You are on page 1of 186

file:///C|/JPSP/index.

txt

Journal of Personality and Social Psychology


Volume: 94, Issue: 5
May, 2008

Table of Contents:

1) Hedonic Tone and Activation Level in the Mood-Creativity Link: Toward a Dual Pathway to Creativity Model
De Dreu, C.K.W.; Baas, M.; Nijstad, B.A. pp. 739-756
(184 KB)

2) Judgments of the Lucky Across Development and Culture


Olson, K.R.; Dunham, Y.; Dweck, C.S.; Spelke, E.S.; Banaji, M.R. pp. 757-776
(348 KB)

3) How to Heat Up From the Cold: Examining the Preconditions for (Unconscious) Mood Effects
Ruys, K.I.; Stapel, D.A. pp. 777-791
(129 KB)

4) Forming Implicit and Explicit Attitudes Toward Individuals: Social Group Association Cues
McConnell, A.R.; Rydell, R.J.; Strain, L.M.; Mackie, D.M. pp. 792-807
(308 KB)

5) Maintaining Sexual Desire in Intimate Relationships: The Importance of Approach Goals


Impett, E.A.; Strachman, A.; Finkel, E.J.; Gable, S.L. pp. 808-823
(298 KB)

6) Receiving Support as a Mixed Blessing: Evidence for Dual Effects of Support on Psychological Outcomes
Gleason, M.E.J.; Iida, M.; Shrout, P.E.; Bolger, N. pp. 824-838
(255 KB)

7) Nomina Sunt Omina: On the Inductive Potential of Nouns and Adjectives in Person Perception
Carnaghi, A.; Maass, A.; Gresta, S.; Bianchi, M.; Cadinu, M.; Arcuri, L. pp. 839-859
(173 KB)

8) Taking the Easy Way Out: Preference Diversity, Decision Strategies, and Decision Refusal in Groups
Nijstad, B.A.; Kaps, S.C. pp. 860-870
(102 KB)

9) Distinguishing Between Silent and Vocal Minorities: Not All Deviants Feel Marginal
Morrison, K.R.; Miller, D.T. pp. 871-882
(113 KB)

10) Making Choices Impairs Subsequent Self-Control: A Limited-Resource Account of Decision Making, Self-Regulation, and Active Initiative
Vohs, K.D.; Baumeister, R.F.; Schmeichel, B.J.; Twenge, J.M.; Nelson, N.M.; Tice, D.M. pp. 883-898
(138 KB)

11) Adolescent Personality Moderates Genetic and Environmental Influences on Relationships With Parents
South, S.C.; Krueger, R.F.; Johnson, W.; Iacono, W.G. pp. 899-912
(138 KB)

12) Societal Threat, Authoritarianism, Conservatism, and U.S. State Death Penalty Sentencing (1977-2004)
McCann, S.J.H. pp. 913-923
(269 KB)

file:///C|/JPSP/index.txt [25/05/2008 5:49:12 PM]


ATTITUDES AND SOCIAL COGNITION

Hedonic Tone and Activation Level in the MoodCreativity Link:


Toward a Dual Pathway to Creativity Model
Carsten K. W. De Dreu, Matthijs Baas, and Bernard A. Nijstad
University of Amsterdam

To understand when and why mood states influence creativity, the authors developed and tested a dual
pathway to creativity model; creative fluency (number of ideas or insights) and originality (novelty) are
functions of cognitive flexibility, persistence, or some combination thereof. Invoking work on arousal,
psychophysiological processes, and working memory capacity, the authors argue that activating moods
(e.g., angry, fearful, happy, elated) lead to more creative fluency and originality than do deactivating
moods (e.g., sad, depressed, relaxed, serene). Furthermore, activating moods influence creative fluency
and originality because of enhanced cognitive flexibility when tone is positive and because of enhanced
persistence when tone is negative. Four studies with different mood manipulations and operationaliza-
tions of creativity (e.g., brainstorming, category inclusion tasks, gestalt completion tests) support the
model.

Keywords: mood, creativity, cognitive flexibility, emotions, arousal

What enables scientists to make notable contributions, engineers mood stands out as one of the most widely studied and least
to develop innovative products, and work teams to creatively solve disputed predictors (e.g., George & Brief, 1996; Isen & Baron,
their problems? What hinders stand-up comedians from being 1991; Mumford, 2003). For example, Ashby, Isen, and Turken
funny and refrains poets from being original? When are people (1999) noted that
creative, and why? What hinders creativity, and when? Partly
It is now well recognized that positive affect leads to greater cognitive
because of the importance of creativity for human progress and
flexibility and facilitates creative problem solving across a broad
adaptation, these questions are as old as the human sciences
range of settings. These effects have been noted not only with college
(Simonton, 2003). Apart from its obvious, problem-solving func- samples but also in organizational settings, in consumer contexts, in
tion (Mumford & Gustafson, 1988), creative ideation allows indi- negotiation situations . . . and in the literature on coping and stress. (p.
viduals to remain flexible (Flach, 1990), giving them the capacity 530)
to cope with the advantages, opportunities, technologies, and
changes that are a part of their day-to-day lives (Runco, 2004). In a similar vein, Lyubomirksy, King, and Diener (2005) con-
Accordingly, creativity is studied in a variety of disciplines, in- cluded that people in a positive mood are more likely to have
cluding psychology, organizational behavior, and communication richer associations within existing knowledge structures, and thus
sciences. are likely to be more flexible and original. Those in a good mood
Creativity is usually defined as the generation of ideas, insights, will excel either when the task is complex and past learning can be
or problem solutions that are new and meant to be useful (Amabile, used in a heuristic way to more efficiently solve the task or when
1983; Paulus & Nijstad, 2003; Sternberg & Lubart, 1999). Among creativity and flexibility are required. (p. 840)
the many variables that have been shown to predict creativity, Although many studies support the idea that positive mood
states trigger more creative responses than do neutral mood control
conditions, studies in which positive and negative mood states
were compared appear to be less conclusive: There is also a large
Carsten K. W. De Dreu, Matthijs Baas, and Bernard A. Nijstad, De- literature on negative affect, which indicates that the impact of
partment of Psychology, University of Amsterdam, Amsterdam, the Neth- negative affect is more complex and difficult to predict than is the
erlands. case for positive affect (Ashby et al., 1999, p. 532). Indeed,
We thank Joyce Jacobs for help in coding the data of Study 4 and
whereas some studies suggest that positive mood states trigger
Gerben van Kleef, Mark Rotteveel, and Richard Ridderinkhof for com-
ments and suggestions.
more creativity than do negative mood states (e.g., Grawitch,
Correspondence concerning this article should be addressed to Carsten Munz, & Kramer, 2003; Hirt, Levine, McDonald, Melton, &
K. W. De Dreu, Department of Psychology, University of Amsterdam, Martin, 1997; Hirt, Melton, McDonald, & Harackiewicz, 1996),
Roetersstraat 15, 1018 WB Amsterdam, the Netherlands. E-mail: other studies report similar levels of creativity (Bartolic, Basso,
c.k.w.dedreu@uva.nl Schefft, Glauser, & Titanic Schefft, 1999), and still other studies
Journal of Personality and Social Psychology, 2008, Vol. 94, No. 5, 739 756
Copyright 2008 by the American Psychological Association 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.739

739
740 DE DREU, BAAS, AND NIJSTAD

report that negative moods promote creative performance more Sujan, Hirt, & Sujan, 1990; also see Isen & Daubman, 1984;
than do positive or neutral moods (e.g., Carlsson, 2002; Gasper, Mikulincer, Paz, & Kedem, 1990; Rietzschel, De Dreu, &
2003; Kaufmann & Vosburg, 1997; Madjar & Oldham, 2002). Nijstad, 2007). It is important to note that besides being a
This has led some to call into question the general conclusion that measure of creative performance, flexibility also refers to a
positive mood states produce more creativity than do negative cognitive process. Many researchers have argued that in order
mood states (Shalley, Zhou, & Oldham, 2004) or that negative to be creative (i.e., produce novel and appropriate products)
mood states undermine creative performance (Gasper, 2003; people must think flexibly, must break set (e.g., Duncker, 1945;
George & Zhou, 2007). Smith & Blankenship, 1991; Smith, Ward, & Schumacher,
In this article, we reconsider the link between mood and cre- 1993), and need flat associative hierarchies (e.g., Eysenck,
ativity and try to reconcile the seemingly contradictory findings 1993; Mednick, 1962; Simonton, 1999) to arrive at uncommon
and conclusions reviewed above. First, we argue that creativity can and disparate (and thus original) associations. Cognitive flexi-
be a function of cognitive flexibility and of cognitive perseverance bility can thus not only be seen as a measure of creativity but
and persistence. Second, we argue that mood states can be con- also as a precursor of the production of many (fluency) and
ceptualized in terms of two underlying dimensions hedonic tone original responses.
(positive vs. negative) and activation (activating vs. deactivating). However, in addition to cognitive flexibility, it is also possible
Whereas past work on mood states and creativity has predomi- to achieve creative fluency and originality through hard work,
nantly focused on hedonic tone dimension and on cognitive flex- perseverance, and more or less deliberate, persistent, and in-depth
ibility, we argue that the extent to which mood states activate or exploration of a few cognitive categories or perspectives (Boden,
deactivate and the tendency toward cognitive perseverance and 1998; Dietrich, 2004; Finke, 1996; Schooler, Ohlsson & Brooks,
persistence need to be taken into account also. More specifically, 1993; Simonton, 1997). Perseverance will manifest itself not in the
we propose that cognitive activation is a necessary precondition use of many or broad cognitive categories but rather in the gen-
for creativity to come about and that hedonic tone determines the eration of many ideas within a few categories or in longer time-
route the flexibility route or the perseverance routethrough on-task. All other things being equal, generating many ideas in a
which creative fluency and originality is achieved. In four studies, few categories will also lead to more ideas overall (i.e., fluency;
we tested (aspects of) the general idea that activating moods with
Nijstad et al., 2002). Furthermore, recent work suggests that flu-
positive tone are linked to cognitive flexibility and thereby pro-
ency within categories is associated with originality of ideas within
mote creative performance, whereas the creativity enhancing ef-
these categories: Because only a limited number of conventional
fects of activating moods with negative tone are due to persever-
and unoriginal ideas are possible in each category, perseverance
ance.
within categories eventually leads to original ideas (Rietzschel,
Nijstad, & Stroebe, 2007). Such within-category fluency (e.g.,
A Dual Pathway to Creative Performance Nijstad & Stroebe, 2006; Nijstad et al., 2002; Nijstad, Stroebe, &
Lodewijkx, 2003) can be illustrated with the example of an indi-
Creativity researchers often operationalize creativity with mea-
vidual who generates ideas as to how to improve health. This
sures of fluency, originality, and flexibility (Guilford, 1967; Tor-
person may think about physical exercise and sport and may start
rance, 1966). Because we will also use these measures in our
studies, it is important to conceptually relate them to each other as out with common ideas like, people should spend more time
well as to the general concept of creativity. Fluency is a measure doing physical exercise. However, provided he or she continues
of creative production and refers to the number of nonredundant generating ideas within this category, he or she might proceed to
ideas, insights, problem solutions, or products that are being gen- more unusual ideas within that category, like, putting a strong
erated. Originality is one of the defining characteristics of creativ- string in your computer keyboard to make typing very hard work.
ity and refers to the uncommonness or infrequency of the ideas, In previous work in which both flexibility (number of used cate-
insights, problem solutions, or products that are being generated gories) and within-category fluency were established, no system-
(Amabile, 1983; Guilford, 1967; Paulus & Nijstad, 2003; Stern- atic correlation between the two was observed (Nijstad et al., 2002,
berg & Lubart, 1999; Torrance, 1966). Fluency and originality 2003).
may be correlated (e.g., quantity breeds quality; Diehl & Stroebe, Taken together, creativity can be achieved through enhanced
1987; Osborn, 1953), but they need not be. For example, creative cognitive flexibility, set-breaking, and cognitive restructuring,
fluency may manifest itself in a relatively large number of solved which manifests itself in the use of many, broad, and inclusive
insight or perception problems, with the solutions themselves not cognitive categories. It can equally well be achieved through
being particularly new or uncommon (cf. Forster, Friedman, & enhanced persistence and perseverance, which manifests itself in a
Liberman, 2004). Moreover, states or traits that influence creative higher number of ideas and insights within a relatively low number
fluency do not necessarily also influence originality and vice versa. of cognitive categories, prolonged effort, and relatively long time-
Flexibility as a measure of creativity manifests itself in the on-task. This may apply to idea generation and divergent thinking
use of different cognitive categories and perspectives and of tasks, as well as to insight tasks that are typically characterized by
broad and inclusive cognitive categories (Amabile, 1983; Med- being ultimately soluble by the average problem solver. Such
nick, 1962). Generating ideas in many different categories will, insight tasks are likely to produce an impasse and a state of high
all other things being equal, be associated with more ideas uncertainty as to how to proceed and to produce a kind of aha
overall (i.e., increased fluency; cf. Nijstad, Stroebe, & Lodewi- experience when the impasse is suddenly overcome and the solu-
jkx, 2002) as well as with the generation of ideas in categories tion is discovered after prolonged efforts at solution (Forster et al.,
that are not usually thought of (i.e., originality; cf. Murray, 2004; Schooler et al., 1993).
MOODCREATIVITY LINK REVISITED 741

Discrete Moods, Creative Fluency, and Originality major role in regulating the excitability of the cortical circuitry on
which the working memory function of the prefrontal cortex
A critical implication of the dual pathway model is that any trait depends (Dreisbach et al., 2005; Goldman-Rakic, 1996). Moderate
or state influencing cognitive flexibility or cognitive persistence levels of dopamine associate with improved working memory
and perseverance may lead to novel yet appropriate insights and performance (Floresco & Phillips, 2001; Kimberg, DEsposito, &
ideas. With regard to the influence of mood on creative fluency Farah, 1997), more efficient processing of task-relevant informa-
and originality, it may thus be that mood states influence creativity tion (Drabant et al., 2006), increased maintenance of task-relevant
to the extent that they enhance cognitive flexibility, perseverance, information (Colzato, Van Wouwe, & Hommel, 2007), and better
or both; perhaps both positive and negative mood states lead to switching between tasks (Dreisbach & Goschke, 2004). Moderate
creative fluency and originality, but through different routes. (but not extremely high) levels of noradrenalin enhance prefrontal
When thinking about mood states, valence, or hedonic tone, cortex control of behavior, including (short-term) working mem-
most readily comes to mind. Discrete moods such as anger, anx- ory (Robbins, 1984; Usher et al., 1999) and sustained selective
iety, sadness, and depression all have negative valence, or tone.
attention on task-relevant information (Chamberlain, Muller,
Discrete mood states such as happiness, elation, and feeling re-
Blackwell, Robbins, & Sahakian, 2006).
laxed and calm all have positive valence, or tone. However, in
Apart from a simple motivating effect of activation, the above
addition to hedonic tone, discrete moods differ in the extent to
indicates that activating mood states rather than deactivating mood
which they activate or deactivate (Barrett & Russell, 1998; Gray,
states come together with higher levels of dopamine and noradren-
1982; Green, Goldman, & Salovey, 1993; Posner, Russell, &
alin and greater working memory capacity. Working memory
Peterson, 2005; Thayer, 1989; Watson, Clark, & Tellegen, 1988).
capacity is often taken as a prerequisite for cognitive flexibility,
Some mood states are positive in tone and deactivating (calm,
abstract thinking, strategic planning, processing speed, access to
relaxed), whereas others are positive in tone and activating (happy,
long-term memory, and sentience (Baddeley, 2000; Damasio,
elated). Likewise, some mood states are negative in tone and
2001; Dietrich, 2004). In terms of the dual pathway model outlined
deactivating (sad, depressed), whereas others are negative in tone
and activating (angry, fearful). This applies to temporarily acti- in the previous section, it thus appears that both for the cognitive
vated and experimentally manipulated mood states (Russell & flexibility route and for the persistence route, working memory
Barrett, 1999; Watson, Wiese, Vaidya, & Tellegen, 1999), as well capacity is required and beneficial. Activating rather than deacti-
as to trait-related differences in mood (Filipowicz, 2006). For vating moods increase working memory capacity, thereby facili-
example, trait extraversion is often equated with positive affectiv- tating cognitive flexibility and restructuring, as well as more
ity (positive, activating), and trait neuroticism is equated with deliberate, analytical, and focused processing and combining of
negative affectivity (negative, activating; Cropanzano, Weiss, information. Indeed, affect intensity, measured with both negative
Hale, & Reb, 2003; Eysenck, 1993). and positive high arousing terms, relates to higher levels of cre-
ativity in children (Russ & Grossman-McKee, 1990) as well as
employees (George & Zhou, 2007).
Activation, Hedonic Tone, and Creativity Whether activating mood states produce creative fluency and
originality through enhanced cognitive flexibility or perseverance
Whether mood states are activating or deactivating may have
important effects on creative performance. According to both may depend on that mood states hedonic tone. According to the
classic and contemporary work on threat rigidity (Carnevale & cognitive tuning model (Clore, Schwarz, & Conway, 1994;
Probst, 1998; Staw, Sandelands, & Dutton, 1981) and stress-per- Schwarz & Bless, 1991), a positive affective state leads individuals
formance linkage (Broadbent, 1972; Yerkes & Dodson, 1908), an to experience their situation as safe and problem free, to feel
individuals capacity for complex thinking is altered in a curvilin- relatively unconstrained, to take risks, and to explore novel path-
ear fashion as arousal and activation increases. Low levels of ways and new possibilities in a relatively loose way, relying on
arousal lead to inactivity and avoidance, neglect of information, heuristic processing styles (Fiedler, 2000; George & Zhou, 2007;
and low cognitive and motor performance. Extremely high levels Schwarz & Clore, 1988). Positive affect facilitates primary process
of arousal reduce the capacity to perceive, process, and evaluate cognition in the right hemisphere, which is holistic and analogical
information. However, at moderate levels of arousal, individuals (Martindale & Hasenfus, 1978; Martindale, Hines, Mitchell, &
will be motivated to seek and integrate information and to consider Covello, 1984; also see Derryberry, 1989; Faust & Mashal, 2007;
multiple alternatives. Provided they are not associated with intense Fink & Neubauer, 2006). Consistent with this is a classic study on
arousal, activating moods are thus more likely than deactivating positive affect and creativity (Isen & Daubman, 1984). Participants
mood states to increase attention to and integration of information. in a state of mild happiness were asked to rate how prototypical
That activating mood states may foster creativity also follows several exemplars (e.g., bus, camel) were for a particular category
from work on the interrelations among arousal, release of specific (e.g., vehicle), with higher ratings for the weak exemplar (camel)
neurotransmitters such as dopamine and noradrenalin, and working indicating broad cognitive categories (Amabile, 1983; Eysenck,
memory capacity (cf., Ashby, Valentin, & Turken, 2002; Flaherty, 1993). Results showed that compared with the control condition,
2005; Nieuwenhuis, Aston-Jones, & Cohen, 2005; Usher, Cohen, happy participants had higher prototypicality ratings, that is, had
Servan Schreiber, Rajkowski, & Aston Jones, 1999). Working broader and more inclusive cognitive categories (also see Isen,
memory capacity refers to the ability to hold information tran- Niedenthal, & Cantor, 1992; Mikulincer & Sheffi, 2000; Murray et
siently in mind in the service of comprehension, thinking, and al., 1990). Other work showed that individuals in happy moods
planning (Baddeley, 2000). Activation and arousal associate with choose a global rather than a local visual configuration and per-
the release of dopamine and noradrenalin, which in turn play a form faster on visual insight tasks that require set-breaking
742 DE DREU, BAAS, AND NIJSTAD

Cognitive Flexibility;
Positive
Inclusiveness

Activation Motivation; Working Tone Creative Fluency


Memory Capacity and Originality

Negative Cognitive Persistence;


Perseverance

Figure 1. Schematic overview of the roles of activation and tone in the dual pathway to creativity model.

(Fredrickson & Branigan, 2005; Gasper, 2003; Wadlinger & Isaa- (Schwarz & Bless, 1991), the broaden-and-build perspective
cowitz, 2006). (Fredrickson, 1998), and the work on visual and conceptual fo-
The cognitive tuning model, and related accounts, thus posits cusing (e.g., Derryberry, 1989), it follows that activating moods
that positive affect allows individuals to be inclusive in their that are positive in tone increase creative fluency and originality
thinking, to switch cognitive categories, and to explore uncommon primarily through enhanced cognitive flexibility, whereas activat-
perspectives; positive affect, in other words, increases cognitive ing moods that are negative in tone increase creative fluency and
flexibility (cf., Ashby et al., 1999). Negative affect, in contrast, originality primarily through enhanced persistence and persever-
informs the individual that his or her situation is problematic, ance. Put differently, whereas we would not necessarily expect
threatening, and troublesome. Specific action must be taken to differences in creative fluency and originality between activating
remedy the current situation, and this calls for a more constrained, positive (e.g., happy, elated) moods and activating negative (angry,
systematic, and analytical approach (Ambady & Gray, 2002; fearful) moods, we would expect activating positive moods to
Chaiken, Liberman, & Eagly, 1989; Schwarz & Bless, 1991). associate with broader and more inclusive cognitive categories,
Negative affect enhances risk aversion and bolsters detail-oriented with greater diversity in the cognitive categories used to generate
processing. It facilitates left hemispherical, secondary process cog- ideas, and with fast completion times in creative insight tasks. Vice
nition, which is more verbal, sequential, and analytical (Martindale versa, we would expect activating negative moods to associate
& Hasenfus, 1978; Martindale, Hines, Mitchell, & Covello, 1984; with more ideas within specific cognitive categories and with
also see Derryberry, 1989; Faust & Mashal, 2007; Fink & relatively long completion times in creative insight tasks.1
Neubauer, 2006). Negative mood states such as anxiety promote Figure 1 provides a schematic overview of the way activation
narrow perceptual processing, resulting in impaired detection of and hedonic tone influence the two routes toward creative fluency
peripheral (but not central) visual information and impaired per- and originality. As can be seen, the level of activation associated
formance on secondary (but not primary) tasks; provided it does with a particular mood state serves as the critical entry point, with
not become too extreme, such narrowed processing accompanying higher activation leading to greater fluency and originality. How-
negative mood states may be adaptive in that it helps prevent ever, which pathway is used depends on a mood states hedonic
distraction while focusing attention on the most important infor- tone, with positive tone facilitating the cognitive flexibility route
mation (Derryberry & Reed (1998). Indeed, negative activating and negative tone facilitating the cognitive perseverance route.
moods such as fear and anxiety lead to narrow cognitive categories Some indirect evidence for our model is available, albeit outside
(Mikulincer et al., 1990), lowered ability to shift attention (Der- the domain of creative performance. In their review of the psy-
ryberry & Reed, 1998), and reduced cognitive flexibility (e.g., chological, neurochemical, and functional neuroanatomical medi-
Carnevale & Probst, 1998). It is important to note that negative ators of the effects of positive and negative mood on executive
activating moods also increase persistence and perseverance
(Gasper & Clore, 2002; Gray & Braver, 2002; Strauss, Hadar, 1
Activation not only varies as a function of mood but also, for example,
Shavit, & Itskowitz, 1981; but see Baumann & Kuhl, 2005). For as a function of physical exercise (see also Kaufmann & Vosburg, 1997).
example, Verhaeghen, Joormann, and Khan (2005) showed that Work on physical exercise and creativity is somewhat inconclusive, how-
rumination (persisting, conscious, and negatively valenced self- ever, with some finding no differences between exercise and baseline
related thoughts) correlated with creative fluency and originality conditions (Isen et al., 1987; Vosburg, 1998), and others finding physical
and that this relationship appeared to be due to greater seriousness exercises to lead to more divergent thinking (Blanchette, Ramocki, ODel
about and more time spent on creative activities. & Casey, 2005; Steinberg, Sykes, Moss, Lowery, & LeBoutillier, 1997).
According to our dual pathway model, creative fluency and Unfortunately, in most of these studies, no manipulation checks for
originality may be achieved through enhanced cognitive flexibil- exercise-induced arousal or activation and no controls for participants
physical condition were included, and it is unclear whether the exercise
ity, increased persistence and perseverance, or some combination
induced low, moderate, or high physical arousal. Furthermore, the tasks
thereof. On the basis of stress-performance literature and psycho- used in these experiments capitalized on cognitive flexibility (e.g., func-
physiological and neuroimaging work on arousal and working tional fixedness, remote associations), which may explain why, in a few
memory capacity, we argued that activating moods enhance cre- cases (e.g., Isen et al., 1987; Vosburg, 1998), happiness (activating positive
ative fluency and originality more than do deactivating moods. mood) produced more creativity than did exercise-induced arousal. We
From a combination of this with the cognitive tuning model return to this in the Conclusions and General Discussion section.
MOODCREATIVITY LINK REVISITED 743

functions, Mitchell and Phillips (2007) concluded that negative participants to brainstorm on ways to improve teaching at their
mood effects on executive functioning are mediated by serotonin, university. We predicted that both positive and negative activating
whereas positive mood effects may be mediated by dopamine, with moods (happy, angry) would be related to greater creative fluency
serotonin being particularly involved in effortful processes asso- and originality than would both positive and negative deactivating
ciated with goal-directed activity and dopamine being particularly moods (sad, relaxed; Hypothesis 1), that activating positive moods
involved in switching flexibly between categories and tasks (e.g., (happy) would be related to greater category diversity than would
Ashby et al., 1999). Spering, Wagener, and Funke (2005) found no any other mood state (Hypothesis 2), and that activating negative
overall differences in complex problem solving between positive moods (angry) would be related to greater within-category fluency
and negative mood states but did find that negative mood states than would any other mood state (Hypothesis 3).
produced a stronger focus on seeking and using information.
Brand, Reimer, and Opwis (2007), finally, showed that partici-
Method
pants in a negative mood solved transfer tasks less efficiently than
did those in a positive mood; negative mood participants needed Design and participants. Undergraduate students (N 58,
more repetitions to reach a mastery level but did not differ from 73% women, 27% men) at the University of Amsterdam partici-
those in a positive mood in their ultimate problem-solving ability. pated for 5 (approximately U.S. $6.50), and participants were
Thus, indeed, there is some evidence that a mood states hedonic randomly assigned to one of four different mood conditions (anger,
tone alters the processes by which individuals perform cognitive sadness, happiness, relaxation). Gender had no effects, and it is not
tasks and solve problems. discussed further. Dependent variables were self-reported activa-
tion and hedonic tone, as checks for the mood manipulation, and
The Present Study: Overview and Hypotheses creative performance during brainstorming as reflected in number
of unique ideas, originality of the ideas, number of cognitive
To test our model on creative fluency and originality as a
categories used (cf. cognitive flexibility), and within-category flu-
function of a mood states activation and tone, we conducted four
ency (cf. cognitive persistence).
studies. In the first three studies, we used self-generated imagery to
Procedure and manipulation of discrete moods. Participants
induce different mood states (cf., DeSteno, Petty, Rucker, Wege-
came to the laboratory, and they were seated in individual cubicles
ner, & Braverman, 2004; Strack, Schwarz, & Gschneidinger,
equipped with a chair, a desk, and a computer with keyboard.
1985), some of which were negative in tone (anger, fear, sadness,
Participants were told that they would be asked to participate in
depression) and some of which were positive in tone (happiness,
two different and independent studies; one was an autobiograph-
elation, calm, relaxation). Apart from a hedonic tone contrast, this
ical memory task (the task used to manipulate discrete moods) and
design allowed us to compute an activation contrast (activating
the other was a brainstorming task about possible ways to improve
moods [angry, fearful, happy, elated] versus deactivating moods
the quality of teaching in the psychology department (the task to
[sad, depressed, calm, relaxed]) that is orthogonal to the hedonic
assess creativity). Participants were then asked to write down their
tone contrast or their interaction. In Study 4, we surveyed individ-
gender and age and to write a short essay about a situation that
uals self-reported mood states negative activating, positive
happened to them and that made them feel really _____ (depend-
activating, negative deactivating, or positive deactivatingand
ing on discrete mood condition: angry, sad, happy, relaxed). They
used regression analyses to relate these mood dimensions to cre-
were given an entire page to report their situation and were asked,
ative performance. We also used, across studies, different tasks to
after finishing their autobiographical story, to report those key-
assess creative performance. In Studies 1 and 4, we engaged
words or (parts of) phrases they considered vital in making them
participants in a brainstorming task. Apart from creative fluency
feel _____ (depending on discrete mood condition: angry, sad,
and originality, from coded ideas we also derived indices of
happy, relaxed; In this experiment, and subsequent ones, the con-
cognitive flexibility (i.e., the number of cognitive categories from
tent of the stories participants wrote always adhered to instruc-
which ideas were sampled) and perseverance (i.e., the number of
tions. Furthermore, we were unable to discern systematic differ-
ideas within a particular cognitive category; cf., Rietzschel, De
ences between conditions in length of stories or particular topics
Dreu, & Nijstad, 2007). In Study 2, we focused on cognitive
participants wrote about.)
inclusiveness and breadth of cognitive categories that people use,
Upon completion of the mood manipulation task, participants
and in Study 3, we assessed performance on a Gestalt Completion
were asked to brainstorm about possible ways to improve the
Test (Ekstrom, French, Harman, & Dermen, 1976; Friedman &
quality of teaching in the psychology department. Participants
Forster, 2000; Schooler & Melcher, 1995), a classical insight
were reminded that the department attracted more and more new
problem in which participants view a series of fragmented pictures
students each year and that this put some pressure on the quality of
of familiar objects and attempt to perceptually integrate and rec-
teaching, as some of you may have already experienced. They
ognize them, to close each gestalt. According to Forster et al.
were further told that the departmental teaching staff was inter-
(2004), this task may also be seen as requiring visual insight
ested in their problem solutions and that they would be given 8 min
inasmuch as each item is ultimately soluble by the average prob-
to type in as many ideas, solutions, or suggestions as they could
lem solver and is likely to produce an impasse that may be
think of. We emphasized that idea generation would be anonymous
suddenly overcome after continued efforts at solution (p. 179).
and that no one would ever be able to link ideas to names or
student identification numbers. Hereafter, participants were asked
Study 1
to start generating ideas. They could type in an idea, and by hitting
In Study 1, we induced one of four different mood states the Enter key, they could submit this idea and receive a new
anger, sadness, happiness, and relaxationand subsequently asked opportunity to type in an idea. This procedure continued for 8 min,
744 DE DREU, BAAS, AND NIJSTAD

after which participants were informed that the brainstorming happiness) produced somewhat higher activation than deactivating
session was over, and they were asked to answer a few questions. moods (sadness, relaxation; M 3.62 vs. M 3.12), F(1, 54)
Then, they were told that the experiment was over, and they were 3.78, p .06 (marginal). Such a 2 2 ANOVA on self-reported
debriefed, paid, and dismissed. tone revealed only that positive moods (happiness, relaxation)
Dependent variables. The ideas, problem solutions, and sug- produced more positive feelings than did negative moods (anger,
gestions generated by the participants were coded and/or trans- sadness; M 2.43 vs. M 1.94), F(1, 54) 4.12, p .05. We
formed into four different components of creativity. First, inde- conclude that our manipulations were successful.
pendent coders counted the number of unique ideas generated per Creative fluency and originality. We submitted the number of
participant (Cohens K .98). This was our measure of creative unique ideas to a four level (angry, sad, happy, relaxed) one-way
fluency. To obtain a measure of originality, independent coders ANOVA. No effects were significant, but a directional test of
rated each unique idea for originality, defined as an idea or
Hypothesis 1 with planned comparisons showed that more ideas
suggestion that is infrequent, novel, and original (1 not original
were generated when participants were in an activating mood
at all to 5 very original). Interrater agreement was satisfactory
rather than in a deactivating mood, t(54) 1.65, p .05, 2 .05
following criteria as per Cicchetti & Sparrow (1981; intraclass
(see also Table 1) Hedonic tone had no effects (ts 1). We
correlation, ICC[1] .69) and we used the aggregation across
conclude that creative fluency is a function of the extent to which
raters as an indicator of originality.
To get at cognitive flexibility, we assigned each unique idea to a mood activates or deactivates (cf., Hypothesis 1).
one of the following seven categories: Ideas having to do with (a) We submitted the averaged originality of ideas to a four level
university environment, such as (architecture of) lecture halls, (angry, sad, happy, relaxed) one-way ANOVA. As predicted,
seminar rooms, and opening hours; (b) student facilities, such as mood influenced originality, F(3, 54) 3.42, p .025. A
extracurricular activities, library access, and classroom interiors; follow-up comparison showed that activating moods (happy, an-
(c) student quality, including selecting better students and increas- gry) produced more original ideas than did deactivating moods
ing cooperation and contact among students; (d) teaching materi- (sad, relaxed), t(54) 3.12, p .003, 2 .15 (for cell means,
als, such as readers, textbooks, handouts of PowerPoint presenta- see Table 1). Hedonic tone did not matter: The planned compari-
tions, examination issues, and grading systems; (e) teachers, such son of positive states (happy, relaxed) with negative states (angry,
as teacher training and selection, use of teaching evaluations, and sad) was not significant (M 2.52 vs. M 2.39), t(54) 1, ns,
use of mentors and coaches; (f) policy, such as scholarships and nor was the interaction between tone and activation, F(1, 54) 1,
other financial issues, information distribution, and reduced bu- ns. From these results, we conclude that originality of produced
reaucracy; and (g) other issues. The higher the number of catego- ideas is a function of the extent to which a mood activates or
ries used, the greater the participants cognitive flexibility (e.g., deactivates. This supports Hypothesis 1.
Nijstad et al., 2002, 2003). Interrater agreement was good (Co- Cognitive flexibility. We submitted the number of categories
hens K .71), and differences were solved through discussion. from which ideas were sampled to a four level (angry, sad, happy,
To get at perseverance, we assessed within-category fluency: the relaxed) one-way ANOVA. Means were in the predicted direction
number of unique ideas divided by the number of categories from (also see Table 1), but there were no significant effects to support
which these ideas were sampled. the hypothesis (Hypothesis 2) that cognitive flexibility is highest
To check the manipulation of hedonic tone and level of activa- among activating, positive moods.
tion, we asked participants how positive they felt (1 not positive
Persistence. A four level ANOVA on persistence showed a
at all to 5 very positive) and how activated they felt (1 not
trend for mood, F(3, 54) 2.44, p .075. Planned contrasts were
very activated to 5 very activated).
computed to examine effects of activation, effects of hedonic tone,
and their interaction on persistence. Neither the simple activation
Results contrast nor the simple hedonic tone contrast was significant,
Manipulation checks. A 2 (activating vs. deactivating) 2 t(54) 1.52, p .14, 2 .04, and t(54) 1.05, p .43, 2
(negative tone vs. positive tone) analysis of variance (ANOVA) on .01, respectively. However, the Tone Activation contrast was
self-reported activation revealed only that activating moods (anger, significant, t(54) 2.66, p .025, 2 .06, showing that anger

Table 1
Means and Standard Deviations for Fluency, Originality, Flexibility, and Perseverance as a Function of Mood (Study 1)

Mood state

Angry Sad Happy Relaxed

Variable M SD M SD M SD M SD

Creative fluency 13.32 5.11 10.16 5.67 11.88 5.32 10.42 5.27
Originality 2.73a 0.71 2.06b 0.79 2.77a 0.66 2.26b 0.74
Flexibility 3.65 1.11 3.56 1.19 4.01 1.52 3.46 1.77
Perseverance 3.64a 1.09 2.85b 0.71 2.96b 0.69 3.01b 0.90

Note. Means within a row not sharing the same subscript differ significantly at p .05.
MOODCREATIVITY LINK REVISITED 745

produced greater persistence than did the other three mood states in separate booklets. In the control condition, the self-generated
(see also Table 1). This supports Hypothesis 3. imagery task was not included, and participants immediately went
on with the study about object perception.
Discussion and Introduction to Study 2 Upon completion of the mood manipulation task, participants
handed in their booklet, and they were asked to turn to their
The results support the hypothesis (Hypothesis 1) that activating computer to continue with the next experiment about object per-
mood states produce greater creative fluency and originality than ception. First, they once again filled in their gender and age (to
do deactivating mood states, and the results support the hypothesis increase the suggestion that indeed a new and independent exper-
(Hypothesis 3) that activating negative moods produce greater iment had started), and participants were given the category inclu-
persistence than does any other mood state. One potential limita- sion task to assess their cognitive flexibility (see below). There-
tion of this support, which pertains to the persistence effect in after, they completed several manipulation checks, and
particular, is that effects are tied to one specific mood state (anger). participants were fully debriefed, paid for participation, and dis-
In the next studies, we deal with this by inducing multiple mood missed.
states that are similar in tone and activation (e.g., anger and anxiety Dependent variables. To assess cognitive inclusiveness, par-
vs. sadness and depression; elation and happiness vs. relaxation ticipants were asked to rate how prototypical exemplars were of a
and calm). Replicating support for Hypothesis 1 and 3 would particular category (1 not at all to 10 very prototypical). For
reduce the concern that effects are tied to aspects of a specific each of the four categories we used, three exemplars were pre-
mood state other than activation and tone. sented, one being strongly, one being moderately, and one being
Although means were in the predicted direction, Study 1 did not weakly prototypical (see Rosch, 1975). Specifically, the four cat-
support the hypothesis (Hypothesis 2) that activating positive egories (with strong, intermediate, and weak exemplars) were
moods produce greater cognitive flexibility than does any other vehicle (bus, airplane, camel), vegetable (carrot, potato, garlic),
mood state. Given the strong support in the literature that positive clothes (skirt, shoes, handbag), and furniture (couch, lamp, tele-
mood states foster cognitive flexibility (e.g., Ashby et al., 1999), phone). Inclusion ratings across the four categories were aggre-
the current failure may reflect a Type II error, and a conceptual gated into separate indices for strong, moderate, and weak exem-
replication is needed before concluding anything with regard to plars ( .78, .82, and .74, respectively). Cognitive flexibility
Hypothesis 2. Accordingly, in Study 2, we asked participants to usually shows up in prototypicality ratings for the weak exemplars
complete the category inclusion task previously used by Isen and more than in ratings for the moderate or strong exemplars (Isen,
Daubman (1984). We predicted greater category inclusiveness for Daubman, & Nowicki, 1987; Rosch, 1975).
activating than for deactivating moods when tone is positive (cf., Upon completion of the category inclusion task (and because of
Hypothesis 2). Given that negative tone has been related to narrow an administrative error in the mood conditions, only), we measured
perceptual focus (e.g., Derryberry, 1988; Mikulincer et al., 1990), hedonic tone by asking participants to rate their affective state on
it may be that activating moods produce reduced category inclu- three items (how do you feel: very positivevery negative; very
siveness when tone is negative. In other words, we expected pleasantvery unpleasant; very nicenot at all nice). Ratings were
greater category inclusiveness among activating moods than aggregated ( .89) and coded so that higher scores indicated
among deactivating moods when tone is positive rather than neg- more positive (and less negative) tone. In addition, we included a
ative. We included a mood-neutral control condition in which measure of activation level. Specifically, we asked participants to
participants did not do the self-generated imagery and only per- rate the following: (a) how energetic do you feel, (b) how engaged
formed the category inclusion task. This permitted us to explore are you, and (c) how active are you presently? (1 not at all to
whether (de)activation and positive (negative) tone promote (in- 5 very much). Ratings were averaged into one activation level
hibit) category inclusiveness. index ( .79).

Method Results
Design and participants. Undergraduate students (N 179, Manipulation checks. Ratings for the activation level measure
73% women, 27% men) participated for 5 (approximately U.S. were tested in two planned comparisons, one testing all four
$6.50), and they were randomly assigned to one of eight different negative mood states (anger, fear, sadness, depression) against all
mood conditions (anger, fear, sadness, depression, happiness, ela- four positive mood states (happiness, elation, relaxation, calm) and
tion, relaxation, calm) or to the mood-neutral control condition. one testing all four deactivating mood states (sadness, depression,
Gender had no effects, and it is not discussed further. Dependent relaxation, calm) against all four activating mood states (anger,
variables were self-reported activation, hedonic tone, and category fear, happiness, elation). Results were as expected: Whereas the
inclusiveness. hedonic tone contrast was not significant, t(155) 1, ns, the
Procedure and manipulation of discrete moods. Participants activation contrast was, t(155) 1.97, p .05. Participants
were seated in individual cubicles and told that they would par- reported more activation when activating moods had been induced
ticipate in two independent studies: one about autobiographical (M 4.73) than when deactivating moods had been induced (M
memory (the task used to manipulate discrete moods) and one 4.53).
about object recognition (the task used to assess cognitive flexi- Ratings for the tone measure were tested in the same two
bility). Participants were then given a booklet with instructions planned comparisons. Results were as expected: Whereas the
about the autobiographical memory study. Discrete moods were activation contrast was not significant, t(155) 1, ns, hedonic tone
manipulated as before, except that participants wrote their essays contrast was, t(155) 4.03, p .01. Participants reported more
746 DE DREU, BAAS, AND NIJSTAD

positive tone when positive moods had been induced (M 4.23) lower inclusiveness ratings, t(170) 1, ns. This supports the idea
than when negative moods had been induced (M 2.53). We that activating moods promote cognitive flexibility and inclusive-
conclude that our manipulations were successful. ness when tone is positive. However, as mentioned, because the
Cognitive flexibility. Table 2 gives the mean prototypicality of same (nonsignificant) trend emerged for negative activating moods
strong, intermediate, and weak exemplars per condition. Hypoth- versus deactivating moods, we cannot conclude that Hypothesis 2
esis 1 was tested in a planned contrast grouping all activating received support.
moods versus all deactivating moods. This contrast was not sig-
nificant for the strong and intermediate exemplars, ts(170) 1, ns, Discussion and Introduction to Study 3
but was significant for weak exemplars, t(170) 2.10, p .037,
2 .03. Prototypicality ratings for weak exemplars were higher Study 2 shows that activating moods increase category inclu-
in activating mood conditions (M 6.43) than in deactivating siveness. Together with Study 1, we thus have reasonable support
mood conditions (M 5.98). for the dual pathway model, which indicates that activating mood
Hypothesis 2 predicted that activating mood states lead to states promote creative fluency and originality more than do de-
greater inclusiveness, especially when tone is positive. A direc- activating mood states and that perseverance is higher among
tional contrast showed that positive activating moods produced activating moods that are negative in tone (cf. Study 1). Although
higher inclusiveness (M 6.51) than did all of the negative mood trends in the data suggested that cognitive flexibility was higher
states and the two deactivating positive mood states (M 6.10), among activating moods that are positive in tone (cf. Study 2),
t(170) 1.82, p .035, 2 .025. Furthermore, among the these tendencies for cognitive flexibility were fairly weakin
positive moods, the two activating moods produced greater cate- Study 1, means were as predicted but were not statistically reliable;
gory inclusiveness than did the two deactivating conditions and the in Study 2, the critical Activation Tone interaction was not
control condition, t(170) 1.98, p .05, 2 .033, whereas both significant.
of the negative activating moods did not produce greater category At present, it cannot be excluded that category diversity (Study
inclusiveness than did the two deactivating moods and the control 1) and category breadth and inclusiveness (Study 2) reflect not
condition, t(170) 1.48, p .14, 2 .016. However, these only cognitive flexibility but also persistence and perseverance.
patterns notwithstanding, the Tone Activation contrast was not Those in an activating positive mood may be cognitively flexible
significant, and Hypothesis 2 received no support; the trend for and may, therefore, include peripheral exemplars (e.g., camel) in a
negative activating moods to produce greater inclusiveness is particular category (e.g., vehicle). Those in an activating negative
weaker but is otherwise in the same direction as the trend for mood may persevere and systematically explore possibilities, ul-
positive activating moods. timately concluding that peripheral exemplars fit into a particular
Comparisons involving the mood-neutral baseline. The con- category. This possibility implies that those in an activating pos-
clusions emerging from the above analyses are further supported itive mood are faster than those in an activating negative mood,
by specific contrasts involving the mood-neutral control condition which indeed fits the results of Isen et al. (1987). In Study 2, we
(for cell means, see Table 2). First, activating moods (positive and did not track time-on-task and cannot examine this possibility. In
negatives together) produced higher inclusiveness ratings for weak Study 3, however, we included time-on-task as a key variable.
exemplars than did mood-neutral control condition (M 6.43 vs. The evidence for our model thus far pertains to cognitive and
M 5.71), t(170) 1.98, p .05, 2 .04. It is interesting to conceptual material (idea generation, cognitive category inclusive-
note that deactivating moods (positives and negatives together) did ness), and an issue is whether our dual pathway model also
not produce lower inclusiveness ratings for weak exemplars than predicts perceptual insights and creativity. Creative insight prob-
did mood-neutral control condition (M 5.98 vs. M 5.71), lems differ from the tasks used thus far in that they are soluble, are
t(170) 1, ns. likely to produce an impasse and a state of high uncertainty as to
Second, consistent with past work (e.g., Isen & Daubman, how to proceed, and are likely to produce a kind of aha expe-
1984), we found that happy participants had higher prototypicality rience when the impasse is suddenly overcome and the solution is
ratings for weak exemplars than did control participants, t(170) discovered after prolonged efforts at solution (Forster et al., 2004;
2.03, p .044, 2 .051. Positive activating moods (happy and Schooler et al., 1993). Such tasks can be solved heuristically,
elated) produced higher inclusiveness ratings than did the mood- through loose and detached processing, which is relatively effort-
neutral control condition, t(170) 1.96, p .05, 2 .031, less and fast (Brand et al., 2007). Alternatively, they can also be
whereas positive deactivating moods did not produce higher or solved through persevering and analytical probing of a series of

Table 2
Mean Prototypicality Ratings as a Function of Experimental Manipulations (Study 2)

Experimental condition

Exemplars Angry Fearful Depressed Sad Happy Elated Relaxed Calm Control

Strong 9.64 9.42 9.57 9.68 9.48 9.56 9.56 9.75 9.62
Intermediate 7.53 7.71 7.86 7.76 7.36 6.93 6.90 7.17 7.25
Weak 6.51 6.22 5.95 5.88 6.63 6.38 6.13 5.88 5.71

Note. Higher numbers indicate greater category inclusiveness.


MOODCREATIVITY LINK REVISITED 747

hypotheses. This is a relatively effortful and time-consuming pro- Results


cess.
From our dual pathway model it follows that activating moods, Manipulation checks. Ratings for the activation level measure
more than deactivating moods, lead to greater creative fluency and, were tested in two directional comparisons. The first tested all four
accordingly, that individuals in activating moods perform better on negative mood states (anger, fear, sadness, depression) against all
creative insight tasksthey close more gestalts (see below) than four positive mood states (happiness, elation, relaxation, calm),
do those in deactivating moods (cf. Hypothesis 1). Because posi- and the second tested all four deactivating mood states (sadness,
tive affective tone increases cognitive flexibility and restructuring depression, relaxation, calm) against all four activating mood
and pairs with a broader visual field, we further expected that states (anger, fear, happiness, elation). Results were as expected:
individuals in positive activating moods would be able to perform Whereas the hedonic tone contrast was not significant, t(81) 1,
creative insight tasks in relatively short time and would not benefit ns, the activation contrast showed a trend in the predicted direc-
from longer time-on-task. But, because negative affective tone tion: Participants reported more activation when activating moods
increases persistence and more effortful processing and pairs with had been induced (M 3.35) than when deactivating moods had
attentional focus, we expected that individuals in negative activat- been induced (M 2.69), t(81) 1.53, p .06. The control
ing moods benefit from longer time-on-task when performing condition fell in between (M 3.11) and did not differ from the
creative insight tasks. Put differently, whereas we did not expect activating or deactivating mood conditions, ts(81) 1, ns.
differences in creative fluency between positive and negative For the tone measure, results were also as expected: Whereas the
mood states, we did expect longer time-on-task to associate with activation contrast was not significant, t(81) 1, ns, the hedonic
creative fluency among (activating) negative mood states more
tone contrast was, t(81) 2.04, p .025. Participants reported
than among (activating) positive mood states.
more positive tone when positive moods had been induced (M
2.89) than when negative moods had been induced (M 2.53).
Method The control condition fell in between (M 2.69) and did not differ
Design and participants. Undergraduate students (N 90, from both the positive and the negative mood conditions, both
66% women, 34% men) participated for 5 (approximately U.S. t(81) 1.20, ns.
$6.50) and were randomly assigned to one of eight different mood Creative fluency. The number of correctly closed gestalts was
conditions (anger, fear, sadness, depression, happiness, elation, analyzed using the same set of a priori contrasts as used in Study
relaxation, calm) or to the mood-neutral control condition. Gender 2. Means and standard deviations, broken down for experimental
had no effects, and it is not discussed further. Dependent variables condition, are given in Table 3.
were manipulations checks, number of correctly closed Gestalts, The planned comparison grouping all positive mood states ver-
and time-on-task. sus all negative mood states was not significant, t(81) 1, ns.
Procedures, mood manipulations, and creativity task. These However, consistent with Hypothesis 1, a planned contrast group-
were the same as in Study 2, except that all materials were provided ing all activating moods versus all deactivating moods was signif-
through computers, and responses had to be given using a keyboard icant, t(81) 2.13, p .036. Participants in activating mood
and a computer mouse. Furthermore, to enhance comparability be- conditions had more correctly closed gestalts (M 7.02) than did
tween mood conditions, we also asked participants in the mood- those in deactivating mood conditions (M 6.25). Directional
neutral control condition to perform a task about autobiographical tests within the negative mood states showed that activating moods
memory. Participants were asked to write a short essay about the route produced more correct responses than did deactivating moods,
they took to the psychology department. They were specifically asked t(81) 1.85, p .05. Likewise, within the positive mood states,
to pay attention to the buildings they passed and to write their essay activating moods produced more correct responses than did deac-
in such a way that another person could imagine the route they took. tivating moods, t(81) 1.75, p .05.2
After finishing their autobiographical story, they were asked to report
Cognitive flexibility and persistence. The time participants
the major building that they passed. Third and finally, we replaced the
needed to correctly close the gestalts was log-transformed to deal
category inclusion task with the gestalt completion task, adapted from
with skewness. A 2 2 (Tone Activation) ANOVA on log-
Forster et al. (2004), which involves recognizing fragmented pictures
transformed time revealed the expected interaction between tone
of familiar objects. After the gestalt completion task, participants
answered a short questionnaire, were debriefed, and were paid for
participation.
Dependent variables. The hedonic tone and activation manip-
2
ulations were checked, as in Study 2. We coded the number of Differences in correctly closed gestalts may be due to differences in
closed gestalts as correct, incorrect, or missed. Although we had 10 incorrectly closed gestalts, and/or differences in number of nonresponses.
gestalts, initial analyses revealed one picture to be unsolvable (less For incorrect responses, no a priori contrasts were significant, ts(81) 1,
than 30% correctly closed, and over 50% missed). We decided to but participants in the activating mood conditions tended to miss fewer
than did those in the deactivating mood conditions (M 1.10 vs. M
base analyses on the remaining 9 pictures (including the tenth
1.73), t(81) 1.84, p .07. This suggests that the lower number of
gestalt produced similar results and identical conclusions). For correct closures in the deactivating mood conditions is due to a higher
each gestalt, we tracked the time in seconds between the appear- number of missed responses. Furthermore, inspection of Table 3 may
ance of the gestalt on the computer screen and the response (either suggest that anger and fear differ in terms of correctly closed gestalts and
a word or a hard return indicating a miss). The total time across the in number of misses. Statistically, however, this is not the case, ts(81)
nine gestalts served as our second dependent measure. 0.92 and 1.32, ps .22, respectively.
748 DE DREU, BAAS, AND NIJSTAD

Table 3
Means and Standard Deviations for Creative Performance and Time-on-Task as a Function of Experimental Manipulations (Study 3)

Experimental condition

Angry Fearful Depressed Sad Happy Elated Relaxed Calm Control


Dependent
variable M SD M SD M SD M SD M SD M SD M SD M SD M SD

Correct 7.36 1.20 6.70 0.94 6.11 1.73 6.37 1.19 7.00 1.00 7.00 1.41 6.00 2.01 6.55 1.58 6.67 0.89
Missed 0.90 0.90 1.40 0.69 1.67 1.50 2.00 1.85 1.34 1.13 1.25 1.04 1.90 1.37 1.58 1.48 1.53 0.99
Time-on-task 1.80 0.13 1.79 0.12 1.74 0.16 1.78 0.08 1.73 0.12 1.78 0.18 1.79 0.18 1.76 0.14 1.80 0.17
Time for
correctly
closed
gestalts 1.68 0.21 1.64 0.15 1.54 0.17 1.53 0.21 1.57 0.12 1.58 0.13 1.58 0.14 1.57 0.15 1.57 0.15

Note. Data for time-on-task and time on correctly closed gestalts are log-transformations of seconds across nine trials.

and activation, F(1, 71) 2.69, p .10 (marginal).3 Participants moods. We also found that higher creativity associated with en-
spent a significantly longer time on the task in the negative hanced perseverance in the case of negative tone. However, with
activating mood conditions than in the negative deactivating mood regard to the idea that cognitive flexibility is enhanced in the case
conditions (M 4.01 min vs. M 3.11 min), F(1, 71) 3.98, p of activating positive moods, evidence was less strong and, in
.05, and a nonsignificantly shorter time in the positive activating Study 1, statistically not reliable. Furthermore, we did not test the
mood conditions than in the positive deactivating mood conditions idea that cognitive flexibility (persistence) mediates between pos-
(M 3.01 vs. M 3.22, F 1). It thus appears that longer itive (negative) activating moods on the one hand and creative
time-on-task benefits participants in activating negative moods, fluency and originality on the other. In Study 4, we used the
who tend to persist, but not those in activating positive moods (see brainstorming task of Study 1 and, to enable formal tests of
Table 3 for log-transformed overall time-on-task, and time needed mediation, engaged a much larger number of participants. We
to correctly close). expected higher creative fluency and originality among activating
A complementary perspective is obtained by regressing creative moods than among deactivating moods to be due to greater cog-
fluency (number of correct responses) on level of activation, nitive flexibility when mood states are positive in tone (Hypothesis
hedonic tone, time-on-task, and their interactions. This produced a 4) and to greater persistence when mood states are negative in tone
significant regression model, R2 .16, F(6, 68) 2.23, p .05. (Hypothesis 5).
Consistent with the contrast analyses reported before, the main Another goal of Study 4 was to replicate results with a different
effects for hedonic tone ( .039, t 1) and for time-on-task assessment of mood states. Whereas the first three studies provided
( .11, t 1) were not significant, whereas the activation good evidence for the causal effects of discrete moods, we cannot
main effect was ( .21, t 1.98, p .05). Furthermore, the exclude a monomethod/operation biasthe possibility that our
interaction between hedonic tone and time-on-task was significant findings are limited to the specific ways we manipulated mood
in the activating mood conditions ( .52, t 3.72, p states in Studies 13. In Study 4, we therefore used a different
.001) and not in the deactivating mood conditions ( .15, t method: Participants rated their current mood state on a number of
1). Among activating negative mood states, longer time-on-task adjectives that were grouped according to their being positive in
associated with more correct responses ( .32, p .05); among tone or negative in tone and, independently, activating or deacti-
activating positive mood states, shorter time-on-task associated vating. These four dimensions were correlated with creativity
with more correct responses ( .76, p .01). This pattern of indices.
results strongly suggests that cognitive processes underlying cre-
ative performance qualitatively differ between positive and nega-
tive activating mood states. This is consistent with our notion that
Method
positive tone impacts creative performance because it allows for Design and participants. We used a correlational design with
cognitive flexibility and set-breaking (cf. Hypothesis 2), whereas measures of discrete moods as predictor variables and brainstorm-
negative tone impacts creative performance because it engenders ing performance creative fluency, originality, cognitive flexibil-
cognitive persistence and perseverance (cf. Hypothesis 3). Further- ity, and within-category fluencyas dependent variables. Partic-
more, results suggest that participants in a negative activating ipants were 546 first year psychology students (74% women, 26%
mood profited from longer time-on-task, whereas those in a pos-
itive activating mood or those in a (positive or negative) deacti- 3
An alternative approach would be to compute planned comparisons
vating mood did not. and to use the overall error terms and degrees of freedom (i.e., those of the
control condition as well). Doing so yields a highly significant contrast of
Discussion and Introduction to Study 4 negative activating moods against all others, t(81) 2.28, p .025, or
against the negative deactivating moods, t(81) 2.27, p .03. No such
Across a variety of tasks, results showed that activating moods effects were obtained when activating positive moods were contrasted
produce more creative fluency and originality than do deactivating against all others or against positive deactivating mood, ts(81) 1, ns.
MOODCREATIVITY LINK REVISITED 749

men) at the University of Amsterdam. They participated for partial Table 4


fulfillment of a course requirement. Factor Solution and Loadings for Mood Items (Study 4)
Procedure and independent variables. The study was included
in mass testing sessions (approximately 50 participants per ses- Mood item Factor I Factor II Factor III Factor IV
sion). Participants were seated in large lecture halls behind per- Disgusted .76 .31 .29 .22
sonal computers, which displayed all materials. Responses to Fearful .75 .31 .34 .57
questions could be typed in using the computer keyboard. Partic- Ashamed .74 .22 .48 .22
ipants were not allowed to talk and were required to work indi- Disdainful .73 .26 .33 .12
Worried .71 .20 .28 .43
vidually, at their own pace, and without consulting others. Exper- Afraid .71 .28 .33 .59
imenters supervised testing sessions and, when necessary, helped Guilty .68 .21 .46 .27
participants or enforced the above rules (this happened rarely). Angry .63 .17 .38 .37
Discrete moods were assessed by asking participants to com- Upset .62 .28 .57 .54
Happy .40 .83 .39 .40
plete a series of items that we derived from the PANAS (Watson Elated .21 .79 .17 .21
et al., 1988) or generated for the specific purpose of this study. In Excited .14 .75 .18 .08
total, participants indicated for each of 29 mood items how much Drained .36 .30 .85 .27
of the mood they had experienced since they got up that morning Lifeless .27 .40 .79 .24
(1 not at all to 5 very much so). Thereafter, supposedly as part Fatigued .26 .14 .78 .23
Depressed .36 .37 .72 .37
of a new and unrelated testing session, we introduced the brain- Discouraged .48 .35 .64 .35
storming task (for further detail, see the Method section of Study Failed .46 .24 .62 .53
1). When time was over, participants were informed that the test Sad .58 .33 .62 .51
was completed, and they continued with another, unrelated test. At Calm .37 .24 .37 .86
Relaxed .34 .54 .36 .77
the end of the semester, all participants received a written debrief- At ease .51 .36 .37 .65
ing along with a mailing address for further questions, and a
complaint form to be submitted when they did not want their data Eigenvalue 9.59 1.97 1.57 1.10
to be used (no additional questions or complaints were received). % variance 41.68 8.59 6.77 4.76
Independent variables. Initial factor analysis of the mood rat- Note. Numbers are factor loadings. Factor loadings in bold within one
ings revealed a six-factor solution, with four factors being readily column are grouped together in subsequent analyses. Factor I negative
interpretable and two factors grouping 5 items that had high activating moods; Factor II positive activating moods; Factor III
cross-loadings with other factors. These 5 items were dropped, and negative deactivating moods; Factor IV positive deactivating moods.
the remaining 24 items were submitted to a principal component
analysis. Because, in theory, dimensions could be correlated, we
applied oblimin rotation with Kaiser normalization. As expected, Creative fluency and originality. To test Hypothesis 1, we
we found a four-factor solution, explaining a total of 62% of the regressed creative fluency and originality on the four mood di-
variance. Table 4 summarizes the factor loadings and cross- mensions. Results are summarized in Table 6 and showed that first
loadings for all items on all four factors. The first factor groups of all, both negative and positive activating moods predicted
negative activating moods (e.g., angry, guilty), the second factor creative fluency. Second, inspection of the regression weights
groups positive activating moods (e.g., happy, elated), the third further reveals that positive activating moods significantly pre-
factor groups negative deactivating moods (e.g., depressed, dis- dicted originality. Because neither positive nor negative deactivat-
couraged), and the fourth factor groups positive deactivating ing mood states were related to creative fluency and originality,
moods (e.g., calm, relaxed). Ratings within each factor were av- these results provide new support for the hypothesis (Hypothesis
eraged to form one index. Internal reliabilities (Cronbachs alphas) 1) that activating mood states associate with more fluency and
were acceptable to good (see the Results section). originality than do deactivating mood states.
Dependent variables. The ideas, problem solutions, and sug- Cognitive flexibility and perseverance. For cognitive flexibil-
gestions generated by the participants were coded and/or trans- ity (i.e., category diversity), regression weights in Table 6 revealed
formed into the same components of creativity as used in Study 1 that only positive activating moods predicted flexibility; no other
(i.e., creative fluency, originality, cognitive flexibility, and perse- predictor was significant. This supports the hypothesis (Hypothesis
verance; .76 Cohens K .98). For originality, interrater 2) that activating moods promote cognitive flexibility, especially
agreement was satisfactory, ICC(1) .67, and we used the aggre- when tone is positive. Regression weights in Table 6 also reveal
gation across raters as an indicator of originality. that only negative activating moods predicted within-category
fluency. This supports the hypothesis (Hypothesis 3) that activat-
ing moods lead to greater persistence, especially when tone is
Results negative.
Mediation tests. To test for mediation (i.e., Hypothesis 4 and
Table 5 gives the descriptive statistics for all study variables. As 5) we computed a series of regression along the criteria set forth by
can be seen, we found moderate to strong correlations between the Kenny, Kashy, and Bolger (1998). We first tested whether cogni-
four mood dimensions and strong correlation between our four tive flexibility mediates the effects of positive activating moods on
indicators of creativity. Zero-order correlations between mood- creative fluency and originality (Hypothesis 4). When we re-
dimensions and indicators of creativity were low and generally gressed originality on positive activating moods after controlling
nonsignificant. for flexibility, the originally significant effect of positive activating
750 DE DREU, BAAS, AND NIJSTAD

Table 5
Descriptive Statistics for Dependent and Independent Variables (Study 4)

Variable M SD 1 2 3 4 5 6 7 8

1. Negative activating moods 2.03 0.69 .88 .45 ***


.71 ***
.59 ***
.06 .01 ***
.11 ***
.01
2. Positive activating moods 3.57 0.72 .80 .49*** .61*** .07 .09* .01 .07
3. Negative deactivating
moods 2.58 0.81 .88 .56*** .02 .01 .03 .01
4. Positive deactivating moods 3.67 0.78 .81 .02 .02 .06 .02
5. Creative fluency 5.06 3.66 .75*** .63*** .56***
6. Flexibility 2.30 1.49 .08 .76***
7. Within-category fluency 1.80 0.97 .05
8. Originality 2.68 0.77

Note. N 546. Scale reliabilities (Cronbachs ) are on the diagonal. Dashes indicate that there is no scale reliability to report.

p .10. * p .05. ** p .01.

moods dropped to a nonsignificant level ( .01, t 1), whereas lates with both within-category persistence and category diversity
flexibility was highly significant ( .76, t 26.90, p .001). (i.e., multiplying these results in creative fluency). However, neg-
A Sobel test confirmed that the mediation was significant, z ative activating moods only affected creative fluency through
2.45, p .015. In other words, consistent with Hypothesis 4, increased within-category fluency, and neither effect of negative
flexibility fully mediated the effect of positive activating moods on activating moods on category diversity nor mediation of category
originality (see also Figure 2a; We explored whether positive diversity was found. In the case of positive tone, results showed
activating moods relate to higher fluency because of greater cog- that activating moods have their effects on originality because of
nitive flexibility. This was not the case.) greater cognitive flexibility. These findings fit well with those of
We also examined whether persistence (i.e., within-category the previous studies and support our theoretical framework. Fur-
fluency) mediated effects of negative activating moods on creative ther, Study 4 showed that creativity was related to activating
fluency. When we regressed creative fluency on negative activat- moods and not to deactivating moods. This suggests that activation
ing moods after controlling for within-category fluency, the ini- stimulates creative performance rather than that deactivation un-
tially significant effect of negative activating moods dropped to a dermines creative performance.
nonsignificant level ( .02, t 1), whereas the effect of
within-category fluency was highly significant ( .63, t Conclusions and General Discussion
18.93, p .001). A Sobel test confirmed that the mediation was
significant (z 2.46, p .015). In other words, consistent with In their Annual Review of Psychology article, Brief and Weiss
Hypothesis 5, perseverance fully mediated the effect of negative (2002, p. 297) stated,
activating moods on creative fluency (see also Figure 2b).
It is apparent that discrete emotions are important, frequently occur-
ring elements of everyday experience. Even at workperhaps espe-
Discussion cially at workpeople feel angry, happy, guilty, jealous, proud,
etcetera. Neither the experiences themselves, nor their consequences,
Activating mood states related to a greater overall number of can be subsumed easily under a simple structure of positive or
unique ideas and, when mood states were positive, to higher levels negative states.
of originality. In the case of negative tone, results further showed
that activating moods have their effects on creative performance Quite consistent with this observation, the current study indeed
(i.e., creative fluency) because they enhance within-category per- showed that positive and negative mood states differentiate in
sistence. It should be noted though that fluency necessarily corre- terms of activating or deactivating nature (cf. Russell & Barrett,

Table 6
Regression of Cognitive Flexibility, Creative Fluency, Within-Category Fluency, and Originality on Positive Activating, Positive
Deactivating, Negative Activating, and Negative Deactivating Moods (Study 4)

Creative fluency Flexibility Within-category fluency Originality

Predictor variable t R2 t R2 t R2 t R2

Negative activating moods .12 2.00* .02 1 .16 2.48** .01 1


Positive activating moods .13 2.29** .14 2.47** .04 1 .11 1.98**
Negative deactivating
moods .02 1 .04 1 .08 1.30 .06 1
Positive deactivating moods .03 1 .02* 0.2 1 .015 .65 1 .02* .02 1 .01

Note. N 545.
*
p .05. ** p .025.
MOODCREATIVITY LINK REVISITED 751

A
Cognitive
Flexibility
= .14* = .75***

Positive Originality
Activating
Moods
= .11 * (.01)

B Within-Category
Fluency
= .16 * = .63***

Negative Number of
Activating Unique Ideas
Moods (Fluency)
= .12* (.02)

Figure 2. A: Path of positive activating mood states on originality mediated by cognitive flexibility (category
diversity). B: Path of negative activating mood states on creative fluency mediated by cognitive perseverance
(within-category fluency). Numbers in brackets are regressions weights after the mediator has been controlled
for. *p .05; ***p .01.

1999), and our results indicate that when it comes to creative 4 provided good evidence for mediation Hypotheses 4 and 5:
performance, both activation and hedonic tone are important. Negative activating moods related to higher fluency because of
Across four studies, findings were consistent with our dual path- increased persistence, whereas positive activating moods related to
way to creativity model, which indicates that only activating, and higher originality because of increased flexibility. We thus take
not deactivating, mood states lead to higher levels of creative these results as quite supportive of our dual pathway to creativity
fluency and originality, that activating positive mood states lead to model and its specific application to the mood creativity link.
creativity through higher levels of cognitive flexibility, and that All in all, results support four conclusions. First, activating
activating negative mood states lead to higher creativity through moods lead to more creativity than do deactivating moods, most
increased perseverance within thought categories and longer time- likely because activation stimulates creativity rather than because
on-task. Below, we discuss implications of these findings for deactivation undermines it. Second, activating moods with positive
research on mood and on creativity. We also discuss some limi- tone lead to creative performance through enhanced cognitive
tations to our findings and highlight avenues for future research. flexibility and inclusiveness. Third, activating moods with nega-
tive tone lead to creative performance through enhanced cognitive
Summary of Results and Theoretical Implications perseverance and persistence. Fourth, and finally, the effects of
mood on creativity cannot solely be understood in terms of acti-
From our dual pathway to creativity model, we derived five vation or in terms of hedonic tone; both dimensions are needed to
hypotheses about the effects of mood states on particular facets of understand how mood states influence creative performance. As
creativity. Hypothesis 1, predicting higher levels of creativity discussed below, these conclusions imply that different dimensions
when moods are activating rather than deactivating, received of creative performance, such as cognitive flexibility, inclusive-
strong supportwith regard to creative fluency it was supported in ness, or perseverance, cannot and should not be used interchange-
all three tests (i.e., Study 1, 3, and 4), and with regard to original- ably. Further, these conclusions imply that the task used to study
ity, it was supported in two out of two tests (i.e., in Study 1 and 4). creativity may determine the likelihood that some traits or states
Hypothesis 3, that negative activating moods positively associate do, and others do not, appear to successfully predict creativity.
with cognitive persistence, also received good support; direct
evidence was obtained in Study 1 and 4, and indirect evidence was
Mood States and Creativity
obtained in Study 3. Hypothesis 2, that activating positive moods
primarily associate with higher levels of cognitive flexibility, We began this research with the observation that there seems to
received less support; no evidence was obtained in Study 1 and 2, be general consensus that positive affect leads to more creativity.
indirect evidence was obtained in Study 3, and only in Study 4 Contemporaries tend to explain this effect in terms of hedonic
were statistical tests were supportive. This notwithstanding, Study tone. For example, Ashby et al. (1999) noted,
752 DE DREU, BAAS, AND NIJSTAD

There is substantial reason to believe that affect and arousal are not bility, divergent thinking, and the use of broad and inclusive
synonymous . . . and that the increases in cognitive flexibility and cognitive categories (cf., Murray et al., 1990). The present work
creative problem solving reported in so many articles are indeed due shows that in such tasks positive moods have an advantage over
to positive affect, not simply to increases in arousal. (p. 532) negative moods in producing creative ideas, insights, and problem
solutions. Other work has relied on tasks such as brainstorming
Current findings qualify these conclusions. When it comes to that allow creativity through persistence and perseverance to come
discrete mood states, we noted that not only hedonic tone but also about. The present analysis shows that in such tasks negative
activation matters and that tone and activation may take on differ- moods have an equal or perhaps even better chance than positive
ent roles in the creativity processactivation determines the like- moods of predicting creative performance. In short, an important
lihood of creative performance, and tone determines whether cre- insight that derives from our research is that the creativity task
ative performance comes about because of enhanced cognitive used may be a critical moderator of the relationship between mood
flexibility (in the case of positive tone) or because of enhanced (or any other trait or state for that matter) and creativity.
perseverance and persistence (in the case of negative tone). The currently proposed dual pathway to creativity model cap-
As mentioned in footnote 2, this is not the first study to examine tures past and current findings on the effects of positive moods on
the role of activation in the mood creativity link. Indeed, a num- creativity quite well. However, things are less clear cut when
ber of other studies focused on the role of arousal, typically considering the effects of negative moods, most notably those for
induced through some form of physical exercise. This past work sadness. Although current findings are supportive of the idea that
produced inconsistent results, sometimes showing that physical sadnessa negative and deactivating mood stateneither pro-
exercise produces more creativity than no exercise and sometimes duces nor inhibits creative performance, past work has revealed
showing that it has no effects. Obviously, there are important that sadness can actually stimulate creativity. For example, when
differences between activation induced through physical exercise the task is being framed as serious, important, and extrinsically
and activation associated with a particular mood state. This not- rewarding, sadness leads to more creativity than do mood-neutral
withstanding, it is important to note that past work on physical control conditions (Gasper, 2003; Hirt et al., 1997; also see, Martin
exercise and creativity did not differentiate between cognitive & Stoner, 1996). One could argue that such task framing is
flexibility and persistence and did not examine possible interac- motivating and activating and, as such, is doing what sad people
tions with hedonic tonefor some participants, physical exercise needthey need to be activated to perform because their mood
may have been a pleasant task, putting them in a good mood state in and by itself will not drive them toward (creative) perfor-
(happy, upbeat, relaxed) and thus, at best, facilitating cognitive mance.
flexibility. For some participants, however, physical exercise may Although we believe that the dual pathway to creativity model
have been an unpleasant task that put them in a negative mood has promise, we readily accept that invoking moderators may be
(upset, frustrated, worried, depressed) and thus, at best, facilitating needed to understand how particular (mood) states influence cre-
cognitive perseverance. Seen this way, it is not surprising that past ative performance. Important moderators may include task framing
work on physical exercise produced inconsistent results. Future and, as we elaborate on below, specific creativity task used. And
work on (physical) activation needs to take into account the although the current analysis focused on hedonic tone and activa-
possible side effects that manipulations have on participants mood tion as critical dimensions underlying discrete mood states, mood
as well as the dependent variables assessed (flexibility and/or states differ on other dimensions as well, and these may meaning-
persistence). fully relate to creativity. For example, Higgins (2006) has argued
Related to this is that past work has revealed an inverted that some mood states, such as happiness and anger, associate with
U-shape relationship between level of activation and arousal on the approach motivation and promotion focus, whereas other moods,
one hand and cognitive and motor performance on the other. Thus, such as fear and feeling relaxed, associate with avoidance moti-
at very low or extremely high levels of activation and arousal, vation and prevention focus (also see Amodio, Shah, Sigelman,
working memory capacity is much lower and relevant brain re- Brazy, & Harmon Jones, 2004; Carver, 2004; Higgins, Shah, &
gions function less effectively than at moderate levels of activation Friedman, 1997). Promotion focus relates to more creativity than
and arousal (cf., Yerkes & Dobson, 1908). An implicit assumption prevention focus (Friedman & Forster, 2001), and this together
in our work thus far has been that the variation in activation related may suggest that mood states associated with promotion focus
to particular mood states is in the lower range of this inverted produce more creativity than do mood states associated with
U-shape relation; only intense emotions may temporarily produce prevention focus. Future research may delve further into these
the level of activation and arousal that shuts down the system and possibilities, keeping in mind that a combination of hedonic tone,
prohibits people from performing cognitive and motor tasks. activation, and, perhaps also, regulatory focus better explains
Clearly, research is needed to further examine this issue. It would creative performance than do any of these dimensions alone.
be particularly interesting to see whether exceedingly high levels
of activation and arousal undermine cognitive flexibility and per- Study Limitations and Avenues for Future Research
sistence to the same degree or to different degrees. Intuitively, it
seems that flexibility is more vulnerable than persistence but, once Before concluding, a few limitations of our study design need
again, research is needed to examine this further. comment. First of all, our evidence for mediation is based on
That mood states impact creativity through different routes correlational designs, and future work is needed to unequivocally
cognitive flexibility or perseverance has an important method- establish the causal links. Second, the support for our dual pathway
ological implication. Some tasks used in creativity research, such model was stronger and more consistent for the negative activating
as Roschs category inclusion task, capitalize on cognitive flexi- moodpersistence creativity pathway, than for the positive acti-
MOODCREATIVITY LINK REVISITED 753

vating moodflexibility creativity pathway. To some, this may be 99% perspiration and 1% inspiration may reflect not only that
surprising because quite some evidence has been gathered showing Edison had apt intuition about the psychology of creativity but also
that positive tone relates to cognitive flexibility. Recent work by that Edison resembled an angry young man more than a happy
Hirt, Devers, and McCrea (2008) invoked hedonic contingency camper.
theory (Wegener & Petty, 1994), which posits that individuals in a
positive mood use greater scrutiny in activity choice than do those References
in neutral or negative moods because fewer activities will be able Amabile, T. M. (1983). The social psychology of creativity: A componen-
to maintain or improve their current mood. Hirt, Devers, and tial conceptualization. Journal of Personality and Social Psychology, 45,
McCrea showed that positive mood is indeed related to cognitive 357376.
flexibility to the extent that it allowed mood maintenance, and their Ambady, N., & Gray, H. M. (2002). On being sad and mistaken: Mood
results thus suggest that positive mood effects may be limited to effects on the accuracy of thin-slice judgments. Journal of Personality
tasks that allow participants to maintain their positive feeling. and Social Psychology, 83, 947961.
In a way, this work is consistent with the more general idea Amodio, D. M., Shah, J. Y., Sigelman, J., Brazy, P. C., & Harmon Jones,
E. (2004). Implicit regulatory focus associated with asymmetrical frontal
underlying the current work that specific tasks may facilitate or
cortical activity. Journal of Experimental Social Psychology, 40, 225
inhibit mood effects on creativity because some tasks provide 232.
more room for persistence and other tasks, for cognitive flexibility Ashby, F. G., Isen, A. M., & Turken, A. U. (1999). A neuropsychological
to come about. Future research could more systematically explore theory of positive affect and its influence on cognition. Psychological
the role of task environment (also see Kaufmann, 2003). For Review, 106, 529 550.
example, some studies on visual perception and set-breaking tend Ashby, F. G., Valentin, V. V., & Turken, A. U. (2002). The effects of
to provide limited time (e.g., 3 min; e.g., Forster et al., 2004) to positive affect and arousal on working memory and executive attention:
complete the task. Our Study 3 revealed that participants in an Neurobiology and computational models. In S. Moore & M. Oaksford
activating negative mood benefited from longer time-on-task (and (Eds.), Emotional contagion: From brain to behavior (pp. 245287).
spent, on average more than 3 min) whereas those in an activating Amsterdam, the Netherlands: Benjamins.
Baddeley, A. (2000). The episodic buffer: A new component of working
positive mood did not. An implication of our work thus is that
memory. Trends in Cognitive Sciences, 4, 417 423.
setting time limits may lead to misguided conclusions about the Barrett, L. F., & Russell, J. A. (1998). Independence and bipolarity in the
creative potential of particular states or traits. structure of current affect. Journal of Personality and Social Psychol-
A final avenue for future research is to analyze creative fluency ogy, 74, 967984.
and originality as a function of variables other than mood. We Bartolic, E. I., Basso, M. R., Schefft, B. K., Glauser, T., & Titanic Schefft,
already discussed regulatory focus and global versus local infor- M. (1999). Effects of experimentally induced emotional states on frontal
mation processing tendencies (cf., Forster et al., 2004; Friedman & lobe cognitive task performance. Neuropsychologia, 37, 677 683.
Forster, 2001). Other candidates for such analyses are the role of Baumann, N., & Kuhl, J. (2005). Positive affect and flexibility: Overcom-
intrinsic motivation versus extrinsic motivation (Amabile, 1983), ing the precedence of global over local processing of visual information.
achievement motivation, and traits such as openness to experience Motivation and Emotion, 29, 123134.
Blanchette, D. M., Ramocki, S. P., ODel, J. N., & Casey, M. S. (2005).
and conscientiousness (e.g., McCrae, 1987). It would be interest-
Aerobic exercise and creative potential: Immediate and residual effects.
ing to examine to what extent these and other variables known to Creativity Research Journal, 17, 257264.
affect creative fluency and originality do so because of enhanced Boden, M. A. (1998). Creativity and artificial intelligence. Artificial Intel-
flexibility, greater persistence, or some combination. ligence, 103, 347356.
Brand, S., Reimer, T., & Opwis, K. (2007). How do we learn in a negative
mood? Effects of a negative mood on transfer and learning. Learning
Concluding Thoughts
and Instruction, 17, 116.
Creativity researchers have long argued that positive mood Brief, A. P., & Weiss, H. M. (2002). Organizational behavior: Affect in the
increases creative performance and have implicitly or explicitly workplace. Annual Review of Psychology, 53, 279 307.
assumed this to be due to enhanced cognitive flexibility and Broadbent, D. E. (1972). Decision and stress. New York: Academic Press.
Carlsson, I. (2002). Anxiety and flexibility of defense related to high or low
reliance on broad, inclusive cognitive categories. Our results sup-
creativity. Creativity Research Journal, 14, 341349.
ported this idea and provided first time evidence for the notion that Carnevale, P. J., & Probst, T. M. (1998). Social values and social conflict
effects of positive mood states are limited to activating moods. in creative problem solving and categorization. Journal of Personality
Creativity researchers have long struggled with the effects of and Social Psychology, 74, 1300 1309.
negative moods on creativity, with some arguing and finding that Carver, C. S. (2004). Negative affects deriving from the behavioral ap-
negative moods undermine creativity and others arguing that it proach system. Emotion, 4, 322.
enhances creative performance. The present work clarified, first of Chaiken, S., Liberman, A., & Eagly, A. H. (1989). Heuristic and systematic
all, that negative moods enhance creative performance when mood information processing within and beyond the persuasion context. In
states are activating rather than deactivating. Second, our results J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 212252).
permit the conclusion that negative activating moods lead to cre- New York: Guilford Press.
Chamberlain, S. R., Muller, U., Blackwell, A. D., Robbins, T. W., &
ative performance because of enhanced cognitive perseverance
Sahakian, B. J. (2006). Noradrenergic modulation of working memory
and persistence more than because of cognitive flexibility and and emotional memory in humans. Psychopharmacology, 188, 387 407.
inclusiveness. Thus, provided some activation, both positive and Cicchetti, D. V., & Sparrow, S. A. (1981). Developing criteria for estab-
negative moods engender creative performance, but through cog- lishing interrater reliability of specific items: Applications to assessment
nitive flexibility and cognitive perseverance, respectively. As such, of adaptive behavior. American Journal of Mental Deficiency, 86, 127
our work suggests that Edisons famous quote that creativity is 137.
754 DE DREU, BAAS, AND NIJSTAD

Clore, G. L., Schwarz, N., & Conway, M. (1994). Affective causes and Flaherty, A. W. (2005). Frontotemporal and dopaminergic control of idea
consequences of social information processing. In R. S. Wyer & T. K. generation and creative drive. Journal of Comparative Neurology, 493,
Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 323 417). 147153.
Hillsdale, NJ: Erlbaum. Floresco, S. B., & Phillips, A. G. (2001). Delay-dependent modulation of
Colzato, L. S., Van Wouwe, N. C., & Hommel, B. (2007). Feature binding memory retrieval by infusion of a dopamine D1 agonist into the rat
and affect: Emotional modulation of visuo-motor integration. Neuropsy- medial prefrontal cortex. Behavioral Neuroscience, 115, 934 939.
chologia, 45, 440 446. Forster, J., Friedman, R. S., & Liberman, N. (2004). Temporal construal
Cropanzano, R., Weiss, H. M., Hale, J. M. S., & Reb, J. (2003). The effects on abstract and concrete thinking: Consequences for insight and
structure of affect: Reconsidering the relationship between negative and creative cognition. Journal of Personality and Social Psychology, 87,
positive affectivity. Journal of Management, 29, 831 857. 177189.
Damasio, A. R. (2001). Some notes on brain, imagination, and creativity. Fredrickson, B. L. (1998). What good are positive emotions? Review of
In K. Pfenninger & V. R. Shubik (Eds.), The origins of creativity (pp. General Psychology, 2, 300 319.
59 68). Oxford, England: Oxford University Press. Fredrickson, B. L., & Branigan, C. (2005). Positive emotions broaden the
Derryberry, D. (1989). Hemispheric consequences of success-related emo- scope of attention and thought-action repertoires. Cognition and Emo-
tional states: Roles of arousal and attention. Brain and Cognition, 11, tion, 19, 313332.
258 274. Friedman, R. S., & Forster, J. (2000). The effects of approach and avoid-
Derryberry, D., & Reed, M. A. (1998). Anxiety and attentional focusing: ance motor actions on the elements of creative insight. Journal of
Trait, state, and hemispheric influences. Personality and Individual Personality and Social Psychology, 79, 477 492.
Differences, 25, 745761. Friedman, R. S., & Forster, J. (2001). The effects of promotion and
DeSteno, D., Petty, R. E., Rucker, D. D., Wegener, D. T., & Braverman, J. prevention cues on creativity. Journal of Personality and Social Psy-
(2004). Discrete emotions and persuasion: The role of emotion-induced chology, 81, 10011013.
expectancies. Journal of Personality and Social Psychology, 86, 4356. Gasper, K. (2003). When necessity is the mother of invention: Mood and
Diehl, M., & Stroebe, W. (1987). Productivity loss in brainstorming problem solving. Journal of Experimental Social Psychology, 39, 248
groups: Toward the solution of a riddle. Journal of Personality and 262.
Social Psychology, 53, 497509. Gasper, K., & Clore, G. L. (2002). Attending to the big picture: Mood and
Dietrich, A. (2004). The cognitive neuroscience of creativity. Psychonomic global versus local processing of visual information. Psychological
Bulletin & Review, 11, 10111026. Science, 13, 34 40.
Drabant, E. M., Hariri, A. R., Meyer Lindenberg, A., Munoz, K. E., George, J. M., & Brief, A. P. (1996). Motivational agendas in the work-
Mattay, V. S., Kolachana, B. S., et al. (2006). Catechol place: The effects of feelings on focus of attention and work motivation.
o-methyltransferase valsuperscript 1-sup-5-sup-8Met genotype and neu- In B. M. Staw & L. L. Cummings (Eds.), Research in organizational
ral mechanisms related to affective arousal and regulation. Archives of behavior: An annual series of analytical essays and critical reviews
General Psychiatry, 63, 1396 1406. (Vol. 18, pp. 75109). Greenwich, CT: JAI Press.
Dreisbach, G., & Goschke, T. (2004). How positive affect modulates George, J. M., & Zhou, J. (2007). Dual tuning in a supportive context: Joint
cognitive control: Reduced perseveration at the cost of increased dis- contributions of positive mood, negative mood, and supervisory behav-
tractibility. Journal of Experimental Psychology: Learning, Memory, iors to employee creativity. Academy of Management Journal, 50, 605
and Cognition, 30, 343353. 622.
Dreisbach, G., Muller, J., Goschke, T., Strobel, A., Schulze, K., Lesch, Goldman-Rakic, P. S. (1996). Regional and cellular fractionation of work-
K. P., et al. (2005). Dopamine and cognitive Control: The influence of ing memory. Proceedings of the National Academy of Sciences, USA,
spontaneous eyeblink rate and dopamine gene polymorphisms on per- 93, 1347313480.
severation and distractibility. Behavioral Neuroscience, 119, 483 490. Grawitch, M. J., Munz, D. C., & Kramer, T. J. (2003). Effects of member
Duncker, K. (1945). On problem-solving. Psychological Monographs, mood states on creative performance in temporary workgroups. Group
58(5). Dynamics: Theory, Research, and Practice, 7, 4154.
Ekstrom, R. B., French, J. W., Harman, H. H., & Dermen, D. (1976). Gray, J. A. (1982). Precis of the neuropsychology of anxiety: An enquiry
Manual for kit of factor-referenced cognitive tests. Princeton, NJ: Edu- into the functions of the septo-hippocampal system. Behavioral and
cational Testing Service. Brain Sciences, 5, 469 534.
Eysenck, H. J. (1993). Creativity and personality: Suggestions for a theory. Gray, J. R., & Braver, T. S. (2002). Integration of emotion and cognitive
Psychological Inquiry, 4, 147178. control: A neurocomputational hypothesis of dynamic goal regulation.
Faust, M., & Mashal, N. (2007). The role of the right cerebral hemisphere In S. C. Moore & M. Oaksford (Eds.), Emotional cognition: From brain
in processing novel metaphoric expressions taken from poetry: A di- to behaviour (pp. 289 316). Amsterdam, the Netherlands: Benjamins.
vided visual field study. Neuropsychologia, 45, 860 870. Green, D. P., Goldman, S. L., & Salovey, P. (1993). Measurement error
Fiedler, K. (2000). Toward an integrative account of affect and cognition masks bipolarity in affect ratings. Journal of Personality and Social
phenomena using the BIAS computer algorithm. In J. P. Forgas (Ed.), Psychology, 64, 1029 1041.
Feeling and thinking: The role of affect in social cognition (pp. 223 Guilford, J. P. (1967). The nature of human intelligence. New York:
252). New York: Cambridge University Press. McGraw-Hill.
Filipowicz, A. (2006). From positive affect to creativity: The surprising Higgins, E. T. (2006). Value from hedonic experience and engagement.
role of surprise. Creativity Research Journal, 18, 141152. Psychological Review, 113, 439 460.
Fink, A., & Neubauer, A. C. (2006). EEG alpha oscillations during the Higgins, E. T., Shah, J., & Friedman, R. (1997). Emotional responses to
performance of verbal creativity tasks: Differential effects of sex and goal attainment: Strength of regulatory focus as moderator. Journal
verbal intelligence. International Journal of Psychophysiology, 62, 129 Personality and Social Psychology, 72, 515525.
141. Hirt, E. R., Devers, E. E., & McCrea, S. M. (2008). I want to be creative:
Finke, R. A. (1996). Imagery, creativity, and emergent structure. Con- Exploring the role of hedonic contingency theory in the positive mood
sciousness & Cognition, 5, 381393. cognitive flexibility link. Journal of Personality and Social Psychology,
Flach, F. (1990). Disorders of the pathways involved in the creative 94, 214 230.
process. Creativity Research Journal, 3, 158 165. Hirt, E. R., Levine, G. M., McDonald, H. E., Melton, R. J., & Martin, L. L.
MOODCREATIVITY LINK REVISITED 755

(1997). The role of mood in quantitative and qualitative aspects of Murray, N., Sujan, H., Hirt, E. R., & Sujan, M. (1990). The influence of
performance: Single or multiple mechanisms? Journal of Personality mood on categorization: A cognitive flexibility interpretation. Journal of
and Social Psychology, 33, 602 629. Personality and Social Psychology, 59, 411 425.
Hirt, E. R., Melton, R. J., McDonald, H. E., & Harackiewicz, J. M. (1996). Nieuwenhuis, S., Aston-Jones, G., & Cohen, J. D. (2005). Decision mak-
Processing goals, task interest, and the mood-performance relationship: ing, the P3, and the locus coeruleusnorepinephrine system. Psycholog-
A mediational analysis. Journal of Personality and Social Psychology, ical Bulletin, 131, 510 532.
71, 245261. Nijstad, B. A., & Stroebe, W. (2006). How the group affects the mind: A
Isen, A. M., & Baron, R. A. (1991). Positive affect as a factor in organi- cognitive model of idea generation in groups. Personality and Social
zational behavior. Research in Organizational Behavior, 13, 153. Psychology Review, 10, 186 213.
Isen, A. M., & Daubman, K. A. (1984). The influence of affect on Nijstad, B. A., Stroebe, W., & Lodewijkx, H. F. M. (2002). Cognitive
categorization. Journal of Personality and Social Psychology, 47, 1206 stimulation and interference in groups: Exposure effects in an idea
1217. generation task. Journal of Experimental Social Psychology, 38, 535
Isen, A. M., Daubman, K. A., & Nowicki, G. P. (1987). Positive affect 544.
facilitates creative problem solving. Journal of Personality and Social Nijstad, B. A., Stroebe, W., & Lodewijkx, H. F. M. (2003). Production
Psychology, 52, 11221131. blocking and idea generation: Does blocking interfere with cognitive
Isen, A. M., Niedenthal, P. M., & Cantor, N. (1992). An influence of processes? Journal of Experimental Social Psychology, 39, 531548.
positive affect on social categorization. Motivation and Emotion, 16, Osborn, A. F. (1953). Applied imagination. New York: Scribner.
6578. Paulus, P. B., & Nijstad, B. A. (2003). Group creativity: Innovation
Kaufmann, G. (2003). Expanding the mood-creativity equation. Creativity through collaboration. New York: Oxford University Press.
Research Journal, 15, 131135. Posner, J., Russell, J. A., & Peterson, B. S. (2005). The circumplex model
Kaufmann, G., & Vosburg, S. K. (1997). Paradoxical mood effects on of affect: An integrative approach to affective neuroscience, cognitive
creative problem-solving. Cognition and Emotion, 11, 151170. development, and psychopathology. Development and Psychopathology,
Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in social 17, 715734.
psychology. In D. T. Gilbert, S. T. Fiske, & G. Gardner (Eds.), The Rietzschel, E. F., De Dreu, C. K. W., & Nijstad, B. A. (2007). Personal
handbook of social psychology (pp. 233265). New York: McGraw-Hill. need for structure and creative performance: The moderating influence
Kimberg, D. Y., DEsposito, M., & Farah, M. J. (1997). Effects of
of fear of invalidity. Personality and Social Psychology Bulletin, 33,
bromocriptine on human subjects depend on working memory capacity.
855 866.
Neuroreport, 8, 35813585.
Rietzschel, E. F., Nijstad, B. A., & Stroebe, W. (2007). Relative accessi-
Lyubomirsky, S., King, L., & Diener, E. (2005). The benefits of frequent
bility of domain knowledge and creativity: The effects of knowledge
positive affect: Does happiness lead to success? Psychological Bulletin,
activation on the quantity and quality of generated ideas. Journal of
131, 803 855.
Experimental Social Psychology, 43, 933946.
Madjar, N., & Oldham, G. R. (2002). Preliminary tasks and creative
Robbins, T. W. (1984). Cortical noradrenaline, attention, and arousal.
performance on a subsequent task: Effects of time on preliminary tasks
Psychological Medicine, 14, 1321.
and amount of information about the subsequent task. Creativity Re-
Rosch, E. (1975). Cognitive representations of semantic categories. Jour-
search Journal, 14, 239 251.
nal of Experimental Psychology: General, 104, 192233.
Martin, L. L., & Stoner, P. (1996). Mood as input: What we think about
Runco, M. A. (2004). Creativity. Annual Review of Psychology, 55, 657
how we feel determines how we think. In L. L. Martin & A. Tesser
687.
(Eds.), Striving and feeling: Interactions among goals, affect, and self-
Russ, S. W., & GrossmanMcKee, A. (1990). Affective expression in
regulation (pp. 279 301). Hillsdale, NJ: Erlbaum.
Martindale, C., & Hasenfus, N. (1978). EEG differences as a function of childrens fantasy play, primary process thinking on the Rorschach, and
creativity, stage of the creative process, and effort to be original. Bio- divergent thinking. Journal of Personality Assessment, 54, 756 771.
logical Psychology, 6, 157167. Russell, J. A., & Barrett, L. F. (1999). Core affect, prototypical emotional
Martindale, C., Hines, D., Mitchell, L., & Covello, E. (1984). EEG alpha episodes, and other things called emotion: Dissecting the elephant.
asymmetry and creativity. Personality and Individual Differences, 5, Journal of Personality and Social Psychology, 76, 805 819.
77 86. Schooler, J. W., & Melcher, J. (1995). The ineffability of insight. In S. M.
McCrae, R. R. (1987). Creativity, divergent thinking, and openness to Smith, T. B. Ward, & R. A. Finke (Eds.), The creative cognition
experience. Journal of Personality and Social Psychology, 52, 1258 approach (pp. 97133). Cambridge, MA: MIT Press.
1265. Schooler, J. W., Ohlsson, S., & Brooks, K. (1993). Thoughts beyond
Mednick, S. A. (1962). The associative basis of the creative process. words: When language overshadows insight. Journal of Experimental
Psychological Review, 69, 220 232. Psychology: General, 122, 166 183.
Mikulincer, M., Paz, D., & Kedem, P. (1990). Anxiety and categorization: Schwarz, N., & Bless, H. (1991). Happy and mindless, but sad and smart?
II. Hierarchical levels of mental categories. Personality and Individual The impact of affective states on analytic reasoning. In J. P. Forgas
Differences, 11, 815 821. (Ed.), Emotion and social judgments (pp. 5571). Elmsford, NY: Per-
Mikulincer, M., & Sheffi, E. (2000). Adult attachment style and cognitive gamon Press.
reactions to positive affect: A test of mental categorization and creative Schwarz, N., & Clore, G. L. (1988). How do I feel about it? The infor-
problem solving. Motivation and Emotion, 24, 149 174. mative function of affective states. In K. Fiedler & J. Forgas (Eds.),
Mitchell, R. L. C., & Phillips, L. H. (2007). The psychological, neuro- Affect, cognition, and social behavior (pp. 44 62). Toronto, Ontario,
chemical, and functional neuroanatomical mediators of the effects of Canada: Hogrefe.
positive and negative mood on executive functions. Neuropsychologia, Shalley, C., Zhou, J., & Oldham, G. (2004). The effects of personal and
45, 617 629. contextual characteristics on creativity: Where should we go from here?
Mumford, M. D. (2003). Where have we been, where are we going? Taking Journal of Management, 30, 933958.
stock in creativity research. Creativity Research Journal, 15, 107120. Simonton, D. K. (1997). Creative productivity: A predictive and explana-
Mumford, M. D., & Gustafson, S. B. (1988). Creativity syndrome: Inte- tory model of career trajectories and landmarks. Psychological Review,
gration, application, and innovation. Psychological Bulletin, 103, 27 43. 104, 66 89.
756 DE DREU, BAAS, AND NIJSTAD

Simonton, D. K. (1999). Origins of genius: Darwinian perspectives on Torrance, E. P. (1966). Torrance tests of creativity. Princeton, NJ: Person-
creativity. New York: Oxford University Press. nel Press.
Simonton, D. K. (2003). Scientific creativity as constrained stochastic Usher, M., Cohen, J. D., Servan Schreiber, D., Rajkowski, J., & Aston
behavior: The integration of product, person, and process perspectives. Jones, G. (1999). The role of locus coeruleus in the regulation of
Psychological Bulletin, 129, 475 494. cognitive performance. Science, 283, 549 554.
Smith, S. M., & Blankenship, S. E. (1991). Incubation and the persistence Verhaeghen, P., Joormann, J., & Khan, R. (2005). Why we sing the blues:
of fixation in problem solving. American Journal of Psychology, 104, The relation between self-reflective rumination, mood, and creativity.
61 87. Emotion, 5, 226 232.
Smith, S. M., Ward, T. B., & Schumacher, J. S. (1993). Constraining Vosburg, S. K. (1998). Mood and the quantity and quality of ideas.
effects of examples in a creative generations task. Memory and Cogni- Creativity Research Journal, 11, 315324.
tion, 21, 837 845. Wadlinger, H. A., & Isaacowitz, D. M. (2006). Positive mood broadens
Spering, M., Wagener, D., & Funke, J. (2005). The role of emotions in visual attention to positive stimuli. Motivation and Emotion, 30, 89 101.
complex problem solving. Cognition and Emotion, 19, 12521261. Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
Staw, B. M., Sandelands, L. E., & Dutton, J. E. (1981). Threat-rigidity dation of brief measures of positive and negative affect: The PANAS
effects in organizational behavior: A multilevel analysis. Administrative scales. Journal of Personality and Social Psychology, 54, 10631070.
Science Quarterly, 26, 501524. Watson, D., Wiese, D., Vaidya, J., & Tellegen, A. (1999). The two general
Steinberg, H., Sykes, E. A., Moss, T., Lowery, S., & LeBoutillier, N. activation systems of affect: Structural findings, evolutionary consider-
(1997). Exercise enhances creativity independently of mood. British ations, and psychobiological evidence. Journal of Personality and Social
Journal of Sports Medicine, 31, 240 245. Psychology, 76, 820 838.
Sternberg, R. J., & Lubart, T. I. (1999). The concept of creativity: Pros- Wegener, D., & Petty, R. (1994). Mood management across affective
pects and paradigms. In R. J. Sternberg (Ed.), Handbook of creativity states: The hedonic contingency hypothesis. Journal of Personality and
(pp. 315). New York: Cambridge University Press. Social Psychology, 66, 1034 1048.
Strack, F., Schwarz, N., & Gschneidinger, E. (1985). Happiness and Yerkes, R. M., & Dodson, J. D. (1908). The relation of strength of stimulus
reminiscing: The role of time perspective, affect, and mode of thinking. to rapidity of habit formation. Journal of Comparative Neurology and
Journal of Personality and Social Psychology, 49, 1460 1469. Psychology, 18, 459 482.
Strauss, H., Hadar, M., Shavit, H., & Itskowitz, R. (1981). Relationship
between creativity, repression, and anxiety in first graders. Perceptual
and Motor Skills, 53, 275282. Received February 9, 2007
Thayer, R. (1989). The biopsychology of mood and arousal. New York: Revision received December 7, 2007
Oxford University Press. Accepted December 7, 2007
Journal of Personality and Social Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 94, No. 5, 757776 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.757

Judgments of the Lucky Across Development and Culture

Kristina R. Olson Yarrow Dunham


Harvard University University of California, Merced

Carol S. Dweck Elizabeth S. Spelke and Mahzarin R. Banaji


Stanford University Harvard University

For millennia, human beings have believed that it is morally wrong to judge others by the fortuitous or
unfortunate events that befall them or by the actions of another person. Rather, an individuals own
intended, deliberate actions should be the basis of his or her evaluation, reward, and punishment. In a
series of studies, the authors investigated whether such rules guide the judgments of children. The first
3 studies demonstrated that children view lucky others as more likely than unlucky others to perform
intentional good actions. Children similarly assess the siblings of lucky others as more likely to perform
intentional good actions than the siblings of unlucky others. The next 3 studies demonstrated that children
as young as 3 years believe that lucky people are nicer than unlucky people. The final 2 studies found
that Japanese children also demonstrate a robust preference for the lucky and their associates. These
findings are discussed in relation to M. J. Lerners (1980) just-world theory and J. Piagets (1932/1965)
immanent-justice research and in relation to the development of intergroup attitudes.

Keywords: preference for the lucky, immanent justice, evaluative contagion, social cognitive develop-
ment, cross-cultural psychology

In many societies and legal systems across time, one moral tenet time and place, from Aristotles Nicomachean Ethics, Roman law
has reigned supreme: Individuals are to be judged by the purpose- (e.g., animus nocendi), and English law (e.g., mens rea) to the
ful actions they commit and not by the random events that befall modern penal law in the United States and the Rome Statute of the
them. This understanding has been broad and deep, evident across International Criminal Court (United Nations, 2003). This funda-
mental moral dictum was most clearly described in the 13th
century by Henry Bracton (13th century/1968 1977): A crime is
Kristina R. Olson, Elizabeth S. Spelke, and Mahzarin R. Banaji, De- not committed unless an intention to injure exists. From it we
partment of Psychology, Harvard University; Yarrow Dunham, Depart- have the practice that volitional and premeditated behaviors, such
ment of Psychology, University of California, Merced; Carol S. Dweck, as stealing and cheating, are punished, just as hard work and
Department of Psychology, Stanford University. helping others are rewardedthese actions speak to the character
This research was supported by funding from the Beinecke Fellowship
of the person performing them. On the other hand, we treat
and National Science Foundation Graduate Research Fellowship to Kris-
tina R. Olson, National Science Foundation Grant BCS-02-1725 to Carol S.
differently those behaviors that involve accidental, unintentional,
Dweck, National Institute of Health Grant HD23103 to Elizabeth S. and random causes. Whether the outcomes themselves are good or
Spelke, funding from the Reischauer Institute for Japanese Studies to bad, such as winning a lottery or being hit by a tornado, we are not
Yarrow Dunham, and funding from the Third Millennium Foundation to to attribute these to the character of the actor.
Mahzarin R. Banaji. This research was conducted as part of Kristina R. Even when it comes to intentional behavior, we hold that it is
Olsons doctoral dissertation. those who are involved in producing it who should be held re-
A full list of items and supplementary statistics for all eight studies are sponsible or praised, not those who happen to be associated with
available at http://www.people.fas.harvard.edu/banaji/research/
the perpetrators via group membership. The Bible supports this
olson_luck.htm
We thank Alexa Reynolds for the drawings used in Study 8; the Harvard belief clearly: The fathers shall not be put to death for the
Museum of Natural History, Highland Park Elementary School, and the children, neither shall the children be put to death for the fathers:
Bing Nursery School for providing spaces for conducting research; and Every man shall be put to death for his own sin (Deuteronomy
Ann Marie Russell, Carla Borras, Rohini Rau-Murthy, Roy Ruhling, 24:16, King James Version). The belief that deems guilt by asso-
Marion Mahone, Valerie Loehr, Patrick Hayden, Lauren Hay, Rachel ciation to be immoral is also broad and deep, being upheld by the
Montana, Kimihiro Shiomura, Christopher Dial, Katie Lancaster, and oldest moral codes from Ptahotep and the Assize of Clarendon to
Marin Tanaka for assistance in data collection. We also thank the parents, most modern legal doctrine (Banaji & Bhaskar, 2000).
teachers, administrators, and children who participated and assisted in this
Our research concerns the dissociation between these ratified
research and Paul Jose for sending us the immanent-justice stories used in
Study 2. codes of conduct and the behavior of ordinary humans. It seeks to
Correspondence concerning this article should be addressed to Kristina understand the disparity between belief and action, between ab-
R. Olson, Department of Psychology, Harvard University, 33 Kirkland stractly held ideals and everyday moral judgments of good and
Street, Cambridge, MA 02138. E-mail: krolson@wjh.harvard.edu bad. In these studies, we investigated the developmental aspects of

757
758 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

such dissociations by analyzing the relatively early manifestation lucky could be that children are making dispositional explanations
of such discrepancies in childhood. Do children recognize that the for the lucky events. Such an explanation leads to the prediction
random bad events that befall others do not make them blamewor- that children growing up in a country that tends to use situational
thy? Do they understand that other people who are associated with explanations for behavior will not show this preference. Therefore,
an unlucky individual are not blameworthy? Observing children as in an initial exploration of the universality of the preference for the
they grapple with such questions can provide an understanding of lucky and its contagious nature, we presented young Japanese
the developmental origins of adult minds that routinely offer such children with the same tasks we presented to American children.
judgments with consequence.
We explored this question in the context of two empirical Immanent Justice and Belief in a Just World
phenomena: preference for the lucky (over the unlucky) and the
evaluation of an individual based on his or her association with As early as 6 months of age, children appear to have a basic
another actorwhat we call evaluative contagion. Preference for understanding of differences between intentional action and unin-
the lucky is simply the greater liking, greater preference, or more tentional action (Woodward, 1998), and by 3 years of age, children
positive attitude toward those who experience randomly occurring are able to distinguish intentional from unintentional actions in
good or lucky events (e.g., finding $5 on the sidewalk) than toward linguistic tasks (Shultz & Wells, 1985; Shultz, Wells, & Sarda,
those who experience random bad or unlucky events (e.g., getting 1980). Nevertheless, considerable evidence suggests that this dis-
splashed by a passing car). More complex evaluations, which we tinction is not the only guide to childrens evaluations of others.
call judgments of the lucky, involve thinking, for example, that Childrens tendency to evaluate others on the basis of uninten-
lucky people are more likely to perform good actions than are tional acts has been stated or implied by several prominent theo-
unlucky people. We use the term random as the overarching ries, most notably in work on immanent justice (Piaget, 1932/
term for lucky and unlucky events, standing in clear contrast with 1965) and belief in a just world (BJW; Lerner, 1980).
actions that we term intentional. Whereas intentional actions
tend to be deliberate and foreseen, random events, for our pur- Immanent Justice
poses, are those that are not intended or foreseen by the targets of
those actions. In his groundbreaking work on moral development, Piaget
Evaluative contagion refers to the extension of evaluations of (1932/1965) described the belief that a fault will automatically
one actor to his or her associates, such as family or social-group bring about its own punishment (p. 256). A classic example is
members. Disliking the sibling of someone who was splashed by evident in childrens responses to the following story: After steal-
a passing car would be an example of evaluative contagion, be- ing apples from an orchard, a boy rides his bike over a rotting
cause the negative evaluation of the target of the action (the person bridge and falls into the water. Piaget asked 6- to 12-year-old
splashed) has spread to the sibling of that target. Such evaluations children why the boy fell into the water and whether the boy would
are not only theoretically important but also may have important have fallen into the water had he not stolen the apples. A sizeable
implications for work on the development of prejudice toward number of young children reported that the perpetrator fell into the
disadvantaged groups. That is, insofar as members of disadvan- water because he stole the apples. In other words, the random bad
taged groups tend to experience more unlucky events, a dislike of event (falling into the water) was viewed as a direct consequence
people associated with others who experience unlucky events of an intentional bad action (stealing the apples). Other research
could lead to prejudice against members of families or social extended Piagets findings to positive events, showing that chil-
groups who themselves have not experienced bad or unlucky dren believe that a positive random event will occur as the con-
events. sequence of an intentional good action (Fein & Stein, 1977).
We seek to establish the generality and breadth of these phe- It is important to note that immanent-justice reasoning is a
nomena across age and culture. As we discuss below, there are mistaken belief about the nature of causation. That is, people who
several theories relevant to a preference for the lucky. One way to endorse immanent-justice reasoning are arguing that a good or bad
evaluate how well these theories explain the preference for the action can cause a lucky or unlucky event and, consequently, that
lucky is to examine the developmental predictions of these theories the lucky or unlucky event would not have occurred if the good or
and to look for convergence or divergence between these theories bad action had not occurred. For our purposes, the most important
and the preference for the lucky across development. Therefore, result is the developmental trend of this belief. Piaget (1932/1965)
one goal of this article is to investigate how the evaluations of the found a decline in immanent-justice reasoning across the elemen-
lucky and evaluative contagion might increase or decrease across tary school years. Subsequently, other researchers have confirmed
development and what these changes imply for alternative expla- the general decline of immanent-justice reasoning throughout
nations of these effects. childhood (Jahoda, 1958; Jose, 1991; Percival & Haviland, 1978;
In addition, we seek to understand whether these phenomena are Suls & Kalle, 1979; but cf. Karniol, 1980; Najarian-Svajian, 1966).
cross-culturally invariant or whether something about American or This work has been extended more recently into samples of older
Western culture might lead young children to prefer the lucky and teens, generally finding that immanent-justice reasoning further
their associates. Previous research (Masuda & Kitayama, 2004; decreases in middle and high school (Johnson, 1962; Najarian-
Morris & Peng, 1994) has demonstrated that Westerners tend to Svarian, 1966), although there is new evidence suggesting that
use dispositional attributions to explain behavior (e.g., he tripped immanent-justice reasoning may reemerge in adulthood (Callan,
because he was clumsy), whereas Easterners tend to use situational Ellard, & Nicol, 2006; Raman & Winer, 2004).
explanations (e.g., he tripped because there was a cord on the In our current research, we examined whether young children
floor). One possible explanation for childrens preference for the prefer lucky to unlucky individuals and whether they use evidence
JUDGMENTS OF THE LUCKY 759

of lucky or unlucky events to predict an actors future good or bad blame and evaluation (but see Fein, 1976). Although little research
behavior. That is, do children think a lucky child is more likely has been conducted on younger children, Lerner (1977) articulated
than an unlucky child to perform a good action in the future? a theoretical argument about the development of just-world think-
Although this question is clearly related to immanent justice, there ing. Most notable, he argued that children move from a focus on
are differences between these procedures. Immanent-justice re- getting what they want immediately to understanding that their
search focuses on how children reason about the causal conse- actions at Time A can be rewarded or punished at Time B. Lerner
quences of intentional good and bad actions. It shows a general related this transition to the development of delay of gratification
decline in immanent justice with age, presumably because children (Long & Lerner, 1974), arguing that once this action now
integrate and articulate a more diverse set of causal principles consequence later rule is understood, children begin to apply this
governing the behavior of agents (Schult & Wellman, 1997). In understanding to other people, recognizing that a persons actions
contrast, judgments of the lucky concern the evaluative conse- now will produce consequences for him or her later. These argu-
quences of having viewed anothers experience of a lucky or ments suggest that children may begin showing just-world beliefs
unlucky event. In Study 2, we investigated the relationship be- in mid childhood, somewhere around age 6 or 7 years. Lending
tween judgments of the lucky and immanent justice by testing the further credence to this approximate age prediction, Lerners own
developmental trajectories of both patterns of reasoning in the research tended to involve children in middle to late elementary
same participants. school, although he demonstrated related principles, such as an
understanding of parity and equity, in kindergarteners and first
BJW graders (Lerner, 1974).
The current work is aimed at testing the core proposition that
The idea that people get what they deserve is at the heart of children, starting early in childhood, prefer the lucky over the
Lerners BJW theory (Furnham, 2003; Lerner, 1980; Montada & unlucky. If BJW is indeed the mechanism by which preference for
Lerner, 1998). One classic demonstration of BJW involved asking the lucky emerges, then preference for the lucky should emerge
participants about the blameworthiness of a rape victim (Jones & sometime after BJW reasoning has developed. However, an alter-
Aronson, 1973). Experimenters manipulated whether the police native possibility is that the tendency to prefer the lucky precedes
report revealed that the victim was a virgin, a married woman, or the more elaborate sort of reasoning described in BJW; if so, it
a divorcee and then asked participants how much the victim was to should emerge earlier in development. Indeed, preference for the
blame for the rape. Counterintuitively, the finding was that partic- lucky might be a core, early-developing tendency that is later
ipants blamed the virgin and married woman the most and the justified via just-world beliefs. The present studies tested the
divorcee the least, though they still blamed the latter. The authors development of the preference for the lucky and assess its origin in
interpreted this finding and many others in terms of their self- relation to just-world beliefs.
protective function: If we believe the world to be a just and fair
place, we can reinterpret or explain good and bad events that seem The Current Work
to befall individuals for no reason at all and, as a result, still feel
personally safe. In this case, the authors argued that the idea of an One of the most important tasks children face in navigating their
innocent virgin or married woman being raped was simply so social world is determining who to approach and who to avoid,
inconsistent with participants view of a just world that they who is a friend and who is a foe. Therefore, we were interested in
derogated the victim, whereas a divorcee being raped is not as whether children would assume that lucky people are more likely
inconsistent with a view of the world as just, and therefore, less to engage in intentional good behaviors and unlucky people are
blame was necessary to maintain a sense of the world as just. more likely to engage in intentional bad behaviors. In an initial set
BJW colors not only peoples beliefs about others but also their of studies, we demonstrated that 5- to 7-year-olds prefer lucky to
attitudes. In another study, participants liked a person who was unlucky people and prefer members of a lucky group to members
randomly assigned to be electrically shocked with no compensa- of an unlucky group (Olson, Banaji, Dweck, & Spelke, 2006).
tion less than a person who was randomly assigned to be shocked Here, we pursued these findings by asking whether children make
for a payment of $30 (Lerner, 1971). The logic of BJW predicts deeper inferences about lucky and unlucky individuals, such as
that the victim of uncompensated shocks was denigrated because whether they believe lucky people are more likely to perform
of the underlying belief that a truly blameless person would not be intentional good actions, whether the preference for the lucky is
so treated. observed across cultures, and when this preference begins in
Traditionally, BJW researchers have tested these questions by childhood.
examining adults responses to extreme events that were presum- In the first two studies, we examined whether children judge
ably strong violations of a sense of justice (e.g., rape, electric lucky people as more likely to perform intentional good actions
shock, etc.). Less is known about whether more everyday events than unlucky people and unlucky people as more likely to perform
(e.g., seeing someone get splashed by a passing car) would trigger intentional bad actions than lucky people. In both studies, we
just-world beliefs. examined the developmental trajectory of these evaluations, and in
In addition, the developmental origins of BJW have not been the second study, we compared this trajectory with performance on
closely studied, as most developmental research has either focused an immanent-justice task.
on older children and teens (Furnham, 1985; Furnham & Raja- In the third study, we investigated whether young children show
manickam, 1992) or has involved tasks that have an uncertain evaluative contagion for behavioral predictions, asking whether
relationship to just-world beliefs themselves, such as distribution children believe that the siblings of lucky individuals are more
of resources (Lerner, 1974; Long & Lerner, 1974), rather than likely to engage in intentional good actions than the siblings of
760 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

unlucky individuals. This study, along with the final study, which person who found $5 on the sidewalk (random good event) is more
examines evaluative contagion in novel social groups, suggests likely to read a story to her little brother (intentional good event)
that evaluative contagion exists and is not limited to the American than a person who was rained on while walking home (random bad
context. Placed alongside the work on preference for the lucky, event)? Similarly, is the person who was rained on seen as more
these data suggest that the development of prejudice against mem- likely to lie to his mother (intentional bad event) than the person
bers of disadvantaged groups may be fueled by (a) the presence of who found $5? We tested this hypothesis and included comparison
negative evaluations of individuals who experience unlucky events items in which actors were described as having previously per-
and (b) the presence of negative evaluations of people merely formed intentional good or bad actions. Because such actions
associated with those who experience misfortune, together result- invite dispositional attributions and so should motivate consistent
ing in prejudice against the disadvantaged either because they predictions about future actions, these items served as a standard
themselves experienced bad luck or because they are associated against which to compare the impact of random events.
with others who have. In addition, this study investigated whether there are develop-
The second emphasis of the current article is an investigation of mental changes in the behavioral predictions of the lucky and
the developmental course of preferences and judgments favoring unlucky across middle childhood.1 Previous research has sug-
the lucky. In the first three studies, we tested children aged 4 to 12 gested that childrens moral reasoning changes considerably be-
years to assess the developmental trajectory of judgments of the tween the ages of 4 and 12 years and, most relevant, that
lucky. Previously, this question has only been addressed using an immanent-justice reasoning declines across this age range (Jose,
attitude measure in children aged 5 to 7 years (Olson et al., 2006). 1991; Piaget, 1932/1965). Therefore, we investigated possible age
In Studies 4, 5, and 6, we investigated the basic preference for the differences in childrens judgments of lucky and unlucky targets.
lucky in preschool-aged children. Testing such young children Although we provided conceptual arguments for why these two
allowed us to investigate whether the developmental predictions of phenomena are different, we sought to bolster this contention with
Lerner (1977), and therefore just-world beliefs, might explain such a direct test. If behavioral predictions following observation of
a preference. random events stem from the same underlying process as imma-
A final question this work seeks to examine is whether our nent justice, we would expect to see an age-related decrease in
initial discoveries of preference for the lucky are the result of some childrens tendency to think that lucky people perform good ac-
culture-specific teaching or whether this preference might be in- tions and unlucky people perform bad actions.
variant across cultures. A cross-cultural test, coupled with work
with very young children, can indicate whether a tendency or
evaluation might be universal or whether it is the result of specific Method
experiences or antecedents. Thus, in Studies 7 and 8, we investi- Participants and recruitment. Participants included 57 chil-
gated whether preference for the lucky and evaluative contagion dren (18 female) aged 4 to 12 years (M 7, SD 2) who
are seen cross-culturally or whether they are the result of a cul- participated while visiting the Harvard Museum of Natural History
turally specific experience or process. in Cambridge, Massachusetts, with their parents or guardians. For
Taken together, these studies have the potential to deepen our complete age breakdowns for this and subsequent studies, see
understanding of preference for the lucky, judgments of the lucky, Table 1. One additional participant began the study but quit after
and evaluative contagion effects. They can inform our understand- completing less than half of the study and is therefore excluded
ing of whether children merely prefer lucky people or whether they from analyses. Participants were approached by an experimenter
make corresponding behavioral predictions about lucky and un- who asked if they would be interested in participating in a study
lucky targets. These studies also clarify the relationship between lasting 5 to 10 min. Interested parents were asked to complete a
preference for the lucky, immanent justice, and just-world beliefs. short consent form, any questions from the child or parent were
Finally, these studies allow a new understanding of the develop- answered by the experimenter, and the child and parent were
mental trajectory and cross-cultural generality of these effects. escorted to the testing area. The experimenter then explained to the
child that he or she was free to stop participation at any time and
Study 1: Behavioral Predictions of the Lucky and asked the child if he or she was ready to begin. Although race
Unlucky information was not asked of participants in this study, experi-
menters observed that the sample was predominately White and
Learning to decide who is good and who is bad is a major middle to upper-middle class.
component of successful functioning in the social world. Previous
research suggests that even 6-month-old infants can distinguish an
1
agent that helps from an agent that harms and can use this infor- Social cognition research with adult participants has found that adults
mation to form preferences for the former (Hamlin, Wynn, & overestimate their ability to predict the behavior of individuals, for exam-
Bloom, 2007). In addition, by the age of 18 months, children prefer ple, thinking that they can predict ones year-long performance in the Peace
to accept a toy from a helpful rather than a harmful actor (Nurock, Corps from a single interview when in fact, interviews are poor predictors
of actual performance (participants estimated r .59 between interview
Jacob, Margules, & Dupoux, 2008). This evidence suggests that
performance and Peace Corps performance, when in fact, r .10; Kunda
very young children evaluate agents on the basis of their helpful or & Nisbett, 1986). The above-mentioned work differs from the work pro-
harmful behavior. posed here. Those authors were concerned with accuracy of predictions
What do children think when they observe something good or compared with reality, whereas the current work is focused on whether
bad befalling someone? Do they form expectations about that children make systematic predictions in a particular direction to reveal an
persons future behavior? For example, do children believe that a underlying belief that lucky people perform good actions.
JUDGMENTS OF THE LUCKY 761

Table 1 (lucky), intentional good versus random good, and intentional bad
Sample Size Broken Down by Age for Each Study (After versus random bad. These four types of learning pairings were
Exclusions) crossed with each of the two possible types of test items (inten-
tional good, intentional bad) in the test phase, resulting in eight
Age (years) types of items. A tabular representation of the design is depicted in
Experiment 2.5 3 3.5 4 4.5 5 6 7 8 9 10 11 12 N Table 2. Finally, we made two examples of each type of item (e.g.,
two items that were random good vs. random bad learning trials
1 6 9 6 10 6 10 6 3 1 57 with an intentional bad question item), resulting in 16 unique
2 8 25 23 19 14 17 14 7 127 questions. The order of mention of the targets (e.g., mentioning
3 14 11 19 7 8 4 7 6 2 78
4 12 31 29 27 16 115 lucky vs. unlucky first) was counterbalanced across items.
5 23 17 9 49 Experimenters in this study and all subsequent studies were
6 25 25 trained to state each item in a neutral or slightly positive tone, even
7 7 9 4 3 23 when the item was negative in valence. Although this was a less
8 21 30 28 8 87
natural way to state the items, this allowed us to be certain that
children were not using the experimenters tone as information in
their responses, so it provided a more conservative test of our
Materials. Thirty-two pictures of White children were se- hypotheses.
lected from the Internet and were arranged into 16 same-sex pairs. Data preparation and analyses. For each item, participants
Pictures were paired such that both pictures had been rated by were given a one if they selected the predicted choice (e.g., the
several adult raters as equal in attractiveness and approximate age. lucky target in the lucky vs. unlucky target predicting an inten-
Of the 16 pairs, eight were pairs of boys and eight were pairs of tional good action item) and a zero if they selected the other choice
girls. Adults estimates of age ranged from 4 to 12 years. These 16 (e.g., the unlucky target). The four items involving the same
pairs were then arranged into four same-sex sets of four pairs (two learning pairs (e.g., lucky vs. unlucky) were combined such that
all male, two all female). Four different versions of the task were each participant then had a prediction score between 0 (never
created, one beginning with each set of four pairs, alternating picked the predicted answer) to 4 (always selected the predicted
between four pairs of boys and four pairs of girls. The side of answer). Thus, if a participant said that a child who turned on the
presentation of each picture was orthogonally counterbalanced television and found no cartoons on was more likely to cheat on a
across participants, yielding eight versions of the task to which test than a child who turned on the television and found an extra
participants were sequentially assigned. hour of cartoons on, the participant was given one point for the
Procedure. Participants were presented with 16 trials. In each lucky versus unlucky prediction score. Similarly, if the participant
trial, they were first shown two photographs of children and were said that the child who walked to school while it was sunny was
told their names and one fact about them (e.g., This is John. John more likely to bake a cake for his grandma than the child who
stole a cookie from his brother). As each person was mentioned, walked to school while it was rainy, that participant scored one
a picture of a child appeared on the screen. Pictures were approx- point for the lucky versus unlucky prediction score. Each child
imately 2 in. 3 in. and appeared on either the right or left side ended up with four prediction scores (lucky vs. unlucky, inten-
of the screen. This part of the trial is called the learning phase
and always consisted of a learning pair (one fact about each of two
children). After the learning phase, participants were asked to 2
An anonymous reviewer raised an important concern that perhaps
guess which of the two children engaged in another action (e.g., children saw lucky items as good, rather than lucky, converting what is a
On Sunday, one of these children got into a fight. Which child got preference for the good into a preference for the lucky. For example, this
into a fight?). Henceforth, this part of the trial is called the test reviewer pointed out that getting to eat cake for a classmates birthday
phase. Participants were instructed to point to the child that they might be seen as good by children and not as lucky. Similarly, one could
believed engaged in the action in question, and their responses argue that children perhaps see unlucky events as bad, rather than unlucky.
One piece of evidence against this argument is the finding throughout this
were recorded.
article that children differentially evaluated lucky and good actors and that
The facts used in the learning phase were all either intentionally they distinguished unlucky and bad actors. However, as a more direct test
caused (by the actor) or randomly caused (not by the actor) and of this concern, we conducted a small-scale pilot study. We presented a
were either good or bad. For example, getting rained on or turning new group of 26 children aged 510 years with each of the lucky and
on the television to discover that no cartoons were on are examples intentional good items from Study 1 and asked them to state whether each
of random bad events, whereas finding $5 on the sidewalk or item was something lucky that happened to the target or whether the actor
getting to eat cake in school because it was a classmates birthday meant to do good (we used this phrase because intentional good is
are examples of random good events.2 In contrast, pulling a confusing for 5-year-olds). A partially overlapping group of 26 children
classmates hair or cheating on a test were examples of intentional aged 510 years completed the parallel task for unlucky and intentional bad
actions. Overall, the pilot participants identified 75.5% of the lucky targets
bad actions, and helping to bake cookies for ones grandmother or
as lucky, 74.0% of the intentional good targets as intentional good, 88.5%
sharing toys with ones little brother were considered intentional
of the unlucky targets as unlucky, and 83.3% of the intentional bad targets
good actions. Each pair of learning items was created to be as as intentional bad. Chance responding would have been 50%, and the
parallel as possible (e.g., accidentally hitting someone vs. inten- reviewers predictions would have suggested results significantly lower
tionally hitting someone). Across the 16 trials, four kinds of than 50% for the lucky and unlucky items, which we did not find. We are
pairings were made in the learning phase: intentional bad versus therefore confident that children did understand that the lucky events were
intentional good, random bad (unlucky) versus random good lucky and that the unlucky events were unlucky.
762 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

tional good vs. intentional bad, intentional good vs. lucky, and
intentional bad vs. unlucky) unless that child failed to complete
one or more questions required to complete a score. At most, one
participant was excluded from each prediction score. We analyzed
prediction scores using one-sample t tests, comparing childrens
prediction scores with chance (2.0). Finally, to examine possible
age changes in predictions, we correlated prediction scores with
age.

Results
Lucky versus unlucky. The comparison between lucky and
unlucky targets was the primary result of interest. Our hypothesis
Figure 1. Mean proportion of responses in which participants selected the
was that children would believe that the unlucky target was more predicted response in Study 1. The predicted response was selecting the
likely to perform an intentional bad action and less likely to lucky or intentional good actor to perform a good action and selecting the
perform an intentional good action than the lucky target, consistent unlucky or intentional bad actor to perform a bad action. The proportion of
with predictions of judgments of the lucky. A one-sample t test unpredicted responses is simply 1 the proportion of predicted responses.
comparing childrens mean prediction score (M 2.47) with Because including both the predicted and unpredicted bars is redundant, we
chance (2.0) supported this hypothesis, t(56) 3.57, p .001. only included the predicted responses in the graph.
Figure 1 shows the proportion of predicted responses made for
lucky versus unlucky items and intentional good versus bad items.
Using a paired t test, we found no effect of the valence of the that a person who did something intentionally good one time
question; that is, participants were just as likely to think a lucky would do so a second time and that someone who did an inten-
target would perform a good action as they were to think an tional bad action would do another. Previous work has suggested
unlucky target would perform a bad action, t(56) 1.53, p .13. that young children do not tend to believe that a person will
In addition, there was a nonsignificant but positive relationship necessarily do the same intentional action a second time (Kalish,
between age and prediction score (r .18, p .18), indicating 2002), but here we asked whether the individual would do a
that this was not likely to be related to immanent-justice reasoning, different intentional action of the same valence. We predicted that
which typically shows a decline with age (Jose, 1991; Piaget, they would expect valence consistency, and indeed, children
1932/1965). In addition, although previous work has suggested viewed an actor who had committed an intentional bad action as
that young children have a poorer understanding of randomness in more likely to perform a different intentional bad action and an
general than do older children (Weisz, 1980), the increase in actor who had committed an intentional good action as more likely
predictions based on lucky and unlucky events with age suggests to perform a different intentional good action as indicated by a
that our result is not due to limitations in childrens understanding one-sample t test comparing childrens average prediction score
of randomness. (M 3.19) with chance (2.0), t(56) 9.27, p .001 (see Figure 1
Intentional good versus intentional bad. We were interested for proportion of predicted responses). Children were equally
in whether children believed in behavioral consistency, thinking likely to think an intentional good actor would do another good
action as to think that an intentional bad actor would do another
bad action, as indicated by a paired t test in which responses did
Table 2 not differ by valence, t(56) 1.18, p .24. Prediction scores were
A Schematic Representing the Items Presented in Studies 1 and 3 correlated with age (r .45, p .001), suggesting that older
children were more likely to predict consistency in the behavior of
Test phase (Who intentional good and bad actors. Also as expected, childrens
Learning would perform Predicted
phase an. . . ?) response prediction scores were higher for the intentional good versus bad
comparison (M 3.19) than for the lucky versus unlucky com-
Intentional good vs. parison (M 2.47), at least indirectly indicating that children
intentional bad* Intentional good action Intentional good understand a distinction between intentional and random behavior,
Intentional good vs.
intentional bad* Intentional bad action Intentional bad
t(56) 4.45, p .001, although the next two comparisons test this
Lucky vs. unlucky* Intentional good action Lucky question more directly.
Lucky vs. unlucky* Intentional bad action Unlucky Intentional good versus lucky. To examine whether children
Intentional good vs. lucky Intentional good action Intentional good distinguish intentional behavior from random behavior, we asked
Intentional good vs. lucky Intentional bad action Lucky whether children would select an intentional good actor as more
Unlucky vs. intentional bad Intentional good action Unlucky
Unlucky vs. intentional bad Intentional bad action Intentional bad likely to perform a different intentional good action than a lucky
target and whether they would select a lucky target as more likely
Note. In the learning phase, participants were introduced to two charac- to perform an intentional bad action than an intentional good actor.
ters (and their siblings, in the case of Study 3). Participants were then asked We found that children make these selections, as evidenced by a
which of the two characters (or their siblings in Study 3) would perform a
different intentional good or bad action in the test phase. We have also
one-sample t test comparing participants average prediction score
listed the predicted response used to conduct analyses. The types of items (M 2.51) with chance (2.0), t(56) 3.90, p .001. Participants
with asterisks were included in Study 2. were equally likely to think that an intentional good actor would
JUDGMENTS OF THE LUCKY 763

perform a good action as they were to think that a lucky target good actions compared with intentional bad actors. Indeed, the
would perform a bad action, t(56) 0.14, p .89. Age was not trends for these cases were even stronger than in the case of
significantly correlated with prediction score (r .14, p .31), random events, demonstrating that children recognize a difference
suggesting that children of all ages distinguished intentional good between intentional and random actions.
and lucky actors. One possible concern regarding these results is that our partic-
Intentional bad versus unlucky. In the final set of compari- ipants in this study, as well as those in previous studies examining
sons, we paired intentional bad actors with unlucky targets and had a preference for the lucky (Olson et al., 2006), came from largely
children report which of these targets would commit other inten- advantaged populations (i.e., White, middle- to upper-middle-class
tional good or intentional bad tasks. We asked whether children children with parents willing and able to take them to a museum,
would select an intentional good actor as more likely to engage in etc.). Perhaps it is because they themselves are lucky or fortunate
a different intentional good action than a lucky target and whether that they show these effects. To test this possibility, we conducted
they would select a lucky target as more likely to perform an a pilot study with a sample of 23 participants (aged 6 12 years)
intentional bad action than an intentional good actor. We found who were all Black and all of low socioeconomic status, many
support for this prediction, as evidenced by a one-sample t test living at or below the poverty line. We found that these children,
comparing the average prediction score (M 2.69) with chance like the children in Study 1, predicted that a lucky target would
(2.0), t(55) 5.17, p .001, suggesting that children do in fact perform a good action more than an unlucky target and, similarly,
distinguish between actors who perform intentional bad actions that an unlucky target would perform a bad action more than a
and those who experience unlucky events, seeing the former as lucky target, suggesting that one does not need to be a member of
more likely to perform an additional bad action and the latter as a lucky group to make these evaluations.3
more likely to perform an additional good action. There was no As previously mentioned, Piaget (1932/1965) found a decrease
significant effect of the valence of the question asked, indicating in immanent-justice reasoning across childhood. In contrast, we
that children were equally likely to think that an intentional bad did not find such a pattern. If anything, the general trend was for
actor would perform other bad actions as to think that an unlucky older children to show behavior more in line with preference for
target would perform more good actions, as indicated by a one- the lucky than younger children showed. At the very least, these
sample t test, t(56) 0.65, p .52. In addition, older children results suggest a developmental dissociation between immanent
were more likely to demonstrate this prediction, as evidenced by a justice and preference for the lucky, militating against the idea that
significant correlation between age and prediction score (r .49, these phenomena arise from the same mental process or belief.
p .001), a somewhat surprising finding given that the previous However, to test the relationship between immanent justice and
comparison of intentional good with random good demonstrated predictions about lucky and unlucky targets behavior more di-
no significant relationship between an intentional and random rectly, in Study 2, we tested both phenomena in the same sample,
distinction and age. Because this finding was unexpected and did allowing us to empirically evaluate the relationship between im-
not replicate across these conceptually similar comparisons, we do manent justice and judgments of the lucky.
not address this issue further.
Nonparametric analyses. Because of possible concerns about
Study 2: The Dissociation of Immanent Justice and
the use of parametric statistics throughout this and subsequent
studies, we also conducted analyses throughout this article using
Behavioral Predictions of the Lucky
nonparametric statistics. However, because it is more common to Researchers since Piaget (1932/1965) have found that children
use parametric responses and because of limited space, parametric believe that intentional bad actions can cause unlucky events to
tests are always reported in this article. The relevant nonparametric occur, and this thinking has been applied to intentional good
tests are available at Mahzarin R. Banajis Web site (see the URL actions and lucky events (Fein & Stein, 1977). In these studies,
in the author note). The findings reported in the text are identical, children are often told about a person who has performed, for
regardless of our use of parametric versus nonparametric statistics. example, a bad action and who has then experienced an unlucky
event (e.g., a boy who stole apples from an orchard and then fell
Discussion through a bridge on his way home). Children are then asked why
the unlucky event happened and/or whether the unlucky event
This study provided evidence that children make behavioral would have happened if the child had not performed the intentional
predictions for lucky and unlucky targets. Children judged unlucky
targets as more likely to commit intentional bad actions and less
3
likely to commit intentional good actions than lucky targets. Thus, These participants were presented with only two items comparing
children do not simply prefer lucky to unlucky targets but make lucky and unlucky targets. In one item, they were asked which target would
different predictions about lucky and unlucky targets. These dif- perform an intentional good action, and in the other item, they were asked
fering predictions may suggest that children make enduring dis- which target would perform an intentional bad action. Participants were
positional inferences about actors and may rely on these inferences given one point if they selected the lucky target to perform the intentional
good action and one point if they selected the unlucky target to perform the
to motivate future predictions, although alternative accounts could
intentional bad action, resulting in scores of 0, 1, or 2 for each subject. The
explain these findings. distribution of these scores was compared with a binomial distribution
Study 1 also provided assurance that our basic method was valid (25% chance of 0, 50% chance of 1, 25% chance of 2). We found that this
in that our clearest case, comparisons between intentional good pilot sample selected the predicted responses more often than chance, as
and bad actors, showed the expected results. Participants judged indicated by a chi-square goodness-of-fit test, 2(2, N 23) 7.09, p
intentional good actors as more likely to perform other intentional .029.
764 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

bad action. The main result is that young children often say that the Procedure. Participants were brought into a conference room
unlucky event happened because of the intentional bad action and in the school and were greeted by an experimenter. They were told
that it would not have happened had the target not performed the they would be playing two games and that in both games there
intentional bad action. This type of reasoning is referred to as were no right or wrong answers. They were also informed they
immanent-justice reasoning. Studies have largely shown that as could quit at any time (although none of the children did). Partic-
children get older, their immanent-justice reasoning declines (Ja- ipants were sequentially assigned to complete either the immanent-
hoda, 1958; Jose, 1991; Percival & Haviland, 1978; Suls & Kalle, justice task and then the preference for the lucky task or vice versa.
1979). For the immanent-justice items, children were read a story while
As we have discussed, there is a similarity, although more being shown a photograph of a boy (the protagonist) and were then
superficial than it might seem, between the procedures that test the asked to recall as much as they could about the story. They were
idea of immanent justice and the present work. Whereas immanent then asked why the good or bad action from the end of the story
justice concerns reasoning about causation, the present research is happened (e.g., Why did Joey fall into the river?). Finally, they
not interested in causal relations. If anything, the causal pathway in were asked if the final action would have happened if the initial
our studies must be reversed, as children are told about lucky and action had not (e.g., Would Joey have fallen into the river if he
unlucky events and then infer intentional good and bad behavior. hadnt stolen the apples?). For the luck-prediction items, a picture
Although we have argued that these are conceptually distinct was presented to represent each of the two targets mentioned, and
phenomena, in Study 2, we tested this dissociation directly by on the test trials, participants were asked to indicate their answers
asking the same children to perform both an immanent-justice task by pointing to the target who they believed had performed the
and a judgment of the lucky task. Although age should be nega- action, as they did in Study 1. After completion of both tasks,
tively correlated with immanent-justice reasoning, as previous participants were thanked for their time and returned to class.
research suggests, age should be uncorrelated or even slightly
positively correlated with behavioral predictions of the lucky, as
Results
demonstrated in Study 1.
Data preparation. Responses to the why question (e.g., Why
did Joey fall into the river?) were coded by two judges, using
Method
predetermined categories from Jose (1991). The categories in-
Participants. Participants included 127 children (63 male, 64 cluded immanent-justice reasoning (e.g., He fell into the water
female; 118 White, 3 Asian, 2 Hispanic, 1 Middle Eastern, and 1 because he stole the apples); mediated causality, including phys-
Black/White biracial; 2 were not identified by parents and race or ical mediation (e.g., He fell through the bridge because he was
ethnicity could not be identified by experimenters) between the carrying so many apples) and psychological mediation (e.g., He
ages of 5 and 12 years (M 8.7, SD 2.0) in a suburban fell because he was feeling badly about stealing the apples and did
elementary school in Utah from a mostly middle-class background. not see the old board); chance contiguity (e.g., He fell into the
Materials. Participants completed two tasks, including an river because the bridge was old and the boards on the bridge were
immanent-justice task taken from Jose (1990) and the judgments falling apart); dont-know responses (e.g., I have no idea); and
of the lucky task from Study 1. Across participants, there were a uncodable responses (e.g., The boy didnt fall in the water).
total of eight immanent-justice stories: In four stories, the protag- Overall, raters agreed on categorization 96% of the time, and in
onist performed a bad action and then experienced a negative those cases in which they disagreed, the coders discussed their
event, and in four parallel stories, the protagonist performed a responses and came to an agreement on a final categorization.
good action and then experienced a positive event. For example, in Participants answers to the why and yes or no (i.e., would x
the orchard stories, the negative version involved a boy stealing have happened if y had not?) immanent-justice questions were
apples from an orchard and then falling through a board on a used for statistical analyses only if they had correctly answered the
bridge and into a river on his way home, whereas the positive memory question. A correct memory answer required the partici-
version involved a boy helping a farmer to pick apples and then pant to accurately recall the initial action and the final action in a
finding a wristwatch on a bridge on his way home. Four scripts given story (e.g., remembering that the boy stole apples and fell
were created, each consisting of one version of each of the four into the river). Overall, children passed the memory requirement
base stories. Each script contained two positive and two negative 89% of the time.
stories. Participants were randomly assigned a script. For the luck-prediction items, prediction scores were computed
The second task involved eight of the judgments of the lucky for each category (lucky vs. unlucky and intentional good vs. bad)
prediction items from Study 1. These included the four items from as was done in Study 1, resulting in a total score ranging from 0
the intentional good versus intentional bad set and the four items (never selected the expected response) to 4 (always selected the
from the lucky versus unlucky set. The other eight items from expected response) for each participant. Children who completed
Study 1 (the intentional good vs. lucky and intentional bad vs. all eight items (n 126) were included in all analyses.
unlucky items) were not used, as they made the task too long for Immanent justice. On the why question, children gave
the youngest children and were unrelated to the question of inter- immanent-justice responses 47% of the time, mediated-causality
est. The items were randomized into three scripts, each containing responses 3% of the time, chance-contiguity responses 43% of the
all eight items. Participants were randomly assigned a script. time, dont-know responses 4% of the time, and uncodable re-
Across items, the order of mention of lucky and unlucky targets sponses 3% of the time. On the yes or no questions, participants
and the order of mention of intentional good and bad targets was said yes 37% of the time and no (the immanent-justice re-
counterbalanced. sponse) 63% of the time.
JUDGMENTS OF THE LUCKY 765

The proportion of immanent-justice responses to the why ques- Table 3


tion was negatively correlated with age (r .19, p .033), Proportion of Responses on the Immanent Justice Task by Type
indicating that younger children gave more immanent-justice re- and Mean Prediction Score on Lucky/Unlucky and Intentional
sponses than did older children. Dont-know responses and uncod- Good/Bad Behavioral Prediction Items by Participant Age for
able responses were also negatively correlated with age (dont Study 2
know, r .22, p .018; uncodable, r .16, p .078). In
contrast, mediated-causality and chance-contiguity responses were Age (years)
positively correlated with age, indicating that older children were 5 6 7 8 9 10 11
more likely to use these explanations (mediated causality, r .26,
p .004; chance contiguity, r .25, p .004). Age was n 7, 8 24, 25 22, 23 19 14 16, 17 13, 14
negatively correlated with no answers on the yes or no question, No responses 0.44 0.79 0.71 0.67 0.61 0.60 0.54
Whyimmanent 0.49 0.44 0.69 0.47 0.42 0.42 0.30
again suggesting that younger children supplied more immanent-
justice
justice responses than did older children (r .22, p .016). Whymediation 0.00 0.01 0.01 0.02 0.05 0.00 0.09
Judgments of the lucky. Overall, children were more likely to Whychance 0.36 0.39 0.27 0.41 0.47 0.54 0.59
believe that an intentional good actor would perform an intentional Whydont know 0.12 0.09 0.03 0.04 0.00 0.03 0.00
good action and an intentional bad actor would perform an inten- Whyuncodable 0.04 0.07 0.00 0.05 0.05 0.00 0.02
Lucky/unlucky 2.75 2.28 3.22 3.05 3.36 3.06 2.85
tional bad action (M 3.39) than expected by chance, t(125) Intentional good/bad 2.62 3.08 3.22 3.47 3.64 3.88 3.54
19.74, p .001, one-sample t test. Children were also more likely
to believe a lucky target would perform a good action and that an Note. When two numbers are present, the valid sample size varied
unlucky target would perform a bad action (M 2.94) than depending on the dependent variable. No responses indicated proportion
of participants saying no on the yes/no immanent-justice question. The
expected by chance, t(125) 10.33, p .001, one-sample t test.
rows starting with why include the proportion of participants at each age
Performance on the intentional good versus bad items was corre- giving each of the possible responses to these items. The lucky/unlucky and
lated with performance on the lucky versus unlucky items (r .24, intentional good/bad rows indicate the average number of times (out of
p .007), although overall, children selected the predicted re- four) that participants at each age selected the good/lucky does good or
sponses more for the intentional good or bad items than for the bad/unlucky does bad responses.
lucky or unlucky items, t(125) 4.50, p .001, paired-samples
t test, as was the case in Study 1. In addition, age was positively Whereas immanent-justice reasoning relies on a misunderstanding
correlated with both composites (intentional good or bad, r .37, about causation (believing that performing a good or bad action
p .001; lucky or unlucky, r .19, p .032), suggesting that can cause a lucky or unlucky event to occur), predictions of the
older children were more consistent in their responses. It is im- behavior of lucky and unlucky people are not claims about cau-
portant to note that although the latter correlation was not signif- sality. One could imagine, for example, a person who does not
icant in Study 1 and is significant here, the effect sizes in both believe in immanent-justice reasoning but does believe that an
cases were nearly identical (r .18 in Study 1, and r .19 in unlucky person is more likely to perform a bad action.
Study 2), suggesting that the sample size explains this difference. With this effect established, we moved on to ask whether
Thus, age was negatively correlated with immanent-justice re- childrens inferences about the actions of lucky and unlucky tar-
sponses and positively correlated with judgments about the lucky gets are confined to the targets as isolated individuals or whether
and unlucky. associates of lucky and unlucky targets are also affected by the
The relationship between immanent justice and judgments of the targets circumstances. In other words, are those who are related to
lucky. Immanent justice was not related to predictions about the unlucky people seen as more likely to engage in bad actions? And
lucky and unlucky, as indicated by nonsignificant correlations is the converse true of someone who is the relative of a lucky
between the why and yes or no immanent-justice questions and the individual?
lucky or unlucky prediction composite (immanent-justice re-
sponses on the why question, r .06, p .50; no answers on
Study 3: Evaluative Contagion
yes or no question, r .05, p .5). Indeed, as noted, the age
trends for immanent justice and judgments of the lucky were in Although most people would argue that it is acceptable to judge
opposite directions (see Table 3 for all means by age). Thus, both someone on the basis of his or her intentions, almost nobody
conceptual and empirical arguments strongly suggest a distinct believes it to be fair to judge another by the random events that
basis for each phenomenon. befall him or her. Similarly, some believe it is undesirable to judge
an actors associate by the actions of the actor, even if the actor has
Discussion performed a premeditated crime but especially if the actor has been
the victim of a random negative or positive event. That is, making
Despite surface similarities between the judgments of the lucky negative inferences about the sibling of a known thief is not
task and immanent-justice reasoning, these two underlying phe- deemed right by some people, but making negative inferences
nomena are quite distinct. We found no significant relationship about the sibling of someone who was the victim of a robbery
between these measures. In addition, whereas immanent-justice seems even less permissible.
reasoning decreased with age, predictions about the lucky in- In Study 3, we tested whether this is indeed the case in the
creased with age, providing further evidence that the mechanism actions of children. We asked whether childrens behavioral pre-
responsible for these effects is not the same. In addition to an dictions of the lucky extend beyond evaluations of individuals to
empirical dissociation, we see a theoretical dissociation as well. evaluations of the associates of lucky and unlucky targets. That is,
766 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

if Jan found $5 on the sidewalk, would children believe that Jans stopped, stopped after eight items. Therefore, we also alternated
sister is more likely to perform a good action than Susans sister if whether children started at Item 1 or Item 9 to maximize the
Susan was splashed by a passing car? We compared these evalu- number of children completing each item. In total, 20 participants
ations with evaluations of the siblings of individuals who perform did not complete all 16 items; however, all participants, except the
intentional good and bad behaviors. Such a study can provide ones described below, completed at least six items.
initial information about whether evaluative contagion occurs in Data preparation and analyses. Several participants were
the prediction of behavior. If children evaluate people on the basis dropped from analyses for the following reasons: Participants
of the events that their associates experience, consistent with always picked the same side of the screen or picked the same side
predictions of evaluative contagion, this may illuminate how ste- of the screen on 15 of 16 trials (n 10), the parent interfered
reotypes and prejudice toward social groups, some of whom ex- during the task (n 2), the participant quit after one item (n 1),
perience more unlucky events, develop. or the child clearly did not understand the task (n 3). After these
exclusions, our sample included 78 participants (41 female), aged
4 to 12 years (M 7, SD 2).
Method
We then computed prediction scores in the same manner as in
Participants. Ninety-four participants (48 female) between Study 1; however, because 20 participants did not complete all of
the ages of 4 and 12 years (M 7, SD 2) were recruited to the items, we had to exclude these participants from any prediction
participate in this study, in the same manner as in Study 1. score in which they did not answer all four items, which resulted
Participant race and socioeconomic status were not requested, but in a sample of 58 63 participants for each prediction score (com-
experimenters reported that participants were largely White and, parable to the number of subjects in Study 1).4
because of location (campus museum), largely middle and upper- Data were prepared and analyzed using the methods described
middle class. in Study 1.
Stimuli. The exact items and pictures from Study 1 were used
in Study 3 with a few additions: We doubled the number of Results
pictures because a sibling was added for each actor. These pictures Lucky versus unlucky. In our main comparison of interest, we
were drawn from the same database of pictures used in Study 1. As found that participants were significantly more likely to pick the
in Study 1, the side of presentation of pictures was counterbal- sibling of the unlucky target to perform an intentional bad action
anced across participants, and the mention of targets (e.g., lucky than the sibling of a lucky target, who was in turn selected to be
first vs. unlucky first) was counterbalanced across items. more likely to perform a good action, as indicated by a one-sample
Procedure. Participants were read a script that included 16 t test comparing the mean prediction score (M 2.48) with chance
items. On each trial, participants were told the names of two (M 2.0), t(57) 3.51, p .001 (see Figure 2 for proportion of
children and a fact about each of them that was classified as either predicted responses). That is, children generalized evaluations of
intentional good, intentional bad, lucky, or unlucky (identical to an actor to the moral behavior of his or her siblings. It was possible
Study 1). Participants were also shown pictures of the siblings of that this significant effect was driven largely by children believing
each of the children. They were then told about another action either that the siblings of lucky people would do more good things
(intentional good or bad) and were told that the sibling of one of or that the siblings of unlucky people would do more bad things;
the two initial actors had performed that action. Participants were however, a paired-samples t test indicated that there was no sig-
asked to point to the sibling who they believed performed the nificant difference on the basis of the valence of the prediction
action. As in the previous studies, when a child or his or her sibling question, t(57) 0.90, p .37. As in Study 1, we found that age
was mentioned, a picture of that child appeared on the screen. did not correlate significantly with prediction score for the lucky
Below is a complete example of an intentional comparison: Inten- versus unlucky comparison, although it was in the positive direc-
tional good: This is Ross [picture appears] and his brother [picture tion (r .15, p .25).
appears]. Ross shared his toys with his neighbor. Intentional bad: Intentional good versus bad. The sibling of the intentional
This is Liam [picture appears] and his brother [picture appears]. bad actor was judged as more likely to perform another intentional
Liam stole a toy from his neighbor. Intentional bad: The brother of bad action than was the sibling of the intentional good actor (and
either Ross or Liam punched a classmate. Which brother punched vice versa for a different intentional good action), as indicated by
his classmate?
Below is a complete example of a random comparison: Unlucky:
This is Jeff [picture appears] and his brother [picture appears]. On 4
Because of concerns about the number of participants excluded in these
Saturday, Jeff turned on the television and found that there were no analyses, we reanalyzed the data using proportions, so that a child who
cartoons on. Lucky: This is Todd [picture appears] and his brother completed only two lucky versus unlucky items but selected the predicted
[picture appears]. On Saturday, Todd turned on the television and response on both items would get a score of 1.0, the same as a child who
found that there was an extra hour of cartoons on. Intentional good: completed four items and always selected the predicted response. Note,
Either Jeffs or Todds brother helped the teacher clean up after art. however, that a child who was distracted for one item would look very
different when he or she had answered two questions than when he or she
Which brother helped his teacher?
had answered four questions. In the former case, the child would get a
This study took slightly longer than previous studies, and there- score of .5 (chance), whereas in the latter case, the child would get a score
fore, halfway through the script (after eight items), we routinely of .75 (better than chance). This is why we did not analyze the data using
asked participants if they wanted to keep playing. Often children, this strategy in the text; this limitation not withstanding, the results of these
especially the younger ones, wanted to stop. We always allowed proportion-based analyses were almost identical to those reported in the
children to stop whenever they asked, and the majority, if they text using overall scores.
JUDGMENTS OF THE LUCKY 767

Study 1). Again, there was no significant difference between


prediction scores for good and bad prediction items, t(62) 0.11,
p .91.

Discussion
Study 3 demonstrated that children are willing to evaluate
people on the basis of the actions and experiences of their siblings.
The negative evaluation of unlucky people observed in Studies 1
and 2 rubs off on childrens evaluations of their siblingsthey
are seen as more likely to perform other bad actions. In the same
vein, siblings of lucky people are viewed as more likely to perform
intentional good actions. These findings provide evidence that
Figure 2. Mean proportion of responses in which participants selected the evaluative contagion exists and that childrens preference extends
predicted response in Study 3. The predicted response was selecting the to the associates of lucky and unlucky people.
sibling of the lucky or intentional good actor to perform a good action and Surprisingly, children seemed to lose the distinction they made
selecting the sibling of the unlucky or intentional bad actor to perform a in Study 1 between intentional and random events when evaluating
bad action. The proportion of unpredicted responses is simply 1 the the siblings of targets. Although children view siblings of inten-
proportion of predicted responses. Because including both the predicted tional good actors as likely to engage in intentional good actions
and unpredicted bars is redundant, we only included the predicted re-
when compared with intentional bad actors, they do not believe
sponses in the graph.
that siblings of intentional good actors are more likely to do so
than siblings of lucky targets. Similarly, although siblings of
intentional bad actors are seen as likely to engage in intentional
a one-sample t test comparing the average prediction score (M bad actions themselves when compared with siblings of intentional
2.66) with chance, t(58) 4.52, p .001 (see Figure 2 for good actors, they are not seen as more likely to do so than the
proportion of predicted vs. unpredicted responses). This result siblings of their unlucky counterparts. Lending further evidence to
suggests that, barring the presence of other information, children this claim is the fact that, unlike in Study 1, there is no significant
use the purposeful behavior of one sibling to predict the purposeful difference between the mean scores on the intentional good versus
behavior of another sibling. A nonsignificant paired-sample t test bad items and the mean scores on the lucky versus unlucky items,
indicated that this effect was equally driven by participants ten- t(56) 0.67, p .51. In other words, whether a child was robbed
dency to see the sibling of an intentional good actor as likely to or was a robber, the sibling was viewed equally negatively despite
perform a good action and by participants tendency to see the the fact that the evaluations of the actual robbed child or robber
sibling of an intentional bad actor as likely to perform a bad action, child may have differed. A possible explanation for this pattern is
t(58) 0.15, p .89. As in Study 1, age was correlated with that siblings merely get tagged with a valence (good vs. bad), and
performance on the intentional good versus bad comparison (r the nature of the original source event is not involved in the
.44, p .001), again suggesting that older children show more subsequent evaluation. We return to this affective tagging hypoth-
consistency across trials than do younger children. esis in the General Discussion, but given past findings (e.g., Olson
Intentional good versus lucky. Surprisingly, given the previ- et al., 2006), one could see how being a member of an unlucky or
ous results, participants did not distinguish between the siblings of otherwise disadvantaged group could lead to being negatively
intentionally good and lucky targets in predicting behavior of evaluated, even if the member being evaluated was not the person
siblings, as indicated by a one-sample t test comparing the mean involved in the original negative event (such as the siblings in this
prediction score (M 2.15) with chance (2.0; p .30). Thus, study).
children did not make a significant distinction between whether a These first three studies stand as evidence of the breadth of the
targets sibling performed an intentional good action or experi- preference for the lucky and evaluative contagion effects. These
enced a lucky event in evaluating that targets future behavior. phenomena extend beyond judgments of preference to beliefs
There was no significant difference between predictions of good about the likelihood of future action, including predictions of both
versus bad actions, as indicated by a paired-samples t test, t(59) a targets actions and the actions of a targets sibling.
0.74, p .46. Just as in Study 1, the correlation between perfor- All three studies also demonstrated a small increase in the
mance on this comparison and age was not significant (r .06, consistency of behavioral predictions for lucky and unlucky targets
p .63). over development, from roughly age 5 through age 12. In the next
Intentional bad versus unlucky. In a similar vein, across all three studies we further investigated the development of prefer-
participants, children did not distinguish between intentional bad ence for the lucky by testing whether even younger, preschool-
actors and unlucky targets in predicting sibling behavior (M aged children showed a preference for the lucky.
2.17, p .15). They did not evaluate the sibling of an intentional
bad actor to be any more or less likely to perform a different Study 4: Preference for the Lucky in Preschoolers
intentional bad action than the sibling of an unlucky target. In
addition, there was a correlation between age and this comparison The question of how early this preference emerges has not been
(r .31, p .013), suggesting that older children tended to show broached. In the current research, we tested this question directly
this expectation more than did younger children (as they did in by creating a simple task that very young children could perform.
768 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

Namely, we presented children with pairs of targets and asked Stimuli. Twenty-four pictures of children (12 male, 12 fe-
them simply whos nicer? As in Study 1, we compared chil- male) were selected from a larger database of photographs and
drens evaluations of lucky and unlucky targets but also compared made into 12 same-sex pairs, matched on adult ratings of attrac-
intentional good with intentional bad actors, intentional good with tiveness and age. Twenty-four statements were also created such
lucky targets, and intentional bad with unlucky targets. that six involved intentional good events, six involved intentional
Thus, we explored the emergence of these distinctions in 2.5- to bad events, six involved random good experiences, and six in-
4.5-year-old children. Evidence of a failure (random performance) volved random bad experiences. An object was used to represent
at one age and a success at the following age would suggest that each statement (to minimize memory demands), and a photograph
either a distinction begins to be made during this period or the task of a child was included to represent the target. For example, for the
is too hard for children below this age. To differentiate between item [John] helped his parents with the chores, a vacuum cleaner
these two possibilities, we compared performance on the compar- icon was presented along with a unique picture of a boy. Each
ison of interest (lucky vs. unlucky) with the other three compari- statement/object/photograph set was paired with another on a page
sons (intentional good vs. bad, intentional good vs. lucky, inten- of a flipbook. In total, participants saw three intentional good
tional bad vs. lucky). If children performed above chance in an intentional bad pairs, three random goodrandom bad pairs, three
expected direction on at least one comparison, this would suggest intentional goodrandom good pairs, and three intentional bad
that children understood the task and simply failed to make the random bad pairs.
lucky versus unlucky distinction. If they failed at all tests, it would In total, we created eight versions of the task to counterbalance
either mean that young children fail to make all distinctions or, for gender of targets, item effects, and the side of the flipbook on
more likely, that the children failed to understand the task. which each photograph appeared. The order of mention of targets
Because so many cognitive and social psychological changes (e.g., lucky vs. unlucky) varied across items. All subjects com-
occur in children during this time (e.g., emergence of reasoning pleted items in the following order, although the exact items
about false beliefs; Gopnik & Astington, 1988; Wellman, Cross, & differed across version: IG (intentional good)IB (intentional bad),
Watson, 2001), we placed children into narrower 6-month age RG (random good)RB (random bad), IGIB, RGRB, IGIB,
ranges to determine exactly how they perform at each age. For ease RGRB, IGRG, IBRB, IGRG, IBRB, IGRG, IBRB. This
of discussion, we label each age group by its lower bound (e.g., order was selected because the first six items were the primary
children aged 36 41 months are called 3.0-year-olds). This sam- ones of interest, and we were initially concerned that the younger
ple included not only younger children than the first three samples children might not sit through 12 items (although they did). Par-
but also more racial and ethnic diversity. This allowed us to test ticipants were sequentially assigned to one of the eight versions.
(beyond the pilot study mentioned in the discussion of Study 1) Procedure. Participants were brought to a small room and
whether our results are limited to majority group participants. were seated next to the experimenter. Children were told Were
Finally, by using such young children in this study, we shed going to play a game. This game is called the whos nicer? game.
light on whether BJW is a likely explanation for childrens pref- I will tell you about some people and then Ill ask you whos
erence for the lucky. As was described in our introduction, Lerner nicer? Does that make sense? Are you ready to play? Once
(1977) predicted that the origins of just-world beliefs are tied to children indicated that they were ready, the experimenter began
learning to delay gratification and a transition away from egocen- reading the pairs one at a time until participants completed all 12
trism. In addition, his theory postulated that children must under- items or something caused the participant to finish early (e.g., a
stand the relationship between their behavior and the consequences fire alarm). Five participants (4%) participated but were excluded
that occur later and then must apply this understanding to the because they failed to complete all 12 items. Failure to complete
behavior of others. All of these abilities are beyond the scope of the study was the result of accidents, such as fire drills, or a child
young preschoolers (Harris, 1992; Kurdek, 1979; Kurdek & Rod- needing to use the bathroom during the task.
gon, 1975; Mischel & Mischel, 1983), so evidence of a preference
for the lucky in young preschoolers would call into question BJW
Results
theory as an explanation for preference for the lucky in young
children. Data preparation. For each type of comparison, we computed
a separate score, giving children one point each time they selected
the predicted response (intentional good for the IGIB comparison,
Method
random good for RGRB comparison, intentional good for IGRG
Participants. Twelve 2.5-year-olds (5 female; 33.236.9 comparison, and random bad for IBRB comparison). Each child
months, M 35.1, SD 0.77), thirty-one 3.0-year-olds (16 therefore had a score that ranged from 0 (never picked the pre-
female; 36.3 41.9 months, M 39.0, SD 1.8), twenty-nine dicted response) to 3 (always picked the predicted response) for
3.5-year-olds (16 female; 42.0 47.8 months, M 45.2, SD each type of comparison. Scores were always compared with
1.6), twenty-seven 4.0-year-olds (18 female; 48.0 53.6 months, chance (1.5) using a one-sample t test.
M 50.4, SD 1.9), and sixteen 4.5-year-olds (8 female; 54.1 Overall results. Across all participants, responses for all
59.9 months, M 56.7, SD 2.1) participated. This sample was composites differed from chance in the predicted direction: inten-
considerably more diverse than those in the previous studies, as it tional good versus intentional bad (M 2.06, SD 0.88),
included 44 White, 11 Black, 19 Asian, 10 Hispanic, 2 Native t(114) 6.82, p .001; lucky versus unlucky (M 1.87, SD
American, and 17 biracial participants and 12 participants whose 0.88), t(114) 4.48, p .001; intentional good versus lucky (M
parents did not specify race or ethnicity. All participants were 1.86, SD 0.94), t(114) 4.10, p .001; intentional bad versus
recruited while attending a university preschool in California. unlucky (M 1.81, SD 0.90), t(114) 3.69, p .001.
JUDGMENTS OF THE LUCKY 769

Having a range of ages also allowed us to look for correlations between those who experienced lucky versus unlucky events, and
between age and performance. Age was correlated with all of the they made further distinctions between those involved in inten-
composites such that older children had higher scores on all tional versus random actions. In particular, it is interesting that
composites (intentional good vs. intentional bad, r .46, p these distinctions seem to emerge at approximately the same age.
.001; lucky vs. unlucky, r .31, p .001; intentional good vs. Of course, one possible explanation remains that the task was
lucky, r .46, p .001; intentional bad vs. unlucky, r .35, p simply too difficult for younger children. In Study 6, we addressed
.001). See Figure 3 for a breakdown of responses by age. We found this possibility by testing children in an even simpler task.
no significant effect of gender on any of the four composites ( ps Our previous study of attitudes toward the lucky and unlucky
.10). looked exclusively at children over the age of 5.0 years. The
2.5-year-olds and 3.0-year-olds. None of the indices differed current study allowed us to see that 4-year-olds do in fact demon-
significantly from chance for 2.5-year-olds ( ps .35) or for strate this preference and that even 3.5-year-olds do. In addition,
3.0-year-olds ( ps .15). this study newly examined the age at which children begin to
3.5-year-olds. Participants who were 3.5 years old judged the distinguish between evaluations of intentional actors and random
intentional good actors to be nicer than the intentional bad actors targets. Most of the studies that have compared intentional with
(M 1.93, SD 0.88), t(28) 2.63, p .014; the lucky targets accidental events have examined older children (Elkind & Dabek,
to be nicer than the unlucky targets (M 2.03, SD 0.78), 1977; Surber, 1982) or collapsed over large age ranges and there-
t(28) 3.70, p .001; the intentional good actors to be nicer than fore have not conclusively demonstrated that 3.5-year-olds show
the lucky targets (M 1.86, SD 0.88), t(28) 2.23, p .034; this distinction between the intentional and accidental (e.g., Shultz
and the unlucky targets to be marginally nicer than the intentional & Wells, 1985; Shultz et al., 1980; Yuill & Perner, 1988). We have
bad actors (M 1.79, SD 0.82), t(28) 1.93, p .064. demonstrated that the distinction between intentional and random
4.0-year-olds. Four-year-old participants judged the inten- is made reliably around age 3.5 years. At this age, children
tional good actors to be nicer than the intentional bad actors (M recognize that an individual is more good or more bad if he or
2.52, SD 0.75), t(26) 7.03, p .001; the lucky targets to be she acted with intent than if he or she happened to be a mere
nicer than the unlucky targets (M 2.15, SD 0.91), t(26) recipient of such events.
3.71, p .001; the intentional good actors to be nicer than the Another explanation for this effect needs to be addressed. It is
lucky targets (M 2.52, SD 0.64), t(26) 8.23, p .001; and possible that these studies created a preference for the lucky by
the unlucky targets to be nicer than the intentional bad actors (M forcing such a response. That is, perhaps young children actually
2.07, SD 1.00), t(26) 3.40, p .002. preferred the lucky and unlucky targets equally but merely dem-
4.5-year-olds. The 4.5-year-old participants also viewed the onstrated this bias because they had to select an answer, a judg-
intentional good actors as nicer than the intentional bad actors ment they would not have offered if left alone. Such a result is still
(M 2.63, SD 0.72), t(15) 6.26, p .001; the lucky targets interesting, and this possibility is worth testing, so we did so in
as nicer than the unlucky targets (M 2.13, SD 0.81), t(15) Study 5.
3.10, p .007; the intentional good actors as nicer than the lucky
targets (M 2.19, SD 0.91), t(15) 3.02, p .009; and the Study 5: Do Young Children Actually Think the Lucky
unlucky targets as nicer than the intentional bad actors (M 2.25,
and Unlucky Are Equally Nice?
SD 0.77), t(15) 3.87, p .002.
In Study 5, we examined the possibility that our findings in
Discussion Study 4 were the result of an experimental demand. To test this
possibility, in Study 5, we presented children with a forced-choice
Across ages, a consistent pattern emerged such that around the task that included a third alternative: Theyre exactly the same.
age of 3.5 years, children were able to make distinctions between Children were presented with the same stimuli as in Study 4, but
those who performed intentional good versus bad actions and were instead asked Whos nicer, [Johnny], [Jimmy], or theyre
exactly the same? If anything, this option should have created a
new demand: to employ use of the exactly the same response,
given its neutral stance and presence as an option. If childrens
natural inclination is not to differentiate between lucky and un-
lucky targets, then we should have seen children providing more
exactly the same responses than lucky responses, and we should have
seen the difference between lucky and unlucky disappear. In contrast,
if children believe that lucky is better than unlucky, we should
have continued to see the lucky targets selected more often than
the unlucky, in spite of the presence of an exactly the same
response.

Method
Figure 3. Mean number of times in Study 4 that children at each age
selected the lucky or intentional good actor as nicer than the unlucky or Participants. Participants included 49 children (33 female)
intentional bad actor. A score of 1.5 equals chance. Asterisks indicate that between the ages of 4.0 and 5.5 years (M 54.7 months, SD 5.1
the mean differed significantly from chance. Intent intentional. months) attending the same university preschool as those in Study
770 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

4. Two subjects were excluded from analyses because they did not dotally, we have observed that children have a tendency to pick the
understand the task, resulting in 47 children (33 female; M 55.0 last option given. Children indicated their responses by either
months, SD 5.1 months). This study also employed a diverse pointing or stating their response.
sample (racial/ethnic breakdown: 19 White, 4 Black, 7 Asian, 7 Data preparation. For the six items of each type (random or
Hispanic, 5 biracial or multiracial, and 5 did not specify). We intentional), we computed a score, tallying the number of times the
included children in this age group because they had most clearly good, bad, or exactly the same response was given. We then
demonstrated the effects in Study 4. compared each score with chance using a one-sample t test.
Stimuli and design. The stimuli were identical to those pre-
sented in Study 4, although the exact items were randomized. This Results and Discussion
time, the 12 items included six that compared lucky to unlucky and
six that compared intentional good to bad (because these were the Contrary to a task-demand explanation, we found that children
primary questions of interest, and we wanted to collect more data continued to select the lucky targets (M 3.14 out of 6) as nicer
on these items from each participant). The items of most interest more often than chance (2.0), t(48) 4.11, p .00l (see Figure 4).
(lucky vs. unlucky) were presented first. We included the inten- Because these responses were interdependent, it is not surprising
tional good versus intentional bad items to test whether the new that the unlucky targets were selected less often than chance (M
option, exactly the same, changed performance on the clearest 1.14), t(48) 4.83, p .001, and the exactly the same response
type of comparison and to test whether children understood the did not differ from chance (M 1.71), t(48) 0.97, p .34. It
task. We reasoned that if a child always responded exactly the is not surprising that children also selected the intentional good
same for each lucky versus unlucky item, this could be either actor most often for the intentional items (M 4.59) and that this
because he or she thought that random events were not indicative was selected more often than chance (M 2.0), t(48) 10.48,
of niceness or because he or she did not understand the task. We p .001 (see Figure 4). The intentional bad actors were selected
therefore included intentional good versus intentional bad to dis- less often than chance would predict (M 0.63), t(48) 10.07,
tinguish between these two cases. If a child used the exactly the p .001, and the exactly the same response was also selected less
same response for every random comparison but stopped using the than chance (M 0.78), t(48) 5.92, p .001. A summary of
exactly the same response for any of the intentional good versus the results can be seen in Figure 4.
intentional bad items, as a few children did, then we kept the child One possible explanation for these results is that children were
in the data set because it seemed clear that he or she understood the simply reluctant to use the exactly the same response and that this
task. If a child used the exactly the same response for all 12 items may therefore have been an unfair test. However, 55% of children
(including intentional and random), we hypothesized that the child used this response at least once during the task. For these partic-
did not understand the task, because it seemed unlikely that a child ipants, we computed a preference for the lucky score by subtract-
would believe that in all cases an intentional bad actor was just as ing the number of times they selected the unlucky as nice com-
nice as an intentional good actor. The latter situation occurred only pared with the number of times they selected the lucky as nice. We
one time, and this child was excluded from the dataset (one of the compared this value with zero using a one-sample t test and found
two excluded above). In total, four versions of the task were that even those participants who used the exactly the same re-
created, consisting of two different scripts, each with the order sponse at least once selected the lucky more than the unlucky,
described above. We then counterbalanced the scripts to control t(26) 2.06, p .05.
for the side of the page on which a given photograph appeared. In These results suggest that the findings in Study 4 were not
all scripts, the order of mention was varied across items. simply the result of a forced-choice task. Instead, we found that
Procedure. First, children were given three training trials for young children continued to articulate that the lucky target was
what we described as the first game, the whos taller game. The nicer than the unlucky target. Of course, children did use the
experimenter explained that two people would appear and the task exactly the same response from time to time, but they did not do
would be to say which one was taller: the first one, the second one, so more often than they selected the lucky target, and the addition
or they are exactly the same. In these trials, two stick figures were
presented. In the first two trials, one was clearly larger than the
other, and the experimenter indicated which one she would select
if asked who is taller. In the third trial, two stick figures of the
same size were presented, differing in color, and the experimenter
indicated that she would say Theyre exactly the same. Data
were not recorded for this task, but anecdotally, children seemed to
understand and often shouted their (always correct) responses
before the experimenter had a chance to say her opinion. Finally
the experimenter said that the subject would get to play a game but
that it was a little different from the whos taller game. Instead, the
game would be the whos nicer game, and children could select
either person or say theyre exactly the same. All children said
they understood, and the experimenter began. Children were read Figure 4. Mean proportion of times in Study 5 that each actor was
each of the 12 pairs of items and were then asked, Whos nicer, selected as the nicer one across six lucky versus unlucky items and across
[Alex], [Andrew], or theyre exactly the same? As a conservative six intentional good versus intentional bad items in which the option
test, we added exactly the same as the final response, as anec- exactly the same was also given. Int. intentional.
JUDGMENTS OF THE LUCKY 771

of this option did not undermine the preference for the lucky over selected the intentional good actors as nice more often than the
the unlucky. lucky targets, t(24) 2.22, p .036, and the unlucky targets as
nice more often than the intentional bad actors, t(24) 3.69, p
.001. Thus, by simplifying the attentional and memory demands of
Study 6: Preference for the Lucky in a Simplified Task
the task, we demonstrated that even children aged 3.0 years prefer
A remaining concern from Study 4 is the difficulty of the task lucky to unlucky individuals.
for children younger than 3.5 years of age. This possibility is In a pilot version of this study with 2.5-year-olds, we found that
suggested by evidence that 2.5- and 3.0-year-old children were no this task was too difficult for them. Children at this age either said
more likely to select the intentional good actor than the intentional nice for every item or simply refused to provide an answer,
bad actor as nicer. Despite our attempts to make Study 5 simple, suggesting that to ask whether children younger than 3.0 years
perhaps the memory and attention load (learning and then remem- show a preference for the lucky, a completely new, perhaps non-
bering what two different people did before making a response) verbal task needs to be created.
was simply too great for our youngest participants. Thus, in Study Across Studies 4 6, our results indicated that even very young
6, rather than presenting pairs of targets and asking children to preschoolers demonstrate a preference for the lucky over the
remember both before selecting an answer, we presented one target unlucky. This preference appears when lucky and unlucky indi-
at a time and simply asked children whether each target was nice viduals are pitted against each other in a forced choice, when
or not nice. Because Study 4 demonstrated a preference for the children have an explicit option to like lucky and unlucky targets
lucky in children beginning at age 3.5, in this study, we tested equally, and when the targets are presented serially. Such a finding
3.0-year-old children. causes some problems for the fullest just-world explanation.
Lerners (1977) hypotheses about the emergence of just-world
thinking suggests that children need to be many months if not
Method
years older to show the earliest evidence of just-world thinking.
Participants. Participants included twenty-five 3.0-year-old His explanation requires that children move beyond the pleasure
children (11 female, aged 36.7 41.8 months, M 39.3, SD 1.5) principle to the reality principle, which is expected to occur around
recruited from either the same campus nursery school in California the age of 6 or 7 years. Even with the most generous definition, 3
as the children in Studies 4 and 5 (n 12) or from a lab database years of age is clearly too young for such a transition. It seems
in Massachusetts (n 13). Participant race was not recorded, but highly unlikely, given other results in cognitive development, that
we estimated that the final sample was approximately 60% White 3.5-year-olds have the cognitive capacities and awareness, such as
and that the remaining 40% were evenly distributed among Black, perspective taking and delay of gratification, required for just-
Asian and multiracial participants. world types of reasoning (Kurdek, 1979; Kurdek & Rodgon, 1975;
Procedure. Participants were brought to a small testing room Mischel & Mischel, 1983).
and were seated next to the experimenter. The experimenter told These results provide clear support for the hypothesis that a
the child that he or she was going to see some other kids and be preference for the lucky is in place by age 3.0 and that it may be
asked whether each target was nice or not nice. Participants were present prior to that age. It is possible that in future research with
presented with a total of 24 targets, six of each type (lucky, new procedures, such a preference may be detected even earlier. It
unlucky, intentional good, intentional bad) in one of four possible is also possible that a simpler version of the just-world belief (e.g.,
scripts (2 randomized orders 2 gender orders). a basic belief that good things happen to good people and bad
things happen to bad people, without a deeper understanding of the
complexity of the world or an ability to inhibit their actions) is held
Results and Discussion
by very young children, although establishing that such a theory is
Data preparation and analyses. In general, children were in place would require additional work.
more like to say nice than expected by chance (50%), t(24) One of the major undertakings of the current article was to
2.27, p .032, and this was true for some participants more than investigate how widespread the preference for the lucky is. We
others. We were not concerned with this fact, however, given that have now demonstrated that it is seen in children ranging in age
this bias should have been equally prevalent across item types and from 3 to 12 years, that older children extend this preference to
that our analyses were within subject. A composite score was predictions of the intentional behavior of lucky and unlucky tar-
created for each type of item such that the total number of nice gets, and that they extend the preference to the siblings of the
judgments (out of six possible) was computed for each subject. We targets. In the final two studies, we investigated whether prefer-
then compared these means using paired-sample t tests. ence for the lucky and evaluative contagion appear cross cultur-
Participants most often designated the intentional good actors as ally.
nice (M 5.0, SD 2.1), followed by the lucky targets (M 4.2,
SD 2.0), then the unlucky targets (M 3.4, SD 2.1), and then Study 7: Cross-Cultural Evidence of the Preference for
the intentional bad actors (M 2.4, SD 2.2); this pattern was the Lucky
demonstrated by a significant linear trend, F(1, 24) 24.12, p
.001. In addition, all paired t tests indicated differences. Most In two final studies, we asked whether the preference for the
notable, the lucky targets were more often labeled as nice com- lucky is constrained to the minds of young children from Western
pared with the unlucky targets, t(24) 2.16, p .041, and the cultures or whether it might instead be a preference held by young
intentional good actors were labeled as nice more often than the children across very different cultures. As our first test of this
intentional bad actors, t(24) 5.17, p .001. Children also question, we investigated whether young children who were raised
772 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

in a culture that appears to promote fewer trait inferences than that


of the United States show this same preference. Research on causal
attributions has examined cross-cultural differences in adults ten-
dency to use situational versus dispositional (trait) explanations of
human behavior (Masuda & Kitayama, 2004; Morris & Peng,
1994).5 Although sometimes the findings have been mixed, when
differences have been found, they have tended to fall along Eastern
(Japan, India, etc.) versus Western (United States, England, etc.)
lines, with Easterners tending to use more situational explanations
for behavior and Westerners using more dispositional explanations
(Krull et al., 1999; Masuda & Kitayama, 2004; Morris & Peng,
1994). As noted, it is possible that dispositional attributions are at
Figure 5. Mean liking rating for intentional good, lucky, unlucky, and
the heart of childrens preference for the lucky. That is, perhaps intentional bad actors in Study 7, as rated by Japanese children. Higher
because American children live in a culture that values disposi- scores indicate greater liking, and error bars indicate standard error of the
tional attributions, they tend to blame unlucky targets and credit mean.
lucky targets, essentially overextending dispositional explanations
to random events. Therefore, Japan stands as an interesting test
case for examining cultural variability in the preference for the
asked the child if he or she understood how to use the scale.
lucky over the unlucky. If children in both cultures show a pref-
Children were then read one of the scripts that included 10 items
erence for the lucky over the unlucky, then cultural differences in
describing the actions of an individual or an event experienced by
attributions likely do not explain the preference-for-the-lucky ef-
an individual (e.g., Tarou helped his teacher). After reading each
fect.
vignette, the experimenter asked children to indicate how much
For this study, we employed a simple test of preference for the
they liked each actor using the 6-point smiley-to-frowny-face
lucky in the form of liking judgments of lucky and unlucky targets.
scale. These scores were then converted to a 6-point numeric scale.
Such a test allowed us to examine whether Japanese children
Data preparation. Following the procedure of Olson et al.
differentially evaluated lucky and unlucky targets, rather than
(2006), the average rating for each type of target (intentional good,
asking them to predict behavior (a task that we reasoned required
intentional bad, lucky, unlucky) for each participant was com-
a more elaborated judgment, so if differences occurred, they could
puted. We used paired t tests to compare ratings of targets.
be explained by several factors). In addition, by employing the
selected method, we could compare the results with published
findings with an equivalent American sample (Olson et al., 2006). Results and Discussion
Ratings. Japanese children preferred intentional good targets
Method (M 4.68) to intentional bad targets (M 3.06), t(22) 3.61,
p .002, and lucky targets (M 4.24) to unlucky targets (M
Participants. Twenty-six children from rural Japan partici- 3.12), t(22) 2.87, p .009 (see Figure 5). These results support
pated; 3 were excluded because of poor participation (e.g., giving the claim that Japanese children have a preference for the lucky
the same response to every item), resulting in 23 participants (10 over the unlucky, despite living in a culture that tends to use fewer
female; aged 4 7 years; M 5). We selected these ages to dispositional attributions and despite our use of a non-forced-
approximately match those used in Olson et al. (2006), which choice method. It is interesting that the difference between inten-
employed an identical method. In actuality, this sample was ap- tional bad actors and unlucky targets was not significant ( p .75)
proximately 6 months younger than the Olson et al. sample. and that the difference between intentional good actors and lucky
Stimuli. In total, 40 vignettes were created, 10 of each type targets was only marginally significant, t(22) 1.85, p .08. In
(intentional good, intentional bad, lucky, unlucky). These items addition, we computed the difference between evaluations of in-
were scrambled and divided into lists of 10 items each. The items tentional good and bad actors and the difference between evalua-
were then scrambled again, and four more lists were created, tions of lucky and unlucky targets. We then compared these
making a total of eight lists. Each list contained at least one item differences and found that there was no significant difference,
of each type, with the gender of the targets alternating by item. t(22) 1.32, p .20. That is, Japanese children made almost no
Participants were sequentially assigned to a list. All items were
taken from Olson et al. (2006) and were translated into Japanese by
5
Yarrow Dunham and then checked and back translated by two The few studies that have directly examined cross-cultural causal
native Japanese speakers to ensure accuracy. The only changes attributions in children have been conducted by Miller (1984, 1986). In
made were those necessary to maintain understanding (e.g., in the those studies, she asked children to spontaneously name examples of
American version, the target found $5 on the sidewalk, whereas in intentional good and bad actions from their lives, and she tested whether
their explanations for these actions were more situational or dispositional.
the Japanese version, the target found 500 yen, and names were
Her studies differed in several significant ways from the current work:
changed from Mike to Minoru). There was no investigation of random events; the events were produced by
Procedure. First, children were trained to use a 6-point the subjects, not the experimenters; the children were older than those
smiley-to-frowny-face scale that they were to use later in the study. examined here; and her sample was Indian, not Japanese. She found no
The experimenter gave examples of how he would evaluate dif- significant differences across Indian and American cultures in childrens
ferent people (e.g., his mother vs. his neighbor) using the scale and tendency to use situational versus dispositional explanations.
JUDGMENTS OF THE LUCKY 773

distinction between whether targets engaged in intentional behav- groups were never explicitly labeled and were distinct only be-
ior or experienced random events. cause of shirt color and the side of the screen on which they
It is important to note that the effect size of the preference for appeared (e.g., cartoons in blue shirts were always on the right side
the lucky in this sample is very similar to the equivalent effect size of the screen, and cartoons in green shirts were always on the left
in the U.S. sample reported in Olson et al. (2006), which used the side of the screen). Cartoon children appeared on the screen one at
same task and a comparable age range (d 1.07 in United States, a time, alternating groups (e.g., first a child in a blue shirt ap-
d 0.93 in Japan). However, the comparison of intentional good peared, followed by a child in a green shirt). As each picture
and intentional bad actors suggests a large difference between the appeared, participants were told one fact about that child. For
Japanese and American samples. Although both groups preferred Group A, three of the five children were described as experiencing
the intentional good to intentional bad targets, the effect size in the lucky events and two were associated with neutral facts. In Group
American sample is more than twice as large as the effect size in B, three of five children were described as experiencing unlucky
the Japanese sample (d 3.04 in United States, d 1.30 in events and two were associated with neutral facts. After the 10
Japan), a result perhaps related to past findings of differences group members had appeared on the screen, two final children
between American and Japanese peoples use of dispositional and appeared, one from each group. These two children, the targets,
situational explanations. In addition, in this sample, participants were identical except for shirt color, and each appeared on the
made no significant distinction between the intentional bad and same side of the screen as had the other members of their group.
unlucky targets and only a marginal distinction between the inten- Participants were asked which of these two targets they liked
tional good and lucky targets. In comparison, American children better. Two unique trials like the ones described above were
made a large distinction between intentional bad and unlucky created, and two additional trials were created substituting inten-
targets and also a marginal distinction between intentional good tional good actions for lucky events and intentional bad actions for
and lucky targets. unlucky events, resulting in four final trials. The lucky group, the
These results suggest that young children in cultures that vary in unlucky group, the intentional good group, and the intentional bad
evaluations of intentional acts nonetheless blame victims of bad group each appeared on the left once and on the right once.
fortune and reward recipients of good fortune in similar ways. Data preparation. Data preparation and analysis was identical
Both show a preference for the lucky. In this result, we have initial to that used in Study 2 of Olson et al (2006). The two lucky versus
evidence of cross-cultural generalizability of the preference for the unlucky items were combined into a composite, and the two
lucky from a country with a culture that provides a meaningful test. intentional good versus bad items were combined into a separate
In our final study, we took this initial result one step further and composite. Each composite was computed by giving the subject
asked whether Japanese children also show evaluative contagion. one point each time they picked the good or lucky actor, resulting
in an index score between 0 (never picked the good or lucky actor)
Study 8: Cross-Cultural Similarity in Evaluative to 2 (always picked the good or lucky actor). Because only three
scores were possible (0, 1, or 2), nonparametric tests were neces-
Contagion
sary. Overall results were analyzed using chi-square goodness-of-
We asked whether Japanese children extend their preference for fit tests (chance was computed to be 25% for 0, 50% for 1, and
the lucky to entire social groups. Children were presented with 25% for 2).
members of two novel groups, one group that contained some
members who had experienced lucky events and one group that Results and Discussion
contained some members who had experienced unlucky events. It
is critical to note that both groups had some members who had Findings. Childrens preferences differed significantly from
experienced neither lucky nor unlucky events (see Levy & Dweck, chance for both the intentional good versus bad comparison, 2(2,
1999, for a similar procedure). Children were then introduced to N 87) 6.45, p .040, and the lucky versus unlucky com-
these new members of each group and were asked to indicate parison, 2(2, N 87) 16.22, p .001 (see Figure 6). Inspec-
which group member they preferred.

Method
Participants. Eighty-seven participants (49 female, aged 4 7
years, M 5.8, SD 1) from rural Japan completed the study.
One other participant completed the study but had to be removed
from the sample because of experimenter error.
Stimuli. An artist drew cartoons of six children, three boys
and three girls. The same six pictures were used to represent
members of each of the groups; the only difference across groups
was the color of their shirts. The lucky and unlucky events were
taken from Study 7, and the neutral items were either described as Figure 6. Proportion of Japanese childrens responses across items in
something the actor liked to eat (e.g., Yuko likes oatmeal) or an which they preferred the new member of the intentional good or intentional
activity in which the actor engaged (e.g., Ayumi rides her bike). bad group in the intentional good versus intentional bad comparisons (left
Procedure. Participants were presented with four trials. On side) and the member of the lucky or unlucky group in the lucky versus
each trial, they were told about members of two groups. The unlucky comparisons (right side) in Study 8.
774 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

tion of the data indicated that children were more likely to prefer erence is innate or it grows readily out of early cognition, perhaps
the member of the intentional good group than the member of the in conjunction with early socialization.
intentional bad group and, consistent with evaluative contagion, The current studies also provided initial evidence that the pref-
also preferred the new member of the lucky group to the new erence for the lucky is not constrained to Western societies by
member of the unlucky group. showing the same tendencies in Japanese school children as in
These results demonstrate that Japanese children evaluate indi- their American counterparts. Although our results are suggestive
viduals on the basis of the actions and experiences of others who of cultural invariance, this preference should be examined in
are socially associated with them. Again, a comparison with the countries that differ from the United States and Japan in meaning-
corresponding U.S. sample is in order. As in Study 7, the effect ful ways, such as in beliefs about or experience with luck to test
sizes for these two samples are nearly identical for evaluations of further for cultural invariance. For example, do children who live
people associated with those experiencing random good and bad in surroundings in which they have very little control over their
events (w .45 in United States, from Olson et al., 2006; w .43 environments and therefore experience unlucky events frequently
in Japan). Also as in Study 7, American children showed three (e.g., children in refugee camps in Sudan) prefer those who expe-
times as large an effect size for intentional good versus bad items rience lucky events to those who experience unlucky events, or
compared with Japanese children (w .83 for United States, w does their own experience attenuate or even reverse this prefer-
.27 for Japan). Once again, we found that Japanese children ence?
showed less of a bias against intentional bad groups than did The value of these results is based both on the empirical dem-
American children, just as they showed less of a bias against onstrations themselves as well as on the theoretical questions they
intentional bad individuals. resolve. Two theories stood out as deserving a test alongside the
In sum, children growing up in Japan, where dispositional phenomena of preference for the lucky and evaluative contagion:
attributions have been observed to be weaker than in Western immanent justice and BJW. Despite a similarity in the structure of
cultures, showed a preference for the lucky over the unlucky as the test of immanent justice and the present studies, the results
well as evaluative contagion in the first cross-cultural tests of these demonstrated a clear dissociation between the two. Whereas im-
phenomena. These results suggest that evaluative contagion is manent justice decreased across age, judgments of the lucky did
generalizable across cultures. Japanese children performed nearly not and, if anything, increased across childhood. In addition, by
identically to American children on this task, preferring members demonstrating a preference for the lucky in very young children,
of predominantly lucky groups to members of predominantly we minimized the likelihood that just-world beliefs, as they have
unlucky groups. been previously described (Lerner, 1977), drive the preference for
the lucky in young children.
In a set of related studies in progress, we are now investigating
General Discussion the hypothesis that the preference for the lucky is not driven by
justice-related concerns at all but, rather, that a simpler mechanism
Across eight studies, we have demonstrated that children show may be responsible for these effects (Olson, Heberlein, Kensinger,
a robust tendency to judge the lucky positively. This preference Spelke, Dweck, & Banaji, 2008). In particular, we are investigat-
was revealed by a variety of methods and is present in children ing the possibility that the affect associated with a good or bad
from a wide range of ethnicities, races, towns, states, countries, event (whether intended or not) rubs off on the individuals expe-
and social classes, including predominantly White middle- to riencing those events, resulting in evaluations of the individuals
upper-middle-class elementary school children in Utah and Mas- that are consistent in valence with the events, a process we call
sachusetts, low-income Black children in Massachusetts, pre- affective tagging. It is important to note that this hypothesis is
school children from a wide range of ethnicities in California, and more parsimonious than many of the justice theories and makes
rural Japanese children. Across these many samples and tasks, some differing predictions. For example, whereas just-world the-
several results emerged clearly. Young children prefer lucky indi- ory predicts that a preference for the lucky should primarily occur
viduals to unlucky ones, children predict that the lucky are more when the events described are extreme and threaten a persons
likely to perform intentional good actions and that the unlucky are sense of justice, the affective-tagging hypothesis predicts that
more likely to perform intentional bad actions, and children extend lucky individuals will always be associated with some positivity
these predictions and evaluations to the siblings and group mem- and unlucky individuals will always be associated with some
bers of lucky and unlucky individuals. negativity (although in some cases, other factors, such as empathy
Another major finding of these studies was that the preference or impression formation, may work in opposition to these evalu-
for the lucky appeared at a very young age. We cannot conclude ations). This prediction is relevant to the current studies because
that children below the age of 3 years do not prefer the lucky, only the items we selected in these studies are trivial events, hardly the
that they did not do so in our task, which may simply have been too events likely to violate ones sense that the world is just. There-
hard for them. Measures such as looking time and reaching be- fore, the fact that we see a preference for the lucky even for these
havior have been used successfully with young infants and even events provides some initial evidence in favor of the affective-
nonhuman primates in other cognitive and social cognitive tasks tagging hypothesis.
(Baillargeon, Spelke, & Wasserman, 1985; Nurock et al., 2007; One may wonder whether children grow out of the preference
Santos & Hauser, 1999), and perhaps creative researchers can for the lucky or, alternatively, whether this preference continues
design studies to test whether children (or even other species) across childhood into adulthood, increasing as the trajectory of the
prefer the lucky. If evidence for this effect in young infants or data in this article might suggest. One could imagine that after the
other primates were found, it would suggest that either this pref- age of 12 years, the developmental trajectory shifts and adoles-
JUDGMENTS OF THE LUCKY 775

cents grow out of this belief. Even if this were the case, a dislike some groups, we may be unwittingly providing our children with
of particular unlucky groups may nonetheless become entrenched the evidence they use to infer that groups inferiority. This means
in childhood and continue into adulthood, long after the mecha- that parents, teachers, and society must not only come to under-
nism that formed them has ceased to operate. Another possibility stand the preferences young children hold but also must under-
is that adults continue to hold these judgments or even increase stand that if they wish to change the impact of these preferences,
them, leading to a continuation of prejudice toward unlucky and society needs to rectify the injustices that cause disadvantage
disadvantaged people and groups. Our research in progress, in and/or develop strategies to counteract young childrens early
which we use a similar paradigm, suggests that these preferences preferences.
seem to continue through adulthood, although they abate consid- Thus, these preferences may be one of the origins of or contrib-
erably; this apparent abatement may be due to adults becoming uting factors to the development of stereotyping, prejudice, and
more reluctant to express the preference publicly. In a simple discrimination, perhaps via the development and maintenance of
replication of Study 1, we found that adults show the same pattern group hierarchies. Such a conclusion is relevant to social psycho-
of believing that lucky targets are more likely to perform good logical discourse on system-justification theory (Jost & Banaji,
actions and that unlucky targets are more likely to perform bad 1994) and social dominance theory (Sidanius & Pratto, 1999).
actions. A preference for the lucky was also found in American Both theories suggest that people are motivated to maintain the
adults in a conceptual replication of Study 7, showing that they status quo in which some social groups have a higher status than
prefer people who experience lucky events to those who experi- others; a preference for the lucky may be one such attitude that
ence unlucky events, even when we used a non-forced-choice contributes to the maintenance of group hierarchies. It is possible
design. that the preference for the lucky is a mechanism for the develop-
As discussed above, this liking of the lucky and disliking of the ment and maintenance of system-justifying and social dominance
unlucky is similar to many related findings that suggest that people beliefs as well as more specific social-group attitudes. We believe
and things are evaluatively tagged on the basis of the valence of this to be a promising avenue of future research.
other information associated with that individual or thing. For
example, research has shown that adults tend to dislike the bearer References
of information with which they disagree, even when the bearer of Baillargeon, R., Spelke, E. S., & Wasserman, S. (1985). Object perma-
the information disagrees with the information being shared (Ma- nence in five-month-old infants. Cognition, 20, 191208.
nis, Cornell, & Moore, 1974), and that adults see an individual as, Banaji, M. R., & Bhaskar, R. (2000). Implicit stereotypes and memory: The
for example, more angry if that individual has described another bounded rationality of social beliefs. In D. L. Schacter & E. Scarry
person as angry (Skowronski, Carlston, Mae, & Crawford, 1998). (Eds.) Memory, brain, and belief (pp. 139 175). Cambridge, MA:
In addition, even novel objects elicit rapid evaluation (Duckworth, Harvard University Press.
Bargh, Garcia, & Chaiken, 2002), and it is therefore not unrea- Bracton, H. (1968 1977). On the laws and customs of England (S. E.
sonable to think that quick evaluations occur when humans ob- Thorne, Trans.). Cambridge, MA: Harvard University Press. (Original
serve other humans, a prediction at the heart of the affective- work published 13th Century) Retrieved August 3, 2006, from Bracton
Online, Harvard Law School Library: http://hlsl5.law.harvard.edu/
tagging hypothesis (Olson et al., 2008).
bracton/
Although the preference for the lucky may seem to be an Callan, M. J., Ellard, J. H., & Nicol, J. E. (2006). The belief in a just world
innocent bias, it is possible that it has important and insidious and immanent justice reasoning in adults. Personality and Social Psy-
repercussions, in particular because those expressing it are young chology Bulletin, 32, 1646 1658.
children. In the real world, random events are, by definition, out of Duckworth, K. L., Bargh, J. A., Garcia, M., & Chaiken, S. (2002). The
the control of the individuals experiencing them, but they are not automatic evaluation of novel stimuli. Psychological Science, 13(6),
completely random in whom they affect. Rather, some groups 513519.
(those who are disadvantaged) tend to experience these types of Elkind, D., & Dabek, R. F. (1977). Personal injury and property damage in
events more than do others. Hurricane Katrina, which hit the the moral judgments of children. Child Development, 48, 518 522.
United States Gulf Coast in August of 2005, stands as a striking Fein, D. (1976). Just world responding in 6- and 9-year-old children.
Developmental Psychology, 12, 79 80.
example of the unequal impact of random events on members of
Fein, D., & Stein, G. M. (1977). Immanent punishment and reward in six-
advantaged and disadvantaged groups. A disproportionate number and nine-year-old children. Journal of Genetic Psychology, 131, 9196.
of those who were stranded in New Orleans were disadvantaged, Furnham, A. (1985). Just world beliefs in an unjust society: A cross
a disproportionate number of those who died were disadvantaged, cultural comparison. European Journal of Social Psychology, 15, 363
and the impact on the lives of those who survived was greater for 366.
the victims who were members of disadvantaged groups. There- Furnham, A. (2003). Belief in a just world: Research progress over the past
fore, what at first appears to be an innocuous beliefthat lucky decade. Personality and Individual Differences, 34, 795 817.
people are better than unlucky peoplemay actually lead to a Furnham, A., & Rajamanickam, R. (1992). The Protestant work ethic and
systematic bias against disadvantaged people and groups, resulting just world beliefs in Great Britain and India. International Journal of
in both inculcation and perpetuation of prejudice in children. Psychology, 27, 401 416.
Gopnik, A., & Astington, J. W. (1988). Childrens understanding of rep-
If it is true that the preference for the lucky and the contagion of
resentational change and its relation to the understanding of false belief
these judgments play a role in the development and maintenance of and the appearancereality distinction. Child Development, 59, 26 37.
prejudice, then this would suggest that to fight prejudice and its Hamlin, J. K., Wynn, K., & Bloom, P. (2007, November 22). Social
development, it is not enough to censor racist remarks, do sensi- evaluation by preverbal infants. Nature, 450, 557560.
tivity training in schools, and read politically correct stories. As Harris, P. L. (1992). From simulation to folk psychology: The case for
long as negative outcomes continue to fall disproportionately on development. Mind & Language, 7, 120 144.
776 OLSON, DUNHAM, DWECK, SPELKE, AND BANAJI

Jahoda, G. (1958). Immanent justice among west African children. Journal Najarian-Svajian, P. H. (1966). The idea of immanent justice among
of Social Psychology, 47, 241248. Lebanese children and adults. Journal of Genetic Psychology, 109,
Johnson, R. C. (1962). A study of childrens moral judgments. Child 57 66.
Development, 33, 327354. Nurock, V., Jacob, P., Margules, S., & Dupoux, E. (2008). A precursor of
Jones, C., & Aronson, E. (1973). Attribution of fault to a rape victim as a moral judgments in infants. Manuscript in preparation.
function of respectability of the victim. Journal of Personality and Olson, K. R., Banaji, M. R., Dweck, C. S., & Spelke, E. S. (2006).
Social Psychology, 26, 415 419. Childrens evaluations of lucky versus unlucky people and their social
Jose, P. E. (1990). Just-world reasoning in childrens immanent justice groups. Psychological Science, 17, 845 846.
judgments. Child Development, 61, 1024 1033. Olson, K. R., Heberlein, A., Kensinger, E., Spelke, E. S., Dweck, C. S., &
Jose, P. E. (1991). Measurement issues in childrens immanent justice Banaji, M. R. (2008). Preferring the lucky to the unlucky, but not
judgments. Merrill-Palmer Quarterly, 37, 601 617. wanting to: An investigation of the role of affective associations in the
Jost, J. T., & Banaji, M. R. (1994). The role of stereotyping in system- preference for the lucky. Manuscript in preparation.
justification and the production of false consciousness. British Journal of Percival, P., & Haviland, J. M. (1978). Consistency and retribution in
Social Psychology, 33, 127. childrens immanent justice decisions. Developmental Psychology, 14,
Kalish, C. W. (2002). Childrens predictions of consistency in peoples 132136.
actions. Cognition, 84, 237265. Piaget, J. (1965). The moral judgment of the child (M. Gabain, Trans.).
Karniol, R. (1980). A conceptual analysis of immanent justice responses in New York: Free Press. (Original work published 1932)
children. Child Development, 51, 118 130. Raman, L., & Winer, G. A. (2004). Evidence of more immanent justice
Krull, D. S., Loy, M. H., Lin, J., Wang, C., Chen, S., & Zhao, X. (1999). responding in adults than children: A challenge to traditional develop-
The fundamental fundamental attribution error: Correspondence bias in mental theories. British Journal of Developmental Psychology, 22, 255
individualist and collectivist cultures. Personality and Social Psychol- 274.
ogy Bulletin, 25, 1208 1219. Santos, L. R., & Hauser, M. D. (1999). How monkeys see the eyes:
Kunda, Z., & Nisbett, R. E. (1986). The psychometrics of everyday life. Cotton-top tamarins reaction to changes in visual attention and action.
Cognitive Psychology, 18, 195224. Animal Cognition, 2, 131139.
Kurdek, L. A. (1979). Childrens coordination of differing cognitive per- Schult, C. A., & Wellman, H. M. (1997). Explaining human movements
spectives. Journal of Genetic Psychology, 135, 279 285. and actions: Childrens understanding of the limits of psychological
Kurdek, L. A., & Rodgon, M. M. (1975). Perceptual, cognitive, and explanation. Cognition, 62, 291324.
affective perspective taking in kindergarten through sixth-grade chil- Shultz, T. R., & Wells, D. (1985). Judging the intentionality of action-
dren. Developmental Psychology, 11, 643 650. outcomes. Developmental Psychology, 21, 83 89.
Lerner, M. J. (1971). Observers evaluation of a victim: Justice, guilt, and Shultz, T. R., Wells, D., & Sarda, M. (1980). Development of the ability to
veridical perception. Journal of Personality and Social Psychology, 20, distinguish intended actions from mistakes, reflexes, and passive move-
127135. ments. British Journal of Social and Clinical Psychology, 19, 301310.
Lerner, M. J. (1974). The justice motive: Equity and parity among Sidanius, J., & Pratto, F. (1999). Social dominance: An intergroup theory
children. Journal of Personality and Social Psychology, 29, 539 550. of hierarchy and oppression. New York: Cambridge University Press.
Lerner, M. J. (1977). The justice motive: Some hypotheses as to its origins Skowronski, J. J., Carlston, D. E., Mae, L., & Crawford, M. T. (1998).
and forms. Journal of Personality, 45, 152. Spontaneous trait transference: Communicators take on the qualities
Lerner, M. J. (1980). The belief in a just world: A fundamental delusion. they describe in others. Journal of Personality and Social Psychology,
New York: Plenum Press. 74, 837 848.
Levy, S. R., & Dweck, C. S. (1999). The impact of childrens static versus Suls, J., & Kalle, R. J. (1979). Childrens moral judgments as a function of
dynamic conceptions of people on stereotype formation. Child Devel- intention, damage, and an actors physical harm. Developmental Psy-
opment, 70, 11631180. chology, 15, 9394.
Long, G. T., & Lerner, M. J. (1974). Deserving, the personal contract, Surber, C. F. (1982). Separable effects of motives, consequences, and
and altruistic behavior by children. Journal of Personality and Social presentation order on childrens moral judgments. Developmental Psy-
Psychology, 29, 551556. chology, 18, 257266.
Manis, M., Cornell, S. D., & Moore, J. C. (1974). Transmission of United Nations. (2003). Rome Statute of the International Criminal Court.
attitude-relevant information through a communication chain. Journal of Retrieved February 22, 2008, from http://untreaty.un.org/cod/icc/
Personality and Social Psychology, 30, 8194. index.html
Masuda, T., & Kitayama, S. (2004). Perceiver-induced constraint and Weisz, J. R. (1980). Developmental change in perceived control: Recog-
attitude attribution in Japan and the US: A case for the cultural depen- nizing noncontingency in the laboratory and perceiving it in the world.
dence of the correspondence bias. Journal of Experimental Social Psy- Developmental Psychology, 16, 385390.
chology, 40, 409 416. Wellman, H. M., Cross, D., & Watson, J. (2001). Meta-analysis of theory-
Miller, J. G. (1984). Culture and the development of everyday social of-mind development: The truth about false belief. Child Development,
explanation. Journal of Personality and Social Psychology, 46, 961 72, 655 684.
978. Woodward, A. L. (1998). Infants selectively encode the goal object of an
Miller, J. G. (1986). Early cross-cultural commonalities in social explana- actors reach. Cognition, 69, 134.
tion. Developmental Psychology, 22, 514 520. Yuill, N., & Perner, J. (1988). Intentionality and knowledge in childrens
Mischel, H. N., & Mischel, W. (1983). The development of childrens judgments of actors responsibility and recipients emotional reaction.
knowledge of self-control strategies. Child Development, 54, 603 619. Developmental Psychology, 34, 358 365.
Montada, L., & Lerner, M. J. (1998). Responses to vicitimizations and
belief in a just world. New York: Plenum Press.
Morris, M. W., & Peng, K. (1994). Culture and cause: American and Received June 21, 2007
Chinese attributions for social and physical events. Journal of Person- Revision received December 15, 2007
ality and Social Psychology, 67, 949 971. Accepted December 16, 2007
Journal of Personality and Social Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 94, No. 5, 777791 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.777

How to Heat Up From the Cold: Examining the Preconditions for


(Unconscious) Mood Effects
Kirsten I. Ruys and Diederik A. Stapel
Tilburg University

What are the necessary preconditions to make people feel good or bad? In this research, the authors aimed
to uncover the bare essentials of mood induction. Several induction techniques exist, and most of these
techniques demand a relatively high amount of cognitive capacity. Moreover, to be effective, most
techniques require conscious awareness. The authors proposed that the common and defining element in
all effective mood induction techniques is the dominating salience of evaluative tone over descriptive
meaning. This evaluative-tone hypothesis was tested in two paradigms in which the evaluative meaning
of the primed concept was more salient than its descriptive meaning (i.e., when subliminal stimulus
exposure was so short that mainly the evaluative meaning was activated [see D. A. Stapel, W. Koomen,
& K. I. Ruys, 2002] and when the primed concepts were sufficiently extreme such that evaluative
meaning always dominated descriptive meaning). Explicit and implicit mood measures showed that the
activation of a dominating evaluative tone affected peoples mood states. Implications of these findings
for theories on unconscious mood induction are discussed.

Keywords: subliminal perception, priming, mood, affect, need for cognition

What makes people feel good or bad? What is needed to put affected only through the intensive and conscious experience (or
someone in a positive or a negative mood? What does it take to recall) of real (or imagined) mood-eliciting stimuli? Are those the
influence peoples affective states? A quick look at the relevant essential ingredients of the mood induction recipe? We think not.
literature does not really suggest a simple answer to these ques- We propose that the common and defining element in all effec-
tions. Past research has shown that there is a myriad of successful tive mood induction techniques is a dominating salience of eval-
techniques to induce positive or negative mood states in people. uative tone.1 Thus, watching a fragment from When Harry Met
Recollecting pleasant or unpleasant memories, listening to uplift- Sally and looking at the local weather report forecasting sunny
ing or depressing music, reading reports of happy or sad events, spells are similarly successful ways to induce a positive mood
watching funny or sad film clips, receiving positive or negative state, even though the descriptive content (i.e., falling in love vs.
performance feedback, imagining wonderful or horrible life predicted hours of sun) of these mood inducers is very different.
eventsall these manipulations can be and have been used suc- What these mood induction methods have in common is their
cessfully to influence how positive or negative individuals feel strong, positive evaluative tone. What differs between these tech-
(Fiedler, 2001; Forgas, 1992; Isen, 1987; Schwarz, 1990). niques is their specific descriptive content. Thus, the crucial in-
But what do these techniques have in common? What makes
gredient for effective mood induction seems to be a strong eval-
them successful mood induction methods? First, it should be noted
uative meaning, rather than specific descriptive meaning.
that all these techniques demand a relatively high amount of
One could argue in more technical terms that successful mood
cognitive capacity. Furthermore, all these techniques require con-
induction techniques cognitively activate (prime) positive or
scious awareness to be effective. It is the conscious recalling,
negative evaluative meaning more strongly than specific descrip-
listening, reading, watching, or imagining that induces a positive
tive information. It is not necessarily the specific descriptive
or negative mood. But are these characteristics really necessary to
elicit an affective state in an individual? Can ones mood be content of the memories one recalls, the movie one watches, or the
music one listens to that makes one feel good or bad. After all, a
memory, movie, or piece of music cannot affect ones mood.
Specific content does not matter: It is the global, diffuse, nonspe-
Kirsten I. Ruys and Diederik A. Stapel,Tilburg Institute for Behavioral cific, overall evaluative tone that is primed while one is recalling
Economics Research (TIBER), Department of Social Psychology, Tilburg memories that produces mood effects. Taking this notion to its
University, Tilburg, the Netherlands. extreme, one could argue that whenever the evaluative features of
This research was supported by a Pionier grant from the Netherlands cognitively activated (primed) information strongly dominate
Organization for Scientific Research awarded to Diederik A. Stapel. We the descriptive meaning, mood states are likely to be affected.
thank Katie Lancaster for her insightful comments on an earlier version of
When the evaluative meaning of primed information dominates the
this article.
Correspondence concerning this article should be addressed to Kirsten I.
descriptive information, then mood states are likely to be affected.
Ruys, Tilburg Institute for Behavioral Economics Research (TIBER),
1
Department of Social Psychology, Tilburg University, P.O. Box 90153, See Hugenberg (2005) and Stapel and Koomen (2005) for a similar use
Tilburg 5000 LE, the Netherlands. E-mail: k.i.ruys@uvt.nl of the term evaluative tone.

777
778 RUYS AND STAPEL

All that is needed to induce a mood state is the dominating activated, the evaluative tone is no longer diffuse because in that
activation of diffuse, nondescriptive, evaluative information. In case it becomes bound to the descriptive information.
short, all you need is evaluative tone. Thus, although intense, long, Although one can use this all-you-need-is-evaluative-tone logic
conscious exposure to or experience of mood-relevant stimuli may to explain the variable successfulness of hot versus cold priming
be a sufficient precondition for the production of mood effects, it techniques to produce actual changes in mood states, one should
is not a necessary precondition. not take this to mean that it is impossible for cold priming tech-
niques to produce mood effects. Rather, our logic suggests that it
A Cold Recipe for Hot Mood Effects should be possible for cold concept priming to produce mood
effects, given that the evaluative side of a stimulus or concept can
A highly relevant question for defining the essential ingredients be primed without activating its descriptive features. In other
for producing mood effects is whether it is necessary to expose words, when exposure to cold concepts activates information that
people to hot emotional material like music and movies or whether is cognitively unconstrained (i.e., without descriptive meaning,
it is possible to elicit mood states by exposing people to cold Clore & Colcombe, 2003), mood effects should be possible. Thus,
semantic concepts. Some mood effects are easy to explain in terms when priming honest versus dishonest activates merely or mainly
of the activation of declarative knowledge (i.e., semantic priming). the cognitively unconstrained, evaluative meaning of these words
For instance, the retrieval of a positive experience can encourage (positive vs. negative) rather than their evaluative descriptive
the recollection of other positive autobiographical memories by meaning (friendly vs. aggressive), mood effects should occur.
semantic association. However, content-free mood effects such as The question is then, under what circumstances, does stimulus
the influence of Barbers Adagio for Strings on peoples pro- exposure result in a relatively strong activation of evaluative
cessing style are more difficult to explain without the assumption (rather than evaluative descriptive) information? Affective pri-
of a hot mood state. macy theory (Zajonc, 1980) provides a possible answer to this
To investigate the influence of hot states versus cold concepts, question because it holds that when people are exposed to a
Innes-Ker and Niedenthal (2002) directly compared the effects of stimulus, affective reactions (i.e., reactions based on an evaluation
hot emotional states (I feel happy vs. I feel sad) versus cold of the stimulus) occur prior to nonaffective reactions (i.e., reac-
emotion concepts (happy vs. sad) on subsequent social judgments. tions based on descriptive stimulus features). This theory has
In one study, those researchers showed that priming cold emotion received support from neurological research showing that indepen-
concepts increased the cognitive accessibility of congruent emo- dent systems exist for coarse evaluative processing and detailed
tion concepts but had no impact on the emotional state of the perceptual processing (e.g., Adolphs, 2003; LeDoux, 1989;
participants. Another study showed that priming an emotional state Zajonc, 2000). The primacy of affective processing was recently
(induced by music) influenced self-reported feelings and produced corroborated by researchers studying event-related brain potentials
emotion-congruent judgments of an ambiguous target person in response to emotional faces (Palermo & Rhodes, 2007). This
whose feelings could either be interpreted in terms of happiness or work shows that crude affective categorization can often occur
sadness. The priming of emotional concepts had no such effects. rapidly, whereas fine-grained processes necessary for recognition
Innes-Ker and Niedenthal (2002) took their results to mean that of the identity of a face or for discriminatation between basic
the mere activation of cold concepts is not sufficient to produce hot emotional expressions typically need more time.
emotional states in the perceiver and that the presence of a hot This suggests, as Stapel et al. (2002) have recently shown that
emotional state is necessary to produce emotion-congruent judg- even in the realm of subliminal perception, it is indeed possible to
ments (see also Maringer & Stapel, 2007). Such an interpretation separate evaluation-based and description-based reactions to stim-
of their results is definitely warranted as well as intuitively ap- uli (see also Ruys & Stapel, in pressa, in press b). Stapel and
pealing. It makes sense that mere priming of emotional concepts colleagues (2002) demonstrated that when a picture of a happy
like happy and uplifting versus sad and melancholy is less female face is primed subliminally, evaluative reactions (posi-
likely to affect peoples mood states than listening to happy, tive) are typically triggered earlier than descriptive reactions
uplifting music or sad, depressing music. (female), but neither type of reaction needs awareness to occur.
The question remains, however, what exactly it is then that Similarly, when primed with the words honest versus dishonest,
makes this hot mood induction procedure, in fact, hot? Applying people pick up the evaluative meaning of these words (positive vs.
the present evaluative-tone perspective, we propose that it is the negative) prior to their descriptive meaning (honest vs. dishonest;
global evaluative tone of information activated through so-called see also Bargh, Litt, Pratto, & Spielman, 1989; Stapel & Koomen,
hot induction procedures versus activation of specific descriptive 2005). Thus, both evaluative and descriptive stimulus cues can be
meaning of so-called cold induction procedures that makes the detected without awareness, but evaluative cues are often detected
difference. We thus argue that movies or music are often more earlier (Stapel, 2003).
effectively used to induce mood effects because, with these tech- The notion that descriptive meaning is picked up later than
niques, global nonspecific evaluative information is more strongly evaluative meaning has important consequences for the question of
activated than its concrete specific descriptive counterpart. Prim- whether (unconsciously) primed cold concepts such as happy,
ing specific emotion concepts is probably less successful because friendly, and honest or sad, aggressive, and dishonest
it is likely that the descriptive meaning of the primed concepts will can affect mood states. As Zajoncs (1980) affective primacy
overshadow their evaluative tone. Thus, we argue, when evaluative hypothesis suggests and recent work on unconscious affect prim-
as well as descriptive information is activated, mood effects are ing (Ruys & Stapel, in pressa, in press b; Stapel et al., 2002;
less likely than when merely or mainly evaluative information is Stapel & Koomen, 2005, 2006) demonstrates, when one is sub-
activated. When evaluative as well as descriptive information is liminally priming concepts (e.g., honest vs. dishonest), exposure
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 779

time may determine what type of information is actually activated. positive or negative affective state is the conscious experience of
At very short exposures, the evaluative meaning or tone of these positive or negative feelings (e.g., Clore, Storbeck, Robinson, &
concepts is activated (positive vs. negative). At longer exposures, Centerbar, 2005). Thus, in Study 1a and Study 1b, we started by
however, both the descriptive and the evaluative meaning may examining explicit mood judgments and asking people how posi-
become available (e.g., honest vs. dishonest). This implies that not tive or negative their mood was at the current moment. In Studies
only subliminally presented, emotionally charged primes (e.g., 2-4, we turned to more indirect measures of mood, in addition to
happy faces vs. sad faces) but also less emotional, colder primes our explicit mood measure. A well-known indirect consequence of
(e.g., honest vs. dishonest) may affect mood judgments, given that moods is that they may influence peoples processing styles: When
these primes are flashed sufficiently quickly to activate mainly people feel good, they are more likely to rely on heuristic, easy,
their evaluative tone (see Stapel et al., 2002). and global processing strategies, whereas when people feel bad,
In sum then, cold concept priming may produce mood effects they tend to use more demanding, systematic, and local processing
when the evaluative meaning of these concepts is more salient and strategies (Fiedler, 1990, 1991; Forgas, 1995; Gaspar & Clore,
more strongly activated than their descriptive meaning. Zajoncs 2002). Thus, in the present research, we also focused on peoples
(1980) affective primacy theory suggests that one way to achieve processing styles as a measure of peoples mood states.
this is by flashing concepts sufficiently quickly such that mainly
their evaluative tone is activated (see Stapel & Koomen, 2005;
Study 1
Stapel et al., 2002). The increased activation of information that is
merely or mainly evaluative and thus diffuse and cognitively Study 1a
unbounded may then spill over to peoples moods (Zajonc, 1980,
2000). In the words of Clore and Colcombe (2003), [R]epeated We started the investigation of our all-you-need-is-evaluative-
suboptimal presentation of positive or negative stimuli activates tone hypothesis with a study of the impact of subliminally primed
evaluative meaning. Being objectless, it may become attached to trait concepts on explicit mood judgments. We expected that when
whatever comes to mind next (p. 343). Thus, objectless evaluative prime exposures were short, explicit mood judgments might be
meaning may also become attached to a persons mood. On a more affected by the evaluative tone of the primes. However, when
general level, Bargh (1997, 2006) has noted and demonstrated prime exposures were long, explicit mood judgments might not be
repeatedly that the cognitive activation of semantic concepts is affected because the activation of descriptive meaning could over-
likely to simultaneously impact all kinds of psychological systems. shadow the evaluative tone. We measured mood immediately after
Priming aggressive or old may influence perception, motiva- the priming episode by asking people how positive or negative
tion, behavior, and evaluation. Interaction and exchange between their mood was at that moment.
the different psychological systems (Treisman, 1996) may further
explain why the cognitive activation of evaluatively toned infor-
Method
mation may affect and color a persons mood state.
Participants, design, priming stimuli, and measures. Partici-
The Present Studies pants (N 98) were undergraduates who took part in exchange for
partial course credit. The participants were randomly assigned to
In the present studies, we tested our all-you-need-is-evaluative- the conditions of a 2 (prime exposure: long or short) 2 (prime
tone hypothesis, using the same subliminal priming paradigm that valence: positive or negative) between-participant design or to a
we have used in prior work (Stapel & Koomen, 2005, 2006; Stapel control condition, in which participants were subliminally primed
et al., 2002). As these recent studies on the affective primacy with neutral traits.
hypothesis have suggested, nondescriptive, nonspecific evaluative Overview. Upon arrival, participants were shown into one of
stimulus features are especially likely to be picked up when eight cubicles in the experimental room and seated in front of a
stimulus exposure is very short. Our logic thus suggestsperhaps computer. They were then told that they would be involved in a
somewhat counterintuitivelythat mood effects from cold con- series of unrelated studies. First, participants performed a parafo-
cept priming are especially likely when priming occurs uncon- veal vigilance task (modeled after a task used by Stapel et al.,
sciously. However, the aim of the present research was not only to 2002) in which trait concepts were presented outside of partici-
show that subliminally presented information may produce mood pants awareness. Participants were told that very short flashes
effects but also to test some of the boundary conditions of these would appear on the screen at unpredictable places and times and
effects. In accordance with Clore and Colcombe (2003) and Zajonc that their task was to decide as quickly and accurately as possible
(1980; see also Stapel et al., 2002), we hypothesized that when whether the flash appeared on the left or right side of the screen.
priming activates information that has a strong evaluative tone and After having completed the vigilance task, participants were
is cognitively unconstrained, peoples moods may be affected. thanked for their participation and given the next task. The exper-
Because of this influence, even cold concept priming may induce imenter told participants, A colleague of mine, from another
mood effects, as long as the evaluative tone is sufficiently strong university, would like you to complete this simple questionnaire.
and salient. Participants were then given a one-page one-question (Rate how
We investigated our evaluative-tone hypothesis with a variety of positive or negative your mood is at this moment) questionnaire
mood measures: We examined several indicators of peoples ex- that measured mood; they responded using a scale ranging from 1
plicit mood judgments and information-processing styles. A mul- (negative) to 7 ( positive). Next, participants received a funnel
titude of mood measures is crucial to revealing the bare essentials debriefing procedure, in which they were probed for awareness of
of mood induction. However, central for the occurrence of a the priming stimuli, awareness of the influence of the priming task
780 RUYS AND STAPEL

on later judgments, and general suspicion concerning the goal of conditions (Fs 1). Finally, there were no participants who
the study (see Stapel et al., 2002). Finally, participants were thought the vigilance and evaluation tasks were related. Thus, we
thanked and debriefed. can safely conclude that we were successful in presenting our
Priming. The priming task was modeled after Stapel et al.s priming stimuli outside of awareness and in not alerting partici-
(2002) parafoveal priming task. Once participants were seated in pants to the actual relation between the vigilance and judgment
front of their computer, the experimenter explained the vigilance tasks. This was also true for the other studies presented here, in
task. Participants were seated so that the distance between their which we used the same paradigm as in the present study.
eyes and the computer screen was 80 100 cm. This distance
ensured that the priming stimuli were presented outside of partic- Results
ipants perceptual field. The experimenter then instructed partici-
pants to place their index fingers on two keys of the keyboard and A Prime Valence Prime Exposure analysis of variance
to press the left key, labeled L, if a flash appeared on the left side (ANOVA) on the mood measurement revealed the predicted in-
of the screen and the right key, labeled R, if a flash appeared on teraction, F(1, 76) 3.94, p .05, 2 .05. The two main effects
the right side of the screen. A fixation point consisting of one X did not reach significance (Fs 1). As can be seen in Table 1, the
was presented continually in the center of the screen. Participants interaction effect reflects that, as expected, in the short exposure
were given 10 practice trials to become familiar with the procedure conditions, participants in the positive priming condition reported
and to ensure that they understood it. After answering any ques- feeling more positive (M 5.60, SD 1.05) than participants in
tions, the experimenter began the 60 experimental trials of the the negative priming condition (M 4.90, SD 1.02), F(1, 76)
vigilance task, which took participants approximately 10 min to 4.43, p .05, 2 .06, whereas in the long exposure conditions,
complete. priming had no effect on experienced affect (F 1). In those
Priming stimuli were trait concepts that were printed in black conditions, participants mood judgments were similar to those in
Times New Roman letters (12 point) printed on a white screen. the control condition (M 5.28, SD 1.13).
The words that were flashed in the 10 practice trials and in 40 of In addition, to provide an overall test of the predicted pattern of
the experimental trials were neutral words (e.g., table, chair, results including the control condition, we performed a contrast
tree). In the remaining 20 experimental trials, in the positive analysis. On the basis of the predictions, we assigned weights of 1
priming conditions, the following words were each flashed five to the cells that we expected not to differ (the long positive, long
times: confident, persistent, honest, pleasant. In the neg- negative, and control conditions), a weight of 4 to the cell in which
ative priming conditions, the following words were each flashed we expected the most positive mood (the short positive condition),
five times: arrogant, stubborn, dishonest, wrong. The or- and a weight of 7 to the cell in which we expected the most
der in which these words were flashed was random. In the long negative mood (the short negative condition). It should be noted
conditions, words were flashed for 120 ms. In the short conditions, that this a priori contrast reached significance, t(93) 2.01, p
words were flashed for 40 ms. In all conditions, these words were .05. We also performed two contrast analyses to directly compare
immediately followed by a 120-ms mask (for details, see Stapel et the short exposure conditions with the control condition. Unfortu-
al., 2002). nately, neither contrast one comparing positive with control,
Awareness and suspicion. Previous subliminal priming studies t(93) 0.91, p .37, and one comparing negative with control,
have shown that the paradigm used here provides sufficient safe- t(93) 1.08, p .29 reached significance. To test whether the
guards to prevent participants from becoming aware of the priming conditions in which we did not expect mood changes to occur (the
stimuli (see Chartrand & Bargh, 1996; Erdley & DAgostino, long exposure conditions) differed from the control condition, we
1988; Stapel et al., 2002). However, to ensure that participants performed two additional contrast analyses. As expected, neither
were not aware of the priming stimuli, we used an extensive funnel contrast one comparing positive with control, t(93) 0.48, p
debriefing procedure in which participants were asked increasingly .64, and one comparing negative with control, t(93) 0.21, p
specific questions about the study (see Stapel et al., 2002). Partic- .84 was significant.
ipants were asked what they thought the purpose of the study had The results of Study 1a showed that short subliminal exposures
been, whether they thought any of the tasks they had performed to valenced concepts are more likely to influence peoples mood
had been related, whether they thought their performance on one
task might have affected their performance on a next task, whether Table 1
anything about the study seemed strange or suspicious to them, and Means and Standard Deviations for Mood Judgments as a
what they thought the content of the flashes had been during the Function of Prime Exposure and Prime Valence (Study 1a)
task. If participants indicated knowledge that the flashes consisted
of words, they were further probed for general or specific meaning Prime exposure
of these words. Next, in several multiple-choice trials, participants
were given the priming stimuli used in this experiment (the posi- Long Short
tive words or the negative words) and were told that at some of the Prime valence M SD M SD
trials, one of those words was flashed. Participants were then asked
to choose (guess) which word was flashed. All participants re- Positive 5.10 1.17 5.60 1.05
ported that they had seen flashes. Although some reported seeing Negative 5.35 1.04 4.90 1.02
words, no participant could report on the contents of the primes. Note. Scale range is from 1 to 7. Higher scores indicate more positive
Furthermore, participants guesses of which of the two words they judgments. Mean mood judgment in control condition was 5.28 (SD
had seen did not exceed chance, nor did they differ between 1.13).
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 781

states than long subliminal exposures to these concepts. This Valence Prime Exposure ANOVA on this measure revealed the
finding supported our hypothesis that a dominating positive or predicted interaction, F(1, 45) 4.21, p .05, 2 .09. The two
negative evaluative tone is sufficient to elicit a corresponding main effects did not reach significance ( ps .17). As can be seen
mood state. However, compared with our control condition, the in Table 2, the interaction effect reflects that, as expected, in the
obtained mood effects were not very strong. Therefore, we con- short exposure conditions, participants in the positive condition
ducted Study 1b. reported feeling more positive (M 7.00, SD 0.74) than
participants in the negative condition (M 6.15, SD 0.98), F(1,
Study 1b 47) 6.18, p .05, 2 .12, whereas in the long exposure
conditions, priming had no effect on experienced affect (F 1). In
This study replicated Study 1a with a more extensive explicit those conditions, participants mood judgments were similar to
mood measure, the Brief Mood Inspection Scale (BMIS; Mayer & those of in the control condition (M 6.61, SD 1.01).
Gaschke, 1988). In addition, we investigated whether our mood We performed a contrast analysis to provide an overall test of
effects could be explained in terms of response mode effects. the predicted pattern of results. Using the same weights as in Study
Therefore, participants also rated themselves and a good friend on 1a, we found that the a priori contrast was significant, t(60)
specific trait dimensions. We expected that in contrast to the mood 2.13, p .05. In addition, we performed two contrast analyses to
judgments, these self and other trait ratings would be unaffected by directly compare the short exposure conditions to the control
the subliminally presented trait terms because the judgments condition. However, both contrasts one comparing positive with
would be relatively specific and descriptive (see Keltner, Locke, & control, t(60) 1.11, p .27, and one comparing negative with
Audrain, 1993). control, t(60) 1.36, p .18 did not reach significance. We
also conducted two additional contrasts, testing the control condi-
Method tion against the long exposure conditions. As predicted these
contrasts one comparing positive with control, t(60) .56, p
Participants (N 65) were undergraduates who took part in .58, and one comparing negative with control, t(60) 0.07, p
exchange for partial course credit. The participants were randomly .94 were not significant.
assigned to the conditions of a 2 (prime exposure: long or short) Together, the findings of Study 1a and Study 1b support the idea
2 (prime valence: positive or negative) between-participant design that a dominating positive or negative evaluative tone may elicit a
or to a control condition, in which participants were subliminally corresponding mood state. Specifically, subliminal exposure to
primed with neutral traits. valenced (trait) information is most likely to affect mood judg-
The procedure was similar to that used in Study 1a. However, ments when prime exposures are sufficiently short to activate
different dependent measures were used. Immediately after the evaluative reactions that have no specific descriptive content.
priming procedure, participants completed a self-report measure of Thus, these two studies suggest that not all subliminal priming
emotional state that consisted of selected items from the BMIS. effects are created equal. They show that sometimes longer primes
This scale listed nine feeling states: five positive states (happy, have less impact. The longer one is exposed to evaluatively toned
content, preppy, lively, active) and four negative states trait information, the less likely it is that such traits may affect
(sad, gloomy, tired, drowsy). Participants indicated how ones mood. The results of Study 1b suggest that such carryover
much they were feeling each state using a scale ranging from 1 effects are most likely to occur on relatively general and evaluative
(definitely do not feel) to 9 (definitely do feel). After participants mood judgments but not on relatively specific descriptive self-
had completed this mood inspection scale, we asked them to rate judgments and other-person judgments. This suggests that sublim-
themselves and a good (same sex) friend on the rating dimen- inally activated affective information is most likely to spill over
sions of friendly, smart, physically attractive, and athletic, into conscious judgment when the target of this judgment is
using a scale ranging from 1 (not at all applicable) to 7 (very evaluatively ambiguous (see also Stapel et al., 2002).
applicable). The order of the mood inspection scale on the one In sum, Study 1a and Study 1b provide the first evidence for our
hand and the self and good friend ratings on the other hand were all-you-need-is-evaluative-tone hypothesis. We showed that eval-
counterbalanced to control for possible order effects.
Table 2
Results and Discussion Means and Standard Deviations of Scores for Items on Brief
Mood Inspection Scale as a Function of Prime Exposure and
ANOVAs showed no main or interaction effects of the order of Prime Valence (Study 1b)
measures variable on any of the dependent measures (Fs 1).
ANOVAs also showed that there were no main or interaction Prime exposure
effects of the primed information on participants self-ratings or
ratings of participants friends (Fs 1), indicating that self-ratings Long Short
or other-ratings on specific trait dimensions (friendly, smart, phys- Prime valence M SD M SD
ically attractive, athletic) were not affected by subliminally primed
trait concepts, independent of whether prime exposure time was Positive 6.42 0.90 7.00 0.74
relatively short or long. Negative 6.58 0.79 6.15 0.98
Reliability analyses of the nine items on the mood inspection Note. Scale range is from 1 to 9. Higher scores indicate more positive
scale were conducted (after reverse scoring the four negative judgments. Mean mood judgment in control condition was 6.61 (SD
items) to form a composite scale (Cronbachs .82). A Prime 1.01).
782 RUYS AND STAPEL

uative tone was critical in influencing peoples explicit mood expected that peoples need for cognition could depend on situa-
judgments. However, we needed additional evidence for several tional factors. We proposed that analogous to the effect of mood on
reasons: First, the mood effects were not very strong (compared peoples information-processing styles (i.e., more systematic in-
with control conditions) in either study. Second, to increase the formation processing in a negative mood and more heuristic pro-
generalizability of our hypothesis, we also needed to demonstrate cessing in a positive mood), people may experience a higher need
these effects on indirect mood measures. Third, our mood effects for cognition when they feel bad than when they feel good. Thus,
could be explained in terms of semantic priming. The exposure to need for cognition may indirectly reflect peoples mood states.
positive (or negative) concepts during the priming episode could In Study 2, we used need for cognition to show that a dominat-
have activated other related positive (or negative) concepts in ing evaluative tone is also essential in evoking effects on indirect
memory, for instance, concepts representing positive (or negative) mood measures. Before turning to this main objective, we con-
affective states. This might have increased the tendency for par- ducted a pretest to demonstrate that a well-known conscious mood
ticipants to agree with experiencing these positive (or negative) induction technique (recalling positive or negative life events)
states. To address these three issues, we performed Study 2. indeed affects individuals need for cognition. The aim of this
pretest was thus to show that peoples mood states affect their need
Study 2 for cognition.

An interesting consequence of moods is their impact on peoples


processing styles. Several studies have shown that people who feel
Pretest
good tend to rely on heuristic, easy, and global processing strate- Participants, Design, Mood Induction, and Measures
gies, whereas people who feel bad are more likely to use demand-
ing, systematic, and local processing strategies (Fiedler, 1990, Participants (N 38) were undergraduates who took part for
1991; Forgas, 1995; Gaspar & Clore, 2002). Thus, for example, partial course credit. The participants were randomly assigned to
people dining out in a restaurant tend to choose the surprise of the the conditions of a three-factor (mood induction: positive, nega-
chef when they are in a good mood but tend to scrutinize the tive, or neutral) between-participant design.
menu and analyze the ingredients of each course in detail before Participants received a booklet consisting of the mood induc-
making a choice when they are in a bad mood. Fiedler (2001) has tion, the need-for-cognition items, and a mood question. We asked
aptly explained the differences between the two processing styles them, dependent on the mood induction condition, to remember a
in his adaptive learning viewpoint, which assumes that the pro- positive, a negative, or a neutral event from the past and to try to
cessing styles associated with positive and negative moods have relive this experience (see, for instance, Bless et al., 1996; Fiedler
respectively evolved in appetitive and aversive situations to cope & Stroehm, 1986, who have successfully used a similar procedure
most effectively with the demands of the situation.2 A friendly, to induce mood). Then, participants completed the following four
appetitive situation demands exploratory and knowledge-driven need-for-cognition items (selected from the Need for Cognition
processing. A threatening, aversive situation requires careful and Scale, Cacioppo & Petty, 1982), using a scale ranging from 1
more systematic processing strategies. (completely disagree) to 5 (completely agree): The idea of relying
Although semantic priming could serve as alternative explana- on thought to make my way to the top appeals to me; I would
tion for the results in Study 1, semantic priming cannot easily prefer a task that is intellectual, difficult, and important to one that
explain mood effects on peoples information-processing styles. is somewhat important but does not require much thought;
For this reason, we mainly focused on mood effects related to Thinking is not my idea of fun (reverse coded); and Learning
information processing to advance our evaluative-tone hypothesis. new ways to think doesnt excite me very much (reverse coded).
In Study 2, we explored and further tested the effects of mood on We selected four representative items because of time concerns.
peoples need for cognition. Previous research has assumed that Next, participants received the same mood question as in Study 1a.
individuals high in need for cognition naturally tend to seek,
acquire, think about, and reflect back on information to make sense
of stimuli, relationships, and events in their world; individuals low Results
in need for cognition, in contrast, are more likely to rely on others A mood induction ANOVA performed on the mood question
(e.g., experts), cognitive heuristics, or social comparisons to pro- demonstrated that our mood induction method was indeed suc-
vide this structure (Cacioppo, Petty, Feinstein, & Jarvis, 1996, p. cessful, F(2, 57) 6.33, p .01, 2 .27. A further contrast
243). analysis showed that participants who remembered and relived a
Although need for cognition has often been shown to be a stable positive life event indicated that they felt more positive (M 5.00,
personality trait that one can measure reliably with a personality SD 0.89) than participants who remembered and relived a
scale (Cacioppo & Petty, 1981; Cacioppo et al., 1996), there are no negative event (M 3.46, SD 1.20), with mood rating of the
indications that need for cognition is also (at least to a certain
extent) context dependent. The features that characterize how 2
Note that although the typical finding is that people in a good mood
people process information as a function of their need for cogni-
rely more on heuristic, global processing strategies and people in a bad
tion remind us of the two processing styles between which, ac- mood rely more on systematic, detailed ways of processing, researchers
cording to most dual-process models, people alternate depending also have reported more complex findings. For example, mood can have
on the situation: a systematic, effortful information-processing motivational consequences because of mood management pressures during
style and a heuristic, easy information-processing style (e.g., positive, negative, and neutral mood states (Gervey, Igou, & Trope, 2005;
Chaiken, 1980; Petty & Caccioppo, 1986). In a similar vein, we Isen, 1987; Wegener & Petty, 1994).
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 783

participants in the neutral condition lying between these two Table 3


extremes (M 4.43, SD 1.09), t(35) 3.53, p .05. Means and Standard Deviations of Scores for Items on Need for
Reliability analyses of the four need-for-cognition items were Cognition Scale and for Mood Judgments as a Function of
conducted to form a composite scale (Cronbachs .87). A Prime Exposure and Prime Valence (Study 2)
mood induction ANOVA on this measure revealed the predicted
main effect, F(2, 57) 4.62, p .05, 2 .21. A contrast Prime exposure
analysis showed that, as expected, in the negative mood condition, Long Short
participants reported a higher need for cognition (M 4.23, SD
0.97) than participants from the positive mood condition (M Prime valence M SD M SD
3.21, SD 0.99), with the neutral condition lying between these
Need for cognition score
two extremes (M 3.59, SD 0.53), t(35) 3.02, p .05).
Further analyses showed that the partial correlation (controlling for Positive 3.46 0.59 3.13 0.49
experimental condition) for these two dependent measures was Negative 3.57 1.00 4.27 1.07
high, r .80 ( p .01).
Mood judgment

Method Positive 4.85 0.90 5.42 0.79


Negative 4.73 1.01 3.83 0.94
Participants (N 60) were undergraduates who took part in the
study for partial course credit. The participants were randomly Note. The Need for Cognition Scale ranges from 1 to 5. Higher scores
indicate a higher need for cognition. Mean need for cognition score in
assigned to the conditions of a 2 (prime exposure: long or short) control condition was 3.63 (SD 0.53). The mood judgment scale ranges
2 (prime valence: positive or negative) between-participant design from 1 to 7. Higher scores indicate more positive judgments. Mean mood
or to a control condition, in which participants were subliminally judgment in control condition was 4.67 (SD 0.78).
primed with neutral traits.
The procedure was similar to that used in Study 1a. However,
different dependent measures were used. Immediately after the trasts one comparing positive with control, t(55) 0.53, p
priming procedure, participants completed the four need-for- .60, and one comparing negative with control, t(55) 0.18, p
cognition items that were used in the pretest. We again included a .86 were not significant.
mood question to determine whether our mood induction worked. A Prime Valence Prime Exposure ANOVA on the mood
judgments also revealed the predicted interaction, F(1, 44) 7.74,
Results and Discussion p .01, 2 .15, and a main effect of prime valence, F(1, 44)
10.45, p .01, 2 .19. There was no effect of prime exposure
A Prime Valence Prime Exposure ANOVA on the need-for- (F 1). The means and standard deviations are depicted in Table
cognition composite scale (Cronbachs .85) revealed the 3. Equivalent to the pattern of results for need for cognition, the
predicted interaction, F(1, 44) 4.80, p .05, 2 .10, and a interaction effect indicated that for the short exposure conditions,
main effect of prime valence, F(1, 44) 6.97, p .05, 2 .14. participants in the positive condition reported feeling more posi-
There was no effect of prime exposure (F 1). As can be seen in tive (M 5.42, SD 0.79) than participants in the negative
Table 3, the interaction effect reflects that, as expected, in the short condition (M 3.83, SD 0.94), F(1, 46) 18.76, p .01, 2
exposure conditions, participants in the positive condition reported .29, whereas for the long exposure conditions, priming had no
a lower need for cognition (M 3.13, SD 0.49) than partici- effect on experienced affect (F 1). Again in those conditions,
pants in the negative condition (M 4.27, SD 1.07), F(1, 46) participants mood judgments were similar (Fs 1) to those in the
12.04, p .01, 2 .21, whereas in the long exposure conditions, control condition (M 4.67, SD 0.78). To provide an overall
priming had no effect on reported need for cognition (F 1). In test of the predicted pattern of results, we performed a contrast
those conditions, participants need for cognition judgments were analysis (with the same weights as in Study 1a) that was signifi-
similar (Fs 1) to those in the control condition (M 3.63, SD cant, t(55) 4.22, p .05. Contrast analyses comparing the
0.53). To provide an overall test of the predicted pattern of results, control condition with the short positive condition, t(55) 2.08,
we additionally performed a contrast analysis. On the basis of the p .05, and comparing the control condition with the short
predictions, we assigned weights of 1 to the cells that we expected negative condition, t(55) 2.31, p .05, were also significant.
not to differ (the long positive, long negative, and control condi- Additional contrast analyses showed that in line with our hypoth-
tions), a weight of 4 to the cell in which we expected the highest eses, neither the long positive condition, t(55) 0.51, p .62, nor
need for cognition (the short negative condition), and a weight of the long negative condition, t(55) 0.16, p .87, differed from
7 to the cell in which we expected the lowest need for cognition the control condition.
(the short positive condition). This a priori contrast was highly Further analyses showed that the partial correlation (controlling
significant, t(55) 3.58, p .05. In addition, we performed two for experimental condition) for our two dependent measures was
contrast analyses to directly compare the short exposure conditions high, r .62 ( p .01). This allowed us to test the robustness of
with the control condition. The contrast comparing positive with our findings by computing a composite scale of the z-transformed
control reached significance, t(55) 2.40, p .05, whereas the need-for-cognition scores and the recoded and then z-transformed
contrast comparing negative with control was marginal, t(55) mood judgments. We performed the same contrast analyses as
1.87, p .08. We also tested whether the long exposure conditions described earlier on our composite scale. A contrast analysis
differed from the control condition. As predicted, these two con- testing the expected overall pattern of results was highly signifi-
784 RUYS AND STAPEL

cant, t(55) 4.24, p .05. Contrast analyses performed to under relatively long but subliminal exposure. Prime exposure
compare the short positive conditions with control and the short duration was the same in all conditions and was similar to the long
negative conditions with control were also significant, t(55) exposure conditions of Study 1a, 1b, and 2. In contrast to these
2.17, p .05, and t(55) 2.58, p .05, respectively. Again as previous studies, we expected that the relatively longer exposures
expected, additional contrast analyses showed that neither the long to trait concepts might affect peoples moods but only when the
positive condition, t(55) 0.62, p .54, nor the long negative evaluative tone was sufficiently dominant. Therefore, this priming
condition, t(55) 0.20, p .84, differed from the control condi- should only work with extreme traits.
tion.
The results thus show that when prime exposures are short, the Method
evaluative tone of the primes affects peoples reported need for
cognition, whereas when prime exposures are long, the evaluative Participants (N 57) were undergraduates who took part for
tone of the primes does not affect peoples reported need for partial course credit. The participants were randomly assigned to
cognition. The same pattern of results was obtained on the mood the conditions of a 2 (prime extremity: moderate or extreme) 2
question. These findings indicate that need for cognition is a stable (prime valence: positive or negative) between-participant design or
personality trait that also depends on situational factors like mood. to a control condition in which participants were subliminally
More important for the present purposes, people may experience a primed with neutral traits.
higher need for cognition when they feel bad than when they feel The priming stimuli presented in the experimental trials were
good. This finding is equivalent to mood effects on peoples trait concepts with a moderate or extreme valence, taken from (and
information-processing styles. pretested by) Stapel and Koomen (2000). In the moderately pos-
In sum, Study 2 provides support for our evaluative-tone hy- itive priming condition, the following words were each flashed on
pothesis on an indirect mood measure. Thus far, we have demon- the computer screen five times: thrifty, reasonable, agree-
strated mood effects when prime exposure is relatively short. The able, pleasant. In the extremely positive priming condition, the
results of Studies 1a, 1b, and 2 strongly suggest that under short following words were each flashed five times: wonderful,
exposure conditions, primarily the evaluative tone of the primed sweet, good, positive. In the moderately negative priming
information is activated. However, the evaluative-tone logic also condition, the following words were each flashed five times:
suggests another possible method of testing our hypothesis. An stingy, weak, plain, unpleasant. In the extremely negative
alternate way to manipulate the dominance of evaluative tone priming condition, the following words were each flashed five
would be to use primed concepts that are sufficiently extreme in times: horrific, cruel, bad, negative.
evaluative tone to dominate descriptive meaning, independent of The procedure was similar to that used in Study 1a, except that
whether prime exposure is extremely or moderately short. Thus, to the time participants were exposed to the primes was always long
expand our evidence, we tested this implication of our line of (120 ms). The dependent measures were similar to those in Study
reasoning in Study 3. 2: Participants reported their need for cognition and indicated to
what extent they felt positive or negative.
Study 3
Results
In this study, we demonstrated in a different way that evaluative
tone is essential to induce a mood state. What is crucial in our A Prime Valence Prime Extremity ANOVA on the need-for-
all-you-need-is-evaluative-tone hypothesis is that evaluative cues cognition composite scale (Cronbachs .83) revealed the
of a concept are more salient, or more strongly activated, than are predicted interaction, F(1, 41) 5.28, p .05, 2 .11, and a
its descriptive cues. One way to achieve this state is to present the main effect of prime valence, F(1, 41) 6.62, p .05, 2 .14.
information for a sufficiently short duration that primarily the There was no effect of extremity (F 1). As can be seen in Table
evaluative features are activated (Stapel et al., 2002), which we did 4, the interaction effect reflects that, as expected, when the prime
in Studies 1a, 1b, and 2. Another way to achieve this is to use words were extreme, participants in the positive condition reported
concepts that are extreme in their valence and thus strong in their a lower need for cognition (M 3.07, SD 0.56) than partici-
(un)desirability (see Stapel & Koomen, 2000). Such concepts have pants in the negative condition (M 4.27, SD 1.07), F(1, 43)
a strong evaluative meaning that could dominate their specific 12.66, p .01, 2 .23, whereas when the prime words were
descriptive meaning. Extremely valenced concepts, such as won- moderate, priming had no effect on reported need for cognition
derful versus horrific, for example, have a strong evaluative (F 1). In those conditions, participants need-for-cognition
tone (i.e., positive vs. negative), whereas moderately valenced judgments were similar (Fs 1) to those in the control condition
concepts, such as pleasant versus unpleasant, although per- (M 3.56, SD 0.59). To provide an overall test of the predicted
haps similarly specific in their descriptive meaning (see Hampson, pattern of results, we also performed a contrast analysis. On the
John, & Goldberg, 1986; Stapel & Koomen, 2000), have a weaker basis of the predictions, we assigned weights of 1 to the cells that
evaluative tone. We therefore expected that when extremely va- we expected not to differ (the moderate positive, moderate nega-
lenced concepts were primed (i.e., with short and long exposures), tive, and control conditions), a weight of 4 to the cell in which we
their evaluative tone would be more likely to dominate their expected the highest need for cognition (the extreme negative
descriptive meaning and thus yield mood effects than when mod- condition), and a weight of 7 to the cell in which we expected the
erately valenced concepts were primed. lowest need for cognition (the extreme positive condition). This a
In the present study, we primed participants with trait concepts priori contrast reached significance, t(52) 3.76, p .05. Con-
of extreme valence or with trait concepts of moderate valence, both trast analyses performed to compare the extreme positive condition
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 785

Table 4 positive with control and one contrasting moderate negative with
Means and Standard Deviations of Scores for Items on Need for controlwas significant, t(52) 0.10, p .93, and t(52) 0.13,
Cognition Scale and for Mood Judgments as a Function of p .90, respectively.
Prime Extremity and Prime Valence (Study 3) Further analyses showed that the partial correlation (controlling
for experimental condition) for our two dependent measures (i.e.,
Prime extremity need for cognition and mood judgment) was high, r .62 ( p
Moderate Extreme .01). This allowed us to test the robustness of our findings by
computing a composite scale of the z-transformed need-for-
Prime valence M SD M SD cognition scores and the recoded and then z-transformed mood
judgments. Next, we performed the same contrast analyses as
Need for cognition score
described earlier on our composite scale. The contrast analysis
Positive 3.86 0.64 3.07 0.56 testing the expected overall pattern of results was highly signifi-
Negative 3.75 0.90 4.27 1.07 cant, t(52) 3.83, p .05. Contrast analyses performed to
compare the extreme positive conditions with control and the
Mood judgment
extreme negative conditions with control were also significant,
Positive 4.55 0.82 5.27 1.01 t(52) 1.97, p .05, and t(52) 2.11, p .05 respectively. As
Negative 4.64 0.92 4.08 1.17 expected, two additional contrasts comparing moderate positive
with control and comparing moderate negative with control were
Note. The Need for Cognition Scale ranges from 1 to 5. Higher scores
indicate a higher need for cognition. Mean need for cognition score in
not significant, t(52) 0.28, p .78, and t(52) 0.26, p .80.
control condition was 3.56 (SD 0.59). The mood judgment scale ranges The results of this study are important for two reasons: First, the
from 1 to 7. Higher scores indicate more positive judgments. Mean mood findings replicate the effect of mood on peoples reported need for
judgment in control condition was 4.58 (SD 0.79). cognition. Second, the results support our evaluative-tone hypoth-
esis because mood was only affected when the evaluative meaning
of the primed trait concepts dominated their descriptive meaning.
with control and the extreme negative condition with control were Thus, participants who were primed with extreme positive trait
respectively significant, t(52) 2.07, p .05, and marginal, concepts reported a lower need for cognition and indicated they
t(52) 2.01, p .06. As expected, additional contrast analyses felt better than participants who were primed with extreme nega-
testing the moderate positive condition against control and the tive trait concepts. As expected, reported need for cognition and
moderate negative condition against control were not significant, mood were not affected in participants who were primed with
t(52) 0.37, p .72 and t(52) 0.57, p .57 respectively. moderate trait concepts.
A Prime Valence Prime Extremity ANOVA on the mood Together, the first three studies demonstrate that a dominating
judgments also revealed the predicted interaction, F(1, 41) 4.68, evaluative tone is essential for influencing peoples mood states.
p .05, 2 .10, and a marginal effect of prime valence, F(1, We have provided evidence on explicit mood measures and on
41) 3.44, p .07, 2 .08. There was no effect of prime peoples motivations (i.e., need for cognition). To complete the
extremity (F 1). The means and standard deviations are depicted picture, we set as our final goal providing support for our all-you-
in Table 4. Equivalent to the pattern of results for need for need-is-evaluative-tone hypothesis using a different processing-
cognition, the interaction effect indicates that when the prime style measure than need for cognition. It was in this spirit that we
words were extreme, participants in the positive condition reported conducted Study 4.
feeling more positive (M 5.27, SD 1.01) than participants in
the negative condition (M 4.08, SD 1.17), F(1, 43) 8.57,
p .01, 2 .17, whereas when the prime words were moderate, Study 4
priming had no effect on experienced affect (F 1). Again in
those conditions, participants mood judgments were similar (Fs As mentioned previously, moods may influence peoples pro-
1) to those in the control condition (M 4.58, SD .79). We then cessing styles. People are more likely to rely on heuristic, easy, and
performed a contrast analysis to provide an overall test of the global processing strategies when they feel good and tend to use
predicted pattern of results. On the basis of the predictions, we more demanding, systematic, and local processing strategies when
assigned weights of 1 to the cells that we expected not to differ (the they feel bad. Bless, Bohner, Schwarz, and Strack (1990) showed
moderate positive, moderate negative, and control conditions), a the effects of mood on processing style in the realm of persuasion
weight of 4 to the cell in which we expected the most positive (see also Petty, DeSteno, & Rucker, 2001). The persuasion liter-
mood (the extreme positive condition), and a weight of 7 to the ature informs us that the impact of argument strength may differ
cell in which we expected the most negative mood (the extreme depending on peoples processing styles. Strong arguments are
negative condition). This a priori contrast reached significance, more convincing than weak arguments for people who use sys-
t(52) 2.73, p .05. Contrast analyses performed to compare the tematic processing strategies, whereas strong and weak arguments
extreme positive condition with control and the extreme negative are equally convincing for people who use heuristic processing
condition with control were respectively marginal, t(52) 1.73, strategies. Building on these findings, Bless and colleagues (1990)
p .09, and nonsignificant, t(52) 1.28, p .21. We then showed that for participants who felt good, argument strength was
conducted two additional contrasts, comparing the control condi- not important in convincing them. For people in a negative mood,
tion with conditions in which we did not expect mood changes to however, argument strength was important: People who felt bad
occur. As expected, neither analysis one contrasting moderate were more influenced by strong rather than weak arguments.
786 RUYS AND STAPEL

The aim of the current research was to provide support for our arguments were indeed judged as stronger (M 5.00, SD 0.83)
evaluative-tone hypothesis on a processing-style measure. We than the weak arguments (M 3.80, SD .079).
used the impact of argument quality on persuasion to show that a Next, a Prime Valence Prime Exposure Argument Strength
dominating positive evaluative tone leads to a good mood, whereas ANOVA on the attitude judgment revealed a marginal three-way
a dominating negative evaluative tone leads to a bad mood. Thus, predicted interaction, F(1, 88) 3.32, p .07, 2 .04, a
because people in a positive mood process information heuristi- marginal Prime Valence Argument Strength effect, F(1, 88)
cally, they are equally likely to be convinced by weak or strong 2.82, p .10, 2 .04, a Prime Valence Prime Exposure effect,
arguments. People in a negative mood tend to process information F(1, 88) 6.61, p .05, 2 .07, a prime valence effect, F(1,
systematically and therefore are more likely to be convinced by 88) 10.21, p .01, 2 .10, and an argument strength effect,
strong rather than weak arguments. We primed participants with F(1, 88) 21.71, p .01, 2 .20. As can be seen in Table 5,
positive or negative trait concepts using long or short subliminal the pattern of means behind these effects strongly supports our
exposures, similar to those used in Studies 1 and 2. After the predictions. As expected, in the short exposure conditions, partic-
priming episode, participants were presented with several strong ipants attitudes were similarly positive in the strong (M 5.14,
and weak arguments in favor of an attitude object and were then SD 0.66) and weak (M 4.86, SD 0.86) conditions (F 1),
asked to indicate their attitude. whereas in the long exposure conditions, participants attitudes
were more positive in the strong condition (M 4.75, SD 0.87)
than in the weak condition (M 3.25, SD 0.75), F(1, 94)
Method 15.39, p .01, 2 .14. In other words, after short negative
priming, argument strength mattered (suggesting a negative mood
Participants (N 108) were undergraduates who took part in
was induced), whereas it did not matter after short positive priming
exchange for partial course credit. The participants were randomly
(suggesting that a positive mood was induced). In the long expo-
assigned to the conditions of a 2 (prime exposure: long or short)
sure conditions, prime valence had, as predicted, no (main or
2 (prime valence: positive or negative) 2 (argument strength:
interaction) effect (Fs 1). Here, strong arguments led to more
weak or strong) between-participant design or to a control condi-
positive attitudes (M 4.78, SD 0.87) than weak arguments
tion, in which participants were subliminally primed with neutral
(M 4.05, SD 0.90), F(1, 94) 6.07, p .05, 2 .06.
traits and read strong and weak arguments.
Participants attitudes in the long exposureweak arguments con-
The procedure was similar to that used in Study 1a, but we relied
ditions were similar (F 1) to participants attitudes in the control
on a different measure of processing style. Immediately after the
condition (M 4.08, SD 0.52). However, to provide an overall
priming procedure, participants read two arguments in favor of
test of the predicted pattern of results, we again performed a
using English as the official language of the Dutch university at
contrast analysis. On the basis of the predictions, we assigned
which this experiment was conducted. This would mean that all
weights of 1 to the cells that we expected to have a positive attitude
classes would be conducted in English rather than in Dutch. The
(all strong argument conditions and the short weak positive con-
arguments were pretested to be either strong or weak. The strong
dition), weights of 1 to the cells in which we expected a rela-
arguments were as follows: If English is the official language, the
tively negative attitude (the long weak conditions and the control
university can attract more international students, which will in-
condition), and a weight of 2 to the cell in which we expected the
crease its status and make it easier to improve facilities and
most negative attitude (the short weak negative condition). This a
teaching resources and The international employability of uni-
priori contrast was highly significant, t(99) 6.87, p .05.
versity graduates will increase if English is the official language.
The weak arguments were as follows: Improving the English of
the universitys students may be helpful to them on holidays and Table 5
If English is the official language at our university, it will be Means and Standard Deviations for Attitude as a Function of
easier to understand Anglo-Saxon movies and television series. Prime Exposure, Argument Strength, and Prime Valence
After reading the strong or weak arguments, participants were (Study 4)
asked to indicate their attitude (1 completely disagree, 7
completely agree) toward the statement that it is a good idea to Prime exposure
make English the official language at our university and have all
Long Short
teaching in English. Similar to the procedure in our previous
studies, we included a mood question to assess the successfulness Prime valence M SD M SD
of our mood induction. Finally, to check the successfulness of our
argument strength manipulation, we asked participants to indicate Strong arguments
the strength of the arguments they had read on a 7-point scale Positive 4.82 0.87 5.14 0.66
ranging from 1 (completely disagree) to 7 (completely agree). Negative 4.70 0.95 4.75 0.87

Weak arguments
Results
Positive 4.08 0.90 4.86 0.86
First, a Prime Valence Prime Exposure Argument Strength Negative 4.00 0.94 3.25 0.75
ANOVA on the argument strength manipulation check revealed Note. Scale ranges from 1 to 7. Higher scores indicate a more positive
that the predicted main effect of argument strength, F(1, 88) attitude. Mean attitude judgment in the control condition was 4.08 (SD
55.67, p .01, 2 .39 (other Fs 1), indicating that the strong 0.52).
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 787

A Prime Valence Prime Exposure Argument Strength Summary of Findings


ANOVA on the mood judgments revealed the predicted Prime
Valence Prime Exposure interaction, F(1, 88) 19.06, p .01, A critical reader could, perhaps, argue that in some of our
2 .18, and a main effect of prime valence, F(1, 88) 15.13, studies, direct comparisons between the control condition and the
p .01, 2 .15 (other Fs 1). As can be seen in Table 6, the conditions in which we expected mood changes to occur were not
always significant. However, we presented multiple studies and
interaction effect reflects that, as expected (as in the other exper-
replicated our main finding across measures (two types of mood
iments), in the short exposure conditions, participants in the pos-
measures, need-for-cognition scale, argument strength logic) and
itive condition reported feeling more positive (M 5.57, SD
across a large number of studies (Study 1a, Study 1b, Study 2,
0.69) than participants in the negative condition (M 4.13, SD
Study 3, and Study 4). This clearly and convincingly shows that
0.90), F(1, 94) 39.68, p .01, 2 .30, whereas in the long our effect is real and robust. To provide concrete statistical support
exposure conditions, priming had no effect on experienced affect for this claim, we performed additional contrast analyses across
(F 1) and participants moods were similar (F 1) to those in Studies 1a, 1b, 2, 3, and 4 on the z-transformed mood judgments
the control condition (M 4.75, SD 0.87). Again, we performed and across Studies 2 and 3 on the z-transformed need-for-cognition
a contrast analysis to provide an overall test of the predicted scores.
pattern of results. Using the same weights as in Study 1a, we found First, we performed an overall test of the predicted pattern on
that the a priori contrast was significant, t(103) 5.40, p .05. the mood judgments. Using the same weights as in Study 1a, we
Contrast analyses performed to compare the short positive condi- found that the a priori contrast was highly reliable, t(383) 7.48,
tions with control and the short negative conditions with control p .05. In addition, we performed two contrast analyses to
were also significant, t(103) 2.83, p .05, and t(103) 2.10, directly compare the control condition with the conditions in
p .05, respectively. However as expected, contrast analyses which we expected mood changes to occur (the short exposure and
performed to compare the long positive conditions with control extreme conditions). Both contrasts one comparing positive with
and the long negative conditions with control were not significant, control, t(383) 4.04, p .05, and one comparing negative with
t(103) 0.28, p .78, and t(103) 0, p 1, respectively. control, t(383) 3.76, p .05were highly significant. To make
In a nutshell, this study provides additional support for our sure that the conditions in which we did not expect mood changes
all-you-need-is-evaluative-tone hypothesis. The results show that to occur (the long exposure and moderate conditions) did not differ
in the long prime exposure conditions, participants were influ- from the control condition, we performed two additional contrast
enced more by the strong than by the weak arguments. Their analyses. In line with our hypotheses, neither of the contrasts
one comparing positive with control, t(383) 0.57, p .57, and
moods were unaffected. More important, in the short prime expo-
one comparing negative with control, t(383) 0.14, p .89
sure conditions, participants primed with positive trait concepts
was significant.
were influenced by both strong and weak arguments, whereas
Second, we performed similar contrast analyses on the need-
participants primed with negative trait concepts were influenced
for-cognition scores, starting with an overall test of the predicted
only by strong arguments. Those participants in the short exposure pattern. Using the same weights as in Study 2, we found that the
conditions also reported the expected mood states. Thus, a domi- a priori contrast was highly reliable, t(112) 4.76, p .05. Next,
nating evaluative tone influenced peoples moods and, therefore, we performed the two contrast analyses to directly compare the
peoples processing styles. conditions in which we expected mood changes to occur with the
control condition. Both contrasts one comparing positive with
control, t(112) 2.23, p .05, and one comparing negative with
control, t(112) 3.08, p .05were highly significant. As
Table 6 expected, neither of the contrasts testing the control condition
Means and Standard Deviations for Mood Judgments as a against the conditions in which we did not expect mood changes to
Function of Prime Exposure, Argument Strength, and Prime occur one comparing positive with control, t(112) 0.18, p
Valence (Study 4) .86, and one comparing negative with control, t(112) 0.28, p
Prime exposure
.78 was significant.
Together, these contrast analyses provide strong support for our
Long Short claim that only very brief presentations of evaluative information
or presentations of very extreme evaluative information may in-
Prime valence M SD M SD fluence mood and cognition. The fact that in some of our studies,
Strong arguments single low-power comparisons between individual cells did not
reach ordinary levels of significance is thus completely surpassed
Positive 4.67 1.16 5.50 0.52 by the robustness and reliability of our hypothesized effect across
Negative 4.90 0.88 4.08 0.87 a large number of studies and measures.
Weak arguments

Positive 4.67 0.65 5.64 0.84


General Discussion
Negative 4.60 0.84 4.17 0.84
Most successful mood induction techniques rely on the con-
Note. Scale ranges from 1 to 7. Higher scores indicate a more positive scious experience or recall of real or imagined mood-eliciting
mood. Mean mood judgment in the control condition was 4.75 (.87). stimuli and demand a relatively high amount of cognitive capacity.
788 RUYS AND STAPEL

But what exactly are the necessary preconditions to influence trait concepts was short, then the activated evaluative tone spilled
peoples mood states? What are the ingredients of a minimal mood over into participants explicit mood reports. In contrast, when
induction paradigm? That is the central question of the present subliminal exposure to the valenced trait concepts was relatively
research. Together, the results of four studies indicate that suc- long, no effects were found in the participants explicit mood
cessful mood induction is much more basic and simple than reports. Thus, mood changes only took place in the case in which
previous mood research would suggest. First, our mood induction the evaluative tone dominated the descriptive meaning of the
paradigm does not demand a high amount of cognitive capacity, prime. The results of Study 1b suggest that such carryover effects
considering that participants only responded to the location of our are most likely to occur on relatively general and evaluative mood
masked priming materials. Second, the present studies indicate that judgments but not on relatively specific, descriptive self-
moods can be induced without participants awareness of the judgments and person judgments.
mood-eliciting stimuli. Thus, participants were not conscious that Study 2 extends the findings of Study 1 to an indirect mood
mood states were being elicited. Third, we used valenced trait measure: The results illustrate that subliminally primed trait con-
concepts as mood-eliciting stimuli, which most researchers regard cepts affected the participants reported need for cognition when
as cold priming materials (see e.g., Clore & Colcombe, 2003; the evaluative tone dominated the descriptive meaning of the
Niedenthal, Rohman, & Dalle, 2003). primes. Thus, the activation of a positive evaluative tone resulted
Recently, it has been argued that the activation of cold semantic in a more positive mood than the activation of a dominating,
concepts is insufficient to induce mood states and thus hot stimulus negative evaluative tone. Similar to our findings in Study 1, we
materials are required to produce mood effects (Clore & Col- obtained no mood effects when the descriptive meaning was
combe, 2003; Innes-Ker & Niedenthal, 2002; Niedenthal et al., sufficiently activated (i.e., under relatively long stimulus expo-
2003). At first glance, the results of the present research seem to sures). Besides supporting our evaluative-tone hypothesis, Study 2
contradict this idea that hot materials are necessary because our is the first study to demonstrate that need for cognition can
findings show that the activation of semantic concepts can produce fluctuate due to situational factors. Specifically, people experi-
affective experiences. However, a closer look supports the view enced a higher need for cognition when they felt bad than when
that hot materials are crucial. Our results clearly reveal that mood they felt good. This finding demonstrates that need for cognition
effects occur when mainly the hot features of these cold concepts may serve as an indirect mood measure, equivalent to peoples
are activated. Our results support the notion that the essential hot information-processing styles. Moreover, this finding shows for
ingredient in mood-induction procedures is a dominating evalua- the first time that need for cognition can be used as a dependent
tive tone. measure. In previous studies, need for cognition always served as
The present studies illustrate two ways in which evaluative tone a mediating or moderating variable. Thus, the present findings
may dominate descriptive cues. According to the affective primacy show that it makes sense to distinguish between state and trait
hypothesis, nondescriptive, nonspecific evaluative stimulus cues need for cognition in future research.
are especially likely to be picked up when stimulus exposure is Study 3 provides additional evidence for our evaluative-tone
very short. Thus, we can separate the activation of evaluative and hypothesis. In the previous studies, we separated the activation of
descriptive meaning with subliminal, very short stimulus expo- evaluative and descriptive meaning by presenting the information
sures, activating mainly the evaluative tone of the stimulus (see for a sufficiently short duration that primarily the evaluative fea-
also Stapel et al., 2002). The evaluative tone of a stimulus also tures were activated. In Study 3, we used trait concepts with an
dominates descriptive meaning when the stimulus is sufficiently evaluative meaning that always dominated their descriptive mean-
extreme. Evaluative meaning is generally more salient in ex- ing. Thus, we primed participants with trait concepts of extreme
tremely valenced trait concepts than descriptive meaning is. valence or with trait concepts of moderate valence, both under
We conclude from our studies that what all successful mood- relatively long, but subliminal, exposure durations. Consistent
induction techniques have in common is that they prime positively with our evaluative-tone hypothesis, reported need for cognition
or negatively toned information at the expense of specific mean- and explicit mood judgments indicated that participants who were
ing. All you need to produce a mood state is a dominating activa- primed with extreme positive trait concepts felt more positive than
tion of global, diffuse, nonspecific evaluative information. The participants who were primed with extreme negative trait concepts.
activation of mainly evaluative, and thus diffuse, information may As expected, the mood states of participants who were primed with
then spill over to peoples moods (Zajonc, 2000; see also Forgas, moderate trait concepts were unaffected.
1995; Schwarz & Clore, 1996). However, mood effects become Studies 1, 2, and 3 provided evidence for our evaluative-tone
less likely when evaluative as well as descriptive information is hypothesis on explicit mood measures and on peoples motiva-
activated, cognitively constraining the evaluative tone. tional states (i.e., need for cognition). To complete the picture, we
conducted Study 4, which directly measured processing style.
Summary of Results After the priming episode, we offered participants at a Dutch
university either strong or weak arguments in favor of using
In Studies 1a and 1b, we tested our evaluative-tone hypothesis English as the official university language. In line with our pre-
using very short versus relatively long subliminal exposures to diction, strong arguments were more persuasive than weak argu-
positive and negative trait concepts. Both studies demonstrate that ments for participants who were exposed for a very short time to
subliminal exposure to valenced information is most likely to negative trait concepts, whereas strong and weak arguments were
affect explicit mood judgments when prime exposures are short equally persuasive for participants who were exposed for a very
enough to activate evaluative reactions that have no specific de- short time to positive concepts. However, when prime exposure
scriptive content. Thus, when subliminal exposure to the valenced was relatively long, strong arguments were more persuasive than
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 789

weak arguments for all participants. Together, these findings in- Wellens, 1987; Winkielman & Berridge, 2004; Winkielman, Ber-
dicate that participants in a good mood used heuristic, global ridge, & Wilbarger, 2005b), until the present studies hardly any
processing styles to process the arguments, and participants in a reliable empirical evidence explained how and when moods could
bad mood used systematic, detailed processing styles. Thus, the be unconsciously induced. Theoretically, it makes sense to assume
results expand our evaluative-tone hypothesis to yet another indi- that people can be aware of their emotional states without being
rect mood measure: information-processing style. aware of the antecedents that evoked these states. Often, people do
not have conscious access to the antecedents that evoked their
Affective or Semantic Primacy? feelings, thoughts, motivations, and behaviors (Nisbett & Wilson,
1977).
In sum, the results of our studies strongly support the all-you- Nevertheless, most previous research has pointed out that un-
need-is-evaluative-tone hypothesis. Our findings also provide ad- consciously presented stimuli do not affect peoples mood states.
ditional support for affective primacy, the hypothesis that people Subliminal priming may have an impact on the preference for and
extract the evaluative meaning of a stimulus before its descriptive liking of neutral target stimuli, without having an impact on
meaning. We showed several times, using explicit and implicit peoples affective state (e.g., Banse, 2001; Clore & Colcombe,
mood indicators, that subliminal presentations of valenced trait 2003; Edwards, 1990; Krosnick, Betz, Jussim, & Lynn, 1992;
concepts only affect peoples mood states when exposure is very Murphy & Zajonc, 1993; Winkielman, Berridge, & Wilbarger,
short. Subliminal presentations of these trait concepts do not affect 2005a). These null findings on mood measures probably led
peoples mood states when exposure is relatively long but still Schwarz and Clore (1996) to conclude that in general, affective
subliminal. We can only conclude from our findings that people priming studies are irrelevant to the study of mood effects. Such
extracted evaluative meaning before descriptive meaning. It seems studies demonstrate effects of subliminal priming on judgments of
very unlikely that specific content or descriptive information in- words, people, or Chinese characters, but they do not show sub-
duced good and bad moods in the participants. liminal priming effects on experienced moods. Thus, Schwarz and
Recently, researchers have argued that it is a sin (Clore et al., Clore (1996) concluded, In the absence of experienced feelings,
2005, p. 394) to assume that affect may precede semantic analysis affective priming studies may indeed be better conceptualized as
and that the evidence for affective primacy is weak (see also reflecting automatic evaluation processes . . . , which have been
Storbeck & Robinson, 2004). These conclusions were based on (a observed with materials unlikely to elicit any feelings . . . , rather
subset of) previous affective priming studies and research that than feeling-based inferences (p. 440).
compared semantic priming and affective priming (Storbeck & We posit that Schwarz and Clore (1996) may have been right in
Robinson, 2004). Storbeck and Robinson (2004) showed that their conclusion that to date, there is no consistent evidence that
semantic priming was consistent and more robust than affective subliminal priming may affect peoples affective feelings. How-
priming. They took their results to mean that semantic analysis is ever, this should not be taken to mean that it is impossible for
more obligatory at encoding than affective analysis is. However, subliminally presented information to influence peoples mood
another (and less sectarian) way to interpret these results is that (see e.g., Chartrand et al., 2006). There does not seem to be an a
(automatic) information processing is flexible (see also Stapel & priori reason to assume that peoples moods cannot be affected by
Koomen, 2006): Sometimes, people extract evaluative meaning subliminally primed information. Why should the subliminally
before descriptive meaning, whereas at other times, people extract primed information only be capable of affecting a perceivers
descriptive meaning before evaluative meaning. Sometimes it judgments of Chinese characters (e.g., Murphy & Zajonc, 1993;
makes sense to see immediately whether an animal is a snake or a Winkielman, Zajonc, & Schwarz, 1997) or other peoples behav-
spider (by extracting descriptive information), whereas at other iors (e.g., Stapel et al., 2002) but not also affect the perceiver him-
times it is more functional to see immediately whether an animal or herself? We think that spreading activation, the mechanism
is cute or threatening (by extracting evaluative information). In the presumed to underlie evaluative priming (Bargh, 1997; Ferguson
specific, minimal mood induction paradigm we used in the current & Bargh, 2003; Wentura, 2000), may equally relate to peoples
studies, people were more likely to extract evaluative meaning moods as to peoples liking judgments. However, mood effects
before descriptive meaning when subliminal exposure to the need a stronger evaluative tone, one that dominates the descriptive
primed information was extremely short. However, we emphasize cues.
that even though we expected affective primacy to occur in our Thus, we explain the lack of mood effects in previous subliminal
current studies, we do not argue that affect always precedes cog- affective priming research by our hypothesis that evaluative tone
nition. Rather, we applied the affective primacy logic to create the needs to overrule the descriptive cues. It seems likely that in
circumstances to test our evaluative-tone hypothesis. previous research, the activation of evaluative cues was sufficient
to produce affective priming but insufficient to affect peoples
Unconscious Moods moods because descriptive cues were also activated. In most of the
studies that have shown dissociation between effects on liking
Our findings show that mood effects are especially likely when judgments and mood, researchers used facial expressions as affec-
we expose people for a very short time to the subliminal primes. tive primes. Participants were subliminally primed with happy,
Put differently, shorter flashes lead to stronger feelings. Although neutral, or angry faces, which resulted in evaluatively congruent
some emotion researchers have posited that it is indeed possible to liking judgments of Chinese ideograms or an unfamiliar beverage
elicit mood effects outside of conscious awareness (Chartrand, (e.g., Winkielman et al., 2005a). We think that a possible expla-
Van Baaren, & Bargh, 2006; Kihlstrom, Mulvaney, Tobias, & nation for the null findings on mood measures might lie in the fact
Tobis, 2000, O hman & Soares, 1994; Robles, Smith, Carver, & that people are very efficient at face perception (Farah, Wilson,
790 RUYS AND STAPEL

Drain, & Tanaka, 1998, p. 482). Some researchers have even quences for experienced affect, impression formation, and stereotyping.
assumed a specialized module for the perception of faces. Other Journal of Experimental Psychology: General, 135, 70 77.
research has demonstrated, for example, that even the social cat- Chartrand, T. L., & Bargh, J. A. (1996). Automatic activation of impres-
egory of a subliminally presented face may have an impact on sion formation and memorization goals: Nonconscious goal priming
subsequent judgments (Ruys, Spears, Gordijn, & De Vries, 2007; reproduces effects of explicit task instructions. Journal of Personality
and Social Psychology, 71, 464 478.
Stapel et al., 2002). Thus, activation of specific descriptive face
Clore, G., & Colcombe, S. (2003). The parallel worlds of affective con-
information is likely to occur early in the information-processing cepts and feelings. In J. Musch & K. C. Klauer (Eds.), The psychology
chain. When evaluative cues face information are activated, of evaluation (pp. 169 188). Mahwah, NJ: Erlbaum.
effects on liking judgment might still occur. It seems functional to Clore, G. L., Storbeck, J., Robinson, M. D., & Centerbar, D. B. (2005).
assume that something is pleasant when other people enjoy it. Seven sins in the study of unconscious affect. In L. F. Barrett, P. M.
However, it seems less functional that mood states would be Niedenthal, & P. Winkielman (Eds.), Emotion and consciousness (pp.
affected in this way. Evaluative cues only influence mood when 384 408). New York: Guilford Press.
the activation of evaluative cues is cognitively unbound. Edwards, K. (1990). The interplay of affect and cognition on attitude
In sum then, the current studies clearly demonstratevia a formation and change. Journal of Personality and Social Psychology,
minimal mood induction paradigmthat all you need to influence 59, 202216.
peoples mood states is evaluative tone. The essential ingredient Erdley, C. A., & DAgostino, P. R. (1988). Cognitive and affective com-
ponents of automatic priming effects. Journal of Personality and Social
for the genesis of mood effects is a dominance of evaluative
Psychology, 54, 741747.
meaning over descriptive meaning. Whereas most previous re-
Farah, M. J., Wilson, K. D., Drain, M., & Tanaka, J. N. (1998). What is
search findings may be interpreted as suggesting that hot, special about face perception? Psychological Review, 105, 482 498.
cognitive-capacity-demanding, and conscious experience of Ferguson, M., & Bargh, J. A. (2003). The constructive nature of automatic
mood-eliciting stimuli are necessary for successful mood induc- evaluation. In J. Musch & K. C. Klauer (Eds.), The psychology of
tion, the results of the present studies support the notion that evaluation (pp. 169 188). Mahwah, NJ: Erlbaum.
peoples mood states can be influenced without their awareness of Fiedler, K. (1990). Mood-dependent selectivity in social cognition. In W.
the mood-eliciting stimuli. Thus, relatively cold concepts can Stroebe & M. Hewstone (Eds.), European Review of Social Psychology
induce hot mental states. (Vol. 1, pp. 132). New York: Wiley.
Fiedler, K. (1991). On the task, the measures, and the mood in research on
affect and social cognition. In J. P. Forgas (Ed.), Emotion and social
References judgments (pp. 83104). Cambridge, United Kingdom: Cambridge Uni-
versity Press.
Adolphs, R. (2003). Cognitive neuroscience of human social behavior.
Fiedler, K. (2001). Affective influences on information processing. In J. P.
Nature Reviews Neuroscience, 4, 165178.
Forgas (Ed.), Handbook of affect and social cognition. Mahwah, NJ:
Banse, R. (2001). Affective priming with liked and disliked persons: Prime
Erlbaum.
visibility determines congruency and incongruency effects. Journal of
Experimental Social Psychology, 15, 501520. Fiedler, K., & Stroehm, W. (1986). What kind of mood influences what
Bargh, J. A. (1997). The automaticity of everyday life. In R. S. Wyer (Ed.), kind of memory: The role of arousal and information structure. Memory
Advances in social cognition: Vol. 10 (pp. 1 61). Mahwah, NJ: Erl- and Cognition, 14, 181188.
baum. Forgas, J. P. (1992). Affect in social judgments and decisions: A multi-
Bargh, J. A. (2006). What have we been priming all these years? On the process model. In M. P. Zanna (Ed.), Advances in experimental social
development, mechanisms, and ecology of nonconscious social behav- psychology (Vol. 25, pp. 227275). San Diego, CA: Academic Press.
ior. European Journal of Social Psychology, 36, 147168. Forgas, J. P. (1995). Mood and judgment: The affect infusion model
Bargh, J. A., Litt, J., Pratto, F., & Spielman, L. A. (1989). On the (AIM). Psychological Bulletin, 117, 39 66.
preconscious evaluation of social stimuli. In A. F. Bennett & K. M. Gaspar, K., & Clore, G. L. (2002). Attending to the big picture: Mood and
McConkey (Eds.), Cognition in individual and social contexts: Proceed- global versus local processing of visual information. Psychological
ings of the XXV International Congress of Psychology. (Vol. 3, pp. Science, 13, 34 40.
357370). Amsterdam: Elsevier/North-Holland. Gervey, B., Igou, E. R., & Trope, Y. (2005). Positive mood and future-
Bless, H., Bohner, G., Schwarz, N., & Strack, F. (1990). Mood and oriented self-evaluation. Motivation and Emotion, 29, 269 295.
persuasion. A cognitive response analysis. Personality and Social Psy- Hampson, S. E., John, O. P., & Goldberg, L. R. (1986). Category breadth
chology Bulletin, 16, 331345. and hierarchical structure in personality: Studies in asymmetries in
Bless, H., Clore, G. L., Schwarz, N., Golisano, V., Rabe, C., & Woelk, M. judgments of trait implications. Journal of Personality and Social Psy-
(1996). Mood and the use of scripts: Does a happy mood really lead to chology, 51, 3754.
mindlessness? Journal of Personality and Social Psychology, 71, 665 Hugenberg, K. (2005). Social categorization and the perception of facial
679. affect: Target race moderates response latency advantage for happy
Cacioppo, J. T., & Petty, R. E. (1982). The need for cognition. Journal of faces. Emotion, 5, 267276.
Personality and Social Psychology, 42, 116 131. Innes-Ker, A., & Niedenthal, P. M. (2002). Emotion concepts and emo-
Cacioppo, J. T., Petty, R. E., Feinstein, J. A., & Jarvis, W. B. G. (1996). tional states in social judgment and categorization. Journal of Person-
Dispositional differences in cognitive motivation: The life and times of ality and Social Psychology 83, 804 816.
individuals varying in need for cognition. Psychological Bulletin, 119, Isen, A. M. (1987). Positive affect, cognitive processes, and social behav-
197253. ior. In L. Berkowitz (Ed.), Advances in experimental social psychology
Chaiken, S. (1980). Heuristic versus systematic information processing and (Vol. 20, pp. 203253). San Diego, CA: Academic Press.
the use of source versus message cues in persuasion. Journal of Per- Keltner, D., Locke, K. D., & Audrain, P. C. (1993). The influence of
sonality and Social Psychology, 39, 752766. attributions on the relevance of negative feelings to personal satisfaction.
Chartrand, T. L., Van Baaren, R. B. V., & Bargh, J. A. (2006). Linking Personality and Social Psychology Bulletin, 19, 2129.
automatic evaluation to mood and information processing style: Conse- Kihlstrom, J. F., Mulvaney, S., Tobias, B. A., & Tobis, I. P. (2000). The
PRECONDITIONS FOR (UNCONSCIOUS) MOOD EFFECTS 791

emotional unconscious. In E. Eich, J. F. Kihlstrom, G. H. Bower, J. P. Stapel, D. A. (2003). Making sense of hot cognition: Why and when
Forgas, & P. M. Niedenthal (Eds.), Cognition and emotion. New York: description influences our feelings and judgments. In J. P. Forgas, K. D.
Oxford University Press. Williams, and W. von Hippel (Eds.), Social judgments: Implicit and
Krosnick, J. A., Betz, A. L., Jussim, L. J., & Lynn, A. R. (1992). Sublim- explicit processes (pp. 227250). New York: Cambridge University
inal conditioning of attitudes. Personality and Social Psychology Bulle- Press.
tin, 18, 152162. Stapel, D. A., & Koomen, W. (2000). How far do we go beyond the
LeDoux, J. E. (1989). Cognitive emotional interactions of the brain. information given? The impact of knowledge activation on interpretation
Cognition and Emotion, 3, 267289. and inference. Journal of Personality and Social Psychology, 78, 19 37.
Maringer, M., & Stapel, D. A. (2007). Unfinished business: How com- Stapel, D. A., & Koomen, W. (2005). When less is more: The conse-
pleteness affects the impact of emotional states and emotion concepts on quences of affective primacy for subliminal priming effects. Personality
social judgment. Journal of Experimental Psychology, 43, 712718. and Social Psychology Bulletin, 9, 1286 1295.
Mayer, J. D., & Gaschke, Y. N. (1988). The experience and meta- Stapel, D. A., & Koomen, W. (2006). The flexible unconscious: Investi-
experience of mood. Journal of Personality and Social Psychology, 55, gating the judgmental impact of varieties of unaware perception. Journal
102111. of Experimental Social Psychology, 42, 112119.
Murphy, S. T., & Zajonc, R. B. (1993). Affect, cognition, and awareness: Stapel, D. A., Koomen, W., & Ruys, K. I. (2002). The effects of diffuse and
Affective priming with optimal and suboptimal stimulus exposures. distinct affect. Journal of Personality and Social Psychology, 83, 60 74.
Journal of Personality and Social Psychology, 64, 723739. Storbeck, J., & Robinson, M. D. (2004). Preferences and inferences in
Niedenthal, P. M., Rohman, A., & Dalle, N. (2003). What is primed by encoding visual objects: A semantic comparison of semantic and affec-
emotion concepts and emotion words? In J. Musch & K. C. Klauer tive priming. Personality and Social Psychology Bulletin, 30, 8193.
(Eds.), The psychology of evaluation (pp. 307333). Mahwah, NJ: Law- Treisman, A. (1996). The binding problem. Current Opinion in Neurobi-
rence Erlbaum. ology, 6, 171178.
Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Wegener, D. T., & Petty, R. E. (1994). Mood management across affective
Verbal reports on mental processes. Psychological Review, 84, 231259.
states: The hedonic contingency hypothesis. Journal of Personality and
hman, A., & Soares, J. J. (1994)Unconscious anxiety: Phobic responses
O
Social Psychology, 66, 1034 1048.
to masked stimuli. Journal of Abnormal Psychology, 103, 231240.
Wentura, D. (2000). Dissociative affective and associative priming effects
Palermo, R., & Rhodes, G. (2007). Are you always on my mind? A review
in lexical decision task: Yes versus no responses to word targets reveal
of how face perception and attention interact. Neuropsychologia, 45,
evaluative judgment tendencies. Journal of Experimental Psychology:
7592.
Learning, Memory, and Cognition, 26, 456 469.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of
Winkielman, P., & Berridge, K. C. (2004). Unconscious emotion. Current
persuasion. In L. Berkowitz (Ed.), Advances in experimental social
Directions in Psychological Science, 13, 120 123.
psychology (Vol. 19, pp. 123205). New York: Academic Press.
Winkielman, P., Berridge, K. C., & Wilbarger, J. L. (2005a). Unconscious
Petty, R. E., DeSteno, D., & Rucker, D. D. (2001). The role of affect in
affective reactions to masked happy versus angry faces influence con-
attitude change. In J. P. Forgas (Ed.), Handbook of affect and social
sumption behavior and judgments of value. Personality and Social
cognition. Mahwah, NJ: Erlbaum.
Robles, R., Smith, R., Carver, C. S., & Wellens, A. R. (1987). Influence of Psychology Bulletin, 31, 121135.
subliminal visual images on the experience of anxiety. Personality and Winkielman, P., Berridge, K. C., & Wilbarger, J. L. (2005b). Emotion,
Social Psychology Bulletin, 13, 399 410. behavior, and conscious experience. Once more without feeling. In L. F.
Ruys, K. I., Spears, R., Gordijn, E. H., & De Vries, N. K. (2007). Barrett, P. M. Niedenthal, & P. Winkielman (Eds.), Emotion and con-
Automatic contrast: Evidence that automatic comparison with the social sciousness (pp. 335362). New York: Guilford Press.
self affects evaluative responses. British Journal of Psychology, 98, Winkielman, P., Zajonc, R. B., & Schwarz, N. (1997). Subliminal affective
361374. priming resists attributional intervention. Cognition and Emotion, 11,
Ruys, K. I., & Stapel, D. A. (in pressa). Emotion elicitor or emotion 433 465.
messenger?: Subliminal exposure to two faces of facial expressions. Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences.
Psychological Science. American Psychologist, 35, 151175.
Ruys, K. I., & Stapel, D. A. (in press b). The secret life of emotions. Zajonc, R. B. (2000). Feeling and thinking: Closing the debate over the
Psychological Science. independence of affect. In J. Forgas (Ed.), Feeling and thinking: The
Schwarz, N. (1990). Feelings as information: Informational and motiva- role of affect in social cognition (pp. 3158). New York: Cambridge
tional functions of affective states. In E. T. Higgins & R. M. Sorrentino University Press.
(Eds.), Handbook of motivation and cognition: Foundations of social
behavior (Vol. 2, pp. 527561). New York: Guilford Press.
Schwarz, N., & Clore, G. (1996). Feelings and phenomenal experiences. In Received January 8, 2007
E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook Revision received December 10, 2007
of basic principles (pp. 433 465). New York: Guilford Press. Accepted December 18, 2007
Journal of Personality and Social Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 94, No. 5, 792 807 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.792

Forming Implicit and Explicit Attitudes Toward Individuals:


Social Group Association Cues

Allen R. McConnell Robert J. Rydell


Miami University University of MissouriColumbia

Laura M. Strain Diane M. Mackie


Miami University University of California, Santa Barbara

The authors explored how social group cues (e.g., obesity, physical attractiveness) strongly associated
with valence affect the formation of attitudes toward individuals. Although explicit attitude formation has
been examined in much past research (e.g., S. T. Fiske & S. L. Neuberg, 1990), in the current work, the
authors considered how implicit as well as explicit attitudes toward individuals are influenced by these
cues. On the basis of a systems of evaluation perspective (e.g., R. J. Rydell & A. R. McConnell, 2006;
R. J. Rydell, A. R. McConnell, D. M. Mackie, & L. M. Strain, 2006), the authors anticipated and found
that social group cues had a strong impact on implicit attitude formation in all cases and on explicit
attitude formation when behavioral information about the target was ambiguous. These findings obtained
for cues related to obesity (Experiments 1 and 4) and physical attractiveness (Experiment 2). In
Experiment 3, parallel findings were observed for race, and participants holding greater implicit racial
prejudice against African Americans formed more negative implicit attitudes toward a novel African
American target person than did participants with less implicit racial prejudice. Implications for research
on attitudes, impression formation, and stigma are discussed.

Keywords: attitudes, implicit attitudes, impression formation, prejudice, stigma

People would like to believe that their attitudes toward others be viewed differently when their race or ethnicity influences
reflect their careful evaluation of others unique and individual perceptions and interpretations of ambiguous behaviors and events
merits. Although this undoubtedly occurs in some cases, social (e.g., Bodenhausen & Wyer, 1985; Duncan, 1976; Sagar &
psychology research raises questions about the pervasiveness of Schofield, 1980). Although these cues typically do not influence
such a reasoned approach to understanding others (Bargh, 1999; perceptions retroactively, they can influence interpretations of
Bargh & Chartrand, 1999; Brewer, 1988; Fiske & Neuberg, 1990; ambiguous acts during encoding (e.g., Bodenhausen & Wyer,
Nisbett & Wilson, 1977; Schwarz & Bohner, 2001). At times, 1985). However, when a targets actions are clear-cut instead of
people are evaluated by the content of their character, but in other ambiguous, accessible social group categories produce little biased
situations, this content can seem largely irrelevant. For example, assimilation, reducing the influence of groups on perceptions of
individuating information about a person can often be relatively target individuals (Bruner, 1957; Higgins, 1989; Srull & Wyer,
inconsequential when perceivers base their evaluations of a person 1979).
on information associated with the individuals social group Of course, the degree to which a targets actions can shape ones
(Fiske, 1998; Fiske & Neuberg, 1990). attitude is determined, in part, by the extent to which perceivers
Indeed, research has demonstrated that cues providing informa- process individuated information about the target. In fact, Fiske
tion about social groups (e.g., obesity, physical attractiveness, and Neubergs (1990) continuum model of impression formation
race) can impact social perceptions. For example, target people can considers the extent to which a targets behaviors guide social
perception instead of information associated with a targets social
group. They proposed that people rely on piecemeal integration
Allen R. McConnell and Laura M. Strain, Department of Psychology, (e.g., the behaviors performed by an individual) instead of cate-
Miami University; Robert J. Rydell, Department of Psychological Sci- gorization (e.g., knowledge associated with the group as a whole)
ences, University of MissouriColumbia; Diane M. Mackie, Department under conditions where perceivers are able and willing to devote
of Psychology, University of California, Santa Barbara. cognitive resources to understanding target individuals. For exam-
These studies were completed while Allen R. McConnell and Laura M. ple, when motivated and presented with a number of behaviors, a
Strain were supported by National Institute of Mental Health Grant perceiver may come to hold a positive attitude toward a target
MH068279 and National Science Foundation Grant BCS 0601148 and
person who is obese (i.e., a member of a social group associated
Robert J. Rydell and Diane M. Mackie were supported by National Insti-
tute of Mental Health Grant MH63762. We thank John Cacioppo and
with negativity) whose behaviors are predominantly positive in
Penny Visser for their comments on this work. nature. Indeed, there is considerable support for the continuum
Correspondence concerning this article should be addressed to Allen R. model (for an overview, see Fiske, Lin, & Neuberg, 1999).
McConnell, Department of Psychology, Miami University, Oxford, OH However, one interesting feature of this work is its focus on how
45056. E-mail: mcconnar@muohio.edu people use categorization and piecemeal integration in the forma-

792
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 793

tion of explicit attitudes (i.e., evaluations that people can report form differently. As an example, Rydell et al. (2006) showed that
and for which expression can be consciously controlled) toward implicit attitudes were formed in response to the valence of sub-
individual group members. Yet, it is an open question as to how an liminal primes presented prior to the visual appearance of a target
individuals social group and individuated behaviors contribute to individual, whereas explicit attitudes were formed in response to
the formation of implicit attitudes (i.e., evaluations for which consciously available descriptions of that targets behaviors. For
people may not initially have conscious access and for which instance, when concurrently presented with a series of negative
activation cannot be controlled) toward individuals. Within the subliminal primes and positive behavioral statements performed
context of the continuum model, the impact of social group knowl- by a target person, participants implicit attitudes toward the target
edge has been assumed to result from less effortful consideration person were negative but explicit attitudes toward the same person
of individuated information (Fiske et al., 1999). But in the current were positive. Consistent with a systems of evaluation account,
work, we suggest that group knowledge may impact implicit implicit and explicit attitudes were formed relatively indepen-
attitude formation even when perceivers devote considerable cog- dently of each other, with each responding to the type of informa-
nitive resources to understanding social targets. Specifically, we tion assumed to influence the associative and rule-based systems,
propose that many social groups are strongly associated with respectively. Although implicit attitudes can, given a sufficient
valence and that the nature of this knowledge (i.e., its associative amount of information, be responsive to verbal information about
basis) may also have important implications for attitude formation, a target person (Rydell & McConnell, 2006), implicit attitudes are
especially for implicit attitudes. In the current work, we examine more responsive to information that is associative in nature (in the
how group association cues affect attitude formation (implicit and case of Rydell et al., 2006, associations that were subliminally
explicit) toward individuals. Although these cues may have impli- paired with the target individual). In the current work, we again
cations for other aspects of impression formation (e.g., stereotypes, focus on how implicit and explicit attitudes (based on different
attributions), here we focus exclusively on how these social group systems of evaluation) can be differentially sensitive to distinct
cues shape attitude formation toward novel individuals. forms of social information.
These demonstrations of differences in implicit and explicit
Systems of Evaluation attitude change notwithstanding, much remains to be determined
about the nature of implicit and explicit attitudes (see Gawronski
Recent work has established that the processes underlying the & Bodenhausen, 2006). For example, in our previous work, we
formation and change of implicit attitudes differ considerably from have only explored attitude formation and change for relatively
those involved in explicit attitudes (e.g., Rydell & McConnell, impoverished targets (e.g., a nondescript White man named Bob).
2006; Rydell, McConnell, Mackie, & Strain, 2006; Rydell, Mc- However, when perceivers encounter social targets, many group
Connell, Strain, Claypool, & Hugenberg, 2007). Specifically, we association cues such as skin color may be available. Although
(Rydell & McConnell, 2006; Rydell et al., 2006) have advanced a Bob could potentially be viewed as a member of several social
systems of evaluation approach to attitudes, proposing that there categories, it is likely that such categorizations are not especially
are two independent systems of evaluation that differ in both what salient to our participants for several reasons. First, because they
information they use and how they act on it (see also Greenwald & only meet one person instead of a target person in a context of
Banaji, 1995; Sloman, 1996; Smith & DeCoster, 2000; Strack & differentiated others, Bobs race, sex, or other possible categories
Deutsch, 2004). The first system of evaluation, the associative (e.g., his age, his hairstyle) should not be distinctive. Indeed, social
system, operates using paired associations based on similarity and categorization is inherently contextual (e.g., an overweight person
contiguity. In this case, learning is based on the accumulation of may be categorized differently in a group of morbidly obese
information over time to form and strengthen associations in others), which means encountering a White male target in isolation
memory. The second system of evaluation, the rule-based system, reduces the number of salient social categories available to a
relies on logic and symbolic representations at a relatively higher perceiver. Further, a college-age, White male target is not likely to
order level of cognitive processing. be viewed as deviant or as a member of a minority social group
On the basis of a systems of evaluation account, one can category (e.g., Miller, Taylor, & Buck, 1991), especially to
delineate evaluations that tap into the associative and rule-based college-age participants who themselves are predominantly Euro-
systems of evaluation: implicit and explicit attitudes, respectively pean American (e.g., McGuire, McGuire, Child, & Fujioka, 1978).
(Rydell et al., 2006). This approach is congruent with current However, the introduction of a target individual with more dis-
conceptualizations of how implicit and explicit attitudes operate, tinctive social group association cues (e.g., an African American
allowing one to generate novel predictions about how evaluations Bob) could presumably have a considerable effect on attitude
are formed and changed in memory (cf. Gawronski & Boden- formation.
hausen, 2006). The associative system of evaluation is relevant to If so, might these group association cues have different impli-
the understanding of how implicit attitudes form and function cations for implicit and explicit attitude formation? On the basis of
because implicit attitudes are posited to follow the basic principles a systems of evaluation analysis, we believe the answer is yes. As
of similarity and association (Smith & DeCoster, 2000). The Rydell et al. (2006) showed, implicit attitudes were primarily
rule-based system, however, fits with a conceptualization of ex- affected by associative information rather than by detailed state-
plicit attitudes as evaluations based on conscious deliberation or ments about the targets behaviors, whereas explicit attitudes were
syllogistic reasoning, which can reveal quick changes in expres- shaped by the valence of the behavioral information instead of the
sion (Fazio, 1995) but require cognitive resources in their forma- valence of subliminal primes. Because of the sensitivity exhibited by
tion and change (Petty & Wegener, 1998). implicit attitudes to information based on associations (see also Slo-
This systems of evaluation approach has proven useful in un- man, 1996), we reasoned that when group association cues are pre-
derstanding how implicit and explicit attitudes toward individuals sented about a target person, such as being overweight, being physi-
794 MCCONNELL, RYDELL, STRAIN, AND MACKIE

cally attractive, or being African American, these cues, because they person, consistent with many existent findings showing that social
are association based in nature, would be used more strongly by the groups can bias judgments in ambiguous situations.
associative system of evaluation and thus influence implicit attitudes In the current work, we examined visual cues strongly associ-
toward the target in proportion to how strongly they are associated ated with positivity or negativity. Specifically, we explored obe-
with positivity or negativity (i.e., stronger cues should have a greater sity, physical attractiveness, and race. We were drawn to these
impact). However, in the absence of such group association cues, cues because obesity and race have been studied extensively in
implicit attitudes toward the person should eventually reflect the research on stigma. For example, people avoid stigmatized group
valence of target-relevant behavioral information (Rydell & McCon- members (e.g., Pryor, Reeder, Yeadon, & Hesson-McInnis, 2004),
nell, 2006). That is, implicit attitudes are sensitive to verbally con- devalue items associated with them (e.g., Neuberg, Smith, Hoff-
veyed information about a targets behavior, but they will be more man, & Russell, 1994; Rozin, Markwith, & Nemeroff, 1992), and
strongly influenced by group cues that have strong valence associa- evaluate them negatively on implicit (e.g., Castelli et al., 2004;
tions. Indeed, Castelli, Zogmaister, Smith, and Arcuri (2004) showed Fazio, Jackson, Dutton, & Williams, 1995; Greenwald et al., 1998;
that implicit attitudes can be formed simply by linking a person with Nosek, 2005; Nosek & Banaji, 2001; Wittenbrink, Judd, & Park,
a group very strongly associated with valence (e.g., child molesters) in 1997) and explicit (e.g., Crocker, Major, & Steele, 1998; Dovidio,
the absence of behavioral information. However, if a person is a Kawakami, & Gaertner, 2002; Plant & Devine, 1998) measures.
member of a social group more weakly associated with valence (or if Thus, being a member of a stigmatized group provides a strong,
no group association cues are available at all), implicit attitudes negative group association cue.
toward the target will reflect the individuals behaviors (e.g., Rydell & Physical attractiveness can also serve as a strong group associ-
McConnell, 2006; Rydell et al., 2007). ation cue (for many of the same reasons as stigmatized group
In contrast, we expected that the valence of the verbal state- membership), but, unlike stigma, a persons physical attractiveness
ments presented about the target persons behaviors would deter- can serve as either a positive or a negative cue. For instance,
mine explicit attitudes toward the individual (Rydell & McCon- people who are physically attractive are assumed to be competent
nell, 2006; Rydell et al., 2006) regardless of the group association and positive in domains unrelated to their looks (e.g., Chaiken,
cues presented. That is, when unambiguous statements clearly 1979; Dion et al., 1972; Eagly, Ashmore, Makhijani, & Longo,
describe a target person who performs positive or negative behav- 1991), whereas those who are physically unattractive are viewed
iors, the likelihood that a group association cue can assimilate such quite negatively (e.g., Ambady & Rosenthal, 1993; Berscheid &
clear-cut behaviors is exceedingly low (Higgins, 1989; Srull & Walster, 1974), even by infants (e.g., Dion, 1973). Thus, whereas
Wyer, 1979). However, if a target persons individual behaviors obesity and race provide ways to instantiate negative group asso-
are ambiguous with respect to valence, a targets group association ciation cues about target individuals, variability in attractiveness
cue may serve to disambiguate each behavior, exerting an assim- can provide positive and negative group association cues.1
ilative effect and thus influencing explicit attitude formation to-
ward the individual in these cases. Overview of the Current Work
We conducted four experiments to evaluate whether group
Group Association Cues association cues would, in general, have a stronger impact on
implicit attitudes than on explicit attitudes when forming attitudes
It has been shown that people have strong negative evaluations toward a group member, as anticipated by a systems of evaluation
with groups ranging from the obese and the unattractive (e.g., Nosek, account. The basic paradigm and the attitudes measures used were
2005; Rudman, Feinberg, & Fairchild, 2002) to racial outgroups (e.g., the same as those applied in previous research (e.g., Rydell &
Greenwald, McGhee, & Schwartz, 1998; McConnell & Leibold, McConnell, 2006; Rydell et al., 2006). Specifically, participants
2001). But in the current work, we were interested in whether these received detailed, verbal information about the behaviors of a
negative group associations would impact attitude formation about novel target person (Bob or Bobbie, depending on the experiment)
individual targets and, in particular, implicit attitudes toward them. prior to reporting their implicit and explicit attitudes toward the
Clearly, obesity (e.g., Crandall et al., 2001), attractiveness (e.g., Dion, target. Initially, a number of trials were presented featuring a target
Berscheid, & Walster, 1972), and race (e.g., Sagar & Schofield, 1980) photo and behavioral statements about the person to induce either
can impact deliberate evaluations and judgments. Yet, it is important a positive or a negative initial attitude toward the target. New to the
to note that many studies showing the impact of groups on perceptions current work were manipulations of target photos (see Figure 1 for
and judgments involve situations engineered to be equivocal (e.g., an examples) that allowed us to present a target with negative or
ambiguous shove in the hallway between two students, student court positive group association cues (or no salient group association cue
cases that present a mixture of guilt-suggestive and guilt-exonerating
details about defendants) to maximize the likelihood that the cue (e.g.,
1
a sketch involving an African American child) will influence percep- In the current work, we use the term group association cue because
tions. participants are never directly told anything about the target persons
Thus, in the current study, we expected that group association cues membership in a social category (e.g., physical attractiveness is inferred
from a visual image of the target person, who, on the basis of pretesting,
would have a far greater impact on implicit attitudes than on explicit
was reliably viewed as normatively attractive). Also, we do not propose
attitudes when a substantial amount of unambiguous verbal informa- that these cues cannot affect explicit attitudes. For instance, group associ-
tion was presented about the target persons behaviors. However, in ation cues are especially likely to affect deliberate evaluations in circum-
cases where the behavioral information about the target person was stances where the cue is perceived to be germane to ones impression (e.g.,
ambiguous with respect to valence, we anticipated that group associ- physical attractiveness is likely to influence explicit attitudes toward a
ation cues would also influence explicit attitudes toward the target potential dating partner; Petty & Wegener, 1998).
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 795

strong group association cues were present, we expected implicit


attitudes to primarily reflect the valence associated with the social
group and thus not be strongly moderated by the CA information.
We tested these predictions by manipulating group association
cues related to obesity (Experiments 1 and 4), physical attractive-
ness (Experiment 2), and race (Experiment 3).
Finally, we anticipated that the group association cue would
impact explicit attitudes toward the target when the behavioral
statements describing the person were relatively uninformative
with respect to valence. Thus, in Experiment 4, we manipulated
whether the target individual was or was not obese, and we crossed
this factor with another manipulation that varied whether the
behavioral statements were clear-cut or ambiguous in terms of
valence. As in the previous studies, we expected implicit attitudes
to be influenced by the presence of a strong group association cue.
However, we also anticipated that the group association cue would
impact explicit attitudes toward the target under conditions where
the individuals behaviors were ambiguous (but not when they
were unambiguous). As noted previously, group membership
should have an impact on explicit attitudes toward the target only
when each behavior encountered is ambiguous with respect to
valence and, thus, the cue can influence how each behavior is
encoded at the time of encounter (e.g., Bodenhausen & Wyer,
1985). However, group cues should not affect explicit attitudes
Figure 1. Sample stimuli used to manipulate obesity (Experiments 1 and
toward the target when each behavior is clear-cut in terms of
4, top row) and race (Experiment 3, bottom row). Top-row photos are from valence (because each action is not subject to interpretation) even
Nosek et al. (2004) and bottow-row photos are from Minear and Park if, ultimately, the final attitude toward the individual is relatively
(2004). mixed in nature (which is more likely in the CA conditions).

Experiment 1
in some conditions). Next, participants received either additional
neutral (control) statements about the target or additional counter- In Experiment 1, we examined how implicit and explicit atti-
attitudinal (CA) statements about the target (i.e., the valence as- tudes formed and changed for members of a stigmatized group
sociated with these subsequent statements was the opposite of the (i.e., those who are overweight) relative to targets who are not
valence of the behavioral statements in the initial learning trials). stigmatized (i.e., those who are not overweight). This study repli-
The CA conditions allowed us to examine how attitudes would cated the basic experimental design of Rydell and McConnell
change in the face of new and conflicting behavioral information (2006), but it also manipulated a group association cue for the
about the target. Past research has shown that presenting a con- target. Specifically, on a between-subjects basis, participants
siderable number of CA behaviors (such as in the current work) formed attitudes toward a person, Bob, who appeared to be either
results in a much more moderated attitude toward the target person overweight or not overweight. In addition to seeing a photo of
(e.g., Kerpelman & Himmelfarb, 1971; Rydell & McConnell, Bob, participants were presented with a number of positive and
2006). Whether such revised attitudes toward the individual reflect negative verbal behavioral statements about him and asked to
relatively neutral or relatively ambivalent attitudes toward the determine whether each statement was characteristic of him. All
target person is less important for the present concerns than is the participants received the same behavioral statements; however,
fact that the introduction of CA information should produce mean- whether a behavior was characteristic or uncharacteristic of Bob
ingful shifts in attitudes toward the target. More important, we was manipulated systematically to indicate that Bob acted posi-
predicted that the introduction of CA information about a target tively (positive behaviors were characteristic and negative behav-
presented with strong group association cues would have a differ- iors were uncharacteristic of him) or negatively (negative behav-
ential impact on explicit and implicit attitudes toward the target iors were characteristic and positive behaviors were
person. uncharacteristic of him). Finally, participants implicit and explicit
In general, we expected that explicit attitudes toward the target attitudes were assessed using the exact same measures as were
person would respond to the valence described in the behavioral used in past research (e.g., Rydell & McConnell, 2006; Rydell et
statements and that they would change after the presentation of CA al., 2006).
information. Also, when no salient group association cue was In line with the prediction that the associative system would
present or when the cue was weakly associated with valence, we reflect the negativity associated with a group association cue and
expected that implicit attitudes toward the target person would the rule-based system would be sensitive to the valence of the
show a pattern similar to the pattern of explicit attitudes. That is, behavioral information provided when forming an attitude toward
similar to Rydell and McConnell (2006), when large amounts of an individual, it was expected that (a) explicit attitudes toward Bob
CA information are presented, implicit attitudes should eventually would reflect the valence suggested by the verbal statements
change in the absence of group association cues. However, when presented, (b) implicit attitudes would reflect the valence of the
796 MCCONNELL, RYDELL, STRAIN, AND MACKIE

group association cue that was salient (i.e., the overweight condi- in the CA condition (100 CA) received CA feedback about Bob on
tion would lead to negative implicit attitudes toward Bob regard- 100 trials (i.e., the behaviors that were described as characteristic
less of the valence of his behaviors), and (c) implicit attitudes or uncharacteristic of Bob were opposite of the valence presented
would be based on the behavioral information when no group during the initial learning trials).3 After completing the second
association cue was salient (i.e., in the not-overweight condition; block of 100 trials, participants completed measures assessing their
Rydell & McConnell, 2006). attitudes toward Bob.4
Explicit attitude measure. To assess explicit attitudes, we had
Method participants judge how likable Bob was on a scale ranging from 1
(very unlikable) to 9 (very likable). In addition, the participants com-
Participants. A sample of 133 undergraduates at Miami Uni- pleted five semantic differential scales, each using a 9-point scale to
versity participated in return for research credit in their introduc- describe Bob with anchors of good bad, pleasantmean, agreeable
tory psychology courses. They were randomly assigned to a 2 disagreeable, caring uncaring, and kind cruel. Further, participants
(Bobs weight: not overweight, overweight) 2 (valence of the
provided their evaluation of Bob on a feeling thermometer that ranged
initial verbal behaviors: positive, negative) 2 (CA condition:
in temperature from 0o to 100o. Following past research (e.g., Rydell
control [0 CA], CA conditioning [100 CA]) between-subjects
& McConnell, 2006; Rydell et al., 2006), we standardized the re-
factorial.
sponses for each explicit measure and computed an overall mean (in
Learning task. A modified version of Kerpelman and Him-
all experiments, s .90). Thus, higher scores indicated more pos-
melfarbs (1971) attitude learning paradigm was used (see Rydell
& McConnell, 2006; Rydell et al., 2006). In this learning task, itive explicit attitudes toward Bob.
participants received information about Bob on a computer over Implicit attitude measure. The Implicit Association Test (IAT;
the course of 200 trials. On the basis of random assignment, one of Greenwald et al., 1998) was used to assess implicit attitudes toward
four different White men served as the target Bob.2 On each trial, Bob, as implicit attitudes have been studied in previous research (e.g.,
participants were concurrently presented with a picture of Bob and Rydell & McConnell, 2006; Rydell et al., 2006). In this study, the IAT
verbal statements of behavior that might be characteristic of him. had 25 stimuli: 1 picture of Bob (Bob was either overweight or not
After reading each behavior, participants indicated whether they overweight), 4 different pictures of White men who were not Bob (2
believed that behavior was characteristic or uncharacteristic of were overweight and 2 were not), 10 positive adjectives (e.g., won-
Bob by pressing the c key or the u key, respectively. After each derful), and 10 negative adjectives (e.g., disgusting). All stimuli were
response, participants were given feedback about whether each presented in the center of the monitor and the adjectives were always
behavior was characteristic of Bob. Specifically, feedback con- presented in lowercase letters.
sisted of the word correct (in blue text) or incorrect (in red text) As in past work (e.g., Rydell & McConnell, 2006; Rydell et al.,
positioned in the center of the computer monitor and, at the same 2006), the IAT task featured seven blocks with 20 trials per block.
time, the behavior was stated correctly, on the basis of the assigned Participants were informed that the task involved making category
condition, at the bottom of the monitor (e.g., Helping the neigh- judgments using one of two responses (the d or k keys on the
borhood children is characteristic of Bob or Helping the neigh- keyboard) for a variety of stimuli (photos or words) presented on
borhood children is uncharacteristic of Bob). Thus, through sys- a computer monitor. During each block, verbal category label
tematically differing feedback (to be described), participants were reminders appeared on the left and right sides of the display
exposed to the same behaviors, but the reinforcement was designed (assignment of particular labels to the d and k keys was counter-
to indicate that Bob performed positive or negative acts. balanced across participants and produced no effects). Participants
Manipulation of Bobs weight. To manipulate whether Bob were instructed to complete that task quickly while also minimiz-
was perceived as overweight or not overweight, the picture of Bob ing errors, and they were told to keep their index fingers on the d
differed as a function of condition. Half of the participants saw a
and k keys throughout the experiment to minimize delays in
picture of Bob during each trial that showed he was not over-
responding. There was a 250-ms intertrial interval.
weight, but the rest saw a picture of Bob during each learning trial
where the photo of Bob had been morphed from the original (i.e.,
the picture in which Bob was not overweight) so that Bob appeared 2
This counterbalancing procedure produced no effects on any of the
to be overweight (see Nosek, Banaji, & Greenwald, 2004). Thus, results. Similar counterbalancing was used in the other experiments and
each not-overweight face was used to create an overweight face produced no effects as well.
that was almost identical except for apparent weight. 3
In the current work, we contrasted the 0 CA control condition (where
Manipulation of valence of the initial verbal information. Dur- no CA information was presented) with the 100 CA condition (where 100
ing the first 100 trials, half of the participants received feedback CA items were presented). We selected 100 CA for our comparison
that positive behaviors were characteristic of Bob and negative because past research (Rydell & McConnell, 2006) has shown that even
behaviors were uncharacteristic of Bob (positive initial verbal slow-changing implicit attitudes change after such a large number of CA
behaviors. Thus, if implicit attitudes continue to reflect group association
information). The remaining participants received feedback that
cue evaluations under conditions where, without the cue, they would be
negative behaviors were characteristic of Bob and positive behav-
significantly moderated, it would be an especially compelling demonstra-
iors were uncharacteristic of Bob (negative initial verbal informa- tion of the unresponsiveness of implicit attitudes to changing behavioral
tion). information that has been shown, in the absence of such cues, to produce
Manipulation of CA condition. After the first 100 trials, par- markedly changed implicit attitudes.
ticipants in the control condition received 100 neutral trials (i.e., 4
In all experiments reported in the current work, the order of attitude
the behavior characteristic of Bob was neither positive nor nega- measure (i.e., implicit before explicit vs. explicit before implicit) was
tive; e.g., Bob waited at the street corner). However, participants counterbalanced, and this factor did not qualify any of the results.
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 797

Explicit Attitudes Implicit Attitudes


1.5 0 CA
1.5 0 CA
100 CA 100 CA
1.0 1.0

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

6.8 4.1 1.9 4.4 6.6 4.7 2.3 3.5 221 -56 -24 118 -150 -179 -107 -141
-1.5 Positive Negative Positive Negative
-1.5 Positive Negative Positive Negative
Not overweight Overweight Not overweight Overweight

Figure 2. Explicit and implicit attitudes as a function of Bobs weight, valence of the initial verbal behaviors,
and counterattitudinal condition in Experiment 1. Standardized means are presented on the y-axis, and non-
standardized means are listed along the abscissa. CA counterattitudinal statements.

In Block 1, participants judged photos of Bob or not Bob, and in of each figure. Because the standardized measures provide the
Block 2, they judged whether the adjectives were negative or positive. most direct tests of the theoretical predictions in the current work,
In Blocks 3 and 4 (Combination 1), participants judged whether the in the Results and Discussion section, we focus on these data.7
stimuli were Bob or negative or not Bob or positive. In Block 5,
participants performed the same judgment task as they did in Block 2
except the assignment of response keys to the two valence categories Results and Discussion
was reversed. Finally, in Blocks 6 and 7 (Combination 2), participants
The attitude measures were examined with a 2 (Bobs weight)
judged whether the stimuli were Bob or positive or not Bob or
2 (valence of the initial verbal information) 2 (CA condition)
negative. As in past IAT research, half of the participants performed
2 (standardized attitude measure: implicit vs. explicit) mixed-
Combination 1 in Blocks 3 4 and Combination 2 in Blocks 6 7,
model analysis of variance (ANOVA), with the latter factor within
whereas the rest performed Combination 2 in Blocks 3 4 and Com-
subjects. Several results obtained, but of greatest importance was
bination 1 in Blocks 6 7 (this counterbalancing manipulation pro-
the four-way interaction that approached significance (see Figure
duced no effects).5
2), F(1, 125) 3.01, p .08.8 To better understand these data, we
To assess implicit attitudes toward Bob, we subtracted the mean
examined the three-way interactions of Bobs Weight Valence
response latencies of Combination 2 from the mean response
of the Initial Verbal Information CA Condition separately for
latencies of Combination 1 (see Greenwald et al., 1998, for de-
implicit and explicit attitudes.
tailed scoring information).6 As in past work (e.g., Rydell &
Explicit attitudes. For explicit attitudes, a main effect of va-
McConnell, 2006; Rydell et al., 2006, 2007), these difference lence of the initial verbal information was found, F(1, 125)
scores were standardized, with greater values indicating relatively 89.51, p .001. Specifically, for those initially receiving positive
more positive implicit attitudes toward Bob. Because IAT scores verbal information, participants reported more positive attitudes
have long been viewed as relative (rather than absolute) measures
of attitudes (e.g., Greenwald et al., 1998; Nosek, Greenwald, &
Banaji, 2006), standardization maintains their relativistic nature. 5
Within each block, an equal number of relevant stimuli were presented,
Moreover, by standardizing the implicit and explicit attitude mea- with the particular order of presentation being randomly determined for
sures and treating the type of attitude (implicit vs. explicit) as a each participant. Thus, in Blocks 1, 2, and 5, ten stimuli from the relevant
within-subjects factor, we can evaluate how implicit and explicit two categories were presented. In Blocks 3, 4, 6, and 7, five stimuli from
attitudes respond differently to the between-subjects manipula- the relevant four categories (i.e., Bob, not Bob, positive, negative) were
tions, testing the central predictions that group association cues presented. With the exception of the inclusion of group association cues,
the current IAT is identical to that used in past research (e.g., Rydell &
have differential effects on implicit and explicit attitudes. Thus, the
McConnell, 2006; Rydell et al., 2006).
discussion of the results focuses on analyses of these data. How- 6
Alternative IAT scoring approaches (e.g., Greenwald, Nosek, & Ba-
ever, to provide readers with a better sense of how measures varied
naji, 2003) produced identical results in the current work.
within and across experiments (where standardization makes com- 7
Parallel analyses conducted on the nonstandardized measures produced
parisons more difficult), each figure in the current study displays
similar results.
both the means for the standardized explicit and implicit attitude 8
Although only marginal in this study, the same group association cue
measures along the y-axis (because the inferential statistics were manipulation (i.e., obesity) was used again in Experiment 4 and revealed the
conducted on these values) and the means for nonstandardized predicted significant interaction. Also, this four-way interaction (using other
explicit (the means of the liking and semantic differential re- group association cues) was significant at conventional levels in both Exper-
sponses, each assessed on 9-point scales) and implicit (the IAT iments 2 and 3. However, because the current four-way interaction was
difference score, in milliseconds) attitude measures along the base marginal, some degree of caution should be exercised in its interpretation.
798 MCCONNELL, RYDELL, STRAIN, AND MACKIE

toward Bob (M 0.54) than did those initially receiving negative established association between obesity and negativity and were
verbal information (M 0.51). In addition, this effect was unaffected by the valence of the verbal information (initial or CA)
qualified by the expected interaction with CA condition, F(1, about him. Thus, when the group association cue was present,
125) 59.08, p .001. Simple effect analyses showed that for implicit attitudes toward the target reflected the valence of the
participants initially receiving positive verbal information, those in group association cue instead of the valence of the behavioral
the 0 CA condition had more positive attitudes toward Bob (M information provided.
1.01) than did those in the 100 CA condition (M 0.06), F(1,
125) 22.27, p .001. For those initially receiving negative Experiment 2
verbal information, the exact opposite pattern emerged, with those
in the 0 CA condition evaluating Bob more negatively (M Experiment 1 supported the systems of evaluation prediction
0.88) than those in the 100 CA condition (M 0.13), F(1, that strong group association cues (in this case, being overweight),
125) 44.10, p .001. The three-way interaction was not when present, would influence implicit attitudes toward a novel
significant, F(1, 125) 2.30, ns (see Figure 2, left panel). Thus, target. On the one hand, the valence of the behavioral statements
the CA information reversed the explicit attitudes that were determined explicit attitudes (in all cases) and implicit attitudes
strongly reflective of the initial verbal information. Also, note that when no salient group association cue was provided. This work
Bobs weight did not play any role in explicit attitudes toward him expands our earlier research (e.g., Rydell & McConnell, 2006;
whatsoever. Rydell et al., 2006) by showing that social groups with a strong
Implicit attitudes. In contrast, implicit attitudes showed a main association value (e.g., obesity is negative) are used by the asso-
effect of Bobs weight, F(1, 125) 32.43, p .001. That is, ciative system when available. In other words, the negativity
participants had more negative implicit attitudes toward the over- associated with Bob being overweight led to negative implicit
weight Bob (M 0.42) than toward the not-overweight Bob attitudes toward him even when the statements about his actions
(M 0.43). Thus, the group association cue had a direct impact on conveyed exclusively positive behavioral information. On the
implicit attitudes. Also, the two-way interaction between the va- other hand, because the behavioral statements were unambiguous
lence of the initial verbal information and CA condition was with respect to valence, the group association cue had no impact on
significant, F(1, 125) 8.30, p .001. For those who initially explicit attitudes toward the target person.
received positive verbal information, participants in the 0 CA Although this provides strong support for our predictions de-
condition had more positive implicit attitudes toward Bob (M rived from a systems of evaluation perspective, we anticipate that
0.30) than did those in the 100 CA condition (M 0.30), F(1, positive group association cues should produce similar outcomes
125) 6.01, p .02. For those who initially received negative for implicit attitudes as well, leading people to hold positive
verbal information, the opposite pattern emerged, as those in the 0 implicit attitudes toward a target described as performing numer-
CA condition held more negative implicit attitudes toward Bob ous negative behaviors. Thus, in Experiment 2, we examined a
(M 0.13) than did those in the 100 CA condition (M 0.12), different group association cue, physical attractiveness, which can
although this difference was not significant, F(1, 125) 1.98, ns. provide positive (attractive) and negative (unattractive) associa-
Although this two-way interaction suggests that implicit atti- tions. Indeed, a considerable amount of research on persuasion
tudes followed the valence of the initial verbal information and (Chaiken, 1979) and the what is beautiful is good effect (Dion et
subsequently were changed by the CA information just like ex- al., 1972) shows that attractive people are evaluated more posi-
plicit attitudes were, this two-way interaction was qualified by the tively than average individuals are and that unattractive people are
predicted three-way interaction with Bobs weight, F(1, 125) evaluated more negatively than average or attractive individuals
8.38, p .005 (see Figure 2, right panel). Specifically, the two- are.
way interaction between initial valence of the behavioral informa- In Experiment 2, participants learned about either an attractive
tion and CA information held for the not-overweight Bob, F(1, female, an average female, or an unattractive female named Bob-
125) 16.40, p .001, but was absent for the overweight Bob, bie. On the basis of our reasoning about which types of informa-
F(1, 125) 0.00, ns. In other words, for the not-overweight Bob, tion the associative and rule-based systems of evaluation would
those initially receiving positive verbal information about Bob had use, we expected that explicit attitudes toward Bobbie would
more positive implicit attitudes toward him in the 0 CA condition reflect the valence of the behavioral statements provided about her
(M 1.06) than did those in the 100 CA condition (M 0.03), and that implicit attitudes toward her in the absence of a distinctive
F(1, 125) 7.65, p .01. In the condition where initial verbal group association cue (i.e., when Bobbie was of average attrac-
information was negative, the opposite pattern emerged, with those tiveness) would do the same. However, when strong salient group
in the 0 CA condition having more negative implicit attitudes association cues were present (i.e., her physical attractiveness is
toward Bob (M 0.02) than those in the 100 CA condition (M salient), we expected that implicit attitudes toward Bobbie would
0.66), F(1, 125) 8.22, p .01. reflect the valence associated with the cue instead of the valence of
As expected, when no salient group association cue was present her behaviors, leading to relatively positive implicit attitudes to-
(i.e., not-overweight Bob), implicit attitudes toward Bob followed ward her when she was presented as being very attractive and
the same pattern as explicit attitudes, tracking the valence of the relatively negative implicit attitudes toward her when she was
large amount of behavioral information provided (Rydell & Mc- presented as being very unattractive.
Connell, 2006). That is, attitudes reflected the valence of the initial
verbal information, and these attitudes reversed after the presen-
Method
tation of a considerable amount of CA information. However,
when the group association cue of Bobs being overweight was Participants. A sample of 185 undergraduates at Miami Uni-
displayed, implicit attitudes toward him reflected the well- versity participated in return for research credit. They were ran-
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 799

Explicit Attitudes Implicit Attitudes


1.5 0 CA
1.5 0 CA
100 CA 100 CA
1.0 1.0

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

7.2 4.7 1.9 4.9 7.6 3.8 2.0 5.1 7.8 4.4 1.7 4.4 208 241 208 194 201 30 -29 149 12 4 -74 -2
-1.5 Positive Negative Positive Negative Positive Negative
-1.5 Positive Negative Positive Negative Positive Negative
Attractive Average Unattractive Attractive Average Unattractive

Figure 3. Explicit and implicit attitudes as a function of Bobbies attractiveness, valence of the initial verbal
behaviors, and counterattitudinal condition in Experiment 2. Standardized means are presented on the y-axis, and
nonstandardized means are listed along the abscissa. CA counterattitudinal statements.

domly assigned to a 3 (Bobbies attractiveness: attractive, average, more positively (M 0.50) than did those initially receiving
unattractive) 2 (valence of the initial verbal information: posi- negative verbal information about her (M 0.51). Also repli-
tive, negative) 2 (CA condition: 0 CA, 100 CA) between- cating the results of Experiment 1, this effect was qualified by CA
subjects factorial. condition, F(1, 173) 236.80, p .001. For those who initially
Procedure. The procedure for Experiment 2 was the same as received positive verbal information about Bobbie, participants in
the procedure in Experiment 1 with three exceptions. First, a the 0 CA condition had more positive explicit attitudes toward her
female target person, Bobbie, was used. Second, the not-Bobbie (M 1.14) than did those in the 100 CA condition (M 0.13),
pictures used in the Bobbie IAT were a mixture of other attractive, F(1, 173) 108.77, p .001. For those who initially received
average, and unattractive women. Third, to manipulate whether negative verbal information about her, the exact opposite pattern
Bobbie was attractive, average, or unattractive, we chose images of emerged, with those in the 0 CA condition evaluating Bobbie more
Bobbie that differed in their level of physical attractiveness. Spe- negatively (M 1.06) than those in the 100 CA condition (M
cifically, pictures were taken from an Internet dating Web site and 0.05), F(1, 173) 136.92, p .001. The three-way interaction
a face database (Minear & Park, 2004) and rated by a separate was not significant, F(2, 173) 0.80, ns (see Figure 3, left panel).
group of 40 participants from the same university (none of whom These analyses revealed two effects that paralleled those of Ex-
participated in the current study). On the basis of these ratings, on periment 1. First, explicit attitudes were very responsive to the
a scale ranging from 1 (extremely unattractive) to 9 (extremely valence of the initial verbal information and changed dramatically
attractive), two images were selected to be attractive Bobbies after participants received the CA information. Second, the group
(M 7.61), two images were selected to be average Bobbies (M association cue manipulation (i.e., Bobbies physical attractive-
5.38), and two images were selected to be unattractive Bobbies ness) did not qualify any of these effects.
(M 3.15).9 The attractiveness of these pictures differed signif- Implicit attitudes. However, implicit attitudes showed a main
icantly, F(2, 76) 205.68, p .001, with all three levels of effect of Bobbies attractiveness, F(2, 173) 46.04, p .001.
attractiveness being significantly different, ps .001. Specifically, participants had more positive implicit attitudes to-
ward the attractive Bobbie (M 0.71) than toward the unattractive
Results and Discussion Bobbie (M 0.66) or the average Bobbie (M 0.05), with the
latter two also differing significantly. In addition, there was a
The attitude measures were examined with a 3 (Bobbies phys-
Valence of the Initial Verbal Information CA Condition inter-
ical attractiveness) 2 (valence of the initial verbal informa-
action, F(2, 173) 8.61, p .005. This interaction showed that a
tion) 2 (CA condition) 2 (standardized attitude measure:
simple effect of CA condition was not significant for those in the
implicit vs. explicit) mixed-model ANOVA, with the latter factor
positive condition (0 CA M 0.25, 100 CA M 0.04), F(1,
being within subjects. Several results obtained, but of greatest
173) 1.92, ns, but it was significant in the negative condition (0
importance was the predicted four-way interaction, F(2, 173)
CA M 0.33, 100 CA M 0.11), F(1, 173) 4.63, p .04.
4.52, p .02, which is presented in Figure 3. To examine this
Most important, this effect was qualified by the predicted three-
effect, we examined the three-way interactions of Bobbies Phys-
way interaction, F(2, 173) 8.87, p .001 (see Figure 3, right
ical Attractiveness Valence of the Initial Verbal Information
panel). To explore this effect, we examined the interaction be-
CA Condition separately for implicit and explicit attitudes.
Explicit attitudes. Explicit attitudes once again showed a main
effect of valence of the initial verbal information, F(1, 173) 9
There was one blonde and one brunette Bobbie for each of the three
169.99, p .001. Similar to the results of Experiment 1, those who levels of attractiveness. The choice of target Bobbie (blonde vs. brunette)
initially received positive verbal information evaluated Bobbie was randomly determined, and this factor did not qualify any of the results.
800 MCCONNELL, RYDELL, STRAIN, AND MACKIE

tween valence of the initial verbal information and CA information evaluation associations in memory. In other words, as the cue-to-
for the attractive, average, and unattractive Bobbie conditions valence association grows weaker, implicit attitudes toward the
separately. For the attractive and the unattractive Bobbies, the individual should be increasingly reflective of the behavioral state-
two-way interactions were not significant, Fs 1, ns. Instead, ments about the person.
implicit attitudes toward the attractive Bobbie were positive re- With this logic in mind, in Experiment 3, we examined another
gardless of the valence of the verbal information, and implicit group association cue, a targets race. Specifically, we replicated
attitudes toward the unattractive Bobbie were negative regardless Experiment 1 but manipulated target race to either provide a
of the valence of the verbal information. However, the two-way distinctive group association cue (i.e., an African American Bob)
interaction was significant for the average Bobbie, F(1, 173) or not provide a distinctive group association cue (i.e., a White
33.34, p .001. For those who received positive verbal informa- Bob). In addition, we also assessed participants evaluative asso-
tion initially, those in the 0 CA condition had more positive ciations with the cue (i.e., their implicit attitudes toward African
implicit attitudes toward the average Bobbie (M 0.55) than did Americans in general) to examine the relation between their im-
those in the 100 CA condition (M 0.47), F(1, 173) 18.08, plicit evaluations of the social group cue and their attitudes toward
p .001. For those who received negative verbal information
a group target member in particular. We expected to replicate the
initially, the opposite pattern was found, with those in the 0 CA
findings of Study 1 using race as the group association cue, and we
condition having more negative implicit attitudes toward the av-
anticipated that implicit prejudice against African Americans
erage Bobbie (M 0.64) than those in the 100 CA condition
would account for the magnitude of negative implicit attitudes
(M 0.34), F(1, 173) 15.50, p .001.
toward Bob when he was African American. In other words,
Thus, Experiment 2 replicated the findings of Experiment 1.
participants with stronger racial prejudice should be less influ-
First, explicit attitudes and implicit attitudes (in the absence of a
salient group association cue) followed the valence of the verbal enced than those with less prejudice by the behavioral statements
information. Yet, when a distinctive group association cue was about an African American target when forming implicit attitudes
present, implicit attitudes reflected the evaluation associated with toward him. Therefore, we predicted an inverse relation between
that cue and not the behaviors performed by the target person. implicit racial prejudice and implicit (but not explicit) attitudes
Similar to Experiment 1, when the group association cue was toward Bob, but only when he was African American and not
negative (in this case, when the cue was the unattractive Bobbie), when he was White.
implicit attitudes toward her were negative even when the behav-
ioral statements conveyed positivity. Moreover, Experiment 2
Method
showed that when the group association cue was positive (i.e.,
when the cue was the attractive Bobbie), implicit attitudes toward Participants. A sample of 94 White undergraduates at the
her were positive, even in cases when the behavior statements University of California, Santa Barbara, participated in return for
suggested negativity. Once again, implicit attitudes reflected the research credit in their introductory psychology courses. They
valence of the salient group association cue when present, whereas were randomly assigned to a 2 (Bobs race: African American,
explicit attitudes toward the target were unaffected by this group White) 2 (valence of the initial verbal information: positive,
association cue and instead reflected the valence of the unambig- negative) 2 (CA condition: 0 CA, 100 CA) between-subjects
uous actions performed by the individual. factorial.
Procedure. The procedure was similar to the procedure of
Experiment 3 Experiment 1 with a few exceptions. First, in the current experi-
ment, we examined the group association cue of race by presenting
So far, we have shown that implicit attitudes can be unrespon-
an African American Bob to half of the participants or a White
sive to behavioral information when strong group association cues
Bob (as in Experiment 1) to the rest. Several minutes before
are available. We contend that the evaluations associated with
engaging in the learning task, participants completed a racial IAT
these cues dominate implicit attitudes because those attitudes are
where African American and White names were presented with
determined by a system of evaluation that is especially sensitive to
positive and negative adjectives using the same trial and block
associative information (Rydell et al., 2006). If this reasoning is
structure as was used with the IAT in Experiment 1 (see also
correct, the extent to which implicit attitudes are driven by these
group association cues should be related to the strength of the McConnell & Leibold, 2001). Thus, in one set of critical blocks of
association between the cue and evaluations of it. For example, the this racial IAT, participants judged whether the stimuli were Af-
overweight Bob in Study 1 revealed negative implicit attitudes rican American or negative or White or positive. In the other set
even in circumstances when he performed many positive behav- of critical blocks, they judged whether the stimuli were African
iors, presumably because most participants had strong associations American or positive or White or negative. The difference in
between obesity and negativity in memory (Nosek, 2005). Yet, mean response latencies for the critical blocks was computed, with
group association cues can be linked with valence to varying higher scores indicating relatively greater implicit prejudice
degrees. For example, although many individuals in American against African Americans (McConnell & Leibold, 2001). After
culture exhibit strong automatic associations between African the learning task (involving either an African American or a White
Americans and negativity (Devine, 1989; Greenwald et al., 1998), Bob target), participants completed the same implicit and explicit
there is meaningful variability in the extent to which people hold attitude measures used in Experiment 1, with the exception that the
such associations (Fazio et al., 1995; McConnell & Leibold, 2001). IAT presented non-Bob targets of the same race as the Bob target
Thus, we would anticipate that group association cues influence (to ensure it assessed implicit attitudes toward Bob specifically and
implicit attitudes more strongly for those with stronger cue- not racial prejudice more generally).
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 801

Results and Discussion toward the target. Also, an interaction between the valence of the
initial verbal information and CA condition was found, F(1, 86)
Attitudes toward Bob were examined with a 2 (Bobs race) 2 15.49, p .001. To examine this interaction, we analyzed the
(valence of the initial verbal information) 2 (CA condition) 2
simple effects of CA condition as a function of the valence of the
(standardized attitude measure: implicit, explicit) mixed-model
initial verbal information. When the valence of the initial verbal
ANOVA, with the latter factor being within subjects. Several
information was positive, participants in the 0 CA condition had
results obtained, but of greatest importance was the predicted
more positive implicit attitudes toward Bob (M 0.51) than did
four-way interaction, F(1, 86) 4.13, p .05, which is presented
those in the 100 CA condition (M 0.21), F(1, 86) 9.28, p
in Figure 4. To explore this outcome, we examined the three-way
.01. For those receiving initially negative verbal information, the
interactions of Bobs Race Valence of the Initial Verbal Infor-
opposite pattern emerged, with those in the 0 CA condition re-
mation CA Condition separately for implicit and explicit atti-
vealing more negative implicit attitudes toward Bob (M 0.40)
tudes.
than those in the 100 CA condition (M 0.24), F(1, 86) 7.12,
Explicit attitudes toward Bob. Replicating the results of Ex-
p .02. Thus, overall, implicit attitudes toward Bob reflected the
periments 12, explicit attitudes showed a main effect of valence
of the initial verbal information, F(1, 86) 24.83, p .001. Once valence of the verbal information (i.e., the valence of the initial
again, those who initially received positive verbal information behavioral information, which was undercut by the CA informa-
about Bob reported more favorable attitudes toward him (M tion), similar to the explicit attitudes toward Bob.
0.40) than did those who initially received negative verbal infor- But, unlike the explicit attitudes, this two-way interaction was
mation about him (M 0.43). In addition, this main effect was qualified by Bobs race in the predicted three-way interaction, F(1,
qualified by the interaction with CA condition, F(1, 86) 10.24, 86) 9.54, p .005 (see Figure 4, right panel). To explore this
p .005. Specifically, when initially receiving positive verbal effect, we examined the interaction between the valence of the
information about Bob, those in the 0 CA condition had more initial verbal information and CA condition for the White and
positive attitudes toward him (M 0.77) than did those in the 100 African American Bobs separately. For the White Bob, the two-
CA condition (M 0.02), F(1, 86) 10.24, p .005. However, way interaction was significant, F(1, 86) 16.20, p .001;
when initially receiving negative behavioral statements about Bob, however, it was not significant for the African American Bob, F(1,
those in the 0 CA condition had more negative attitudes toward 86) 0.01, ns. To examine this interaction for the White Bob, we
Bob (M 1.06) than did those in the 100 CA condition (M analyzed the simple effects of the CA condition as a function of the
0.21), F(1, 86) 28.31, p .001. The three-way interaction was valence of the initial verbal information. For those initially receiv-
not significant, F(1, 86) 0.04, ns (see Figure 4, left panel). Thus, ing positive verbal information about the White Bob, those in the
the two-way interaction revealed that CA information reversed the 0 CA condition had more positive implicit attitudes toward him
attitudes that reflected the valence of the initial information about (M 0.95) than did those in the 100 CA condition (M 0.20),
Bob. Yet, similar to the findings of Experiments 1 and 2, Bobs F(1, 86) 12.45, p .005. For those initially receiving negative
race did not play a role in any of these outcomes. verbal information about the White Bob, the opposite pattern
Implicit attitudes toward Bob. In stark contrast, implicit atti- emerged, with those in the 0 CA condition having more negative
tudes toward Bob revealed a main effect of Bobs race, F(1, 86) implicit attitudes toward him (M 0.53) than those in the 100
6.07, p .02. That is, participants had more negative implicit CA condition (M 0.68), F(1, 86) 12.24, p .005.
attitudes toward Bob when he was African American (M 0.19) In sum, these effects revealed that implicit attitudes toward a
than when he was White (M 0.23). Thus, as in Experiments 1 and target without a distinctive group association cue (i.e., the White
2, the group association cue had a direct effect on implicit attitudes Bob) reflected the valence of the verbal behaviors presented about

Explicit Attitudes Implicit Attitudes


1.5 0 CA
1.5 0 CA
100 CA 100 CA
1.0 1.0

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

5.9 4.4 1.7 5.4 7.0 4.7 1.8 4.5 229 9 -25 175 73 18 -9 32
-1.5 Positive Negative Positive Negative
-1.5 Positive Negative Positive Negative
White African-American White African-American

Figure 4. Explicit and implicit attitudes as a function of Bobs race, valence of the initial verbal behaviors, and
counterattitudinal condition in Experiment 3. Standardized means are presented on the y-axis, and nonstand-
ardized means are listed along the abscissa. CA counterattitudinal statements.
802 MCCONNELL, RYDELL, STRAIN, AND MACKIE

him, replicating the results of Experiments 1 and 2 and past work bias the encoding of ambiguous actions (e.g., Bodenhausen &
involving nondescript targets (Rydell & McConnell, 2006). How- Wyer, 1985).
ever, when a group association cue was present (i.e., the African In Experiment 4, we revisited the group association cue used in
American Bob), the implicit attitudes toward him were reflective Experiment 1 by manipulating Bobs apparent weight in a more
of the valence of the group association cue, as found in Experi- simplified experimental design. Specifically, participants were
ments 1 and 2. only presented with 100 statements about Bob (whose weight was
Prejudice against African Americans. To examine if negative manipulated between subjects) and were told that each statement
associations with the cue (i.e., prejudice against African Ameri- was characteristic of him. As part of another between-subjects
cans) can account for the implicit attitudes toward the African factor, half of the participants read statements indicating that Bob
American Bob being negative, we explored the extent to which performed unambiguous positive acts whereas the rest read state-
participants implicit prejudice toward African Americans pre- ments that were relatively ambiguous (i.e., not strongly valenced).
dicted their attitudes toward Bob. In our sample, the average In the latter case, we predicted that the group association cue
participant revealed relatively strong implicit racial prejudice would have an assimilative effect, resulting in a relatively negative
against African Americans (M 207.88 ms IAT effect, d 1.35). explicit attitude toward Bob when he was obese. Such a finding
In essence, this effect reaffirms the relative negativity participants would not only be valuable to test the importance of behavioral
associated with the group association cue (i.e., being African ambiguity in how group association cues affect explicit attitudes,
American). Next, we examined the correlations among partici- but it would also demonstrate that participants in our studies are
pants racial prejudice, explicit attitudes toward Bob, and implicit not reticent to report negative explicit attitudes toward members of
attitudes toward Bob separately as a function of the race condition. stigmatized groups (i.e., perhaps the lack of effect of cues in
As expected, when Bob was White, there were no relations be- previous studies reflects engaging in positive impression manage-
tween implicit racial prejudice and implicit attitudes toward Bob ment). However, when Bobs behaviors were unambiguously pos-
(r .05, ns) or explicit attitudes toward him (r .07, ns). itive, we expected relatively positive explicit attitudes toward Bob
However, as predicted, a different pattern emerged when Bob was regardless of his weight, replicating the results of Experiment 1.
African American. Although participants implicit racial prejudice We chose positive unambiguous behaviors in this study to provide
was unrelated to their explicit attitudes toward Bob (r .14, ns), the best opportunity for Bobs stigma to impact attitudes toward
implicit racial prejudice was significantly negatively correlated to him (i.e., avoid floor effects).
their implicit attitudes toward him (r .50, p .001). That is, In addition to testing our reasoning that group association cues
the more negativity they associated with African Americans, the could impact explicit attitudes toward an individual whose behav-
less positive their feelings toward Bob were on implicit (but not ior was relatively ambiguous, we also modified our IAT task in the
explicit) attitude measures, but only when he was Black. As current experiment. In the previous three experiments, we used
expected, the relation between racial prejudice and implicit atti- images of people as IAT stimuli to render Bob (or Bobbie) versus
tudes toward Bob differed as a function of race, z 2.81, p .01, not-Bob (or not-Bobbie) categorizations. It is possible that when
but there were no race condition differences in the relation be- making these judgments, participants could have been led by each
tween racial prejudice and explicit attitudes toward Bob, z 1. presentation of Bob (or Bobbie) in conditions involving a group
These data indicate that the magnitude of the valence associated association cue to strengthen their association between the target
with the group association cue can account for how Bobs race led person and the targets stigma. For example, for participants ex-
to relatively negative implicit attitudes toward him when he was a posed to an overweight Bob, the IAT task continually re-presents
member of that social group. images of an obese Bob throughout the IAT task, which may have
served to further reinforce negativity toward Bob. Also, the alter-
Experiment 4 native targets (e.g., the not Bobs) provided distractors that some-
times did and sometimes did not present the group association cue
To this point, we have shown in three different experiments as well, which could introduce unwanted context effects. To any
using three different group association cues that implicit attitudes extent that the group association cue was re-presented during the
toward an individual reflect the valence (and, in Experiment 3, the IAT task, the possibility that the implicit attitude measure toward
extremity of valence) of a salient group association cue when such Bob (or Bobbie) reflects a confound of attitudes toward the target
cues are present but that they are responsive to the valence of the and attitudes toward the group association cue itself exists. To
behaviors describing the target when such cues are absent or when eliminate this possibility, in the current experiment, we used an
the cues have relatively weaker associations with valence. Yet, in IAT task that presented names and not images of the target and
each of these studies, explicit attitudes were unaffected by the five nontargets. Because the current IAT task did not present visual
group association cues. At first blush, these results may seem images of people, we avoided the possibility that the implicit
difficult to reconcile with findings in the literature showing that attitude measure was assessing a blend of attitudes toward the
group membership can impact judgments. We have argued that target and attitudes toward the group association cue.
because the behavioral statements ascribed to the target individuals To summarize, the overall design of the study crossed Bobs
in the current experiments were both numerous and clear-cut with weight (overweight vs. not overweight) with type of behavioral
respect to valence, the ability of the group association cue to information (100 positive vs. 100 ambiguous) in a more simplified
induce assimilation effects on explicit attitudes was effectively experimental paradigm and with a modified IAT task. For implicit
curtailed. However, we would anticipate that if the target-relevant attitudes, a main effect of Bobs weight was expected, revealing
behaviors were more ambiguous with respect to valence instead of more negative implicit attitudes toward Bob when he was over-
being clear-cut, group association cues would have an impact on weight than when he was not (thus replicating the results of
explicit attitudes toward the individual by providing a means to Experiment 1). However, for explicit attitudes, we predicted an
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 803

interaction, such that the group association cue (i.e., overweight subjects. As expected, we observed the predicted three-way inter-
Bob) would reduce the positivity of explicit attitudes toward Bob action, F(1, 43) 4.16, p .05, which is illustrated in Figure 5.
when his behaviors were ambiguous with respect to valence. To explore this effect, the two-way interaction of Bobs weight and
statement type were examined separately for implicit and explicit
Method attitudes.
Explicit attitudes. The Bobs Weight Statement Type
Participants. A sample of 47 undergraduates at Miami Uni- ANOVA yielded three effects. First, not surprisingly, there was a
versity participated in return for research credit in their introduc- main effect of statement type, F(1, 43) 15.45, p .001,
tory psychology courses. They were randomly assigned to a 2 revealing that explicit attitudes toward Bob were more positive
(Bobs weight: not overweight, overweight) 2 (statement type: when the 100 statements suggested positivity (M 0.55) than
positive, ambiguous) between-subjects factorial. when they were ambiguous with respect to valence (M 0.57).
Procedure. Participants were presented with 100 behavior Also, there was a main effect of Bobs weight, F(1, 43) 10.65,
statements about Bob and told that they were all characteristic of
p .01, indicating that explicit attitudes toward Bob were more
him. Each statement was presented on the monitor for 8 s. On the
negative when he was overweight (M 0.29) than when he was
basis of pretested norms, participants assigned to the positive
not (M 0.30). It is important to note that this effect was qualified
statement condition read 100 statements that implied positivity
by an interaction with statement type, F(1, 43) 6.53, p .02
(e.g., Bob helped friends move into a new house), whereas those
(see Figure 5, left panel). Although explicit attitudes toward the
in the ambiguous statement condition read 100 statements that
Bob described by unambiguous positive behaviors did not differ as
were relatively valence neutral (e.g., Bob watched TV with
a function of his weight, F(1, 22) 0.75, ns, the same was not true
friends). The image of Bob presented on the monitor and asso-
when his behaviors were ambiguous, F(1, 21) 9.84, p .01,
ciated with each statement was either overweight or not over-
with Bob being viewed more negatively when he was overweight
weight, depending on condition assignment (using the same stim-
(M 1.19) than when he was not (M 0.01). Thus, as
uli as Experiment 1).
hypothesized, the group association cue did impact explicit atti-
Next, participants completed implicit and explicit attitude mea-
tudes toward the target individual, but only when his behaviors
sures toward Bob (once again counterbalanced). The explicit mea-
were not clear-cut with respect to valence.
sures were identical to those used in Experiment 1. However, the
Implicit attitudes. In contrast to the explicit attitudes data, the
implicit measure was a slightly modified version of the IAT.
Specifically, it was identical to the IAT used in Experiment 1 Bobs Weight Statement Type ANOVA for implicit attitudes
except that rather than the presented images being Bob and not- toward Bob revealed only a main effect of Bobs weight, F(1,
Bob targets, the person-related stimuli were names presented in 43) 13.86, p .001. Figure 5 (right panel) shows that partici-
uppercase font (positive and negative adjectives were presented in pants had more negative implicit attitudes when Bob was over-
lowercase), either BOB or five not-Bob names that began with the weight (M 0.49) than when he was not (M 0.48). Thus,
same letter (e.g., BEN). There were an equal number of presenta- impact of the group association cue was strong in all conditions.
tions of Bob and non-Bob names in each block. In sum, these findings show that a group association cue (i.e.,
Bobs weight) can affect explicit attitudes, but only when the
targets actions are relatively ambiguous and thus capable of being
Results and Discussion
assimilated by the group association cue. However, when the
Attitudes toward Bob were examined with a 2 (Bobs weight) targets behaviors were unequivocal, the group association had no
2 (statement type) 2 (standardized attitude measure: implicit, impact on explicit attitudes, replicating the results of Experiments
explicit) mixed-model ANOVA, with the latter factor being within 13. These findings obtained using a more simplified attitude-

Explicit Attitudes Implicit Attitudes


1.5 Positive
1.5 Positive
Ambiguous Ambiguous
1.0 1.0

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

7.5 6.5 7.4 5.0 207 189 36 25


-1.5 -1.5
Not overweight Overweight Not overweight Overweight

Figure 5. Explicit and implicit attitudes as a function of Bobs weight and statement type in Experiment 4.
Standardized means are presented on the y-axis, and nonstandardized means are listed along the abscissa. CA
counterattitudinal statements.
804 MCCONNELL, RYDELL, STRAIN, AND MACKIE

learning paradigm and using a modified IAT designed to circum- toward targets is influenced by them. On the basis of a systems of
vent possible confounds that might exist with presenting target evaluation perspective on attitudes (e.g., Rydell & McConnell,
stimuli with the group association cue. It is interesting to note that 2006; Rydell et al., 2006), a number of novel hypotheses were
there was no evidence that the type of statement (unambiguously advanced. For example, because many social group cues have
positive vs. ambiguous) qualified the main effect of Bobs weight strong associations with valence (e.g., being physically attractive is
on implicit attitudes. Although the type of statement did impact desirable, being obese is undesirable), we anticipated that such
explicit attitudes toward Bob, it appears that the strong group cues would have an especially strong impact on implicit attitude
association cue (i.e., Bobs obesity) had a greater impact on formation for individuals because such attitudes rely on associative
implicit attitudes toward him. These data may, at first glance, seem knowledge.
at odds with the earlier studies showing that the valence of the Indeed, across four experiments, we found strong and consistent
behaviors impacted implicit attitudes toward targets when group support for this prediction. For example, implicit attitudes toward
association cues were absent. However, it should be noted that the members of stigmatized groups were negative regardless of the
current study differed from the first three experiments in that the valence of the behaviors attributed to these individuals. These
valence manipulations in the former studies pitted two starkly outcomes were observed for a wide variety of stigmas, including
different valence conditions (i.e., 100 positive vs. 100 negative) being overweight (Experiments 1 and 4), being physically unat-
against each other, whereas the current manipulation (designed to tractive (Experiment 2), and being African American (Experiment
introduce ambiguity rather than polar-opposite valences) was far 3). However, when groups were associated with positivity (i.e.,
more modest. Thus, it appears that information (i.e., group asso- being physically attractive in Experiment 2), implicit attitudes
ciation cues) that is especially attuned to the system of evaluation toward the target were positive, again regardless of the nature of
underlying implicit attitudes (i.e., the associative system) has a the individuals actions. Overall, these data suggest that group
greater impact on the attitudes produced. association cues have an especially strong impact on implicit
In a similar vein, readers comparing the outcomes for explicit attitudes because such evaluations are based on a system of eval-
attitudes toward the not-overweight Bob between Experiments 1 uation that uses associative knowledge (Rydell & McConnell,
and 4 might conclude that similar explicit attitudes resulted despite 2006; Sloman, 1996; Smith & DeCoster, 2000).
very different behavior presentations (i.e., 0 CA vs. 100 CA in Further, these group association cues did not have much impact
Experiment 1 when positive information was initially presented, on explicit attitudes when the targets behavioral descriptions were
positive vs. ambiguous information in Experiment 4). However, clear-cut with respect to valence. However, when the targets
this apparent similarity actually reflects a by-product of the stan- actions were ambiguous, group association cues influenced ex-
dardization process. That is, although the standardized explicit plicit attitudes toward targets as well (Experiment 4). Thus, in
attitudes between the two studies were nearly identical for the cases where ambiguity exists, social groups can serve as an acces-
not-overweight Bob in Experiment 1 (i.e., initially positive behav- sible construct to promote assimilation effects (Bruner, 1957;
ioral characteristics followed by the 100 CA condition) and the Higgins, 1989; Srull & Wyer, 1979). Yet, when a targets actions
not-overweight Bob in Experiment 4 (i.e., the ambiguous behavior were unequivocal in their implications, group cues did not play a
condition), mean nonstandardized explicit attitudes were much role in explicit evaluations.
more positive in the latter case (M 6.5) than the former case It is also important to note that although we used group asso-
(M 4.1), t(26) 4.49, p .001, reflecting the absence of ciation cues that are widely held in society in the current research,
negative behavioral information about Bob in Experiment 4. In there will be meaningful variability in the extent to which these
other words, although the standardization process appears to sug- cues are associated with valence. Accordingly, in Experiment 3,
gest similar explicit attitudes when comparing between studies we saw that individual differences in the extent to which African
(which is not what one would expect from such markedly different Americans as a group were associated with negativity directly
behavioral presentations), inspection of the nonstandardized values predicted the degree to which implicit attitudes toward a novel
(see the bottom of Figures 2 and 5) indicates that, indeed, attitudes African American target were negative as well. This finding fur-
toward the not-overweight Bob who performed ambiguous behav- ther reaffirms that it is the overarching association between group
iors in Experiment 4 were far more positive than were attitudes cues and valence that determines how a targets group identity
toward the not-overweight Bob in Experiment 1 who performed shapes the formation of implicit attitudes toward the individual.
100 positive behaviors followed by 100 negative behaviors. Al- Stronger cue-to-valence associations should result in more extreme
though the focus on the standardized data in the current analyses implicit attitudes being formed more quickly. Further, meaningful
can make comparisons between studies somewhat more difficult, differences in ones past history of group associations will play an
the value they provide within studies to directly test the key important role in how implicit attitudes are affected by group
theoretical questions of interest (i.e., how group association cues association cues. Moreover, this outcome indicates that implicit
differentially impact implicit and explicit attitudes) is substantial. attitudes toward an individual should be more sensitive to the
valence of a targets behavioral information when cue-to-valence
General Discussion associations are relatively weak (just like they are in the absence of
salient group association cues).
In the current work, we explored how a targets social group that The current work sheds light on a number of important issues
is strongly associated with valence affects the formation of atti- involved in understanding others. For instance, most research
tudes toward the individual. Whereas a considerable amount of exploring group prejudice has focused on its pervasiveness, ex-
research has focused on the impact of social categorization on pression, and assessment (e.g., Devine, 1989; Dovidio et al., 2002;
explicit attitudes toward people (Fiske et al., 1999), the current Greenwald et al., 1998; McConnell & Leibold, 2001) rather than
study is the first to consider how the formation of implicit attitudes on its specific impact when people are forming attitudes toward
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 805

individuals. And although some important work, both empirical relevant cue. Thus, not all cues will have the same impact across
and theoretical, has been directed at considering the impact of individuals.
social groups on attitude formation (e.g., Brewer, 1988; Fiske & In addition to these important conceptual issues, the current
Neuberg, 1990), past work has not considered the implications of findings suggest a number of sizable roadblocks in reducing prej-
social groups for the formation of implicit attitudes toward indi- udice and discrimination. First, to the extent that stigmas impact
viduals. Given many striking demonstrations of how implicit atti- implicit attitudes more strongly than they do explicit attitudes, it
tudes uniquely predict behavior toward members of social groups may often be the case that people will be unaware of their stigma-
(e.g., Fazio et al., 1995; McConnell & Leibold, 2001) and toward related biases because the biases are associative in nature, which,
individuals without any distinctive group identification (Rydell & in turn, makes it less likely that correction processes will be used
McConnell, 2006), in the current work, we engaged an underex- (Wegener & Petty, 1995). Further, such nonconscious biases may
amined intersection of important issues with a framework for elicit behavioral confirmation from targets (e.g., Chen & Bargh,
considering how particular types of information (i.e., associative 1997), perpetuating such evaluations. Also, situational factors that
vs. rule based) are especially likely to impact particular types of reduce the impact of the rule-based system of evaluation (e.g.,
attitudes (Rydell & McConnell, 2006; Rydell et al., 2006). In distraction, off-peak circadian rhythms) should exacerbate the
general, the systems of evaluation perspective anticipated how influence of the associative system in directing behavior, increas-
group association cues would impact implicit and explicit attitude ing the likelihood that negative implicit attitudes will guide actions
formation toward a novel individual quite well. toward stigmatized targets (especially negative, nonverbal behav-
More generally, the current work points to the need to further iors; see McConnell & Leibold, 2001; Richeson & Shelton, 2003).
develop models of impression formation to include implicit atti- It is interesting that these group association cues had little effect
tudes toward individuals, and we believe the systems of evaluation on explicit attitudes except under conditions where behavioral
approach provides a compelling framework for doing so. In addi- information was ambiguous. This suggests that although group
tion to considering the extent to which perceivers expend cognitive association cues may play an important role in deliberate judg-
resources in attitude formation (e.g., Fiske & Neuberg, 1990), the ments and evaluations, their impact may be reduced in situations
systems of evaluation perspective suggests that the fit between that are less ambiguous in nature. Does this mean that the conse-
information type and attitude type matters too. That is, correspon- quences of stigma are less important than the literature suggests?
dence between the type of knowledge (i.e., associative vs. rule We believe the answer to this question is no. First, although most
based) and the type of attitude (i.e., implicit vs. explicit) is an of the behavioral statements presented about the targets in the
important dimension to consider in the attitude formation process. current study were clear-cut, most social interactions contain con-
Although in the current work we focused on social group cues as siderable ambiguity, increasing the likelihood of biased assimila-
one form of associative knowledge, we would anticipate that many tion (see Experiment 4). Further, in the current work, participants
other types of cues strongly associated with valence (e.g., experts were compelled to process a large number of statements about the
are good) would be especially influential in implicit attitude for- target individuals. However, in real life, people may be far less
mation as well. attentive to individuating information, especially for members of
Extending this point, we believe that the current work can shed stigmatized groups (Kurzban & Leary, 2001). Thus, in many cases,
new light on processes involved in attitudes and persuasion. For stigmas may dissuade perceivers from encountering information
example, several models of attitudes propose that heuristic and that could present targets in a much more positive light. Moreover,
peripheral cues influence attitudes and behavior more strongly even if such behaviors are encountered, the extent to which people
when ones motivation to process detailed information is low (e.g., effortfully individuate such information in such cases may be
Chaiken, 1979; Petty & Wegener, 1998). But what underlies this limited (Fiske & Neuberg, 1990). At any rate, it is clear that
outcome? When one considers that nonconscious associations (Ry- additional work is needed to examine how these trade-offs operate
dell et al., 2006) and association-based cues (the current work) in more complex social interaction situations.
play critical roles in determining implicit attitudes and that recent To conclude, the current work shows that the formation of
work reveals that implicit attitudes are more likely than explicit implicit attitudes toward members of social groups may often
attitudes to determine spontaneous behaviors that do not involve reflect the valence of group association cues instead of the behav-
deliberation and planning (e.g., Dovidio et al., 2002; Jellison, ioral data available to the perceiver. Explicit attitudes, however,
McConnell, & Gabriel, 2004; McConnell & Leibold, 2001; Rydell were less influenced by these cues and more determined by de-
& McConnell, 2006), a mechanism to account for how these scriptions of the target persons behaviors, unless the available
association-based cues influence behavior in low-effort situations behavioral information was ambiguous with respect to evaluations.
becomes apparent. That is, cues such as physical attractiveness or This research shows that stigmatized people may face challenges
others not considered in the current work (e.g., expertise) will in changing others implicit attitudes toward them, even when they
shape implicit attitudes, which, in turn, are more likely to guide perform good deeds, whereas members of highly valued groups
behavior in situations where the rule-based system does not (e.g., can behave badly and still enjoy others implicit approbation.
no verbal information is available for central route persuasion) or Although knowledge of the impression formation process with
cannot (e.g., limited cognitive resources) operate. Thus, a systems respect to explicit attitudes is well-developed, the current study
of evaluation perspective suggests that implicit attitudes may serve reveals that the formation of implicit attitudes toward individuals
as a mechanism to explain how association-based cues impact can operate quite differently (see also Rydell & McConnell, 2006;
behavior in some situations. Moreover, as the current Experiment Rydell et al., 2006). In sum, we believe that a systems of evalu-
3 suggests, the impact of these cues in shaping implicit attitudes ation perspective offers a very insightful theory and useful tools
varies on the basis of idiosyncratic associations with the target- for building an understanding of implicit attitude formation pro-
806 MCCONNELL, RYDELL, STRAIN, AND MACKIE

cesses, and it demonstrates the importance of appreciating the Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). The continuum model: Ten
consequences of these nonconscious evaluations as well. years later. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in
social psychology (pp. 231254). New York: Guilford Press.
References Fiske, S. T., & Neuberg, S. L. (1990). A continuum of impression forma-
tion, from category based to individuating processes: Influences of
Ambady, N., & Rosenthal, R. (1993). Half a minute: Predicting teacher information and motivation on attention and interpretation. In M. P.
evaluations from thin slices of nonverbal behavior and physical attrac- Zanna (Ed.), Advances in experimental social psychology (Vol. 23, pp.
tiveness. Journal of Personality and Social Psychology, 64, 431 441. 174). San Diego, CA: Academic Press.
Bargh, J. A. (1999). The cognitive monster. In S. Chaiken & Y. Trope Gawronski, B., & Bodenhausen, G. V. (2006). Associative and proposi-
(Eds.), Dual process theories in social psychology (pp. 361382). New tional processes in evaluation: An integrative review of implicit and
York: Guilford Press. explicit attitude change. Psychological Bulletin, 132, 692731.
Bargh, J. A., & Chartrand, T. L. (1999). The unbearable automaticity of Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition:
being. American Psychologist, 54, 462 479. Attitudes, self-esteem, and stereotypes. Psychological Review, 102,
Berscheid, E., & Walster, E. (1974). Physical attractiveness. Advances in 4 27.
Experimental Social Psychology, 7, 211276. Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring
Bodenhausen, G. V., & Wyer, R. S., Jr. (1985). Effects of stereotypes on individual differences in implicit cognition: The implicit association test.
decision making and information-processing strategies. Journal of Per- Journal of Personality and Social Psychology, 74, 1464 1480.
sonality and Social Psychology, 48, 267282. Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding
Brewer, M. B. (1988). A dual process model of impression formation. In and using the Implicit Association Test: I. An improved scoring algo-
T. K. Srull & R. S. Wyer (Eds.), Advances in social cognition (Vol. 1, rithm. Journal of Personality and Social Psychology, 85, 197216.
pp. 136). Hillsdale, NJ: Erlbaum. Higgins, E. T. (1989). Knowledge accessibility and activation: Subjectivity
Bruner, J. S. (1957). On perceptual readiness. Psychological Review, 64, and suffering from unconscious sources. In J. S. Uleman & J. A. Bargh
123152. (Eds.), Unintended thought (pp. 75123). New York: Guilford Press.
Castelli, L., Zogmaister, C., Smith, E. R., & Arcuri, L. (2004). On the Jellison, W. A., McConnell, A. R., & Gabriel, S. (2004). Implicit and
automatic evaluation of social exemplars. Journal of Personality and
explicit measures of sexual orientation attitudes: Ingroup preferences
Social Psychology, 86, 373387.
and related behaviors and beliefs among gay and straight men. Person-
Chaiken, S. (1979). Communicator physical attractiveness and persuasion.
ality and Social Psychology Bulletin, 30, 629 642.
Journal of Personality and Social Psychology, 37, 13871397.
Kerpelman, J. P., & Himmelfarb, S. (1971). Partial reinforcement effects in
Chen, M., & Bargh, J. A. (1997). Nonconscious behavioral confirmation
attitude acquisition and counterconditioning. Journal of Personality and
processes: The self-fulfilling consequences of automatic stereotype ac-
Social Psychology, 19, 301305.
tivation. Journal of Experimental Social Psychology, 33, 541560.
Kurzban, R., & Leary, M. R. (2001). Evolutionary origins of stigmatiza-
Crandall, C. S., DAnello, S., Sakalli, N., Lazarus, E., Nejtardt, G. W., &
tion: The functions of social exclusion. Psychological Bulletin, 127,
Feather, N. T. (2001). An attribution-value model of prejudice: Anti-fat
187208.
attitudes in six nations. Personality and Social Psychology Bulletin, 27,
McConnell, A. R., & Leibold, J. M. (2001). Relations among the Implicit
30 37.
Association Test, discriminatory behavior, and explicit measures of
Crocker, J., Major, B., & Steele, C. (1998). Social stigma. In D. T. Gilbert,
racial attitudes. Journal of Experimental Social Psychology, 37, 435
S. T. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th
442.
ed., Vol. 2, pp. 504 553). New York: McGraw-Hill.
Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and McGuire, W. J., McGuire, C. V., Child, P., & Fujioka, T. (1978). Salience
controlled components. Journal of Personality and Social Psychology, of ethnicity in the spontaneous self-concept as a function of ones ethnic
56, 680 690. distinctiveness in the social environment. Journal of Personality and
Dion, K. (1973). Young childrens stereotyping of facial attractiveness. Social Psychology, 36, 511520.
Developmental Psychology, 9, 183188. Miller, D. T., Taylor, B., & Buck, M. L. (1991). Gender gaps: Who needs
Dion, K., Berscheid, E., & Walster, E. (1972). What is beautiful is good. to be explained? Journal of Personality and Social Psychology, 61,
Journal of Personality and Social Psychology, 24, 285290. 512.
Dovidio, J. F., Kawakami, K., & Gaertner, S. L. (2002). Implicit and Minear, M., & Park, D. C. (2004). A lifespan database of adult facial
explicit prejudice and interracial interaction. Journal of Personality and stimuli. Behavior Research Methods, Instruments, & Computers, 36,
Social Psychology, 82, 62 68. 630 633.
Duncan, B. L. (1976). Differential social perception and attribution of Neuberg, S. L., Smith, D. M., Hoffman, J. C., & Russell, F. J. (1994).
intergroup violence: Testing the lower limits of stereotyping of Blacks. When we observe stigmatized and normal individuals interacting:
Journal of Personality and Social Psychology, 34, 590 598. Stigma by association. Personality and Social Psychology Bulletin, 20,
Eagly, A. H., Ashmore, R. D., Makhijani, M. G., & Longo, L. C. (1991). 196 209.
What is beautiful is good, but . . . : A meta-analytic review of research Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know:
on the physical attractiveness stereotype. Psychological Bulletin, 110, Verbal reports on mental processes. Psychological Review, 84, 231259.
109 128. Nosek, B. A. (2005). Moderators of the relationship between implicit and
Fazio, R. H. (1995). Attitudes as object-evaluation associations: Determi- explicit evaluation. Journal of Experimental Psychology: General, 134,
nants, consequences, and correlates of attitude accessibility. In R. E. 565584.
Petty & J. A. Krosnick (Eds.), Attitude strength: Antecedents and con- Nosek, B. A., & Banaji, M. R. (2001). The Go/No-Go Association Task.
sequences (pp. 247282). Mahwah, NJ: Erlbaum. Social Cognition, 19, 625 666.
Fazio, R. H., Jackson, J. R., Dutton, B. C., & Williams, C. J. (1995). Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2004). Project Implicit.
Variability in automatic activation as an unobtrusive measure of racial Retrieved October 15, 2005, from http://implicit.harvard.edu
attitudes: A bona fide pipeline? Journal of Personality and Social Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2006). The Implicit
Psychology, 69, 10131027. Association Test at age 7: A methodological and conceptual review. In
Fiske, S. T. (1998). Stereotyping, prejudice, and discrimination. In D. T. J. A. Bargh (Ed.), Social psychology and the unconscious: The automa-
Gilbert, S. T. Fiske, & G. Lindzey (Eds.), Handbook of social psychol- ticity of higher mental processes (pp. 265292). New York: Psychology
ogy (4th ed., Vol. 2, pp. 357 411). New York: McGraw-Hill. Press.
IMPLICIT AND EXPLICIT ATTITUDE FORMATION 807

Petty, R. E., & Wegener, D. T. (1998). Attitude change: Multiple roles for Sagar, H. A., & Schofield, J. W. (1980). Racial and behavioral cues in
persuasion variables. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), Black and White childrens perceptions of ambiguously aggressive acts.
Handbook of social psychology (4th ed., Vol. 1, pp. 323390). New Journal of Personality and Social Psychology, 39, 590 598.
York: McGraw-Hill. Schwarz, N., & Bohner, G. (2001). The construction of attitudes. In A.
Plant, E. A., & Devine, P. G. (1998). Internal and external motivation to Tesser & N. Schwarz (Eds.), Blackwell handbook of social psychology:
respond without prejudice. Journal of Personality and Social Psychol- Intrapersonal processes (pp. 436 457). Oxford, UK: Blackwell.
ogy, 75, 811 832. Sloman, S. A. (1996). The empirical case for two systems of reasoning.
Pryor, J. B., Reeder, G. D., Yeadon, C., & Hesson-McInnis, M. (2004). A Psychological Bulletin, 119, 322.
dual-process model of reactions to perceived stigma. Journal of Person- Smith, E. R., & DeCoster, J. (2000). Dual-process models in social and
ality and Social Psychology, 87, 436 452. cognitive psychology: Conceptual integration and links to underlying
Richeson, J. A., & Shelton, J. N. (2003). When prejudice does not pay: memory systems. Personality and Social Psychology Review, 4, 108
Effects of interracial contact on executive function. Psychological Sci- 131.
ence, 14, 287290. Srull, T. K., & Wyer, R. S. (1979). The role of category accessibility in the
Rozin, P., Markwith, M., & Nemeroff, C. (1992). Magical contagion interpretation of information about persons: Some determinants and
beliefs and fear of AIDS. Journal of Applied Social Psychology, 22, implications. Journal of Personality and Social Psychology, 37, 1660
10811092. 1672.
Rudman, L. A., Feinberg, J., & Fairchild, K. (2002). Minority members Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of
implicit attitudes: Automatic ingroup bias as a function of group status. social behavior. Personality and Social Psychology Review, 8, 220 247.
Social Cognition, 20, 294 320. Wegener, D. T., & Petty, R. E. (1995). Flexible correction processes in
Rydell, R. J., & McConnell, A. R. (2006). Understanding implicit and social judgment: The role of naive theories in corrections for perceived
explicit attitude change: A systems of reasoning analysis. Journal of bias. Journal of Personality and Social Psychology, 68, 36 51.
Personality and Social Psychology, 91, 9951008. Wittenbrink, B., Judd, C. M., & Park, B. (1997). Evidence for prejudice at
Rydell, R. J., McConnell, A. R., Mackie, D. M., & Strain, L. M. (2006). Of the implicit level and its relationship with questionnaire measures.
two minds: Forming and changing valence-inconsistent implicit and Journal of Personality and Social Psychology, 72, 262274.
explicit attitudes. Psychological Science, 17, 954 958.
Rydell, R. J., McConnell, A. R., Strain, L. M., Claypool, H. M., &
Hugenberg, K. (2007). Implicit and explicit attitudes respond differently Received May 26, 2007
to increasing amounts of counterattitudinal information. European Jour- Revision received December 11, 2007
nal of Social Psychology, 37, 867 878. Accepted December 22, 2007

E-Mail Notification of Your Latest Issue Online!


Would you like to know when the next issue of your favorite APA journal will be available
online? This service is now available to you. Sign up at http://notify.apa.org/ and you will be
notified by e-mail when issues of interest to you become available!
INTERPERSONAL RELATIONS AND GROUP PROCESSES

Maintaining Sexual Desire in Intimate Relationships: The Importance of


Approach Goals
Emily A. Impett Amy Strachman
University of California, Berkeley University of Southern California

Eli J. Finkel Shelly L. Gable


Northwestern University University of California, Santa Barbara

Three studies tested whether adopting strong (relative to weak) approach goals in relationships (i.e., goals focused
on the pursuit of positive experiences in ones relationship such as fun, growth, and development) predict greater
sexual desire. Study 1 was a 6-month longitudinal study with biweekly assessments of sexual desire. Studies 2 and
3 were 2-week daily experience studies with daily assessments of sexual desire. Results showed that approach
relationship goals buffered against declines in sexual desire over time and predicted elevated sexual desire during
daily sexual interactions. Approach sexual goals mediated the association between approach relationship goals and
daily sexual desire. Individuals with strong approach goals experienced even greater desire on days with positive
relationship events and experienced less of a decrease in desire on days with negative relationships events than
individuals who were low in approach goals. In two of the three studies, the association between approach
relationship goals and sexual desire was stronger for women than for men. Implications of these findings for
maintaining sexual desire in long-term relationships are discussed.

Keywords: sexual desire, motivation, close relationships, gender differences, daily experience methods

I know nothing about sex, because I was always married.Zsa Zsa Gabor help couples to maintain sexual desire over the course of their relation-
With this statement, Zsa Zsa highlights a common belief about the ships.
decline of sexual interest and activity in long-term relationships. Lack of Sexual Desire and Relationship Quality
sexual desire is the most common presenting problem at sex therapy Although there is no widely accepted definition of sexual desire among
clinics (e.g., Beck, 1995; Hawton, Catalan & Fagg, 1991). In the Amer- researchers and theorists (Levine, 2003), central to many definitions is the
ican survey conducted by Laumann and his colleagues, a lack of sexual need, drive, or motivation to engage in sexual activities (Brezsnyak &
desire was reported by 32% of women and 15% of men between the ages Whisman, 2004; Clayton et al., 2006; Diamond, 2004).1 Several large-
of 18 and 29 years (Laumann, Gagnon, Michael, & Michaels, 1994). scale surveys have shown that sexual desire as well as the related con-
Recent books by sex therapists and clinicians with such titles as Rekin- structs of sexual satisfaction and sexual frequency decline with the length
dling Desire: A Step-by-Step Program to Help Low-Sex and No-Sex of time that partners have been in a relationship (e.g., Johnson, Wads-
Marriages (McCarthy & McCarthy, 2003) and Reclaiming Desire: Four worth, Wellings, & Field, 1994; Klusmann, 2002). One large survey of
Keys to Finding Your Lost Libido (Goldstein & Brandon, 2004) target German college students revealed that as duration of partnership in-
couples who seek to rekindle sexual intimacy and passion in their rela- creased, the frequency of sexual intercourse and sexual satisfaction de-
tionships. In this article, we introduce and test approach relationship clined in both women and men. Further, whereas mens sexual desire
goals (i.e., goals focused on the pursuit of positive experiences in ones remained relatively stable over the course of a relationship, womens
relationship such as fun, growth, and development) as a factor that may sexual desire dropped steadily after about 1 year of dating (Klusmann,

1
Emily A. Impett, Institute of Personality and Social Research, University of Although we were centrally concerned with the motivational component of
California, Berkeley; Amy Strachman, Institute for Health Promotion & Disease sexuality (i.e., sexual desire), there is substantial overlap between sexual desire and
Prevention Research, University of Southern California, and eHarmony Labs, the related constructs of sexual arousal and enjoyment. The traditional human sex
Pasadena, California; Eli J. Finkel, Department of Psychology, Northwestern response cycle of Masters and Johnson (1966) and Kaplan (1979) depicts sexual
University; Shelly L. Gable, Department of Psychology, University of California, desire as a spontaneous force that itself triggers sexual arousal. In more recent
Santa Barbara. years, however, therapists and researchers have begun to challenge this model,
Preparation of this article was supported by a fellowship awarded to Emily A. particularly Basson and colleagues (Basson et al., 2004) who have suggested that
Impett from the Sexuality Research Fellowship Program of the Social Science sexual arousal, desire, and enjoyment co-occur and can reinforce each other. Many
Research Council and by a Ruth L. Kirschstein National Research Service Award people report that their sexual desire increases during sexual intercourse; that is, as
to Amy Strachman. We thank Amie Gordon, Anne Peplau, and Deborah Schooler they begin to be aroused and enjoy the sexual experience, they recognize that their
for helpful comments on earlier versions of this article. sexual desire increases, and they become motivated to become even more aroused
Correspondence concerning this article should be addressed to Emily A. Impett, (Levine, 2002). For these reasons, in some of the studies in the current article, we
Institute of Personality and Social Research, 4143 Tolman Hall, 5050, University assessed sexual arousal and enjoyment in addition to sexual desire in order to more
of California, Berkeley, CA 94720-5050. E-mail: eimpett@gmail.com fully capture the interrelated components of sexual desire.

Journal of Personality and Social Psychology, 2008, Vol. 94, No. 5, 808 823
Copyright 2008 by the American Psychological Association 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.808

808
APPROACH GOALS AND SEXUAL DESIRE 809

2002). Another study documented that the association between relation- higher BIS sensitivity reported experiencing more daily negative
ship duration and reduced frequency of intercourse was stronger than the affect than those with lower BIS sensitivity.
association between age and sexual frequency (Johnson et al., 1994). In The approachavoidance motivational distinction has been par-
short, sexual desire typically peaks at the beginning of relationships when ticularly helpful in understanding motivation in interpersonal re-
partners are just getting to know each other and often decreases over the lationships. Basing their work on that of early social motivation
course of relationships (Basson, 2002; Levine, 2003). theorists (e.g., Boyatzis, 1973; Mehrabian, 1976), Gable and col-
Because both sexual desire and sexual satisfaction play key leagues have recently distinguished between approach and avoid-
roles in determining the quality of intimate relationships, relation- ance social goals (Elliot, Gable, & Mapes, 2006; Gable, 2006b).
ship scholars and therapists should care about the decline of sexual Whereas approach social goals direct individuals toward potential
desire. Many studies of couples who voluntarily attend sex therapy positive outcomes such as intimacy or growth in their relation-
clinics provide support for the idea that low sexual desire is ships, avoidance social goals direct individuals away from poten-
associated with decreased levels of relationship satisfaction, both tial negative outcomes such as conflict or rejection. For example,
for individuals with low desire and for their partners (e.g., Mc- in a discussion about child care, a husband who has strong ap-
Cabe, 1997; Trudel, Landry, & Larose, 1997). More recent studies proach goals may be concerned with wanting the discussion to go
have also documented similar associations between sexual desire smoothly and wanting both partners to be happy with the outcome.
and relationship satisfaction in community samples of married In contrast, a husband with strong avoidance goals may be more
couples (Brezsnyak & Whisman, 2004) and dating couples concerned with avoiding conflict about child care and preventing
(Regan, 2000; Sprecher, 2002). Further, many empirical studies both partners from being unhappy with the outcome (Gable,
have documented a significant positive association between sexual 2006b). These goals are flexible forms of regulation that may take
satisfaction and dating and marital quality (Yeh, Lorenz, Wick- on diverse manifestations; they may focus on a specific relation-
rama, Conger, & Elder, 2006; see review by Sprecher & Cate, ship or relationships in general, they may focus on close relation-
2004). Sex therapists have similarly noted that when sexuality ships or acquaintances, and they may focus on a variety of rela-
functions well in a marriage, it contributes substantially to the tional concerns such as sexuality, intimacy, and parenting
marital bond. However, dysfunctional or nonexistent sexuality (Dowson & McInerney, 2003; Sorkin & Rook, 2004). In this
robs the marriage of intimacy, satisfaction, and stability (Mc- article, we are specifically concerned with individuals goals in
Carthy, 1999). their romantic relationships (in Studies 1 and 2) and in their
While a wealth of research has shown that sexual desire con- interpersonal relationships more generally (in Study 3).
tributes to relationship quality and stability, less research has Just as BAS and BIS are associated with distinct emotional
investigated the factors that help promote and sustain sexual desire outcomes, research has also shown that approach and avoidance
in relationships. In this article, we suggest that the adoption of social goals predict different social outcomes (Gable, 2006b; Im-
approach relationship goals may help couples to maintain sexual pett, Gable, & Peplau, 2005). In three short-term longitudinal
desire over the course of their relationships. We first introduce the studies, approach goals were associated with more positive social
approachavoidance theoretical perspective guiding this research attitudes, more satisfaction with social bonds, and less loneliness,
and apply this theory to the study of sexuality in intimate relation- whereas avoidance goals were associated with more negative so-
ships. We then present the results of three studies designed to test cial attitudes, relationship insecurity, and more loneliness (Gable,
our hypotheses concerning the link between approach relationship 2006b). In a daily experience study of dating couples, on days
goals and sexual desire. Finally, we discuss the implications of this when individuals made sacrifices for approach motives, they ex-
research for couples and sex therapists who wish to promote perienced greater positive affect and relationship satisfaction; on
healthy sexual functioning in long-term relationships. days when they sacrificed for avoidance motives, they experienced
greater negative affect, less relationship satisfaction, and more
ApproachAvoidance Motivational Framework conflict (Impett, Gable, & Peplau, 2005). In short, the approach
system (but not the avoidance system) is associated with positive
Several theories of motivational processes postulate the exis- emotional and social outcomes.
tence of distinct approach (also called appetitive) and avoidance
(also called aversive) motivational systems (see reviews in Carver, Applying the Approach-Avoidance Motivational
Sutton, & Scheier, 2000; Elliot & Covington, 2001). For instance, Framework to Sexuality
Grays (1987) neuropsychological model of motivation posits ap-
petitive and aversive motivational systems, referred to as the Central to many definitions of sexual desire is the need or
behavioral approach system (BAS) and the behavioral inhibition motivation to engage in sexual activities or the pleasurable antic-
system (BIS; see also Carver & White, 1994). Specifically, the ipation of such activities in the future (Brezsnyak & Whisman,
BAS is an appetitive system that is primarily sensitive to positive 2004; Clayton et al., 2006; Diamond, 2004). In short, sexual desire
stimuli or signals of reward, whereas the BIS is an aversive system involves the potential rewards and the positive emotional experi-
that is primarily sensitive to negative stimuli or signals of punish- ence that are characteristic of the approach motivational system. In
ment. Gray (1990) has shown that the BAS is associated with addition to predicting positive affect and relationship satisfaction
feelings of hope, whereas the BIS is associated with feelings of (Gable, 2006b; Gable et al., 2000; Impett, Gable, & Peplau, 2005),
anxiety. In a study of motivational dispositions and daily events, approach relationship goals may also be associated with daily
Gable, Reis, and Elliot (2000) found that participants with higher sexual desire and the maintenance of sexual desire over time. One
BAS sensitivity reported experiencing more daily positive affect possible reason why approach relationship goals may promote
than those with lower BAS sensitivity, while participants with sexual desire concerns peoples motives or reasons for engaging in
810 IMPETT, STRACHMAN, FINKEL, AND GABLE

sexual activity with a partner (Cooper, Shapiro, & Powers, 1998; research suggest that womens sexual desire may be more sensitive
Impett, Peplau, & Gable, 2005). People who pursue positive ex- than mens to relationship dynamics, and in particular, to womens
periences, such as growth and development, in their relationships goals for the relationship. For this reason, a secondary goal of the
may view sexual activity as one way to create positive, intimate current research was to explore whether the association between
experiences with a partner. Therefore, compared with people with approach relationship goals and sexual desire is stronger for
weak approach relationship goals, those with strong approach women than for men.
relationship goals may think more about sex, be more sensitive to
their partners cues, create environments that promote intimate Hypotheses and Research Overview
interaction, and act more readily upon potential sexual encounters.
Previous research has shown that approach motives and goals are We conducted three studies of individuals in dating relation-
primarily linked to outcomes through an exposure process (Elliot ships to test several predictions from approachavoidance motiva-
et al., 2006; Gable, 2006b; Gable et al., 2000). That is, individuals tional theory about the maintenance of sexual desire in dating
with strong approach goals or motives tend to report experiencing relationships. Study 1 was a 6-month longitudinal study of indi-
a greater number of positive events (but not fewer negative viduals in dating relationships that included biweekly assessments
events). Therefore, individuals with strong approach goals are of sexual desire. In this study, we tested the hypothesis that the
likely to experience a greater number of positive events and adoption of approach relationship goals would buffer against de-
positive emotions (including desire) with their partners. Because of clines in sexual desire over time. Study 2 was a 2-week daily
these previous findings, we believe that approach goals in close experience study of individuals in dating relationships designed to
relationships will be more strongly related to sexual desire than extend the findings from Study 1 by testing whether approach
will avoidance goals because approach goals are primarily asso- sexual goals would mediate the link between approach relationship
ciated with positive events (likely through processes that lead to goals and sexual desire. Study 3 was an additional 2-week daily
increased exposure) and avoidance goals have not been linked experience study that included (a) a more general measure of
reliably to positive events. social (as opposed to relationship-specific) approach goals, (b) a
It is also likely that people with strong approach goals for their more detailed measure of sexual goals that distinguished between
relationships in general may also engage in daily sexual activity self-focused and other-focused goals, and (c) measures of positive
for approach reasons such as pleasing a partner or enhancing and negative relationship events. These last measures enabled us to
intimacy in the relationship. Repeatedly engaging in sex for ap- examine how perceptions of the daily relationship climate influ-
proach reasons, in turn, may promote greater sexual desire. A ence sexual desire and whether relationship events moderate the
recent cross-sectional study of late adolescent girls showed that association of approach goals with sexual desire. In all three
engaging in sex for approach goals (e.g., to express love, for studies, we conducted additional analyses to examine the effects of
physical attraction) was positively associated with sexual satisfac- approach goals on sexual desire beyond the influence of how long
tion (Impett & Tolman, 2006). Based on this research, we pre- people have been involved in their relationships, how satisfied they
dicted that individuals with strong approach relationship goals are with their partners, and how frequently they engage in sexual
would report engaging in sexual activity for approach reasons, in activity. Finally, in all three studies, we examined gender as a
turn, promoting greater sexual desire. moderator of the link between approach goals and sexual desire,
exploring the possibility that the association may be stronger for
women than for men as womens sexual desire may be particularly
Gender and Sexual Desire sensitive to womens goals for their relationships.
Many lines of research demonstrate that men show more interest
in sex than do women (see review by Baumeister, Catanese & Study 1
Vohs, 2001). For example, men think about sex more often than
We tested three main predictions in a 6-month longitudinal
women do (Laumann et al., 1994), report more frequent sexual
study of college students in dating relationships: (a) Individuals
fantasies (Beck, Bozman, & Qualtrough, 1991), and report greater
with strong approach goals would report higher sexual desire at
feelings of sexual desire (Leitenberg & Henning, 1995). Further,
study entry than individuals with weak approach goals; (b) indi-
men and women differ in their preferred frequency of sex; when
viduals who began the study with weak approach relationship
dating and marriage partners disagree about sexual frequency, men
goals would experience decreases in sexual desire over the course
usually want to have sex more often than women (Julien, Bou-
of the study, whereas individuals who began the study with strong
chard, Gagnon, & Pomerleau, 1992; Sprecher & Regan, 1996).
approach relationship goals would not experience such decreases;
Complementing these descriptive gender differences is research
and (c) avoidance relationship goals would not be significantly
demonstrating that womens sexual desire may be more closely
associated with sexual desire. Finally, we explored gender as a
tied to the interpersonal aspects of the relationships than is mens
moderator of the association between approach relationship goals
desire (see review by Peplau, 2003). For instance, when Regan and
and sexual desire.
Berscheid (1999) asked young adults to define sexual desire, men
were more likely than women to emphasize physical pleasure and
sexual intercourse, whereas women were more likely than men to Method
emphasize the emotional or relational side of sexual desire.
Participants and Procedure
Women are more likely than men to engage in sex to enhance
commitment and express love for their partners (Basson, 2002; Sixty-nine Northwestern University undergraduate students (34
Impett, Peplau, & Gable, 2005). Taken together, these lines of men, 35 women) were recruited via flyers posted around campus
APPROACH GOALS AND SEXUAL DESIRE 811

to participate in a 6-month longitudinal study of dating processes. activities were not limited to sexual intercourse.) Participants com-
Eligibility criteria required that each participant be: (a) a first-year pleted a two-item partner-specific measure of sexual desire, answer-
undergraduate at Northwestern University, (b) involved in a dating ing the questions I feel a great deal of sexual desire for my partner
relationship of at least 2 months duration, (c) between 17 and 19 and When my partner and I have sexual contact, I enjoy it a great
years old, (d) a native English speaker, and (e) the only member of deal on 7-point scales (1 strongly disagree, 7 strongly agree).
a given couple to participate in the study. Participants who com- Within each of the 14 waves of online data collection, the correlations
pleted all components of the study were paid $100; those who between these two items were quite high (rs .69 .98, ps .001,
missed some were paid a prorated amount. At the beginning of the with an average correlation across these waves of .91). We also
study, most participants were 18 years old (7% were 17, 81% were assessed frequency of sexual contact with ones partner. Participants
18, and 12% were 19) and White (74% White, 12% Asian Amer- answered the question How many times did you have sexual contact
ican, 3% Hispanic, 1% African American, and 10% other). On with your partner over the last 2 weeks? by typing in a number rather
average, participants had been dating their partner for a little over than answering on a response scale.
1 year (M 13 months; range 2 42 months). During the Relationship satisfaction. Participants answered one question
6-month study, 26 participants (38%) broke up with their romantic designed to measure relationship satisfaction as part of the 14
partner; they were included in the analyses until the breakup.2 biweekly online questionnaires. Specifically, they responded to the
This study was part of a larger investigation of dating processes that statement I am satisfied with my relationship on 7-point scales
was divided into four parts: (a) an initial 60-min questionnaire sent via (1 strongly disagree, 7 strongly agree).
campus mail, (b) a 90-min lab-based session involving additional
questionnaires and training for the online sessions, (c) a 10- to-15-min
online questionnaire every other week for 6 months (14 in total), and Results
(d) a 60-min lab-based session at the end of the 6-month period.
During the training for the online sessions, a researcher reviewed the Participants reported an average of 3.16 (SD 3.92) acts of
procedures for the completion of the biweekly surveys, specifically sexual contact with their partner per 2-week time period of the
emphasizing that participants should complete their surveys every study. The central hypotheses guiding this study were that ap-
other Wednesday evening and that their responses were confidential proach relationship goals would predict elevated sexual desire at
(i.e., they used a password to log onto the server). To bolster and study entry and buffer against declines in sexual desire over time.
verify compliance, we sent participants reminder e-mails if they The two-level data structure included measures assessed on each
forgot to complete a survey on time, and financial incentives were of the online questionnaires (Level 1) nested within each partici-
linked to completing each survey. Only surveys received within 48 hr pant (Level 2). For example, participants who completed all waves
of when they were due were retained in the data set. Participant of online data collection reported their level of sexual desire on 14
retention was excellent: All 69 participants completed the study, and different occasions. Traditional ordinary least squares regression
67 of them completed at least 12 of the 14 online measures on time. methods assume independence of observations, a criterion that is
Fourteen participants failed to complete the measure of approach and typically violated when the same individual completes the same
avoidance relationship goals correctly, leaving the final sample at 55 measures repeatedly. Therefore, we analyzed the data using mul-
participants.3 tilevel modeling techniques (Raudenbush & Bryk, 2002) with the
MIXED procedure in SAS (Littell, Milliken, Stroup, & Wolfinger,
1996). Multilevel modeling approaches provide unbiased hypoth-
Measures esis tests by simultaneously examining variance associated with
Approach and avoidance relationship goals. As part of the each level of nesting. A strength of multilevel modeling techniques
initial questionnaire, participants completed a 4-item measure as- is that they can readily handle an unbalanced number of cases per
sessing approach relationship goals (e.g., I will be trying to person (i.e., number of surveys completed), giving greater weight-
deepen my relationship with my romantic partner and I will be ing to participants who provide more data (Snijders & Bosker,
trying to move toward growth and development in my romantic 1999). Following Singer and Willett (2003), we permitted the
relationship; .86) and another assessing avoidance relation- intercept and slope terms for approach relationship goals to vary
ship goals (e.g., I will be trying to avoid disagreements and randomly; the slope terms for the other predictors were treated as
conflicts with my romantic partner and I will be trying to make fixed. Finally, all variables were standardized prior to analyses;
sure that nothing bad happens in my romantic relationship; consequently the coefficients represented changes in standard de-
.66; Gable, 2006a). All questions were answered on 7-point scales viation units of the dependent variable (i.e., sexual desire) associ-
(1 strongly disagree, 7 strongly agree), and each scale was ated with a standard deviation unit of the predictor variable. Thus,
calculated as an average score of the ratings on the 4 items. In the the coefficients are a convenient measure of effect size.
current study, a two-factor-solution principal components analysis
with varimax rotation explained 60% of the scale variance. The 2
Participant sexual orientation was not assessed in Studies 1 and 3.
first factor (35% of explained variance) included the four approach 3
These 14 participants responded to the approach relationship goals
relationship goals items, and the second factor (25% of explained
questionnaire items with check marks rather than with the 17 rating scale,
variance) included the four avoidance goals items. The correlation which meant that we were not able to calculate a score for them. The 55
between the two subscales was .21, p .12. participants who completed the goals measure correctly did not differ
Sexual desire. As part of the 14 biweekly online questionnaires, significantly from the 14 who did not on the initial measures of relationship
participants answered questions about their sexual desire for and duration, relationship satisfaction, or sexual desire. This problem with the
participation in sexual activities with their dating partner. (These goals measure was subsequently rectified in Studies 2 and 3.
812 IMPETT, STRACHMAN, FINKEL, AND GABLE

Figure 1. Approach relationship goals as a moderator of the intercept and slope of sexual desire (Study 1).
Note: The means were estimated with 1 standard deviation on approach goals.

Before testing our specific hypotheses concerning approach and the intercept, .28, t(470) 2.64, p .01, and slope of sexual
avoidance relationship goals and sexual desire, we conducted a pre- desire, .015, t(470) 2.16, p .05. In the second analysis, we
liminary analysis to determine if, on average, participants experienced controlled for the frequency with which participants engaged in sex-
a decline in sexual desire over the course of the study. This analysis ual intercourse across the 14-day study, and approach relationship
included time as a predictor of the intercept and the slope of sexual goals remained significant predictors of both the intercept, .34,
desire. In this and all subsequent analyses, time was coded such that t(469) 3.03, p .01, and slope of sexual desire, .02, t(469)
the first wave of data collection was 0 and the final wave was 13. A 2.34, p .01, pointing to the robust nature of these findings.4,5
significant effect of time on sexual desire, .02, t(66) 2.52, The final goal of this study was to explore whether the
p .02, showed that sexual desire decreased significantly over time association between approach relationship goals and sexual
at a rate of .02 standard deviation units every 2 weeks; this rate of desire is stronger for women than for men. To examine this
biweekly decline would lead to an annual decline in sexual desire of possibility, we included six additional terms involving gender
approximately half a standard deviation (.52 standard deviation units, (coded as 1 men and 1 women) to the primary analysis
to be precise). This decline over time in sexual desire mirrors a similar described above. Specifically, we examined whether gender
decline in desire in samples of married couples (e.g., Johnson et al., moderated the intercept and slope effects for sexual desire and
1994; Klusmann, 2002).
whether gender moderated any of the associations of approach
Next, we tested the hypothesis that approach relationship goals
or avoidance relationship goals with the intercept and slope of
would moderate the intercept and slope of sexual desire. We predicted
sexual desire. This analysis revealed a significant gender effect
that participants with strong approach goals would begin the study
on the sexual desire intercept, .18, t(591) 2.04, p .04,
higher in sexual desire and would not experience the decline in sexual
indicating that men reported greater sexual desire than did
desire that characterized the sample as a whole. To test this hypoth-
esis, we simultaneously entered time, approach goals, and avoidance women at the beginning of the study. There was no significant
goals to predict both the intercept and the slope of sexual desire. The
results showed that approach goals predicted the intercept of sexual 4
We conducted additional analyses in which we analyzed both of the
desire, .35, t(625) 3.17, p .01, providing support for the individual-item dependent measures of sexual desire separately (i.e., sexual
hypothesis that participants with strong approach goals would report desire and enjoyment). When we included both approach and avoidance goals
greater sexual desire at study entry relative to those with weak in an equation simultaneously, approach goals significantly predicted the
approach goals. Approach goals also (marginally) moderated the intercepts of both sexual desire, .33, t(469) 3.05, p .01, and
effect of time on sexual desire, .014, t(625) 1.87, p .06. The enjoyment, .24, t(467 2.17), p .05; further, approach goals margin-
results showed that whereas participants with low approach relation- ally predicted the slope of sexual desire, .013, t(469) 1.77, p .077,
ship goals experienced declines in sexual desire over the course of the and significantly predicted the slope of sexual enjoyment, .02, t(467)
2.96, p .003, when both the intercept and the slope of relationship satisfac-
study, participants with strong approach goals retained relatively high
tion and duration were controlled. There were no significant associations
levels of sexual desire over the course of the study. Figure 1 depicts
between avoidance relationship goals and either of the individual item-
both of these effects. Consistent with our hypotheses, avoidance goals dependent measures of sexual desire.
predicted neither the intercept, .17, p .13, nor the slope, 5
In all of the studies reported in this article, we tested for interactions
.01, p .42, of sexual desire. We then conducted two sets of between approach and avoidance goals, and none of these effects was
follow-up analyses. In the first analysis, we controlled for relationship significant. Furthermore, once the interaction terms were added, the effects
satisfaction and duration (both the intercept and slope terms), and for the intercept and the slope of approach goals (in Study 1) and the effects
approach relationship goals remained significant predictors of both for approach goals in Studies 2 and 3 remained significant.
APPROACH GOALS AND SEXUAL DESIRE 813

gender effect on the sexual desire slope, however, .01, Method


t(591) 0.67, p .51. Gender also moderated the effect of
approach goals on the intercept of sexual desire, .33, Participants and Procedure
t(591) 3.63, p .001, suggesting that the association be-
tween approach goals and sexual desire at study entry was The study was advertised as an examination of dating rela-
stronger for women than for men. Finally, gender did not tionships, and participants received credit toward psychology
significantly moderate the effect of approach goals on the coursework at the University of California, Los Angeles, in
slope of sexual desire, .012, t(591) 1.43, p .15. exchange for participation. To be eligible, participants had to:
This result suggests that men and women with weak ap- (a) be currently involved in a dating (not a marital) relationship,
(b) see their partner at least 5 days per week (i.e., no long-
proach goals did not significantly differ in the tendency to
distance relationships), and (c) be the only member of a given
experience decreased sexual desire over time, although this
couple to participate in the study. Of the 121 participants (55
nonsignificant effect trended in the direction of approach goals
men, 66 women) who completed the study, 2 were engaged to
more positively predicting the slope of sexual desire for women
be married, and 18 were cohabitating; the mean relationship
than for men.
length for all participants was 18 months (range 1 month 8
years). Participants ranged in age from 18 to 38 years (M
20.2, SD 2.6). The sample was ethnically diverse: 5% were
Brief Discussion African American, 36% were Asian or Pacific Islander, 15%
Study 1 provided evidence for the two hypotheses linking were Hispanic, 37% were White, and 7% self-identified as
multi-ethnic or other. In addition, all participants identified as
approach relationship goals and sexual desire. Not only did
heterosexual except one gay man, and he was included in the
approach relationship goals predict greater sexual desire at
study.
study entry, but having strong approach relationship goals buff-
During an initial session, each participant was given 14
ered against declines in sexual desire over a 6-month period.
surveys, each containing the daily measures, one for each night
Avoidance goals were not significantly associated with sexual
of the week. A researcher then reviewed the procedures for
desire at the beginning of the study or trajectories of sexual
completion of the daily surveys, specifically emphasizing that
desire over time. Finally, the association between approach
participants should begin completing their surveys that evening,
relationship goals and sexual desire at the beginning of the
that they should complete one survey each night before going to
study was stronger for women than for men, pointing to the
bed (even if they did not engage in sex on that particular day),
particular importance of goals focused on obtaining positive that their responses were confidential, that they should not
outcomes in romantic relationships for enhancing womens discuss their surveys with their partner, and that if they missed
sexual desire. Why do approach relationship goals buffer a day, they should leave that particular survey blank. To bolster
against declines in sexual desire over time? Study 2 tested the and verify compliance with the daily schedule, we asked par-
hypothesis that approach relationship goals promote increased ticipants to return completed surveys every 23 days to a locked
sexual desire during daily sexual interactions, given that people mailbox located outside the laboratory. As an incentive, each
who typically pursue approach goals in their relationships may time participants handed in a set of surveys on time, they
also be highly motivated to pursue shorter term approach goals, received a lottery ticket for one of several cash prizes ($100,
such as enhancing intimacy and closeness, during their sexual $50, $25) to be awarded after the study. Participants who did
interactions with a partner (Gable, 2006b). In addition, Study 2 not return a particular set of surveys on time were reminded by
tested approach sexual goals as a possible mediator of the phone or e-mail. Only daily surveys returned on time were
association between approach relationship goals and sexual treated as valid and retained in the data set. In total, participants
desire. completed 1,549 daily surveys on time, an average of 12.8 days
per person. Ninety percent of the participants completed all 14
daily reports on time.
Study 2

We tested three main predictions in a 2-week daily experi-


Background Measures
ence study of college students in dating relationships: (a) Ap-
proach relationship goals would be associated with increased In their initial session in the laboratory, participants com-
sexual desire in day-to-day sexual interactions, (b) approach pleted a questionnaire with basic demographic information
sexual goals would mediate the association between ap- (i.e., gender, age, ethnicity, relationship duration), as well as
proach relationship goals and sexual desire; and (c) avoidance the same measure of approach and avoidance relationship
relationship goals would not be significantly associated with goals used in Study 1 (Gable, 2006a). They were instructed to
daily sexual desire. In addition, as in Study 1, we examined answer the questions about their goals for their relationships
gender as a possible moderator of the association between over the next few months. In the present study, .78 for
approach relationship goals and sexual desire, exploring the approach social goals and .79 for avoidance social goals.
possibility that the association would be stronger for women The correlation between the two subscales was .57, p .001.
than for men. In addition, participants completed a standard 5-item measure
814 IMPETT, STRACHMAN, FINKEL, AND GABLE

of relationship satisfaction (Rusbult, Martz, & Agnew, Relationship Goals and Daily Sexual Desire
1998). Participants responded to such statements as Our
relationship makes me happy on 9-point scales (0 do not The first major hypothesis guiding this study was that ap-
agree at all, 8 agree completely). In this sample, .89. proach relationship goals would predict increased daily sexual
desire. To test this hypothesis, we entered approach and avoid-
ance relationship goals as simultaneous predictors of daily
Daily Measures sexual desire. The results showed that approach relationship
goals were positively associated with sexual desire, .20,
If participants had engaged in sexual intercourse since they had
t(117) 2.84, p .01. In contrast, avoidance goals were not
completed the previous days survey, they completed measures of
significantly associated with sexual desire, .07, p .31.7
sexual desire and sexual goals.6
As in Study 1, we then conducted two sets of follow-up anal-
Sexual desire. Each time that they engaged in sexual inter-
yses. In the first analysis, we controlled for relationship satis-
course, participants answered two questions designed to measure
faction and duration, and the association between approach
their sexual desire on 7-point scales (1 very low, 7 very high).
relationship goals and sexual desire remained significant,
More specifically, they responded to the following two items:
.17, t(115) 2.23, p .05. In the second analysis, we con-
Rate your own level of sexual desire just prior to engaging in
sex, and Rate your own level of sexual desire during sex. A trolled for the frequency with which participants engaged in
composite sexual desire variable was created by averaging the sexual intercourse across the 14-day study, and the association
responses to these two questions ( .64). between approach goals and desire also remained significant,
Sexual goals. Each time they engaged in sexual intercourse, .20, t(116) 2.79, p .01, pointing to the robust nature
participants responded to a nine-item measure of sexual goals of these findings.
adapted from Cooper et al. (1998) and used by Impett, Peplau, We also explored gender as a moderator of the association
& Gable (2005). Participants rated the importance of five ap- between approach relationship goals and sexual desire. Similar
proach and four avoidance goals in influencing their decision to to the way we conducted analyses in Study 1, we simulta-
engage in sex on 7-point scales (1 not at all important, 7 neously entered approach relationship goals, avoidance rela-
extremely important). The approach items were to pursue tionship goals, gender, and two interaction terms (Approach
my own sexual pleasure, to feel good about myself, to Relationship Goals Gender; Avoidance Relationship Goals
please my partner, to promote intimacy in my relationship, Gender) to predict daily sexual desire. Although there was no
and to express love for my partner. The avoidance items were main effect of gender, .06, p .28, the interaction
to avoid conflict in my relationship, to prevent my partner between gender and approach relationship goals significantly
from becoming upset, to prevent my partner from getting predicted sexual desire, .21, t(114) 2.08, p .05. As
angry at me, and to prevent my partner from losing interest in shown in Figure 2, the association between approach relation-
me. The within-person correlation between approach and ship goals and sexual desire was stronger for women than for
avoidance sexual goals was .03, p .49. The reliability coef-
ficients were .71 for approach goals and .90 for avoidance
goals. 6
Although people certainly experience sexual desire in the absence
of having sexual contact with their partner, we focused only on days on
which participants reported engaging in sexual contact with the partner
Results for several reasons. First, we focused on desire just before a concrete
event (e.g., sexual activity) to improve recall and lessen retrospective
Participants reported a total of 480 sexual interactions. On biases (Reis & Gable, 2000). Second, because there are multiple reasons
average, participants reported engaging in sexual intercourse on that partners may not have engaged in sexual activities, some benign or
4 days during the 2-week study (SD 2.3; range 110 related to circumstance (e.g., schedule, proximity of partner) and some
days). A central goal of this study was to test predictions related to self or partner desire (e.g., rebuffed sexual advances), it
would be extremely difficult to compare event days to nonevent days.
about the associations between approach and avoidance rela-
Finally, we wanted to examine goals for each sexual event to capture
tionship goals and sexual desire. To address the data noninde- the full range of goals that participants experienced across days, and we
pendence, analyses were performed using multilevel modeling suspected that asking participants to report sexual goals in the absence
techniques in the hierarchical linear models (HLM) com- of sexual activity would have been difficult and would have produced
puter program (HLMwin, Version 5.02; Raudenbush, Bryk, unreliable data.
Cheong, & Congdon, 2000). Level-1 (i.e., daily) predictors 7
We conducted analyses in which we analyzed both of the
were centered around each individuals mean across the 14-day individual-item dependent measures of sexual desire separately (i.e.,
study. This technique, known as group-mean centering, ac- sexual desire just prior to engaging in sex and sexual desire during sex).
counts for differences between-persons in the sample and as- When we included both approach and avoidance goals in an equation
simultaneously, approach relationship goals significantly predicted sex-
sesses whether day-to-day changes from a participants own
ual desire just prior to engaging in sex, .20, t(117) 3.23, p .01,
mean are associated with changes in the outcome variable, and marginally predicted sexual desire during sex, .14, t(117)
consequently unconfounding between- and within-person ef- 1.69, p .10. There were no significant associations between avoid-
fects. As in Study 1, all variables were standardized prior to ance relationship goals and either of the individual item-dependent
analyses. measures of sexual desire.
APPROACH GOALS AND SEXUAL DESIRE 815

Figure 2. Gender as a moderator of the association between approach relationship goals and sexual desire
(Study 2).

men.8 Neither avoidance relationship goals, .04, p .68, (z 2.90, p .01), providing evidence for mediation. In other
nor the interaction between gender and avoidance goals, words, participants with strong approach relationship goals also
.06, p .56, significantly predicted daily sexual desire. tended to engage in sexual activity to pursue positive outcomes, in
turn promoting greater daily sexual desire.
Approach Sexual Goals as a Mediator
Another hypothesis was that approach sexual goals would me- Brief Discussion
diate the association between approach relationship goals and
Study 2 replicated and extended the findings from Study 1 in
sexual desire (see Figure 3). Standard (ordinary least squares
several important ways. First, the results showed that approach
[OLS]) hierarchical regression analysis based on the principles of
relationship goals promoted greater sexual desire during day-to-
Baron and Kenny (1986) was used to test mediation. Data were
day sexual interactions. Second, Study 2 demonstrated that ap-
aggregated across days such that each person received summary
proach sexual goals may be an important mechanism by which
scores for approach sexual goals and sexual desire. The first
approach relationship goals promote sexual desire. That is, indi-
requirement in demonstrating mediation is that the predictor vari-
viduals who are generally oriented toward promoting positive
able be associated with the outcome variable. Indeed, approach
experiences in their relationships also engage in sex to pursue
relationship goals were significantly associated with sexual desire
positive outcomes such as a partners happiness or increased
(r .24, p .01). The second requirement is to show that
intimacy. Approach sexual goals were, in turn, associated with
approach relationship goals predict the putative mediator, ap-
sexual desire. Third, this study showed that the association be-
proach sexual goals; indeed they did (r .45, p .001). The third
tween approach relationship goals and sexual desire was stronger
requirement is that the mediator predicts the outcome variable (i.e.,
for women than for men, providing further evidence that womens
sexual desire) after the predictor variable is controlled and that this
sexual desire is more closely tied to relationship dynamics than is
effect could plausibly account for the direct effects between the
mens sexual desire. Fourth, this study showed that avoidance
predictor and the outcome variable. Approach sexual goals signif-
goals were not significantly associated with daily sexual desire.
icantly predicted sexual desire, .33, p .01, and the direct
effect from approach relationship goals to sexual desire dropped to
nonsignificance, .09, p .36). A significant Sobel (1982) test Study 3
indicated that the drop in the value of the latter beta was significant
Study 3 was another daily experience study of college students
in dating relationships, but Study 3 differed from Study 2 in three

8
In addition to testing for interactions with participant gender, we also
tested for interactions with ethnicity by creating two dummy-coded vari-
ables: 1 White versus not White, and 2 Asian versus not Asian. When
each of these dummy-coded variables was (separately) added as a covari-
ate, approach relationship goals remained a significant predictor of daily
sexual desire. More important, we also created interaction terms between
approach goals and both of the dummy-coded variables. Neither of these
Figure 3. Approach sexual goals as a mediator between approach rela- interaction terms was associated with daily sexual desire. We also tested
tionship goals and sexual desire (Study 2). Note: All numbers are ordinary for interactions with ethnicity in Study 3 using the same strategy; none of
least squares regression coefficients. *p .05. **p .01. ***p .001. the interactions was significant.
816 IMPETT, STRACHMAN, FINKEL, AND GABLE

important ways. First, the approach relationship goals measure was server each day. The daily survey was posted on a Web site, and
replaced with a more general measure of approach social goals, participants were given a login name and password to use each
allowing us to determine whether the effects were specific to a time they entered the site. Participants were asked to complete the
measure of romantic relationships. Second, Study 3 included a survey at the beginning of each day for 14 consecutive days. The
longer, more refined measure of sexual goals that enabled us to survey asked about the previous days relationship and sexual
determine the relative contributions of both self-focused sexual activities. Participants were instructed to complete the survey by
goals (e.g., to pursue my own sexual pleasure) and other-focused 1 p.m. each day. The date and time of survey completion were
sexual goals (e.g., to please my partner) to daily sexual desire. automatically recorded by the Web site, and research assistants
Third, Study 3 also included measures of positive and negative checked this log each morning and e-mailed reminders to partic-
relationship events to enable us to examine how peoples percep- ipants who had not yet completed their daily surveys. Only surveys
tions of the daily relationship climate relate to their levels of sexual completed on time were accepted and included in the data analy-
desire. Each day poses an opportunity for positive events (e.g., ses. As an incentive for on-time completion of surveys, partici-
partners compliment each other, express their love, or do fun pants who completed between 11 and 14 diaries (N 71) were
things together) as well as negative events (e.g., they criticize, entered into a lottery drawing for $100. Participants completed a
disagree, or give each other the silent treatment). On the basis of total of 1,182 daily surveys on time, an average of 13 days per
previous research showing a link between relationship satisfaction person. Ninety percent of participants completed all their surveys
and increased sexual desire (e.g., Brezsnyak & Whisman, 2004; on time.
Sprecher, 2002), we predicted that individuals would experience
greater sexual desire on days with more frequent positive events
Background Measures
and also on days with less frequent negative events in their
relationships. We also predicted that approach social goals would In their initial session in the laboratory, participants completed
moderate these associations, such that people with strong approach a questionnaire with basic demographic information (i.e., gender,
goals would experience even greater sexual desire on days with age, ethnicity, relationship duration), as well as a measure of
many positive events because the approach system is sensitive to approach and avoidance social goals (Elliot et al., 2006). Partici-
the presence and absence of positive goal-relevant events (Gable et pants responded to four approach statements (e.g., I will be trying
al., 2000). This prediction is also consistent with work on the to move toward growth and development in my friendships, and
upward spiral effect of positive emotions (Fredrickson & Joiner, I will be trying to deepen my relationship with my friends) and
2002) and on the role of positive-arousing activities in relationship four avoidance statements (e.g., I will be trying to make sure
satisfaction and passionate love (Aron, Norman, Aron, McKenna, nothing bad happens to my close relationships, and I will be
& Heyman, 2000). Finally, as in Studies 1 and 2, we explored trying to avoid getting embarrassed, betrayed, or hurt by any of my
gender as a moderator of the association between approach social friends) on 7-point scales (1 not at all true of me, 7 very true
goals and sexual desire. of me). They were instructed to answer the questions about their
goals for their relationships over the next few months. In the
Method present study, .78 for approach social goals and .79 for
avoidance social goals. The correlation between the two subscales
Participants and Procedure was .36, p .05. Relationship satisfaction was also assessed using
the same measure as in Study 2 (Rusbult et al., 1998; .94).
The study was advertised as an examination of relationships,
sexuality, and health, and participants received credit toward
psychology coursework at the University of California, Los An- Daily Measures
geles, in exchange for participation. Participants were told that the If participants had engaged in sexual intercourse since they had
study was about daily events in relationships, including sexual completed their previous days survey, they completed measures
interactions. To be eligible, participants had to be: (a) currently of sexual desire and sexual goals.
involved in a dating relationship, (b) sexually active with their Sexual desire. Each time that they engaged in sexual inter-
partner, (c) see their partner at least 5 days per week (i.e., no course, participants answered three questions designed to measure
long-distance relationships), and (d) the only member of a given their sexual desire on 5-point scales (1 not at all, 5 very
couple to participate in the study. Ninety participants (60 women, much). The questions were: How much did you want to have
29 men, 1 did not report gender) completed the study. Twelve of sex? How much did you enjoy the sexual experience? and
the participants did not engage in sexual intercourse during the How sexually aroused were you during this sexual experience?
study; therefore, the final sample consisted of the remaining 77 A composite variable called sexual desire was created by averag-
participants (55 women, 22 men). Two of the participants were ing the responses to these three questions ( .95).
married, and 8 were cohabitating; the mean relationship length for Sexual goals. Cooper et al.s (1998) sexual motivation scale
all participants was 21 months. Participants ranged in age from 17 was used to measure approach and avoidance sexual goals. The
to 44 years (M 20.3, SD 3.6).9 The sample was ethnically scale consists of 29 items and was modified to assess the partici-
diverse: 4% were African American, 32% were Asian or Pacific pants most recent sexual experience. This measure categorizes
Islander, 35% were White, 26% identified as multi-ethnic or other,
and 3% did not report their ethnicity.
During an initial session, participants were given instructions 9
The 44 year-old participant was an outlier in terms of age. All analyses
about how to complete an online survey by logging onto a secure yielded identical conclusions when this person was excluded.
APPROACH GOALS AND SEXUAL DESIRE 817

sexual goals using the approach/avoidance distinction as well as a sis, we entered approach and avoidance social goals as simulta-
self-focused/other-focused distinction. These two dimensions are neous predictors of daily sexual desire. The results showed that
crossed to yield four categories of goals (six discrete goals) for approach social goals were positively associated with sexual de-
engaging in sex: (a) approach self-focused goals (e.g., I have sex sire, .19, t(69) 2.50, p .05. In contrast, avoidance goals
because it feels good [Enhancement]), (b) approach other-focused were not associated with sexual desire, .01, t(69) .71, p
goals (e.g., I have sex to feel emotionally close to my partner .84.10 As in Studies 1 and 2, we then conducted two sets of
[Intimacy]), (c) avoidance self-focused goals (e.g., I have sex to follow-up analyses. In the first analysis, we controlled for rela-
reassure myself that I am attractive [Self-Affirmation], and I tionship satisfaction and duration, and the association between
have sex to help me deal with disappointments in my life [Cop- approach relationship goals and sexual desire remained significant,
ing]), and (d) avoidance other-focused goals, I have sex because .17, t(67) 2.08, p .05. In the second analysis, we
I dont want my partner to be angry with me [Partner Approval]), controlled for the frequency with which participants engaged in
and I have sex just because all of my friends are having sex [Peer sexual intercourse across the 14-day study, and the association
Approval]). The reliability coefficients for Enhancement, Inti- between approach goals and desire also remained significant,
macy, Self-Affirmation, Coping, Partner Approval, and Peer Ap- .19, t(67) 2.22, p .05, pointing to the robust nature of these
proval were .90, .93, .84, .91, .84, and .78, respectively. findings.
Positive and negative relationship events. Participants com- As in Studies 1 and 2, we explored gender as a moderator of the
pleted measures of positive and negative events adapted from association between approach social goals and sexual desire. As in
previous research (Gable, Reis, & Downey, 2003). Each day, Studies 2 and 3, approach social goals, avoidance social goals,
participants indicated whether they experienced each of nine pos- gender, and two interaction terms (Approach Goals Gender;
itive relationship events and nine negative relationship events. Avoidance Goals Gender) were used to predict sexual desire.
Positive event items included: My partner told me that he/she Neither interaction term reached significance (Approach Gen-
loves me, My partner and I participated in an activity that I really der: .05, p .76; Avoidance Gender: .14, p .65).
enjoy, During a discussion, I felt understood and appreciated by This result suggests that the gender effects found in Studies 1 and
my partner, My partner did something that made me feel 2 may be specific to approach goals in romantic relationships, not
wanted, My partner and I did something fun, My partner did in social relationships in general.
something special for me, My partner complimented me, My
partner made me laugh, and My partner and I talked about Sexual Goals and Daily Sexual Desire
making our relationship more serious or committed. Negative
event items included: My partner and I had a minor disagree- The second major goal of this study was to determine which
ment, My partner was inattentive and unresponsive to me, My specific sexual goals were associated with daily sexual desire.
partner tried to control what I did, We had a major disagree- Although we were primarily interested in the distinction between
ment, My partners behavior made me question his or her self-focused and other-focused approach sexual goals, we also
commitment to me, My partner criticized me, My partner went examined the four different measures of avoidance sexual goals.
out with his/her friends instead of spending time with me, My Therefore, we simultaneously entered all six types of sexual goals
partner did something that made me feel irritated or angry, and (Enhancement, Intimacy, Self-Affirmation, Coping, Partner Ap-
My partner gave me the silent treatment. Responses to these proval, and Peer Approval) as well as the control variables (rela-
questions were summed to create separate indices of the total tionship duration, relationship satisfaction, and sexual frequency)
number of positive events and the total number of negative events to predict daily sexual desire. Table 1 displays the results of this
that participants experienced in their relationships each day. analysis. When all six sexual goals were entered simultaneously,
both of the measures of approach sexual goals (i.e., Enhancement
Results and Intimacy) significantly predicted daily sexual desire. On days
when participants engaged in sexual intercourse more often to
Participants reported a total of 283 sexual interactions. On pursue positive outcomes either for themselves (i.e., for enhance-
average, participants reported engaging in sexual intercourse on ment goals) or for their relationships (i.e., for intimacy goals), they
3.4 days during the 2-week study (SD 2.0; range 114 days). reported increased sexual desire. Furthermore, these associations
As in Study 2, the data set was hierarchically nested, with days remained significant even after we controlled for relationship
nested within persons. Multilevel modeling in the HLM computer satisfaction, relationship duration, and frequency of sexual inter-
program (HLMwin, Version. 5.02; Raudenbush et al., 2000) was course over the course of the 14-day study. In contrast, two of the
used to examine the hypotheses linking social goals, sexual goals, measures of avoidance sexual goals (i.e., Self-Affirmation and
relationship events, and sexual desire. Level-1 (i.e., daily) predic- Coping) were not significantly associated with sexual desire, and
tors were centered around each individuals mean across the 14-
day study, enabling us to determine whether day-to-day changes
from a participants own mean were associated with changes in the 10
We conducted additional analyses in which we analyzed each of the
outcome variable. As in Studies 1 and 2, all variables were stan- three sexual desire items separately (i.e., wanting sex, enjoying sex, being
dardized prior to analyses. sexually aroused). When we included both approach and avoidance goals
simultaneously, approach social goals significantly predicted daily arousal,
Social Goals and Daily Sexual Desire .18, t(69) 2.39, p .05, and desire for sex, .15, t(69) 2.04,
p .05, and marginally predicted daily enjoyment, .15, t(69) 1.91,
As in Studies 1 and 2, we predicted that approach social goals p .10. There were no significant associations between avoidance social
would predict increased daily sexual desire. To test this hypothe- goals and any of the individual-item dependent measures of sexual desire.
818 IMPETT, STRACHMAN, FINKEL, AND GABLE

Table 1 Relationship Events and Daily Sexual Desire


Associations Between Sexual Goals and Daily Sexual Desire in
Study 3 A third aim of this study was to examine whether daily rela-
tionship events predicted daily sexual desire and whether approach
Outcome: daily social goals moderated these associations. Participants reported an
sexual desire average of 4.23 positive events and 1.25 negative events each day.
Variable t
The most common positive events included my partner told me
that he/she loves me and my partner made me laugh. The most
Approach sexual goals common negative events included my partner did something that
Enhancement goals .65*** 7.87a made me feel irritated or angry and my partner and I had a minor
Intimacy goals .21** 2.95a
disagreement. As predicted, on days when participants reported
Avoidance sexual goals
Self-affirmation goals .12 1.74a more frequent positive events (than their own average across the
Coping goals .04 0.86a 14-day study), they reported significantly greater sexual desire,
Partner approval goals .11 1.24a .26, t(249) 2.69, p .01. On days when participants
Peer approval goals .02 0.11a reported more frequent negative events, they reported significantly
Control variables
Relationship duration .01 0.08b less sexual desire, .12, t(249) 2.34, p .05.
Relationship satisfaction .12* 1.88b We further predicted that approach social goals would moderate
Sexual frequency .17* 1.85b the association between daily positive relationship events and
sexual desire. To test this hypothesis, we predicted daily sexual
Note. All numbers are standardized hierarchial linear model coefficients.
desire from positive events at Level 1. At Level 2, we included
a
df 244. b df 68.

p .10. * p .05. ** p .01. *** p .001. approach social goals (grand mean centered) as a predictor of both
the intercept of sexual desire and the slope of positive events with
sexual desire. A similar model was used for negative events.
For positive events, approach social goals were a marginally
the other two measures of avoidance sexual goals (i.e., Self- significant predictor of the slope between sexual desire and posi-
Affirmation and Partner Approval) were marginally negatively tive events, .18, t(247) 1.83, p .07, such that compared
associated with sexual desire. with those with weak approach goals, individuals with strong
approach goals experienced a marginally greater increase in sexual
desire on days when they reported more positive events (see Figure
Approach Sexual Goals as a Mediator 4). When relationship duration, sexual frequency, and relationship
satisfaction were added as covariates to Level 2, this effect re-
Another hypothesis, supported in Study 2, was that approach
mained marginally significant, .16, t(244) 1.65, p .10.
sexual goals would mediate the association between approach
We also tested whether approach social goals moderated the as-
social goals and sexual desire. Study 3 used a more detailed
sociation between negative events and desire. Approach social
measure of sexual goals than that included in Study 2, with many goals were a significant predictor of the slope between sexual
items distinguishing between self-focused (i.e., Enhancement) and desire and negative events, .12, t(247) 2.14, p .05, such
other-focused (i.e., Intimacy) approach sexual goals. Before test- that compared with those with weak approach goals, individuals
ing for mediation, we examined associations between approach with strong approach goals experienced less of a decrease in sexual
social goals and both types of approach sexual goals. Approach desire on days when they reported more negative events (see
social goals were not associated with enhancement sexual goals, Figure 5). When relationship duration, sexual frequency, and re-
.04, p .72, but were associated with intimacy sexual goals, lationship satisfaction were added as covariates to Level 2, this
.40, t(69) 4.43, p .001. Therefore, in the following effect remained significant, .12, t(244) 2.08, p .05.
analyses, we examined intimacy sexual goals as a mediator of the Additional analyses conducted to determine whether the interac-
association between approach social goals and sexual desire. We tions between approach goals and positive and negative events
used the aggregated data and the composite measure of sexual were further moderated by gender revealed no significant effects.
desire.
As in Study 2, standard (OLS) hierarchical regression analysis
based on the principles of Baron and Kenny (1986) was used to
Brief Discussion
test mediation. Approach social goals were marginally associated Study 3 extended the results of the previous two studies in
with sexual desire (r .20, p .08). Approach social goals were several important ways. First, it showed that approach social goals
significantly associated with intimacy sexual goals (r .42, p measured more generally predict daily sexual desire. Second, it
.001). Finally, intimacy sexual goals significantly predicted sexual extended the findings from Study 2 by showing that approach
desire after we controlled for approach social goals, .39, p sexual goals that focus on the self (e.g., to pursue ones own sexual
.01, and the marginally significant direct effect from approach pleasure) and approach sexual goals that focus on the partner/
social goals to sexual desire dropped to nonsignificance, .04, relationship (e.g., to please ones partner or enhance intimacy)
p .76. A significant Sobel (1982) test indicated that the drop in were both associated with increased sexual desire. Third, it re-
the value of the betas was significant (z 2.36, p .05), vealed that other-focused approach sexual goals (intimacy goals)
providing evidence for mediation. The pattern of results is similar mediated the association between approach social goals and daily
to the one found in Study 2, which is displayed in Figure 3. sexual desire. Fourth, it replicated the results of Studies 2 and 3
APPROACH GOALS AND SEXUAL DESIRE 819

Figure 4. Approach social goals as a moderator of the association between positive events and sexual desire
(Study 3).

showing that avoidance social and sexual goals were not signifi- in enabling dating couples to maintain high levels of sexual desire.
cantly associated with daily sexual desire. Finally, it showed that Study 1 showed that the adoption of approach relationship goals
relationship events are an important moderator of the link between buffered against declines in sexual desire over a 6-month period in
approach social goals and daily sexual desire. More specifically, relationships. Whereas people with weak approach goals (i.e.,
people with strong approach goals experienced even greater sexual those possessing a lack of interest in pursuing growth, fun, and
desire on days that they reported many positive events and less of development in their relationships) experienced declines in sexual
a decrease in sexual desire on days that they reported many desire over the course of the 6-month study, people with strong
negative events than individuals with weak approach social goals. approach goals (i.e., those who possessed a great deal of interest in
pursuing positive outcomes in their relationships) maintained high
General Discussion levels of sexual desire over the course of the study.
Numerous studies have documented the importance of sexual Study 2 showed that approach relationship goals predicted ele-
desire in promoting satisfaction and stability in long-term relation- vated sexual desire during daily sexual interactions and that this
ships (e.g., Yeh et al., 2006). Many individuals report that their association was mediated by approach sexual goals. That is, people
own or a partners low sexual desire creates problems for their who are generally oriented toward creating positive outcomes in
relationships (Laumann et al., 1994), and some couples seek sex their relationships may view sexual interactions as one way to
therapy in order to deal with one or both partners lack of sexual create closeness and intimacy, and their approach sexual goals
desire (McCarthy, 1999). The three studies described in this article may, in turn, predict greater desire during daily sexual interactions.
provide converging support for the importance of approach goals Finally, Study 3 showed that approach sexual goals that focus on

Figure 5. Approach social goals as a moderator of the association between negative events and sexual desire
(Study 3).
820 IMPETT, STRACHMAN, FINKEL, AND GABLE

the self (e.g., to pursue ones own sexual pleasure) and approach of the couple. Oftentimes, the assessment of sexual desire in
sexual goals that focus on others (e.g., to please ones partner or to couples is relative; that is, people perceive that their sexual desire
enhance intimacy) were both associated with daily sexual desire. is either too low or too high only after comparing their own desire
Moreover, Study 3 showed that people with strong approach goals with their partners desire (Davies, Katz, & Jackson, 1999; Ellison,
experienced even greater sexual desire on days that they reported 2001). Future research should obtain sexual desire reports from
many positive events. This effect is consistent with previous re- both members of dating or married couples. We also measured the
search that has shown that strong approach tendencies are associ- relationship and sexual goals of only one member of the couple;
ated with even greater increases in approach behaviors when however, goals in relationships are different than goals in other
signals of movement toward the goal (i.e., gains) are experienced domains such as achievement or other life tasks in that they
(Forster, Higgins, & Idson, 1998). Finally, approach goals seemed involve coordinating with another person who has his or her own
to buffer against the deleterious effect that negative relationship goals. For example, what are the implications for sexual function-
events had on sexual desire, such that individuals with strong ing in a relationship if one partner has strong approach goals and
approach goals reported less of a decrease in desire on days that the other partner has weak approach goals? Future research should
they reported many negative events than individuals with weak examine the joint contribution of both partners goals to both
approach goals. Although we did not specifically predict this partners sexual desire, sampling the partners feelings at specific
finding, it is consistent with previous work that has found that moments in their daily lives as well as over longer periods of time.
approach goals are associated with interpreting ambiguous or Most of the participants in these studies were college students in
neutral information in a positive manner (Strachman & Gable, relatively new relationships in which sexual desire may have been
2006). Thus, those with strong approach goals may have reframed near its peak. It is possible that the effect of relationship goals on
negative events more positively, which may have attenuated the desire might be even more magnified in relationships of greater
association between negative events and sexual desire. duration and commitment, such as in married couples. It is also
The current studies used a combination of longitudinal and daily possible that approach goals fail to promote sexual desire in
experience methods to examine the link between approach goals long-term couples who have already experienced steep declines in
and sexual desire. In Study 1, participants provided biweekly sexual desire. Future research focusing on relationships of greater
assessments of sexual desire, enabling us to examine the influence duration and commitment is needed to examine these ideas. An-
of approach goals measured early in relationships on the mainte- other important direction for future research to examine is the
nance of sexual desire over a 6-month period. In Studies 2 and 3, benefits of adopting approach goals for other aspects of relation-
participants provided daily accounts of their sexual desire, en- ships in addition to sexuality. Desires to pursue growth and de-
abling us to obtain accurate, daily accounts of sexual desire. The velopment in relationships may also be associated with other
use of a daily experience method enabled us to study relationship positive behaviors and processes such as relationship commitment
processes within the context of daily life in a way that is not (Strachman & Gable, 2006), willingness to sacrifice (Impett, Ga-
possible with more traditional, cross-sectional designs (Bolger, ble, & Peplau, 2005), and willingness to forgive a partners wrong-
Davis, & Rafaeli, 2003). doings (Finkel, Rusbult, Kumashiro, & Hannon, 2002).
These studies contribute to a growing body of research demon- Another limitation stemmed from the fact that participants
strating the utility of approachavoidance models of motivation in sexual desire was only assessed on days when they engaged in
understanding a broad range of phenomena in everyday life (e.g., sexual intercourse (in Studies 2 and 3). Individuals sometimes
Elliot & Sheldon, 1997; Gable et al., 2000). More specifically, the choose to engage in sexual activity in order to please the partner or
current studies are part of an emerging area of research that to avoid conflict rather to out of personal sexual interest (Levine,
focuses on motivation and close relationships (Gable, 2006b; Ga- 2002). Indeed, research has shown that both men and women
ble & Strachman, 2007; Impett, Gable, & Peplau, 2005). Previous report having engaged in sexual behavior in the absence of desire
research guided by Gables (2006b) model of social motivation has (see review by Impett & Peplau, 2003), suggesting that the expe-
shown that approach (but not avoidance) motives and goals are rience of sexual desire does not entirely overlap with sexual
associated with positive outcomes including positive emotions and behavior. An interesting direction for future research would be to
relationship satisfaction (Gable, 2006a, 2006b; Impett, Gable, & assess sexual desire both on days that couples engage in sexual
Peplau, 2005). Our results show that sensitivity to positive rela- activity and on days that sexual activity does not occur. Future
tionships processes, such as intimacy, growth, and fun, has impor- research would also benefit from the use of a measure of sexual
tant implications for close relationships that are independent and desire that distinguishes between desire in solitary and dyadic
separate from sensitivity to negative processes, such as conflict contexts (e.g., Spector, Carey, & Steinberg, 1996). Finally, this
and rejection. It is particularly important to note that there are far research was centrally concerned with the motivational component
fewer studies focusing on the role of positive processes than those of the human sexual response (i.e., sexual desire), but the measures
focused on negative processes in the field of close relationships, used in each of the studies included related sexual constructs such
reflecting possible empirical and theoretical oversights (Gable & as sexual enjoyment and arousal. Although auxiliary analyses
Haidt, 2005; Reis & Gable, 2003). using individual items (i.e., sexual arousal, desire, and enjoyment)
did not change the pattern of results, it will be important for future
Limitations and Future Directions research to include more nuanced measures to capture possibly
meaningful distinctions among these interrelated sexual constructs.
The results of these studies provide several interesting directions Although our theoretical framework proposes that motivation
for future research. First, similar to most of the available research influences sexual desire, our data do not provide a definitive test of
on sexual desire, the current research focused on only one member this direction of causality. It is also possible that experiencing high
APPROACH GOALS AND SEXUAL DESIRE 821

levels of sexual desire may also cause people to pursue approach studies documented the utility of approachavoidance motiva-
goals in their relationships. Future research in which both approach tional theory as well as the important roles of both approach
and avoidance goals are experimentally manipulated (Strachman relationship and sexual goals in helping individuals to maintain
& Gable, 2006) would provide a more refined explanation of the high levels of sexual desire.
findings reported in the current studies. Finally, while the results of
the two daily experience studies suggest that the association be- References
tween approach goals and sexual desire applies equally well to Aron, A., Norman, C. C., Aron, E. N., McKenna, C., & Heyman, R. E.
White and Asian participants, it will be important for future (2000). Couples shared participation in novel and arousing activities
research to replicate these effects in a sample of greater racial/ and experienced relationship quality. Journal of Personality and Social
ethnic diversity and in non-Western cultures. Psychology, 78, 273284.
Baron, R. M., & Kenny, D. A. (1986). The moderatormediator variable
distinction in social psychological research: Conceptual, strategic, and
Implications statistical considerations. Journal of Personality and Social Psychology,
51, 11731182.
The results from three studies document the importance of
Basson, R. (2002). Womens sexual desire: Disordered or misunderstood?
approach goals for predicting elevated levels of sexual desire on a Journal of Sex and Marital Therapy, 28, 1728.
daily basis and maintaining desire over time. Unfortunately, the Basson, R. (2006). Sexual desire and arousal disorders in women. New
current study cannot address the question of whether it is possible England Journal of Medicine, 354, 14971506.
for people with chronically low levels of approach goals to learn to Basson, R., Leiblum, S., Brotto, L., Derogatis, L., Fourcroy, J., Fugl-
focus on the positive things to be experienced in their relation- Meyer, K., et al. (2004). Revised definitions of womens sexual dys-
ships. Nevertheless, it is important to note that, by definition, goals functions. Journal of Sexual Medicine, 1, 40 48.
are short-term cognitive representations of wants and fears that Baumeister, R. F., Catanese, K. R., & Vohs, K. D. (2001). Is there a gender
should be malleable and sensitive to situational cues (Gable, difference in strength of sex drive? Theoretical views, conceptual dis-
2006b). Moreover, previous research has shown that goals can be tinctions, and a review of relevant evidence. Personality and Social
Psychology Review, 5, 242273.
experimentally manipulated in the achievement domain (e.g., El-
Beck, J. G. (1995). Hypoactive sexual desire disorder: An overview.
liot & Harackiewicz, 1996) and in the highly similar area of Journal of Consulting and Clinical Psychology, 63, 919 927.
regulatory focus research (e.g., Shah, Higgins, & Friedman, 1998). Beck, J. G., Bozman, A. W., & Qualtrough, T. (1991). The experience of
Experimental evidence for changing relationship goals in situ has sexual desire: Psychological correlates in a college sample. Journal of
yet to be conducted, but on the basis of theory and previous Sex Research, 28, 443 456.
experimental research, we expect that it is possible for peoples Bolger, M., Davis, A., & Rafaeli, E. (2003). Diary methods: Capturing life
goals in their relationships to change over time. as it is lived. Annual Review of Psychology, 54, 579 616.
This research also points to the central importance of consider- Boyatzis, R. E. (1973). Affiliation motivation. In D. C. McClelland & R. S.
ing the role of gender in understanding sexual desire in intimate Steele, (Eds.) Human motivation: A book of readings (pp. 252276).
relationships. Approach goals were a stronger predictor of sexual Morristown, NJ: General Learning Press.
Brezsnyak, M., & Whisman, M. A. (2004). Sexual desire and relationship
desire for women than for men in the two studies that used a
functioning: The effects of marital satisfaction and power. Journal of Sex
measure of approach goals in romantic relationships. Thus, the and Marital Therapy, 30, 199 217.
results of this study support a growing body of research that Carver, C. S., Sutton, S. K., & Scheier, M. F. (2000). Action, emotion, and
demonstrates the importance of relationship dynamics for wom- personality: Emerging conceptual integration. Personality and Social
ens sexual desire relative to mens (e.g., Basson, 2002, 2006; Psychology Bulletin, 26, 741751.
Peplau, 2003). In recent years, publicity about treatments for Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral
mens erection problems focused attention on womens sexuality activation, and affective responses to impending reward and punishment:
and provoked a competitive commercial hunt for the female The BIS/BAS scales. Journal of Personality and Social Psychology, 67,
Viagra (Tiefer, 2002). This hunt reflects a general medicalization 319 333.
of sexuality; many people are trying to find pharmaceutical so- Clayton, A. H., Segraves, R. T., Leiblum, S., Basson, R., Pyke, R., Cotton,
D., et al. (2006). Reliability and validity of the Sexual Interest and
lutions to what in some cases may be relationship problems.
Desire InventoryFemale (SIDIF), a scale designed to measure sever-
Because womens sexual desire is much more closely tied to their ity of female hypoactive sexual desire disorder. Journal of Sex and
goals in relationships than mens sexual desire, attempts to boost Marital Therapy, 32, 115135.
womens sexual desire through pharmacological intervention may Cooper, M. L., Shapiro, C. M., & Powers, A. M. (1998). Motivations for
be misguided. The results of this research highlight the importance sex and risky sexual behavior among adolescents and young adults: A
of considering the interpersonal aspects of relationships when functional perspective. Journal of Personality and Social Psychology,
thinking about how to treat problems of low sexual desirefor 75, 1528 1558.
women but also for men. Davies, S., Katz, J., & Jackson, J. L. (1999). Sexual desire discrepancies:
Effects on sexual and relationship satisfaction in heterosexual dating
couples. Archives of Sexual Behavior, 28, 553567.
Concluding Comments Diamond, L. M. (2004). Emerging perspectives on distinctions between
romantic love and sexual desire. Current Directions in Psychological
While a wealth of research has documented important links Science, 13, 116 119.
between sexual desire and relationship quality in intimate relation- Dowson, M., & McInerney, D. M. (2003). What do students say about their
ships (e.g., Regan, 2000; Sprecher, 2002; Yeh et al., 2006), much motivational goals? Towards a more complex and dynamic perspective
less research has investigated factors that may help promote and on student motivation. Contemporary Educational Psychology, 28, 91
sustain sexual desire in relationships over time. In this article, three 113.
822 IMPETT, STRACHMAN, FINKEL, AND GABLE

Elliot, A., & Harackiewicz, J. M. (1996). Approach and avoidance achieve- views of marital sex: A dyadic analysis. Journal of Sex Research, 29,
ment goals and intrinsic motivation: A mediational analysis. Journal of 343360.
Personality and Social Psychology, 70, 461 475. Kaplan, H. S. (1979). Hypoactive sexual desire. Journal of Sex and Marital
Elliot, A. J., & Covington, M. V. (2001). Approach and avoidance moti- Therapy, 19, 324.
vation. Educational Psychology Review, 13, 7392. Klusmann, D. (2002). Sexual motivation and the duration of partnership.
Elliot, A. J., Gable, S. L., & Mapes, R. R. (2006). Approach and avoidance Archives of Sexual Behavior, 31, 275287.
motivation in the social domain. Personality and Social Psychology Laumann, E. O., Gagnon, J. H., Michael, R. T., & Michaels, S. (1994). The
Bulletin, 32, 378 391. social organization of sexuality: Sexual practices in the United States.
Elliot, A. J., & Sheldon, K. M. (1997). Avoidance achievement motivation: Chicago: University of Chicago Press.
A personal goals analysis. Journal of Personality and Social Psychol- Leitenberg, H., & Henning, K. (1995). Sexual fantasy. Psychological
ogy, 73, 171185. Bulletin, 117, 469 496.
Ellison, C. R. (2001). A research inquiry into some American womens Levine, S. B. (2002). Reexploring the concept of sexual desire. Journal of
sexual concerns and problems. Women and Therapy, 24, 147159.
Sex and Marital Therapy, 28, 39 51.
Finkel, E. J., Rusbult, C. E., Kumashiro, M., & Hannon, P. A. (2002).
Levine, S. B. (2003). The nature of sexual desire: A clinicians perspective.
Dealing with betrayal in close relationships: Does commitment promote
Archives of Sexual Behavior, 32, 279 285.
forgiveness? Journal of Personality and Social Psychology, 82, 956
Littell, R. C., Milliken, G., Stroup, W. W., & Wolfinger, R. D. (1996). SAS
974.
system for mixed models. Cary, NC: SAS Institute.
Forster, J., Higgins, E. T., & Idson, L. C. (1998). Approach and avoidance
Masters, W. H., & Johnson, V. E. (1966). Human sexual response (p. 311).
strength during goal attainment: Regulatory focus and the goal looms
Boston: Little, Brown.
larger effect. Journal of Personality and Social Psychology, 75, 1115
1131. McCabe, M. P. (1997). Intimacy and quality of life among sexually
Fredrickson, B. L., & Joiner, T. (2002). Positive emotions trigger upward dysfunctional men and women. Journal of Sex and Marital Therapy, 23,
spirals toward emotional well-being. Psychological Science, 13, 172 276 290.
175. McCarthy, B. W. (1999). Marital style and its effects on sexual desire and
Gable, S. L. (2006a). Approach and avoidance relationship goals. Unpub- functioning. Journal of Family Psychotherapy, 10, 112.
lished manuscript, University of California, Los Angeles. McCarthy, B. W., & McCarthy, E. (2003). Rekindling desire: A step-by-
Gable, S. L. (2006b). Approach and avoidance social motives and goals. step program to help low-sex and no-sex marriages. New York:
Journal of Personality, 74, 175222. Brunner-Routledge.
Gable, S. L., & Haidt, J. (2005). What (and why) is positive psychology? Mehrabian, A. (1976). Questionnaire measures of affiliative tendency and
Review of General Psychology, 9, 103110. sensitivity to rejection. Psychological Reports, 38, 199 209.
Gable, S. L., Reis, H. T., & Downey, G. (2003). He said, she said: A Peplau, L. A. (2003). Human sexuality: How do men and women differ?
quasi-signal detection analysis of daily interaction between close rela- Current Directions in Psychological Science, 12, 37 40.
tionship partners. Psychological Science, 14, 100 105. Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models:
Gable, S. L., Reis, H. T., & Elliot, A. J. (2000). Behavioral activation and Applications and data analysis methods. Thousand Oaks, CA: Sage.
inhibition in everyday life. Journal of Personality and Social Psychol- Raudenbush, S. W., Bryk, A. S., Cheong, Y. F., & Congdon, R. T. (2000).
ogy, 78, 11351149. HLM5: Hierarchical linear and nonlinear modeling. Chicago, Scientific
Gable, S. L., & Strachman, A. (2007) Approaching social rewards and Software.
avoiding social punishments: Appetitive and aversive social motivation. Regan, P. C. (2000). The role of sexual desire and sexual activity in dating
In J. Shah & W. Gardner (Eds.) Handbook of motivation science (pp. relationships. Social Behavior and Personality, 28, 5159.
561575). New York: Guilford. Regan, P. C., & Berscheid, E. (1999). Lust: What we know about human
Goldstein, A., & Brandon, M. (2004). Reclaiming desire: Four keys to sexual desire. Thousand Oaks, CA: Sage.
finding your lost libido. New York: Rodale. Reis, H. T., & Gable, S. L. (2000). Event-sampling and other methods for
Gray, J. (1987). The psychology of fear and stress (2nd ed.). New York: studying everyday experience. In H. T. Reis & C. M. Judd (Eds.),
Cambridge. Handbook of research methods in social and personality psychology (pp.
Gray, J. A. (1990). Brain systems that mediate both emotion and cognition.
190 222). New York: Cambridge University Press.
Cognition and Emotion, 4, 269 288.
Reis, H. T., & Gable, S. L. (2003). Toward a positive psychology of
Hawton, K., Catalan, J., & Fagg, J. (1991). Low sexual desire: Sex therapy
relationships. In C. L. Keyes & J. Haidt (Eds.). Flourishing: The positive
results and prognostic factors. Behavioral Research and Therapy, 29,
person and the good life (pp. 129 159). Washington, DC: American
217224.
Psychological Association.
Impett, E. A., Gable, S. L., & Peplau, L. A. (2005). Giving up and giving
Rusbult, C. E., Martz, J. M., & Agnew, C. R. (1998). The investment
in: The costs and benefits of daily sacrifice in intimate relationships.
model scale: Measuring commitment level, satisfaction level, quality of
Journal of Personality and Social Psychology, 89, 327344.
Impett, E. A., & Peplau, L. A. (2003). Sexual compliance: Gender, moti- alternatives, and investment size. Personal Relationships, 5, 357391.
vational, and relationship perspectives. Journal of Sex Research, 40, Shah, J., Higgins, T., & Friedman, R. S. (1998). Performance incentives
87100. and means: How regulatory focus influences goal attainment. Journal of
Impett, E. A., Peplau, L. A., & Gable, S. L. (2005). Approach and Personality and Social Psychology, 74, 285293.
avoidance sexual motivation: Implications for personal and interpersonal Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis.
well-being. Personal Relationships, 12, 465 482. New York: Oxford University Press.
Impett, E. A., & Tolman, D. L. (2006). Late adolescent girls sexual Snijders, T., & Bosker, R. (1999). Multilevel analysis: An introduction to
experiences and sexual satisfaction. Journal of Adolescent Research, 6, basic and advanced multilevel modeling. London: Sage.
628 646. Sobel, M. E. (1982). Asymptotic confidence intervals for indirect effects in
Johnson, A. M., Wadsworth, J., Wellings, K., & Field, J. (1994). Sexual structural equation models. In S. Leinhardt (Ed.), Sociological method-
attitudes and lifestyles. London: Blackwell. ology (pp. 290 312). Washington, DC: American Sociological Associ-
Julien, D., Bouchard, C., Gagnon, M., & Pomerleau, A. (1992). Insiders ation.
APPROACH GOALS AND SEXUAL DESIRE 823

Sorkin, D. H., & Rook, K. S. (2004). Interpersonal control strivings and affects what you see (and do not see): Avoidance social goals and social
vulnerability to negative social exchanges in later life. Psychology and events. Personality and Social Psychology Bulletin, 32, 113.
Aging, 19, 555564. Tiefer, L. (2002). Beyond the medical model of womens sexual problems:
Spector, I. P., Carey, M. P., & Steinberg, L. (1996). The Sexual Desire A campaign to resist the promotion of female sexual dysfunction.
Inventory: Development, factor structure, and evidence of reliability. Sexual and Relationship Therapy, 17, 127135.
Journal of Sex and Marital Therapy, 22, 175190. Trudel, G., Landry, L., & Larose, Y. (1997). Low sexual desire: The role
Sprecher, S. (2002). Sexual satisfaction in premarital relationships: Asso- of anxiety, depression, and marital adjustment. Sexual and Marital
ciations with satisfaction, love, commitment and stability. Journal of Sex Therapy, 12, 9599.
Research, 39, 190 196. Yeh, H.-C., Lorenz, F. O., Wickrama, K. A. S., Conger, R. D., & Elder,
Sprecher, S., & Cate, R. M. (2004). Sexual satisfaction and sexual expres- G. H. (2006). Relationships among sexual satisfaction, marital quality,
sion as predictors of relationship satisfaction and stability. In J. Harvey, and marital instability at midlife. Journal of Family Psychology, 20,
A. Wenzel, & S. Sprecher (Eds.), Handbook of sexuality in close 339 343.
relationships. Mahwah, NJ: Erlbaum.
Sprecher, S., & Regan, P. C. (1996). College virgins: How men and women Received April 3, 2007
perceive their sexual status. Journal of Sex Research, 33, 315. Revision received August 7, 2007
Strachman, A., & Gable, S. L. (2006). What you want (and do not want) Accepted August 23, 2007
Journal of Personality and Social Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 94, No. 5, 824 838 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.824

Receiving Support as a Mixed Blessing: Evidence for Dual Effects of


Support on Psychological Outcomes

Marci E. J. Gleason Masumi Iida and Patrick E. Shrout


Wayne State University New York University

Niall Bolger
Columbia University

Although social support is thought to boost feelings of closeness in dyadic relationships, recent findings
have suggested that support receipt can increase distress in recipients. The authors investigated these
apparently contrary findings in a large daily diary study of couples over 31 days leading up to a major
stressor. Results confirm that daily support receipt was associated with greater feelings of closeness and
greater negative mood. These average effects, however, masked substantial heterogeneity. In particular,
those recipients showing greater benefits on closeness tended to show lesser cost on negative mood, and
vice versa. Self-esteem was examined as a possible moderator of support effects, but its role was evident
in only a subset of recipients. These results imply that models of dyadic support processes must accord
a central role to between-individual heterogeneity.

Keywords: close relationships, daily diaries, social support, reciprocity, multilevel models

Perceived availability of social support (the belief that social ships are positively associated with relationship satisfaction (Reis
support has been available to one in the past and will be in the & Patrick, 1996; Sanderson & Cantor, 1997), it would seem that
future) has been linked to a variety of beneficial outcomes (Sara- instances of support receipt could lead to lowered relationship
son, Sarason, & Pierce, 1994). However, the beneficial effects of closeness as a secondary outcome.
perceived support stand in contrast to those for concrete acts of We review the strength of the evidence that support receipt has
support (Lakey & Lutz, 1996). Although a few studies have shown adverse effects on individual well-being, then review the possible
that support receipt is related to some positive outcomes (Feeney effects of support receipt on relationship closeness and intimacy,
& Collins, 2001, 2003), many studies of actual support transac- and finally outline the strengths of considering both negative mood
tions find that support receipt is associated with negative outcomes and relationship closeness outcomes in a single study of social
and in particular with increased negative mood (Barrera, 1981; support receipt.
Bolger, Zuckerman, & Kessler, 2000; Liang, Krause, & Bennett,
2001; Shrout, Herman, & Bolger, 2006).
How Strong Is the Evidence That Support Receipt Can
Studies of support transactions also have a tendency to focus on
Lead to Increased Distress?
individual emotions and well-being and are less likely to measure
relationship-level variables such as closeness and felt intimacy. As discussed above, there is an established literature linking
Given that negative affect is inversely associated with relationship generalized perceived support to better outcomes, including re-
satisfaction (Bradbury & Fincham, 1989; Edwards, Nazroo, & duced distress (e.g., Cohen, 2004), but daily support receipt is
Brown, 1998; Gottman, 1979; Uebelacker, Courtnage, & Whis- frequently linked to negative outcomes. Given the seemingly con-
man, 2003) and feelings of closeness and intimacy within relation- tradictory nature of these findings, the association between daily
support and negative mood has been questioned. Specifically, it
has been suggested that the apparent negative effects of support
Marci E. J. Gleason, Communication and Behavioral Oncology, Kar- receipt could be due to (a) reverse causation, that is, distress
manos Cancer Institute, Wayne State University; Masumi Iida and Patrick leading to support provision, or (b) a common third cause, such as
E. Shrout, Department of Psychology, New York University; Niall Bolger, stress leading to both distress and support, which would create a
Department of Psychology, Columbia University. spurious association
This research was supported by National Institute of Mental Health Two kinds of evidence argue against reverse causation. One is
Grant MH60366 to Niall Bolger. We gratefully acknowledge the invalu- the use of lagged variables so that the association of distress on one
able contributions of Amie Rapaport, Diane Ruble, Hiro Yoshikawa, Tom
day is related to support on the previous day (e.g. Bolger et al.,
Tyler, Gwen Seidman, Christopher Burke, and the Couples Lab of New
York University.
2000; Shrout et al., 2006). In these lagged models, yesterdays
Correspondence concerning this article should be addressed to Marci negative mood is adjusted for statistically such that the adverse
E. J. Gleason, Karmanos Cancer Institute, 4100 John R Street, ROC effects of yesterdays support on todays mood cannot be due to
Room 336, Detroit, MI 48201; email: marci.gleason@gmail.com. Further simple build-up of negative mood. Bolger et al. (2000; and also
information can also be found at http://www.psych.nyu.edu/couples. Shrout et al., 2006, in a somewhat more elaborate model) found

824
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 825

that the negative effects of yesterdays support remain after ad- support than one has provided) and underbenefit (providing more
justment. The second source of evidence is from an examination of support than one has received) are psychologically distressing and
simulated data that were constructed to conform to a pattern in that individuals are motivated to restore equity either behaviorally
which distress elicits support (Seidman, Shrout, & Bolger, 2006). by providing or eliciting aid from caregivers or cognitively by
Seidman et al. (2006) constructed the fictitious data using a sim- psychologically justifying the inequity (Buunk & Schaufeli, 1999;
ulation strategy that was first outlined by Abelson (1968) and then Uehara, 1995; Walster et al., 1973). Buunk and Schaufeli (1999)
analyzed this reverse causation data with Bolger et al.s analytic took an evolutionary approach to reciprocity, suggesting that it is
approach. They concluded that the effects obtained empirically by a basic psychological mechanism that developed to maintain social
Bolger et al. could not have been obtained with data that were relationships and indicate individuals importance in their social
constructed under the reverse causation model. groups. Research on couples in which one member is seriously ill
The other alternative model is that of a third variable that leads has shown that both the ill spouse and the caregiving spouse suffer
to both distress and support provisiona spurious association from frustration, anger, depression, and resentment when the re-
model. An example of this alternative model is the support- lationship is not judged to be reciprocal (Thompson, Medvene, &
seekingtriage model, which posits that the negative outcomes Freedman, 1995).
from receiving support are due not to the support itself, but instead From the perspective of reciprocity theory, Uehara (1995) spe-
to support and psychological distress being simultaneously caused cifically argued that it is being overbenefitedreceiving support
by a precipitating stressful event (Barrera, 1986). According to this without returning itthat is particularly psychologically distress-
model, it is not receiving support that causes distress, but the ing. In this case, recipients are likely to feel obligated to repay
stressor that simultaneously evokes both psychological distress what was given to them, and when they cannot, they begin to doubt
and increased support from others. Support and negative mood their status and usefulness in the relationship (see also Roberto &
coincide because both are responses to negative events, not be- Scott, 1986). In a daily diary study of committed couples, we
cause they are causally linked to negative mood. found that individuals reported increased negative affect and de-
However, the association between support and negative out- creased positive affect on days on which they reported receiving
comes typically remains even after adjustment for relevant third support from, but not providing support to, their partners (over-
variables. Krause (1997), in a nationwide study of 60-year-olds in benefit) as compared with days when they only provided support
Great Britain, found that even when adjusting for health status, to their partners (underbenefit) or both provided support to and
individuals who received support had an increased mortality risk, received support from their partner (equitable or reciprocal ex-
and those who had high perceived availability of support had changes; Gleason, Iida, Bolger, & Shrout, 2003).
decreased mortality risk. Experimental studies have also demon- A different explanation for a tendency for distress to increase
strated that support receipt and not just the precipitating stressor with received support is one that focuses on possible effects of
can have deleterious effects on mood. Bolger and Amarel (2007) support on the recipients self-esteem. Several studies have re-
found that students asked to give an impromptu speech were more ported that being helped is associated with decreased self-esteem
anxious when they received explicit, visible support from a con- and depressed mood in the recipient (Nadler, 1987; Nadler &
federate than were students who did not receive support. Further- Fisher, 1976). There is some evidence that this explanation is
more, the Seidman et al. (2006) simulation study discussed above especially relevant in close relationships (Nadler, Fisher, & Itzhak,
also investigated the third-variable explanation. They created fic- 1983). Fisher, Nadler, and Whitcher-Alagna (1982) proposed the
titious data in which the level of distress today was caused by threat-to-self-esteem model of aid or support receipt that posits
yesterdays distress, as well as adversity experienced today and that helping consists of both self-threatening and supportive com-
yesterday. Similarly, the support transactions today were modeled ponents. The self-threatening components can undermine the re-
to be more likely when support was provided yesterday and when cipients evaluation of their self-efficacy, competence, and coping
adversity was experienced either today or yesterday. Unlike the abilities, which can in turn lead to increased psychological distress.
results of the reverse causation simulation study, Seidman et al. On the other hand, Fisher et al. theorized that the supportive
found that when Bolger et al.s (2000) analysis strategy was used, components could provide comfort and a sense of being cared for
some spurious association was created by the omitted third vari- by the support provider.
able (adversity). However, the size of the bias was very small, even
when the magnitude of the effect of adversity was made to be Social Support as Relationship Enhancer
unrealistically large. Seidman et al. concluded that Bolger et al.s
effect sizes were unlikely to be due to an omitted third variable. It is perhaps this potential sense of being cared for by ones
Taken together, these findings suggest that the association between partner bolstered by the positivity of the perceived availability of
psychological distress and support receipt is not spurious. support that gives social support its positive reputation. Reis,
Clark, and Holmes (2004) and Cutrona (1996) have related the
Support Receipt and Increased Distress positive findings associated with perceived availability of social
support to a global construct called perceived responsiveness of the
Equity theory and reciprocity research have also been invoked partner to the self. This perception of partner responsiveness is the
to explain why the receipt of support can be negative (Uehara, central path to the development and maintenance of closeness and
1995; Walster, Berscheid, & Walster, 1973). Both approaches intimacy in relationships. Like perceived responsiveness to the
suggest that people will be most satisfied when they perceive their self, perceived availability of support seems to be based on both
supportive relationships as being equitable or reciprocal. Equity personality characteristics of the perceiver and actual supportive
and reciprocity theories posit that both overbenefit (receiving more interactions. People who judge themselves as being highly sup-
826 GLEASON, IIDA, SHROUT, AND BOLGER

ported are also judged by observers as being more supported, but Specifically, people who trust their partners may benefit from
support recipients who perceived their relationships as more pos- support (Group A in Figure 1), whereas people who lack that trust
itive also judge the support they receive more positively than may experience costs that have been described above (Group B in
observers (Collins & Feeney, 2000). Figure 1).
Given the research indicating that support receipt increases In contrast to the individual differences model, Model 2 in
negative mood, it is surprising that it is judged as positive by the Figure 1 represents a differential effects model, whereby a single
recipient at all. One possible explanation for this contradiction is support event leads to improved relationship closeness and in-
that support receipt makes one feel closer to the provider of that creased individual distress. We might imagine a stressed worker
support because it makes one feel cared for or responded to (Reis who comes home to a well-intending partner who attempts to
et al., 2004) even while increasing personal distress. Gable, provide him or her with a break. The worker might appreciate the
Gonzago, and Strachman (2006) found that when individuals were good intentions and feel closer to his or her partner but be dis-
supportive when talking with their partners about their partners tressed by the loss of an evening of productivity. If this were the
successes, the partners (i.e., the support recipients) rated the rela- typical pattern of support provision in the couple, then one would
tionship as more satisfying. Although a positive association be- witness simultaneous positive and negative effects of support acts.
tween receipt of support and positive relationship variables such as Although the individual differences model and the differential
satisfaction, closeness, and intimacy has been found in a few effects model appear to be discrete alternatives, they can actually
studies (Acitelli & Antonucci, 1994; Hagedoorn et al., 2000), it is be viewed as examples of a range of processes that might vary
unclear whether the support being assessed was actual support from couple to couple. For some pairs of partners, support events
received, perceived support, or some combination of both. Regard- could lead to closeness but not to distress; for other partners,
less, this research on support receipt and relationship variables support events could lead to distress but no closeness; and for still
raises the question of whether support has differential effects on others, there could be dual effects. In the population, some patterns
individual-level variables (e.g., personal distress) when compared of these relations might be more common than others.
with relationship-level variables (e.g., relationship closeness). Only three studies that we know of have reported on these two
processes in the same samples of partners. The Bolger and Amarel
Understanding Dual Effects of Support on Personal (2007) study cited earlier did find evidence for mixed effects of
Distress and Relationship Closeness support receipt. Students who received visible support experienced
larger increases in negative emotion than those in the nonsupport
The idea that support or aid can produce both increased psy- condition, and they also felt that their partners were more con-
chological distress and a sense of being cared for by the provider cerned, considerate, and supportive than those in the invisible
is particularly intriguing. There are at least two possible models of support condition. However, this study had only a single support
this pattern of effects: an individual differences model and a event and was not designed to examine individual differences in
within-person differential effects model. An individual differences response to support events. Gable, Reis, and Downey (2003) were
model would mean that support increases personal distress for able to study repeated support events among dating couples. They
some people but increases relationship closeness for others. A found that support events that were reported by both partners and
differential effects model would mean that support receipt leads to recipients (called hits) were related to both relationship well-
both increased personal distress and increased relationship close- being and recipient distress. Even though the data were based on
ness in the same person. Model 1 in Figure 1 shows a represen- diary reports that allow the study of individual differences, the
tation of an individual differences model of support receipt. In one authors did not include these individual differences (which are
group (Group A), there is no effect of support on individual called random effects in the multilevel statistics literature) in their
distress, but there is a strong and positive effect on relationship statistical model.
closeness. In the other group (Group B), there may be a strong Gleason, Iida, Bolger, and Shrout (2003) examined the effect of
effect of support on individual distress, but no improvement in imbalance in support provision and receipt on recipients negative
relationship closeness. If data from these two groups are combined mood, and Gleason (2005) analyzed data from the same daily diary
without a formal model of the nature of the moderation (individual study with a focus on relationship closeness outcomes. In both of
differences), then one might conclude from the mixed analysis that these analyses (i.e., for both outcomes), Gleason and her col-
couples are likely to experience both relationship exhilaration and leagues found evidence that the effects of unreciprocated support
individual distress. events varied from couple to couple, but on average unreciprocated
An individual differences model of the effects of support receipt support was associated with an increase in negative mood and in a
is consistent with aspects of the relationship enhancement model separate analysis with an increase in relationship closeness. How-
of social support (Cutrona, Russell, & Gardner, 2005). This model ever, it was not possible to determine from these two analyses how
explains the association between actual support receipt, perceived often the pattern in Model 2 (differential effects model) occurred
availability of support, relationship satisfaction, and health. It in the sample.
suggests that the perceived availability of support is directly re-
lated to instances of received support, particularly when the pro- The Current Study
vider is seen as a caring and committed partner, and that this
process is cyclical: People who receive consistent beneficial sup- To tease apart the effects of support receipt on negative mood
port will trust their partners, and people who trust their partners and relationship closeness, we analyzed data from a large daily
will benefit from support. Perhaps the negative effects of the diary study of nearly 300 cohabitating couples in which one
receipt of support can be explained by individual differences: partner was approaching a stressful event, the bar examination (a
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 827

Model 1: Individual Differences (Moderation) Relationship


Closeness
+

Group A Support
Event

Individual
Distress

Relationship
Closeness

Support
Group B Event
+
Individual
Distress

Model 2: Differential Effects

Relationship
Closeness
+
Support
Event
+
Individual
Distress

Figure 1. Possible models for understanding the effects of support receipt on individual distress and relation-
ship closeness. In Model 1, support receipt increases relationship closeness in some individuals (Group A) and
increases distress in others (Group B). In Model 2, support receipt increases both distress and closeness in all
individuals. The weak or missing effect is represented by a dashed arrow; the strong effect is represented by a
solid arrow.

difficult-to-pass licensing examination for lawyers that they must there was significant variation around it (a significant random
pass to practice). Using a dataset in which one member of the effect), we would know that individuals negative mood was
couple is approaching a significant stressor allowed us to investi- differentially affected by receipt of support. Furthermore, we could
gate whether responses to support receipt are affected by overall obtain estimates of each individuals receipt of support effect,
stress level and ensured that we captured couples at a time when which would reveal whether for some people receipt of support
support exchanges should have occurred frequently. decreased negative mood despite the average effect being an
A typical analysis would involve estimating and interpreting increase in negative mood or whether receipt of support increased
only the fixed or average effects. Although the fixed effects give negative mood in all individuals but to lesser and greater degrees.
valuable information about the predominant pattern of the data, In the current study, we took such an analysis one step further
fixed effects alone are unable to distinguish between models like and built a model in which we simultaneously modeled the effects
those discussed above. Estimating the random effects of the receipt of support receipt on negative mood and relationship closeness.
of support on negative mood and relationship closeness will pro- This special multilevel model is what Raudenbush and Bryk (2002,
vide evidence as to whether the effects of receipt of support on the pp. 185199) called a multivariate repeated measures model. The
outcomes differ between individuals. If receipt of support in- model and the large sample size allowed us to estimate the random
creased negative mood on average (a significant fixed effect) and effects of support receipt on both negative mood and closeness and
828 GLEASON, IIDA, SHROUT, AND BOLGER

then estimate the correlation between them. A significant correla- (SD 4.9). The composition of the sample was 76.8% White,
tion between the random effects of support receipt on negative 7.1% Asian, 4.4% Latino, 2.1% Black, 0.6% American Indian,
mood and relationship closeness would suggest that the effects are 5.1% other, and 3.9% not specified for examinees; 78.8% White,
systematically linked across individuals, whereas a null correlation 5.2% Asian, 4.8% Latino, 3.8% Black, 0.9% American Indian,
would suggest that the association between the effects of support 3.0% other, and 3.5% not specified for partners. Couples over-
receipt on negative mood and relationship closeness vary by indi- whelmingly agreed on how long they had been romantically in-
vidual but are not linked. This analysis is particularly powerful for volved (mean difference between estimates 0.03 years), with the
two reasons: (a) It allowed us to model the effects of receipt of average length of relationship being 6.5 years (SD 5.5), the
support on negative mood and closeness simultaneously, thereby minimum 8 months, and the maximum 48.8 years. This is a highly
allowing us to investigate how these effects are associated within educated sample, and therefore is not representative of the popu-
individuals, and (b) it allowed us to see whether and how people lation as a whole.
systematically differ in their reactions to support receipt without Couples were paid $150 for participation, and each couple was
having to identify an explanatory moderator. given a chance to win $1,000 on the completion of the study.
This second strength is particularly important. Conceptually, we Couples received an initial payment of $10, two consent forms,
tend to think about the heterogeneity of the responses to support two background questionnaires, and two return envelopes when
events as possible moderation, as illustrated in Model 1 of Fig- they agreed to participate in the study. They returned the com-
ure 1, but, as stated above, in practice the multilevel models do not pleted background questionnaires an average of 3 weeks before the
require that we specify the variables that distinguish Group A from start of the diary period. The diary period consisted of the 5 weeks
Group B. Given the difficulty of measuring all possible moderators before the bar exam, the 2 days on which the exam took place, and
and the fact that moderation is often difficult to find, it is partic- the week after the exam. Between 1 and 2 weeks before the start
ularly useful that we can identify systematic variation without of the diary period, both members of each couple received an
having to identify its specific source. initial packet containing a batch of daily diary questionnaires, a
In the current study, we first determined the average response to return envelope, and instructions regarding the diary question-
support receipt, then whether there was reliable variation in those naires. Packets were mailed to each participant on a weekly basis
responses; as a third step, we attempted to identify the variables (six packets over the 6 weeks of the study). Each batch consisted
that can account for such variation. The literature suggests two of seven identically structured daily diaries with the exception of
important candidates for moderating variables that we could ex- the last batch, which consisted of nine daily diaries. The diary form
amine. One is derived from the Cutrona et al. (2005) theory that included questions regarding mood, relationship closeness, daily
suggests that support receivers who are in more trusting and troubles or difficulties, relationship conflicts, and support transac-
satisfying relationships will find support to be more effective in tions. Participants were asked to complete the questionnaires sep-
reducing personal distress. Another is the proposition by Fisher et arately and not to share or discuss their answers with their partners.
al. (1982) that support can be a threat to self-esteempersons who Participants were also asked to complete the diaries on the days
have compromised self-esteem might be more vulnerable to a assigned and to indicate whether each diary had been completed on
threat associated with support acts. the correct day. Only entries that indicated that they had been
completed on the correct day were included in the analyses (88%
of completed diaries).
Method
Of those who agreed to participate, 89% returned their back-
Design and Participants ground questionnaire (372 couples in which both members re-
turned the background and 16 in which only one member did).
The data were collected in the summers of 2001, 2002, and 2003 Two hundred eighteen couples returned all materials (476 partic-
by contacting more than 100 law schools in the continental United ipants2). The final sample consisted of 293 examinees and 290
States. In 2001, 14 schools agreed to participate by allowing their partners who completed at least 1 week of the daily diaries.
graduating students to be contacted; in 2002, 27 schools partici-
pated; and in 2003, 30 schools participated. Because access to Measures: Dependent Variables
students marital or cohabitation status was unavailable before
recruitment, the school representatives were asked to distribute Closeness. Each evening, participants indicated separately
either a letter or an e-mail to their entire graduating class. Across how emotionally close and how physically intimate they were with
the 3 years, more than 15,000 students were contacted. To be their partner on a scale ranging from 0 to 4 with midpoints. High
eligible for participation, couples had to be married or cohabiting numbers indicated more emotional closeness and increased phys-
for at least 6 months at the time of the recruitment, and only 1 ical closeness, and low numbers indicated emotional distance and
member of the couple could be planning on taking the July bar
exam. There were 765 eligible couples who contacted us to par- 1
Because this sample is part of a larger study that focused on method-
ticipate, and of those 552 were assigned to the diary condition.1 Of
ology, interested couples were randomly assigned to different conditions.
those, 472 couples agreed to participate, resulting in an 86% Only 72% of interested couples were assigned to the diary design, and our
agreement rate. analyses use only those couples.
The average age of the examinee was 28.9 years (SD 6.4), and 2
In some couples, only 1 member returned materials, so the participant
the average age of the partner was 28.4 (SD 7.8). Of the numbers reflect how many individual participants returned the materials,
examinees, 46% were male. Sixty-four percent of the participants and couple numbers reflect the number of couples in which both members
were married, and the average length of cohabitation was 4.2 years returned the materials.
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 829

lack of physical closeness. Cronbachs alphas for the two items of data, but analyses including these days did not differ from the
were .71 for examinees and .68 for partners.3 Items were averaged findings presented. The 4th day of the study was also dropped to
to create the closeness scale (examinee: M 2.27, SD 1.07; account for the use of lagged variables (relationship closeness).
partner: M 2.25, SD 1.06). We adjusted for yesterdays The variable representing duration of time in the study was created
relationship closeness by including lagged relationship closeness such that Day 5 was coded 0, Day 6 was coded 1, and so on up to
as a predictor in the model. Controlling for lagged closeness results Day 35 (coded 31), resulting in values from 0 to 31 being included
in the outcome variable being residualized change in relationship in the analyses. Hereinafter, we refer to the days by their code
closeness (todays relationship closeness adjusting for yesterdays number. Day 31 was the day before the bar examination. We did
relationship closeness). This strengthens the claim that any change not include the days of or the days following the examination in
in relationship closeness is due to the events of the day in question the study to limit the sample to persons approaching a stressor.
rather than lingering effects from the day before. However, additional analyses revealed that the days following the
Negative mood. Anger, depressed mood, and anxiety were bar examination did not differentiate in important or dramatic
measured using items from the Profile of Mood States (Lorr & ways from the data reported below.
McNair, 1971). For each mood, at least three high-loading items Weekend. We have found closeness to be systematically
from a factor analysis conducted by Lorr and McNair (1971) were higher on weekends than on weekdays, and we therefore adjusted
used. Anger and anxiety consisted of three items, and depressed for the effects of the weekend. We represented weekend with a
mood consisted of four items. For each of these items, participants variable that was coded 1 for Saturday and Sunday and 0 for
rated how they felt right now on a 5-point scale ranging from 1 Monday through Friday.
(not at all) to 5 (extremely). The scores were rescaled to a 0 4 Daily stressors. As discussed above, the support-seeking
interval, and a mean for each mood was obtained by averaging the triage model has suggested that negative effects of support receipt
rescaled values of the relevant items. Anger, depressed mood, and are due to the fact that stress and support co-occur (stress-eliciting
anxiety were highly related (between-person reliability estimate support), and therefore when distress increases after support, it is
.60; within-person reliability estimate .624) and were therefore not because of the support, but because of the stressor that
averaged to form a single negative mood scale that was then prompted that support. This same potential confound exists for
centered on the respective overall means (partners: M 0.31 relationship closeness and stressors, given that stressful life events
before centering, SD 0.52; examinees: M 0.72 before center- are associated with declines in relationship satisfaction (Tesser &
ing, SD 0.74). It can be seen that on average, examinees as Beach, 1998) and an increase in social support. To rule out these
compared with partners reported more than twice as much negative alternative hypotheses, a count of participants daily stressors was
mood. Negative mood was measured twice each day, once in the included in the model. Each day, participants were asked to indi-
morning and once at night. Therefore, we adjusted for morning cate whether any of 21 possible stressful events had occurred and
negative mood, and again our outcome variable was residualized to indicate any stressful event that occurred that did not correspond
change in negative mood, again strengthening the claim that any to one on the list. The number of events indicated was summed and
change in negative mood was due to events of the day in question
rather than lingering effects from the day before.5
3
In an effort to lessen the burden of taking the daily diaries, we
shortened a closeness scale used in previous studies by one item. We chose
Measures: Predictor Variables to eliminate an item that asked how connected one felt to their partner that
Support provision and receipt. Participants provision of emo- day because it appeared redundant with the emotional closeness item
(Cronbachs .92). By doing so, we lowered the alpha of the scale, but
tional support to their partner and receipt of emotional support
this is to be expected because Cronbachs alpha underestimates reliability
from their partner was assessed in the evening each day. Each
when items tap different aspects of a construct (Raykov, 1998).
measure consisted of a single item in which participants reported 4
The between-person reliability is interpreted as the between-person
whether they had provided emotional support to their partner and, reliability of the average of the measures taken on the same day; the
separately, whether they had received emotional support from their within-person reliability is interpreted as the reliability of change within
partner. Specifically, participants were asked to indicate, by cir- person throughout the study (see Cranford et al., 2006).
cling yes or no, whether they had received (provided) any help 5
Positive mood was also measured using the Profile of Mood States
from (to) their partner for a worry, problem, or difficulty in the (Lorr & McNair, 1971), but as positive and negative mood have been
past 24 hr. Examples of support such as listening and comforting shown to operate independently of each other we did not condense them
were given to clarify the question. Support receipt was coded 1 and into one scale. We did conduct separate analyses looking at positive mood
a lack of receipt was coded 0; similarly, support provision was and the effect of receipt of support and found that in both partners and
coded 1 and a lack of provision was coded 0.6 examinees positive mood was not negatively affected by support receipt
and was positively affected by giving support. Gleason, Iida, et al. (2003),
on the other hand, found that positive and negative mood behaved similarly
Covariates (support-only days resulted in an increase in negative mood and a decrease
in positive mood).
Time. Temporal effects of being in the study were adjusted for 6
Practical support was also measured and behaved very similarly to
by including time in the study as a predictor of both outcomes. It emotional support (smaller beta coefficients, but in the same direction) in
has been shown that the first 3 days of diary studies generally have both bar examinees and their partners; however, when both emotional and
elevated levels of negative reports, but not of positive reports practical support were entered into the models, practical support no longer
(Gleason, Bolger, & Shrout, 2003). Given this potentially bias- had any explanatory power, but the results for emotional support remained.
inducing tendency in our analyses, we eliminated the first 3 days Given this pattern, we report only the effects of emotional support.
830 GLEASON, IIDA, SHROUT, AND BOLGER

centered on the grand mean (partner M 1.61 before centering, b5nGik b6niRik b7nGik Rik eijk]
SD 1.55; examinee M 1.76 before centering, SD 1.65).
Gender. Gender (coded .5 for men and .5 for women) was Cijk b0ci b1cYijk1 b2cDik b3cWik b4cSik
originally included as a covariate, but because it did not affect the
b5cGik b6ciRik b7c(Gik Rik) eijk]. (1)
variables of interest (receiving and providing support) and to
simplify the model presented, it was not included in the analyses The dependent variable, Yijk, is the outcome for participant i for
reported below. outcome j (when j 1 it is negative mood; when j 2 it is
closeness) on day k. Thus, there were two records for each day
Moderating Variables within participant, so the maximum number of records that a
participant contributed was 62. When the outcome is negative
Both potential moderating variables were measured in the back- mood, Nijk 1 and Cijk 0, and the first part of the model is
ground questionnaire, which both members of the couple com- selected and all of the b coefficients have the subscript n. When the
pleted approximately 3 weeks before starting the diary portion of outcome is closeness, Nijk 0 and Cijk 1, and the second part
the study (see above for more details about the background ques-
of the model is selected and each of the b coefficients have a
tionnaire administration).
subscript c. Yijk 1 is morning negative mood for individual i when
Relationship satisfaction. Overall relationship satisfaction was
j is equal to 1; Yijk 1 is yesterdays closeness for the same
measured with one item taken from the Dyadic Adjustment Scale
individual i when j is equal to 2; Dik is the number of days in the
(Spanier, 1976), on which 0 extremely unhappy, 3 happy, and
study; Wik indicates whether it is a weekend day or not; Sik adjusts
6 perfectly happy. Relationship satisfaction was generally high
for the number of stressors experienced; Gik is the individuals
among these couples (examinees: M 4.45, SD 1.04; partners:
report of providing (giving) support; Rik is the individuals report
M 4.56, SD 1.03).
of receiving support; Gik Rik is the interaction term for providing
Self-esteem. Self-esteem was measured using the Rosenberg
and receiving support, and the residual components are represented
Self-Esteem scale (Rosenberg, 1965), a 7-item Likert scale ranging
by eijk. The coefficient b0ni is the regression intercept for negative
from 0 (low self-esteem) to 4 (high self-esteem; examinees: M
mood for individual i and represents negative mood on the first
3.17, SD 0.59; partners: M 3.16, SD 0.60). Alpha reliability
weekday of the study when the individual has neither given nor
was .86 for examinees and .88 for partners.
received support and all other variables are at their projected
average level (as morning mood and daily stressors are grand mean
Analytic Approach centered). The coefficient b0ci is the regression intercept for close-
ness for individual i and represents closeness on the first weekday
The goal of the current analysis was to examine the effects of
of the study when the individual has neither given nor received
receiving support from and providing support to ones partner on
support and all other variables are at their projected average level
both an individuals evaluation of the degree of closeness in the
(as yesterdays closeness and daily stressors are grand mean cen-
relationship and simultaneously on an individuals level of nega-
tered).
tive mood. We used a multilevel statistical model to investigate
As Bolger and Shrout (2007) discussed, the mixed-model ap-
these relationships separately for partners (less stressed) and ex-
proach can be specified to acknowledge that the residuals on
aminees (highly stressed). The models had two levels: a within-
adjacent days are likely to be correlated, and we used this speci-
individual level (over time) and a between-individuals level. The
fication in the analysis we report here. This specification allowed
model also took into account the fact that outcomes, negative
us to account for dependency between outcomes in individuals and
mood and closeness, were clustered within individuals.7 Using the
within individuals across time.
multivariate approach described by Raudenbush and Bryk (2002),
The between-individual level of the analysis allows us to model
we included both closeness and negative mood in a single multi-
possible individual differences in the coefficients specified in
level analysis. The multivariate approach allowed us to estimate
Equation 1. We fit a model that considered intercepts for both
the correlation between the random effects for negative mood and
closeness and negative mood to be random (i.e., to vary across
closeness and to examine the frequencies of participants showing
persons) and the effect of support receipt on each of the two
a moderated pattern (see Figure 1, Model 1) and a differential
effects pattern (see Figure 1, Model 2). All analyses were con- outcomes. The formal specification of these models involves the
ducted using the MIXED procedure in SAS (SAS Institute, 2003). inclusion of random effects in the Level 2 equation. These have a
The within-individual level of the analysis allowed each indi- mean of zero but variance that is assumed to be nonzero. For
viduals relationship closeness and negative mood to be modeled example, the between-individuals level of the model for the inter-
as a function of receipt of support. We predicted a given days cepts involves the sum of overall means () and random effects
closeness and negative mood for a particular individual; we ad- (u). Our analytic model also allowed the random effects for the
justed for either yesterdays closeness or same-day morning neg- intercepts and the support receipt effects to be correlated across
ative mood, respectively; number of days in the study; and week-
end effects. Given that support transactions may be more likely to 7
Kenny, Kashy, and Bolger (1998) provided a general description of
take place on days when an individual experiences stressful events, multilevel statistical models. Raudenbush and Bryk (2002) showed that
a count of daily stressors was included to adjust for the effects of these models can be influenced by both within- and between-individuals
stressful events as a third variable. The equation was as follows: variation. When within-person and between-person effects are predicted to
be the same, the multilevel analysis that combines the effects is recom-
Yijk Nijk b0ni b1nYijk1 b2nDik b3nWik b4nSik mended.
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 831

both effect type and outcome variable. For those interested in the Table 1
details of this analysis, the syntax used is available from Marci Multilevel Analysis Results Relating Daily Support to Negative
E. J. Gleason. The Level 2 equations were Mood and Closeness for Partners and Examinees: Fixed Effects

b0ni 0n u0ni Partners Examinees


(n 290) (n 293)
b0ci 0c u0ci
Variable SE SE
b6ni 6n u6ni
Negative mood
b6ci 6c u6ci. (2) Intercept 0.343** 0.016 0.589** 0.021
Day 10 0.015* 0.005 0.102** 0.007
Weekend 0.016 0.009 0.016 0.011
In addition, we tested the moderation hypothesis in two separate Daily stressors 0.027** 0.003 0.064** 0.004
multivariate, multilevel analyses. The same Level 1 equation de- Morning negative mood 0.428** 0.012 0.472** 0.011
scribed in Equation 1 was used for each analysis. The Level 2 Receiving emotional support 0.075* 0.021 0.037* 0.018
equations were modified when testing for moderation to include Giving emotional support 0.001 0.014 0.048* 0.021
Receiving Emotional
the moderators (self-esteem or relationship satisfaction), resulting
Support Giving
in an additional predictor in each of the four equations. We did not Emotional Support 0.093* 0.023 0.080* 0.027
alter the specifications of the random effects for the moderation Closeness
tests. Intercept 1.989** 0.037 1.971** 0.046
Day 10 0.057** 0.010 0.048** 0.001
Weekend 0.113** 0.019 0.135** 0.019
Results Daily stressors 0.043** 0.007 0.065** 0.007
Yesterdays closeness 0.318** 0.011 0.140** 0.011
Support Patterns Receiving emotional support 0.248** 0.039 0.411** 0.034
Giving emotional support 0.245** 0.028 0.319** 0.037
Receiving Emotional
Examinees reported receiving support on 50% of days, whereas Support Giving
partners reported receiving support on only 40% of days. Exam- Emotional Support 0.080 0.046 0.011 0.046
inees reported giving support on 37% of days, whereas partners
reported giving support on 53% of days. Examinees support

p .10. *
p .05. **
p .001.
receipt increased over time (r .09, p .01), and their reports of
giving support decreased (r .03, p .01). The opposite
Figure 3 shows the results for closeness. Support receipts
pattern is observed for partners, that is, partners received less and
positive effects on closeness were evident despite its also being
provided more support as the bar exam approached (receiving, r
associated with an increase in negative mood. Although there was
.03, p .05; giving, r .11, p .01).
a marginal interaction between receipt and provision on closeness
for partners, suggesting that supportive equity days were particu-
Fixed Effects larly positive for partners, the interaction does not diminish the
beneficial effects of support receipt on closeness.
Table 1 presents the fixed-effect results for both outcomes for Partners effects of receiving and giving support on relationship
partners and examinees. Only the variables of interest are reported closeness did not differ (difference between estimates 0.002),
here. The main effect of support receipt was significant for both t(289) 0.10, ns. However, examinees effect of receiving sup-
negative mood, partners: b6n 0.075, t(289) 3.54, p .001; port was greater than the effect of giving support (difference
examinees: b6n 0.037, t(292) 2.04, p .05, and closeness, between estimates 0.09), t(292) 2.09, p .05. Days on which
partners: b6c 0.248, t(289) 6.37, p .001; examinees: b6c support was received and not given were significantly more neg-
0.411, t(292) 12.19, p .001. The main effect of giving support on ative when compared with the other three types of days for both
negative mood was significant for examinees, but not for partners, partners (difference between estimates 0.08), t(289) 4.45, p
partners: b5n 0.001, t(289) 0.10, ns; examinees: b5n .001, and examinees (difference between estimates 0.08),
0.048, t(292) 2.24, p .05. The main effect for giving support t(292) 5.19, p .001.
was significant for both partners and examinees on closeness, part-
ners: b5c 0.245, t(289) 8.73, p .001; examinees: b5c 0.319,
Random Effects
t(292) 8.63, p .001. For negative mood, these effects have to be
interpreted in the context of a significant interaction between receipt The random effects covariance matrix for both partners and
and provision (see Table 1). When one takes the interaction into examinees is displayed in Table 2. The model generated random
account, the current findings replicate those of Gleason, Iida, et al. effects for the intercepts of closeness and negative mood both
(2003), in which it was found that supportive equity days (days in between and within level. The within-level random effects provide
which support is both received and provided) are associated with the information about what is occurring in individuals lives that
lowest levels of negative mood and that receipt-only days are asso- affects their levels of negative mood and closeness that was not
ciated with the highest levels of negative mood (see Figure 2). As the captured by our models. As can be seen, the variances for both the
figure shows, the receipt of support is detrimental to negative mood, intercept for negative mood and closeness are significant, suggest-
but only on days in which the recipient of support did not also provide ing that the model does not account for all the variation in these
support to his or her partner. variables. In addition, the evidence for a negative covariance
832 GLEASON, IIDA, SHROUT, AND BOLGER

0.65

Examinees
0.60

0.55 No Giving
Negative Mood

Giving
0.50

0.45

0.40
Partners

0.35

0.30
No Receiving Receiving

Figure 2. The effects of support receipt and provision on evening negative mood for both partners and
examinees.

between these intercepts suggests that whatever is causing that A representation of these negative correlations of the slopes can
within-level daily variation affects closeness and negative mood in be found in Figure 4. Each point on the scatterplots represents the
opposite ways. On a given day, when negative mood is increased, estimated random effect of receipt on negative mood (x-axis) and
closeness is decreased and vice versa. on closeness (y-axis) for a single individualin other words, each
The between-person random effects for the intercepts of close- point represents how support receipt typically affects a particular
ness and negative mood provide information about how negative individuals negative mood and feelings of closeness. As can be
mood and closeness behave across individuals. By generating seen, there are individuals in three of the four quadrants of the
random effects, we were able to obtain estimates of each individ- scatterplots. Most individuals fall in the upper right quadrant
uals intercepts for negative mood and closeness. If the random (examinees 209 individuals; partners 245 individuals); mem-
effects are positively correlated, it suggests that individuals who bers in this quadrant experience something akin to the fixed
have generally higher levels of closeness also have higher levels of effects: an increase in both closeness and negative mood. How-
negative mood; if the correlation between them is negative, it ever, a sizable portion of individuals also fall into the upper left
suggests that individuals who have generally higher levels of quadrant (examinees 80; partners 40); support receipt is only
closeness have lower levels of negative mood and vice versa. positive for members of this quadrant: It decreases negative mood
There are also between-person random effects for the slopes (the and increases closeness. Finally, a few individuals fall into the
effect of receipt on both negative mood and closeness for each lower right quadrant (examinees 4; partners 5); support
individual), and their correlation will give us information as to receipt is only negative for members of this quadrant: It increases
whether these effects are systematically linked across individuals. negative mood and decreases closeness.
Using the random effects variances and covariances, we calcu- Selecting individuals at both the positive and the negative ends
lated the correlations between the intercepts for negative mood and of the scatterplots and plotting their individual data allows us to
closeness and the effects of receipt on negative mood and close- see how support differentially affects different people. Figure 5
ness (the slopes) at the within level. As can be seen in Table 2, displays the data for 2 partners and 2 examinees. The top two
partners covariance between the random effect for the intercept of graphs display data for a partner and examinee for whom the
negative mood (0n .035) and the intercept for closeness (0c receipt of support is negative. Notice that on days on which they
.212) is 0.027. This results in a correlation of .31 ( p .05) for receive support, their closeness ratings are low and their negative
partners, and similarly the correlation for examinees is .25 ( p mood ratings are high. The bottom two graphs show a strikingly
.05). These correlations suggest that people who are higher in different patternthese are of individuals who are positively af-
negative mood tend to be lower in closeness. The correlation fected by the receipt of support. Notice that on days on which they
between the random effects of support receipt on both outcomes is receive support, their closeness ratings are high and their negative
.36 ( p .05) for partners and .31 ( p .05) for examinees, mood ratings are low. These extreme individuals are good ex-
suggesting systematic differences across individuals. amples of how people differentially react to the receipt of support,
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 833

Examinees
3.0

Relationship Closeness 2.5

2.0
No Giving

Giving
1.5

1.0
No Receiving Receiving

Partners
3.0

2.5
Relationship Closeness

2.0

No Giving

1.5 Giving

1.0
No Receiving Receiving

Figure 3. The effects of support receipt and provision on relationship closeness for both partners and
examinees.

but it is important to note that most individuals fall in the middle mood and closeness for both examinees and partners. Self-esteem
of the distribution and experience something more akin to the fixed did not moderate the receipt of support for examinees on either
effects results when they receive support: an increase in negative negative mood (bn .024, SE .027, ns) or closeness (bc
mood and an increase in feelings of closeness. .056, SE .049, ns). It did moderate the effect of support receipt
on negative mood for partners (bn 0.067, SE .026), t(289)
2.58, p .05, but not the effect of support receipt on closeness
Moderation
for partners (bc .005, SE .045, ns). These results suggest that
The moderators included in the analyses did not explain why when one is not approaching a large stressor, the negative effects
individuals react differently to support receipt. Relationship satis- of support receipt may be tempered for those with above-average
faction failed to moderate the effects of support receipt on negative self-esteem (self-esteem was group-mean centered). However,
834 GLEASON, IIDA, SHROUT, AND BOLGER

Table 2 individuals who experience a larger increase in closeness when


Multilevel Analysis Results Relating Daily Support to Negative they receive support experience less of an increase in negative
Mood and Closeness for Partners and Examinees: Random mood. Although this does not eliminate support receipts average
Effects effect on outcomes across people, it does indicate that the duality
observed for an average individual is limited. The majority of
Partners Examinees participants were in the middle of the spectrum of support reac-
(n 290) (n 293)
tivity and reported experiencing increases in both outcomes fol-
Level and variable SE SE lowing support receipt, whereas those on the ends of the spectrum
either benefited from support or suffered from support. It seems
Level 1
that support not only differentially affects closeness and negative
Variance of negative mood (NM) 0.128** 0.002 0.176** 0.003
Variance of closeness (CL) 0.538** 0.010 0.508** 0.009 mood, but also operates differently across individuals.
Covariance of NM and CL 0.054** 0.003 0.057** 0.004 Several possible explanations for the overall finding that support
Level 2 increases both negative mood and relationship closeness seem likely,
Variances
including characteristics of the recipient, characteristics of the pro-
NM 0.035** 0.004 0.074** 0.009
CL 0.212** 0.023 0.444** 0.045 vider, and characteristics of the relationship. The characteristics of the
Receipt Negative Mood recipients that we tested here, self-esteem and relationship satisfac-
(RNM) 0.030** 0.006 0.021** 0.005 tion, did not explain our pattern of findings. However, there are many
Receipt Closeness (RCL) 0.053** 0.016 0.097** 0.020 other individual difference constructs, such as attachment style, that
Covariances
NMCL 0.027** 0.007 0.045** 0.014 are plausible moderators but were not included in this study. An
NMRNM 0.007* 0.004 0.007* 0.005 intriguing candidate for moderation, which has intuitive ties to Nadler
NMRCL 0.010 0.010 0.019* 0.009 and colleagues work on self-esteem (Fisher et al, 1982; Nadler, 1987;
CLRNM 0.004 0.008 0.027** 0.012 Nadler & Fisher, 1976), is perceived respect from ones partner or the
CLRCL 0.052** 0.015 0.128** 0.025
RNMRCL 0.014* 0.007 0.014* 0.008 extent to which one feels respected and self-efficacious in ones
relationship. This construct may be more directly implicated than
Note. Significance tests of Level 1 effects are constricted ratio Wald self-esteem when considering support exchanges between partners.
tests; significance tests of Level 2 variances are chi-squares (df 4); For instance, work on self-efficacy has demonstrated that judging
significance tests of Level 2 covariances are chi-squares (df 1).
*
p .05. ** p .001. oneself as being inefficacious can impair coping and goal achieve-
ment (Bandura, 1982), and the receipt of support may lead some
individuals to doubt their ability to accomplish goals on their own.
Although we were not able to test this idea in the current study, it is
given that the effect of receipt of support on negative mood is bn bolstered by the current findings that providing support is beneficial,
.075 and there are no participants who are more than 0.90 units particularly when one has received support. Demonstrating ones
above the mean in self-esteem (the mean self-esteem for partners efficacy through the provision of support may allow one to accept
was 3.16 and the maximum score was 4.0), this moderation makes support from ones partner without experiencing efficacy declines.
receipt of support less detrimental to negative mood, but not on Should this be the case, we would then want to determine why
average positive, for those high in self-esteem. receiving support signals lack of efficacy for some individuals but not
others.
Discussion A second explanation for the differential effect documented in
The results support Model 2 (differential effects) from Figure 1 this study could be characteristics of the provider. Characteristics
in that receiving support simultaneously increased relationship such as being a high self-monitor might result in an individuals
closeness and negative mood. However, this was only true on days being a particularly skilled support provider (Flynn, Reagans,
when support was not provided (22% of days for partners and 21% Amanatullah, & Ames, 2006), one whose support may be less
of days for examinees). On days when support was both received detrimental to recipients moods and feelings of efficacy. Work by
and providedsupportive equity days (31% of days for partners, Lakey, Lutz, and Scoboria (2004; Lakey & Scoboria, 2005) on
29% of days for examinees)support receipt increased relation- perceived support suggests that the benefits of perceived social
ship closeness and decreased negative mood. This was true for support may be derived not only through personality characteris-
individuals who were approaching a major stressor (examinees) tics of the perceiver, but also through relationship factors such as
and for their less stressed partners. The varying stress level did not the perceived similarity of the provider to the recipient. Perhaps
substantially affect the influence of support receipt or provision on the same is true in actual support transactionsthe more a recip-
relationship closeness and negative mood. ient feels similar to a provider, the more positive the support given
Although this pattern is evident on average, it is not the whole by that provider would be. Other relationship characteristics, such
story. Evidence from the random effects analysis suggests that as the match or mismatch of communication styles (see Swann,
support also differentially affects individuals, and this is in line Rentfrow, & Gosling, 2003), could also lead to the effects docu-
with Model 1 (individual differences) from Figure 1. Namely, the mented here. Although we cannot offer definitive evidence sup-
negative correlations between the random effects of receipt on porting a particular explanation for these patterns, given that
negative mood and on closeness suggest that individuals who support is an exchange between at least two individuals it seems
experience a larger increase in negative mood when they receive likely that at least part of the explanation for these effects will
support experience less of an increase in closeness; conversely, come from dyad-level variables or processes.
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 835

Partners
1.4

1.2

Closeness X Receipt of Support


1.0

0.8

0.6

0.4

0.2

0.0

-0.2

-0.4
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Negative Mood X Receipt of Support

Examinees
1.4

1.2
Closeness X Receipt of Support

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

-0.4
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Negative Mood X Receipt of Support

Figure 4. Scatterplots of the random effects of Receipt Closeness and Receipt Negative Mood for each
partner and examinee. Individuals whose point lies in the upper left-hand quadrant are those for whom support
receipt decreases negative mood and increases closeness; those in the upper right-hand quadrant are those for
whom support receipt increases negative mood and also increases closeness; and those in the lower right-hand
quadrant are those for whom support receipt increases negative mood and decreases closeness.

Also worth noting was the benefit of support provision for uals use to prove their worth in their social group (Buunk &
providers, which, regardless of support receipt, improved mood Schaufeli, 1999). It may also be that giving support boosts self-
(for examinees) and relationship closeness. Considering provi- esteem by making one feel more competent and needed (Fisher et
sions beneficial qualities, it is important that social support re- al., 1982). Provisions benefits need to be explored more thor-
searchers include it in their studies to understand why giving oughly, particularly in the context of close relationships.
appears to be better than receiving. Reciprocity research would These results speak directly to two recently proposed theories in
suggest that giving is positive because it repays our debts or puts the close relationships literature: Cutrona et al.s (2005) relation-
others in our debt (Uehara, 1995) and that it is a tool that individ- ship enhancement model of social support and Reis et al.s (2004)
836 GLEASON, IIDA, SHROUT, AND BOLGER

Partner Examinee
4
Closeness 3
2
1
0
Yes
Receipt of
Support
No
4
Negative 3
Mood 2
1
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
4
Closeness 3
2
1
0
Yes
Receipt of
Support
No
4
Negative 3
Mood 2
1
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Day in Study Day in Study

Figure 5. Top: Partner and examinee for whom support decreases closeness and increases negative mood.
Bottom: Partner and examinee for whom support increases closeness and decreases negative mood

perceived partner responsiveness model. The relationship enhance- receipt (see Antonucci & Akiyama, 1987; Cutrona, 1996). It is im-
ment model of social support suggests that consistent supportive portant to note that we did find some gender effectsfor instance,
responses can lead to higher perceived partner support, which men tended to experience more negative mood and to be less nega-
leads to greater trust, relationship satisfaction, and ultimately bet- tively affected by troublesome events but not on the variables of
ter health. Although this model is widely supported by research on interest (support receipt and provision). Perhaps gender differences in
perceived support availability, it has been less clear how to rec- support processes would have emerged if we had investigated the
oncile findings regarding the negativity of social support receipt amount of support requested or received by men and women.
with these ideas. The current research suggests that supportive
acts, despite causing personal distress for some individuals, en- Limitations
hance relationship closeness. Perhaps it is this positive effect of
support that leads to the positive effects of perceived support and This study had several limitations, some of which have been
ultimately to relationship satisfaction and health. discussed above. The sample was not randomly chosen and con-
Reis et al. (2004) also referred to the benefits of perceived avail- sisted of well-educated individuals. It is possible that this pattern
ability of support in their discussion of perceived partner responsive- of results would not be present in a less privileged population. In
ness and also noted that the research on support receipt itself has future studies, more diverse samples should be sought to determine
questioned its benefit. They suggested that the field needs to examine the generalizability of these findings.
how the effects of social support differentially affect outcomes and In addition, given that this was a nonexperimental study we
individuals. The current research does both of these and suggests that were unable to completely adjust for the level of stress participants
the benefits of support are largely due to relationship enhancement experienced and to definitively establish that support increased
(i.e., increased closeness), although it is important to note that this negative mood instead of just co-occurring with it. We took steps
association itself varies by individual. Actual instances of support, to limit this concern: We included a count of stressful events in all
despite some negative side effects, may be one of the more impor- analyses and adjusted for morning negative mood. In addition, by
tant ways that partners establish responsiveness. demonstrating the effects in both highly stressed individuals (ex-
The lack of gender effects on the variables of interest in both aminees) and less stressed individuals (partners), we feel confident
studies may seem surprising, but it is consistent with the support that support receipt can have these mixed effects regardless of
literature that has found similarities in support processes for men and overall stress level. Furthermore, the simulation study by Seidman
women (Neff & Karney, 2004; Porter et al., 2000). However, a few et al. (2006) concluded that the negative effects associated with
studies have shown that men and women react differently to support support receipt in naturalistic studies could not reasonably be
DUAL EFFECTS OF SUPPORT ON PSYCHOLOGICAL OUTCOMES 837

caused by co-occurring negative eventsthe parameter values can be included in studies, of course, but the intensive longitudinal
needed for such an association were unrealistic. Finally, a recent approach taken here allows one to quantify the extent of hetero-
experimental study demonstrated the negative effects of support geneity without having identified moderators a priori. Thus, these
receipt on mood (Bolger & Amarel, 2007). Given these previous intensive designs have the possibility to demonstrate that many of
findings and the precautions we took to minimize this concern, we the findings in the field reflect average effects, averages that can
feel confident that support is causally linked to negative mood in obscure important, consequential variability.
this study. In the future, however, we hope to demonstrate this
pattern of findings in an experimental setting. References
Our analyses of support receipt and provision included only
emotional support. Although we also asked about practical and Abelson, R. P. (1968). Simulation of social behavior. In G. Lindzey & E.
instrumental support, we found that emotional support was the Aronson (Eds.), Handbook of social psychology (Vol. 2, pp. 274 356).
driving force behind our findings and therefore included only Reading, MA: Addison-Wesley.
emotional support in our models. However, research by other Acitelli, L. K., & Antonucci, T. C. (1994). Gender differences in the link
between marital support and satisfaction in older couples. Journal of
individuals has shown that emotional support can be further ex-
Personality and Social Psychology, 67, 688 698.
panded into such things as esteem support, companionship, and Antonucci, T. C., & Akiyama, H. (1987). An examination of sex difference
caring (Kang & Rafaeli, 2007). Perhaps such finer distinctions in in social support among older men and women. Sex Roles, 17, 737749.
the type of support received would shed some light on the current Bandura, A. (1982). The assessment and predictive generality of self-
findings. For instance, it may be that esteem support is both percepts of efficacy. Journal of Behavior Therapy and Experimental
common and unhelpful, whereas companionship is less common Psychiatry, 13, 195199.
but beneficial. Also, we had participants indicate whether they had Barrera, M. (1986). Distinctions between social support concepts, mea-
received support or not, but did not obtain information on the sures, and models. American Journal of Community Psychology, 14,
amount of support they received. Future studies should include a 413 445.
more detailed collection of support that may shed light on both Barrera, M., Jr. (1981). Social support in the adjustment of pregnant
adolescents: Assessment issues. In B. H. Gottlieb (Ed.), Social networks
differential effects of support receipt and individual differences in
and social support (pp. 69 96). Beverly Hills, CA: Sage.
the effects of support receipt. Bolger, N., & Amarel, D. (2007). Effects of support visibility on adjust-
Finally, although we did include theoretically important con- ment to stress: Experimental evidence. Journal of Personality and Social
structs as moderators, there are several others that would be of Psychology, 92, 458 475.
interest that were not included. These include not only character- Bolger, N., & Shrout, P. E. (2007). Accounting for statistical dependency
istics of the recipient (i.e., perceived competence), but also of the in longitudinal data on dyads. In T. D. Little, J. A. Bovaird, & N. A. Card
provider (i.e., empathy or responsiveness) and of the relationship (Eds.), Modeling ecological and contextual effects in longitudinal stud-
itself (i.e., communication style). Future studies would do well to ies of human development (pp. 285298). Mahwah, NJ: Erlbaum.
examine such possible moderators. Bolger, N., Zuckerman, A., & Kessler, R. C. (2000). Invisible support and
adjustment to stress. Journal of Personality and Social Psychology, 79,
953961.
Concluding Thoughts and Implications Bradbury, T. N., & Fincham, F. D. (1989). Behavior and satisfaction in
marriage: Prospective mediating processes. In C. Hendrick (Ed.), Close
The effects of support receipt in this study were consistent and relationships (pp. 119 143). Thousand Oaks, CA: Sage.
compelling. Although actual support receipt has been linked to neg- Buunk, B. P., & Schaufeli, W. B. (1999). Reciprocity in interpersonal
ative outcomes (Barrera, 1981; Bolger et al., 2000; Gleason, Iida, et relationships: An evolutionary perspective on its importance for health
al., 2003), this study demonstrated that this does not appear to be true and well-being. European Review of Social Psychology, 10, 259 291.
for relationship closeness. These findings highlight how a single act Cohen, S. (2004). Social relationships and health. American Psychologist,
(receiving support) can make one feel better in one domain (relation- 59, 676 684.
ship closeness) but worse in another (personal mood). Perhaps it is Collins, N. L., & Feeney, B. C. (2000). A safe haven: An attachment theory
this duality that allows social support to be considered positive by perspective on support seeking and caregiving in intimate relationships.
Journal of Personality and Social Psychology, 78, 10531073.
laypersons even though psychologists have documented that it can be
Cranford, J. A., Shrout, P. E., Iida, M., Rafaeli, E., Yip, T., & Bolger, N.
ineffective and even detrimental. Furthermore, evidence of random (2006). A procedure for evaluating sensitivity to within-person change:
effects showed that this duality itself showed substantial between- Can mood measures in diary studies detect change reliably? Personality
individual heterogeneity, which implies that models of support pro- and Social Psychology Bulletin, 32, 917929.
cesses need to accord a more central role to heterogeneity than they Cutrona, C. E. (1996). Social support in couples: Marriage as a resource
have done heretofore. At this time, we have not identified moderators in times of stress. Thousand Oaks, CA: Sage.
of this heterogeneity and therefore cannot explain why people react Cutrona, C. E., Russell, D. W., & Gardner, K. A. (2005). The relationship
differently to support, but understanding that individuals react differ- enhancement model of social support. In T. A. Revenson, K. Kayser, &
ently is an important step toward understanding social support pro- G. Bodenmann (Eds.), Couples coping with stress: Emerging perspec-
cesses in close relationships. tives on dyadic coping (pp. 7395). Washington, DC: American Psy-
chological Association.
In conclusion, it is worth considering that the implications of our
Edwards, A. C., Nazroo, J. Y., & Brown, G. W. (1998). Gender differences
heterogeneity findings may not be limited to the social support in marital support following a shared life event. Social Science &
literature. Although many of the constructs studied in psychology Medicine, 46, 10771085.
are likely to have differential effects across individuals, dyads, and Feeney, B. C., & Collins, N. L. (2001). Predictors of caregiving in adult
social contexts, rarely are studies designed in ways that allow such intimate relationships: An attachment theoretical perspective. Journal of
heterogeneity to be reliably distinguished. Potential moderators Personality and Social Psychology, 80, 972994.
838 GLEASON, IIDA, SHROUT, AND BOLGER

Feeney, B. C., & Collins, N. L. (2003). Motivations for caregiving in adult friend: Effect of single or multiple act aid as a function of donor and task
intimate relationships: Influences on caregiving behavior and relationship characteristics. Journal of Personality and Social Psychology, 44, 310 321.
functioning. Personality and Social Psychology Bulletin, 29, 950 968. Neff, L. A., & Karney, B. R. (2004). The dynamic structure of relationship
Fisher, J. D., Nadler, A., & Whitcher-Alagna, S. (1982). Recipient reac- perceptions: Differential importance as a strategy of relationship main-
tions to aid. Psychological Bulletin, 91, 2754. tenance. Personality and Social Psychology Bulletin, 29, 14331446.
Flynn, F. J., Reagans, R. E., Amanatullah, E. T., & Ames, D. R. (2006). Porter, L. S., Marco, C. A., Schwartz, J. E., Neale, J. M., Shiffman, S., &
Helping ones way to the top: Self-monitors achieve status by helping Stone, A. A. (2000). Gender differences in coping: A comparison of trait
others and knowing who helps whom. Journal of Personality and Social and momentary assessments. Journal of Social & Clinical Psychology,
Psychology, 91, 11231137. 19, 480 498.
Gable, S. L., Gonzaga, G. C., & Strachman, A. (2006). Will you be there Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models:
for me when things go right? Supportive responses to positive event Applications and data analysis methods. Thousand Oaks, CA: Sage.
disclosures. Journal of Personality and Social Psychology, 91, 904 917. Raykov, T. (1998). Coefficient alpha and composite reliability with inter-
Gable, S. L., Reis, H. T., & Downey, G. (2003). He said, she said: A related nonhomogeneous items. Applied Psychological Measurement,
quasi-signal detection analysis of daily interactions between close rela- 22, 375385.
tionship partners. Psychological Science, 14, 100 105. Reis, H. T., Clark, M. S., & Holmes, J. G. (2004). Perceived partner
Gleason, M. E. J. (2005). Is receiving support a mixed blessing? Evidence responsiveness as an organizing construct in the study of intimacy and
for dual effects of support on psychological outcomes (Doctoral disser- closeness. In D. J. Mashek & A. P. Aaron (Eds.), Handbook of closeness
tation, New York University, 2005). Dissertation Abstracts Interna- and intimacy (pp. 201225). Mahwah, NJ: Erlbaum.
tional, 65, 4896. Reis, H. T., & Patrick, B. C. (1996). Attachment and intimacy: Component
Gleason, M. E. J., Bolger, N., & Shrout, P. E. (2003, February). Effects of processes. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology:
study design on reports of mood: Understanding differences between Handbook of basic principles (pp. 523563). New York: Guilford Press.
cross-sectional, panel, and diary designs. Poster session presented at the Roberto, K. A., & Scott, J. P. (1986). Equity considerations in the friend-
annual meeting of the Society for Personality and Social Psychology, ships of older adults. Journal of Gerontology, 41, 241247.
Universal City, CA. Rosenberg, M. (1965). Society and the adolescent child. Princeton, NJ:
Princeton University Press.
Gleason, M. E. J., Iida, M., Bolger, N., & Shrout, P. E. (2003). Daily
Sanderson, C. A., & Cantor, N. (1997). Creating satisfaction in steady
supportive equity in close relationships. Personality and Social Psychol-
dating relationships: The role of personal goals and situational affor-
ogy Bulletin, 29, 1036 1045.
dances. Journal of Personality and Social Psychology, 73, 1424 1433.
Gottman, J. M. (1979). Detecting cyclicity in social interaction. Psycho-
Sarason, I. G., Sarason, B. R., & Pierce, G. R. (1994). Social support:
logical Bulletin, 86, 338 348.
Global and relationship-based level of analysis. Journal of Social and
Hagedoorn, M., Kuijer, R. G., Buunk, B. P., DeJong, G. M., Wobbes, T.,
Personal Relationships, 11, 295312.
& Sanderman, R. (2000). Marital satisfaction in patients with cancer:
SAS Institute. (2003). The SAS system for Windows (Version 8.02). Cary,
Does support from intimate partners benefit those who need it the most?
NC: Author.
Health Psychology, 19, 274 282.
Seidman, G., Shrout, P. E., & Bolger, N. (2006). Why is enacted social
Kang, N. J., & Rafaeli, E. (2007). Show, dont tell: Evidence for a
support associated with increased distress? Using simulation to test two
hierarchy of support types. Manuscript submitted for publication.
possible sources of spuriousness. Personality and Social Psychology
Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in social Bulletin, 32, 52 65.
psychology. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), Handbook of Shrout, P. E., Herman, C., & Bolger, N. (2006). The costs and benefits of
social psychology (4th ed., pp. 233265). New York: McGraw-Hill. practical and emotional support on adjustment: A daily diary study of
Krause, N. (1997). Received support, anticipated support, social class, and couples experiencing acute stress. Personal Relationships, 13, 115134.
mortality. Research on Aging, 19, 387 422. Spanier, G. B. (1976). Measuring dyadic adjustment: New scales for
Lakey, B., & Lutz, C. J. (1996). Social support and preventive and assessing the quality of marriage and similar dyads. Journal of Marriage
therapeutic interventions. In G. R. Pierce, B. R. Sarason, & I. G. Sarason and Family, 38, 1528.
(Eds.), Handbook of social support and the family (pp. 435 465). New Swann, W. B., Jr., Rentfrow, P. J., & Gosling, S. D. (2003). The precarious
York: Plenum Press. couple effect: Verbally inhibited men critical, disinhibited women bad
Lakey, B., Lutz, C. J., & Scoboria, A. (2004). The information used to judge chemistry. Journal of Personality and Social Psychology, 86, 10951106.
supportiveness depends on whether the judgment reflects the personality of Tesser, A., & Beach, S. R. H. (1998). Life events, relationship quality, and
perceivers, the objective characteristics of targets, or their unique relation- depression: An investigation of judgment discontinuity in vivo. Journal
ships. Journal of Social and Clinical Psychology, 23, 817 835. of Personality and Social Psychology, 74, 36 52.
Lakey, B., & Scoboria, A. (2005). The relative contribution of trait and Thompson, S. C., Medvene, L. J., & Freedman, D. (1995). Caregiving in the
social influences to the links among perceived social support, affect, and close relationships of cardiac patients: Exchange, power, and attributional
self esteem. Journal of Personality, 73, 361388. perspectives on caregiver resentment. Personal Relationships, 2, 125142.
Liang, J., Krause, N. M., & Bennett, J. M. (2001). Social exchange and Uebelacker, L. A., Courtnage, E. S., & Whisman, M. A. (2003). Correlates
well-being: Is giving better than receiving? Psychology and Aging, 16, of depression and marital satisfaction: Perceptions of marital communi-
511523. cation style. Journal of Social and Personal Relationships, 20, 757769.
Lorr, M., & McNair, D. M. (1971). The Profile of Mood States manual. San Uehara, E. S. (1995). Reciprocity reconsidered: Gouldners moral norm of
Diego, CA: Educational & Industrial Testing Service. reciprocity and social support. Journal of Social and Personal Rela-
Nadler, A. (1987). Determinants of help seeking behaviour: The effects of tionships, 12, 482502.
helpers similarity, task centrality and recipients self esteem. European Walster, E., Berscheid, E., & Walster, W. G. (1973). New directions in equity
Journal of Social Psychology, 17, 57 67. research. Journal of Personality and Social Psychology, 25, 151176.
Nadler, A., & Fisher, J. D. (1976). When helping hurts: Effects of donor-
recipient similarity and recipient self-esteem on reactions to aid. Journal Received March 3, 2006
of Personality, 44, 392 409. Revision received August 28, 2007
Nadler, A., Fisher, J. D., & Itzhak, S. B. (1983). With a little help from my Accepted August 30, 2007
Journal of Personality and Social Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 94, No. 5, 839 859 0022-3514/08/$12.00 DOI: 10.1037/0022-3514.94.5.839

Nomina Sunt Omina: On the Inductive Potential of Nouns and Adjectives


in Person Perception

Andrea Carnaghi Anne Maass and Sara Gresta


University of Trieste University of Padova

Mauro Bianchi Mara Cadinu and Luciano Arcuri


University of Jena University of Padova

Six studies (N 491) investigated the inductive potential of nouns versus adjectives in person
perception. In the first 5 studies, targets were either described by an adjective (e.g., Mark is homosexual)
or by the corresponding noun (e.g., Mark is a homosexual) or by both (Study 3). The authors predicted
and found that nouns, more so than adjectives, (a) facilitate descriptor-congruent inferences but inhibit
incongruent inferences (Studies 13), (b) inhibit alternative classifications (Study 4), and (c) imply
essentialism of congruent but not of incongruent preferences (Study 5). This was supported for different
group memberships and inclinations (athletics, arts, religion, sexual preference, drinking behavior, etc.),
languages (Italian and German), and response formats, suggesting that despite the surface similarity of
nouns and adjectives, nouns have a more powerful impact on person perception. Study 6 investigated the
inverse relationship, showing that more essentialist beliefs (in terms of a genetic predisposition rather
than training) lead speakers to use more nouns and fewer adjectives. Possible extensions of G. R. Semin
and K. Fiedlers (1988) linguistic category model and potential applications for language use in
intergroup contexts are discussed.

Keywords: language, essentialism, stereotyping

In his seminal work, The Nature of Prejudice, Allport (1954) As for labels that are associated with categories in general and
argued that social category labels such as Jew, Black, or gay, are social categories in particular, there is little doubt that they provide
unusually potent both as cognitive organizing principles and as perceivers with important and useful information as they indicate
evaluative reference points. In a similar vein, Rothbart and Taylor to which class of objects a given instance belongs (Krueger &
(1992) have argued that social categories (such as Jew) are often Clement, 1994; Putnam, 1975; Rothbart, Davis-Stitt, & Hill,
perceived much like natural kinds (such as birds or fish) rather 1997). Moreover, even when labels, by virtue of their absolute
than cultural constructions. People tend to attribute a deep under- meaning (e.g., nominal categories such as A and B), are totally
lying essence to social categories, analogous to biological species irrelevant to the classified objects (e.g., lines that differ in their
that are genetically defined. Perceiving social categories as quasi- length), they functionally represent some underlying property of
biological concepts has a number of important implications, in- the objects in question (e.g., ordinal categories such as shorter and
cluding the inexorability and exclusiveness of group membership.
longer; Tajfel & Wilkes, 1963).
Indeed, thinking of social categories as natural kinds implies not
There is, indeed, evidence that social labels carry additional
only that membership remains stable over time but that it practi-
information that the object itself does not convey (Bruner, Good-
cally excludes simultaneous membership in other categories con-
now, & Austin, 1956). For instance, observers who have to guess
sidering that each object has only a single essence. Furthermore,
such natural kind categories also tend to have great inductive the weight of visually presented target persons are greatly influ-
potential, allowing inferences about a wealth of related attributes. enced by the labels attached to the targets (such as obese vs. above
average; see Foroni & Rothbart, 2006). Along the same line,
derogatory group labels (e.g., fag, nigger) carry affective informa-
tion that goes beyond the mere description of the group and that
Andrea Carnaghi, Department of Psychology, University of Trieste, produces distinctly negative associations, quite different from the
Trieste, Italy; Anne Maass, Sara Gresta, Mara Cadinu, and Luciano Arcuri, corresponding neutral group labels (e.g., gay, Afro-American; Car-
Department of Psychology, University of Padova, Padova, Italy; Mauro naghi & Maass, in press; Carnaghi, Maass, Bianchi, Castelli, &
Bianchi, International Graduate College, University of Jena, Jena, Ger- Brentel, 2005; Simon & Greenberg, 1996). In summary, there is
many.
evidence for Allports (1954) claim that noun labels serve both
We are grateful to L. Castelli, G. B. Flores DArcais, R. Job, L. Lotto,
and V. Yzerbyt for their helpful comments on our studies.
as organizing devices and as carriers of affective information.
Correspondence concerning this article should be addressed to Andrea However, do they fulfill these functions to a greater extent than
Carnaghi, University of Trieste, Via S. Anastasio 12, 34134, Trieste, Italy. other linguistic forms? In particular, how do they compare with
E-mail: acarnaghi@units.it adjectives, the word form that is dominant in spontaneous

839
840 CARNAGHI ET AL.

person description as well as in personality research in Western target, but with the same color). Depending on experimental con-
languages?1 dition, participants were asked to point either to the blue (i.e., the
Note that despite Allports (1954) early intuition regarding the noun form) or to the blue one (i.e., the adjectival form). Results
distinct role of nouns, they have remained a neglected word class indicated that in the adjective condition, participants chose the
in social and personality psychology. Although extensively inves- property-matching object whereas in the noun condition partici-
tigated by linguists, nouns (unlike adjectives) have attracted sur- pants typically pointed to the target-kind matching object. In other
prisingly little attention in research on person perception and words, Hall and Moore (1997) showed that preschoolers used noun
personality. To cite but two examples, practically all Big Five labels as event-organizing devices based on category membership,
research conducted in different countries and on different lan- whereas they used attributes to indicate a contingent property of
guages has relied exclusively on adjectives, with only a few the object.
researchers considering the possibility that people may also be Third, whereas adjectives tend to refer to a single property,
described by nouns (e.g., De Raad & Hoskens, 1990; Saucier, nouns imply a complex cluster of characteristics, although the
2003; see also Angleitner, Ostendorf, & John, 1990). In a similar noun is generally considered more than the sum of these properties
vein, Semin and Fiedlers (1988) influential linguistic category (Wierzbicka, 1986; see also Jespersen, 1968). An example cited by
model (LCM) distinguishes four word classes (including three Wierzbicka (1986) may illustrate this point: Whereas blond refers
types of verbs and adjectives) relevant to the interpersonal domain, to a single property of the target person, namely his or her hair
without considering nouns. The scope of the present research is to color, the noun blonde implies multiple properties (including being
start filling this lacuna by investigating the distinct role that nouns female and probably a host of additional characteristics including
play, compared with adjectives, in person description. We first sexy, glamorous, and not particularly intelligent). It is therefore
briefly review some of the linguistic literature concerning differ- not surprising that nouns elicit richer associations than adjectives.
ences between nouns and adjectives and then speculate about the Indeed, Loftus (1972) found that participants listed more associ-
implications for the domain of person perception. ated items when cued by a noun (e.g., fish, clam, etc., when
prompted by seafood) than when cued by an adjective (e.g., brick,
Nouns Versus Adjectives: Similarities and Differences cement, etc., when prompted by hard).
Fourth, and closely related to the above points, adjectives tend
At the surface, nouns and adjectives appear functionally similar, to allow distinctions of degree (e.g., a person may be more or less
as in the case of statements like Shira is a Jew versus Shira is athletic), whereas nouns tend to have an either or quality (e.g., a
Jewish. In many IndoEuropean languages, nouns and adjectives person either is or is not an athlete). This difference is clearly
may share the same word stem and differ only in suffix (athlete demonstrated by the fact that adjectives, but not nouns, allow
athletic, drunkdrunkard), implying, among other things, consid-
comparative and superlative forms (e.g., athletic, more athletic,
erable phonetic similarity. At times, they are distinguishable only
and most athletic).
by the article (e.g., Marco is an Italian vs. Marco is Italian; Laurie
Fifth, adjectives are parts of speech that modify nouns, suggest-
is an anorexic vs. Laurie is anorexic). At least in IndoEuropean
ing that nouns are primary in information processing and that
languages, they also seem to play a similar role in personality
adjectives can be fully understood only after the noun has been
description (Saucier, 2003).
processed. As argued by Wierzbicka (1986), the noun is semanti-
Despite the striking functional, phonetic, and semantic similar-
cally superordinate, whereas the modifying adjective plays a sub-
ities between nouns and adjectives, there are reasons to believe that
ordinate role. For instance, talking about a Catholic woman or a
nouns have a greater inductive potential and exert a greater effect
female Catholic is not the same, as the focus in the former is on
on impression formation than adjectives. Indeed, linguists have
gender (qualified by religion) and the focus in the latter is on
identified a number of ways in which nouns differ from adjectives
religious affiliation (qualified by gender). Indeed, the former
(e.g., Jespersen, 1986; Lyons, 1977; Wierzbicka, 1986). The first
phrase directs attention to the subportion of all females that is also
basic difference is that nouns represent a universal word class,
whereas adjectives are rare or entirely absent in some languages Catholic, whereas the opposite holds for the latter phrase. That the
(Dixon, 1977). adjective is subordinated with respect to the noun is also exem-
Second, and most important for the aims of our research, nouns plified by the fact that in languages that have nominal gender (such
identify the class to which a given object belongs whereas adjec- as German or Italian), the adjective adapts to the gender of the
tives denote a quality or property of the object. That is, nouns noun rather than vice versa. In line with the idea of superordination
categorize people by assigning them to a specific group or type or versus subordination in sentence processing, existing research
kind of person, which is quite different from adjectives that denote generally shows that nouns, compared with adjectives, have pri-
one of many qualities that a person may possess. It is interesting to macy in speech production and are more potent memory cues. In
note that children are sensitive to this difference between nouns speech production, nouns are chosen before adjectives simply
and adjectives as early as 2 years of age (Gelman & Coley, 1990). because the object has to be defined before its properties can be
For instance, Hall and Moore (1997) exposed preschoolers along
with adults to a target (i.e., a blue creature) that was defined either 1
Throughout this article, we use the terms nouns and adjectives in a less
by an adjective (e.g., This is a blue one) or by a noun (e.g., This is generic sense than is usually done. We are referring only to the subset of
a blue). Participants were then presented with two additional nouns and adjectives that can be applied to human beings and that gener-
pictures, one depicting a target-kind matching object (i.e., the same ally denote humans social membership (including intimacy groups, social
creature but with a different color) and the other depicting a categories, and loose associations, such as professions, see Lickel et al.,
property-matching object (i.e., a novel creature, different from the 2000) as well as their personal qualities (inclinations or personality types).
NOMINA SUNT OMINA 841

delineated, regardless of whether adjectives precede or follow the same noun label (e.g., bird), but not with the same adjective
nouns in a given language (e.g., a red automobile vs. un automo- label (e.g., wide awake), share the same nonobvious properties
bile rossa; see Martin, 1969). Also, there is evidence that nouns are (e.g., lives in a nest) even when the two animals were perceptually
generally more available for recall than adjectives (Lockhart, dissimilar (i.e., a dodo). Thus, nouns, compared with adjectives,
1969; Lockhart & Martin, 1969). For instance, in Lockhart and led preschoolers to go beyond the perceptual exemplar and draw
Martins (1969) study, participants were first asked to memorize a novel inferences on the basis of the animal category membership
series of adjectivenoun pairs, followed by a cued recall test in (see also, Gelman & Markman, 1986, 1987). Along similar lines,
which either the noun or the adjective served as the cue on the Gelman et al. (1986) found that gender nouns, such as boy,
basis of which participants were to recall the missing word. Results triggered richer inferences than gender-matching properties,
supported the idea that nouns are more effective cues than adjec- namely will grow up to be a daddy. Preschoolers appropriately
tives. inferred sex-related properties from a category-defining label (i.e.,
In summary, nouns differ from adjectives in important ways. a boy) but their performance decreased when they accomplished a
They (a) categorize events, rather than describing event properties; gender-categorization task in which they had to focus on sex-
(b) imply multiple qualities, rather than specifying a single prop- related properties (i.e., will grow up to be a daddy). In other words,
erty; (c) have an either or quality, rather than allowing distinctions Gelman et al. (1986) observed an asymmetry between inferring
of degree; and (d) have a superordinate status in speech production properties from a given category (i.e., deduction) and inferring the
and sentence comprehension. Thus, nouns are more efficient mem- category from properties (i.e., induction), with the deductive pro-
ory cues. cess being more viable than the inductive one.
The main thesis of this article is that the above differences Turning to the adult literature, few studies have contrasted the
between nouns and adjectives, identified by research in linguistics, use of and the consequences produced by nouns and adjectives.
may have important implications for impression formation. We Among the few exceptions is Markmans (1989) work demonstrat-
argue that people will form quite different impressions of and ing that nouns, compared with adjectives, trigger a richer repre-
make different inferences about others when nouns rather than sentation of the object that they define. Markman (1989) asked
adjectives are used to describe them. First, we hypothesize that participants to read a series of labels and to list, for each label, their
nouns, compared with adjectives, lead people to draw richer in- defining properties. Note that labels were either adjectives (e.g.,
ferences about the target and that these inferences will be largely intellectual) or nouns (e.g., an intellectual). Quantitative analyses
in line with stereotypical expectancies associated with the label. revealed that participants listed many more properties for nouns
Put simply, nouns will induce greater stereotyping than adjectives. than for adjectives. In a similar study by Markman and Smith (as
Second, we believe that nouns are more likely than adjectives to cited in Markman, 1989), participants were provided with sentence
inhibit alternative classifications of the same person. Thus, once a pairs depicting a person with an adjective (e.g., Alexander is
person has been identified as X (e.g., an athlete), it becomes intellectual) and with a corresponding noun (e.g., Alexander is an
subjectively difficult to believe that the same person may also be intellectual). Participants were asked to list what else might be
Y (e.g., an artist). In contrast, one adjectival quality is unlikely to typically expected of this person. Again, results indicated that
preclude another, unless there is a logical (semantic) contradiction. participants listed more attributes in response to the noun than to
For example, the same person may be athletic as well as artistic. the adjective label. In addition, when asked which sentence of each
Finally, we predict that nouns will convey greater essentialism pair seemed to be more informative, participants systematically
than adjectives, thus implying a more profound characteristic of chose the noun label and justified their choices by claiming that
the person. We explain the reasons for each of these predictions nouns were more lasting and central to the targets identity.
below and then provide a brief overview of the present research Together, the above research suggests that nouns, more than
project. adjectives, lead perceivers to draw inferences that go beyond the
information given. At least in adulthood, perceivers also judge the
Stereotypical Inferences event in question as more stable and enduring when nouns are
used, suggesting that the identity of an object is communicated
The first question to be addressed is whether nouns will induce more forcefully when expressed by nouns than by adjectives.
stronger inferences about the target person than adjectives and Although previous research on both children and adults clearly
whether these inferences will mirror the stereotypes that are asso- shows that nouns stimulate more inferential processing than ad-
ciated with the label. For instance, will people infer typical Jewish jectives, none of the above studies has investigated the degree to
habits more strongly when the target person is described as a Jew which these inferences follow stereotypical beliefs. We predict that
than when he or she is described as a Jewish person? The fact that nouns elicit only inferences that are congruent with what is ste-
nouns assign targets to types and generally delineate multiple reotypically associated with the label, but they may actually inhibit
properties would indeed suggest that they provide a richer source inferences about aspects that are incongruent. Because nouns rep-
for subsequent inferences. resent category-defining labels whereas adjectives stand for
To our knowledge, there are few studies that have investigated category-related properties (Hall & Moore, 1997), we argue that
inferences elicited by nouns and adjectives, and most of them have nouns (e.g., Jew) will make category-related contents more salient
taken a developmental (Gelman & Coley; 1990; Gelman, Collman, than the corresponding adjectives (e.g., Jewish). Moreover, im-
& Maccoby, 1986; Gelman & Markman, 1986, 1987) rather than pression formation research has consistently shown that to the
social psychological or cognitive perspective (Markman, 1989; extent to which a given category is salient, an individual target will
Markman & Smith as cited in Markman, 1989). Gelman and Coley be assimilated to the category-related contents (Fiske & Neuberg,
(1990) reported that preschoolers inferred that two animals with 1990). On the basis of these premises, we hypothesized stronger
842 CARNAGHI ET AL.

stereotypical inferences (e.g., He always goes to the Synagogue), predict that human characteristics that are designated by nouns
but weaker counterstereotypical inferences (e.g., He always goes (rather than adjectives) will be perceived as more permanent
to the Church), when an individual target is described by a noun and/or important.
(e.g., Mark is a Jew) than when the same individual target is To our knowledge, this hypothesis has not been tested in the
described by the corresponding adjective (e.g., Mark is Jewish). past, but indirect evidence for the high degree of essentialism
Thus, unlike prior research, our studies investigate both facilitative implied by nouns comes from studies comparing nouns with verbs
and inhibitory effects of nouns (vs. adjectives) on inferential (rather than adjectives). As a matter of fact, Walton and Banaji
processing. If confirmed, this would have important implications (2004) found that the same behavioral preference was perceived as
for the perpetuation of social beliefs, considering that the facilita- a much deeper seated and more central characteristic when the
tion of stereotypical inferences and inhibition of counterstereotypi- person was described by a noun (e.g., carrot eater) rather than by
cal inferences, typical of nouns, would render stereotype change a verb (e.g., eating carrots). In our studies, we aimed to extend this
rather difficult. The first aim of our research was therefore to line of research by focusing on the (much more subtle) difference
investigate this issue. between adjectives and nouns, hypothesizing that nouns entail
greater essentialism than adjectives. Indeed, we varied the label
Inhibition of Alternative Classifications (i.e., noun, such as artist, vs. adjective, such as artistic) with which
the target was described and then assessed the inferences about the
The second issue to be addressed in this article is whether nouns targets behavioral preferences (such as likes to draw paintings).
are likely to inhibit alternative classifications more so than other If nouns were indeed shown to imply greater essentialism, then
word forms. At a theoretical level, this idea was originally pro- it becomes legitimate to ask whether the inverse relationship may
posed by Allport in 1954 when he argued that nouns inhibit also be true. Not only may nouns induce essentialist perceptions in
alternative classification or even cross-classification (p. 179) of the listener, but it is also possible that speakers who do hold
the same object (or person). In other words, the fact that nouns essentialist beliefs find nouns particularly useful for revealing this
assign the object (in our case, the person) to a specific, all encom- conviction. Therefore, the fourth and final aim of this research
passing class and that this assignment has the quality of an all-or- project was to investigate how a speakers beliefs may guide his or
none statement makes alternative classifications less likely. Al- her linguistic choices and, more specifically, whether nouns would
though this hypothesis is interesting and intuitively appealing, we be considered particularly useful for expressing essentialist beliefs.
are not aware of any empirical proof (or disconfirmation) reported
in the literature. Overview of the Studies and Hypotheses
Although individuals can theoretically be classified according to
many different dimensions (age, sex, ethnic group membership, To address the above questions, we present the following studies
profession, religion, nationality, etc.), once a specific classification addressing the four interrelated issues. First, in Studies 1 and 2, we
has occurred, alternative classifications tend to be inhibited. For tested the hypothesis that nouns (e.g., athlete) trigger more stereo-
instance, Macrae, Bodenhausen, and Milne (1995) have demon- typical inferences and fewer counterstereotypical inferences than
strated such inhibitory mechanisms in their well-known research the corresponding adjectives (e.g., athletic). In other words, com-
involving a target classifiable either as a woman or as Chinese. pared with adjectives, nouns were expected to facilitate stereotyp-
When one of the two categories (either woman or Chinese) became ical inferences but to prevent counterstereotypical inferences. In an
highly accessible, the competing category was inhibited. Although additional experiment (Study 3), we investigated whether this
the role of inhibitory processes in social categorization is well would hold even in the case in which the same person was
documented in the social psychological literature, little is known described both by a noun and by an adjective. Thus, we tested the
about the role that language plays in this process. Despite the lack relative weight of nouns and adjectives in combined noun
of research on this issue, there are reasons to give credence to adjective descriptions such as Marco is an artistic athlete or
Allports (1954) intuition, considering than nouns delineate a Marco is an athletic artist, again hypothesizing a greater impact of
single, all-encompassing type of person, which is less likely to nouns in eliciting stereotype-congruent inferences. This study also
allow competitors than are adjectives describing one out of many allowed us to test whether the descriptor encountered first exerts a
possible qualities. Testing this conjecture was one of the main aims greater influence on the perception of the target person than the
of this research. one encountered later.
The second question we investigated was whether nouns would
Essentialism inhibit alternative classifications more so than adjectives. Because
every person has multiple category membership (e.g., Jew,
The third question addressed in this research program is whether woman, academic, Italian, environmentalist), he or she may legit-
nouns imply greater essentialism than adjectives. For example, imately be described in many different ways. However, following
art-related activities such as painting or visiting art exhibitions are Allports (1954) argument, once targets with multiple member-
certainly typical of both artistic people and artists, but when ships have been assigned to one particular social category (e.g.,
performed by an artist, these behaviors are likely to be interpreted Jew), other classifications (e.g., woman, Italian) should become
as a more profound and enduring behavior tendency than when difficult. Following Allports reasoning, we argue that the inhibi-
performed by an artistic person. Indeed, Wierzbicka (1986) has tion of alternative classifications should be much stronger when
speculated that human characteristics tend to be designated by nouns (Jew), rather than adjectives (Jewish), are used, for the
nouns rather than adjectives if they are seen as permanent and/or simple reason that nouns already define a category membership
conspicuous and/or important (p. 357). Inverting this logic, we whereas adjectives delineate a quality that is compatible with
NOMINA SUNT OMINA 843

many different categories. This issue was investigated in Study 4, Together, we hoped that this research would provide convergent
in which a target person was described either by a noun or by the evidence for the powerful effect of lexicalization (Gelman &
corresponding adjective. Participants were asked to estimate the Heyman, 1999) in the process of impression formation. More than
likelihood that the target would also belong to another, unrelated adjectives, we expected nouns (a) to bolster category-congruent
category (noun) or possess another, unrelated trait (adjective). expectations and to reduce the expectancy of incongruent behav-
The third issue, addressed in Study 5, concerned the differential iors (Studies 1 3), (b) to inhibit possible alternative classifications
likelihood of nouns versus adjectives to elicit essentialist percep- (Study 4), and (c) to suggest greater essence (Study 5). Finally, in
tions of the target. If nouns are more abstract than adjectives, then a complementary vein, the essentialist views of speakers were
the behavioral preferences of a target described by a noun should expected to lead them to use more nouns and fewer adjectives
be perceived as intrinsically linked to the person. In line with (Study 6).
previous empirical evidence on the issue under consideration
(Walton & Banaji, 2004; but see also Gelman & Heyman, 1999;
Study 1: Stereotype-Congruent Inferences
Markman, 1989), we hypothesized that a target described by a
noun (Mark is an athlete), rather than by the corresponding adjec- Study 1 aimed at testing whether the linguistic form (i.e., ad-
tive (Mark is athletic), would allow perceivers to appraise the jectives vs. nouns) in which an individual target is described can
targets preference as stronger, more persistent, and less prone to affect perceivers estimates of the frequency of descriptor-
be modified under situational constraints. This was expected for congruent behaviors. Participants were provided with a short de-
the case in which the targets preference was congruent with the scription of a target person by way of either an adjective (i.e.,
labels meaning (e.g., he runs three times per week). In contrast, athletic) or a noun (i.e., athlete). Participants were then required to
behavioral preferences that are incongruent with the labels mean- estimate the frequency of a descriptor-congruent behavior as a
ing (e.g., he drinks a lot of wine each day) should lead perceivers measure of stereotype application. We expected that participants
to consider the targets preference as less essential, that is as would make stronger inferences from nouns than from adjectives.
weaker, less enduring, and more likely to vary under situational This hypothesis was tested on professions and personality types in
constraints, when the target is described by a noun rather than by Study 1A, on intimacy groups in Study 1B, and on a random
an adjective. Thus, noun (compared with adjective) labels will sample of adjectivenoun pairs in Study 1C.
foster the perception of essence in cases of fit but will undermine
the attribution of essence for characteristics that do not concord
with the label. Study 1A
Finally, in our last experiment (Study 6), we inverted the logic
Method
of the above experiments. Rather than investigating the effects that
nouns and adjectives have on the attribution of essence, we tested Participants. Forty-eight students enrolled at the University of
what words people would use in interpersonal communication, Padova participated in the experiment (30 women and 18 men).
depending on whether they do or do not have essentialist beliefs. Participants were recruited at the university library and completed
Thus, our last experiment focused on the speaker rather than on the the questionnaire individually.
receiver of the communication. The main prediction was that Procedure and materials. Participants were told that we were
people would show a greater preference for nouns when they interested in the way people form an impression about a target
perceive a given characteristic as fundamental, lasting, deep- about which they had only a restricted amount of information
seated, and/or innate. In contrast, when the same characteristic is available. This cover story was used in all studies, unless otherwise
perceived as transient or less essential, people should prefer to use noted. Participants were provided with a questionnaire containing
adjectives. We induced the belief that athletic abilities (Study 6A) three sentences each portraying an individual target. Each sentence
or intelligence (Study 6B) is either genetically determined or contained the name of the target (Paolo, Marco, or Fabio), his age
mainly a result of training. We expected that participants, when (21, 22, or 23 years), and a short description. Depending on the
asked to describe a specific target person engaging in athletic (or experimental manipulation, the description was given by means of
intelligent) behaviors, would use more nouns and fewer adjectives either a noun (e.g., Paul, 22 years old, is an artist) or an adjective
when they had a genetic theory of athletic (or intellectual) abilities. (e.g., Paul, 22 years old, is artistic). Each participant received
The nouns and adjectives used in the different studies of this three descriptors either in noun form (i.e., artista [artist], atleta
research project cover the main types of social groups identified by [athlete], and genio [genius]) or in the corresponding adjective
Lickel et al. (2000) in their seminal work on intuitive theories of form sharing the same word stem to assure a maximum degree of
groups, namely intimacy groups (e.g., father, mother), social cat- phonetic and semantic similarity (i.e., artistico [artistic], atletico
egories (e.g., homosexuals, Catholics), and loose associations [athletic], and geniale [brilliant]).
(e.g., professional groups such as artists or athletes). The only type Participants then had to estimate the frequency of the corre-
of group not considered in this research is task groups (e.g., sponding descriptor-congruent behavior (respectively, How many
students studying for an exam, employees of a local restaurant) paintings does he draw each week? How many kilometers does
that can be defined only in context and hence cannot easily be he run per week? and How many problems does he solve in a
described by a single noun or adjective. Instead, we added a type week?) on 11-point scales ranging from 1 (few) to 11 (many).
of category not considered by Lickel et al., namely nouns and These behaviors were selected on the basis of a small pretest (N
adjectives describing personality characteristics (e.g., individual- 9) showing that each behavior was judged typical of one category
ist, drunkard, genius), because such personality types are particu- label but at the same time unrelated to the remaining two catego-
larly common in person descriptions. ries. To decrease the risk of participants selecting behaviors that
844 CARNAGHI ET AL.

were more typical of nouns than of adjectives, we had participants However, a potential limit of the above findings is that both
in the pretest judge each behavior with respect to category labels Studies1A and 1B deal with restricted samples of the adjective
comprising both the noun and the adjective descriptor (e.g., How noun pairs that may not be representative. As a matter of fact, one
typical is it for an athlete/for an athletic person to run?). may argue that our results reflected the peculiarity of the experi-
The order of the presentation of the sentences was counterbal- mental material rather than representing a robust phenomenon that
anced. We expected participants estimations of the descriptor- can be generalized to any other matched nounadjective pair. In
congruent behaviors to be stronger in the noun condition than in order to bolster the external validity of our results, we therefore
the adjective condition. decided to run a new experiment using a larger and randomly
selected list of adjectivenoun pairs.
Results and Discussion
Study 1C
First, in order to equate the two dependent variables, we stan-
dardized participants ratings for each descriptor-congruent behav- Participants. Thirty-seven students at the University of
ior into z scores and then analyzed these scores by means of a 3 Padova were recruited at the university library and completed the
(type of behavior: draw vs. run vs. solve) 2 (condition: adjective questionnaire individually. Six participants did not entirely fill out
vs. noun) 2 (version: Order 1 vs. Order 2) analysis of variance the questionnaire and were discarded from the analysis, reducing
(ANOVA). An almost significant main effect of condition was the sample to 31 students (24 women and 7 men).
found, F(1, 44) 3.81, p .057, 2p .08. Inspection of the Procedure and materials. The procedure was the same as that
means2 revealed that participants ratings tended to be higher in in Study 1A, except for the stimulus material and for the fact that
the noun (M 7.23, SE 0.33) than in the adjective condition a within-participants design was used. Participants read 24 sen-
(M 6.25, SE 0.36), suggesting that nouns were more likely to tences each portraying an individual target. Each sentence con-
elicit stereotypical inferences than adjectives. No other effects tained the name of the target (e.g., Paolo) and a short description.
were found. Twelve sentences included nouns as target descriptors and twelve
sentences included adjectives as target descriptors.
As for the selection of the adjectivenoun pairs, we used a
Study 1B random number table to identify pages of the Zingarelli (1988)
This study aimed to extend the results of Study 1A to a different dictionary, one of the most widely used Italian dictionaries. Start-
semantic as well as social domain, namely kinship or intimacy ing from each randomly selected page, we selected the first noun
groups (see Lickel et al., 2000). adjective pair that satisfied the following six criteria:

1. Noun and adjective shared the same word stem and


Method
appeared in close proximity (generally as a part of the
Participants. Twenty participants (11 women and 9 men) en- same paragraph).
rolled at the University of Padova participated in the experiment.
Procedure. The experimental procedure was same as in Study 2. Neither noun nor adjective was reported in the dictionary
1A, except for the experimental material. Participants, in a within- as rare or archaic.
participants design, were given two nouns (mother, father) and two
3. Nouns and adjectives were judged as semantically similar
corresponding adjectives (maternal, paternal) as the target de-
according to two independent judges as well as to the
scriptors. They were then asked to rate the frequency of descriptor-
definitions provided by both the Zingarelli (1988) and the
congruent behaviors (i.e., To what degree is the target interested
Devoto and Oli (1971) dictionaries. Note that semantic
in food-related issues?; To what degree is the target engaged in
similarity of the adjectivenoun pairs was essential to
financial matters? respectively, for the mothermaternal and for
assure that the same congruent (or incongruent) behav-
the fatherpaternal pairs) on a 7-point scale, ranging from 1 (not
iors could serve as dependent measures. Unpaired sam-
at all) to 7 (very much). We counterbalanced the order of presen-
ples of nouns and adjectives would necessarily require
tation of the targets. Again, we expected that participants would
that different and hence no longer comparable behaviors
estimate the frequency of a congruent behavior higher when ex-
be used as the dependent measures.
posed to a noun rather than to an adjective as the target descriptor.
4. Nouns and adjectives were applicable to human beings
Results and Discussion according to the judgments of two independent raters.
Participants ratings were first averaged to create two indices of For instance, the word pair archeologistarcheological
participants estimations, one for the nouns ( .61) and the other was excluded because the adjective is not easily applica-
for the adjectives ( .56). In line with our a priori hypothesis, a ble to human beings.
paired-sample t test revealed that participants ratings were higher
5. Nouns and adjectives were distinguishable by suffix. We
in the noun (M 5.88, SE 0.22) than in the adjective condition
did not allow any entries in which the exact same word
(M 5.35, SE 0.22), t(19) 1.71, p .05, one-tailed, d .53.
These results clearly show that participants were more prone to
draw descriptor-congruent inferences when they received a noun 2
Note that all the statistical analyses of Study 1 (but not of Study 1B),
rather than an adjective as the target descriptor, thus expanding Study 2, and Study 3 were computed on z-transformed scores, whereas, for
previous results to the realm of kinship or intimacy groups. the sake of clarity, all means are reported as raw scores.
NOMINA SUNT OMINA 845

was identifiable as noun and as adjective (such as Ab- t(38) 1.18, p .25, d .37. However, the difference between
original, Panamese), because, in such cases, the two nouns and adjectives may not lie so much in the average positivity
linguistic forms are undistinguishable, unless the article of the word but rather in the degree of polarization. Thus, nouns
is added. may imply a more extreme evaluation than adjectives both in the
positive and negative direction. To test this possibility, we also
6. Finally, if no appropriate pairs were found, applying the looked at the absolute distance of each rating from the neutral scale
above criteria, on three consecutive pages, we moved on midpoint (5). The degree of polarization was practically identical
to the next random number. for adjectives (M 1.95) and nouns (M 1.98), t(38) 0.12, p
.90, d .04, suggesting that there was no difference in the degree
Following these criteria, we selected 12 adjectivenoun pairs to which the two types of words carried implicit evaluative (pos-
(see the Appendix). A focus group identified a descriptor- itive or negative) content.
congruent behavior considered diagnostic of both noun and adjec- Although there were no reliable differences in valence or in
tive of each pair. Participants were asked to estimate the occur- evaluative polarization between nouns and adjective, we calculated
rence of the corresponding descriptor-congruent behavior on a the relative valence and the relative evaluative polarization of
7-point scale ranging from 1 (few) to 7 (many). In order to avoid nouns compared with adjectives by subtracting the mean ratings of
any effect due to the order of presentation, we counterbalanced the each adjective from that of the corresponding noun. Thus, the
list of the target descriptions. higher the value, the more positive the noun compared with the
adjective, or the greater the evaluative polarization of the noun
Results and Discussion compared with the adjective. We then added relative valence and,
subsequently, relative evaluative polarization to our main analysis
Participants ratings were first standardized into z scores and involving the stereotypical inferences. Remember that we had
averaged, thus creating two indices of target stereotyping, one found that nouns elicit stronger stereotypic inferences than adjec-
related to adjectives and one related to nouns. Because the order of tives, F(1, 11) 6.74, p .025, 2p .38. This effect continued
presentation did not produce any significant effect, we discarded to be reliable when adding relative valence as a covariate, F(1,
this factor from the current analyses. We analyzed the data first 10) 9.31, p .012, 2p .48, and it approached significance
using the participants and subsequently the stimuli as the unit of when using relative evaluative polarization as a covariate, F(1,
analysis. In line with our a priori hypotheses, participants ratings 10) 4.29, p .066, 2p .30. It is therefore unlikely that the
were higher in the noun (M 4.64, SE 0.11) than in the differential implicit valence of nouns versus adjectives may have
adjective condition (M 4.43, SE 0.12), F(1, 30) 7.34, p played a critical role in our findings. In other words, the fact that
.011, 2p .20. Furthermore, we repeated the analysis using the nouns elicit greater descriptor-congruent inferences than adjectives
nounadjectives pairs as random independent variables. Again, does not seem to be attributable to differences in evaluative content.
findings revealed that nouns (M 4.64, SE 0.14) elicited higher
ratings than adjectives (M 4.36, SE 0.21), F(1, 11) 6.74, p Study 2: Congruent and Incongruent Inferences
.025, 2p .38. Taken together, these results jointly confirmed
that nouns were more likely than adjectives to trigger stereotypical So far our studies show that compared with adjectives, nouns
inferences. Note that the stronger inductive power of nouns versus facilitate congruent inferences, but they do not test the related
adjectives has been found using both participants and stimuli as prediction that nouns also inhibit inferences about behaviors or
random variables.3 habits that are incongruent with the descriptor. Study 2 aimed to
Moreover, the use of an extended and randomly generated list of investigate this issue.
adjectivenoun pairs rules out the possibility that the findings may
simply be a function of biased or unrepresentative stimulus material.
Study 2A
The fact that an identical result pattern was obtained when randomly
selected adjectivenoun pairs were used suggests that the findings of Method
Study 1A can be generalized across different linguistic labels within
the larger word classes of nouns and adjectives. Participants. Sixty-four students (39 women and 25 men)
There is one potential confound that may create problems for the enrolled at the University of Padova were recruited in the univer-
interpretation of our results, namely that nouns and adjectives may sity library and asked to fill out the questionnaire individually.
differ in implicit valence. Studies on the LCM have shown that Two participants were excluded from the analyses because they
with increasing abstraction (e.g., passing from verbs to adjectives), completed questionnaires collectively, resulting in a final sample
words carry more evaluative information, which, in turn, may of 62 participants (38 women and 24 men).
affect stereotypical inferences. For instance, if nouns were to carry Procedure and materials. Half of the participants received
more evaluative meaning than adjectives, then greater inductive information about the two targets in adjective form (i.e., Paolo, 27
power for nouns may simply be a function of the fact that this
category of words describes more extreme (positive or negative) 3
Although several authors argued that minF-test (Clark, 1973) is
characteristics than adjectives. To test this possibility, we had 40 needlessly complicated and highly conservative (Baayen, Feldman, &
volunteers rate the valence of our stimulus material on a 9-point Schreuder, 2006; Brysbaert, 2007), we decided to further compute a
scale from 1 (very negative) to 9 (very positive), half of whom minF-test that turned out to be marginally significant, F(1, 31) 3.51, p
rated the 12 adjectives and the other half the 12 nouns. On average, .07, confirming, at least in part, the generalizability of our findings to
adjectives (M 5.49) were rated as positive as nouns (M 5.23), other sets of adjectivenoun matching pairs.
846 CARNAGHI ET AL.

years old, is artistic; Marco, 27 years old, is drunk), and the other of valence, these results provided additional evidence of the inde-
half received information about the two targets in noun form (i.e., pendence of the effect under consideration from the valence of the
Paolo, 27 years old, is an artist; Marco, 27 years old, is a characteristic that is being described.4
drunkard). The order of presentation of the targets was counter- Together, this study confirms the idea that nouns have a greater
balanced across participants. likelihood than adjectives to induce stereotype-congruent expect-
For each individual description, participants were asked to es- ancies and to inhibit incongruent ones. However, one potential
timate the frequency of occurrence of two behaviors (i.e., How problem is that the specific nouns selected for the above study may
many kilometers does he run per week? and How many glasses have differed semantically from the respective adjectives. Espe-
of wine does he drink per week?) on an 11-point scale ranging cially in the case of athlete, the noun seems to imply a higher
from 1 (few) to 11 (many). On the basis of a pretest (N 9), (professional) level of involvement in the activity and may, indeed,
behaviors were selected so that what was congruent with one evoke different exemplars than the respective adjectives. They
category was incongruent with the other, and vice versa. The order may also differ in degree of mutual inclusiveness such that all
of the behaviors to be estimated was counterbalanced across par- athletes are athletic but not all athletic people are athletes. If so, the
ticipants. We expected participants to judge descriptor-congruent above results could be interpreted in light of the theoretical con-
behaviors more frequent in the noun condition but incongruent jectures and the empirical evidence that fall under the banner of
behaviors more frequent in the adjective condition. correspondence bias (Gilbert & Malone, 1995; Jones, 1990;
Jones & Harris, 1967; Ross, Amabile, & Steinmetz, 1977). In line
Results and Discussion with the literature on the correspondence bias, one may conclude
that higher levels of target stereotyping in reaction to athleteartist
Participants ratings on the two items were first z transformed, than to athleticartistic may not occur by virtue of the differential
thus combining them into two indices of participants ratings, one linguistic properties of these labels but as a function of the higher
for descriptor-congruent behaviors and the other for descriptor- dispositional attribution that the professional role entails. Although
incongruent behaviors. As the order of presentation of the targets this alternative explanation seems to hold mainly for those stim-
and the order of presentation of the to-be-estimated behaviors did ulus words that refer to professions but less so for intimacy groups
not produce any significant effects, we discarded these two factors (madre-materno [mothermaternal], padrepaterno [father
from the current analyses. A 2 (condition: adjective vs. noun) 2 paternal]) and personality types (ubriaco-ubriacone [drunk
(domain: athletics vs. drinking) 2 (behavior: congruent vs. drunkard], geniogeniale[genius brilliant], individualista
incongruent) ANOVA was carried out on participants ratings, individualistic [individualistindividualistic]), we decided to conduct
with the former factor as a between-participants variable and the an additional study in which this potential confound was removed.
latter factors as within-participants variables. In line with our
predictions, the Condition Behavior interaction was significant,
F(1, 58) 14.14, p .001, 2p .20. As Table 1 shows, Study 2B
participants rated descriptor-congruent behaviors more frequent in In Study 2B, we therefore tested the same general hypothesis,
the noun than in the adjective condition, F(1, 58) 9.52, p .003, but this was done (a) using semantically very similar nouns and
2p .14. By contrast, they rated descriptor-incongruent behaviors adjectives; (b) extending the investigation to a different language
less frequent in the noun than in the adjective condition, F(1, community; and (c) relying on labels referring to social groups,
58) 4.12, p .05, 2p .07. rather than professional roles, personality characteristics, or inti-
Conceptually replicating the results of our previous studies, macy groups. We chose two pairs (homosexuala homosexual and
participants expected the target to show more category-congruent Catholica Catholic) that did not make any reference to a profes-
and fewer category-incongruent behaviors when the target was sion and for which nouns and adjectives implied each other in a
labeled by a noun rather than by an adjective. Please note that in symmetrical fashion (a homosexual person is very likely to be a
contrast to Study 1, this study included a positive label (i.e., homosexual and vice versa).
athlete) and a negative label (i.e., drunkard). Because we found Moreover, the extension to a different language, in this case
greater target stereotyping for nouns than for adjectives regardless German, was important to understand whether the systematic
differences between adjectives and nouns observed in our first two
studies can be generalized to other IndoEuropean languages. Note
Table 1 that Italian and German differ in the order in which nouns and
Participants Frequency Estimates as a Function of Behavior adjectives generally appear. In German, just like in English, the
and Condition (Study 2A) adjective generally precedes the noun, whereas the canonical order

Behavior
4
Note that the z transformation does not allow differences between the
Congruent Incongruent two domains, drunkard and athlete, to emerge. We therefore repeated the
ANOVA on the raw scores. Besides replicating the Condition Behavior
Condition M SE M SE interaction, F(1, 58) 12.76, p .001, 2p . 94, we did not find a main
effect of domain. Congruent behaviors were judged more frequent than
Adjective 9.37a 0.22 3.83a 0.32
Noun 10.27b 0.22 2.77b 0.32 incongruent behaviors, both in the athletics domain (respectively, M
9.68 and M 3.93) and in the drinking domain (respectively, M 9.95
Note. Means with different subscripts differ significantly within columns and M 2.77). More important, the Condition Behavior Domain
(t test, ps .05). interaction was not significant, F(1, 58) 0.63, ns, 2p .01.
NOMINA SUNT OMINA 847

in Italian is inverse, although both orders are possible in principle .12. On the basis of the results of Study 1 and Study 2A and in line
(we come back to this issue in Study 3). Theoretically, it is with our theoretical conjectures, we expected that nouns would
therefore possible that the greater inferential potential of nouns in trigger more descriptor-congruent inferences but fewer descriptor-
Italian is, at least in part, a function of the fact that nouns tend to incongruent inferences than adjectives. In keeping with these a
precede adjectives. This would be in line with findings showing priori hypotheses, an inspection of the means revealed that partic-
that the recall advantage of nouns over adjectives is particularly ipants frequency ratings for the descriptor-congruent behaviors
pronounced when nouns precede adjectives (Kusyszyn & Paivio, were higher in the noun condition (M 30.74, SE 4.74) than in
1966; Paivio, 1963). Thus, it was important to test whether nouns the adjective condition (M 19.02, SE 4.67), t(61) 1.75, p
would allow stronger inferences than adjectives even in languages .04, one-tailed, d .31. By contrast, participants ratings for the
(such as German or English) in which they generally have a less
descriptor-incongruent behaviors were lower in the noun (M
favorable (second) position in relation to adjectives.
4.89, SE 1.67) than in the adjective condition (M 8.77, SE
0.15), t(61) 1.76, p .04, one-tailed, d .31.
Method Again, replicating the results of our previous studies, partici-
Participants. Seventy-one students enrolled at the University pants reported stronger target stereotyping when the target was
of Jena participated in this experiment. We selected participants described by a noun rather than by an adjective. Moreover, and in
that were neither Catholic nor homosexual. Our final sample then line with results of Study 2A, participants expected the same target
included 65 participants (31 women and 33 men). One additional to display incongruent behaviors with lower frequency when a
participant was excluded from the current analyses because his noun rather than the corresponding adjective was used as the target
ratings on two items were four standard deviations above the descriptor. In summary, these results provide further support for
mean5 (for a similar procedure, see Walton & Banaji, 2004). our hypothesis that nouns facilitate congruent, but inhibit incon-
Procedure and materials. Similar to the experimental proce- gruent inferences more so than adjectives. It is important to note
dures outlined above, participants read a brief description of two that the findings of this study also suggest that this is true even in
male targets. In one condition, varied between participants, the languages in which adjectives have a primacy advantage as they
noun referred to sexual preference and the adjective to religion (a generally precede nouns. Also, in contrast to our previous studies,
homosexual, Catholic), and in the other, the inverse was true (a the adjectivenoun pairs of this study did not refer to professional
Catholic, homosexual). Note that, unlike English, German adjec- roles, to personality characteristics, or to intimacy groups but
tives and nouns differ not only in terms of the presence or absence instead referred to social categories (see Lickel et al., 2000),
of the article but also with respect to the uppercase versus lower- suggesting that the same processes that were found for assumed
case initial letter and the suffix (noun: ein Homosexueller, ein categories in our first studies are also at work when ascribed
Katholik; adjective: homosexuell, katholisch). The order in which categories are involved (Mae & Carlston, 2005).
the targets were presented was varied between participants. Fol-
Moreover, and in line with the results of Study 2A, it is inter-
lowing each target description, participants had to estimate the
esting to note that facilitation and inhibition were practically
frequency with which the target person engaged in two behaviors,
identical for the category that is positively valued in society
namely How often does he attend church in a year? and How
(Catholic) as for the category that is often subject of discrimination
often does he have one-night stands in a year? In contrast to our
previous studies, an open response format was chosen ( __ times (homosexual).6 Thus, valence of the characteristic and/or the cor-
per year). A pretest session of a small sample of students (N 8) responding behavior (going to church vs. having one-night
indicated that to attend church was considered as more typical of stands) does not seem to moderate the lexicalization effect. Fi-
a Catholic personCatholic (M 5.63, SE 0.50) than of a nally, the robustness of the lexicalization effect is also corrobo-
homosexual person homosexual (M 2.13, SE 0.35), F(1, rated by the observation that it emerges independently of the way in
7) 38.11, p .001, 2p .85, whereas to have one-night which participants report their estimates. Indeed, the current study
stands was judged to be more typical of a homosexual person shows that our previous findings, based on rating scales, extend to a
homosexual (M 5.50, SE 0.42) than of a Catholic person situation in which estimates are generated in a free response format.
Catholic (M 1.89, SE 0.40), F(1, 7) 41.17, p .001, 2p Together, these results suggest that we are dealing with a robust and
.86. Therefore behaviors were selected so that what was congruent pervasive phenomenon that holds across languages, nounadjective
with one descriptor was incongruent with the other, and vice versa. pairs, categories, valence, and types of measurement.

Results and Discussion 5


The Condition Behavior interaction was still significant, F(1, 62)
Because the order of presentation of the targets and the order of 5.87, p .02, 2p .09, even when the outlier was not discarded from the
analysis.
presentation of the to-be-estimated behaviors did not produce any
significant effects, these two factors were not considered in the
6
A 2 (condition: adjective vs. noun) 2 (target: homosexual vs.
Catholic) 2 (behavior: congruent vs. incongruent) ANOVA on the raw
analyses. A 2 (condition: adjective vs. noun) 2(domain: homo-
data confirmed the Condition Behavior interaction, F(1, 61) 5.92, p
sexuality vs. Catholicism) 2 (behavior: congruent vs. incongru- .02, 2p .67, and also showed that this interaction was not moderated by
ent) ANOVA was conducted on participants z-transformed rat- type of target, F(1, 61) 0.17, ns, 2p .07. Corroborating, albeit in a
ings, with the former factor as a between-participants variable and different manner, the results of Study 1C, this pattern shows that higher
the latter as a within-participants variable. The Condition Be- levels of target stereotyping occur in reaction to nouns rather than to
havior interaction was significant, F(1, 61) 8.34, p .005, 2p adjectives, regardless of valence.
848 CARNAGHI ET AL.

Study 3: Inferences When Nouns and Adjectives Co- Italian male name, age, and a short description that in contrast to
Occur the above studies, contained two descriptors, one referring to the
athletic and the other to the artistic inclination of the same person.
The first two studies, conducted in two different languages, Target persons were described by two nouns (e.g., athlete and
mainly focused on perceivers inferences about an individual tar- artist), by two adjectives (e.g., athletic and artistic), by an adjec-
get who was identified by a single label, either a noun or an tive followed by a noun (e.g., an athletic artist), or by a noun
adjective. However, in natural language, adjectives and nouns followed by an adjective. Note that although nouns generally
frequently co-occur, with adjectives qualifying nouns. For exam- precede adjectives, the word order can be inverted in Italian. Also,
ple, we may refer to a person as an athletic woman or as a female the material was organized so that the artistic inclination either
athlete. This offers the possibility for a direct comparative test of preceded or followed the athletic one (e.g., an athletic artist or an
our hypothesis, as the relative weight of nouns and adjectives in artistic athlete).
eliciting stereotype-congruent (and inhibiting incongruent) infer- Participants rated each of the eight target individuals with re-
ences about the target person can be determined within a single spect to the frequency of two behaviors (i.e., How many kilome-
minimal sentence. We know from previous research in which both ters does he run per week? and How many exhibitions does he
word forms were presented that adjectives are generally less avail- visit per week?) using an 11-point scale ranging from 1 (few) to
able for recall than nouns (Lockhart, 1969: Lockhart & Martin, 11 (many). Note that in a pretest, a small sample of students (N
1969). Indeed, if nouns serve as primary conceptual pegs to 7) indicated that to run was considered as more typical of an
which adjectival modifiers are attached (Kusyszyn & Paivio, 1966; athletic personathlete (M 6.43, SE 0.57) than of an artistic
Paivio, 1963), it is not surprising that they play a more important personartist (M 2.86, SE 0.60), F(1, 6) 38.11, p .005,
role in information processing and memory. By extension nouns 2p .75, whereas to visit exhibitions was considered as more
can also be expected to preserve their unique ability to induce typical of an artistic personartist (M 6.00, SE 0.44) than of
congruent (and inhibit incongruent) inferences even in the copres- an athletic personathlete (M 2.86, SE 0.60), F(1, 6) 37.08,
ence of an adjective. Thus, our main aim was to test whether nouns p .001, 2p .86. Therefore, behaviors were selected so that
would induce more stereotypic inferences than adjectives even what was congruent with one descriptor was incongruent with the
when the two were presented together. other, and vice versa. In addition, we varied the order of the
The combination of nouns and adjectives within a single de- stimulus sentences and of the dependent variables between partic-
scription also allowed us to test whether word order would affect ipants.
the inferences that people draw about the target. Although nouns
are superior to adjectives both in facilitating memory (Martin,
Results and Discussion
1969) and in guiding inferences, as shown by our Studies 1 and 2,
regardless of whether they precede or follow adjectives, it is Data were first z transformed and then combined into two
conceivable that this difference is further enhanced when nouns indices of participants ratings, one for the behaviors that were
also enjoy a primacy advantage. This is in line with classical congruent with the first descriptor (and incongruent with the
research showing that recall for adjectivenoun pairs is better second descriptor) and one for the behaviors that were congruent
when nouns precede, rather than follow, adjectives (e.g., Kusyszyn with the second descriptor (and incongruent with the first descrip-
& Paivio, 1966; Paivio, 1963). Extending this reasoning, we ar- tor). Because the order of presentation of the descriptions as well
gued that (a) nouns would induce greater inferences regarding as the order of presentation of the to-be-estimated behaviors did
congruent behaviors than adjectives (replicating Studies 1 and 2) not produce any significant effects, we discarded these factors
and (b) that this advantage of nouns over adjectives would be from the analyses. A 4 (stimulus pairs: nounadjective, adjective
enhanced when nouns precede adjectives. Note that we expected noun, nounnoun, adjectiveadjective) 2 (behavioral congru-
nouns, but not adjectives, to profit from the primacy position. The ency: congruent with first vs. congruent with second) ANOVA
reason for this asymmetrical order effect lies in the fact that nouns was carried out on participants ratings, with all factors as within-
constitute the primary element to which qualifying adjectives are participants variables.
subsequently attached. Indeed, various authors have argued that To facilitate interpretation of the findings, we first report effects
speakers habitually choose adjectives after choosing the modified pertaining to our main hypothesis, namely that nouns elicit more
nouns (Martin, 1969, p. 472; for similar arguments regarding the stereotypical inferences than adjectives, and discuss order effects
listener, see Ehrlich, 1977). Thus, nouns may induce stronger only subsequently. Looking first at the upper portion of Table 2,
stereotype-congruent inferences when they are encountered first, we can see that our main results of Studies 1 and 2 were perfectly
whereas the position of adjectives should be largely irrelevant replicated. In fact, the noun always elicited stronger congruent
because, regardless of order, the qualifier will become informative inferences than the adjective of each pair, regardless of whether it
only after the to-be-qualified noun has been identified. preceded or followed the adjective. In our design, this was re-
flected in a reliable interaction between stimulus pairs and behav-
ior congruency, F(1, 39) 48.07, p .001, 2p .55, when
Method
analyzing the upper portion of Table 2 separately.
Participants. Forty students (23 women and 17 men) enrolled The lower portion of Table 2 shows that on the average, noun
at the University of Padova were recruited in the library and noun pairs (M 4.99) elicited greater behavioral inferences than
completed the questionnaire individually. adjectiveadjective pairs (M 3.88), F(1, 39) 26.33, p .001,
Procedure and materials. Participants read minimal descrip- 2p .40. This replicates our findings obtained on single nouns and
tions of eight distinct individual targets each defined by a common adjectives, showing that the inductive potential of nouns greatly
NOMINA SUNT OMINA 849

Table 2 congruent behaviors from nouns than from adjectives also holds
Participants Frequency Estimates as a Function of Behavior when nouns and adjectives co-occur in the same minimal phrase.
Congruency and Stimulus Pair (Study 3) Second, the information encountered first appears to have a greater
weight as it induces a higher expectancy that the target person will
Behavior congruency engage in behaviors congruent with the descriptor. For example, a
Congruent with Congruent with
person described as an athlete and an artist is believed to engage
first descriptor second descriptor in more fitness and fewer art-related behaviors than a person
described as an artist and an athlete. Third, this primacy effect in
Stimulus pair M SE M SE which the first information affects impression formation more than
information encountered later only holds for nouns but not for
Nounadjective 5.51 0.19 3.59 0.22
Adjectivenoun 3.80 0.22 4.89 0.24 adjectives. This is in line with the idea that in nounadjective pairs,
Nounnoun 5.31 0.21 4.68 0.25 adjectives will acquire their meaning only after nouns have been
Adjectiveadjective 3.91 0.22 3.85 0.20 processed, so it is less important whether they precede or follow
the noun.

exceeds that of adjectives. Together, this suggests that the greater Study 4: Inhibition of Alternative Classifications
inductive potential of nouns compared with adjectives was con-
firmed, regardless of whether mixed pairs (upper portion of Table Whereas Studies 13 focused on stereotypical inferences, the
2) or uniform pairs (lower portion of Table 2) were presented. aim of Study 4 was to corroborate a different hypothesis, namely
In addition, a number of order effects emerged that are in line that nouns, much more than adjectives, would prevent perceivers
with our predictions. First, a main effect of behavioral congruency, from alternative classifications of a given social target. In line with
F(1, 39) 4.53, p .04, 2p .10, indicated that behaviors Allports (1954) suggestions, we hypothesized that nouns are more
congruent with the first descriptor (M 4.63, SE 0.16) were likely than adjectives to reify the individual target into the contents
rated more frequent than those congruent with the second descrip- that they describe, thus impeding alternative classifications. In
tor (M 4.25, SE 0.16). This reflects an order effect, demon- order to test this hypothesis, we conducted a study in which
strating that the first descriptor is primary as an information- participants were presented with a target description either in noun
organizing device. This effect was reliably modified by an form (e.g., athlete) or in adjective form (e.g., athletic). Then
interaction with stimulus pair, F(3, 37) 15.72, p .001, 2p participants had to estimate the likelihood that the target may also
.56, represented in Table 2. The upper portion of Table 2 shows possess another, unrelated characteristic. Please note that this
that the advantage of nouns over adjectives was stronger in the additional characteristic was, again, provided either in noun form
nounadjective condition (M 5.51 vs. M 3.59), F(1, 39) (e.g., artist) or in adjective form (e.g., artistic).
66.31, p .001, 2p .63, than in the adjectivenoun condition We hypothesized that nouns, compared with adjectives, would
(M 4.89 vs. M 3.80), F(1, 39) 22.64, p .001, 2p .37, block alternative classifications of the target, hence making it less
indicative of a strong order effect. likely that the target would also possess the additional, unrelated
Moreover, and as shown in the upper portion of Table 2, we had characteristic. Furthermore, we suspected that this would be
argued that word order would mainly be relevant for nouns but less mainly true in the case in which the alternative classification was
so for adjectives, considering that the noun (regardless of position) provided by a noun rather than an adjective. The reason for this
has to be processed before the adjective becomes meaningful. prediction is straightforward. Assigning a target initially to one
Crossover comparisons confirmed this prediction. Behaviors con- class (e.g., dog) makes it difficult to also assign it to another class,
gruent with the noun were judged much more frequent when the but because every target can have a variety of qualities (tall, dark,
noun appeared in the first (M 5.51) than in the second position fast, etc.), nothing argues against assigning a number of additional
(M 4.89), F(1, 39) 40.50, p .001, 2p .51. In contrast, for qualities to the same object. Indeed, in most IndoEuropean lan-
the adjective, the position was practically irrelevant considering guages, multiple adjectives can legitimately be associated with a
that inferences about congruent behaviors were very similar for the single noun (e.g., a large old house), but not vice versa (e.g., a
adjective in the first (M 3.80) and in the second position (M large house and tree).
3.59), F(1, 39) 0.76, ns, 2p .02.
If the order of presentation was relevant to nouns but not to
Method
adjectives, then this should also emerge for nounnoun and
adjectiveadjective pairs. The lower portion of Table 2 illustrates Participants. Forty students (23 women and 17 men) enrolled
that for the nounnoun pair, behaviors congruent with the first at the University of Padova were recruited in the library and
noun (M 5.31) were judged more frequent than those congruent completed the questionnaire individually.
with the second noun (M 4.68), although this difference fell Procedure and materials. Participants first read the same
slightly short of significance, F(1, 39) 3.43, p .07, 2p .08. cover story as in Study 1. Participants were provided with minimal
In contrast, there was no difference between behaviors congruent descriptions of eight distinct individual targets, each defined by a
with the first and with the second adjective, F(1, 39) 0.18, ns, common Italian male name, age, and a descriptor. For half of the
2p .005, suggesting that the order of presentation was irrelevant targets, the descriptor referred to their athletic inclination, and for
for adjectives. the other half, the descriptor referred to their artistic inclination.
Together, three main findings emerge from this study. First, and Following the instructions outlined by Macrae et al. (1995), we
most important, this study shows that the greater tendency to infer decided to use two categories that have no semantic overlap
850 CARNAGHI ET AL.

(unlike woman and housewife) and for which it is unlikely that one scriptor takes the form of a noun, perceivers are unlikely to include
constitutes a subtype of the other (such as woman and manager). the target in an alternative classification when this potential alter-
More important, regardless of the semantic content of the descrip- native is also a noun, although they are willing to assign the target
tor, each target person was labeled either by a noun (e.g., athlete a new adjective. In contrast, when an adjective is used to describe
and artist) or by an adjective (e.g., athletic and artistic). After the target initially, perceivers are equally likely to assign a new
reading a target description, participants had to estimate the prob- characteristic to the target, regardless of whether the new infor-
ability that the target (e.g., described as athletic) may also have mation is conveyed in adjective or noun form. These results may
another characteristic (e.g., artistic). In particular, participants be summarized very simplistically as follows: Assuming that
were prompted by the following question: Could you indicate the nouns assign targets to classes, whereas adjectives describe qual-
likelihood that [the name of the target] could also be [descriptor].
ities, participants behave as if each target can belong to only one
Participants provided their estimates by means of a percentage
class (be it the initial descriptor or the alternative one) but can have
format, ranging from 0% to 100%. We counterbalanced the order
any number of additional qualities.
of presentation of the target description.

Results and Discussion Study 5: Implicit Essentialism


Data were first combined to create four indeces of participants To this point, our research project has demonstrated two inter-
estimations. Specifically, two indices concerned the condition in related phenomena: First, nouns provide much more powerful
which the initial target descriptor was provided in noun form and descriptions of people than adjectives as they induce greater ex-
the alternative descriptor either in noun form or in adjective form, pectations that the person will engage in descriptor-congruent
and two additional indices were related to the condition in which behaviors and, at the same time, will not engage in incongruent
an adjective was used as the initial target descriptor and the behaviors. In other words, despite their great semantic and pho-
alternative descriptor was provided either in noun form or in
netic similarity, nouns (Carlo is a homosexual) produce more
adjective form.
stereotypical views of the person than do adjectives (Carlo is
A 2 (initial target descriptor: noun vs. adjective) 2 (additional
homosexual). Second, nouns block alternative classifications.
descriptor: noun vs. adjective) ANOVA, with likelihood estimates
However, stereotypical inferences and the inhibition of alternative
as the dependent variable, revealed a main effect for initial target
descriptor, F(1, 39) 4.07, p .05, 2p .094, showing that in classifications may not be the only way in which nouns differ from
line with our hypothesis, alternative classifications were more adjectives. Rothbart and Taylor (1992) have argued that social
likely when the target had initially been described by an adjective categories are cognitively represented by two distinct but related
(M 41.25, SE 2.67) than by a noun (M 36.94, SE 2.72). dimensions, namely stereotypicality and essentialism. Whereas
Also, a main effect for the additional descriptor revealed that Studies 13 dealt with the former dimension, the following three
participants were more likely to consider the target as also pos- studies (Studies 5A, 5B, and 5C) focus primarily on the essential-
sessing an additional characteristic when that characteristic was istic judgments that a noun, compared with the corresponding
expressed in adjective (M 40.63, SE 2.81) rather than in noun adjective, may elicit. Therefore, these studies test whether the
form (M 37.56, SE 2.33), F(1, 39) 4.23, p .05, 2p linguistic form (adjective vs. noun) in which an individual target is
.098. Most important, the predicted interaction emerged, although described can affect the perceived strength of an individuals
it fell slightly short of significance, F(1, 39) 3.7, p .06, 2p preference for descriptor-congruent behaviors (for a similar para-
.087. When the target had initially been described by a noun, the digm, see Gelman & Heyman, 1999; Walton & Banaji, 2004).
likelihood estimate was lower when the alternative classification
was presented in noun form (M 33.87, SE 2.86) than in
adjective form (M 40.0, SE 3.06), F(1, 39) 6.43, p .015, Study 5A
2p .15. Thus, participants were reluctant to assign an unrelated
characteristic to a target that had initially been described by a noun Method
if the new characteristic took the form of another noun but not
Participants. Fourteen female individuals attending the Uni-
when the same characteristic was proposed as an adjective. Please
versity of Padova participated in the experiment.7 Participants
note that this reluctance was not found when the target had initially
were recruited in a seminar class and completed the questionnaire
been described by an adjective. If an adjective had been used as the
in a collective session.
target descriptor, then an alternative classification of the target was
considered equally likely, regardless of whether the alternative Procedure. Participants read five sentences, each portraying a
descriptor took the form of a noun (M 41.25, SE 3.04) or an different individual target. Each sentence contained the individu-
adjective (M 41.25, SE 2.64), F(1, 39) 0.00, ns, 2p .007. als name (e.g., Paul), the individuals description (e.g., is an
In summary, when an individual target has been described by a artistis artistic) and the individuals behavioral preference (e.g.,
noun rather than an adjective, an alternative classification of the
same target is unlikely to occur. In line with our predictions,
nouns, more than adjectives, appear to pervade the identity of the 7
Only female students participated in Studies 5A and 5C for the simple
target, hence impeding a new classification. However, our results reason that the seminar, in which participants were asked to volunteer for
also suggest that this depends greatly on the way in which the the study, was exclusively attended by female students on the day of the
alternative classification is expressed. When the initial target de- data collection.
NOMINA SUNT OMINA 851

he likes to work with plaster).8 For each sentence, the descriptive that experiment. In order to ensure that the findings are not
term could be either an adjective or its corresponding noun (atleta attributable to any unintended bias in the selection of the material
atletico [athleteathletic], ubriacone ubriaco [drunkard drunk], and, hence, to the idiosyncratic characteristics of the selected
genio geniale [genius brilliant], poetapoetico [poetpoetic], and stimulus material, the same randomly selected pairs of adjectives
artistaartistico [artistartistic]). Note that if the descriptive term and nouns of Study 1B were included in the current study. As in
was an adjective [noun], its corresponding noun [adjective] was Study 5A, participants were asked to evaluate the targets prefer-
not included in the same questionnaire. Therefore, we created four ence for a descriptor-congruent behavior. It is worth noticing that
different versions of the questionnaire in which three individuals the targets behavioral preferences used in the current study were
were respectively described by adjectives and two by nouns and exactly the same as the descriptor-congruent behaviors of Study 1B.
four versions of the questionnaire in which two individuals were
described by adjectives and three by nouns.
The participants perception of the targets behavioral prefer- Method
ences were assessed on 9-point scales (see Walton & Banaji, 2004)
assessing strength (How strong is Xs preference for this activ- Participants. Twenty-one students (14 women and 7 men)
ity?), stability (How likely is it that Xs preference for this enrolled at the University of Padova participated in the experiment.
activity will remain the same in the next two years?), and resil- Participants were recruited in the library and completed the ques-
ience (How likely is it that Xs preference for this activity would tionnaire individually. One participant was excluded from the
remain the same if he was surrounded by friends who did not enjoy present sample because she was not an Italian native speaker.
the activity in question?). The first scale was anchored to very Procedure and material. Participants read 24 sentences, each
weak (1) and very strong (7), and the second and third scales were portraying an individual target. Each sentence contained the indi-
anchored to very likely to change (1) and very likely to remain the vidual s name (e.g., Marco), the individuals description (e.g., is
same (7). traditional vs. is a traditionalist), and the individuals behavioral
Note that the individuals behavioral preferences were always preference (e.g., he likes to send postcards for Christmas). In
congruent with respect to the descriptive terms. Therefore, we contrast to the previous studies, the name of the target was kept
expected participants to attribute greater essentialism (greater constant across the descriptions. For each sentence the descriptive
strength, stability, and resilience) to the individuals preference term could be either an adjective or its corresponding noun. The
when the individual was described by a noun rather than by an participants perception of the targets behavioral preferences was
adjective. assessed by means of the same three-item scale that we used in
Study 5A. Finally, participants were asked to evaluate the valence
Results and Discussion of each targets behavioral preference on a 7-point scale ranging
Data were first combined within each type of descriptor condi- from 1 (very negative) to 9 (very positive). Replicating the results
tion (adjective vs. noun) and within each participants ratings on of Study 5A, we expected participants to attribute greater essen-
the three scales. Therefore, we created six indices, namely one for tialism to the individuals preference when the individual was
perceived strength, one for stability, and one for resilience of the described by a noun rather than by an adjective.
behavioral preference separately for the adjective and the noun
stimuli. Because the three measures had reasonable internal con-
sistency in both the adjective (Cronbachs .74) and the noun
condition (Cronbachs .63), we averaged the three measures 8
A small sample of students, coming from the experimental population,
into two overall indices of perceived essentialism, one pertaining judged the typicality of a series of behaviors with respects to the labels we
to the adjectival stimuli and the other to the noun stimuli. used in the current study and in Study 6B. Results indicated that (a) to run
As the version of the questionnaire did not produce any signif- during the week was considered as more typical for an athleteathletic
icant effect, we analyzed participants ratings on the overall indi- person (M 6.67, SE 0.17) than to sit on a sofa while watching TV
ces of perceived essentialism by means of a paired t test. In line (M 2.55, SE 0.41), t(8) 8.48, p .001; (b) to find new solutions
with our hypothesis, participants judged the targets behavioral was judged to be more typical (M 6.67, SE 0.17) for a genius brilliant
preferences in a more essentialist manner when the target person person than to watch reality shows (M 3.44, SE 0.73), t(8) 4.24,