Content uploaded by Joel B. Cohen
Author content
All content in this area was uploaded by Joel B. Cohen
Content may be subject to copyright.
297
11
The Nature and Role of Affect
in Consumer Behavior
J B. C
University of Florida
M T P
Columbia University
E B. A
University of California–Berkeley
In the intervening years since publication of the chapter “A ect and Consumer Behavior” (Cohen
& Areni, 1991) in the Handbook of Consumer Behavior (Kassarjian & Robertson, 1991), research in
consumer behavior dealing with a ect has exploded, making it one of the eld’s central research
topics. Within psychology more generally, Schimmack and Crites (2005) located 923 references
to a ect between 1960 and 1980 and 4,170 between 1980 and 2000. Since research on a ect has
become more specialized, this chapter will concentrate on the various ways a ect in uences judg-
ment and choice rather than on broader and historical perspectives. ese will include the role of
a ect in information retrieval, di erential processing of a ectively colored information (including
the role of a ect in strengthening mental associations and memory consolidation), how and when
a ect provides information that in uences judgments and decisions, and the motivational role of
a ect in guiding behavior and signaling the need for changes in vigilance, intensity, and direction.
We begin, however, with some essential de nitions.
THE NATURE OF AFFECT: FEELINGS, EMOTIONS, AND MOODS
What A ect Means
ere is still some carryover from the use of the term “a ect” to also refer to what is, in essence, the
evaluative aspect of attitudes. is stems from the classic tri-partite depiction of attitudes: cogni-
tive, a ective, and conative (see Eagly & Chaiken, 1993) and a failure to adequately di erentiate
between evaluative measures (e.g., favorable/unfavorable) and antecedent or subsequent processes,
which might be feeling-based. Consistent with most recent scholarly discussions, we reserve the
term “a ect” to describe an internal feeling state. One’s explicit or implicit “li king” for some object ,
person, or position is viewed as an evaluative judgment rather than an internal feeling state. As
Russell and Carroll (1999a) put it:
Haugtvedt_ER5603X_C011.indd 297Haugtvedt_ER5603X_C011.indd 297 7/10/2007 4:47:16 PM7/10/2007 4:47:16 PM
298 JOEL B. COHE N, MICHEL TUAN PHAM, AND EDUAR DO B. ANDRADE
By a ect, we have in mind genuine subjective feelings and moods (as when someone says, ‘I’m feeling
sad’), rather than thoughts about speci c objects or events (as when someone calmly says, ‘ e cru-
sades were a sad chapter in human history’). (pp. 3–4)
is chapter maintains the separation of a ect as a feeling state that is distinct from either liking
or purely descriptive cognition. So when we use the term “a ect” to describe stimuli, internal and
overt responses, it is only in relation to evoked feeling states. Imagine, in contrast, an advertise-
ment whose words or images connote a happy (i.e., successful) outcome. A ective processes cannot
merely be assumed. Alternative explanations (e.g., the advertised product seems likely to produce
favorable outcomes) for so-called “a ective” in uences on subsequent evaluations and behavior
must be ruled out before implicating a ect. ese include semantically associated changes in object
meaning or construct accessibility.
is de nition also raises both philosophical and empirical questions about whether such a feel-
ing state must be consciously experienced or whether we can be unaware that we are experiencing
a ect. Research where subliminally presented smiling or frowning faces were used to prime a ect
(outside of awareness) and bring about subsequent evaluative responses (Winkielman, Zajonc, &
Schwarz, 1997), is a case in point. A ective experience in the absence of an identi ed basis for that
experience has been a staple of psychological research since Zajonc’s (1980) early work on “mere
exposure.” In that program of research, repeated subliminal exposure to unfamiliar stimuli having
neutral valence such as Chinese ideographs has been shown to generate some degree of liking for
the stimuli, possibly as a result of a primitive reward mechanism associated with increasing famil-
iarity or a reduction in uncertainty. Another standard paradigm for investigating precognitive
a ective processes is to present (outside of awareness) a stimulus known to evoke either a negative
or positive a ective response (e.g., a sad face). Following that exposure, people are asked to indicate
how they are feeling (to rule out more conscious a ective responses including inferences) and to
rate the emotional quality of a semantically unrelated object, such as a piece of music. Using such a
procedure, for example, Strahan, Spencer, and Zanna (2002) found that a ective stimuli can in u-
ence positive/negative assessments even without producing a measurable e ect on people’s a ec-
tive experiences (i.e., reported feelings). In a particularly sophisticated study (Schimmack, 2004),
subjects received masked subliminal presentations of pleasant and unpleasant pictures, followed by
supraliminal presentations of an identical picture (the target) paired with a foil whose valence was
either the same as the target or opposite. If the initial subliminal target exposure produced a spon-
taneous a ective experience, participants should be better able—and they were—to discriminate
the target from the foil when they had a di erent valence because only one object should match
the originally experienced a ect. Di erent results have frequently been observed for pictures and
words when used as subliminal stimuli (Schimmack & Crites, 2005). Words have been found to
elicit a skin-conductance response under conditions of very short exposure (suggesting a ec-
tive experience), whereas pictures have not. However, this nding may also be due to the greater
inherent polarity of the selected words relative to pictures, since the interpretation of pictures may
require more cognitive resources than words having relatively xed a ective associations.
Most consumer research on a ect deals with moods (e.g., Barone, Miniard, & Romeo, 2000;
Cohen & Andrade, 2004; Gorn, Goldberg, & Basu, 1993; Pham, 1998), although there has been
growing interest in the study of speci c emotions (e.g., Lerner, Small, & Loewenstein, 2004; Rag-
hunathan & Pham, 1999; Raghunathan, Pham, & Corfman, 2006). Moods are usually thought of
as low intensity and di use a ective states that generally lack source identi cation.1 e individ-
ual, prompted either by physiological or hormonal/chemical activity (such as changes in levels of
serotonin and dopamine) or by external stimuli (music, weather, exposure to happy versus sad
Haugtvedt_ER5603X_C011.indd 298Haugtvedt_ER5603X_C011.indd 298 7/10/2007 4:47:34 PM7/10/2007 4:47:34 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 299
information), experiences a vague sense of feeling good or bad without necessarily knowing quite
why. Some days or a er certain experiences, we are aware of feeling good or bad, optimistic or pes-
simistic, up or down, relaxed or restless, alert or drowsy. Mood states also track our bodily energy
levels (e.g., blood glucose levels), our daily circadian rhythm, and our general wellness or illness,
thereby guiding relatively automatic self-regulatory responses as well as more conscious decisions,
as we shall discuss later on. Emotions, on the other hand, are much more di erentiated and hence
provide more attitude- and behavior-speci c information. Feeling anger, for example, will o en
lead to target and context-speci c responses rather than more general displays of unhappiness
(Bushman & Baumeister, 1999). It should be noted, however, that speci c emotions can produce
mood-like e ects (e.g., being angry or sad can a ect a pattern of behavior) o en without realizing
that one has transferred the emotional response (to an identi ed target) to unrelated behaviors.
Recent studies show that the degree of transfer will be a function of two factors: (1) the salience
of the source of the emotional state—transfer is more likely when the actual source of the a ect is
not salient; and (2) the domain similarity between the actual source of the a ective state and the
objectively unrelated behavior (Raghunathan, Pham, & Corfman, 2006).
Moods have been shown to be easily manipulated through exposure to a ectively charged stim-
uli such as music, videos, and pictures, or through the recall of emotionally involving experiences
(e.g., Cohen & Andrade, 2004). Note that the use of low intensity emotion manipulations, such as
sadness, displeasure, or happiness, to create positive or negative mood states tends to blur the line
between emotions and moods, especially when the source is made salient.
Because a ect is o en used as information about “how things are going” (Schwarz, 1990;
Schwarz & Clore, 1983), the misattribution of incidental a ect may play a powerful role in everyday
life. Even experimentally-induced proprioceptive feedback of head nodding or shaking can lead a
person to conclude that message-related thoughts are positive or negative (Brinol & Petty, 2003).
e duration of mood changes is typically assumed to be short, from a few minutes to a couple of
hours (Isbell & Wyer, 1999), although this duration probably varies with the method of instigation
(Ehrlichman & Halpern, 1988; Isen, Clark, & Schwartz, 1976).
EXPERIMENTAL MANIPULATION OF AFFECTIVE STATES
Multiple techniques have been used to manipulate individuals’ transient a ective states. In most
experiments, participants are exposed to a sequence of ostensibly unrelated studies, where the rst
study is meant to manipulate people’s feelings while the second assesses the dependent variables of
interest. In the rst study, participants might be exposed, for instance, to false positive or negative
performance feedback (Barone, Miniard, & Romeo, 2000; Swinyard, 1993), cheerful or depressing
movies (E. B. Andrade, 2005; Cohen & Andrade, 2004), pleasant or unpleasant music (Gorn, Gold-
berg, & Basu, 1993), positive (PA) or negative a ective (NA) self-referential statements such as the
Velten procedure (Velten, 1968), unexpected gi s (Barone, Miniard, & Romeo, 2000; Isen & Sim-
monds, 1978), or recalling and describing in writing an a ectively charged experience (Pham, 1998).
Due to their transient and mild, hence, short-lived nature, experimentally induced moods may dis-
sipate relatively fast (see Isen, Clark, & Schwartz, 1976). erefore, regardless of the mood induction
procedure, the dependent measure usually is collected not long a er the manipulation. Sometimes
the mood manipulation and dependent measures co-occur. Mood manipulations using background
music or physical ambience, like scent, for instance, allow for a simultaneous assessment of depen-
dent variables (Grunberg & Straub, 1992; Schwarz, Strack, Kommer, & Wagner, 1987).
ere is no single best option among all potential techniques. Di erent techniques raise di erent
issues in terms of potential confounds , control for intensity levels, reliabil ity, demand char acteristics,
Haugtvedt_ER5603X_C011.indd 299Haugtvedt_ER5603X_C011.indd 299 7/10/2007 4:47:34 PM7/10/2007 4:47:34 PM
300 JOEL B. COHEN, MICHEL T UA N PHAM, AND EDUAR DO B. ANDRADE
and motivational requirements. For instance, receiving an unexpected gi , a common manipula-
tion of positive mood (Isen, Shalker, Clark, & Karp, 1978; Kahn & Isen, 1993), can activate norms
of reciprocity independently of a ective changes (e.g., “ e experimenter was nice to me, I’ll be
nice to him/her”). Similarly, false performance feedback can in uence self-esteem or self-e cacy
along with desired changes in a ective states (Hill & Ward, 1989). Such unwanted e ects might be
confounding depending on the research question and other aspects of the procedure.
Asking participants to report an a ectively charged personal experience can avoid some of the
above-mentioned concerns. A major advantage of this method is that because each participant
recruits his or her own personal experience, there is a lesser chance of confounding with the content
of the a ect-inducing event. Content-related confounds are much more likely with manipulations
that involve exposure to a common a ect-inducing stimulus across participants, such as watching
a happy or sad movie. On the other hand, the personal experience method requires relatively high
participant motivation; otherwise, the manipulation tends to be weakened. A second drawback of
this method is that participants usually are explicitly directed to write about experiences that lead
them to feel good or bad, which may enhance the likelihood of hypothesis guessing and demand
artifacts. is concern is heightened if a salient mood manipulation check is administered before
the dependent measures are collected. In addition, there may be extra variability in emotional
states induced—hence higher experimental error—because participants may have di erent inter-
pretations of the type of experience they are supposed to report. Some respondents may interpret
“an event that made you feel bad” as one that made them feel angry, whereas others may interpret it
as one that made them feel sad. It is therefore important that the instructions be very precise when
using this manipulation.
Music does not require highly motivated participants, does not direct participants to speci c
feelings, and has been shown to produce signi cant e ects on judgment and behavioral measures
(Gorn, Goldberg, & Basu, 1993; Gorn, Pham, & Sin, 2001). However, there is signi cant variance in
the population when it comes to music tastes, which can compromise reliability. Exposure to a ec-
tively charged videos has proven to be quite successful due to the general appeal of these stimuli
(l ow mo tiv ation re qui red), their eas y-to -de ter mine va len ce, and their h igh er inten sit y, compared t o
written or audio stimuli. However, as mentioned before, using a common video across participants
within a given mood condition raises the possibility of confounding between a ective experience
and the semantic or episodic content of the video. To try to mitigate this problem, one can consider
using di erent stimulus replicates across conditions or experiments. Another potential drawback
of video-based inductions is that, compared to other procedures, exposure to videos may also facil-
itate hypothesis guessing and, consequently, demand artifacts. To avoid such concern, the cover
story must be “convincing” and, also importantly, the a ect manipulation check disguised. For
example, Cohen and Andrade (2004) used a combined technique of video plus personal experience,
in which participants were informed that the university, in order to augment its web-based teach-
ing environment, attempted to assess the impact of audio and video stimuli transmitted through
the web. Students were informed that they would watch ve minutes of a video and then would
perform memory and judgment recall tasks. A er the video, the “memory task,” instructed them
to write a personal story related to the scenes watched in the clip. A er the “memory task,” a “judg-
ment task” (i.e., the manipulation check) asked participants to assess ten items related to the video.
Only three of them were a ect-related. e other items were in line with the general cover story.
is manipulation has shown strong and reliable e ects on people’s feelings and, importantly, very
low incidence of hypothesis guessing (see also Andrade, 2005).
Videos have also been used to manipulate speci c emotional states, such as anger (Andrade
& Ariely, 2006; Phillipot, 1993) sadness and disgust (Lerner, Small, & Loewenstein, 2004), and
Haugtvedt_ER5603X_C011.indd 300Haugtvedt_ER5603X_C011.indd 300 7/10/2007 4:47:35 PM7/10/2007 4:47:35 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 301
fear (Andrade & Cohen, 2007). Restricting the e ect to one speci c emotion may be challenging;
some video manipulations can enhance more than one speci c a ective state at the same time. For
instance, Gross and Levenson (1995) showed that an anger manipulation tended to increase disgust
levels as well. As they pointed out, “With lms, it appears that there is a natural tendency for anger
to co-occur with other negative emotions” (p. 104). Still, videos and some combined techniques
(video and personal story writing) are relatively successful a ect manipulations (Westermann,
Spies, Stahl, & Hesse, 1996).
PHYSIOLOGICAL AND COGNITIVE ANTECEDENTS OF EMOTION
e in uential James-Lange theory (James, 1884) held that emotional stimuli elicited bodily
responses, that is, peripheral activity such as changes in heart rate, blood pressure, and skin con-
ductance, and that these bodily responses were translated fairly directly into conscious di er-
ences in emotional experience (e.g., fear versus anger). While there was modest success relating
“energetic” physiological responses to higher arousal negative a ect (compared to lower arousal
states such as sadness and guilt), there was no consistent translation of bodily responses into
di erential positive a ect. More generally, such physiological measures do not appear to re ect
essential di erences in the valence of emotion (Bradley, Cuthbert, & Lang, 1993; Schimmack &
Crites, 2005). One response to the failure to support the James-Lange theory was to search for
other, more sensitive indicants of emotional response that could then be interpreted as particular
types of emotion. Facial feedback theories identi ed patterns that corresponded to happiness,
surprise, sadness, fear, anger, and disgust (Ekman, 1973; Izard, 1977; Kleinke, Peterson, & Rut-
ledge, 1998). However, a meta-analysis of these studies (including those where participants were
induced to adopt musculature associated with smiling and frowning) indicated that these e ects
were too weak to perform the central function ascribed to bodily responses in the James-Lange
theory (Matsumoto, 1987).
A more basic challenge to the original theory was to question the central role of bodily response
to subsequent emotional experience. Schachter and Singer (1962) made signi cant inroads by show-
ing (via injections of either epinephrine or a placebo) that peripheral arousal only di erentiated an
emotional response from merely cognitive responses. In their two-factor theory, cognitive pro-
cesses played the decisive role in interpreting the arousal that was being experienced. A substantial
challenge to the bodily arousal component of this theory can be seen in other research conducted at
about the same time. Lazarus and Alfert (1964) asked people to watch a lm depicting a tribal ritual
involving what appeared to be genital mutilation. However, half of those watching were given mis-
information that the experience was actually not painful and that adolescents looked forward to
this initiation into manhood, and signi cant cognitive control over arousal was observed. Subse-
quent research on spinal cord injured patients best supports the view that peripheral arousal is not
necessary to the experience of emotion, but can intensify it (Mezzacappa, Katkin, & Palmer, 1999).
However, the importance of emotional intensi cation should not be minimized. Recent research
on memory, for example, demonstrates the importance of such emotional experience to memory
consolidation, and is, thus, consistent with evolutionary underpinnings of classical conditioning
(Cahill & McGaugh, 1998). More generally, emotional response was shown to be far more under
cognitive control and appraisals of experience than had been imagined.
Since then, cognitive appraisal theories have dominated research on emotion (see Scherer,
Schorr, & Johnstone, 2001, for current theoretical perspectives). While emotional underpinnings
may be somatic, and in that sense have signi cant evolutionary value in predisposing the body
toward approach/appetitive or avoidance/inhibitory action, modern theories point to relatively few
Haugtvedt_ER5603X_C011.indd 301Haugtvedt_ER5603X_C011.indd 301 7/10/2007 4:47:35 PM7/10/2007 4:47:35 PM
302 JOEL B. COHEN, MICHEL T UA N PHAM, AND EDUAR DO B. ANDRADE
hardwired connections to discrete emotional states (Oatley & Johnson-Laird, 1987; Ortony, Clore,
& Collins, 1988). Instead, such theories stress the involvement of cognitive appraisal (Scherer,
Schorr, & Johnstone, 2001). ese appraisal processes assign evaluative meaning to objects and
events (i.e., desirable versus undesirable) and facilitate causal attributions to sources either in the
external world, such as another person’s actions or in our own behavior or thoughts. e combi-
nation of evaluative meaning, assessment of internal versus external causation and responsibility,
and temporal perspective is then assumed to produce such highly di erentiated emotions as hap-
piness, pride, envy, disgust, sadness, anger, and fear (see Lazarus, 1991). While research supports
the role of arousal in the experience of relatively intense emotions, cognitive processes (e.g., telling
someone that an experience is/is not painful) play a major role and have been shown to alter the
experience even when assessed by heart rate and skin conductance.
Building on earlier work by Leventhal (1980), Buck (1985), and Ho man (1986), Cohen and
Areni (1991) advanced a three-phase dynamic model in which activation of a mental concept (e.g.,
identi cation of a ashing red light) produces a largely unconscious and very rapid, sensory-level
a ective response. Some such responses may be largely innate (e.g., surprise) or at least potentiated
by evolutionary processes, for example, (e.g., responses to taste; Steiner, Glaser, Hawilo, & Berridge,
2001). Others require very little learning beyond simple association to become generalized ten-
dencies, such as preferences for smiling and familiar faces. ese phase-one emotional responses
interrupt other cognitive processes, orient attention, and bring resources to bear on the instigating
stimulus. In phase two, the cognitive system attaches somewhat greater meaning to the stimulus by
automatically extracting easily processed stimulus information and associating it with experienced
pleasantness/unpleasantness and arousal. us, the second-phase a ective response becomes more
di erentiated through the operation of associational, rather than reasoning processes. In phase
three, a ective experience results from cognitive elaboration, thereby taking into account con-
text and previous experience. At stage two, and to a much greater degree at stage three, cognitive
appraisal can enhance or suppress arousal (Lazarus & Alfert, 1964) as well as create more nuanced
feeling states, such as disgust rather than sadness.
MEMORY FOR AFFECTIVE EXPERIENCES
ere is considerable evidence that the arousal intensity of an a ective experience increases peo-
ple’s immediate and long-term memory for this experience (Bradley, Greenwald, Petry, & Lang,
1992; Kroeber-Riel, 1979; orson & Friestad, 1989), especially with respect to the central elements
of this experience (Christianson, Lo us, Ho man, & Lo us, 1991). is appears to be the case even
when the source of arousal is unrelated to the material to be learned and comes a er the learning
has taken place, which suggests that the phenomenon may be due, in part, to a better consolida-
tion of memory traces under high emotional arousal (Nielson, Yee, & Erickson, 2005). Emotional
intensity is no guarantee of memory accuracy, however. Biases due to changes in cognitive apprais-
als of the events or revised standards of judgment (e.g., looking back, a person may have a di erent
perspective on the emotion-eliciting event) as well as a desire to see things di erently (e.g., when
anticipating a recurrent experience such as childbirth) may intrude on people’s memory (Levine,
1997; Levine, Prohaska, Burgess, Rice, & Laulhere, 2001). Retrospective assessments of a ective
experiences also seem to be more impacted by intensity at both the peak and the end of the expe-
rience, with duration playing a less signi cant role (Ariely & Loewenstein, 2000; Fredrickson &
Kahneman, 1993; Kahneman, Fredrickson, Schreiber, & Redelmeier, 1993).
Haugtvedt_ER5603X_C011.indd 302Haugtvedt_ER5603X_C011.indd 302 7/10/2007 4:47:36 PM7/10/2007 4:47:36 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 303
THE STRUCTURE AND ASSESSMENT OF AFFECT
ere are two separate research traditions among those whose work involves the assessment of
a ect. e rst is to identify the underlying dimensions of a ect by analyses of judged similarities
and semantic di erential ratings of mood terms, as well as facial and vocal emotional expressions.
e second combines a more functional/motivational analysis with evidence from studies of neu-
rophysiological and hormonal processes.
e rst body of work supports the existence of two general dimensions: pleasantness versus
unpleasantness and activation/arousal/engagement (see Remington, Fabrigar, & Visser, 2000;
Russell & Carroll, 1999a; see Russell & Carroll, 1999b; Watson, Wiese, Vaidya, & Tellegen, 1999).
Research in this tradition is heavily measurement-based and has been extended into several prac-
tical domains to classify a ective responses to stimuli of interest, such as pictures and advertise-
ments, as well as to provide a more general basis for delineating categories of emotional response
(Watson & Tellegen, 1985).
AFFECT TAXONOMIES IN CONSUMER RESEARCH
Researchers with a primary interest in a ective aspects of stimuli, such as advertisements and
how people describe their a ective responses to them, have been less interested in the underlying
dimensionality of a ect and o en prefer to think in terms of a ect taxonomies that correspond
to more macro-level constellations or prototypes (Shaver, Schwartz, Kirson, & Oconnor, 1987).
ey rely on specially constructed inventories of mood and emotion terms, as well as scales that
have been developed for other purposes (e.g., to represent appetitive and aversive motivational sys-
tems). In research on advertising, for example, Holbrook and Batra (1987) began with over 90 items
that combined emotional responses and evaluative reactions to advertising content, and Edell and
Burke (1987) used a 69-item inventory of feelings. In such research, investigators typically attempt
to reduce the individual items to distinct clusters using techniques such as factor analysis and hier-
archical cluster analysis (Shaver, Schwartz, Kirson, & Oconnor, 1987). At times, claims are made
about underlying structure, but the generality of such claims is questionable because of the arbi-
trary selection of items and stimuli, the ambiguity of hierarchical con gurations of emotion terms,
as well as measurement issues to be discussed below (see also Mano, 1991; Schimmack & Crites,
2005). Nevertheless, such research may serve the investigator’s needs in di erentiating between
types of a ective responses to content and situations (e.g., store settings) of particular interest.
Much of the earlier a ect taxonomy research in consumer behavior, at least through 1990, was
reviewed by Cohen and Areni (1991), and so it will not receive explicit attention here.
An extremely comprehensive analysis of many of the emotion measures used in consumer
research was carried out by Richins (1997), who identi ed shortcomings in their ability to address a
greater variety of consumption experiences. ese included the contemplation, purchase, use, and
subsequent reactions to a broad variety of products and services from the mundane to the impor-
tant and sentimental (see also C. Derbaix & Pham, 1991). She identi ed a list of 175 emotion terms
that had been used in consumer research and that satis ed criteria developed by Ortony, Clore,
and Collins (1988) to screen out nonemotion terms focusing on bodily states such as “sleepy,” sub-
jective evaluations such as “feeling con dent,” behaviors and action tendencies such as “crying”
and “hesitant,” and cognitive states such as “interested.” She supplemented this list by prompting
open-ended self reports of positive and negative feelings (most commonly, the positive a ective
terms “happiness,” “relief,” and “excitement,” but also “worry,” “sadness,” and “guilt”) to a vari-
Haugtvedt_ER5603X_C011.indd 303Haugtvedt_ER5603X_C011.indd 303 7/10/2007 4:47:36 PM7/10/2007 4:47:36 PM
304 JOEL B. COHEN, MICHEL T UA N PHAM, AND EDUAR DO B. ANDRADE
ety of consumption experiences. Although the underlying dimensionality of the resulting “Con-
sumption Emotion Set” (CES) is somewhat ambiguous (beyond the traditional positive-negative
axis), the instrument appears to be quite useful for those who wish to assess consumers’ a ective
responses to a more comprehensive set of consumption experiences.
UNDERLYING DIMENSIONS: THE BIPOLARITY OF AFFECT
More basic research on the structure of a ect attempts to identify relationships among two primary
components of a ect, pleasantness, and arousal/activation. Russell (1980) originally proposed that
these two dimensions be viewed as a circumplex, that is, a model in which individual mood and
emotion descriptors are systematically arranged around the perimeter of a circle. Data from Rus-
sell and Feldman Barrett, (1999) indicate that a ective structure actually falls somewhere between
a classic simple structure in which the variables cluster in dense groups around labeled axes and
a true circumplex, as in Figure 11.1, in which the variables are more evenly spaced and de ne a
complete circle.
Within this measurement tradition, there has been a debate over the bipolarity versus indepen-
dence of positive and negative a ect. When people experience and report a ect, and negative a ect
is high, does that mean that positive a ect is low (i.e., bipolarity)? When thinking about implica-
tions of bipolarity, it is important to di erentiate between “core a ect” (Russell & Barrett, 1999)
and evaluative outcomes of a ective processes. e latter are the result of cognitive appraisals and
clearly allow for mixed assessments (i.e., positive in some respects and negative in others). Core
a ect, on the other hand, refers to how a person is feeling emotionally at a point in time. Also, when
Figure 11.1 e circumplex model of a ect. e circumplex model describes a ect in terms of
the two orthogonal dimensions of valence and activation. From “ e structure of currect a ect:
Controversies and emerging consensus,” by L. Feldman Barrett and J. A. Russell, 1999, Current
Directions in Psychological Sciences, 8, 11. Copyright 1999 by the American Psychological Society.
Reprinted with permission.
Haugtvedt_ER5603X_C011.indd 304Haugtvedt_ER5603X_C011.indd 304 7/10/2007 4:47:36 PM7/10/2007 4:47:36 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 305
the analysis shi s to over-time emotional experiences (Diener & Irannejad, 1986; Diener, Smith,
& Fujita, 1995), bipolarity assumptions o er no predictions about independence; people who are
asked to report emotional states using experience sampling diaries are just as likely to have self
reports indicating high average negative a ect, regardless of their average levels of positive a ect.
A general consensus has emerged that a bipolar structure dominates (see Russell & Carroll,
1999a), as acknowledged even by or iginal proponents of the independence assumption (see Watson
& Clark, 1997; Watson, Wiese, Vaidya, & Tellegen, 1999). While interested readers should consult
the extensive literature directly, several key issues must be noted. First, tests of independence versus
bipolarity require truly independent measures of positive and negative a ect. Using bipolar items
(e.g., –3 = sad, 0 = neutral, +3 = happy) arti cially promotes bipolarity. On the other hand, using
unipolar items o en leaves the lower endpoint of the scale ambiguous. For example, if respondents
are asked to rate their happiness on a 1–5 scale going from “not at all” to “extremely,” does a “1”
mean a mere absence of happiness or an opposite state such as sadness? In addition, where does
neutrality lie in such a scale? Such measurement issues can have major in uences on the factor
structure of the resulting scale.
Even assuming independent measurement of positive and negative a ect, the bipolarity of
speci ed circumplex axes (corresponding to particular a ective states shown in Figure 11.1 and
measured by various scales) has proven to be far more problematic. A primary reason is the con-
founding caused by variations in arousal/engagement that prevent the semantic opposites in Fig-
ure 11.1 from lying along a 180° angle. Across a number of such studies, positive and negative
a ect have been found to be moderately negatively correlated (typically around r = –.44) rather
than nearly perfectly negatively correlated, implying bipolarity, or nearly perfectly uncorrelated,
implying independence (Diener, Smith, & Fujita, 1995; Watson, Clark, & Tellegen, 1988; Watson,
Wiese, Vaidya, & Tellegen, 1999). However, the correlation increases dramatically when people
report a ect in the presence of strong, intense emotion (Diener & Irannejad, 1986), demonstrating
the bipolarity of strong emotional states. On the other hand, feelings originally characterized as
low positive a ect (hence low in activation/arousal) and that correspond to quietude and calmness
are not far removed from low negative a ect states (e.g., sluggish) characteristic of mild depression.
Consider a person who views himself as experiencing a lack (or loss) of pleasure or lack of response
to pleasurable stimuli. At low levels, then, self reports of positive and negative a ect can be posi-
tively correlated (Watson, Wiese, Vaidya, & Tellegen, 1999), suggesting independence.
Underlying neurophysiological processes clearly register and potentiate positive and negative
a ect simultaneously. ough con ict and/or ambivalence would seem to be logical consequences
of events and stimuli that simultaneously prompt fear and excitement, making e ective reactions
di cult, emotional paralysis in not the norm. A great deal of current research beyond the scope of
this chapter is providing substantial insight about how these underlying neurophysiological pro-
cesses translate into more molar responses. We will discuss some relevant work on “mixed emo-
tions” a bit later.
THE PANAS SCALES: COMBINING VALENCE AND ACTIVATION
Consumer researchers rely heavily on the positive and negative a ect scales of the PANAS (Wat-
son, Clark, & Tellegen, 1988). e positive cluster consists of active, alert, attentive, determined,
enthusiastic, excited, inspired, interested, proud, and strong. e negative cluster consists of
afraid, ashamed, distressed, guilty, hostile, irritable, jittery, nervous, scared, and upset. While this
instrument has proven to be useful in many studies, there are two signi cant issues that consumer
researchers should consider. First, the two scales do not cover the full range of positive and negative
Haugtvedt_ER5603X_C011.indd 305Haugtvedt_ER5603X_C011.indd 305 7/10/2007 4:47:37 PM7/10/2007 4:47:37 PM
306 JOEL B. COHEN, MICHEL T UA N PHAM, AND EDUAR DO B. ANDRADE
a ect. Low to moderate activation states (e.g., happy, positive, satis ed, serene, pleased) are miss-
ing from the positive a ect scale where high activation states (e.g., active, alert, attentive, excited)
predominate. Low to moderate activation states (e.g., unhappy, negative, depressed, sad) are also
missing from the negative a ect scale where high activation states (distressed, jittery, upset) pre-
dominate. is is not an oversight, though perhaps an inartful choice of labels. Watson and col-
leagues were clear at the outset that they wished to capture a combination of positive a ect and
activation (and the same enhanced activation on the negative side). In e ect, Watson and Tellegen
(1985) rotated the axes 45° to attempt to focus on two orthogonal dimensions that they now term
positive activation and negative activation (Watson, Wiese, Vaidya, & Tellegen, 1999).
e functional signi cance of these two dimensions was recognized by Fowles (1994). PA cor-
responds to a ect that energizes and facilitates approach behavior and reward seeking; whereas
NA corresponds to a ect that inhibits similar behavior and leads instead to avoidance—a “stop,
look and listen” response to the environment. Considerable research in neurophysiology is being
directed to understanding the processes responsible for these appetitive and aversive e ects, but
that is beyond the scope of this chapter (for a review, see Lang, 1995).
Most recently, then, a ect researchers have been rede ning the PANAS instrument to more
appropriately recognize its measurement of positive activation and negative activation (J. T. Larsen,
McGraw, & Cacioppo, 2001; Schimmack & Crites, 2005). e second and related problem with the
PANAS instrument is that it violates the semantic-opposites requirement listed earlier for a test of
bipolarity. Only positive a ective states of high activation are semantically opposite (180° away)
of negative a ective states of low activation. Russell and Carroll (1999a) note that the negative set
includes none of the semantic opposites of the positive set because the opposite of positive activa-
tion is not negative activation, but a state combining negative a ect and low arousal. Accordingly,
PANAS does not have psychometric properties that allow it to be used to investigate questions
involving the independence of positive and negative a ect.
In a reanalysis of available data, Russell and Carroll (1999a) identi ed three clusters of posi-
tive items that could alternatively be viewed as varying continuously in arousal. e rst cluster
involves positive a ective states of high activation such as being enthused, ebullient, excited, and
energetic. A second cluster involves positive a ective states of moderate activation such as being
happy, grati ed, pleased and content. A third cluster involves positive a ective states of low activa-
tion such as being calm, serene, tranquil and relaxed. A parallel clustering was uncovered on the
negative a ect side: (1) negative a ective states of high activation such as being tense, upset, jittery,
and nervous; (2) negative a ective states of moderate activation such as being unhappy, miserable,
discontent and troubled; and (3) negative a ective states of low activation such as being depressed,
bored, lethargic, and glum.
MIXED EMOTIONS
A person can feel sad and guilty or happy and proud at the same time. But is it possible to feel guilty
and proud (or any other combination of oppositely valenced emotions) at the same time? Imagine,
for example, being very successful in a negotiation in a third-world country that deprived the seller
of money that had signi cantly greater value to her than to you. You might feel pride at your skill
(particularly if others in your group did less well in similar negotiations), but you also may experi-
ence guilt. Do we simply shi back and forth in such emotional quandaries (alternating between
positive and negative feelings), or can we actually be happy and sad at the same time? Note that this
is a di erent issue from an evaluation of one’s behavior (which might be tempered because of oppo-
sitely-valenced emotions) or how a person would translate mixed emotions in responding to a sad-
Haugtvedt_ER5603X_C011.indd 306Haugtvedt_ER5603X_C011.indd 306 7/10/2007 4:47:38 PM7/10/2007 4:47:38 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 307
happy scale, since a person would need to resolve that con ict in order to make such judgments.
Russell and Carroll (1999a) argue that the bipolarity of emotional experience implies that when you
are happy, you are not sad, just as when you are hot, you are not cold. Strictly speaking, that claim
is very strong since being at one point on the abscissa of an a ect distribution precludes being at
any other point. As Larsen et al. (2001) point out, bipolarity implies a linear relationship, and thus
a correlation close to -1. Mutual exclusivity of positive and negative a ect, on the other hand, would
produce intermediate degrees of independence. But does any level of correlation between measures
of positive and negative a ect imply that any (even low) levels of happiness preclude experiencing
sadness? Watson and Tellegen (1999) maintained that sadness decreases as happiness increases. In
that sense, happiness and sadness regularly co-occur and are only mutually exclusive when people
are maximally happy or maximally sad.
Williams and Aaker (2002) employed either happy, sad, or mixed emotional appeals by combin-
ing the same picture with di erent characterizations of it. ey found that the acceptance of the
duality of emotions (via mixed emotional appeals) was greater among Asian American than Anglo
Americans (who actually reported discomfort), although both groups reported experiencing a
combination of happiness and sadness when given the mixed emotional appeal. Such self reports
are ambiguous since it is di cult to know whether people assume they should be experiencing
mixed emotions when confronted with a stimulus presenting a happy and sad event or could be
translating their “somewhat sad” (or somewhat happy) feelings on to scales that allow them to
report mixed emotions.
A well-known demonstration of mixed emotions was carried out by Larsen et al. (2001). ey
exposed people to a mixture of happy and sad events in the movie Life is Beautiful in which a
father seeks to keep his child’s spirits up while they are in a concentration camp. Viewers reported
mixed feelings of happiness and sadness. Since people may well have alternated back and forth,
in a subsequent study, Larsen et al. (2004) created a gambling context which led to disappoint-
ing wins (because people expected more favorable outcomes) and relieving losses (because people
expected even worse outcomes). A button-pressing task suggested that people, in fact, experienced
mixed emotions simultaneously. Note, however, that these are unlikely to be extremely positive and
extremely negative feelings—a combination that seems di cult to imagine. So, another possibility
is that people subjectively interpreted these more moderate feelings as though they were part posi-
tive and part negative in light of the information they were given. Such ndings raise important
questions about the nature of a ective experience.
Evidence supporting mixed emotions has also shed new light into the potential processes under-
lying the consumption of products and services that—at least from an outside observer’s point
of view—are expected to produce negative feelings (e.g., watch horror movies, practice danger-
ous sports, etc). Andrade and Cohen (2007) point out that previous theories tended to rely on the
assumption that positive and negative feelings cannot be experienced at the same time. As a result,
people who deliberately exposed themselves to apparent sources of negative feelings either do not
experience much negative a ect (Fenz & Epstein, 1967; Zuckerman, 1996) or focus on its relieving
consequences—a er removal of the aversive stimuli (Solomon & Corbit, 1974; Zillmann, 1980).
Andrade and Cohen showed the importance of relaxing the single a ective valence assumption in
order to provide a more complete understanding of the phenomenon. In a series of four studies,
it was shown that a er exposing participants to horrifying scenes of a lm clip (scenes from e
Exorcist), fear approach (horror movie watchers) and fear avoidance (non-horror movie watch-
ers) consumers displayed strikingly similar levels and patterns of negative a ect. However, fear
approach consumers also showed increased levels of positive a ect, whereas fear avoidance con-
sumers showed no signs of positive experiences. Also importantly, the authors demonstrated that
Haugtvedt_ER5603X_C011.indd 307Haugtvedt_ER5603X_C011.indd 307 7/10/2007 4:47:38 PM7/10/2007 4:47:38 PM
308 JOEL B. COHEN, MICHEL T UA N PHAM, AND EDUAR DO B. ANDRADE
mixed feelings are more likely when individuals are able to place themselves within a protective
(detachment) frame: “An ideal detachment frame gives people the ability to increase psychological
distance from the main actors of the movie, while still absorbing the impact of the scenes” (p. 32).
When presented with the actors’ biography prior to the movie, and the actors’ “regular” pictures
next to the video during the scenes, which was supposed to remind the audience that these were
simply actors playing a role, fear-avoidance and fear-approach participants reacted similarly, both
displaying increased levels of positive or negative a ective states during the movie. People tend to
deliberately choose to expose themselves to sources of negative a ect when a psychological pro-
tective frame is present, as it allows for the co-activation of positive and negative a ect. A similar
rationale has been adopted to show the presence of mixed feelings of disgust and amusement as a
result of video exposure (Hemenover & Schimmack, 2004).
AFFECT AND CONSUMER JUDGMENT AND DECISION MAKING
It is useful to distinguish three types of a ect in consumer judgment and decision making. Integral
a ect refers to a ective responses that are genuinely experienced and directly linked to the object
of judgment or decision.2 Integral a ective responses include momentary feelings experienced
through direct exposure to the object itself (such as the pleasant feeling of tasting a ne wine) and
those experienced in response to some representation of the object—a representation that may be
externally provided (e.g., a TV commercial for a product) or internally generated (e.g., thinking
about a product). ese a ective responses are integral to the extent that they are elicited by fea-
tures of the object, whether these features are real, perceived, or only imagined.
Incidental a ect refers to a ective experiences whose source is clearly unconnected to the object
to be evaluated. Most of the literature on mood e ects on consumer behavior (e.g., Gardner, 1985;
e.g., Kahn & Isen, 1993; Lee & Sternthal, 1999) deals with incidental a ect in that the source of the
mood is typically unrelated to the judgment or decision being made. In addition to a person’s cur-
rent mood, incidental a ect may also come from a person’s emotional dispositions (such as chronic
anxiety or depression) and temperament (such as general optimism or pessimism), or from any
contextual stimuli associated with integral a ect (such as background music, pleasant scent, etc.).
Task-related a ect lies somewhere between integral and incidental a ect. It refers to a ective
responses that are elicited by the task or process of making judgments and decisions, as opposed to
direct, integral responses to features of the target objects or purely incidental feelings. For example,
the emotional stress of having to choose between two very attractive o ers would be considered
task-induced in that it is the process of having to choose between these two o ers that is stressful,
not the o ers themselves. Indeed, decisions may trigger unpleasant task-related a ect even when
the options are associated with pleasant integral a ect, for example, a choice between two vacation
destinations. In the above example, the emotional stress experienced would not be incidental either
because, by de nition, it would not have arisen had a judgment or decision not been required. Each
type of a ect will be discussed separately in relation to consumer judgment and decision making.
INTEGRAL AFFECT IN JUDGMENT AND DECISION MAKING
e In uence of Integral A ect on Target Evaluation
Numerous studies across various disciplines show that integral a ective responses to a target
object—whether the object is a product, a person, or a company—are o en incorporated into a
summary evaluation of the object. In general, though not always, objects that elicit pleasant feel-
Haugtvedt_ER5603X_C011.indd 308Haugtvedt_ER5603X_C011.indd 308 7/10/2007 4:47:39 PM7/10/2007 4:47:39 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 309
ings, such as a beautiful symphony, a sweet dessert, or an attractive singer, are evaluated more
favorably, and objects that elicit unpleasant feelings, such as a noisy apartment, a sour tasting dish
or a rude salesperson, are evaluated less favorably. e relation between integral a ective responses
and object evaluation is so strong that, for a long time, a ect and evaluation (or attitude) were
considered to be synonymous (e.g., Fishbein & Azjen, 1975). Nevertheless, despite the generally
strong positive correlation between measures of integral a ective responses and measures of over-
all evaluation, there is growing consensus that the two constructs are theoretically and empirically
distinct, with integral a ective responses generally conceived as one of several potential anteced-
ents or determinants of overall evaluation or attitude (see Breckler & Wiggins, 1989; Crites, Fabri-
gar, & Petty, 1994; Pham, Cohen, Pracejus, & Hughes, 2001; Wyer, Clore, & Isbell, 1999; Zanna &
Rempel, 1988).
In an early demonstration of the in uence of integral a ective responses on summary evalu-
ations, Abelson, Kindler, Peters, and Fiske (1982) documented in two large surveys that people’s
emotional responses to prominent politicians (notably, Jimmy Carter, Gerald Ford, and Ronald
Reagan) were highly predictive of their overall attitudes and preferences toward these politicians.
Importantly, people’s emotional responses predicted their attitudes over and above their own party
a liation and their assessments of the politicians’ traits and behaviors. A major theme of research
in social psychology and consumer research during the 1980s and 1990s was that integral a ec-
tive responses to various target objects may predict judgment, choice, and behavior toward these
objects over and above assessments based on more descriptive or “cognitive” bases of judgments
such as beliefs, stereotypes, base-rates, prior attitudes, etc. In marketing and consumer research,
this theme was pursued most extensively in the advertising domain. A large number of studies have
indicated that a ective responses to advertisements have direct e ects on consumers’ attitudes
toward the ad (Aad) and at least indirect e ects on consumers’ attitudes toward the brand (Ab)
through the e ects on Aad (cf Aaker, Stayman, & Hagerty, 1986; Batra & Ray, 1986; Brown, Homer,
& Inman, 1998; Brown & Stayman, 1992; Edell & Burke, 1987; Holbrook & Batra, 1987). Some stud-
ies indicate that ad-induced a ective responses may also in uence Ab directly, independently of
Aad. (e.g., Burke & Edell, 1989; C. M. Derbaix, 1995; Morris, Woo, Geason, & Kim, 2002; Stayman
& Aaker, 1988) However, the ndings probably deserve further analysis because, at the time, less
attention was paid to separating actual a ective responses to the ad from sheer liking of the ad. In
addition, the mechanisms explaining how a ect was transferred from the ad to the brand were not
closely examined.
Conceptually related results have been obtained across a variety of other domains of judg-
ment, choice, and behavior. For instance, Bodur, Brinberg, and Coupey (2000) indicated that a ect
toward various AIDS prevention behaviors such as abstinence or condom usage predicted attitudes
and intentions toward these behaviors over and above personal beliefs about these behaviors. Simi-
larly, Allen, Machleit, and Klein (1992) reported that emotional responses to past blood donations
predicted future donation behavior over and above expressed attitudes toward donation. Likewise,
Oliver (1993) indicated that a ective responses to products such as cars and a college class pre-
dicted overall satisfaction with these products over and above one’s satisfaction with the products’
speci c attributes. In the investment domain, MacGregor, Slovic, Dreman, and Berry (2000) found
that investment banking students’ feelings toward various industry sectors (electronics, managed
healthcare, etc.) were strongly predictive of their intentions to invest in these sectors, independent
of the sectors’ nancial fundamentals. In summary, the literature indicates that integral a ective
responses to a variety of target objects in uence consumers’ overall evaluations of and behaviors
toward these objects beyond more descriptive bases of object evaluation such as beliefs, stereo-
types, prior attitude, etc. However, as noted above, some caution may be called for because some of
Haugtvedt_ER5603X_C011.indd 309Haugtvedt_ER5603X_C011.indd 309 7/10/2007 4:47:39 PM7/10/2007 4:47:39 PM
310 JOEL B. COHE N, MICHEL TUAN PHAM, AND EDUAR DO B. ANDRADE
the studies in this line of research may not adequately distinguish actual feeling states from overall
liking responses.
Underlying Processes
ree mechanisms may explain why integral feelings have a direct in uence on overall evaluation
and behavior independent of one’s descriptive knowledge about the target. e rst two mecha-
nisms both imply a noninferential, “automatic” in uence of integral feelings on target evaluation.
e rst possibility is that integral feelings enter evaluations directly through simple evaluative
conditioning. Evaluative conditioning refers to the transfer of evaluative meaning across stimuli
that are presented simultaneously (see Staats & Staats, 1957); it di ers from classical (Pavlovian)
conditioning in that one stimulus does not serve as a signal of another (see De Houwer, omas, &
Baeyens, 2001). A close proximity between a target and an integral feeling experience may result
in the evaluative meaning of the feelings (mostly their valence) being carried over to the target—a
mechanism sometimes called “a ect transfer” in consumer research and marketing (e.g., Macken-
zie, Lutz, & Belch, 1986). A second possibility builds on the idea that a ective experiences are asso-
ciated with par ticu lar action tendencies such as approach, avoidance, withdrawal, or conf rontation
(see Frijda, Kuipers, & Terschure, 1989). ese action tendencies may not only carry over to actual
behavior, but also be translated spontaneously into proxies of behavior such as evaluations and
intentions. is mechanism would be consistent with the emerging view of a ect as an embodied
mode of evaluation (e.g., Clore & Schnall, 2005; Damasio, 1994; Forster, 2004; Zajonc & Markus,
1982).
A third mechanism is based on the hypothesis that integral a ective responses are o en viewed
as sources of information during object evaluation (Schwarz, 1990; Schwarz & Clore, 1996). To eval-
uate a target, people may consciously inspect their feelings to see “how they feel” about it. Pleasant
feelings would be interpreted as evidence of liking, satisfaction, well-being, and so on; unpleasant
feelings would be interpreted as evidence of disliking, dissatisfaction, misery, and so on. is pro-
cess is known as the “how-do-I feel-about-it?” heuristic (Schwarz & Clore, 1988). Numerous studies
have documented the existence and operation of this heuristic both in psychology (e.g., Schwarz &
Clore, 1983; see Schwarz & Clore 1996, for a review) and in consumer research (e.g., Gorn, Gold-
berg, & Basu, 1993; Pham, 1998; Pham, Cohen, Pracejus, & Hughes, 2001). Although the heuristic
was originally proposed as an explanation of incidental mood e ects on judgment (Schwarz &
Clore, 1983, 1988; see the discussion of incidental a ect below), there is growing evidence that the
heuristic is used with integral feelings as well (Pham et al., 2001). In fact, the primary application
of this heuristic—its raison d’être—is in relation to integral a ective feelings (see Pham, 2004, for
a discussion). Some research indicates that the heuristic is also used anticipatorily in consumer
decision making. Sometimes consumers appear to construct “mental pictures” of the alternatives
and assess how they feel as they hold these pictures in their minds (Pham, 1998).3 (See also Gilbert,
Gill, & Wilson, 2002, for conceptually related results). ese pictures appear to be concrete, which
explains why the reliance on momentary feelings in decision making has been found to be more
pronounced among people with high imagery ability (Pham, 1998). Unlike the rst two mecha-
nisms, the “how-do-I-feel-about-it?” heuristic is assumed to be inferential, as opposed to purely
associationistic or mechanistic. at is, people are assumed to re ect on what their integral feelings
mean for the judgment to be made; they do not rely on these feelings automatically (see Avnet &
Pham, 2004, for evidence consistent with this interpretation).
e three mechanisms described above all predict a direct e ect of integral a ective responses
on evaluations. A fourth mechanism posits an indirect e ect. It has been suggested that integral
Haugtvedt_ER5603X_C011.indd 310Haugtvedt_ER5603X_C011.indd 310 7/10/2007 4:47:40 PM7/10/2007 4:47:40 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 311
a ective responses enter evaluations only indirectly by changing the person’s perceptions or beliefs
about the target (e.g., Fishbein & Middlestadt, 1995). For instance, feelings of frustration toward a
service provider might reinforce perceptions that “they are not reliable” or trigger beliefs that “they
don’t care about the customer.” It is these perceptions and beliefs—not the feelings that triggered
them—that are then summarized and integrated into the overall evaluation. Consistent with this
mechanism, some studies indicate that a ective responses to advertisements may also in uence
brand attitudes by changing brand beliefs (see Brown & Stayman, 1992; Mackenzie, Lutz, & Belch,
1986). is mechanism is also consistent with a major explanation of incidental mood-congruency
e ects on evaluations (discussed further below). According to this explanation, evaluations tend to
be assimilated toward incidental mood states because these states cue mood-consistent materials in
memory, which then color perceptions of the target (Isen, Shalker, Clark, & Karp, 1978).
Properties of Evaluations and Decisions based on Integral A ect
It is widely accepted that, in general, judgments and decisions based on integral feelings are reached
more rapidly than are comparable judgments and decisions based on descriptive inputs. Although
this property was originally assumed mostly on theoretical grounds (e.g., Epstein, 1990; Zajonc,
1980), it has since been documented empirically both with stimulus-based evaluations (Pham,
Cohen, Pracejus, & Hughes, 2001) and with memory-based evaluations (Verplanken, Hofstee, &
Janssen, 1998). is property should logically extend to decisions and choices as well, although this
prediction remains to be tested. ree factors account for the generally greater speed of judgments
and decisions based on integral a ect. First, integral a ect o en arises very rapidly (e.g., LeDoux,
1996; Zajonc, 1980). Second, integral a ective responses o en enter evaluations through simple
associations. Finally, even if integral a ective responses have to be interpreted, as they do in the
“how-do-I-feel-about-it?” heuristic, their interpretation is generally very clear (Strack, 1992).
It is also widely accepted that judgments and decisions based on integral a ect generally require
less processing resources (e.g., Epstein, 1990). As a result, any constraint on processing resources
(time pressure, distraction, cognitive load, etc.) tends to increase the reliance on integral a ective
responses both in evaluative judgments (Pham et al., 2001) and in choices (Nowlis & Shiv, 2005;
Shiv & Fedorikhin, 1999). For example, when given a choice between a tempting piece of chocolate
cake (an a ectively attractive option) and a healthier fruit salad (a “cognitively” attractive option),
consumers whose cognitive resources were not constrained tended to choose the healthier fruit
salad. However, when cognitive resources were constrained, consumers tended to choose the more
tempting cake, presumably because a ective drivers of preference still operated while the more
cognitive drivers could not (Shiv & Fedorikhin, 1999). Similarly, Nowlis, and Shiv (2005) found that
distracting consumers while they are sampling a pleasant-tasting but relatively unknown brand of
chocolate increases the likelihood that they will subsequently choose the sampled brand over a bet-
ter-known brand of chocolate. Again, this is presumably because distraction increases the relative
weight attached to the pleasant integral feelings associated with the sampling experience.
Evaluations and decisions based on integral a ect also tend to be myopic. Immediate a ective
rewards and punishments tend to be weighted too heavily, whereas delayed consequences are not
weighted su ciently (see Loewenstein, 1996). is property is very salient in impulse control situa-
tions where people have to trade o the immediate hedonic consequences of an option (such as the
pleasure of eating junk food or the pain of visiting the dentist) against its long-term consequences
(high cholesterol and obesity; healthy teeth and gums). According to Loewenstein (1996), the myo-
pia of a ect-based judgments and decisions is caused by the di erential accessibility of current
and delayed a ective states. Whereas the experience of immediate integral a ect has strong drive
Haugtvedt_ER5603X_C011.indd 311Haugtvedt_ER5603X_C011.indd 311 7/10/2007 4:47:40 PM7/10/2007 4:47:40 PM
312 JOEL B. COHE N, MICHEL TUAN PHAM, AND EDUAR DO B. ANDRADE
properties (e.g., the cathartic anger release of yelling at an uncooperative salesclerk), it is more
di cult to anticipate and vividly picture future a ective states (e.g., the embarrassment of being
escorted out of the store). As a result, a reliance on a ect tends to yield preferences for options that
are more rewarding (or less painful) in the short term even if these options are less desirable in the
long run. Consistent with this proposition, recent brain imaging studies indicate that preferences
for immediate rewards are associated with greater activation in parts of the limbic system that are
associated with a ect (McClure, Laibson, Loewenstein, & Cohen, 2004). A ective responses, it
seems, are part of a decision-making system of the present (Pham, 2004).
A lesser-known property of evaluations based on integral a ect is that they can exhibit relatively
high consensus. Contrary to popular beliefs that a ect is highly subjective, a growing body of evi-
dence suggests that a ective judgments are, in fact, quite consensual, sometimes even more so than
cognitive judgments. For instance, judgments of physical attractiveness, long thought to be purely
subjective (“beauty is in the eye of the beholder”) have recently been shown to be largely universal
(Etco , 1999). Similarly, emotional responses to music have been shown to be largely shared (Peretz,
Gagnon, & Bouchard, 1998). It has also been observed that, although juries may disagree widely on
the amount of punitive damages they are willing to award in legal cases, they tend to agree strongly
on how outraged they feel in response to each case (Kah neman, Schk ade, & Su nstein, 19 98). In fact ,
for a variety of everyday stimuli, people seem to agree more on how they feel toward the stimuli
than on how they would cognitively assess these stimuli (Pham, Cohen, Pracejus, & Hughes, 2001).
According to Pham et al., (2001), a ect-based judgments will be most consensua l when the integral
a ective responses are triggered through hardwired programs involved in bioregulation (such as
the pleasure experience of eating ice cream) or through emotional schemata acquired through con-
ditioning and socialization (e.g., the outrage elicited by an unprovoked insult). A ect-based judg-
ments will be less consensual when based on integral a ective responses arising through controlled
appraisal processes such as the guilt experienced when attributing one’s failure to a lack of e ort.
Evaluations and decisions based on integral a ec t ad dit iona lly tend t o be sensitive to the presence
or absence of a ect-producing stimuli but relatively insensitive to further variations in the magnitude
of these stimuli. is property was recently demonstrated in an interesting series of studies by Hsee
and Rottenstreich (2004). In one study, respondents were asked to assess how much they would pay
for a used collection of either 5 or 10 Madonna CDs. One group of respondents was primed to make
their assessments based on how they felt toward the target (Madonna and her music); the others
were primed to make their assessment in a calculating fashion. Consistent with the researchers’
predictions, respondents’ willingness to pay for the CD collection was much more sensitive to the
size of the collection among respondents who had had been primed to rely on calculation than
among respondents who had been primed to rely on their feelings. In another study, respondents
were asked to assess how much they would be willing to donate for a rescue e ort that would save
either one or four pandas’ lives. For one group of respondents the number of pandas saved was
simply represented by one or four dots. For the other group, the number of pandas saved was repre-
sented by one or four cute pictures of pandas, which was expected to trigger a more a ective mode
of e valuation. Again, a s pred icted, respondents’ donat ions were much more sensitive to the number
of pandas saved in the a ect-poor (dot) condition than in the a ect-rich (picture) condition. Hsee
and Rottenstreich (2004) explain this phenomenon as follows. A ect-based evaluations are o en
based on concrete mental images of the target (see Pham, 1998). Because these images are discrete,
usually consisting of prototypical representations of the target (a lovely panda, a popular Madonna
song), continuous quantitative information tends to be lost (Kahneman, Ritov, & Schkade, 1999).
Similarly, evaluations and decisions based on integral a ect are relatively insensitive to prob-
abilities, except for the presence or absence of uncertainty (Loewenstein, Weber, Hsee, & Welch,
Haugtvedt_ER5603X_C011.indd 312Haugtvedt_ER5603X_C011.indd 312 7/10/2007 4:47:40 PM7/10/2007 4:47:40 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 313
2001; Rottenstreich & Hsee, 2001). In a telling demonstration of this phenomenon, Rottenstreich &
Hsee (2001) asked respondents how much they would be willing to pay to avoid two types of nega-
tive outcomes, either losing $20 or receiving a painful but harmless electric shock, with an either
1% or 99% probability of occurrence. Consistent with standard economic theory, respondents were
willing to pay much more to avoid a 99% probability of losing $20 (Mean = $18) than to avoid a 1%
probability of losing $20 (Mean = $1). However, when the decision was about receiving an electric
shock, a prospect rich in negative a ect, respondents were not willing to pay much more to avoid
a 99% probability of shock (Mean = $10) than to avoid a 1% probability of shock (Mean = $7).
According to Loewenstein et al. (2001), this phenomenon again arises because a ective decisions
and evaluations o en involve anticipatory a ective responses to discrete images of the options that
do not incorporate probabilities. However, a ect-based decisions and evaluations are sensitive to
deviations from absolute certainty (i.e., from impossibility to small probability and vise versa). For
example, many consumers grossly overpay to turn zero probabilities of winning in big lotteries (a
prospect rich in a ect) into probabilities that are in nitesimal. Similarly, most consumers would be
willing to pay large insurance or security premiums to convert minute probabilities of catastrophic
events (prospects rich in a ect) into zero probabilities. Loewenstein and colleagues (2001) argue
that anticipatory a ective responses such as dread (of negative outcomes) or hope (of positive out-
comes) are sensitive to possibility (i.e., de viat ions from cer taint y) rather tha n act ual probability (see
also Slovic, Finucane, Peters, & MacGregor, 2002).
Finally, evaluations and decisions based on integral a ect tend to have a high degree of internal
coherence (Pham, 2004). is is because integral a ective responses to a target, which are o en
immediate and highly accessible, o en trigger a con rmatory search for information that supports
or helps explain these initial feelings (Pham, Cohen, Pracejus, & Hughes, 2001; Yeung & Wyer,
2004). is con rmatory search results in a strong correlation between the immediate a ective
response elicited by a target and the spontaneous thoughts that people associate with the target.4
is strong correlation in turn results in more polarized evaluations (Adaval, 2003). Consistent
wit h this proposition, Pham et al. (2001) found, for instance, that a ective feelings toward a variety
of stimuli (magazine pictures, TV commercials, etc.) are almost perfect predictors of the thoughts
generated spontaneously by the stimuli. Similarly, Yeung and Wyer, (2004) found that that consum-
ers’ initial a ective responses to a product’s appearance make them more likely to attend and weigh
product attribute information that is evaluatively consistent with the valence of the initial a ective
responses. A review of neurophysiological evidence led Damasio (1994) to a similar proposition:
“Somatic states, negative or positive, caused by the appearance of a given representation, operate
not only as a marker for the value of what is represented, but also as a booster for continued work-
ing memory and attention.” (p. 198) at immediate integral a ect directs subsequent thoughts
may partly explain why personal impressions based on very limited samples of expressive behavior
(watching a 30-second video clip of an instructor teaching) are surprisingly predictive of long-term
evaluations (the instructor’s end-of-semester student evaluations) (see Ambady & Rosenthal, 1992).
Pham (2004) speculates that the internal coherence of a ect-based judgments and decisions may
have had the evolutionary purpose of promoting faster and more e cient behavioral responses to
the environment by increasing the intrapsychic consistency of the signals that the person receives.
Determinants of Reliance on Integral A ect
A number of factors have been found to increase people’s reliance on integra l a ective responses in
judgment and decision making. Although some of these factors were actually identi ed in studies
of incidental a ect, they are discussed here to the extent that they apply to integral a ect as well.
Haugtvedt_ER5603X_C011.indd 313Haugtvedt_ER5603X_C011.indd 313 7/10/2007 4:47:40 PM7/10/2007 4:47:40 PM
314 JOEL B. COHE N, MICHEL TUAN PHAM, AND EDUAR DO B. ANDRA DE
Because integral a ective responses are easy to access, monitor, and interpret (e.g., Pham, Cohen,
Pracejus, & Hughes, 2001; Wyer, Clore, & Isbell, 1999), consumers tend to rely on integral a ect
more when (a) their motivation to process information is low (e.g., Miniard, Bhatla, Lord, Dickson,
& Unnava, 1991; Petty, Schumann, Richman, & Strathman, 1993), (b) they are distracted, cogni-
tively constrained, or under time pressure (e.g., Albarracin & Wyer, 2001; Pham, Cohen, Pracejus,
& Hughes, 2001; Shiv & Fedorikhin, 1999), (c) other bases of evaluation are ambiguous (e.g., Gorn,
Pham, & Sin, 2001; e.g., Isen & Shalker, 1982; Miniard, Bhatla, & Sirdeshmukh, 1992), or (d) they
lack expertise in the target domain (Ottati & Isbell, 1996; Srull, 1987). ese types of results have
led some authors to theorize that the reliance on feelings in judgment and decision making is
primarily a low-involvement, simplifying strategy (e.g., Clore, Schwarz, & Conway, 1994; Forgas,
1995; Petty, Schumann, Richman, & Strathman, 1993). However, the reliance on feelings is not
always a low-involvement heuristic. Certain high-involvement decisions, such as who to marry
or which house to buy, o en seem based on a ect (e.g., P. R. Darke, Chattopadhyay, & Ashworth,
2006). Sometimes, integral a ective responses are considered to be very relevant for the judgment
or decision to be made, even if the judgment or decision is highly involving (Pham, 1998).
Several factors have been found to in uence the perceived relevance of a ective responses in
judgments and decisions. A ective responses to the target are perceived to be more relevant (hence,
are relied upon more) under the following conditions: (a) when the consumer has experiential
motives (e.g., evaluating a book as a potential beach read), as opposed to instrumental motives (e.g.,
evaluating a tax manual as potential help for a tax return) (Pham, 1998); (b) when the judgment
or decision is inherently a ective, e.g., evaluating one’s satisfaction (Clore, Schwarz, & Conway,
1994; Wyer, Clore, & Isbell, 1999); (c) when the consumer makes the decision for himself or herself
as opposed to someone else (Raghunathan & Pham, 1999); (d) when the consumer is promotion-
focused (an inclination to use approach strategies in goal pursuit) as opposed to prevention-focused
(an inclination to use avoidance strategies in goal pursuit) (Pham & Avnet, 2004); and (d) when the
consumer generally trusts his or her feelings (Avnet & Pham, 2005). Overall, it appears that con-
sumers are much more exible in their reliance on feelings than previously thought (Pham, 2004).
Emerging Research on Anticipatory and Anticipated Integral A ect
As demonstrated by Pham’s (1998) research on the use of the “how-do-I-feel-about-it?” heuristics
in consumer decision making, consumers o en make decisions based on feelings that they experi-
ence anticipatorily while holding a mental representation of the target in their minds (see also Gil-
bert, Gill, & Wilson, 2002 for conceptually related results ). ese anticipatory feelings function as
valuation proxies for the anticipated consequences of alternative courses of actions (Pham, 2004).
Although the two notions are o en confused, the notion of anticipatory feelings should be dif-
ferentiated from the notion of anticipated or expected a ect. Anticipatory feelings refer to actual
feeling experiences that arise during the decision process in the course of evaluating a target object
(Bechara, Damasio, Tranel, & Damasio, 1997; Pham, 1998). Although anticipatory feelings may be
subtle, they are genuine feelings with a distinct emotional quality (see Pham, 1993, Experiment
3). In contrast, anticipated or expected a ect refers to predictions about potential a ective conse-
quences of the decision (see Loewenstein & Lerner, 2003). For example, in order to make a choice
between two houses, a consumer may try to predict how happy she and her family would be in each
house. is type of a ect, sometimes called “predicted utility” by decision theorists (Kahneman
& Snell, 1990), is central to the standard economic theory of choice, where people are posited to
make choices based on the predicted hedonic consequences of the various options. ese predic-
tions appear to be strong determinants of risky choice in gambling tasks (Mellers, Schwartz, Ho,
Haugtvedt_ER5603X_C011.indd 314Haugtvedt_ER5603X_C011.indd 314 7/10/2007 4:47:41 PM7/10/2007 4:47:41 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 315
& Ritov, 1997; Mellers, Schwartz, & Ritov, 1999). As shall be discussed later in the section on a ect
and motivation, expected a ect also plays an important role in a ect regulation where options
and behaviors are chosen speci cally in terms of their a ect-changing or a ect-maintaining con-
sequences. However, these predictions are a ective beliefs (i.e., cognitions), not genuine visceral
feelings. Although anticipatory feelings and a ective beliefs (expected a ect) will naturally be cor-
related when the former informs the latter, there can be substantial dissociation between the two
(Raghunathan, Pham, & Corfman, 2006; Robinson & Clore, 2002). For example, whereas genuine
feelings of anxiety tend to push decision makers toward low-risk/low-reward options and genuine
feelings of sadness tend to push decision makers toward high-risk/high-reward options, a ective
beliefs about anxiety or sadness do not have the same in uences (Pham & Raghunathan, 2006).
As Pham (2004) recently pointed out, the distinction between genuine anticipatory feelings and
mere a ective beliefs has important methodological implications. Certain methodologies such as
survey questionnaires about future behavior (commonly used in attitude research) and hypotheti-
cal scenarios (commonly used in decision research) may tap into people’s a ective beliefs (and the
intuitive theories that guide these beliefs) rather than genuine integral feelings.
An important type of anticipatory feeling is the fear response. Although such responses were
examined many years ago in connection with fear appeals in persuasion (Dollard & Miller, 1950;
see Eagly & Chaiken, 1993), they remain important in understanding the reluctance of people to
depart from the status quo and to take action when pathways to successful outcomes are uncer-
tain and risky. Leventhal (1970) argued that threat-related cues instigated both a problem-solving
process (“danger control”) and a process that focused on threat avoidance (“fear control”). To the
extent the latter dominates, people may minimize or rationalize the threat and fail to take e ective
action. is parallel-response model in uenced the development of protection motivation theory
(Rogers, 1975) and the health-belief model (Janz & Becker, 1984) in which perceived vulnerability
and response e cacy are key elements of coping behavior. Consistent with the a ect-as-informa-
tion perspective (Schwarz & Clore, 1996), a growing body of evidence suggests that subjective esti-
mates of risks are largely based on anticipatory feelings elicited by the threat—a proposition known
as the “risk-as-feelings” hypothesis (Loewenstein, Weber, Hsee, & Welch, 2001).
INCIDENTAL AFFECT IN JUDGMENT AND DECISION MAKING
Although integral a ect appears to play a major role in consumers’ judgments and decisions, social
and consumer psychologists have generally focused instead on incidental a ect, especially mood
states.
Congruency E ects of Mood and other Forms of Incidental A ect
Numerous studies have shown that mood states and other forms of incidental a ect generally have
assimilative (i.e., a ect-congruent) in uences on evaluations, decisions, and behaviors. Objects are
typically evaluated more favorably when the evaluator is in a good mood than when the evalua-
tor is in a bad mood (for reviews, see Forgas, 1995; Gardner, 1985), and more generally when the
objects are evaluated in the context of pleasant stimuli than when they are evaluated in the context
of unpleasant stimuli. In fact, some of the earliest demonstrations of this phenomenon appeared
in marketing. In a pioneering study, Axelrod (1963) observed that, a er viewing a depressing tele-
vision documentary, consumers evaluated a variety of products more negatively than they did
before they saw the documentary. ese shi s in evaluation were found to be directly related to
changes in participants’ moods. In another early study, Dommermuth and Millars (1967) found
Haugtvedt_ER5603X_C011.indd 315Haugtvedt_ER5603X_C011.indd 315 7/10/2007 4:47:41 PM7/10/2007 4:47:41 PM
316 JOEL B. COHE N, MICHEL TUAN PHAM, AND EDUAR DO B. ANDRADE
that consumers, who one week earlier had watched a pleasant movie a er tasting a so drink,
evaluated the drink more favorably than consumers who had watched an unpleasant movie. Still,
the most widely-cited demonstration of mood-congruency e ects is the one published more than
10 years later by Isen, Shalker, Clark, and Karp (1978), who observed that people who were put in
a good mood through a small gi were more willing to participate in a survey and evaluated prod-
ucts more favorably than control subjects who did not receive a gi . In a debated study, Gorn (1982)
also observed that people who had seen a pen advertised while pleasant music was playing in the
background were more likely to choose this pen than people who had seen the same pen advertised
with unpleasant music playing in the background. Similar congruency e ects of incidental a ect
have been observed on a variety of other evaluative and behavioral responses, including judgments
of life satisfaction (e.g., Schwarz & Clore, 1983), evaluations of brand extensions (Barone, 2005;
Barone, Miniard, & Romeo, 2000; Yeung & Wyer, 2005), evaluations of political candidates (Isbell
& Wyer, 1999), judgments of perceived risks (Johnson & Tversky, 1983), and decisions about future
consumption episodes (e.g., Gilbert, Gill, & Wilson, 2002; Pham, 1998).
Much of the early evidence about this phenomenon was reviewed 20 years ago by Gardner
(1985). Two main types of developments have emerged since. Substantive-oriented investigations
have focused how various aspects of the marketplace can trigger incidental a ect that in uences
evaluations in a congruent fashion. A number of studies have shown that incidental a ect elicited
by the media context of an ad (TV program or magazine) generally has a congruent in uence on
consumers’ evaluations of the ad but less in uence on their evaluations of the advertised brand
(e.g., Gardner & Wilhelm, 1987; Goldberg & Gorn, 1987; Mathur & Chattopadhyay, 1991; Murry &
Dacin, 1996; Yi, 1990); but see Kamins, Marks, & Skinner, 1991 for di erent ndings). Gi wrap-
ping can also enhance the recipient’s evaluation of the gi by elevating the recipient’s mood (How-
ard, 1992). Even the mere action of browsing a series of attractive options may elevate a consumer’s
mood and result in assimilative e ects on subsequent evaluations (Raghunathan & Irwin, 2001).
Incidental a ect can “transfer” to the target object very rapidly. Morales and Fitzsimons (forth-
coming) recently found that a mere physical contact between a new, factory-sealed pack of sanitary
napkins (a product most consumers nd disgusting) and another factory-sealed product is suf-
cient to decrease consumers’ intention to try the latter product even though there is no chance of
real contamination by the former. An important question is whether this type of spontaneous inci-
dental a ect transfer is limited to disgust—as Morales and Fitzsimons (forthcoming) propose—or
whether it applies to other types of incidental a ect as well (e.g., pride, anger, joy).
More theoretically-oriented investigations have focused on clarifying the process (or processes)
that underlie the phenomenon and identifying its boundary conditions. e best-known explana-
tion for the congruency e ects of incidental a ective states attributes the phenomenon to a di er-
ential accessibility of valenced materials in memory under positive versus negative a ective states
(Isen, Shalker, Clark, & Karp, 1978). Consistent with other research on the e ects and representa-
tion of mood states in memory (Bower, 1981), Isen and her colleagues (1978) proposed that positive
moods enhance target evaluations by making positive thoughts about the target object more acces-
sible in memory. Consistent with this explanation, they found that participants who were put in a
good mood through false success feedback had better memory for pleasant words studied earlier
than did participants who were put in a bad mood through false failure feedback. Interestingly,
however, there was no di erence between the two groups in their memory for unpleasant words,
suggesting that the proposed explanation may not equally account for negative mood congruency
e ects.
Building on the idea that a ect is o en seen as having information value—an idea known as
the “a ect-as-information” hypothesis, Schwarz and Clore (1983) proposed a di erent explana-
Haugtvedt_ER5603X_C011.indd 316Haugtvedt_ER5603X_C011.indd 316 7/10/2007 4:47:41 PM7/10/2007 4:47:41 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 317
tion. ey argued that positive and negative a ective states have congruent e ects on evaluations
because, in the course of evaluating objects, people are o en inclined to inspect how they feel
about these objects—a process known as the “how-do-I-feel-about-it?” heuristic (Schwarz & Clore,
1988; see also Pham, 1998). When relying on this heuristic, people may not always realize that the
feelings they experience while evaluating an object may have been a ected by preexisting a ec-
tive states, resulting in assimilation e ects. Consistent with this interpretation, Schwarz and Clore
(1983) found that respondents, who were in a good mood as a result of being interviewed on a
sunny day, reported higher levels of life satisfaction than respondents who were in a bad mood as a
result of being interviewed on a rainy day. Importantly, this mood congruency e ect disappeared
when respondents were made aware of the actual source of their a ective state (i.e., the weather).
is contingency—known as the representativeness principle (Pham, 1998; Strack, 1992)—supports
the interpretation that incidental feelings in uence evaluations only to the extent that they are
perceived to provide information about the target. In a related study in a consumer context (Gorn,
Goldberg, & Basu, 1993), respondents were asked to evaluate a pair of speakers through which
music was being played. Incidental a ect was manipulated by varying the music being played to be
either pleasant or unpleasant. Awareness of the actual source of the feelings was manipulated by
asking respondent to rate the music either before they evaluated the speakers (high awareness) or
a er they had evaluated the speakers (low awareness). As predicted, respondents in the low aware-
ness condition evaluated the speakers more favorably when the music was pleasant than when it
was unpleasant—a classic a ect congruency e ect. In contrast, respondents in the high awareness
condition evaluated the speakers equally whether the music was pleasant or unpleasant—consis-
tent with the idea that the congruency e ect was driven by the perceived informativeness of the
music-induced feelings.
According to the integrative “a ect-infusion” model (Forgas, 1995), the two accounts—di er-
ential memory accessibility and mood-as-information—are not inconsistent; instead, they oper-
ate under di erent conditions. e di erential memory accessibility account is more likely when
the person engages in extensive substantive processing such as full and open search in memory,
which presumes a certain level of involvement. e mood-as-information account is more likely
when the person engages in heuristic processing, which is more common under conditions of low
involvement. Consistent with this proposition, it has been found that positive moods have two
separate e ects in persuasion (Batra & Stayman, 1990; Petty, Schumann, Richman, & Strathman,
1993). Under conditions favoring low elaboration, positive moods have direct assimilative e ects
on attitudes—a nding consistent with a mood-as-information process. Under conditions favoring
high elaboration, the assimilative e ects of positive moods on attitudes appear to be mediated by
the valence of target-related thoughts during message exposure—a nding consistent with a dif-
ferential memory accessibility explanation.
Within the mood-as-information account, there is still uncertainty as to the exact locus of the
process. In conventional interpretations of the process, feelings are assumed to enter judgments
during the formal evaluation stage (Schwarz & Clore, 1988, 1996). However, recent ndings suggest
that the informative in uence of incidental feelings may take place earlier in the process, during an
initial appraisal of the object (Yeung & Wyer, 2004).
Moderators of Congruency E ects of Incidental A ective States
Factors that increase the reliance to integral feelings in evaluations generally also increase the
in uence of incidental a ect on evaluations. Incidental moods are generally found to have stron-
ger assimilative in uences on evaluations when motivation to process information is low (e.g.,
Haugtvedt_ER5603X_C011.indd 317Haugtvedt_ER5603X_C011.indd 317 7/10/2007 4:47:41 PM7/10/2007 4:47:41 PM
318 JOEL B. COHE N, MICHEL TUAN PHAM, AND EDUAR DO B. ANDRADE
Miniard, Bhatla, Lord, Dickson, & Unnava, 1991; Petty, Schumann, Richman, & Strathman, 1993)
and when processing resources are constrained such as under distraction (e.g., Albarracin & Wyer,
2001) or under time pressure (Siemer & Reisenzein, 1998). However, recent studies suggest that
higher levels of motivation and ability to process information do not necessarily result in a mono-
tonic decrease of the in uence of incidental mood on evaluations. e relationship may instead be
curvilinear, with stronger in uence of incidental mood under moderate levels of motivation and
ability (Albarracin & Kumkale, 2003). is is because when motivation and ability to process infor-
mation are very high, people are likely to recognize that their feelings are incidental and therefore
irrelevant for the judgment at hand. Incidental feelings would therefore be discounted. However,
when people’s motivation and ability to process information is very low, they may fail to even
notice their incidental feelings. Incidental feelings would therefore not have any in uence. ere
is also consistent evidence that incidental a ective states have stronger a ect-congruent in uences
on evaluations when other bases of evaluation are ambiguous (e.g., Bakamitsos, 2006; Gorn, Pham,
& Sin, 2001; e.g., Isen & Shalker, 1982; Miniard, Bhatla, & Sirdeshmukh, 1992), or when people lack
expertise with the target domain (Ottati & Isbell, 1996; Srull, 1987).
e a ect-as-information hypothesis implies two additional moderators. Consistent with the
principle of representativeness mentioned earlier, incidental a ective states tend to be more in u-
ential when their actual source is not salient (Gorn, Goldberg, & Basu, 1993; Raghunathan, Pham,
& Corfman, 2006; Schwarz & Clore, 1983; Siemer & Reisenzein, 1998). is is because, when the
actual source of the a ective state is salient, people recognize that the a ective state is unrelated to
(i.e., not representative of) the target and therefore noninformative. Further evidence of a represen-
tativeness interpretation of this contingency comes from the nding that, even when their actual
source is salient, incidental a ective states may still in uence objectively unrelated judgments and
decisions provided that there is a super cial domain similarity between the judgment or decision
and the salient origin of the a ective state (Raghunathan, Pham, & Corfman, 2006). Inferences of
representativeness, we know, are very sensitive to surface similarity (Gilovich, 1981).
Also consistent with the logic of a ect-as-information is the ndings that, even when people
do not recognize that their feelings are truly incidental—that is, even when they assume that their
feelings are representative—they do not seem to use them unless they perceive their feelings to be
a relevant basis for judgment—a contingency known as the relevance principle (Pham, 1998). For
example, people are more in uenced by their mood when making decisions guided by experien-
tial motives, such as assessing a movie for an evening out, than when making decisions guided by
instrumental motives, e.g., assessing the same movie as material for a school project (Pham, 1998).
As a result, they are typically more in uenced by their moods in decisions involving hedonic prod-
ucts than in decisions involving utilitarian products (Adaval, 2001; Yeung & Wyer, 2004). Mood
states have also been found to have greater in uence on global judgments of well-being than on
more speci c judgments such as satisfaction with one’s work or housing (Schwarz, Strack, Kom-
mer, & Wagner, 1987). In addition, incidental a ect ive state s have been found to be more in uential
when people make decisions for themselves as opposed to for someone else (Raghunathan & Pham,
1999). Avnet and Pham (2004) recently suggested and found evidence that the reliance on feelings
in judgments and decisions may involve a meta-cognitive stage in which people assess whether they
should use their feelings in a given judgment or decision. is meta-cognitive assessment appears
to require signi cant cognitive resources; when resources are insu cient, incidental and integral
feelings in uence judgments and decisions more indiscriminately. e dual-process, meta-cogni-
tive model proposed by Avnet and Pham (2004) appears to account for a wide variety of ndings
about the moderators of incidental and integral a ect in judgments and decisions.
Haugtvedt_ER5603X_C011.indd 318Haugtvedt_ER5603X_C011.indd 318 7/10/2007 4:47:42 PM7/10/2007 4:47:42 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 319
According to Forgas’s (1995) a ect-infusion model, incidental a ective states have greater
assimilative in uences on evaluations when the person engages in constructive processing. Evalua-
tions based on nonconstructive processes, such as the retrieval of prior attitudes, are less amenable
to a ect infusion. For example, Fedorikhin and Cole (2004) found that respondents exposed to a
mood-inducing video prior to being exposed to a commercial were more in uenced by their mood
in their evaluation of the advertised product than were respondents exposed to the mood-induc-
ing video a er viewing the same commercial and forming an initial evaluation of the product.
is is presumably because respondents in the mood-before condition had to form their product
evaluations “from scratch,” which required constructive processing, whereas respondents in the
mood-a er condition could simply retrieve their previously-formed initial evaluations. More gen-
erally, Srull (1987) showed, in a very compelling series of studies, that mood congruent evaluation
e ects generally require that the incidental mood be experienced at the same time the evaluation
is constructed.
Beyond Simple Valence Congruency
Recent ndings suggest that incidental mood states do not always in uence evaluations in a mood-
congruent fashion (i.e., higher evaluations under positive moods than under negative moods).
Sometimes, incidental mood states interact with the target object to produce evaluations that are
con gural rather than mood-congruent. Martin, Aben, Sedikides, and Green (1997) observed that,
when asked to evaluate a story that was meant to be happy, participants in a happy mood reported
more favorable evaluations than participants in a sad mood. However, when asked to evaluate a
story that was meant to be sad, participants in a sad mood reported more favorable evaluations
than participants in a happy mood. is nding suggests that people do not literally interpret
the valence of their feelings as meaning “goodness” versus “badness.” Instead, they interpret the
valence of their feelings in light of salient goals and judgment criteria (Pham, 2004). If the criterion
is “Is this a good happy story?”, feelings of happiness will mean “Yes” and feelings of sadness will
mean “No.” On the other hand, if the criterion is “Is this a good sad story?”, feelings of happiness
will mean “No” and feelings of sadness will mean “Yes.” Similar con gural e ects were obtained
by Kamins, Marks, and Skinner (1991) who observed that a happy commercial was evaluated more
favorably and elicited stronger purchase intentions when presented a er a happy TV program than
when presented a er a sad TV program. In contrast, a sad commercial was evaluated more favor-
ably when presented a er a sad program than when presented a er a happy program. Somewhat
related to the idea of con gurality, some ndings suggest that mood states increase the weight
attached to information that is evaluatively consistent with the mood in product evaluations (Ada-
val, 2001). According to Adaval (2001), when there is a match between the consumer’s a ective state
and the valence of a piece of information, this information “feels right” and is therefore relied upon
with greater con dence.
A growing body of research suggests that incidental a ective states also in uence target evalu-
ations when the valence of the a ective state is held constant. Two main types of ndings have
been uncovered in this regard. e rst set of ndings pertain to the e ects of incidental emotional
arousal. e second set of ndings pertain to the di erential in uence of qualitatively distinct
emotional states.
A large number of studies show that, holding valence constant, the arousal component of inci-
dental a ective states tends to polarize the evaluation of objects. Evaluations of target objects are
usually more extreme under incidental states of high arousal than under incidental states of low
Haugtvedt_ER5603X_C011.indd 319Haugtvedt_ER5603X_C011.indd 319 7/10/2007 4:47:42 PM7/10/2007 4:47:42 PM
320 JOEL B . COHEN, MICHEL TUAN PHAM, AND EDUARDO B. ANDR ADE
arousal (Dutton & Aron, 1974; Foster, Witcher, Campbell, & Green, 1998; Gorn, Pham, & Sin, 2001;
Mattes & Cantor, 1982; White, Fishbein, & Rutsein, 1981; Zillmann, 1971). For example, individu-
als aroused by contextual factors such as a roller-coaster ride or crossing a high suspension bridge
have been found to be more attracted to good-looking individuals of the opposite sex and less
attracted to individuals of the same sex (Dienstbier, 1979; Dutton & Aron, 1974; White, Fishbein,
& Rutsein, 1981). Similarly, Gorn, Pham, and Sin (2001) observed that consumers who had recently
listened to an arousing piece of music (in a supposedly unrelated study) reported more polarized
evaluations of an advertisement than consumers who had listened to a less arousing piece of music.
Compared to consumers in the low arousal conditions, consumers in the high arousal condition
repor ted more favorable ad evaluations when the ad ’s a ective tone was positive and more unfavor-
able evaluations when the ad’s a ective tone was negative. erefore, incidental states of arousal
seem to amplify people’s inherent a ective and evaluative responses to a target.
ree main explanations have been o ered for this phenomenon. According to the cognitive-
complexity hypothesis (Paulhus & Lim, 1994), under high arousal, people’s representations of tar-
get objects become simpler. As a result, the evaluative dimension becomes relatively more salient,
causing more polarized judgments. is explanation is consistent with the nding that high arousal
induces a selective reliance on information perceived to be more diagnostic in judgment (Pham,
1996). A second explanation is that states of high arousal encourage dominant responses in judg-
ment and behavior (Zajonc, 1965). Because evaluative responses tend to be dominant, high arousal
induces polarization (e.g., Allen, Kenrick, Linder, & McCall, 1989). A third explanation, consistent
with the a ect-as-information hypothesis, is that individuals misattribute their incidental arousal
as indicating the strength of their integral a ective response to the target, which again results in
more polarized evaluations (Pham, 2004). Overall, it seems that all three explanations are viable
and may be di erentially operative in di erent situations. For instance, Gorn, Pham, and Sin (2001)
suggest that the cognitive-complexity may be more valid under conditions of very strong arousal,
whereas the misattribution explanation may be more applicable under conditions of milder arousal.
is is because strong arousal states are known to narrow people’s focus of attention (Easterbrook,
1959) but are unlikely to be misattributed, whereas milder arousal states may not narrow people’s
attention but are more likely to be misattributed.
More recent demonstrations of the importance of studying incidental a ective states beyond
their valence come from the growing body research on the di erential e ects of distinct emo-
tional states (Lerner & Keltner, 2000, 2001; Lerner, Small, & Loewenstein, 2004; Raghunathan &
Pham, 1999; Raghunathan, Pham, & Corfman, 2006; Tiedens & Linton, 2001). In one of the earliest
demonstrations, Raghunathan and Pham (1999) found that, in choices between high-risk/high-
reward and low-risk/low-reward options, sad individuals consistently favor the former, whereas
anxious individuals consistently favor the latter. is is presumably because, even though their
states are incidental, sad individuals tend to infer that they have lost something of value (a typi-
cal cause of sadness), which activates a goal of reward acquisition that shi s preferences toward
high-reward options. In contrast, anxious individuals tend to infer that the situation is uncertain
and beyond control (typical causes of anxiety), which activates a goal of risk avoidance that shi s
preferences toward low-risk options. Similarly, Lerner, Small, and Loewenstein (2004) found that
incidental states of sadness reverse the classic endowment e ect, that is, the tendency to place a
higher value on objects that are already in our possession compared to identical objects that not in
our possession. In contrast, incidental states of disgust eliminate the endowment e ect. According
to the authors, this is because sadness creates a motivation to change the current situation, which
increases the willingness to pay for objects that are not in our possession (higher purchase prices)
and also increases the willingness to sell objects that currently are (lower selling prices). In con-
Haugtvedt_ER5603X_C011.indd 320Haugtvedt_ER5603X_C011.indd 320 7/10/2007 4:47:42 PM7/10/2007 4:47:42 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 321
trast, disgust triggers an impulse to get rid of objects that are currently in our possession (lower
selling prices) without necessarily distorting the value of objects that are not in our possession
(unchanged purchase prices). Tiedens and Linton (2001) also observed that respondents made pre-
dictions with greater con dence when under states of disgust or happiness than when under states
of fear or hopefulness. is is presumably because both disgust and happiness typically arise in
situations appraised as certain (e.g., witnessing something repulsive or receiving very good news),
whereas fear and hope typically arise in situations appraised as uncertain (e.g., going up for ten-
ure). ese ndings are generally consistent with an a ect-as-information process in which the
information conveyed by a ective states comes from the typical appraisal content of these a ective
states (Pham, 2004). Further evidence that these e ects are driven by an a ect-as-information pro-
cess comes from the nding that they tend to disappear when the true source of the a ective state
is made salient (Raghunathan et al. 2006, Experiment 1), unless there is a surface domain resem-
blance between the source of the incidental a ective state and the target decision (Raghunathan et
al. 2006, Experiment 2).
E ects of Incidental A ect on Judgment and Decision Processes
In addition to shaping the content of consumer’s judgments and decisions, incidental a ective
states can also in uence the process through which these judgments and decisions are made. e
nature of these in uences seems to depend on (a) the intensity of the a ective state, (b) the valence
of the a ective state, and (c) the appraisal content of the emotional state.
Because high arousal is known to impair working memory capacity (S. Darke, 1988a; Hum-
phreys & Revelle, 1984), it is generally believed and o en observed that states of high emotional
arousal interfere with people’s ability to reason and make judgments and decisions. For example,
compared to nonanxious participants, anxious participants have been observed to (a) take longer
to verify logical inferences (S. Darke, 1988b), (b) scan alternatives in a more haphazard fashion
and select options without considering every alternative (Keinan, 1987), (c) commit more errors in
analogical problems (Keinan, 1987; Leon & Revelle, 1985), and (d) process persuasion arguments
less thoroughly (Sanbonmatsu & Kardes, 1988; but see Pham, 1996 for a di erent interpretation)
It should be noted, however, that most of these ndings pertain to the e ects of high anxiety. It is
therefore not clear whether they generalize to other types of emotional arousal such as intense joy,
anger, or pride. It should also be noted that states of intense emotional arousal can bene t judg-
ment processes in at least one respect. Compared to states of lower arousal, states of high emotional
arousal seem to increase the relative reliance on diagnostic versus nondiagnostic information in
persuasion and judgment in general (Pham, 1996). is explains the nding that incidental states
of high anxiety do not decrease message elaboration when the message is related to the source of
anxiety (Sengupta & Johar, 2001).
Interestingly, intense emotional states characterized by low arousal also appear to interfere with
people’s judgment abilities. A number of studies indicate that chronic depression interferes with
people’s reasoning and ability to engage in e ortful processing in judgment (Conway & Giannopou-
los, 1993; Hartlage, Alloy, Vazquez, & Dyk man, 1993; Silberman, Weingartner, & Post, 1983). Over-
all, it appears that intense emotional states, regardless of their associated level of arousal, generally
interfere with people’s reasoning and judgment processes. One exception, however, relates to the
ability to prioritize diagnostic versus nondiagnostic information, which seems to increase under
states of high emotional arousal (Pham, 1996).
Milder incidental a ective states, such as moods, can also in uence people’s judgmental pro-
cesses. Compared to individuals in a neutral mood, individuals in a good mood have been found to
Haugtvedt_ER5603X_C011.indd 321Haugtvedt_ER5603X_C011.indd 321 7/10/2007 4:47:42 PM7/10/2007 4:47:42 PM
322 JOEL B . COHEN, MICHEL TUAN PHAM, AND EDUARDO B. ANDR ADE
(a) categorize objects more broadly (Isen & Daubman, 1984; Isen, Niedenthal, & Cantor, 1992), (b)
generate more creative answers in response-generation tasks (Greene & Noice, 1988; Hirt, Melton,
McDonald, & Harackiewicz, 1996), (c) perform better in problem-solving tasks that require inge-
nuity (Greene & Noice, 1988; Isen, Daubman, & Nowicki, 1987), and (d) solve a multi-attribute
decision problem more e ciently (Isen & Means, 1983). ese and other ndings have been inter-
preted as showing that positive moods have generally bene cial e ects on reasoning, problem solv-
ing, judgment, and decision making (Isen, 2001).
However, a number of studies suggest that positive moods lead to poorer reasoning performance
in a variety of respects. Positive mood individuals are more likely to commit the “fundamental
attribution error” of overestimating the degree to which others’ behaviors are driven by their per-
sonal disposition as opposed to by situational factors (Forgas, 1998). Positive mood participants
have also been found to perform more poorly in deductive reasoning tasks (Oaksford, Morris,
Grainger, & Williams, 1996) and exhibit more intransitive preferences (Fiedler, 1988). Numerous
studies also indicate that positive moods generally decrease the depth with which people process
substantive information in persuasion and attitude formation (Batra & Stayman, 1990; Bless, Boh-
ner, Schwarz, & Strack, 1990; Bless, Mackie, & Schwarz, 1992; Mackie & Worth, 1989; Worth &
Mackie, 1987). Positive mood individuals seem to rely instead on global knowledge structures
and internal cues including scripts (Bless, Schwarz, Clore, Golisano, & Rabe, 1996), stereotypes
(Bodenhausen, Kramer, & Suesser, 1994), and judgmental heuristics such as the ease of retrieval
(Ruder & Bless, 2003). Overall, positive moods seem to have mixed e ects on people’s reasoning
and judgment processes. On the one hand, they seem to promote greater exibility and creativity in
problem solving; on the other hand, they seem to promote a more top-down, less data-driven, and
less thorough mode of processing. Recent studies by Adaval (2003) suggest that one consequence
of this more top-down form of processing is that positive moods polarize the evaluative impact
of judgment inputs involving preexisting knowledge structures (e.g., schemata, stereotypes, cat-
egories, brands, etc.). Adaval’s (2003) studies show that, under positive mood, brand names (and
their associated knowledge structures) have greater in uence on product evaluations than under
negative mood. Speci cally, products from popular brands are evaluated more favorably under
positive mood than under negative mood, whereas products from unpopular brands are evaluated
less favorably under positive mood than under negative mood.
Negative moods, especially those of sad nature, have generally been found to have e ects that
mirror those described above. Compared to neutral and pleasant moods, sad moods have been
found to increase the care with which people process substantive information in persuasion (Bless,
Bohner, Schwarz, & Strack, 1990; Sinclair, Mark, & Clore, 1994), decrease the reliance on general
knowledge structures such as scripts and stereotypes (Bless, Schwarz, Clore, Golisano, & Rabe,
1996; Bodenhausen, Kramer, & Suesser, 1994), increase the ability to estimate covariation from
scatter plot data (Sinclair & Mark, 1995), reduce the susceptibility to halo e ects (Sinclair, 1988),
reduce fundamental attribution errors (Forgas, 1998), and increase the transitivity of preferences
(Fiedler, 1988). Overall, temporary sad moods seem to trigger a more systematic, data-driven, and
analytical form of reasoning. One possible explanation, also based on the a ect-as-information
hypothesis, is that sad moods signal to the individual that the situation is problematic and therefore
requires a more vigilant form of processing, whereas good moods signal that the situation is benign
and allows a more nonchalant form of processing (Schwarz, 2002).
It should be noted, however, that not all negative moods trigger a vigilant form of processing.
States of anger or disgust seem to decrease the depth of processing and increase the reliance on
stereotyping and other heuristic cues, apparently because these states trigger a sense of certainty
(Bodenhausen, Sheppard, & Kramer, 1994; Tiedens & Linton, 2001).
Haugtvedt_ER5603X_C011.indd 322Haugtvedt_ER5603X_C011.indd 322 7/10/2007 4:47:43 PM7/10/2007 4:47:43 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 323
It should also be noted that negative moods do not always increase task e ort and positive
moods always decrease it. Martin, Ward, Achee, and Wyer (1993) asked respondents in a positive
or negative mood to perform various tasks under one of two sets of instructions. One group was
asked to keep working until they were satis ed with their performance. e other group was asked
to keep working until they no longer enjoyed the task. When instructed to keep working until
they were satis ed with their performance, respondents in a negative mood worked longer than
those in a positive mood (a result consistent with the typical nding that negative mood leads to
more careful processing compared to positive mood). However, when instructed to keep working
until they no longer enjoyed the task, the e ect reversed: respondents in a negative mood stopped
sooner than those in a positive mood. Apparently, when the instruction was to keep working until
satis ed with the performance, a negative mood was construed as dissatisfaction with one’s e ort,
producing greater perseverance, whereas a positive mood was construed as satisfaction with one’s
e ort, triggering an early stop. In contrast, when the instruction was to keep working until the
task was no longer enjoyed, a negative mood was construed as the task being not fun, producing an
early stop, whereas a positive mood was construed as the task being fun, producing perseverance.
is interpretation, known as the “mood-as-input” hypothesis (Martin, Ward, Achee, & Wyer,
1993), illustrates a broader principle about the informative role of feelings in judgments. e same
feelings can have very di erent interpretations depending on the question that people are asking
themselves (Pham, 2004). e information value of the feelings lies not so much in the feelings
themselves as in the interaction between these feelings and the questions that people are trying to
answer when consulting their feelings, which depends on situational demands and more generally
on the person’s currently active goals.
E ects of Mood States on Risk-Taking
A number of studies indicate that, compared to neutral moods, positive moods promote risk-tak-
ing when the stakes and chances of loss are low but risk-avoidance when the stakes and chances of
loss are high (Arkes, Herren, & Isen, 1988; Dunegan, Duchon, & Barton, 1992; Isen & Geva, 1987;
Isen, Nygren, & Ashby, 1988; Kahn & Isen, 1993; Nygren, Isen, Taylor, & Dulin, 1996). For example,
Arkes, Herren, and Isen (1988) observed that, compared to control participants, participants who
received a small gi were willing to pay more for lottery tickets, especially when the prize level and
probability of winning were high—indicating greater risk-seeking under positive mood in situa-
tions with only upsides. However, participants who received a small gi were also more willing to
pay more to insure against a variety of losses, especially when potential losses were high—indi-
cating greater risk-aversion under positive mood in situations characterized by downsides. Kahn
and Isen (1993) observed a similar pattern in the e ects of positive mood on variety-seeking. ey
found that, compared to control participants, participants who were in a good mood sampled a
greater variety of products such as crackers, soups, and chips, unless the choice set included items
expected to taste poorly. Again, positive mood appeared to promote risk-seeking in benign settings
and risk-avoidance in settings involving potential risk. According to Isen and her colleagues, when
the decisions entail low risks and stakes, positive mood individuals tend to have more optimistic
(mood-congruent) expectations about the outcomes and, therefore, take greater risks compared to
neutral mood individuals. However, when the stakes are high and the potential for losses signi -
cant, positive mood individuals become risk-averse because they want to maintain their positive
a ective state, which a loss would disrupt (e.g., Isen, Nygren, & Ashby, 1988; Kahn & Isen, 1993) as
will be discussed further in the section on a ect regulation.
Haugtvedt_ER5603X_C011.indd 323Haugtvedt_ER5603X_C011.indd 323 7/10/2007 4:47:43 PM7/10/2007 4:47:43 PM
324 JOEL B . COHEN, MICHEL TUAN PHAM, AND EDUARDO B. ANDR ADE
e e ects of negative a ective states on risk-risking are not as clear. Several studies indicate that
negative emotional states accompanied by strong arousal increase risk-seeking (Fessler, Pillsworth,
& Flamson, 2004; Leith & Baumeister, 1996; Mano, 1992, 1994). For example, Leith and Baumeister
(1996) found that angry participants and participants anticipating impending embarrassment were
mor e like ly to choo se an e conomical ly inferior “long- shot ” gamble with a low pr obability of obtain-
ing a larger amount of money and a high complementary probability of enduring some stressful
noise, over a superior “safe-bet” gamble with a higher probability of obtaining a smaller amount
of money with a low probability of enduring the stressful noise. Sad participants, however, did not
exhibit this bias. Fessler, Pillsworth, and Flamson (2004) also found that anger triggered more risk-
seeking in gambling, especially among men. Similarly, Mano (1994) found that intense emotional
arousal increased people’s willingness to pay for lotteries and decreased their willingness to pay for
insurance; that is, increased risk-taking for both potential gains and potential losses.
However, other ndings indicate that people’s attitude toward risk under negative a ective
states is not just a function of the level of arousal associated with the a ective state, but also a func-
tion of the appraisal content of the a ective state (Lerner & Keltner, 2001; Raghunathan & Pham,
1999). For example, as mentioned previously, in choices between low-risk/low-reward and high-
risk/high-reward options, anxious individuals tend to prefer the former, whereas sad individuals
tend to prefer the latter (Raghunathan & Pham, 1999; Raghunathan, Pham, & Corfman, 2006). e
consistent risk-aversion exhibited by anxious individuals in these studies seems inconsistent with
Leith and Baumeister’s (1996) proposition that high-arousal emotion lead to risk-seeking. In fact,
other studies show that, when the level of arousal is held constant, anxiety reduces risk-seeking
(Mano, 1992, 1994). According to Raghunathan and Pham (1999), this is because anxiety, which is
typically associated with situations of low control and high uncertainty, activates a goal of risk and
uncertainty minimization, whereas sadness, which is typically experienced in response to the loss
of a source of reward, activates a goal of reward maximization. Similarly, Lerner and Keltner (2001)
observed that, even though fear and anger are both high-arousal negative emotions, fear tends to
trigger risk-aversion, whereas anger tends to trigger risk-seeking. is is apparently because fear
is typically associated with situations of uncertainty and low control, whereas anger is typically
associated with situations of certainty and high control. It has also been found that disgust, another
high arousal emotion, decreases risk-seeking in gambling among women (Fessler, Pillsworth, &
Flamson, 2004).
In summary, it appears that intense negative emotions do not have a uniformly positive e ect on
risk-seeking. High emotional arousal seems neither necessary, nor su cient to explain risk-seek-
ing under negative emotions. e e ects of negative emotions on risk-seeking appear to depend not
only on the level of arousal associated with the emotional state, but also on complex interactions
between the goals activated by the emotional state and the nature of the risks to be taken. is
may partly explain why a meta-analysis of published studies relating chronic states of anger, sad-
ness, and anxiety to risky sexual behavior (Crepaz & Marks, 2001) found virtually no correlation
(r = .05).5
TASKRELATED AFFECT IN JUDGMENT AND DECISION MAKING
e process of making judgments or decisions may itself induce feelings and emotions. We call
this type of a ect task-related a ect. A common task-related a ect is the unpleasant feelings that
consumers o en experience when they have to trade o important attributes whose values are
negatively correlated across choice options. Consider a choice between prospective apartments.
Apartment A is much roomier, but is located in a bad neighborhood; whereas Apartment B is
Haugtvedt_ER5603X_C011.indd 324Haugtvedt_ER5603X_C011.indd 324 7/10/2007 4:47:43 PM7/10/2007 4:47:43 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 325
much smaller, but located in a great neighborhood. It has been found that the aversive emotional
experience of having to make trade-o s across important attributes o en lead consumers to prefer
avoidant options such as choosing the status quo or deferring the choice (Luce, 1998). is avoidant
behavior seems to be a deliberate attempt by the consumer to mitigate the unpleasant feelings by
eschewing the trade-o s altogether. is avoidant behavior is attenuated under situations of cog-
nitive load, which reduce the aversiveness of the trade-o s by making them less apparent (Drolet
& Luce, 2004). Another way in which consumers attempt to deal with the emotional aversive-
ness of di cult decisions is to resort to simpler, heuristic decision strategies such as processing by
attribute-by-attribute (as opposed to alternative-by-alternative) and invoking dominance relations
(Luce, 1998; Luce, Bettman, & Payne, 1997). It has also been found that, when trade-o s are emo-
tional, consumers tend to place greater weight on the relative quality of the options, such as the
relative safety of two cars, rather than on their relative resource requirements, e.g., the relative price
of the cars (Luce, Payne, & Bettman, 1999).
Task-related a ect, in the form of stress, can also be induced by giving decision makers some
time pressure and the impression of being monitored during the decision (Stone & Kadous, 1997).
Unlike the manipulations based on trade-o di culty mentioned above, this method presents the
advantage of holding information about the options constant. Stone and Kadous (1997) found that,
under task-related stress, decision makers tend to use a “scanning” strategy of quickly but indis-
criminately examining available information, which may increase choice accuracy in easy choice
environments but decrease choice accuracy in di cult choice environments.
Another common type of task-related a ect in decision making is the unpleasant feeling of
having to forego attractive options. In a choice among a Mercedes-Benz S550, a BWM 750i, and an
Audi A8, for example, choosing the Mercedes-Benz also means forfeiting the BMW and the Audi,
which may be emotionally stressful. Dhar and Wertenbroch (2000) found that the emotional dis-
comfort of forgoing an option is greater when the option is primarily hedonic, such as a fun sports
car, than when the option is primarily utilitarian, such as a functional minivan. It has also been
found that the more consumers deliberate about their choices, the more they become emotionally
attached to the options, which leads to decision-related discomfort once one option has been cho-
sen (Carmon, Wertenbroch, & Zeelenberg, 2003).
A particularly important outcome of a task-related a ect is the transfer of that a ect onto the val-
uation of the chosen alternative. Just like incidental a ective states, task-related a ective responses
may be misconstrued as re ecting genuine integral a ective responses to one of the options—a
phenomenon again broadly consistent with the a ect-as-information framework (Schwarz, 1990;
see also Pham [2004] for a review). For example, Garbarino and Edell (1997) found that reducing
the e ort involved in selecting an alternative (i.e., making a task less unpleasant) can increase the
price respondents are willing to pay for that alternative. e transfer of task-related a ect onto the
valuation of alternatives underlies a growing body of research on the “value-from- t” hypothesis
(Higgins, 2000). According to this hypothesis, a t between the manner in which a decision is made
and the current orientation of the decision maker can produce pleasant task-related feelings of
“being right,” which can then be (mis)attributed to a chosen object, enhancing its perceived value
(Avnet & Higgins, 2003, 2006; e.g., Higgins, Idson, Freitas, Spiegel, & Molden, 2003). is nding
also illustrates the close connection between a ect and motivation, discussed next.
AFFECT AND MOTIVATION
e study of a ect and the study of motivation have traditionally been interrelated (Young, 1961).
Motivation and emotion share the same Latin root, movere, which means “to move.” A ective
Haugtvedt_ER5603X_C011.indd 325Haugtvedt_ER5603X_C011.indd 325 7/10/2007 4:47:43 PM7/10/2007 4:47:43 PM
326 JOEL B . COHEN, MICHEL TUAN PHAM, AND EDUARDO B. ANDR ADE
states are said to stimulate action tendencies (Elster, 1999; Lazarus, 1991), action readiness (Frijda,
1986; Lang, 1995), and goal shi s (Oatley, 1992; Simon, 1967). Some researchers argue that a ec-
tive states function as part of a superordinate program that directs motivational priorities and goal
choice (Cosmides & Tooby, 2000) and motivates individuals to pursue speci c goals. e interac-
tion between motivation and a ect is bidirectional: (1) a ective states set goals; and (2) goal pursuit
(achievement, blockage, progress rate, etc.) triggers speci c a ective reactions (Carver, Lawrence,
& Scheier, 1996). Although both directions of in uence are of substantial theoretical interest, in
this chapter we focus on the former because a larger number of studies have examined the a ect to
motivation direction.
In general, a ective motivation relies on a ect’s informational role. As previously described,
Schwarz and Clore’s (1983) seminal work represents the cornerstone of the informational tradi-
tion associated with experienced a ective states. When asked to assess their life satisfaction, many
respondents in this study misattributed their weather-induced feelings to the unrelated judgment
at hand. Usually overlooked in the literature is the fact that the study also asked individuals to
assess their “desire to change.” When asked this question on a rainy (versus sunny) day, individ-
uals reported a stronger desire—hence, a stronger motivation—to change. Schwarz and Clore’s
(1983) ndings thus demonstrate not only the informational role of a ect, but also its motivational
consequences.
STRENGTH OF THE SIGNAL, APPROACH, AND AVOIDANCE
When a ect deviates from a homeostatic range, this signals that something has altered the actual
or anticipated state of the environment, thus increasing the likelihood of unanticipated negative
or positive consequences, such as those related to threats or safety. e strength of this signal is
a direct function of how much a ect deviates from its normal homeostatic range. However, the
relation is not symmetric. Cacioppo, Gardner, and Burnsten (1999) suggest that when the a ective
signal is at zero (in terms of motivational input), there is a weakly positive approach tendency (in
terms of motivational output)—a phenomenon, the authors call the “positivity o set.” In other
words, at very low levels of a ective activation, the motivation to approach is stronger than the
motivation to avoid. Cacioppo, Gardner, and Burnsten (1999) further suggest that the evolutionary
function of the positivity o set is to increase the organism’s tendency to approach novel objects and
stimuli in neutral environments. In the absence of such a motivation to explore, organisms would
learn little about novel or neutral-appearing environments and their potential reward value.
However, there is also a well documented negativity bias in human behavior (Baumeister, Brat-
slavsky, Finkenauer, & Vohs, 2001), which cannot be fully explained by the greater diagnosticity
(e.g., reduced frequency and greater information value) of negative information (Skowronski &
Carlston, 1989). Ito, Cacioppo, and Lang (1998) analyzed a ective responses to hundreds of slides
from the International A ective Picture System (Lang, Bradley, & Cuthbert, 1997). In one analysis,
subjective ratings of positivity of the mostly pleasant slides, and subjective ratings of negativity
of the mostly unpleasant slides, were modeled as a function of the level of arousal elicited by each
slide. Not surprisingly, both relationships were positive: pleasant slides were rated more positively
as they arousal levels increased; and unpleasant slides were rated more negatively as they arousal
levels increased. More importantly, ratings of positivity had a higher intercept than ratings of nega-
tivity. When arousal was very low, pleasant slides were rated more positively than unpleasant slides
were rated negatively—a nding consistent with a positivity o set. On the other hand, the slope
of the negativity ratings was steeper than that of the positivity ratings—a nding consistent with
the negativity bias. erefore, at low levels of activation, positive stimuli have a greater impact
Haugtvedt_ER5603X_C011.indd 326Haugtvedt_ER5603X_C011.indd 326 7/10/2007 4:47:44 PM7/10/2007 4:47:44 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 327
on a ective responses (and presumably behavior) than negative stimuli do. However, at higher
levels of activation, negative stimuli have greater impact than positive stimuli do. (See Crites, Fab-
rigar, & Petty, 1994; Taylor, 1991; Wojciszke, Brycz, & Borkenau, 1993 for supportive ndings.) A
recent study by Dijksterhuis and Aarts (2003) provides additional support for these propositions.
Respondents were exposed subliminally to both positive and negative words and asked to guess
whether the words were positive or negative. Performance was better than chance for negative
words, but not for positive words. is nding suggests a lower threshold for unconscious negative
a ect to become accessible. According to Cacioppo, Gardner, and Bernsten (1999), the positiv-
ity o set fosters useful exploratory behavior. However, because exploration can also have aversive
consequences, and it is more di cult to reverse adverse consequences, natural selection may have
favored a capacity to respond more strongly to, and withdraw from, proximate negative stimuli.
is duality of function may have contributed to distinguishable approach and avoidance mecha-
nisms mobilizing the individual toward immediate action.
A similar analysis by Lang and his colleagues (Lang, Bradley, & Cuthbert, 1990, 1992) supports
the distinction between an aversive/defensive/withdrawal system and an appetitive/approach sys-
tem, each with distinct patterns of neural activity. e greater the appetitive input, the stronger
the activation of positivity and approach forces; the greater the aversive input, the stronger the
activation of negativity and avoidance forces. Cacioppo & Berntson’s (1994) evaluative space for-
mulation a llow s for a co-act ivat ion of both sy stems, resu lting i n either directional action (approach
or avoidance), indi erence (low activation of both positivity and negativity), or ambivalence (high
activation of both positivity and negativity).
However, physical behavior is facilitated by integrating the output of these dual processes and
resolving any co-activation of opposing forces to express the dominant tendency and inhibit
the weaker tendency. is is evident in postural support reactions, balance, and dynamic motor
adjustments and implies a neurological substrate for central bivariate control (i.e., coactivity) of
exors and extensors despite the fact that outcomes are physically constrained into bipolar molar
responses. Co-activation is also consistent with ndings indicating that central mechanisms for
reward and aversion can be independently manipulated, since this implies a fundamental dissocia-
bility of related brain systems. Whether this separability can help explain why initiatives designed
to produce positive a ective responses, such as to minorities and organ donation, do not neces-
sarily overcome negative biases is an important research issue (see Sarason et al., 1993; see Scho-
eld, 1991). One important implication of co-activation is that an increase in the intensity of either
positive or negative valence can transform inaction into action, perhaps leading a person to take
risks that had restrained behavior, as when fear of unsafe sex or cigarette addiction are overcome
by either added momentary attractiveness or the perception of reduced likelihood or severity of
consequences (Bolton, Cohen, & Bloom, 2006).
Broadly speaking, then, a ective signals direct attention to both environmental and personal
factors (particular current actions) that seem likely to alter consequences. is also fosters, or miti-
gates, energy and resource expenditure for both mental and behavioral activity. is “e ort to
perform” response is one of the accounts used to explain why people may devote more time or
higher levels of thought in scrutinizing the information available (versus relying on heuristics)
when experiencing a negative (versus positive) a ective state (Schwarz, 2002).
While goal achievement and harm avoidance are particularly responsive to a ective signals,
hedonism (the emphasis on feeling good) can be a default goal. For example, psychological theories
have traditionally stressed the dynamic tension between task success and accuracy motivations, on
the one hand, and ego-bolstering and feeling good on the other (e.g., Lazarus & Folkman, 1984).
Of particular interest in research on a ect is the notion that a currently negative a ective state can
Haugtvedt_ER5603X_C011.indd 327Haugtvedt_ER5603X_C011.indd 327 7/10/2007 4:47:45 PM7/10/2007 4:47:45 PM
328 JOEL B . COHEN, MICHEL TUAN PHAM, AND EDUARDO B. ANDR ADE
motivate individuals to pursue a short-term hedonistic objective, whereas a positive a ective state
can motivate individuals to protect the current given state. People are not necessarily spending
more e ort in order to perform well, but are selectively choosing the stimuli that will regulate their
current a ective states.
AFFECT REGULATION
In the last two decades, a ect regulation has received special attention in the literature (Forgas,
Johnson, & Ciarrochi, 1998; Gross, 1998; Larsen, 2000; Zillmann, 1988a, 1988b). For our pur-
poses, a ect regulation corresponds to people’s spontaneous (conscious or unconscious) attempt
to intensify, attenuate, or maintain a given a ective state, usually in the short-term. It incorporates
related constructs, such as mood regulation (Erber & Erber, 2001; Larsen, 2000), negative state
relief (Cialdini, Darby, & Vincent, 1973), mood management (Forgas, Johnson, & Ciarrochi, 1998;
Wegener & Petty, 1994; Zillmann, 1988b), mood maintenance (Clark & Isen, 1982; Isen, 1984),
emotion regulation (Gross, 1998), and coping (Lazarus, 1991). Although biological drives such as
hunger and thirst (Buck, 1999) could also be incorporated within the a ect regulation umbrella
(Gross, 1998), they are beyond our scope of analysis.
As a basic psychological mechanism, Andrade (2005) proposes that a ect regulation rests on
three principles: dynamic a ect, conditional hedonism, and a ective signaling (see also Cohen
& Andrade, 2004). Dynamic a ect represents individuals’ projected discrepancy between feelings
at two points in time, that is, what they feel now and what they could feel in the future as a result
of the cognitive or behavioral activity. is gap captures the motivational property of a ect in
guiding information processing, judgment, and decision making. Coupled with a basic hedonistic
assumption, when no contingencies are available in the environment, a ect regulation predicts
that people in negative a ective states will the most likely to engage in cognitive or behavioral
activities in anticipation of the mood-li ing consequences of such enterprises, whereas people in
a positive mood will be the most likely to avoid thoughts and actions in anticipation of the mood-
threatening consequences associated with them. In short, as a result of a dynamic analysis, people
are likely to move toward the goal of a more positive a ective state when they feel bad, as well as to
protect a current a ective state when they feel good. At the core of the dynamic a ect principle is
the idea that people’s intuitive theories about the a ective changing properties of the forward-look-
ing cognitions or behaviors are critical determinants of the impact of a ect regulation. For a ect
regulation to guide responses, people must intuitively believe that the forthcoming thoughts and/
or actions will regulate a current state upward or downward (e.g., Manucia, Baumann, & Cialdini,
1984; e.g., Tice, Bratslavsky, & Baumeister, 2001).
Although individuals are predisposed to improve a negative a ective state and/or protect a cur-
rent positive a ective state, there are circumstances in which internal or environmental contingen-
cies convince them to follow a di erent route (e.g., Erber, Wegner, & erriault, 1996). Conditional
hedonism, therefore, implies that both upward and downward a ect regulation, and negative and
positive mood maintenance, represent potential a ect regulation strategies, depending on compet-
ing goals available in the environment. If a performance goal overcomes a short-term hedonistic
goal, the former may be preferred (e.g., Cohen & Andrade, 2004).
Finally, stronger, more accessible a ective signals lead to clearer assessment of the discrepancy
between current and expected a ective states and the appropriateness of the actual state. As polar-
ized a ective states produce stronger signals compared to more neutral feelings, a ect regulation
should lead to stronger impacts when people experience positive or negative a ect versus neutrality
(e.g., Cohen & Andrade, 2004; Wegener & Petty, 1994). If one’s current a ective state is not acti-
Haugtvedt_ER5603X_C011.indd 328Haugtvedt_ER5603X_C011.indd 328 7/10/2007 4:47:45 PM7/10/2007 4:47:45 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 329
vated, a discrepancy is less likely to be identi ed, minimizing the impact of a ect regulation on
thoughts and behavior. Not only accessibility, but also diagnosticity should play a role. Consistent
with the a ect-as-information hypothesis (Schwarz & Clore 1983; Pham 1998), the impact of a ect
regulation can be corrected if the diagnosticity of the current a ective state is called into question.
When a sad consumer realizes that he is suddenly buying cookies in a supermarket in an attempt
to regulate his/her current negative feelings, misattribution correction may mitigate the impact of
a ect regulation on impulse buying, unless the cookies seem somewhat related to the cause of sad-
ness (Raghunathan, Pham, & Corfman, 2006). However, at least two other aspects must be consid-
ered. First, a normative assessment of the a ect-behavior relationship is required. One must believe
that the ongoing/forthcoming cognitive or behavioral activity that results from the transient feel-
ing state is inappropriate (e.g., “I shouldn’t be buying that much just because I feel sad today”). is
may not be always the case. For instance, many consumers report going shopping in a deliberate
attempt at a ect regulation (Babin, Darden, & Gri n, 1994; Mick & Demoss, 1990). Trying to
regulate current negative feelings through a shopping experience may sometimes be perceived as
an appropriate and e ective reaction. In that case, to highlight negative a ect as a potential cause
for the behavior may intensify rather than mitigate the impact of a ect regulation on behavior.
Second, the consumer must have the skills to stop the ongoing action or avoid the forthcoming
action. Behavioral reactions driven by current a ective states are sometimes much more di cult
to control than one would expect, even when the person consciously knows that his/her behavior is
inappropriate (e.g., “I shouldn’t be doing this, but I can’t control myself”; Loewenstein, 1996). e
impact of appropriateness and skills on the correction of a ect regulation attempts remain an open
question in the literature.
A ect Regulation E ects on Information Processing
A ective states have been shown to produce changes in attention, recall, and processing style via
a ect regulation. Most research has so far been conducted in the psycholog y literature (for a review,
see Gross, 1998), although consumer researchers have demonstrated growing interest in the topic
(e.g., Keller, Lipkus, & Rimer, 2003; Meloy, 2000).
Attention. Standard information processing theory assumes limited cognitive resources
and therefore selective processing. us, at the perceptual level, decisions have to be made as to
what pieces of information one attends to. Incidental (dispositional or situational) and integral
a ect have both been shown to in uence such perceptual decision processes via a ect regulation
(Krohne, 2003). First, one can simply avoid the threatening stimulus when there is no reason to
pay attention to it. For instance, MacLeod and Mathews (1988) observed that states of high anxiety
(as assessed 1 week before a exam) shi s attention toward exam-relevant threat words, suggesting
vigilance, whereas, states of low anxiety (as assessed 12 weeks before the exam) shi s attention
away from exam-relevant threat words, suggesting avoidance. Second, recent evidence suggests
that negative stimuli may not necessarily capture, but also hold, people’s attention (Fox, Russo,
Bowles, & Dutton, 2001; Putman, Hermans, & van Honk, 2004). us, those who display stronger
upward a ect regulation tendencies should demonstrate faster disengagement skills. Mather and
Carstensen (2003) adopted such a rationale to provide further evidence that upward a ect regula-
tion skills improve with age. A er being presented with pairs of negative and neutral faces, older
adults responded faster to a subsequent dot probe that appeared on the opposite side of the negative
(versus neutral) face. is attentional bias did not emerge among younger adults. Finally, it has
been suggested that a ect regulation directs people’s attention not only away from negativity, but
also toward relieving cues (Derryberry & Tucker, 1994).
Haugtvedt_ER5603X_C011.indd 329Haugtvedt_ER5603X_C011.indd 329 7/10/2007 4:47:45 PM7/10/2007 4:47:45 PM
330 JOEL B. COHEN, MICHEL TUAN PH AM, AND EDUARDO B. ANDR ADE
Memory. Feelings not only in uence what individuals encode but also how much they store and
retrieve (Blaney, 1986). Initial evidence supported mood congruent recall. However, as data accu-
mulated, the phenomenon has been shown to be more robust on the positive side of the a ective
spectrum—i.e., people in a positive mood were more likely to recall positive events than those in
a negative mood were to recall negative events (Bower & Forgas, 2000). Isen (1984) suggested that
a ect regulation could explain this asymmetry. Negative feelings can sometimes encourage the
recall of positive information because people are naturally motivated to feel better. Considerable
evidence supports Isen’s initial proposition (Erber & Erber, 1994; McFarland & Buehler, 1997,1998;
Parrott & Sabini, 1990; Rusting & DeHart, 2000; Sedikides, 1994; Smith & Petty, 1995). However,
boundary conditions do exist. e impact of upward a ect regulation is weaker among chronically
sad individuals. Josephson and colleagues (1996) found that whereas a negative a ect manipu-
lation increased mood-incongruent recall among nondysphoric individuals, the e ect dissipated
among chronically dysphoric participants. Moreover, further evidence showed that when explicitly
asked to recall more positive autobiographical memories, nondysphoric and dysphoric participants
were equally capable of accomplishing the task. However, only nondysphoric participants felt bet-
ter a erwards (Joormann & Siemer, 2004). ese results suggest that recall of positive memories
is reduced among dysphoric individuals, not because of cognitive inability but probably because
such memories are not be perceived as an a ect regulation opportunity within this group. Accord-
ing to the dynamic a ect assumption described above, if the cognitive or behavioral activity is not
perceived as mood-li ing, or mood-threatening, a ect regulation will be less likely to in uence
information processing.
Whereas the impact of mood incongruent recall has been more consistently explored within the
negative a ect realm, there has been some evidence suggesting that positive a ect can also trigger
a ect regulation and biased recall. Parrott and Sabini (1990) showed in natural (Studies 1 and 2)
and laboratory (Studies 3 and 4) settings that not only did subjects in a negative mood recall more
positive events, but that subjects in a positive mood were also more likely to recall negative events.
e authors do not make strong claims as to the reason for mood incongruent recall but speculate
that people may be engaging in spontaneous, and probably unconscious, downward mood regula-
tion. According to the conditional hedonism assumption described above, positive mood attenu-
ation is possible. However, this usually happens when competing goals—including the need for
accuracy and objectivity—are made available. In this case, individuals have a reason to give up
their short-term happiness in exchange for better task performance and subsequent pleasantness
(e.g., Cohen & Andrade, 2004; Erber, Wegner, & erriault, 1996).
A ect regulation motivates people to protect positive a ect sometimes at the expense of per-
formance. However, when the task at hand produces harmful e ects in the long run, performance
becomes a major concern (a stronger competing goal) and positive a ect can serve as a psycho-
logical bu er to help individuals cope with the negativity in the environment. at is the ratio-
nale behind Trope and colleagues’ mood-as-resource hypothesis and ndings (Raghunathan &
Trope, 2002; Trope & Neter, 1994; Trope & Pomerantz, 1998). ey showed that when informa-
tion is highly self-relevant, those in a positive, hence bolstered, mood were more likely to select
and/or recall negative information. For example, heavy consumers of ca eine in a positive (versus
negative) mood recalled more pieces of negative information, as compared to positive information,
about ca eine consumption. However, a positive mood did not enhance the recall of the negative
information for light consumers of ca eine. Although it is not clear whether this e ect is mediated
solely by motivation (i.e., goal shi from happiness in the short-term to happiness in the long-term)
or also by skill (i.e., higher cognitive capacity to deal with information), the mood-as-a-resource
Haugtvedt_ER5603X_C011.indd 330Haugtvedt_ER5603X_C011.indd 330 7/10/2007 4:47:45 PM7/10/2007 4:47:45 PM
THE NATUR E AND ROEL OF AFFEC T IN CONSUMER BEHAVIOR 331
hypothesis can, under certain circumstances, account for mood incongruent recall by people expe-
riencing positive a ect.
Processing Style. Some authors have suggested that the motivational in uence of a ect may not
only represent a stronger or weaker desire to process the information per se, but also corresponds
to a deliberate attempt at a ect regulation (Ambady & Gray, 2002; Forgas, 1998). People experi-
encing negative a ect may be pursuing more systematic and e ortful processing in an attempt to
improve their negative a ective state (Clark & Isen, 1982; Forgas, 1991a, 1991b; Sedikides, 1994),
whereas those in a good mood may spend less e ort in an attempt to protect their current a ec-
tive state (Clark & Isen, 1982; Isen, 1984). According to the main implication of the dynamic a ect
principle, the impact of a ect regulation on the quantity and quality of information processing is
highly contingent on the expected a ective changes associated with the task (i.e., its mood-li ing
or mood-threatening characteristics). For instance, if people’s unbiased assessment of the attitude
object when in a bad mood represents an e ective mood-li ing opportunity (e.g., “I will feel hap-
pier if do not rely on stereotypes”), sad people should be less likely to rely on such heuristics. Simi-
larly, if the e ort spent processing the information represents a mood-threatening experience (e.g.,
“If I think too much about it, I will feel bad”), happy people should be more likely to rely on heuris-
tics. Although plausible, one of the major challenges in this line of research is to isolate the impact
of the a ect regulation motive from other mediating processes, such as a ect-as-information and
the role of con dence. For instance, when a sad person processes a message more carefully than
a happy person, is it due to (1) a basic inner signal (“ reatening environment, be careful!”), (2) a
lower level of con dence about his/her level of accuracy (“I’m not yet ready to make up my mind.”),
and/or (3) an a ect regulation strategy (“If I accomplish this task accurately, I will feel much bet-
ter!”)? Moreover, these accounts may overlap, which makes claims about a ect regulation e ects
on information processing even more challenging. For a ect regulation to operate and bias thought
and behavior, individuals must perceive the usually short-term, mood-li ing or mood-threatening
cues associated with the upcoming activity. In the information processing world, people have to
recognize that the e ort or performance consequences will regulate their a ective state upwards
when they feel bad; whereas it will attenuate their a ective state when they feel good. Further
research is needed to tackle this issue.
A ect Regulation in Judgment and Decision Making
e impact of a ect regulation on decision making has drawn increasing attention. Consumers
o en take the a ective consequences of decisions (buy vs. not buy; shop vs. not shop; buy product
A vs. B) into account. In the past 25 years, the results converge with the a ect regulation principles
described above.
Negative a ect and a ect regulation. An overview of di erent research streams shows that peo-
ple experiencing negative a ect are more willing to make behavioral choices that will lead to more
positive feelings. us, they will engage in a wide variety of behavior in an attempt to change the
current feelings. Documented regulation behaviors include watching comedies (Weaver & Laird,
1995; Zillmann, 1988a, 1988b), listening to upli ing music (Cohen & Andrade, 2004; Knobloch &
Zillmann, 2002), eating (Grunberg & Straub, 1992; Tice, Bratslavsky, & Baumeister, 2001), exercis-
ing (Hsiao & ayer, 1998), acting in an aggressive way toward others (Bushman, Baumeister, &
Phillips, 2001), reading upli ing news (R. Erber, Wegner, & erriault, 1996), purchasing gi s for
themselves (Luomala & Laaksonen, 1997; Mick & Demoss, 1990), helping others (Bagozzi & Moore,
1994; Cialdini, Darby, & Vincent, 1973), taking greater risks for greater rewards (Raghunathan &
Haugtvedt_ER5603X_C011.indd 331Haugtvedt_ER5603X_C011.indd 331 7/10/2007 4:47:46 PM7/10/2007 4:47:46 PM
332 JOEL B . COHEN, MICHEL TUAN PHAM, AND EDUARDO B. ANDR ADE
Pham, 1999), buying impulsively (Rook & Gardner, 1993), selling unwanted items (Lerner, Small, &
Loewenstein, 2004), choosing the status-quo option (Luce, 1998), or simply procrastinating (Tice,
Bratslavsky, & Baumeister, 2001).
Under the a ect regulation umbrella, these examples all represent mood-li ing alternatives
pursued in an attempt to restore or achieve the desired a ective state (i.e., to close the gap between
current and ideal a ect). At the core of the dynamic a ect principle is the implication that a ect
regulation is contingent on one’s intuitive theories about the a ective changing properties of the
forward-looking cognitions and behavior. us, for a ect regulation to fully operate, sad people
must expect the behavioral activity to improve their current a ective state. If that expectation is
not present, the e ect disappears. Several experiments have intrinsically or extrinsically controlled
for the mood-changing properties associated with the behavioral activity. For example, helping is
usually perceived as a mood-li ing opportunity because most people have learned the self-satisfy-
ing, and therefore upli ing, bene ts associated with altruistic acts. is explains why, when expe-
riencing negative a ect, teenagers and adults are more willing to help compared to children who
have yet to learn these associations (Cialdini & Kenrick, 1976). Food items, such as chocolate, have
also been shown to present di erent mood-li ing expectations. A ect regulation via chocolate has
a greater in uence on women who are more likely to perceive chocolate as a mood-li ing oppor-