using the same paradigm neural responses in cognitive processing regions to display the
(Van Dillen and Koole, 2007), there was no effect of task load in the opposite pattern. To test these predictions, we directly compared
neutral trials, F b 1. By contrast, task load had a significant effect in the the responses to negative pictures followed by either a complex or a
negative trials, F(1, 16) = 11.54, p = .004. Participants reported less simple arithmetic task in the regions that showed interactions
negative emotion when negative pictures were followed by a complex between picture valence and task complexity in the remaining two
arithmetic equation rather than a simple arithmetic equation time frames following task onset (i.e. the third and the fourth time
(M = 4.40, SD = 1.07 versus M = 4.70, SD = 1.10). Moreover, response frame). Note that these comparisons result in a positive t-value if the
time differences were unable to account for the effects of task load response is greater in the high load than the low load trials, and a
on negative moods, F(1,15) b 1. Thus, in line with previous findings negative t-value if the response is greater in the low load trials.
(Van Dillen and Koole, 2007), task load modulated participants' self- During the third time frame (t = 9–12 s), the left amygdala (t(16) =
reported negative emotion. −2.91, p = .010) and the right amygdala (t(16) = −1.97, p = .066) showed
weaker responses to negative pictures followed by a complex task
Brain regions involved in both emotion and task-related processes rather than a simple task. During the fourth time frame (t = 13–16 s),
the left amygdala (t(16) = −4.29, p = .001), right amygdala (t(16) = −5.11,
To investigate whether task load modulated the unfolding of p = .0001), and right insula (t(16) = −2.25, p = .039) showed weaker
participants' emotional brain responses, we first identified brain responses to negative pictures followed by a complex task rather than
regions that responded to negative emotional pictures in our task (see a simple task (see Table 4). These regions are known to be involved in
Method section). This yielded a number of areas, such as the bilateral the processing of emotional stimuli (Ochsner et al., 2004; Phan et al.,
amygdalae, the bilateral inferior insulae, the right dorsolateral frontal 2002). Thus, increased cognitive load led to reduced activation in
cortex and the right superior parietal cortex. In Table 1, the regions of emotion circuits.
interest (ROI's) are given, with their xyz-coordinates and cluster sizes. During the third time frame (t = 9–12 s), right dorsolateral frontal
Subsequently, for each region and participant, we computed the cortex (t(16) = 7.94, p b .0001), right superior parietal cortex (t(16) =
course of the hemodynamic response to each picture/load combina- 6.62, p b .0001), and left dorsal occipital cortex (t(16) = 5.01, p = .0001)
tion by means of a deconvolution analysis, for the first 16 s following showed greater activation to negative pictures followed by a complex
picture onset. That is, we estimated the shape of the hemodynamic task rather than a simple task. During the fourth time frame (t = 13–
response by assigning a GLM regressor to each of 8 fMRI volumes 16 s), right dorsolateral frontal cortex (t(16) = 3.43, p = .003), right
following picture onset. In order to assess the timing of the effects of superior parietal cortex (t(16) = 2.43, p = .027), and left dorsal occipital
our manipulations more precisely, the hemodynamic responses of the cortex (t(16) = 2.52, p = .023) showed greater activation to negative
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx 5
Fig. 2. (a) Two coronal views of the averaged brain (n = 17) in Talairach space on which areas are displayed that responded significantly to negative pictures (p b .05, corrected for
multiple comparisons) and in which activity was either greater when task load was low compared to high (upper row; amygdalae [A] and right inferior insula [I]), or in which activity
was greater when task load was high compared to low (lower row; right dorsolateral frontal cortex [Dlfc]). (b) Deconvolved averaged timecourses of the brain responses in the right
amygdala and right dorsolateral frontal cortex for each trial type; negative (Neg) or neutral (Neu) pictures that were followed by either a high (Hi) or a low (Lo) load task. Picture onset
was at time = 0 s, task onset was at time = 4 s. Due to the hemodynamic delay, effects of pictures valence are observed from time = 6 s onwards and effects of task load from time = 10 s
onwards.
pictures followed by a complex task rather than a simple task (see al., 1999). Thus, as expected, performing complex arithmetic sums
Table 4). These regions are known to be involved in cognitive engaged brain regions that support cognitive processes.
processing and to respond to increasing task demands (De Fockert et Recall that we found an effect of task complexity on participants'
al., 2001; Duncan and Owen, 2000; Prabhakaran et al., 2000; Rypma et response times. To investigate whether the effects of task load could
be attributed to response time differences between the complex and
the simple arithmetic equations, we repeated the analyses with
Table 2
participant's response time difference between complex and simple
Regions of interest showing a greater response to negative than to neutral pictures
between 5 and 8 s after picture onset (p b .05)
trials as a covariate. These analyses did not reveal any differential
findings, except for the left amygdala during the third time frame
Brain region (Brodmann area, hemisphere) Talairach Volume Fval (t = 9–12 s). When entered as a covariate, response time differences
coordinates (mm) (ml)
partially accounted for the effect of task complexity on left amygdala
x y z 5–8 s
responses, F(1,15) = 7.22, p = .017. However, a 2 (task complexity)
Dorsolateral frontal cortex (BA 6/44, right) 40 2 30 0.210 61.56⁎
Superior parietal cortex (BA 7, right) 24 − 59 53 0.331 13.63⁎
Fasciculus uncinatus (BA 34, left) −28 7 −11 0.353 43.23⁎
Dorsal occipital cortex (left) −22 − 79 19 13.979 12.13⁎ Table 3
Dorsal occipital cortex (right) 23 −77 18 13.129 8.63 Regions of interest showing a picture valence × task load interaction between 9 and 12 s
Ventral occipital cortex (left) −22 −64 −5 30.286 6.47 after picture onset
Medial anterior temporal 34 −4 −23 0.221 8.53
Brain region (Brodmann area, hemisphere) Talairach Volume Fint
cortex (BA 38, right)
coordinates (mm) (ml)
Amygdala (left) −19 −5 −9 0.211 51.89⁎
Amygdala (right) 22 −8 −10 0.450 5.32 x y z 9–12 s
Inferior insula (left) −31 −1 −9 465 10.94⁎ Dorsolateral frontal cortex (BA 6/44, right) 40 2 30 0.210 4.69
Inferior insula (right) 31 −3 −9 0.173 5.32 Superior parietal cortex (BA 7, right) 24 −59 53 0.331 5.28
Pulvinar (left) −18 −24 0 0.118 25.89⁎ Dorsal occipital cortex (left) −22 −79 19 13.979 5.16
Pulvinar (right) 19 −24 0 0.986 8.31 Amygdala (left) −19 −5 −9 0.211 4.74
Superior colliculus (left) −6 −24 0 0.934 77.66⁎ Amygdala (right) 22 −8 − 10 0.450 4.82
Superior colliculus (right) 6 −24 0 0.800 57.21⁎ Inferior insula (right) 31 −3 −9 0.173 7.70
Note. Fval = F-value of picture valence, all negative N neutral. Degrees of freedom are 1 Note. Fint = F-value of picture valence × task load interaction. Degrees of freedom are 1
and 16, ⁎p b .01. and 16, p b .05.
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
6 L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx 7
valenced pictures were followed by complex arithmetic tasks. Blair, K.S., et al., 2007. Modulation of emotion by cognition and cognition by emotion.
Neuroimage 35, 430–440.
Accordingly, our results reveal an integration effect for processing Brass, M., Derrfuss, J., Forstmann, B., von Cramon, D.Y., 2005. The role of the inferior
negative emotion and task load in right dorsolateral frontal cortex. frontal junction area in cognitive control. Trends Cogn. Sci. 9, 314–316.
One potential explanation for these findings is that the right Bunge, S.A., Kahn, I., Wallis, J.D., Miller, E.K., Wagner, A.D., 2003. Neural circuits subserving
the retrieval and maintenance of abstract rules. J. Neurophysiol. 90, 3419–3428.
dorsolateral frontal cortex is engaged more with increasing task Canli, T., Desmond, J.E., Zhao, Z., Gabrieli, J.D.E., 2002. Sex differences in the neural basis
load, in order to sustain priority to processing of the central task at the of emotional memories. Proceedings of the National Academy of Sciences of the
cost of the further processing of emotionally salient, but task United States of America 99, 10789–10794.
Cohen, J.D., Perlstein, W.M., Braver, T.S., Nystrom, L.E., Noll, D.C., Jonides, J., et al., 1997.
irrelevant negative stimuli (see Erk et al., 2007, for a similar Temporal dynamics of brain activation during a working memory task. Nature 386,
argument). Theorists have proposed that the lateral frontal/prefrontal 604–608.
cortex may function as a key neural substrate of the central bottleneck Cuthbert, B.N., Schupp, H.T., Bradley, M.M., Birbaumer, N., Lang, P.J., 2000. Brain
potentials in affective picture processing: covariation with autonomic arousal and
of information processing (Dux et al., 2006; Herath et al., 2001; Marois
affective report. Biol. Psychol. 52, 95–111.
and Ivanoff, 2005). As such, the lateral frontal cortex may be critical for De Fockert, J.W., Rees, G., Frith, C.D., Lavie, N., 2001. The role of working memory in
mental operations such as cognitive control, decision-making, and visual selective attention. Science 291, 1803–1806.
modality-independent selection of task-relevant information (Badre Dolcos, F., McCarthy, G., 2006. Brain systems mediating cognitive interference by
emotional distraction. J. Neurosci. 26, 2072–2079.
et al., 2005; Brass et al., 2005; Bunge et al., 2003). The present findings Drevets, W.C., Raichle, M.E., 1998. Reciprocal suppression of regional cerebral blood flow
suggest that the role of the lateral frontal cortex in controlling during emotional versus higher cognitive processes: implications for interactions
information processing may extend beyond the cognitive domain, and between emotion and cognition. Cogn. Emot. 12, 353–385.
Duncan, J., Owen, A.M., 2000. Common regions of the human frontal lobe recruited by
may similarly control the processing of emotional information. diverse cognitive demands. Trends Neurosci. 23, 475–483.
Cognitive emotion regulation strategies may build upon more Dux, P.E., Ivanoff, J.G., Asplund, C.L., Marois, R., 2006. Isolation of a central bottleneck of
general information processing systems such as working memory and information processing with time-resolved fMRI. Neuron 52, 1109–1120.
Erber, R., Tesser, A., 1992. Task effort and the regulation of mood — the absorption
cognitive control that typically engage (frontal) cortical regions such
hypothesis. J. Exp. Soc. Psychol. 28, 339–359.
as the dorsolateral frontal cortex, the superior parietal cortex and the Erthal, F.S., De Oliviera, L., Mocaiber, I., Pereira, M.G., Machado-Pinheiro, W., Volchan, E.,
dorsal occipital cortex (Ochsner and Gross, 2008). Indeed, a recent 2005. Load-dependent modulation of affective picture processing. Cogn. Affect.
Behav. Neurosci. 5, 388–395.
study revealed a role of the lateral frontal cortex in inhibiting
Erk, S., Kleczar, A., Walter, H., 2007. Valence specific regulation effects in a working
emotional distraction in healthy adults during a working memory memory task with emotional context. Neuroimage 37, 623–632.
task (Johnson et al., 2005; Dolcos and McCarthy, 2006). Performing a Friston, K.J., Holmes, A.P., Price, C.J., Buchel, C., Worsley, K.J., 1999. Multisubject fMRI
working memory task engaged the lateral frontal cortex to a greater studies and conjunction analyses. Neuroimage 10, 385–396.
Gerin, W., Davidson, K.W., Christenfeld, N.J.S., Goyal, T., Schwartz, J.E., 2006. The role of
degree when the participants were distracted by negative emotional angry rumination and distraction in blood pressure recovery from emotional
pictures rather than neutral or scrambled pictures. Moreover, arousal. Psychosom. Med. 68, 64–72.
participants who displayed greater activity to emotional distracters Glynn, L.M., Christenfeld, N., Gerin, W., 2002. The role of rumination in recovery from
reactivity: cardiovascular consequences of emotional states. Psychosom. Med. 64,
in the lateral frontal cortex judged emotional distracters as less 714–726.
distracting and less emotional. Goebel, R., Muckli, L., Zanella, F.E., Singer, W., Stoerig, P., 2001. Sustained extrastriate
The present findings potentially have important implications for cortical activation without visual awareness revealed by fMRI studies of
hemianopic patients. Vis. Res. 41, 1459–1474.
psychopathologies that are characterized by both emotional and Gross, J.J., 2001. Emotion regulation in adulthood: timing is everything. Curr. Dir.
cognitive deficits, like depression and anxiety (Harvey et al., 2005; Luu Psychol. Sci. 10, 214–219.
et al., 1998; Mayberg et al., 1999). Emotional stimuli can have a lasting Harvey, P.O., Fossati, P., Pochon, J.B., Levy, R., LeBastard, G., Leheriey, S., et al., 2005.
Cognitive control and brain resources in major depression: an fMRI study using the
effect on amygdala activity (Cuthbert et al., 2000), specifically in
n-back task. Neuroimage 26, 860–869.
depressed individuals (Siegle et al., 2002). Moreover, research has Herath, P., Klingberg, T., Young, J., Amunts, K., Roland, P., 2001. Neural correlates of dual
shown that people are inclined to ponder over a negative experience, task interference can be dissociated from those of divided attention: an fMRI study.
Cereb. Cortex 11, 796–805.
which may subsequently intensify and prolong people's negative
Johnson, M.K., Raye, C.L., Mitchell, K.J., Greene, E.J., Cunningham, W.A., Sanislow, C.A.,
emotional states (Nolen-Hoeksema et al., 1993). Placing people in a 2005. Using fMRI to investigate a component process of reflection: prefrontal
quiet environment in order to let them ‘cool down’ from an emotional correlates of refreshing a just-activated representation. Cogn. Affect. Behav.
response may therefore not necessarily result in a more neutral state of Neurosci. 5, 339–361.
Keightley, M.L., Winocur, G., Graham, S.J., Mayberg, H.S., Hevenor, S.J., Grady, C.L.,
mind. Although more research is needed in this area, it is conceivable 2003. An fMRI study investigating cognitive modulation of brain regions
that cognitively demanding tasks may eventually be used as a associated with emotional processing of visual stimuli. Neuropsychologia 41,
therapeutic tool. Having people perform an engaging task may 585–596.
Lang, P.J., Bradley, M.M., Cuthbert, B.N., 2001. International Affective Picture System
alleviate the intensity of acute emotional responses, such that people (IAPS): Instruction Manual and Affective Ratings. Technical report A-5. The Center
become more receptive to long-term therapeutic interventions. for Research in Psychophysiology, University of Florida.
More broadly speaking, the present work attests to the close Lyubomirsky, S., Tucker, K.L., Caldwell, N.D., Berg, K., 1999. Why ruminators are poor
problem solvers: clues from the phenomenology of dysphoric rumination. J. Pers.
coordination between cognitive and emotional functioning. Whereas Soc. Psychol. 77, 041–1060.
both cognitive and emotional brain circuits are involved in processing Luu, P., Tucker, D.M., Derryberry, D., 1998. Anxiety and the motivational basis of working
negative pictures, our findings demonstrated that further engagement memory. Cogn. Ther. Res. 22, 577–594.
Mackiewicz, K.L., Sarinopoulos, I., Cleven, K.L., Nitschke, J.B., 2006. The effect of
of cognitive circuits by a subsequent task go hand in hand with a anticipation and the specificity of sex differences for amygdala and hippocampus
decrease in activity in emotional circuits (Drevets and Raichle, 1998; function in emotional memory. Proceedings of the National Academy of Sciences of
Keightley et al., 2003; Mayberg et al., 1999), especially when the United States of America 103, 14200–14205.
Marois, R., Ivanoff, J., 2005. Capacity limits of information processing in the brain.
processing load of the task is high. Depending on the threats and
Trends Cogn. Sci. 9, 296–304.
challenges that people face, cognitive and emotional systems are Mayberg, H.S., Liotti, M., Brannan, S.K., McGinnis, S., Mahurin, R.K., Jerabek, P.A., et al.,
recruited in a flexible manner, such that people can deal effectively 1999. Reciprocal limbic-cortical function and negative mood: converging PET
with the ever changing demands of their environment. findings in depression and normal sadness. Am. J. Psychiatry 156, 675–682.
Mitchell, D.G., Nakic, M., Fridberg, D., Kamel, N., Pine, D.S., Blair, R.J., 2007. The impact of
processing load on emotion. Neuroimage 34, 1299–1309.
References Morrow, J., Nolen-Hoeksema, S., 1990. Effects of responses to depression on the
remediation of depressive affect. J. Pers. Soc. Psychol. 58, 519–527.
Ashcraft, M.H., Kirk, E.P., 2001. The relationships among working memory, math Nolen-Hoeksema, S., Morrow, J., Fredrickson, B.L., 1993. Response styles and the
anxiety, and performance. J. Exp. Psychol. Gen. 130, 224–237. duration of episodes of depressed mood. J. Abnorm. Psychology 102, 20–28.
Badre, D., Poldrack, R.A., Pare-Blagoev, E.J., Insler, R.Z., Wagner, A.D., 2005. Dissociable Ochsner, K.N., Ray, R.D., Cooper, J.C., Robertson, E.R., Chopra, S., Gabrieli, J.D.E., et al.,
controlled retrieval and generalized selection mechanisms in ventrolateral 2004. For better or for worse: neural systems supporting the cognitive down- and
prefrontal cortex. Neuron 47, 907–918. up-regulation of negative emotion. Neuroimage 23, 483–499.
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
ARTICLE IN PRESS
8 L.F. Van Dillen et al. / NeuroImage xxx (2009) xxx–xxx
Ochsner, K.N., Gross, J.J., 2005. The cognitive control of emotion. Trends Cogn. Sci. 9, Sapolsky, R.M., 1994. Why Zebras Don't Get Ulcers: A guide to Stress, Stress-related
242–249. Diseases and Coping. WH Freeman, New York.
Ochsner, K.N., Gross, J.J., 2008. Cognitive emotion regulation: insights from social Siegle, G.J., Steinhauer, S.R., Thase, M.E., Stenger, V.A., Carter, C.S., 2002. Can't shake
cognitive and affective neuroscience. Curr. Dir. Psychol. Sci. 17, 153–158. that feeling: event-related fMRI assessment of sustained amygdala activity in
Öhman, A., 2007. Has evolution primed humans to “beware the beast”? Proc. Natl. Acad. response to emotional information in depressed individuals. Biol. Psychiatry 51,
Sci. U. S. A. 104, 16396–16397. 693–707.
Okon-Singer, H., Tzelgov, J., Henik, A., 2007. Distinguishing between automaticity and Talairach, J., Tournoux, P., 1988. Co-planar Stereotaxic Atlas of the Human Brain. M.
attention in the processing of emotionally significant stimuli. Emotion 7, 147–157. Rayport, Transl, Thieme, New York.
Pessoa, L., 2008. On the relationship between emotion and cognition. Nat. Rev., Van Dillen, L.F., Koole, S.L., 2007. Clearing the mind: a working memory model of
Neurosci. 9, 148–158. distraction from negative feelings. Emotion 7, 715–723.
Pessoa, L., McKenna, M., Gutierrez, E., Ungerleider, L.G., 2002. Neural processing of Van Dillen, L.F., and Koole, S.L., in press. How automatic is “automatic vigilance”?. The
emotional faces requires attention. Proc. Natl. Acad. Sci. U. S. A. 99, 11458–11463. role of working memory in interference of negative information. Cognition and
Phan, K.L., Wager, T., Taylor, S.F., Liberzon, I., 2002. Functional neuroanatomy of emotion: a Emotion.
meta-analysis of emotion activation studies in PET and fMRI. Neuroimage 16, 331–348. Van Reekum, C.M., Johnstone, T., Urry, H.L., Thurow, M.E., Schaefer, H.S., Alexander, A.L.,
Prabhakaran, V., Narayanan, K., Zhao, Z., Gabrieli, J.D.E., 2000. Integration of diverse Davidson, R.J., 2007. Gaze fixation predicts brain activation during the voluntary
information in working memory within the frontal lobe. Nat. Neurosci. 3, 85–90. regulation of picture-induced negative affect. Neuroimage 36, 1041–1055.
Rypma, B., Prabhakaran, V., Desmond, J.E., Glover, G.H., Gabrieli, J.D.E., 1999. Load- Wrase, J., Klein, S., Gruesser, S.M., Hermann, D., Flor, H., Mann, K., et al., 2003. Gender
dependent roles of frontal brain regions in the maintenance of working memory. differences in the processing of standardized emotional visual stimuli in humans: a
Neuroimage 9, 216–226. functional magnetic resonance imaging study. Neurosci. Lett. 348, 41–45.
Please cite this article as: Van Dillen, L.F. et al. Tuning down the emotional brain: An fMRI study of the effects of cognitive load on the
processing of affective images. NeuroImage (2009), doi:10.1016/j.neuroimage.2009.01.016
informs ®
Vol. 26, No. 1, January–February 2007, pp. 130–141
issn 0732-2399 ! eissn 1526-548X ! 07 ! 2601 ! 0130 doi 10.1287/mksc.1060.0212
© 2007 INFORMS
Jaap M. J. Murre
Department of Psychonomy, University of Amsterdam, Roetersstraat 15, 1018 WB Amsterdam, The Netherlands,
jaap@murre.com
W e introduce a new (point process) model of learning and forgetting, inspired by the structures of the
brain, that we apply to model long-term memory for advertising and brand name recall. Recall-probability
functions derived from the model are tested with classic data by Zielske (1959), as well as advertisement content
and brand name recall data of a Dutch study that tracked over 40 campaigns of TV commercials. Data fits
and cross-validation results indicate that the recall functions serve as a good first approximation for aggregate
behavior. The shapes of optimal GRP schedules, which are obtained by maximizing a recall measure, are strongly
related to the model parameters and corresponding memory processes. Comparisons with existing models in
the literature indicate that a neurobiologically motivated model may give a more realistic description of memory
for advertisements.
Key words: advertising; memory; impact; scheduling; bursting; dripping; massed and spaced learning; point
processes
History: This paper was received May 28, 2003, and was with the authors 27 months for 3 revisions processed
by Joel Huber.
derive different expressions for learning and forget- also compare the results with a model presented in
ting of both advertisement content and brand name Naik et al. (1998).
recall. The memory chain model has already been
applied as such to hundreds of memory retention data
The Memory Chain Model
sets (e.g., Chessa and Murre 2004, Janssen et al. 2005).
The memory model is built upon two main concepts.
There are a few models in the psychological liter-
During exposure (e.g., to an advertisement), a mem-
ature that have some similarity to the memory chain
ory is encoded as a number of representations, each
model (Atkinson and Shiffrin 1968, Murdock 1974). of which captures characteristics of the memorized
However, these models only describe situations with item. Representations may be activated over time by
a single learning trial and subsequent forgetting, in a memory cue, for instance, by a product category
which the short-term to long-term memory processes when a brand name has to be recalled. We model the
are rather restrictive. While the existing models are process of activated memory representations in time
limited to one or two memory stores, the memory as a point process (Daley and Vere-Jones 1988, Diggle
chain model can handle any number of stores. Two 1983, Stoyan et al. 1987). We will consider a Poisson
other types of modeling approaches are based on: point process; that is, numbers of representations acti-
(1) Markov chains (Murdock 1974), where the states vated in time intervals have a Poisson distribution,
represent the level of mastery of some item memo- and numbers in disjoint intervals are independent.
rized; and (2) regression procedures for fitting func- Point processes have been used for modeling neural
tions to retention data (Rubin and Wenzel 1996). In spike trains (Abeles 1991).
the latter approach, the functions used are not derived The second concept is motivated by neurobiological
from an underlying formal theory of memory, as is evidence that memories may be stored in one or more
the case with the memory chain model, but are sim- parts of the brain, which we will denote in abstract
ply selected and fitted. Markov models only formalize terms by memory stores. Two processes characterize
the effect of learning at successive trials, but do not each store: a process of memory decline or loss, and
model forgetting between intermediate trials, which an induction process, which transfers representations
the memory chain model does take into account. from one store to another in a feed-forward fashion
The data on which the recall functions for adver- (Figure 1). A memory can be strengthened in a sec-
tisement content and brand name will be tested and ond store, for example, through induction by activa-
validated are presented in the third and fourth sec- tion. This is an induction process where, for example,
tions. The results of the data fits and optimization will a rapidly decaying neural group activates neurons in
be discussed in the discussion section, where we will a less rapidly decaying part of the brain. An exam-
Figure 1 (a) Storage Systems for a Memory at Different Time Scales, with Feed-Forward Induction Between and Decline Within Stores. (b) Abstract
Representation Used in the Memory Chain Model
(b)
ple of the latter is the dorsolateral prefrontal cortex, inductions (which in some experiments can be thought
which is assumed to hold a working memory rep- of as rehearsals), of which the intensity is proportional
resentation for a few seconds to minutes (Goldman- to the intensity in the first store, that is, !2 !1 r˜1 "t#,
Rakic 1992, 1995). From a neurobiological point of where !2 is a (time-independent) induction rate. An
view, these processes are part of a cascade of induc- induction that occurs at time % induces a memory rep-
tion and decline processes that take place at different resentation at time t ≥ % according to probability den-
time scales (McGaugh 2000). sity function a2 exp"−a2 "t − %## in Store 2, which we
We will use models with two stores, at most, in the write as f . This function specifies the decline rate in
fits to advertising data in order to limit the number the second store, which we assume to be smaller than
of parameters. Because all the data fall in time ranges in the first store "a2 < a1 #. The intensity function r2 of
in the order of weeks and months, the two stores of the second store is a convolution of !2 !1 r˜1 "t# and f ,
the model will most likely reflect neural processes in which yields the expression:1
the hippocampus (Store 1) and the neocortex (Store 2).
There are no data at time lags shorter than a week, so !1 !˜ 2
r2 "t# = "exp"−a2 t# − exp"−a1 t##$ (2)
that the short-term stores in the upper panel of Fig- a1 − a2
ure 1 are in fact covered by Store 1 in our model fits. where !˜ 2 = !2 a2 , for which we will continue to
It should therefore be noticed that there is not neces- write !2 .
sarily a one-to-one correspondence between the neu-
ral stores (Figure 1(a)) and the stores of the memory Retrieval. Memory stores are searched for evidence
chain model (Figure 1(b)). at a recall test. Stores can be searched either simul-
taneously or sequentially; it can be proved that both
A Single Learning Trial strategies lead to identical expressions for recall prob-
The memory chain model formalizes memory and re- ability. The effectiveness of the search process and of
tention as a process that consists of four stages, which the retrieval cue(s) used for the search is expressed
are described below. by a single parameter q, which can be interpreted
Encoding. During stimulus exposure, the number as the probability that a memory representation will
of memory representations eventually reaches a mean be cued. Cueing has the following effect on the total
value, or intensity, !1 . Encoded memory representa- intensity r1 + r2 over the two stores:2
tions are stored in the first memory store (i.e., Store 1). r"t# = "r1 "t# + r2 "t##q& (3)
Representations are not labeled, so the model for-
malizes how much is stored. Memory representations This model characteristic does not necessarily imply
could be seen as critical features or relevant details of that cue effectiveness is constant over time. The reten-
an advertisement. tion functions that follow from our memory model
Storage. After encoding, a decline process is acti- are invariant under exponentially varying cue effec-
vated, for example, by spontaneous decay, by the tiveness. This assumption does not lead to addi-
overwriting of other learned items, by neural noise, tional degrees of freedom in the model, because cue
or by other factors. An encoded memory represen- effectiveness can be subsumed under the exponential
tation is still available at time t since the end of a memory decline functions of the stores. We will com-
learning trial with probability denoted by r˜1 "t#, which bine q with the encoding parameter !1 , for which we
will be called the decline function of Store 1. Memory will continue to write !1 , because the effect of cue-
in Store 1 is assumed to be a nonhomogeneous Poisson ing is not varied systematically in the studies reported
process with intensity function r1 "t# = !1 r˜1 "t#. We will here.
assume an exponential decline function, so that Recall. Our model includes a fourth stage, which
r1 "t# = !1 exp"−a1 t#$ (1) can be considered a decision process that relates re-
trieval to recall. The number of retrieved representa-
where a1 > 0. Data obtained from laboratory exper- tions is compared with a threshold b. Unless stated
iments that intend to measure short-term memory otherwise, we assume that successful recall takes
decline through the classical Brown-Peterson learning place if a store exists with at least one cued repre-
and distraction task support an exponential decline sentation. In some cases, subjects require more than
(Peterson and Peterson 1959, Murdock 1961). Evi- one representation for recall. From laboratory exper-
dence at the neural level is offered by the long-term
potentiation or synaptic growth and decline data in 1
For more details, see http://www.neuromod.org/staff/murre/
the hippocampus of young and old rats (Barnes and mcmforgettingdraft.doc.
McNaughton 1980). 2
The formalisms of memory decline and cueing, as shown in (1)
Memory representations may be induced from and (3), are in fact independent thinnings of Poisson processes with
Store 1 to Store 2, where they form a new point pro- intensities !1 and r1 + r2 , respectively, which again result in Poisson
cess. This point process arises from a time-series of processes with thinned intensities (Stoyan et al. 1987).
Chessa and Murre: Neurocognitive Model of Advertisement Content and Brand Name Recall
Marketing Science 26(1), pp. 130–141, © 2007 INFORMS 133
Figure 2 Illustration of Memory Processes in a Two-Store Model then describe the initial encoding during the learning
Store 1 Store 2 trial as follows:3
Parameters
!0 Intensity of homogeneous (base rate) Expected number of memory representations still left
Poisson process from previous campaigns
v Learning rate per unit of learning time Expected number of memory representations formed
in time-continuous setting per GRP during encoding of ads
!1 Initial encoding (intensity after first trial) Expected number of memory representations at the
end of the first ad exposure or GRP pulse
rmax Upper bound on the intensity of Store 1 Finite rmax leads to decreasing encoding over ad
"rmax ≥ !1 # exposures and to equal contributions otherwise
!2 Induction rate from Store 1 to Store 2 Constant transfer rate of memory representations
of ads from Store 1 to Store 2
a1 $ a2 Decline rates in Stores 1 and 2, respectively Relative loss of memory representations
in Stores 1 and 2
b Recall threshold Minimum number of retrieved representations, e.g.,
critical features of ads, needed for successful recall
Functions
r "t# Intensity function summed over stores Specifies expected number of memory representations
taken over both stores as a function of test lag t
r1 "t#$ r2 "t# Intensity functions of Stores 1 and 2 Same as r "t#, but for Stores 1 and 2 separately
r˜1 "t#$ r˜2 "t# Decline functions of Stores 1 and 2 Express changes in the encoded number of memory
representations for Stores 1 and 2 at test lags t
!"L# "l# Encoding for trial L with presentation time l Expected number of memory representations for ad
exposure L with l GRPs
p"t# Recall-probability function Probability of successfully recalling the content or
the name of a brand in a commercial at test lag t
Meaning in applied context
Variables
t Retention lag or interval Time since last ad exposure
l Learning or exposure time GRPs in SPOT data
ti Time at which learning trial i is presented Moment at which an ad is exposed for the ith time
L Number of learning trials Number of advertisements or GRP pulses in campaign
the simplifying assumption that memory processes are but that after about 17 weeks, recall improves for the
the same for different subjects. We make this choice four-week schedule and stays at a higher level after
because we propose a new model in this paper, and we the end of the campaign. The average recall percent-
want to find out to what extent this model is already age calculated over a period of one year is higher for
able to fit the data with a limited number of parameters. the four-week schedule. This implies that the spaced
schedule would be preferred on two criteria, namely,
(1) mid-term and long-term recall percentages after
The Zielske Study the end of an advertising campaign, and (2) the aver-
The Data age recall percentage calculated over a period of one
In the literature known to us, the only extensive year.
study about the recall of printed advertising material
for repeated advertisements is the classic study by The Fitted-Recall Functions
Zielske (1959). Housewives were repeatedly sent the We fitted recall-probability function (7), which holds
same printed advertisement, according to two sched- for recall threshold b = 1, and recall-probability func-
ules. One group received the advertisement once a tions with higher recall thresholds, which have the
week for 13 successive weeks, while the second group form of (4). A function was fitted simultaneously to
received the advertisement 13 times with a uniform both schedules. This means that we used the same
spacing of four weeks. To obtain data about the recall parameter values in the fits to both schedules, because
of the advertisement in time, both groups were sub- differences in recall for two different schedules should
divided into smaller groups that were interviewed be explained by one underlying memory mechanism,
in specific weeks. The interviews were organized in and should merely represent the effect of intermediate
such a way that data about learning and forget- forgetting between two successive advertisements.
ting between and after the exposures were obtained. The parameters were estimated by minimizing a
Simon (1979) analyzed the data and concluded that chi-square statistic, which is the sum of conventional
the one-week schedule reaches a higher peak in recall, chi-square statistics computed for every advertising
Chessa and Murre: Neurocognitive Model of Advertisement Content and Brand Name Recall
Marketing Science 26(1), pp. 130–141, © 2007 INFORMS 135
Figure 3 Fits of a Two-Store Model with Saturated Learning to The expected recall percentage over a one-year
Zielske’s Recall Data (Dots) of a Printed Advertisement period can be calculated exactly for the two sched-
(a) 1.0 ules simply by summing the expected recall over
the weeks. On the basis of this criterion, the four-
0.8
week schedule is one-and-a-half times more effec-
tive. An advantage of using a model is that other
Recall probability
Figure 4 Impact (Solid Circles) and Brand Name Recall (Open Circles) as a Function of GRPs (Gray Bars) for Four Brands (a: Meal, b: Detergent,
c: Meal Sauce, d: Candy)
Effectiveness (%)
Effectiveness (%)
50 50
60 40 60 40
GRP
GRP
40 30 40 30
20 20
20 20
10 10
0 0 0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
Week Week
Effectiveness (%)
Effectiveness (%)
50 50
60 40 60 40
GRP
GRP
40 30 30
40
20 20
20 20
10 10
0 0 0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
Week Week
Notes. The model fits are shown as lines (thick line for impact, thin line for brand name recall): (a) a two-store model without saturation, rapid forgetting in
Store 1, slow forgetting in Store 2, and a recall threshold equal to 1; (b) strong forgetting, no saturation; (c) no forgetting, no saturation; (d) strong saturation
with slow forgetting.
with a one-store model. Fifteen brands only required in a composite test. The result will then be a fit with
the stationary base rate process. For these brands, the a low R2 value, which will increase with additional
corresponding advertising campaigns did not have parameters, causing an opposite effect compared to !.
any effect on brand name recall. The results thus show Such fits also influenced the results in Table 2, as
numerous examples of TV commercials that improve a comparison between the values for maximum !
impact but not brand name recall. (which corresponds with the minimized chi-square
The question that we will try to answer in the statistic) and the number of rejected functions on the
next subsection is whether the memory processes that one hand, and the R2 values on the other hand, illus-
underlie the fitted-recall functions imply GRP sched- trates.
ules that improve the average impact and brand name The memory chain model gives better fits than the
recall over the tracking periods. For example, Fig- model by Naik et al. (1998). We will analyze and com-
ure 4(d) shows a brand with strong saturation, which pare the two models in the discussion section. Table 2
is evidenced by the flat part of the impact and brand also indicates that the model used for the cross valida-
name recall functions during repeated advertising. tion gives results that are in close agreement with the
Should one choose a schedule with an increased num- model fitted to the complete data sets. The parameter
ber of exposures, as is shown in Figure 4(d) during estimates turned out to be stable when comparing the
the first half of the campaign period; or a uniformly estimates for the restricted data and the full data. In
spaced schedule over a longer advertising period, more than 70% of the brands, the relative change in
given the decreasing contributions of repeated expo- the parameter values remained within 10%. The fit-
sures to recall? Before answering such questions, we ted functions were hardly affected in the majority of
will subject the recall functions to a cross-validation the remaining cases, because the estimated parameter
study. values were very small.
To assess the predictive performance of recall func-
tions, we perform a cross-validation study, where a Optimization of Advertisement Scheduling
part of the data is used to estimate model parame- We used the recall-probability functions that were fit-
ters and the remaining data is predicted by using the ted to impact and brand name recall to investigate
functions fitted to the restricted data sets. We left out whether more effective campaigns exist with a dif-
the data for the last five campaign weeks for every ferent GRP scheduling, under the same total amount
brand. Table 2 shows the values of three statistics for of GRP and campaign period. Different criteria can
the recall functions fitted to the complete impact data be chosen for optimization: impact and brand name
sets and to the restricted data sets used for the cross recall in a specific target week (e.g., when advertising
validation. We also listed the results for the model a movie premiere), or the average impact and brand
proposed by Naik et al. (1998). Table 2 shows average name recall over a time interval (stimulate sales con-
values over all brands in each of four product classes. sistently). Here we focus on the latter.
We included R2 because it is widely used in model An increase of the average expected impact by more
fits. However, we would like to emphasize that it does than 2% was found for six brands, and by more than
not always give clear indications about the quality of 5% for two brands. Regarding brand name recall,
a fit. A low R2 value may correspond to test sizes two brands improve by 4%–5%, while the remain-
! > 0"05. This depends on factors such as the sample ing brands improve up to 1%. A brand for which the
size per data point and the number of model param- average impact improves significantly under the opti-
eters used in a fit. For example, a data set with small mal GRP distribution is a brand that is fitted with a
variation may be fitted well by a constant function, as one-store model with saturation and rapid forgetting
additional parameters could decrease the test size ! (Figure 5(d)). For this brand, the average expected
Table 2 Values of Three Statistics for Three Models Fitted to the Impact Data (SPOT): Average Maximum
Test Size ! for Minimized Chi-Square Statistics, the Number of Rejected Functions "! = 0#05$, and
Average R2 (for Brands Grouped into Four Product Classes)
Product class Alpha Rejected R-square Alpha Rejected R-square Alpha Rejected R-square
Notes. The number of brands per product class is given between parentheses in the first column. MCM = memory
chain model; NMS = model of Naik et al. (1998).
Chessa and Murre: Neurocognitive Model of Advertisement Content and Brand Name Recall
138 Marketing Science 26(1), pp. 130–141, © 2007 INFORMS
Figure 5 GRP Distributions (Gray Bars) for Maximized Average Expected Impact (Solid Line, Left) and Brand Name Recall (Solid Line, Right) over
the Campaign Period
Impact (%)
20 20
60 60
GRP
GRP
40 40
10 10
20 20
0 0 0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
Week Week
Impact (%)
60 40 60 40
GRP
GRP
40 30 40 30
20 20
20 20
10 10
0 0 0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
Week Week
20 20
600 600
GRP
GRP
400 400
10 10
200 200
0 0 0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
Week Week
40 40
60 60
GRP
GRP
30 30
40 40
20 20
20 10 20 10
0 0 0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
Week Week
Notes. Data are for the same fitted functions as in Figure 4. The original impact and brand name recall data (dots) are also shown.
impact increases from 26.3% to 33%, while brand no additional encoding during the campaign), drip-
name recall increases from 50.1% to 54.8% under the ping schedules with small amounts of GRP will be
optimal schedule. Instead of the bell-shaped GRP most cost effective. This situation arises three times
distribution shown in Figure 4(d), the memory pro- for impact and 15 times for brand name recall. Brands
cesses that underlie this commercial and brand name that are memorized according to a one-store model
rather suggest a uniformly spaced, dripping schedule. with strong forgetting or that suffer from strongly
Significant increases were also found for the brands saturated learning benefit most from a spaced dis-
shown in Figures 5(a) and 5(c). tribution of GRP (dripping). Brands that are learned
There is obviously a strong relation between the without saturation and that show very slow forget-
parameter values of a fitted-recall probability func- ting afterwards benefit most from a single burst at the
tion and the type of optimal advertisement schedul- beginning of a campaign (bursting). The same holds
ing, which can be classified as follows. For brands when the recall threshold increases (new brands or
that only consist of a stationary base rate process (i.e., campaigns). In this way, the increased impact will
Chessa and Murre: Neurocognitive Model of Advertisement Content and Brand Name Recall
Marketing Science 26(1), pp. 130–141, © 2007 INFORMS 139
Table 3 Implications of Different Model Versions for Scheduling may be quite homogeneous over respondents. Depar-
tures from these properties will imply an overesti-
Base Recall
rate Learning Forgetting threshold Schedule mation of the recall functions used here at aggregate
level, because recall probability is a concave function
Exists Very slow Drip of memory intensity. Explicit modeling of variations
Rapid Low Drip
Saturated Low Drip
over respondents should reveal the precise amount of
Not saturated Very slow Burst overestimation. Because the overestimation applies to
Not saturated Moderate, no Low Burst, then drip the entire course of the recall-probability function, we
consolidation expect that the effects on the optimal schedule types
Not saturated Rapid, with Low Burst, then drip shown in Table 3 will be limited.
consolidation High Burst
The fits of our model to the SPOT data have shown
that the feed-forward structure of the notion of mem-
fall within the campaign period as much as possi- ory stores, by which we allow memories to consol-
ble. Other situations give rise to a mixture of bursting idate to one or more stores, is an essential element
and dripping. Examples are brands and commercials in the applicability of our model. There are numer-
that are memorized according to a two-store model ous instances in the SPOT data that show a decline
without saturated learning, with strong forgetting in of impact and brand familiarity during advertising,
the first store, and with slow or no forgetting in the which then stabilize at constant values after a cam-
second store. Here, on the one hand, the memory con- paign. From our model perspective, these cases all
solidation process of the second store favors a single result from a rapid decline in the first memory store,
burst at the start of the campaign, but the strong for- from which a part of information is consolidated to a
getting in the first store requires a uniform distribu- second store.
tion of GRPs over the campaign weeks. A mixture of As a comparison, we implemented and applied the
bursting and dripping also applies to one-store mod- model proposed by Naik et al. (1998) to the SPOT
els with moderate forgetting. Examples of the above- data. Their model gave considerably poorer fits with
mentioned brands and optimal GRP distributions are as many free parameters as in the memory chain
shown in Figure 5. Table 3 summarizes the above model, with about 22% of the fits rejected (Table 2).
implications of the model for optimal advertisement The poorer fits were caused to a large extent by the
scheduling. model rejections for situations described in the previ-
ous paragraph. The use of one forgetting parameter
(! in Naik et al. 1998) is not sufficient to capture com-
Discussion plex behaviors of advertising effectiveness during and
In this paper, we presented and applied a neurobi- after campaigns. Naik et al.’s model must fit varia-
ologically informed model to the recall of advertis- tions in memory decline during and after a campaign
ing material as a function of multiple exposures to with the same value for the forgetting parameter !.
the same advertisement (print or TV). To our knowl- This leads to overly smooth fits, and an inability to
edge, it is the first application of neurobiological pro-
cess models to advertising. The same processes are
Figure 6 Fits of a Two-Store Memory Chain Model (Thick Line) and of
involved in a wide range of learning and forgetting the Model of Naik et al. (Thin Line), with the Same Number
situations to which our model has been fitted success- of Parameters, to Impact Data (Dots) of a Meal Sauce Brand
fully. This study shows that the model also captures 100 40
the learning and forgetting of advertising material.
An important and essential feature of the model is
its ability to describe situations with multiple learn- 80
30
ing trials, which was demonstrated in previous stud-
ies with model fits to psychological laboratory data,
Impact (%)
60
GRP
handle changes in memory decline over time due 2004). We will also consider explicit modeling of com-
to, for instance, long-term memory consolidation (see petition effects. In this respect, it could be interesting
Figure 6). This is also true for other models, such to integrate the model part on copy wearout in Naik
as the Vidale-Wolfe and Nerlove-Arrow models and et al. (1998) with our model.
Brandaid (see the paper by Naik et al., where expres-
sions for different models are listed in Table 2). Acknowledgments
Our cross-validation study showed that the param- This research was supported by NWO, the Netherlands
eter estimates are stable over time. The models thus Society for Scientific Research. The authors are very grate-
ful to Jan Ligthart for providing the SPOT data and to Giep
used to optimize advertisement scheduling showed
Franzen for many helpful discussions. They also thank the
that optimal schedules fall in one of three types: (1) a reviewers for their many useful and detailed comments.
GRP burst at the start of the campaign, (2) a uniform,
dripping GRP distribution, (3) a combination of burst-
ing and dripping, with the greatest GRP expenditure References
at the start of the campaign and a more even distri- Abeles, M. 1991. Corticonics. Cambridge University Press, Cam-
bution of GRP in later weeks. The implications of the bridge, UK.
memory model parameters for optimal scheduling, as Akçura, M. T., F. F. Gönül, E. Petrova. 2004. Consumer learning and
brand valuation: An application on over-the-counter drugs.
shown in Table 3, could thus be used to target adver- Marketing Sci. 23 156–169.
tising to different groups of products or brands (see Atkinson, R. C., R. M. Shiffrin. 1968. Human memory: A pro-
also Iyer et al. 2005). posed system and its control processes. K. W. Spence, J. T.
The optimal schedules that we obtained are all de- Spence, eds. The Psychology of Learning and Motivation: Advances
in Research and Theory, Vol. 2. Academic Press, New York.
creasing or constant in GRP as a function of time. This
Baddeley, A. D., D. J. A. Longman. 1978. The influence of length
does not mean, however, that the schedules implied and frequency of training session on the rate of learning to
by our model are all monotonic (e.g., decreasing). Fur- type. Ergonomics 21 627–635.
ther model analyses have shown that optimal GRP Bahrick, H. P., L. E. Bahrick, A. S. Bahrick, P. E. Bahrick. 1993. Main-
schedules may also be nonmonotonic. The critical tenance of foreign language vocabulary and the spacing effect.
Psych. Sci. 4 316–321.
parameters in this situation are the decline parame-
Barnes, C. A., B. L. McNaughton. 1980. Spatial memory and hip-
ters of the memory stores and the rate of induction pocampal synaptic plasticity. D. Stein, ed. The Psychobiology
between the first and second stores. Nonmonotonic of Aging: Problems and Perspectives. Elsevier/North Holland,
optimal schedules arise when the induction rate is New York, 253–272.
larger than the memory decline rate of the first store. Burtt, H. E., E. M. Dobell. 1925. The curve of forgetting for adver-
tising material. J. Appl. Psych. 9 5–21.
This situation never arose in our fits. Chessa, A. G., J. M. J. Murre. 2004. A memory model for Internet
The contribution of this study from a scheduler’s or hits after media exposure. Physica A 333 541–552.
manager’s point of view is the effect of advertisement Daley, D. J., D. Vere-Jones. 1988. An Introduction to the Theory of
memory parameters on optimal scheduling. Decisions Point Processes. Springer-Verlag, Inc., New York.
concerning the planning of advertisement campaigns Diggle, P. J. 1983. Statistical Analysis of Spatial Point Patterns. Aca-
demic Press, London, UK.
could be supported better when knowledge is avail-
Glenberg, A. M. 1976. Monotonic and nonmonotonic lag effects in
able about relations between memory parameters and paired-associate and recognition memory paradigms. J. Verbal
advertisement characteristics. Although finding these Learn. Verbal Behav. 15 1–16.
relations is left as a challenge for future research, we Glenberg, A. M., T. S. Lehmann. 1980. Spacing repetitions over
could make several hypotheses here. For instance, we 1 week. Memory Cognition 8 528–538.
Goldman-Rakic, P. S. 1992. Working memory and the mind. Sci.
believe that a small saturation of learning parame- Amer. 267 111–117.
ter rmax is related to a more or less worn-out adver- Goldman-Rakic, P. S. 1995. Cellular basis of working memory.
tising concept, that encoding !1 and learning rate v Neuron 14 477–485.
are highly influenced by personal involvement and Iyer, G., D. Soberman, J. M. Villas-Boas. 2005. The targeting of
attention, and that the recall threshold b is influ- advertising. Marketing Sci. 24 461–476.
enced by novelty and the coherence of an advertising Janssen, S. M. J., A. G. Chessa, J. M. J. Murre. 2005. The reminis-
cence bump in autobiographical memory: Effects of age, gen-
message (i.e., small for well-known or coherent mes- der, education, and culture. Memory 13 658–668.
sages). Answers to these open questions (e.g., through McGaugh, J. L. 2000. Memory—A century of consolidation. Science
psychological experiments) could lead to a better and 287 248–251.
more complete understanding of the relation Meeter, M., J. M. J. Murre. 2004. Consolidation of long-term mem-
ory: Evidence and alternatives. Psych. Bull. 130 843–857.
ad properties → memory parameters → scheduling" Meeter, M., J. M. J. Murre. 2005. TraceLink: A model of consolida-
tion and amnesia. Cognitive Neuropsych. 22 559–587.
Merzereau, P., L. Battais. 2000. Continuity or burst? How do TV
In future research, we will extend our model with strategies compare? Admap (October) 30–35.
a decision-analytic component for consumer brand Murdock, B. B., Jr. 1961. The retention of individual items. J. Exper-
choice and buying behavior (e.g., see Akçura et al. iment. Psych. 62 618–625.
Chessa and Murre: Neurocognitive Model of Advertisement Content and Brand Name Recall
Marketing Science 26(1), pp. 130–141, © 2007 INFORMS 141
Murdock, B. B., Jr. 1974. Human Memory: Theory and Data. Lawrence cal Report 116, Psychology Series, Institute for Mathematical
Erlbaum Associates, Inc., Potomac, MD. Studies in the Social Sciences, Stanford University, Stanford,
Murre, J. M. J. 1996. TraceLink: A model of amnesia and consolida- CA.
tion of memory. Hippocampus 6 675–684. Simon, J. L. 1979. What do Zielske’s real data really show about
Murre, J. M. J., K. S. Graham, J. R. Hodges. 2001. Semantic demen- pulsing? J. Marketing Res. 16 415–420.
tia: Relevance to connectionist models of long-term memory. SPOT. 1998. The effectiveness of TV-advertising. Internal report.
Brain 124 647–675. (Available at www.spot.nl/onderzoek/onderzoek/downloads/
Murre, J. M. J., M. Meeter, A. G. Chessa. Modeling amnesia: Effectiviteit.pdf, in Dutch: De effectiviteit van tv-reclame), The
Connectionist and mathematical approaches. M. J. Wenger, Foundation for Promotion and Optimization of Television Ad-
C. Schuster, eds. Statistical and Process Models for Neuroscience vertising (SPOT), Amsterdam, The Netherlands.
and Aging. Erlbaum, Mahwah, NJ. Forthcoming. Stoyan, D., W. S. Kendall, J. Mecke. 1987. Stochastic Geometry and
Naik, P. A., M. K. Mantrala, A. G. Sawyer. 1998. Planning media Its Applications. John Wiley Sons/Akademie Verlag, Berlin,
schedules in the presence of dynamic advertising quality. Mar- Germany.
keting Sci. 17 214–235. Talamini, L. M., M. Meeter, J. M. J. Murre, B. Elvevåg, T. E. Gold-
Peterson, L. R., M. J. Peterson. 1959. Short-term retention of indi- berg. 2005. Integration of parallel input streams in parahip-
vidual verbal items. J. Experiment. Psych. 58 193–198. pocampal model circuits. Arch. General Psychiatry 62 485–493.
Rubin, D. C., A. E. Wenzel. 1996. One hundred years of forgetting: Wickens, T. D. 1982. Models for Behavior: Stochastic Processes in Psy-
A quantitative description of retention. Psych. Rev. 103 734–760. chology. Freeman, San Francisco, CA.
Rumelhart, D. E. 1967. The effects of interpresentation intervals Zielske, H. A. 1959. The remembering and forgetting of advertising.
on performance in a continuous paired-associate task. Techni- J. Marketing 23 239–243.
The Journal of Neuroscience, May 28, 2008 • 28(22):5861–5866 • 5861
Behavioral/Systems/Cognitive
In outcome-specific transfer, pavlovian cues that are predictive of specific outcomes bias action choice toward actions associated with
those outcomes. This transfer occurs despite no explicit training of the instrumental actions in the presence of pavlovian cues. The neural
substrates of this effect in humans are unknown. To address this, we scanned 23 human subjects with functional magnetic resonance
imaging while they made choices between different liquid food rewards in the presence of pavlovian cues previously associated with one
of these outcomes. We found behavioral evidence of outcome-specific transfer effects in our subjects, as well as differential blood
oxygenation level-dependent activity in a region of ventrolateral putamen when subjects chose, respectively, actions consistent and
inconsistent with the pavlovian-predicted outcome. Our results suggest that choosing an action incompatible with a pavlovian-predicted
outcome might require the inhibition of feasible but nonselected action– outcome associations. The results of this study are relevant for
understanding how marketing actions can affect consumer choice behavior as well as for how environmental cues can influence drug-
seeking behavior in addiction.
Key words: instrumental conditioning; decision; fMRI; human; learning; pavlovian conditioning; reward
with an affectively neutral (control) outcome. On the basis of the Table 1. Trial composition for training and transfer sessions
animal studies, we focused our analysis on the striatum, particu- Cues
Number of
larly its ventral aspect, including the nucleus accumbens and ad- Phase presentations Pavlovian Instrumental Outcome
jacent ventral putamen. We also tested for outcome-specific
transfer effects in the amygdala. 1 10 S1 O1
S2 O2
S3 O3
Materials and Methods S4 ON
Subjects 2 6 R1 R2 O1 O2
Twenty-three healthy, right-handed subjects participated in this study R1 R3 O1 ON
(six females), ranging in age from 18 to 40 (mean 24 ! 5.3 SD). One R1 R4 O1 ON
additional subject did not complete the study and was not included in the R2 R3 O2 ON
analysis. All subjects gave informed consent and the study was approved R2 R4 O2 ON
by the Caltech Institutional Review Board. R3 R4 ON ON
3 10 S1 O1
Stimuli S2 O2
Visual stimuli were presented via a projector positioned at the back of the S3 O3
room. Subjects viewed a reflection of the projected image (800 " 600 S4 ON
pixels) in a mirror attached to the scanner head coil. The food rewards 15 R1 R2 O1 O2
were delivered by means of four separate electronic syringe pumps (one R1 R3 O1 ON
for each liquid) positioned in the scanner control room. For each re- R1 R4 O1 ON
warded trial, these pumps pushed 0.6 ml of liquid to the subject’s mouth R2 R3 O2 ON
via #10 m long polyethylene plastic tubes, the other end of which were R2 R4 O2 ON
held between the subject’s lips like a straw while they lay supine in the R3 R4 ON ON
scanner. Stimulus presentation and response recording were controlled 4 25 S1 R1 R2
with the Cogent 2000 Matlab (Mathworks) toolbox. S2 R1 R2
S3 R1 R2
Behavioral procedures S4 R1 R2
During both the pavlovian and instrumental training subjects were ex- S4 R3 R4
plicitly asked to learn the cue– outcome and action– outcome relation- S1–S4, Visual cues; R1–R4, four button response actions; O1–O3, liquid rewards; ON, affectively neutral, tasteless
ships. All four training and test sessions described below were performed control solution.
in the scanner.
Pavlovian training. Pavlovian training consisted of the presentation of 1c), and as in instrumental training, subjects were asked to choose be-
associations between simple geometrical visual stimuli (Fig. 1a) and one tween two available options. This phase was conducted in extinction,
of four liquid outcomes, three of which were rewarding [chocolate milk meaning that no outcomes were delivered. The reason for performing
(Hershey’s, distributed by Dean National Brand Group), cola (Coca- this phase in extinction was to allow assessment of the influence of the
Cola), and orange juice (Trader Joe’s)] and an affectively neutral tasteless pavlovian cues on instrumental responding without the confounding
control solution, which consisted of the main ionic components of hu- effects of the outcomes themselves. Testing for outcome-specific effects
man saliva (25 mM KCl and 2.5 mM NaHCO3) (Fig. 1a). Cues were in extinction is standard in animal learning studies of this phenomenon
presented at the center of the screen for 1.75 s, and then 3 s after cue offset (Rescorla, 1994; Blundell et al., 2001; Corbit et al., 2001).
rewards were delivered with a probability of 50%. The intertrial interval There were five different types of trials. Two of the trial types were
varied uniformly between 1 and 5 s. designed to test for outcome-specific transfer effects. On these trials,
Instrumental training. During instrumental training trials, subjects subjects chose between actions associated with two particular reward
were asked to choose between two button-press actions. Four gray outcomes, O1 and O2 (for example, orange juice and chocolate milk),
squares at the bottom of the screen corresponded to the four buttons on whereas the concurrently presented pavlovian cue was associated with
a response box (Current Designs) that the subjects held in their right one of these specific outcomes. One of the specific trial types involved the
hand. Specific actions were made available for selection when the corre- pavlovian cue paired with outcome O1, and the other specific trial type
sponding squares changed color from gray to brown, two at a time. As in involved the pavlovian cue paired with O2. Evidence for an outcome-
the pavlovian trials, the response cues appeared for 1.75 s. Subjects were specific transfer effect would be seen if the presence of the pavlovian cue
asked to make a choice during this time. The choice was followed by a 3 s biased choice toward the action associated with the same outcome as the
delay before the outcome associated with the chosen action was delivered pavlovian cue. In the subsequent analysis, we pooled over both of the
on 50% of trials (Fig. 1b). The intertrial interval varied uniformly be- specific trial types, but differentiated between trials in which subjects
tween 1 and 5 s. Responses on each button earned distinct outcomes: two made choices compatible with the pavlovian outcome from those trials in
of the buttons led to rewarding outcomes (for example, orange juice and which subjects made choices that were not compatible.
chocolate milk), and two led to the neutral outcome. Therefore, during In a third “pavlovian reward control” trial type, subjects were again
pavlovian training, subjects experienced four different outcomes, presented with the choice between two reward outcomes (O1 and O2),
whereas in the instrumental trials they experienced only three. but instead the pavlovian cue was previously associated with a different
Training schedule. The first training session consisted entirely of pav- outcome (for example, cola), that was not compatible with either re-
lovian trials, 10 of each type for a total of 40 trials and a duration of #6 sponse option.
min (Table 1). The second training session consisted entirely of instru- In the fourth “pavlovian neutral control” trial type, subjects were again
mental trials, six of each type for a total of 36 trials and a duration of #5 presented with the choice between two reward outcomes (O1 and O2),
min. In the first two sessions, pavlovian and instrumental trials were but the pavlovian cue presented this time was that associated with the
presented separately to enhance learning of the respective associations. In affectively neutral outcome.
the third session, pavlovian and instrumental trials were randomly inter- In the final “neutral choice control” trial type, subjects made choices
mixed, 60 (15 " 4) pavlovian trials and 60 (10 " 6) instrumental trials, between actions associated with the affectively neutral outcome, in the
for a duration of #18 min. Before training and after each session, subjects presence of a pavlovian cue also associated with a neutral outcome. This
rated the pleasantness of the stimuli as described below. last trial type was intended to be a baseline condition for choosing be-
Outcome-specific transfer. After the three training sessions subjects per- tween two options in the presence of a visual cue but in the absence of
formed a series of transfer trials. During transfer trials one of the pavlov- predicted rewards. Each type of trial was presented 25 times, for a total of
ian cues was presented simultaneously with the instrumental cues (Fig. 125 trials and a duration of #20 min.
Bray et al. • Pavlovian Cues Influence Decision Making J. Neurosci., May 28, 2008 • 28(22):5861–5866 • 5863
Figure 2. Behavior during training and test sessions. a, Mean pleasantness ratings for visual
Figure 3. Imaging results from the pavlovian-instrumental transfer (PIT) phase. a, fMRI
cue stimuli after the training sessions, plotted by outcome pairing. The cues paired with the
results from the contrast comparing the outcome-specific transfer trials in which the action
neutral outcome were rated as significantly less pleasant than the cues paired with reward
compatible with the pavlovian cue is selected to those in which the incompatible action is
outcomes (paired t test, t(22) ! #3.0840; p " 0.01). b, Pupil diameter in response to visual
selected (red, p " 0.01; yellow, p " 0.001). At a threshold of p " 0.001, uncorrected, we find
cues. The peak amplitude is significantly smaller for the cues paired with reward outcomes for
significant activation in the ventrolateral putamen (t(21) ! 3.79; p " 0.001, uncorrected; x !
the 16 subjects who showed reliable amplitude changes after cue presentation (paired t test,
27, y ! #3, z ! #3) and bilateral pallidum (t(21) ! 3.81, p " 0.001, uncorrected, x ! 24,
t(15) ! 2.4173; p " 0.05). c, Choice behavior during the second session of instrumental trials,
y ! #18, z ! 0; and t(21) ! 3.82, p " 0.001, uncorrected, x ! #27, y ! #15, z ! #3). b,
above cue invariant responding (50%). Plotted are responses during trials in which subjects
Parameter estimates from the peak putamen voxel for each subject for each of the five experi-
chose between a reward outcome and the neutral outcome. Subjects were significantly more
mental conditions during the transfer phase [specific compatible (comp), specific incompatible
likely to choose the action leading to the reward outcome (one-sided paired t test, t(22) !
(incomp), pavlovian reward (pav rew) control, pavlovian neutral (pav N) control, and neutral (N)
1.8399; p " 0.05). d, Choice data binned into five 10-trial bins. There is no significant linear
choice control]. Parameter estimates in the specific compatible condition do not differ signifi-
trend across the session (linear regression of the percentage of compatible choice allocation
cantly from any condition other than specific incompatible (paired t tests, p $ 0.05). Error bars
onto bin number, p ! 0.239). Error bars indicate SEM.
indicate SEM.
(Fig. 2 d), indicating that the biasing effect of the pavlovian cues in the outcome-specific trials did not differ from the control
on choice persisted for the duration of the extinction test and did conditions (paired t tests, p $ 0.05).
not attenuate.
Discussion
fMRI results Our results provide insights into the neural mechanisms by
To gain insight into the mechanisms underlying outcome- which pavlovian cues can modulate choice between different in-
specific transfer in humans, we performed two analyses. First, we strumental courses of action, in humans. In outcome-specific
compared brain activity during trials assessing outcome-specific transfer, subjects are more likely to choose an action that is asso-
transfer when subjects chose the option compatible with the pav- ciated with a particular outcome in the presence of a pavlovian
lovian cue to trials when they chose the incompatible option (one cue that was previously associated with the presence of that out-
subject who never chose the incompatible cue was excluded from come. We found neural correlates of outcome-specific transfer in
this analysis) (Fig. 3a). We found significant activation in the a very circumscribed region of extended ventral striatum in the
right ventrolateral putamen (t(21) ! 3.79, p " 0.001 uncorrected; ventral caudolateral putamen. This region and an adjacent region
x ! 27, y ! #3, z ! #3) extending posteriorly toward the palli- of ventral pallidum were the only areas to meet our statistical
dum (t(21) ! 3.81, p " 0.001 uncorrected; x ! 24, y ! #18, z ! criterion for significance.
0). The left pallidum also showed a peak at this threshold (t(21) ! These findings add to an accumulating body of evidence from
3.82, p " 0.001 uncorrected; x ! #27, y ! #15, z ! #3). These human fMRI studies of a role for an extended region of ventral
were the only regions to meet our significance criterion in this parts of putamen alongside the nucleus accumbens in functions
contrast. related to reward-learning and prediction errors (O’Doherty et
Second, we plotted the average parameter estimates taken al., 2002, 2003; McClure et al., 2003) and now in interactions
from the general linear model estimates at the peak putamen between pavlovian and instrumental conditioning. Such findings
voxel for each subject (Fig. 3b). We found that the difference resonate with anatomical and histochemical studies in primates
between conditions was caused by a significant decrease in signal that indicate that ventral parts of putamen share many of the
during the outcome-specific trials where the incompatible re- cytoarchitectonic characteristics of nucleus accumbens, as well as
sponse was chosen, relative to the outcome-specific trials when sharing similar inputs (Russchen et al., 1985; Selemon and Gold-
the compatible response was chosen and to the other control manrakic, 1985; Fudge and Haber, 2002; Fudge et al., 2002).
conditions. In fact, activity in the compatible condition did not The present findings do suggest, however, that different parts
differ significantly from activity during any of the other control of the ventral striatum may contribute differentially to distinct
conditions (paired t tests, p $ 0.05) and, more generally, activity forms of pavlovian-instrumental transfer in humans. This sug-
Bray et al. • Pavlovian Cues Influence Decision Making J. Neurosci., May 28, 2008 • 28(22):5861–5866 • 5865
gestion is based on a comparison of our finding that the ventro- This finding provides insight into the computations that
lateral putamen is involved in outcome-specific transfer in hu- might be taking place in the ventral striatum during outcome-
mans with the results of a previous study implicating nucleus specific transfer effects. Outcome-specific transfer effects are
accumbens in the general excitatory effects of pavlovian cues on thought to be mediated by outcome–response (O–R) associa-
instrumental performance (Talmi et al., 2008). It is well estab- tions that are activated by the pavlovian cues (Rescorla, 1994;
lished that pavlovian cues can exert a general, nonspecific excita- Balleine and Ostlund, 2007). A natural hypothesis is that when
tory effect on the performance of instrumental actions (Estes, the action plan activated by the O–R association is feasible (be-
1943, 1948; Rescorla and Solomon, 1967; Colwill and Rescorla, cause such an action is available), it must be inhibited before
1988; Rescorla, 1994; Holland, 2004), an effect that Talmi et al. another action can be taken. Note that, under this hypothesis, the
(2008) demonstrated is mediated by activation of nucleus ac- O–R association needs to be inhibited during the outcome-
cumbens and of amygdala. In the context of the present study, specific transfer trials when the incompatible response is chosen,
these findings suggest that outcome-specific and general transfer but not when the compatible response is selected, or in any of the
may depend on quite distinct neural substrates in humans, mir- other control trials. This provides a computational explanation
roring clear double dissociations between the neural circuits for why suppression of activity in the ventrolateral putamen is
known to be involved in implementing these effects in rodents observed only in the incompatible outcome-specific transfer
(Corbit et al., 2001; Corbit and Balleine, 2005). Although the trials.
present study was not designed to assess the effects of general Specific transfer effects from pavlovian cues have been argued
transfer, in future it will be important to compare and contrast to play a role in addictive behaviors (Ludwig et al., 1974). For
outcome-specific and general transfer effects within the same example, Hogarth et al. (2007) demonstrated specific transfer of a
fMRI study to provide a more direct test of the hypothesis that, as tobacco-seeking response in the presence of a tobacco predicting
in rodents, outcome-specific and general transfer in humans de- cue, relative to a money predicting cue. Here, we demonstrate
pend on distinct components of the ventral striatum. similar behavioral results, using nonaddictive outcomes, indicat-
Note that although we found a remarkably good correspon- ing that the observed transfer effects reflect a general property of
dence between our findings and those from the rodent lesion reward-associated cues that are not specifically related to addic-
studies at the level of the ventral striatum, other regions in addi- tive stimuli. Nonetheless, there are clear parallels between our
tion to the ventral striatum have been implicated in specific experimental design and the potential influence of environmen-
pavlovian-instrumental transfer in rodents, including the baso- tal cues on drug-seeking behavior. Our fMRI results suggest the
lateral amygdala (Corbit and Balleine, 2005) and dorsolateral hypothesis that suppression of an outcome–response association
striatum (Corbit and Janak, 2007). We did not find any evidence might contribute toward biasing behavior away from cue-
for a differential contribution of these regions in the present compatible responding. This raises the possibility of a future
study. One possibility is that these areas do play a role in specific therapeutic intervention in addiction, in which ventrolateral pu-
transfer effects in humans, but this does not result in a global tamen circuitry could potentially be targeted (for instance via a
increase in activity between conditions and, thus, does not be- neurofeedback procedure) (deCharms et al., 2005; Bray et al.,
come manifest with BOLD fMRI. 2007) to suppress effects of environmental drug cues on drug-
The present results go beyond merely pointing to homologies seeking behavior.
between outcome-specific transfer effects in rodents and hu- In this study, we demonstrated an outcome-specific
mans. Previous animal studies on this topic have all involved pavlovian-instrumental transfer effect in humans, which serves
lesion manipulations, which, although important for identifying to bias action choice toward actions associated with an outcome
whether a given region is necessary for implementing specific consistent with a concurrently presented cue. BOLD fMRI mea-
transfer effects, cannot provide insight into the neural computa- sured while subjects performed this task demonstrated a signal
tions underlying such an effect. Here, we measured dynamic decrease in ventrolateral putamen when subjects’ chose the ac-
changes in BOLD responses as subjects made choices that were tion incompatible with the cue. This finding points to a compu-
either consistent or inconsistent with the specific transfer effect. tational role for this region in suppressing outcome–response
Responses consistent with the specific transfer effect occurred associations, necessary to perform an action incompatible with
when subjects chose the outcome compatible with the pavlovian the pavlovian cue only when a compatible action is feasible. This
cue, and inconsistent responses occurred when subjects chose the work adds to our understanding of the neural mechanisms of
incompatible option. Although subjects showed a significant bias stimulus– outcome guided decision making in both animals and
toward the compatible action overall, sometimes they chose the humans, which is fundamental for understanding maladaptive
incompatible action; this allowed us to compare activity when choice behaviors such as addiction.
transfer guided behavior with activity under identical stimulus
conditions when subjects chose independently of the cue. Activ- References
ity in ventrolateral putamen was not significantly elevated on Balleine B (1994) Asymmetrical interactions between thirst and hunger in
pavlovian-instrumental transfer. Quart J Exp Psychol B 47:211–231.
trials when an outcome-specific cue was presented compared
Balleine BW, Ostlund SB (2007) Still at the choice-point: action selection
with control trials where cues for other, unavailable outcomes and initiation in instrumental conditioning. Ann NY Acad Sci
were presented, suggesting that outcome-specific transfer effects 1104:147–171.
are not mediated by an overall increase in activity in this area. Bitsios P, Szabadi E, Bradshaw CM (2004) The fear-inhibited light reflex:
Furthermore, even on outcome-specific trials where subjects importance of the anticipation of an aversive event. Int J Psychophysiol
chose the action compatible with the pavlovian cue, there was no 52:87–95.
increase in activity compared with non-outcome-specific control Blundell P, Hall G, Killcross S (2001) Lesions of the basolateral amygdala
disrupt selective aspects of reinforcer representation in rats. J Neurosci
trials. Instead, we found a significant decrease in signal on those 21:9018 –9026.
outcome-specific trials where subjects chose the action incom- Bray S, Shimojo S, O’Doherty JP (2007) Direct instrumental conditioning of
patible with the outcome, compared with compatible choice neural activity using functional magnetic resonance imaging-derived re-
outcome-specific trials. ward feedback. J Neurosci 27:7498 –7507.
5866 • J. Neurosci., May 28, 2008 • 28(22):5861–5866 Bray et al. • Pavlovian Cues Influence Decision Making
Colwill RM, Rescorla RA (1988) Associations between the discriminative Ludwig AM, Wikler A, Stark LH (1974) First drink: psychobiological as-
stimulus and the reinforcer in instrumental learning. J Exp Psychol Anim pects of craving. Arch Gen Psychiatry 30:539 –547.
Behav Process 14:155–164. McClure SM, Berns GS, Montague PR (2003) Temporal prediction errors in
Corbit LH, Balleine BW (2005) Double dissociation of basolateral and cen- a passive learning task activate human striatum. Neuron 38:339 –346.
tral amygdala lesions on the general and outcome-specific forms of O’Doherty JP, Deichmann R, Critchley HD, Dolan RJ (2002) Neural re-
pavlovian-instrumental transfer. J Neurosci 25:962–970. sponses during anticipation of a primary taste reward. Neuron
Corbit LH, Janak PH (2007) Inactivation of the lateral but not medial dorsal 33:815– 826.
striatum eliminates the excitatory impact of pavlovian stimuli on instru- O’Doherty JP, Dayan P, Friston K, Critchley H, Dolan RJ (2003) Temporal
mental responding. J Neurosci 27:13977–13981. difference models and reward-related learning in the human brain. Neu-
Corbit LH, Muir JL, Balleine BW (2001) The role of the nucleus accumbens ron 38:329 –337.
in instrumental conditioning: evidence of a functional dissociation be- O’Doherty JP, Buchanan TW, Seymour B, Dolan RJ (2006) Predictive neu-
tween accumbens core and shell. J Neurosci 21:3251–3260. ral coding of reward preference involves dissociable responses in human
deCharms RC, Maeda F, Glover GH, Ludlow D, Pauly JM, Soneji D, Gabrieli ventral midbrain and ventral striatum. Neuron 49:157–166.
JD, Mackey SC (2005) Control over brain activation and pain learned by Ostlund SB, Balleine BW (2007) Selective reinstatement of instrumental
performance depends on the discriminative stimulus properties of the
using real-time functional MRI. Proc Natl Acad Sci USA
mediating outcome. Learn Behav 35:43–52.
102:18626 –18631.
Rescorla RA (1994) Transfer of instrumental control mediated by a deval-
Deichmann R, Gottfried JA, Hutton C, Turner R (2003) Optimized EPI for
ued outcome. Anim Learn Behav 22:27–33.
fMRI studies of the orbitofrontal cortex. NeuroImage 19:430 – 441.
Rescorla RA, Solomon RL (1967) Two-process learning theory: relation-
Estes WK (1943) Discriminative conditioning. I. A discriminative property
ships between pavlovian conditioning and instrumental learning. Psychol
of conditioned anticipation. J Exp Psychol 32:150 –155. Rev 74:151–182.
Estes WK (1948) Discriminative conditioning. 2. Effects of a pavlovian con- Russchen FT, Bakst I, Amaral DG, Price JL (1985) The amygdalostriatal
ditioned stimulus upon a subsequently established operant response. J projections in the monkey. An anterograde tracing study. Brain Res
Exp Psychol 38:173–177. 329:241–257.
Fudge JL, Haber SN (2002) Defining the caudal ventral striatum in pri- Selemon LD, Goldmanrakic PS (1985) Longitudinal topography and inter-
mates: cellular and histochemical features. J Neurosci 22:10078 –10082. digitation of corticostriatal projections in the rhesus monkey. J Neurosci
Fudge JL, Kunishio K, Walsh P, Richard C, Haber SN (2002) Amygdaloid 5:776 –794.
projections to ventromedial striatal subterritories in the primate. Neuro- Seymour B, Daw N, Dayan P, Singer T, Dolan R (2007) Differential encod-
science 110:257–275. ing of losses and gains in the human striatum. J Neurosci 27:4826 – 4831.
Hogarth L, Dickinson A, Wright A, Kouvaraki M, Duka T (2007) The role of Smeets PM, Barnes-Holmes D (2003) Children’s emergent preferences for
drug expectancy in the control of human drug seeking. J Exp Psychol soft drinks: stimulus-equivalence and transfer. J Econ Psychol
Anim Behav Process 33:484 – 496. 24:603– 618.
Holland PC (2004) Relations between pavlovian-instrumental transfer and Talmi D, Seymour B, Dayan P, Dolan RJ (2008) Human pavlovian-
reinforcer devaluation. J Exp Psychol Anim Behav Process 30:104 –117. instrumental transfer. J Neurosci 28:360 –368.
The Journal of Psychology, 2007, 141(2), 117–125
Copyright © 2007 Heldref Publications
ABSTRACT. Previous researchers (e.g., J. A. Bargh, 1992, 2002) demonstrated the impor-
tance of nonconscious processes on consumer choice behavior. Using an advertisement, the
authors determined the effect of two nonconscious processes—the mere exposure effect,
which increases object preference by increasing consumer exposure to an object, and the
endowment effect, which increases object valuation by providing consumer possession of
an object—on consumer behavior. Although the mere exposure effect and endowment
effect did not produce an interaction, they produced independent effects. The endowment
effect increased object valuation but not object preference. The mere exposure effect
increased object preference but not object valuation. Thus, at the unconscious level, an
increase in object preference does not lead to an increase in object valuation, nor does an
increase in object valuation lead to an increase in object preference. The authors discuss the
importance of developing measures of unconscious process in advertising effectiveness.
Keywords: consumer decision making, endowment effect, mere exposure, prime
This project was funded by a grant from the College of Business Administration, Califor-
nia State University, Sacramento.
Address correspondence to Gail Tom, College of Business Administration, California
State University, Sacramento, CA 95819; tomgk@csus.edu (e-mail).
117
118 The Journal of Psychology
1992). Consumers choose products with characteristics that match their primed-
activated motives. In one study, thirst-induced consumers subliminally exposed
(i.e., primed) to happy facial expressions subsequently consumed more and were
willing to pay more for a drink as compared with thirst-induced consumers who
were subliminally exposed to angry faces (Winkielman, Berridge, & Wilbarger,
2001). Strahan, Spencer, and Zanna (2001) found that only consumers who were
made to be thirsty and subliminally primed to be thirsty, but not nonthirsty, sub-
liminally primed consumers, opted to drink more of a purported thirst-quenching
drink than of a purported energy drink. Participants were not aware of the
primed-activated motive and its subsequent effect on choice behaviors in either
study. Results of these and other studies (Bargh, Chen, & Burrows, 1996; Bargh,
Gollwitzer, Lee-Chai, Barndollar, & Troetschel, 2001; Chartrand & Bargh, 2002;
Fitzsimons & Shiv, 2001) demonstrated that consumer downstream behaviors are
Downloaded by [Det Kgl Bibl Natl bibl og Kbh Univ bibl] at 23:14 02 October 2011
Method
Materials
We investigated the effect of the endowment effect and mere exposure effect
on object valuation and object preference with a 2 (Ownership: endowment vs.
nonendowment) × 3 (Exposure: no exposure, normal exposure, subliminal expo-
sure) factorial design. In the endowed condition, we gave participants an object
to examine and keep. In the unendowed condition, we gave participants an object
to examine and then return. We created three versions of the same video for the
120 The Journal of Psychology
refrigerator). It can be used, for example, as a memo holder, a novelty item that
sticks to a wall, or an item to place on a bookcase.
We created a fast-paced, 2-min musical promotional video featuring select-
ed alumni and the current year’s recipients of outstanding student awards from
the College of Business. The video opened with the words From here followed
by historical and contemporary scenes of the College of Business: students walk-
ing to the building, instructors presenting to their classes, and students partici-
pating in a sports rally. The word Anywhere then followed. Photos of alumni
were displayed successively. Words framed the portrait: Across the top was the
statement College of Business Administration; at the bottom was the alumnus
name, position, and company, the left side provided the statement class of; and
the right side provided the alumnus graduation class. Then the word Everywhere
was displayed, followed by more portraits of alumni. Last, the words No Limits
where displayed, and the outstanding student award recipients with captions of
their names were successively listed at the end of the promotion. The video
ended with a recap of the words From here…anywhere…everywhere…NO LIM-
ITS College of Business, [name of the university].
The three versions of the promotional spot were identical and differed only
in the inclusion or exclusion and length of exposure time of the promotional item.
In the subliminal piece, the promotional item was shown three times, once each
as the words From here, Anywhere, Everywhere were shown. The promotional
item received three exposures of 3 frames (1/10 s) each. In the normal exposure
video, the promotional item was shown three times at the same locations as in the
subliminal piece but was on the screen each time for 8 frames (approximately 1/4
Downloaded by [Det Kgl Bibl Natl bibl og Kbh Univ bibl] at 23:14 02 October 2011
s). It was also shown at the end of the video, when it flew up from the bottom of
the screen to the top of the screen as the words From here, Anywhere, Every-
where, College of Business Administration, [name of university] were shown.
This Dartman! was on the screen for 9 s (72 frames). The no exposure condition
did not show the promotional item.
Participants
Procedure
Results
Discussion
among the groups’ rating of the video as neutral (3 on the 5-point Likert-type
scale) and suggest that differences in object valuation or object preference ratings
were not caused by differential advertisement preferences.
Our results suggest that the mere exposure effect and the endowment effect
do not interact to affect object preference. Subliminal exposure produced a
stronger mere exposure effect than did supraliminal exposure and no exposure,
and supraliminal exposure produced a stronger mere exposure effect than did no
exposure. Thus, whereas both subliminal and supraliminal exposures produced a
mere exposure effect, the mere exposure effect was stronger in the subliminal
condition. One explanation for this finding is that subliminal exposure reflects
the stronger effect of the occurrence of a nonconscious process (Bornstein &
D’Agostino, 1992). Priming is effective when consumers are not aware that they
are being primed (Strahan et al., 2001; Winkielman et al., 2001). Participants’
nonconscious, subliminally primed motive was then matched with their subse-
quent receipt of the promotional item. The object preference likely reflects the
match between the prime and the object (Ratneshwar, Mick, & Huffman, 2000;
Read & van Leeuwen, 1998). Another explanation for this finding is that the sub-
liminal exposure of the promotional item provided a mere exposure effect while
retaining the novelty characteristics of the promotional item (i.e., although con-
sumers were exposed to the Dartman!, their unawareness of the exposure main-
tained its novelty). In the case of supraliminal exposure, the mere exposure effect
was attenuated by the awareness of the novelty characteristics of the promotion-
al item. Thus, the more favorable rating for the Dartman! in the subliminal con-
dition than in the supraliminal condition may have been a result of maintenance
of its novelty. The ratings of such an item likely reflect its inherent novelty.
We did not find an interaction effect of the mere exposure effect and the
endowment effect on object valuation. However, we found a significant main
effect of the endowed condition on object valuation. These results suggest that
either object valuation is not affected by the mere exposure effect or, alternative-
ly, these results may be the result of an insufficient manipulation of the mere
124 The Journal of Psychology
exposure effect. It is possible that the mere exposure effect can affect object val-
uation but that it was not sufficiently strong in this experiment to do so. Future
researchers should explore this possibility.
Previous researchers in the mere exposure effect used slides, but we used
a video. Video presentation is a realistic representation of advertisements
broadcasted on television and the Internet, but it reduces ability to control par-
ticipants’ attention. When slides are used, the experimenter has greater control
of participants’ attention to the exposed stimuli. In a video, the participant has
greater control of attention. Future researchers using videos should weigh the
advantages of its realism with the limitations of reduced experimental control
of participant attention. Future researchers should compare the effects of
video and slide presentations on audience preference, object valuation, atten-
tion, and retention. The frequency and duration of exposure we used in this
Downloaded by [Det Kgl Bibl Natl bibl og Kbh Univ bibl] at 23:14 02 October 2011
study were within the range used in previous work (Bornstein, 1989). Howev-
er, future researchers should investigate the effect of different frequencies and
durations on consumer behavior. The cover story for our study required the
use of an inexpensive novelty item, a Dartman!, as a promotional item. Future
researchers should use different objects to determine the generalizability of
our findings to items of different monetary values and varying degrees of
familiarity.
We demonstrated the impact of the unconscious processes of the mere expo-
sure effect and endowment effect on consumer decision-making behavior. Along
with studies within this area of research, our results suggest that the effectiveness
of advertisements may be understated if they rely on measures of recall or recog-
nition and point to the need to include instruments that measure the influence of
unconscious processes.
REFERENCES
Bargh, J. A. (1992). Why subliminality does not matter to social psychology: Awareness
of the stimulus versus awareness of its influence. In R. F. Bornstein & T. E. Pittman
(Eds.), Perception without awareness (pp. 236–255). New York: Guilford.
Bargh, J. A. (2002). Losing conscious: Automatic influences on consumer judgment,
behavior, and motivation. Journal of Consumer Research 29, 280–285.
Bargh, J. A., Chen, M., & Burrows, L. (1996). Automaticity of social behavior: Direct
effects of train construct and stereotype priming on action. Journal of Personality and
Social Psychology, 71, 230–244.
Bargh, J. A., Gollwitzer, P. M., Lee-Chai, A., Barndollar, K., & Troetschel, R. (2001). The
automated will: Nonconscious activation and pursuit of behavioral goals. Journal of
Personality and Social Psychology, 81, 1014–1027.
Bornstein, R. F. (1989). Exposure and affect: Overview and meta-analysis of research,
1968–1987. Psychological Bulletin, 106, 265–289.
Bornstein, R. F., & D’Agostino, P. R. (1992). Stimulus recognition and the mere exposure
effect. Journal of Personality and Social Psychology, 63, 545–552.
Chartrand, T. L., & Bargh, J. A. (2002). Nonconscious motivations: Their activation, oper-
ation, and consequences. In A. Tesser, D. Stapel, & J. Woods (Eds.), Self-motivation:
Tom, Nelson, Srzentic, & King 125
effect and the coase theorem. Journal of Political Economy, 98, 1325–1348.
Kahneman, D., & Tverksy, A. (1984). Choices, values and frames. American Psycholo-
gist, 39, 341–350.
Knetsch, J. L. (1989). The endowment effect and evidence of nonreversible indifference
curves. The American Economic Review, 79, 1277–1284.
Knetsch, J. L. (1990). Derived indifference curves. Unpublished manuscript, Department
of Economics, Simon Fraser University, Burnaby, BC, Canada.
Ledoux, J. (1996). The emotional brain. New York: Simon and Schuster.
Ratneshwar, S. D., Mick, G., & Huffman, C. (Eds.). (2000). The why of consumption. New
York: Routledge.
Read, D., & Van Leeuwen, B. (1998). Predicting hunger: The effects of appetite and delay
on choice. Organizational Behavior and Human Decision Process, 76, 189–205.
Simonson, I., Carmon, Z., Dhar, R., Drolet, A., & Nowlis, S. M. (2001). Consumer
research: In search of identity. Annual Review of Psychology, 52, 249–275.
Strahan, E. R., Spencer, S. J., & Zanna, M. P. (2001). Subliminal priming and persuasion:
Striking while the iron is hot. Unpublished manuscript, Department of Psychology, Uni-
versity of Waterloo, Ontario, Canada.
Thaler, R. H. (1980). Toward a positive theory of consumer choice. Journal of Economic
Behavior and Organization, 1, 39–60.
Tom, G. (2004). The endowment-institutional affinity effect. The Journal of Psychology,
136, 160–170.
Wilson, W. R. (1979). Feeling more than we can know: Exposure effects without learn-
ing. Journal of Personality and Social Psychology 37, 811–821.
Winkielman, P., Berridge, K. C., & Wilbarger, J. (2001). Doing without feeling: Uncon-
scious affect controls human consumption. Unpublished manuscript, Department of
Psychology, University of Denver, CO.
Zajonc, R. B. (1968). Attitudinal effects of mere exposure. Journal of Personality and
Social Psychology, 9, 1–28.
Zajonc, R. B. (2001). Mere exposure: A gateway to the subliminal. Current Directions in
Psychological Science 10, 224–228.
Zola-Morgan, S., Squire, L. R., Alvarez-Royo, P., & Clower, R. P. (1991). Independence
of memory functions and emotional behavior. Hippocampus, 1, 207–220.