Content uploaded by Manfred Schmitt
Author content
All content in this area was uploaded by Manfred Schmitt on Apr 02, 2014
Content may be subject to copyright.
Group Processes &
Intergroup Relations
2008 Vol 11(1) 69–87
Implicit and Explicit Attitudes
and Interracial Interaction: The
Moderating Role of Situationally
Available Control Resources
Wilhelm Hofmann
University of Würzburg
Tobias Gschwendner
University of Koblenz-Landau
Luigi Castelli
University of Padova
Manfred Schmitt
University of Koblenz-Landau
The present research examined whether implicit and explicit racial attitudes predict interracial
interaction behavior differently as a function of situationally available control resources.
Specifi cally, we investigated how implicit attitudes (Implicit Association Test) and explicit
attitudes (Blatant/Subtle prejudice) were related to interracial interaction behaviors of Italians
toward an African interviewer (Study 1) and of Germans toward a Turkish interviewer (Study 2).
For half of the interview questions, participants’ control resources were reduced via a memory
task. Across both studies, the Race IAT was more predictive of behavior when participants were
taxed than when untaxed. Conversely, explicit attitudes were somewhat more predictive under
full resources. Taken together, our fi ndings suggest that available control resources moderate
the predictive validity of implicit and explicit attitudes.
keywords control resources, implicit and explicit attitudes, intergroup behavior, interracial
interaction
Author’s note
Address correspondence to Wilhelm
Hofmann, Universität Würzburg, Lehrstuhl
für Psychologie II, Röntgenring 10,
97070 Würzburg Germany [email:
hofmannw@psychologie.uni-wuerzburg.de]
Copyright © 2008 SAGE Publications
(Los Angeles, London, New Delhi and Singapore)
11:1; 69–87; DOI: 10.1177/1368430207084847
Under many circumstances, social behavior
is determined by impulses such as spontaneous
evaluations rather than by reasoned attitudes and
beliefs. Just imagine yourself standing in line
at the supermarket either after an exhausting
workday with a long list of unfi nished business
in mind or after your weekly yoga course and a
relaxing afternoon at the local wellness center.
Being eventually served by a foreign cashier,
in which of the two cases will your interaction
behavior more likely be infl uenced by your
spontaneous evaluation of foreigners, and
Group Processes & Intergroup Relations 11(1)
70
when will it be based instead on your reasoned
attitudes and beliefs?
A significant proportion of psychological
research over the last decade concerns the distinc-
tion between so-called implicit and explicit attitudes
and their roles in behavior determination. Im-
plicit attitudes can be understood as evaluations
that are spontaneously and effortlessly activated
upon the mere encounter of an attitude object
and may serve as ‘quick guides’ to behavior that
do not depend on people’s awareness and control
(e.g. Greenwald & Banaji, 1995). In contrast,
explicit attitudes are assumed to refl ect capacity-
consuming reasoned evaluations that infl uence
action through deliberation and motivation (e.g.
Fazio & Olson, 2003).
Implicit and explicit attitudes seem to
share some systematic overlap on average (e.g.
Hofmann, Gawronski, Gschwendner, Le, &
Schmitt, 2005) and the conditions that deter-
mine the magnitude of the relationship have
become increasingly well understood (Hofmann,
Gschwendner, Nosek, & Schmitt, 2005; Nosek,
2005). Furthermore, preliminary results from
a meta-analysis suggest that both implicit and
explicit measures contribute signifi cantly to the
prediction of a large range of behaviors across
domains (Poehlman, Uhlmann, Greenwald, &
Banaji, 2006). However, even though theoretical
models about the relative infl uence of implicit
and explicit measures on behavior exist, the
situational boundary conditions of these measures’
predictive validity have rarely been investigated.
The purpose of the present research is to em-
pirically test one key moderator of implicit and
explicit measures’ predictive validity, situation-
ally available control resources, in the domain
of interracial interaction.
The distinction between implicit and explicit
attitudes can be linked to a variety of dual-process
(e.g. Fazio & Towles-Schwen, 1999), or dual-
system theories (e.g. Strack & Deutsch, 2004)
that distinguish two modes or two systems of
information processing. These theories share
a conception that the two modes or systems
will be differentially infl uential on perception,
judgment, and action but differ somewhat with
regard to whether implicit and explicit attitudes
are seen as structurally different. For instance,
the MODE model (Fazio & Towles-Schwen,
1999) distinguishes between spontaneous and
deliberate processes by which an attitude may
infl uence judgments and behavior. Spontaneous
processes are understood as immediate reactions
toward an attitude object upon encountering
the object. In contrast, deliberate processes are
characterized by effortful considerations about
the pros and cons of a certain behavior. Whereas
implicit measures are assumed to tap into the
spontaneous processing stage, explicit measures
are assumed to usually refl ect more elaborate
judgments from the deliberate processing stage
(Fazio & Olson, 2003). Hence, the emphasis of
the MODE model is on implicit and explicit
measures tapping into different processing stages
of an underlying attitude. Regarding behavior
prediction, the model assumes that laborious
deliberate processes will only be infl uential in
overriding spontaneous processes and guiding
behavior if a person has suffi cient resources
available to engage in deliberate processing and
is motivated to do so. If a person does not have
suffi cient resources to deliberate or is not motiv-
ated to use them, spontaneous processes driven
by implicit attitudes should prevail.
In contrast to the MODE model, most other
dual-process theories propose that implicit and
explicit measures refl ect structurally distinct
mental representations (e.g. Smith & DeCoster,
2000; Strack & Deutsch, 2004; Wilson, Lindsey, &
Schooler, 2000). From the perspective of Strack
and Deutsch’s (2004) Refl ective-Impulsive Model
(RIM) of social behavior, for instance, implicit
attitudes are part of the impulsive system which
operates via associative processes of spreading
activation and elicits behavior by activating
relevant behavioral schemas. Explicit attitudes,
on the other hand, are part of the refl ective
system in which behavior is elicited as the result
of a higher-order decision or ‘reasoned action’
process (Strack & Deutsch, 2004). Importantly,
the refl ective system may fi nally execute behavior
by activating behavioral schemas relevant to the
intended action plan. Thus, even though both
systems operate in parallel and involve different
processes by which behavior is elicited, the
RIM assumes a fi nal, common pathway of be-
havioral execution: the activation of behavioral
schemas (see also Norman & Shallice, 1986).
Most important for the present considerations,
71
Hofmann et al. control resources and interracial interaction
the relative impact with which the two systems
affect behavior may depend on certain boundary
conditions or moderator variables. Since the
refl ective, but not the impulsive system, re-
quires substantial amounts of executive control
resources for effi cient functioning (Strack &
Deutsch, 2004), behavior should primarily
be elicited by the impulsive system when con-
trol resources are scarce, i.e. when behavioral
schemas are predominantly triggered by spon-
taneously activated attitudes. Conversely, the
refl ective system’s potential to generate action
plans and override behavioral schemas activated
by the impulsive system should increase when
suffi cient control resources are available and,
as a consequence, the relative impact of the
impulsive system on behavior should lessen.
Taken together, despite their differences on the
structural level, both the MODE model and the
RIM converge on the prediction that implicit
and explicit measures should have a differential
impact on behavior as a function of available
control resources.
Indirect evidence for these assumptions can be
derived from research showing the differential
validity of implicit and explicit measures in the
prediction of spontaneous versus controlled
behavior (e.g. Asendorpf, Banse, & Mücke, 2002;
Dovidio, Kawakami, & Gaertner, 2002; Dovidio,
Kawakami, Johnson, & Johnson, 1997; Fazio,
Jackson, Dunton, & Williams, 1995). These
studies demonstrated that implicit attitudes
(and personality traits; Asendorpf et al., 2002)
mainly predict spontaneous behavior, whereas
explicit attitudes mainly predict controlled
behavior. However, this line of research was
primarily concerned with the impact of implicit
and explicit attitudes on various behaviors that
are supposed to differ a priori with regard
to controllability. For example, in the Dovidio
et al. (1997) study on Whites’ interracial inter-
action behavior toward Blacks, the rate of eye
blinking and the percentage of visual contact
toward the Black interaction partner—both
considered nonverbal behaviors that are diffi cult
to control—were best predicted by Whites’
implicit attitudes as measured with a subliminal
priming paradigm. In contrast, participants’ self-
reported evaluations of their Black interaction
partner were best predicted by explicitly assessed
modern racism scores.
Although the distinction between spontan-
eous and controlled behaviors is parsimonious
and intuitively intriguing, such a clear-cut pat-
tern of results did not emerge in all studies
on differential predictive validity. For instance,
McConnell and Leibold (2001) reported
signifi cant positive correlations between implicit
attitudes toward Blacks as measured with the
Implicit Association Test (IAT; Greenwald,
McGhee, & Schwartz, 1998) and a variety of be-
haviors such as speaking time, social comments
and smiling that arguably contain controlled
elements as well. Likewise, a variety of behaviors
usually considered spontaneous such as for-
ward leaning, body openness, or seating distance,
were not reliably related to IAT scores. Findings
like these point to the possibility that behavior
determination by implicit and explicit attitudes
is more dynamic and complex than previously
assumed, and that the above fi ndings on pre-
dictive validity may be somewhat limited by the
fact that ‘there is not an established taxonomy
that identifi es deliberative from spontaneous
behaviors’ (Dovidio et al., 2002, p. 66). Most
important, as available control resources were
not manipulated or assessed in the above line of
research, these studies provide no direct evidence
for the hypothesis that a given behavior may
be infl uenced to different degrees by implicit
and explicit attitudes depending on available
control resources.
The main goal of the present research was to
test this central assumption of dual-process or
dual-system theories in two studies on interracial
interaction behavior of Italians toward Africans
(Study 1) and Germans toward Turks (Study 2).
In each study, we fi rst assessed implicit and ex-
plicit attitudes toward the outgroup by using
an IAT (Greenwald, et al., 1998) as an implicit
measure, and the Blatant and Subtle Prejudice
Scale (BSPS; Pettigrew & Meertens, 1995) as an
explicit measure. Then we created conversations
around race-neutral topics closely following the
Dovidio et al. (1997, 2002) paradigm in which
participants interacted both with an outgroup
and an ingroup conversation partner. As in
those studies, behavior toward the ingroup
Group Processes & Intergroup Relations 11(1)
72
partner was used as a baseline in order to con-
trol for individual differences in interaction
styles. Situationally available control resources
were manipulated as a within-subjects factor by
imposing a memory task (e.g. Reisberg, 1983)
on participants during half of the interview
questions. The interaction sequences were
videotaped and later coded on a variety of be-
havioral measures adapted from prior research
on interpersonal interaction (Asendorpf
et al., 2002; Dovidio et al., 1997; McConnell &
Leibold, 2001).
Our primary interest was in the prediction of
racial biases, i.e. differences in interaction be-
havior toward the outgroup partner relative to
the ingroup partner. In accordance with a dual-
process or dual-system framework, we expected
that the IAT would be a stronger predictor of
behavior in the memory load condition and have
a weaker impact in the high resources condition.
Conversely, we expected that the BSPS would
yield higher predictive validity when full control
resources were available.
Study 1
Method
Participants A total of 86 (72 female, 14 male)
fi rst-year psychology students from an introduc-
tory psychology course at the Department of
Psychology in Padova, Italy, participated in ex-
change for course credit. Gender proportions
mirror the high number of females registered
in psychology at the University of Padova. The
age of participants ranged between 18 and
49 years (M = 21.00, SD = 4.34). For one partici-
pant, the video camera did not register behavior
due to a technical problem.
Materials and procedure The study consisted
of two supposedly unrelated parts. Participants
were informed that they would take part in two
different experiments that contained tasks often
used in social psychological research and were
going to be run in two separate laboratories
by different experimenters. In the fi rst part,
introduced as a computer experiment on per-
ception and classifi cation, participants completed
the implicit and explicit attitude measures. In the
second part, introduced as an interview practice
session, participants’ interracial behavior toward
African males was observed. We used only male
confederates and male stimuli for the attitude
measures because prejudice against males has
been found to be stronger and more pervasive
(Eagly & Kite, 1987), and therefore we reasoned
that using males would increase the power of
the test of our hypotheses.
Implicit attitude measures Implicit attitudes were
assessed with a Race IAT, using facial images
of eight African and eight Italian adult males
as target stimuli and eight positive and eight
negative affective nouns as attribute stimuli (see
Appendix). Images were 90 × 128 pixel color
photos of African and Italian male young adults
taken from different web pages. Following the
Race IAT, a control Flower–Insect IAT (Greenwald
et al., 1998) was administered, using eight
pictures of fl owers and insects, respectively, and
the same affective nouns as for the Race IAT. The
control IAT was included in order to evaluate
the specifi city of the Race IAT with regard to
the prediction of interracial behavior.
Each IAT consisted of seven blocks in a fi xed
sequence: in the fi rst block, participants classifi ed
target stimuli into the categories ‘African’ and
‘Italian’ or ‘fl owers’ and ‘insects’, respectively.
In the second block, participants classified
attribute stimuli into the categories ‘positive’
and ‘negative’. In the third practice and the
fourth test block, targets and attributes had to
be classifi ed simultaneously such that African
faces (or insects, respectively) and positive nouns
were assigned to the same key. In the fi fth block,
the key assignment for the attribute dimension
was reversed. In the sixth practice and seventh
test block, African faces (insects) and negative
nouns were assigned to the same key. As we were
interested in individual differences in implicit
attitudes, order of combined blocks and key
assignment were held constant for both IATs
(Banse, Seise, & Zerbes, 2001).
The practice and test blocks had 32 trials each,
such that each stimulus was presented once.
Target and attribute stimuli were presented
randomly in an alternate fashion without replace-
ment. IAT scores were computed according
73
Hofmann et al. control resources and interracial interaction
to the improved scoring algorithm proposed
by Greenwald, Nosek, and Banaji (2003). In
order to estimate the reliability of the IAT,
we created four mutually exclusive subsets of all
combined-task trials via an odd–even split and
calculated IAT scores separately for each subset.
The internal consistency of subsets amounted
to .90 for the Race IAT and .88 for the Flower–
Insect IAT (see Table 1).
Explicit attitude measure An Italian translation of
the BSPS (Pettigrew & Meertens, 1995) by Arcuri
and Boca (1996) assessed participants’ explicit
attitudes toward Africans on 5-point rating scales.
Since all items loaded on a single common
factor (λ = 5.50) as indicated by a screeplot and
since the BSPS subscales were highly correlated
(r(84) = .69, p < .001), we combined all 20 items
into a single scale score (BSPS).
1
One item (‘If
Africans would only try harder, they could be
as well off as Italian people’) showed a very low
item-total correlation of .08 and was therefore
removed from the scale (α = .85).
Interracial interaction After completing the fi rst
part of the study, participants were met by a second
female experimenter and escorted to another
room with two chairs in it at a standardized dis-
tance of 9 feet (2.8m).
2
Participants were seated
on the chair in the back of the room, facing the
side of the entrance door. A video camera was
situated behind the interviewer’s chair, facing
the participant’s chair, and another camera was
situated behind the participant’s chair directed
toward the interviewer’s chair.
Participants were led to believe that—due
to a collaborative effort of the Psychology and
Sociology Departments, sociology students from
a research methods course were given the pos-
sibility to gain experience in interviewing. For
this reason, two students from this course would
be asking two short questions each about topics
related to studying and living in Padova. A sample
question was given together with the reassurance
that questions would be easy to answer and not
too intimate. Furthermore, participants were
given the following information:
From the view of a social psychologist, one interesting
research question is how an interview is infl uenced
by ongoing mental occupation: in order to study
these effects, you will be given a word list shortly
before some of the interview questions are asked.
Your task will be to read this list and to think of
these words during the interview in order to be
able to recall as many words as possible after the
interview question has been answered.
In sum, the cover story allowed us to confront
participants with two confederates, one African
and one Italian, and to impose half of the
Table 1. Means, standard deviations, reliabilities, and intercorrelations for implicit and explicit measures in
Study 1 and Study 2
(1) (2) (3)
Study 1
(1) RACE IAT (.90)
(2) BSPS .34** (.85)
(3) Flower–Insect IAT –.07 .04 (.88)
Mean –0.62 3.72 –.39
SD 0.48 0.52 0.47
Study 2
(1) RACE IAT (.81)
(2) BSPS .00 (.84)
Mean –0.33 3.11
SD 0.26 0.58
*p < .05; **p < .01.
Notes: N = 86 in Study 1 and N = 77 in Study 2. Reliabilities (α) are in parentheses in the main diagonals. IAT =
Implicit Association Test; BSPS = Blatant Subtle Prejudice Scale.
Group Processes & Intergroup Relations 11(1)
74
interview questions with a memory task aimed
at reducing available control resources.
After the experimenter had given the instruc-
tions and obtained participants’ consent that
their responses may be videotaped, she started
both video cameras and the first interview
sequence began.
3
In the memory load condition
(see below for details), participants were fi rst
presented with a list of words to remember. In the
case of no memory task (full resources condition),
the experimenter immediately moved on to the
next point in the script.
The experimenter then left the room, and
after a brief pause the fi rst confederate entered
the room, introduced himself, sat down on the
confederate chair and asked the fi rst interview
question. Confederates were instructed to let
participants respond without interruption
until they indicated the end of the answer. If
necessary, the confederate asked one or more
follow-up question(s) until approximately two
minutes had passed. At the end of the interview
sequence, the confederate left the room and
the experimenter entered again.
In the case of a preceding memory task, the
experimenter handed participants a blank sheet
and asked them to recall and write down as many
words as possible from the list within one minute.
Then, in all cases, participants were asked to rate
the diffi culty of answering the interview question
and the interviewer’s competence in asking the
question on 7-point scales.
4
After the questionnaire, the next interview
sequence began until all four sequences had
been fi nished. Questions 1 and 2 were asked by
the fi rst confederate, questions 3 and 4 by the
second confederate. The order of confederate
race (African vs. Italian) was counterbalanced.
There was one memory task per confederate,
and the position of the task—at each inter-
viewer’s fi rst or second question—was balanced
across confederates. At the end of the study,
participants were given course credit. A detailed
debriefi ng took place in front of the class after
data collection was over.
Manipulation of control resources A memory
task was chosen to reduce available control re-
sources instead of other possible dual-task
paradigms frequently employed such as tracking
a point of light, because we reasoned that mere
mental occupation would interfere least with
the interracial interaction behavior displayed
concurrently by our participants. In the mem-
ory load condition, participants were given one
minute to read and remember as many words
as possible from one of two parallel lists of 20
words taken from Roediger and McDermott
(1995). There was one list per confederate and
the order of lists was held constant. In the full
resources condition, participants were not
confronted with a memory task. In order to
provide an achievement incentive for the
memory task, participants were told that they
would be receiving sweets at the end of the
interview session (which they actually did)
and that the amount received would depend
on the number of words recalled. We did not
inform participants about the exact minimum
number of words that had to be remembered
for receiving sweets because we wanted to avoid
setting a fi xed goal standard that may have given
rise to success or failure experiences during the
interview sequences.
Confederates The team of confederates consisted
of two male African and three male Italian
confederates who were hired for the duration of
the experiment. Confederates were comparable
in size, age, attractiveness, clothing, and did
not wear glasses. They underwent practice in
order to ask the interview questions in a com-
parable way and to behave similarly toward the
participants. Confederates were unaware of
the hypotheses of the study, memory load con-
dition, and participants’ implicit and explicit
attitude scores.
Interview questions We used four easy and
familiar questions about student life: (a) advant-
ages and disadvantages of eating in the university
dining hall vs. eating at home, (b) advantages
and disadvantages of living with one’s parents
vs. living alone, (c) consequences of crowded
seminars and lectures on the quality of teaching,
and (d) an evaluation of cultural life in Padova.
The order of questions was balanced such that
each question was asked equally often at the
75
Hofmann et al. control resources and interracial interaction
fi rst, second, third, and fourth position of the
interview session. Furthermore, question order
was balanced with regard to memory load
condition and confederate race.
Codings of behavior Participants’ digitized inter-
action behavior from the fi rst 90 after a question
was asked was coded by one trained coder who
was unaware of experimental conditions and
study hypotheses. Coding reliability was as-
sessed via independent coding of 80 randomly
selected interview sequences by a second coder.
Videos were watched once without coding fi rst.
Then coders fi lled out a global rating of the
participants’ behavior consisting of fi ve bipolar
adjective ratings on a 7-point scale (from –3
to +3) with a midpoint of zero. The positive and
negative poles were: polite–impolite, talkative–
quiet, pleasant–unpleasant, relaxed–nervous, and
warm–cold. All items loaded on a single factor
(explaining 49% of the variance) as indicated by
a screeplot. For this reason, we combined these
ratings into a single global rating score. The
interrater correlation of this measure was .75.
In a separate run, the frequencies of speech
illustrators (i.e. hand or arm movements which
accompany speech and which are used to em-
phasize what is being said) and body adaptors
(i.e. touches of the own body or head) were
coded with the help of a coding sheet following
the recommendations by Ekman and Friesen
(1972). The interrater correlations of these
ratings were .78, and .84 respectively. Visual con-
tact was timed in a separate run with the help
of a behavioral observation computer program.
Coders pressed a key on the keyboard whenever.
The participant gazed on the confederate’s
face while answering the question and released
the key whenever he or she gazed away from
the confederate’s face. The video recording
did not make it possible to distinguish between
face-directed gazing and eye contact. Because
participants usually looked at the confederate
whenever the latter asked a follow-up question,
we additionally recorded the time that the con-
federate spent talking and did not record visual
contact during these instances. The confederate
talking time was then subtracted from the total
interaction time (90 s) in order to arrive at
participant answering time. The percentage of
visual contact was computed as the proportion
of answering time participants spent gazing
at the confederate, multiplied by 100 (Dovidio
et al., 1997). Visual contact codings of both
coders were correlated .81.
Results
Implicit and explicit measures The means and
standard deviations of IAT and BSPS scores as
well as their intercorrelations are displayed in
the upper part of Table 1. Attitude constructs
were scaled such that higher values indicate a
more positive attitude toward Africans. The
Flower–Insect IAT was scaled such that higher
values indicate more positive attitudes toward
insects. For both the Race and the Flower–
Insects IAT, negative mean effects emerged,
indicating that participants responded faster
when African faces and negative words shared
the same response key, and when insects and
negative words shared the same key, respect-
ively. Both IATs were almost independent from
each other (r(84) = –.07, p = .50), hence, they
did not seem to share common method-specifi c
sources of variance (but see Mierke & Klauer,
2003). Establishing convergent and discriminant
validity of our attitude measures, the Race
IAT was signifi cantly correlated with the BSPS
(r(84) = .34, p < .001), but the Flower–Insect IAT
was not (r(84) = .04, p = .82).
Memory task For each participant, we counted
the number of words correctly recalled for
the fi rst and second word list. There was no
difference in memory recall between the lists
(M = 8.91, SD = 3.05; M = 8.78, SD = 2.98;
t(85) = .47, p = .64). Furthermore, an analysis
of variance (ANOVA) with memory recall as the
dependent variable revealed no signifi cant main
effects or interaction as a function of position
of the memory task (fi rst or second position
within confederate) and confederate race (all
Fs < 1), indicating that memory performance
was not affected by position and by whether
participants interacted with an Italian or African
confederate.
Group Processes & Intergroup Relations 11(1)
76
Interviewer competence There was no sig-
nifi cant difference in participants’ competence
ratings of the two African (F(1, 84) = 2.07, p = .15),
and three Italian confederates (F(2, 83) < 1),
suggesting that confederates’ interview be-
havior was comparable. On average, African
confederates were rated as competent as Italian
confederates (M = 4.97, SD = 1.40 vs. M = 5.06,
SD = 1.54; t(85) = –.51, p = .61).
Question diffi culty In order to check whether
interview questions were equally diffi cult, we
performed a repeated measures ANOVA on
diffi culty ratings with question as the within-
subjects factor. This analysis revealed a signifi cant
main effect (F(3, 246) = 5.07, p = .002). Simple
contrasts indicated that the question about
cultural life (M = 3.15) differed signifi cantly
(p < .05) from the dining hall (M = 2.48), living
(M = 2.52), and quality of teaching (M = 2.76)
questions, which did not differ signifi cantly
from each other. Thus, one caveat for the pre-
sent analyses is that differences in question dif-
fi culty might have affected control resources
in addition to our experimental memory task
manipulation. However, as question order was
orthogonal to position, memory load condition,
and confederate race, these differences should
not have affected the results systematically.
Behavioral observations Table 2 (upper part)
summarizes the means and standard deviations
for the behavioral observations separately by
condition and confederate race. For each be-
havior, simple contrasts were computed in order
to detect signifi cant differences among means.
These analyses revealed that participants gazed
signifi cantly less toward the African confederate
in the memory load condition compared with
all other cells. Furthermore, participants used
signifi cantly more speech illustrators toward
Africans when resources were available than they
did toward Italians when resources were scarce.
No further reliable mean differences emerged
for body adaptors and the global rating.
Following Dovidio et al. (1997) we computed
relative difference scores for each behavior by
subtracting the behavior displayed toward the
Italian confederate (baseline) from the behavior
displayed toward the African confederate
separate for the memory load and full resources
condition. Hence, a positive difference indicates
that a participant displayed a given behavior
more frequently toward the African relative to
the Italian confederate.
Do implicit and explicit attitudes predict
behavior differently depending on control
resources? In order to test our main hypothesis
Table 2 . Means and standard deviations (in parentheses) for behavioral observations in Study 1 and Study 2
Memory load condition Full resources condition
Study 1: African Italian African Italian
Behavioral observations confederate confederate confederate confederate
Visual contact 54.13
a
(17.56) 57.81
b
(17.02) 57.07
b
(15.36) 57.10
b
(15.51)
Speech illustrators 4.34
ab
(1.83) 4.15
a
(1.91) 4.65
b
(1.75) 4.49
ab
(2.08)
Body adaptors 1.28
a
(1.57) 1.55
a
(1.69) 1.30
a
(1.46) 1.60
a
(1.71)
Global rating 0.66
a
(0.59) 0.65
a
(0.68) 0.65
a
(0.68) 0.55
a
(0.59)
Study 2: Turkish German Turkish German
Behavioral observations confederate confederate confederate confederate
Visual contact 50.66
ab
(17.29) 52.34
ab
(18.31) 53.21
a
(17.67) 49.44
b
(16.74)
Speech illustrators 5.38
ab
(5.28) 4.62
a
(4.47) 7.34
c
(5.86) 5.77
b
(5.02)
Body adaptors 0.76
a
(1.02) 0.94
a
(1.52) 0.91
a
(1.26) 0.81
a
(1.33)
Global rating 0.58
a
(0.77) 0.59
a
(0.75) 0.73
b
(0.73) 0.74
b
(0.67)
Note: Different subscripts indicate signifi cant differences (p < .05) within rows.
77
Hofmann et al. control resources and interracial interaction
that the predictive validity of implicit and explicit
attitudes differs as a function of available con-
trol resources, we performed a series of repeated
measures regression analyses on the various
behavioral indicators, using the General Linear
Model procedure for repeated measures in SPSS.
In each of these equations, behavior in the mem-
ory load and in the full resources condition was
the dependent variable. The continuous IAT
and BSPS scores were included as simultaneous
predictors. The model also specifi ed the inter-
action between the within-subjects factor Con-
dition (memory load vs. full resources) and
the predictors. This procedure allowed us to
estimate the standardized regression weights
of the IAT and BSPS separately for the memory
load and full resources conditions and to test
whether regression weights differed signifi cantly
as a function of condition. According to our
hypotheses and based on how variables were
coded, we expected a negative interaction ef-
fect for the IAT (i.e. a reduction of predictive
validity in the full resources as compared with
the memory load condition) and, conversely,
a positive interaction effect for the BSPS. The
results of the regression analyses for Study 1 are
reported in the upper part of Table 3.
Visual contact As expected, the IAT was a sig-
nifi cant predictor of visual contact in the memory
load condition, indicating that—among taxed
participants, those harboring positive implicit
attitudes toward Africans spent relatively more
time gazing at the African interviewer than those
harboring negative implicit attitudes. In the
full resources condition, however, the regression
weight of the IAT was negative and close to zero,
consistent with our expectations. Moreover, the
interaction of IAT × Condition was signifi cantly
negative (F(1, 82) = 5.02, p = .028), indicating that
regression slopes for the IAT were signifi cantly
smaller in the full resources as compared with
the memory load condition. Conversely, the
weight of the explicit measure was negative (but
not signifi cant) in the memory load condition
and changed to positive (but not signifi cant) in
the full resources condition. As with the implicit
measure, the difference between conditions was
reliable as indicated by the BSPS × Condition
interaction (F(1, 82) = 4.49, p = .037).
Speech illustrators The IAT signifi cantly pre-
dicted the frequency of speech illustrators
under memory load, indicating that positive im-
plicit attitudes were associated with more
frequent use of speech illustrators to emphasize
communication, but did not predict speech
illustrator behavior under full resources. The
difference between conditions was statistically
signifi cant (F(1, 82) = 6.87, p = .010). However,
no simple main or moderator effect was obtained
for the explicit measure.
Body adaptors Contrary to our expectations, the
Race IAT failed to predict body adaptors reli-
ably in the memory load condition. As expected,
BSPS scores predicted the frequency of body
adaptors only when resources were available
but not when resources were restrained by the
memory task. The positive direction of the ef-
fect suggests that it may be best interpreted as
indicating that participants with positive explicit
attitudes felt more comfortable and therefore
touched themselves more often in the presence
of the African confederate (but see Asendorpf
et al., 2002, for a different interpretation of body
adaptors in the context of shyness).
Interviewer competence ratings Although it is not
a behavioral measure in the strict sense, we also
predicted participants’ interviewer competence
ratings (computed as a relative difference score)
by implicit and explicit attitudes (see Table 3).
For both the memory load and the full resources
condition, this judgment was signifi cantly related
only to the explicit measure but not to the im-
plicit measure, indicating that participants with
positive explicit attitudes perceived the African
confederate as more competent in relation to the
Italian confederate than participants harboring
negative explicit attitudes. This fi nding is con-
sistent with our reasoning as the competence
rating was always administered after the memory
task had already been completed, i.e. at a time
Group Processes & Intergroup Relations 11(1)
78
where resources were no longer constrained by
our memory load manipulation.
Control IAT In order to examine the discriminant
validity of the Race IAT and the Flower–Insect
control IAT in predicting interracial interaction,
we entered the Flower–Insect IAT instead of the
Race IAT as the implicit predictor in the regres-
sions above. In line with our expectation that
predictive validity was restricted to the Race
IAT as a thematically corresponding measure,
the noncorresponding Flower–Insect IAT did
not signifi cantly predict any of the behavioral
indicators in either condition, with all regression
coeffi cients ranging between β = –.15, p = .18
and β = .04, p = .70.
Discussion
Study 1 provides initial evidence for the notion
that implicit and explicit attitudes predict inter-
racial interaction behavior differently depending
on situationally available cognitive resources.
Specifi cally, implicit attitudes as assessed with
a Race IAT were reliably related to the amount
Table 3. Prediction of behavioral observations and interviewer competence ratings by implicit (IAT) and
explicit (BSPS) racial attitude measures in Study 1 and Study 2 and for the combined analysis
Predictor
IAT BSPS
β
IAT
β
IAT
η² β
BSPS
β
BSPS
η²
memory full β
IAT
× cond. memory full β
BSPS
× cond.
load resources load resources
Study 1
Behavioral observations
Visual contact .24* –.09 .062 –.18 .12 .053
Speech illustrators .35** –.09 .074 –.10 .00 .004
Body adaptors –.05 .02 .002 .11 .30* .016
Global rating .04 .10 .003 –.06 .10 .017
Interviewer competence rating –.01 .01 .000 .27* .30* .001
Study 2
Behavioral observations
Visual contact .32** –.07 .094 –.03 .14 .014
Speech illustrators .31** –.06 .084 –.09 –.03 .003
Body adaptors .36** –.15 .156 .01 .01 .000
Global rating .06 .13 .003 .04 –.01 .002
Interviewer competence rating –.06 –.06 .000 .15 .29* .018
Combined analysis
Behavioral observations
Visual contact .26** –.08 .063 –.11 .13 .032
Speech illustrators .33** –.07 .083 –.09 –.02 .003
Body adaptors .16* –.04 .020 .02 .17* .011
Global rating .04 .12 .005 –.01 .04 .002
Interviewer competence rating –.03 –.03 .000 .21** .30** .007
Global behavioral indicator .31** –.02 .072 –.08 .17* .041
*p < .05; **p < .01.
Notes: All regression estimates are standardized regression weights. Signifi cant differences in IAT (BSPS) regression
weights between the memory load and full resources condition are indicated in bold. Effect size η² (partial eta
squared) for each predictor × condition interaction is provided in order to indicate the percentage of variance
in behavioral indicators explained by this interaction. IAT = Implicit Association Test; BSPS = Blatant Subtle
Prejudice Scale; cond. = condition (memory load vs. full resources).
79
Hofmann et al. control resources and interracial interaction
of visual contact when participants had to per-
form a memory task while interacting with an
outgroup member, but unrelated to gazing be-
havior when participants were not taxed with
an additional task. Conversely, explicit attitudes
were signifi cantly more positively associated with
visual contact in the full resources condition as
compared to the memory load condition. Partial
support was obtained with regard to speech
illustrators and body adaptors. However, for
these dimensions only the implicit or the explicit
measure proved sensitive to our manipulation of
available control resources. No predictive validity
at all was obtained for judges’ global ratings
of interaction behavior, suggesting that global
impressions may be less valid than our specifi c,
event-oriented codings. Furthermore, the com-
parison with the Flower–Insect control IAT
showed that the predictive validities we obtained
were specifi c to the Race IAT, corroborating the
often neglected discriminant validity of implicit
attitude measures (Gawronski, 2002).
Consistent with our hypotheses, explicit but
not implicit attitudes were associated with higher
competence ratings of the African interviewer,
independent of condition. As these ratings were
always collected after the resource manipulation
was fi nished, this result lends additional support
for the hypothesis that explicit measures are
better predictors of behavior when available
control resources are high rather than low.
Even though there is initial support for our
main predictions, the fi ndings from Study 1 may
have been limited by three possible shortcomings
of our design: fi rst, our manipulation of available
control resources may have been rather weak.
A second possible shortcoming is that the close
temporal and local proximity of the attitude as-
sessment and the behavioral observation phase
may have made at least some of our participants
suspicious of our true study goals, leading to
differences in the motivation to control inter-
racial interaction behavior. And fi nally, one of
the questions asked was significantly more
diffi cult than the other ones, which may have
introduced additional variation in available con-
trol resources in those cases over and above our
situational manipulation.
Study 2
We conducted a second study that was intended
as a conceptual replication of Study 1 with
the following refi nements and modifi cations.
First, we used separate measurement occasions
in order to minimize potential transfer effects
between predictor assessment and behavioral
observation. Second, we attempted to strengthen
our experimental manipulation by having par-
ticipants perform an additional memory retrieval
task in the memory load condition and by grant-
ing them time to prepare their answers in the
full resources condition. Third, we carefully
selected interview questions on the basis of pilot
ratings in order to assure comparable question
diffi culty. Fourth, we switched to attitudes of
Germans toward Turks as the content domain.
In Germany, Turks constitute the largest of
all foreign groups and range among the least
accepted ones (Wagner, van Dick, & Zick, 2001),
making interracial interaction between Germans
and Turks an appropriate content domain to
identify the generalizability of our fi ndings.
Method
Participants Seventy-seven students (27 male,
50 female) of German nationality from the
University of Koblenz-Landau, between 19 and
49 years of age (M = 22.96 years, SD = 5.59),
participated in a longitudinal study either in
exchange for course credit or 10 euros. Students
were recruited from fi rst year psychology classes
(70%) or from other departments (30%).
Materials and procedure The study consisted
of two parts. In the fi rst session, participants
completed the implicit and explicit attitude
measures in this order. At least two weeks later,
participants interacted with male Turkish and
German confederates in an ostensibly unrelated
interview practice session in a different location
on campus. We used the same cover story and
identical procedure as in Study 1, with the fol-
lowing modifi cations.
Implicit attitude measure A Race IAT was admin-
istered, using eight German and eight Turkish
looking facial stimuli as target stimuli and eight
Group Processes & Intergroup Relations 11(1)
80
positive and eight negative affective nouns as
attribute stimuli (see Appendix). Target pictures
were selected from pretested material used
by Neumann and Seibt (2001). The number of
practice and test blocks was increased to 64 trials
each, such that each stimulus was presented
twice per block.
Explicit attitude measure A German translation
of the BSPS by Zick (1997) was used in order to
measure participants’ explicit attitudes toward
Turks. Since all items loaded on a single main
factor (λ = 4.88) as indicated by a screeplot and
both subscales were again highly correlated
(r(75) = .60, p < .001), we computed a single
BSPS index (α = .84).
Manipulation of control resources In the memory
load condition, participants were taxed even
more strongly than in Study 1 by having to gen-
erate as many words as possible starting with a
‘Y’ or ‘X’, in addition to the memory task adapted
from Study 1. In German there are relatively few
words starting with these letters, which should
have led participants to continue ruminating
during the interview instead of being content
with the number of generated words too quickly.
In the full resources condition, the experimenter
handed participants a memo with the interview
question wording. Participants were then granted
one minute of preparation time for the interview
to follow. We reasoned that being able to prepare
one’s answer should free up additional control
resources during the interview.
Confederates The team of confederates con-
sisted of fi ve male German and three male
Turkish confederates. The Turkish and German
confederates introduced themselves with
a stereotypic Turkish (‘Murat’) or German
(‘Michael’) name, respectively, in order to ensure
that participants categorized them according to
their ethnic background.
Interview questions We pretested (N = 11) a
variety of questions about campus life in terms
of familiarity and diffi culty, and selected the
four questions that did not differ signifi cantly
on these dimensions (both Fs < 1): (a) quality
and service in the dining hall, (b) equipment
of the university’s computer pool, (c) services
and equipment of the library, and (d) quality
and equipment of the lecture halls and seminar
rooms.
Codings of behavior The videotaped behavior
of all participants in the fi rst 90 of their re-
sponse was coded twice by a total set of four
trained coders. For the analyses to follow, we
averaged together both codings for each be-
havioral indicator. As in Study 1, we coded
for visual contact (intraclass r = .86 between
coders), speech illustrators (r = .89), and body
adaptors (r = .77). For the global rating, the
same fi ve bipolar adjective ratings as in Study 1
were combined into a single global judgment
(intraclass r = .70).
Debriefi ng In order to ensure that during data
collection all participants were naive to the true
study purpose, we fully debriefed all participants
via telephone after the data had been collected
and gave them the opportunity to learn about
their personal implicit and explicit attitude
scores in our lab which most participants (75%)
utilized.
Results
Implicit and explicit measures All attitude con-
structs were scaled such that more positive values
indicate a more positive attitude toward Turks. As
can be seen from the lower part of Table 1, similar
to Study 1, a negative mean effect emerged on
the implicit measure, indicating that on average
participants implicitly preferred German faces
over Turkish faces. Unlike in Study 1, however,
implicit and explicit attitudes were independent
from each other (r(75) = .004, p = .98).
Memory-task, interviewer competence, and
question diffi culty An ANOVA with memory
recall (combined for the memory and the word
generation task) as the dependent variable
revealed no signifi cant main effects or interaction
as a function of the position of the memory task
and confederate race.
As in Study 1, there was no reliable difference
in competence ratings among the Turkish or
81
Hofmann et al. control resources and interracial interaction
German confederates. Furthermore, participants
rated the outgroup and ingroup interviewers
as equally competent on average (M = 4.58,
SD = .90 vs. M = 4.52, SD = .84; t(76) = .70,
p = .48). Regarding question diffi culty, a re-
peated measures ANOVA yielded no signifi cant
main effect (F(3, 225) < 1), indicating that
participants regarded the four questions as
equally diffi cult.
Behavioral observations Table 2 (lower part)
summarizes the means and standard deviations
for the behavioral observations separately by
condition and confederate race. As can be seen
from a comparison of means, participants gazed
slightly less toward the Turkish confederate in
relation to the German confederate when put
under memory load, but this difference did
not reach statistical signifi cance. However, par-
ticipants spent signifi cantly more time gazing at
the Turkish relative to the German confederate
in the full resources condition. Furthermore,
participants used signifi cantly more speech
illustrators with regard to the Turkish interviewer
when cognitively untaxed rather than taxed.
Finally, the global rating was more positive with
regard to both confederates in the full resources
condition.
Behavior prediction by implicit and explicit
attitude measures As in Study 1, we computed
relative difference scores (Turkish minus German
confederate) in order to control for individual
differences in interaction styles. In order to
investigate differential predictive validities,
multiple repeated measures regression analyses
were performed on behavioral indicators with
IAT and BSPS scores as predictors. Again, inter-
actions among condition and predictors were
specifi ed. Consistent with Study 1, the IAT sig-
nifi cantly positively predicted the percentage
of visual contact in the memory load condition
only but not in the full resources condition, and
this difference was signifi cant (F(1, 73) = 5.79,
p = .017) (see middle part of Table 3). The same
pattern emerged with regard to the prediction
of speech illustrators by our implicit measure
(F(1, 73) = 6.97, p = .010). In addition to Study 1,
the implicit measure’s predictive validity with
regard to the frequency of body adaptors was
also moderated by available control resources
(F(1, 73) = 12.06, p = .001), such that predictive
validity was high in the memory load condition
and low in the full resources condition (see
middle part of Table 3). Conversely, explicit
attitudes toward Turks were related to interviewer
competence ratings only when resources were
available but,unlike in Study 1, not under mem-
ory load. As in Study 1, no signifi cant regression
weights were obtained for the prediction of the
global interaction ratings.
Discussion
Study 2 lends additional support to our hypothesis
that the predictive validity of implicit and
explicit attitudes is moderated by situation-
ally available control resources. Concerning our
main hypothesis, as in Study 1, the percentage
of visual contact and the frequency of speech
illustrators were best predicted by the Race IAT
when participants’ control resources were limited
by a memory task. In addition, the frequency of
body adaptors was substantially predicted by the
implicit measure under memory load. Similar to
Study 1, the positive direction of the effect sug-
gests that a higher frequency of body adaptors
may be interpreted as a sign of comfort and a
relaxed manner rather than as a sign of anxiety
or shyness in interracial interaction. Again, the
global ratings were reliably associated with
neither the implicit nor the explicit measure.
Notwithstanding the many convergent results
across Studies 1 and 2, there are some differences
in fi ndings that deserve special attention. First,
the correlation between implicit and explicit
attitude measures was signifi cantly positive in
Study 1 but zero in Study 2, and the difference
between these two correlations was reliable
(z = 2.21, p = .013). One potential explanation
for this discrepancy is that German participants
may have been less willing to truthfully report
on their automatic reactions toward outgroup
members in their explicit self-reports than
Italians. A second possible explanation is that, in
comparison with the Italian–African intergroup
setting, German participants may harbor more
Group Processes & Intergroup Relations 11(1)
82
cognitively elaborated explicit attitudes about
Turks that correspond less well with more
immediate affective reactions as refl ected in the
IAT (e.g. Hofmann, Gawronski et al., 2005).
Second, there was a different pattern of mean
effects for the behavioral dimension of eye
gaze between the two studies: whereas partici-
pants in Study 1 spent less time gazing at the
outgroup confederate as compared to the
ingroup confederate under memory load, this
difference was not signifi cant—albeit in the right
direction—in Study 2. Instead, participants in
Study 2 spent signifi cantly more time gazing at
the outgroup confederate when they had full
resources available. This fi nding suggests that
under full resources, German participants may
have tried to convey a particularly good im-
pression by allocating even more attention to
the outgroup member, and that this favorable
treatment of the outgroup member vanished
when resources were situationally reduced.
Third, in order to identify reliable differences
with regard to our main hypothesis of differential
predictive validity, we conducted a combined
analysis of both studies in which we predicted the
behavioral dimensions from implicit and explicit
attitudes as a function of control resources (see
next section for details). Including the Study
factor (Study 1 vs. Study 2) as an additional factor
in these analyses showed that (a) Study did not
moderate the BSPS × Condition interaction and
(b) Study did not moderate the IAT × Condition
interaction except in the case of body adaptors.
This result refl ects the fact that body adaptors
were not reliably predicted by the IAT regardless
of Condition in Study 1, whereas in Study 2
body adaptors were signifi cantly predicted by
the IAT, but only when control resources were
scarce (see Table 3). Although we cannot offer
a good explanation for this difference, this fi nd-
ing may indicate a cultural difference such that
body adaptors may be more strongly guided by
automatic processes in German participants,
whereas body adaptors may be used somewhat
more strategically as a means of communication
by Italians (as also suggested by the signifi cant
relationship between body adaptors and the BSPS
in Italians in the full resources condition).
Combined analysis of Study 1 and
Study 2
The high correspondence in design and proced-
ure of both studies enables a joint analysis of our
data. Such an analysis has several advantages.
First, the accumulation of data across studies
strengthens the robustness and, as a conse-
quence, the generalizability of fi ndings. Second,
the power to detect meaningful systematic
variation is improved. Before combining the
data from both studies, we z-transformed all
behavioral indices and predictor measures in
order to equalize the metric of the variables
before pooling the data. We then performed the
above regression analyses for the pooled datasets.
As can be seen from the lower part of Table 3,
this analysis confi rmed the interaction of im-
plicit and explicit measures with available con-
trol resources for visual contact, and also the
interaction effect of the implicit measure with
regard to speech illustrators. Moreover, the
combined estimates across both studies indicated
that the IAT predicts the frequency of body
adaptors only under memory load, whereas the
BSPS scale does so only when suffi cient resources
are available (it is important to keep in mind,
however, that this is the only fi nding that should
be generalized with caution). The global rating
did not appear to be predictable by implicit
and explicit measures across both studies. In
line with our reasoning above, the interviewer
competence rating was only predicted by the
explicit measure, regardless of condition.
Finally, we combined all four behavioral
observations into a broad global behavioral
indicator and subjected it to a repeated measures
regression analysis with IAT and BSPS scores
as predictors. Overall, the implicit measure
predicted the global behavioral indicator sig-
nifi cantly in the memory load condition (β = .31,
p < .001), but not in the full resources condition
(β = –.02, p = .77) (see Table 3, bottom row).
Conversely, the explicit measure predicted the
global indicator in the full resources (β = .17,
p = .04) but not in the memory load condition
(β = –.08, p = .30). Most important, both the
IAT × Condition interaction (F(1, 158) = 12.33,
83
Hofmann et al. control resources and interracial interaction
p = .001), as well as the BSPS × Condition
interaction (F(1, 158) = 6.70, p = .011), were reli-
able. Thus, our predictions were supported on
the level of the global behavioral indicator.
General discussion
The present fi ndings on interracial interaction
extend previous research on the differential
predictive validity of implicit and explicit
attitudes (Dovidio et al., 1997, 2002; Fazio et al.,
1995; McConnell & Leibold, 2001) by inves-
tigating the conditions under which both types
of attitudes best predict behavior. Drawing on
dual-process models such as the MODE model
(Fazio & Towles-Schwen, 1999) or dual-system
models such as the Refl ective-Impulsive Model
(Strack & Deutsch, 2004), we expected the
relative behavioral impact of implicit attitudes
to be stronger when situationally available con-
trol resources are low rather than high. Con-
versely, we expected the relative infl uence of
explicit attitudes on behavior determination
to be relatively stronger when suffi cient control
resources are available.
In line with these assumptions, we found
across both studies that implicit racial attitudes
as measured with a Race IAT predicted visual
contact with an outgroup confederate better
when participants were cognitively taxed than
when untaxed. At the same time the predictive
validity of explicit group attitudes as measured
with the BSPS was more positive under full
resources as compared with the memory load
condition. The IAT also predicted speech illus-
trators reliably and more strongly under memory
load than under full resources in both studies.
Partially supporting our predictions, IAT scores
were reliably associated with the frequency
of body adaptors only in the memory load
condition in German participants (Study 2),
whereas BSPS scores were reliably associated
with body adaptors only in the full resources
condition in Italian participants (Study 1). In
accordance with our hypotheses, only explicit
attitudes were consistently positively related to
the rating of interviewer competence which had
always been administered after the memory task
was completed. And fi nally, our framework was
strongly supported on the level of the composite
behavioral indicator.
By demonstrating the role of control resources
as a moderator of predictive validity, our
findings add an important qualification to
previous research applying the taxonomy of
spontaneous versus deliberative behaviors (e.g.
Asendorpf et al., 2002; Dovidio et al., 1997).
Instead of classifying different behaviors a
priori into ‘spontaneous’ or ‘controlled’, be-
havior prediction may be enhanced by taking
into account key moderators such as momentarily
available control resources. More specifi cally,
it is plausible to assume that control resources
impose a certain threshold on how well and
how many different behaviors can be simultan-
eously controlled by the individual. Such a
view is also mirrored in the so-called resource
or ‘strength’ model of self-regulation (e.g.
Baumeister, Bratslavsky, Muraven, & Tice, 1998),
according to which control resources constitute
a general and limited resource that is subject
to situational and temporal fl uctuation. Note
that the resource model does not contradict
the idea of a behavioral hierarchy in which
some behaviors generally draw more heavily
on available control resources than others.
Rather, this model adds a dynamic element to
our understanding of behavior determination
in that it spells out more clearly the conditions
under which automatic vs. controlled infl uences
may be most likely expected. In the light of the
present framework and fi ndings, one could say
that impulses and automatic predispositions such
as implicit attitudes are not always translated
into spontaneous behavior, and less so when
suffi cient control resources to monitor and
adjust one’s behavior are present. Conversely,
explicit attitudes may even extend to nonverbal
behaviors such as body adaptors given that
suffi cient control is available.
When comparing the predictive validities of
implicit and explicit attitude measures in our
studies, it has to be noted that—taken as a
whole—the IAT appeared to be a stronger, more
reliable predictor of behavior. One plausible
explanation is that implicit and explicit measures
may differ with regard to where they tap into the
processes of behavior determination. The higher
Group Processes & Intergroup Relations 11(1)
84
predictive validities of the implicit measure may
be due to the fact that it assesses automatic
evaluations that are directly associated with
behavioral schemas of approach or avoidance
(Chen & Bargh, 1999; Neumann, Hülsenbeck, &
Seibt, 2004) which then infl uence behavior
quite directly. In contrast, an explicit meas-
ure such as the BSPS refl ects propositionally
held attitudes and beliefs that may have to pass
through a longer sequence of transformations
(e.g. corrections, decisions, intentions) until
they are fi nally converted into concrete action
(Strack & Deutsch, 2004). In other words, the
implicit measure may be ‘closer’ to the be-
havioral output stage than the explicit measure.
Developing and validating implicit and explicit
measures with regard to the exact processing
stages they tap into should become one of the
main objectives of current attempts to link
those measures to dual-process or dual-system
theorizing.
The present data yield fi rst empirical con-
fi rmation for the hypothesis that available control
resources moderate the infl uence of implicit and
explicit attitudes on actual behavior in interracial
interaction. This hypothesis can be derived from
a variety of existing dual-process or dual-system
models in social psychology such as the MODE
or the Refl ective-Impulsive Model. As such, the
present data undergird these models. But they do
not favor one theoretical account over another.
For instance, they cannot help to differentiate
between a dual-process or dual-system view or
to decide whether a single attitude (e.g. Fazio &
Olson, 2003) or dual attitude (e.g. Wilson et al.,
2000) perspective is more appropriate. In prin-
ciple, the data are consistent with any model
positing that (a) implicit and explicit measures
tap into different stages or systems of information
processing and (b) situationally available control
resources affect these different stages or systems
differently such that low resources boost the
infl uence of the stage/system implicit measures
tap into. Therefore, the benefi t of the present
research is clearly empirical in that it tests these
key assumptions with regard to actual behavior
by manipulating available control resources
experimentally, providing a useful starting point
for future investigations.
There are several limitations of this research
that should be kept in mind when generalizing
the present findings to other domains and
samples. First, we used only male confederates
in an attempt to increase the power of our test
because previous research has found that
prejudiced reactions usually are stronger with
regard to males than females (Eagly & Kite,
1987). Hence, it is possible that our effects may
be weaker with regard to female target per-
sons. Second, we could not predict judges’ global
ratings of interaction behavior at all. One ex-
planation for the absence of an effect is that
the global rating may be based to a stronger
extent on judges’ subjective impressions than
the easier to defi ne and operationalize behaviors
such as visual contact. Third, differences in the
mean pattern of behavior utilization between
the studies point to cultural differences in
the usage of nonverbal behavior in interracial
interaction between countries and/or as a
function of specifi c intergroup constellations.
Under memory load, Italian participants pri-
marily reduced their visual contact with the
outgroup member whereas German participants
primarily cut down the frequency of speech
illustrators (see Table 2). Regarding our main
hypothesis (moderated predictive validity) only
body adaptors appeared to be differentially
infl uenced by implicit attitudes across studies
such that body adaptors varied more strongly
as a function of implicit attitudes and control
resources in the German sample. The possibility
of culture dependent behavior determination
by implicit or explicit attitudes is an exciting
avenue for future research. Taking additional
settings into account may help to decompose
the general patterns of behavior determination
from the more cultural or intergroup-specifi c
idiosyncrasies of interracial interaction.
Finally, our fi ndings have applied implications
as well: our results show that the causes that
primarily drive interracial behavior may vary
according to circumstances. For instance, im-
plicit group attitudes or biases will most likely
infl uence behavior when control resources are
scarce. States of reduced control resources may
result when people are concurrently taxed by
additional tasks (as in the present studies), have
85
Hofmann et al. control resources and interracial interaction
been depleted by prior acts of self-control and
are under time pressure, stress, or under the in-
fl uence of alcohol. Identifying the causes (e.g.
implicit vs. explicit attitudes) and the trigger-
ing conditions of impulsive or refl ective inter-
group behavior may eventually result in refi ned
methods to combat prejudice and other societal
problems resulting from frequent competitions
among the automatic and controlled parts of
human nature.
Notes
1. Disaggregating the BSPS into a blatant
and subtle subscale did not affect the
statistical conclusions drawn in both
studies.
2. The distance between chairs was kept constant
in order to control for the effects of spatial
distance on eye gazing behavior (e.g. Argyle &
Dean, 1965).
3. The camera facing the interviewer did not
actually record his behavior.
4. One participant did not provide a competence
rating of the African interviewer for one of the
questions asked; two participants did not provide
one competence rating each with regard to the
Italian interviewer.
Acknowledgements
This research was supported by a grant from the
German Academic Exchange Service to Wilhelm
Hofmann and by a grant from the German Science
Foundation (DFG) to Manfred Schmitt (Schm
1092/5–1). We would like to thank Rainer Banse,
Malte Friese, Bertram Gawronski, Konrad Schnabel,
and Jane Thompson for valuable comments on
an earlier version of this article. We also thank
Valeria Adami, Gabriela Blum, Mauro Catellani,
Franziska Hacke, Nina Heckmann, Gregor Roux,
and Nadine Thomas for their help in collecting
the data.
References
Arcuri, L., & Boca, S. (1996). Pregiudizio e
affi liazione politica: Destra e sinistra di fronte
all’immigrazione dal terzo mondo [Prejudice
and political affi liation: Right- and left-wing
attitudes toward immigration from the third
world]. In P. Legrenzi & V. Girotto (Eds.),
Psicologia e politica [Psychology and politics]
(pp. 241–274). Milano: Cortina.
Argyle, M., & Dean, J. (1965). Eye-contact, distance
and affi liation. Sociometry, 28, 289–304.
Asendorpf, J. B., Banse, R., & Mücke, D. (2002).
Double dissociation between implicit and
explicit personality self-concept: The case of
shy behavior. Journal of Personality and Social
Psychology, 83, 380–393.
Banse, R., Seise, J., & Zerbes, N. (2001). Implicit
attitudes towards homosexuality: Reliability,
validity, and controllability of the IAT. Zeitschrift
für Experimentelle Psychologie, 48, 145–160.
Baumeister, R. F., Bratslavsky, M., Muraven, M., &
Tice, D. M. (1998). Ego depletion: Is the active
self a limited resource. Journal of Personality and
Social Psychology, 74, 1252–1265.
Chen, M., & Bargh, J. (1999). Consequences of
automatic evaluation: Immediate behavioral
predispositions to approach or avoid the
stimulus. Personality and Social Psychology Bulletin,
25, 215–224.
Dovidio, J. F., Kawakami, K., & Gaertner, S. L.
(2002). Implicit and explicit prejudice and
interracial interaction. Journal of Personality and
Social Psychology, 82, 62–68.
Dovidio, J. F., Kawakami, K., Johnson, C., &
Johnson, B. (1997). On the nature of prejudice:
Automatic and controlled processes. Journal of
Experimental Social Psychology, 33, 510–540.
Eagly, A. H., & Kite, M. E. (1987). Are stereotypes
of nationalities applied to both women and
men? Journal of Personality and Social Psychology,
53, 451–462.
Ekman, G., & Friesen, W. V. (1972). Hand
movements. Journal of Communication, 22,
353–374.
Fazio, R. H., Jackson, J. R., Dunton, B. C., &
Williams, C. J. (1995). Variability in automatic
activation as an unobtrusive measure of racial
attitudes: A bona fi de pipeline? Journal of
Personality and Social Psychology, 69, 1013–1027.
Fazio, R. H., & Olson, M. A. (2003). Implicit
measures in social cognition research: Their
meaning and uses. Annual Review of Psychology,
54, 297–327.
Fazio, R. H., & Towles-Schwen, T. (1999). The
MODE model of attitude-behavior processes.
In S. Chaiken & Y. Trope (Eds.), Dual-process
theories in social psychology (pp. 97–116). New
York: Guilford.
Gawronski, B. (2002). What does the Implicit
Association Test measure? A test of the
convergent and discriminant validity of
Group Processes & Intergroup Relations 11(1)
86
prejudice related IATs. Experimental Psychology,
49, 171–180.
Greenwald, A. G., & Banaji, M. R. (1995). Implicit
social cognition: Attitudes, self-esteem, and
stereotypes. Psychological Review, 102, 4–27.
Greenwald, A. G., McGhee, D. E., &
Schwartz, J. L. K. (1998). Measuring individual
differences in implicit cognition: The Implicit
Association Test. Journal of Personality and Social
Psychology, 74, 1464–1480.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R.
(2003). Understanding and using the Implicit
Association Test I: An improved scoring
algorithm. Journal of Personality and Social
Psychology, 85, 197–216.
Hofmann, W., Gawronski, B., Gschwendner, T.,
Le, H., & Schmitt, M. (2005). A meta-analysis on
the correlation between the Implicit Association
Test and explicit self-report measures. Personality
and Social Psychology Bulletin, 31, 1369–1385.
Hofmann, W., Gschwendner, T., Nosek, B. A., &
Schmitt, M. (2005). What moderates implicit–
explicit consistency? European Review of Social
Psychology, 16, 335–390.
McConnell, A. R., & Leibold, J. M. (2001).
Relations among the Implicit Association Test,
discriminatory behavior, and explicit measures
of racial attitudes. Journal of Experimental Social
Psychology, 37, 435–442.
Mierke, J., & Klauer, K. C. (2003). Method-specifi c
variance in the Implicit Association Test. Journal
of Personality and Social Psychology, 85, 1180–1192.
Neumann, R., Hülsenbeck, K., & Seibt, B. (2004).
Attitudes toward people with AIDS and
avoidance behavior: Automatic and refl ective
bases of behavior. Journal of Experimental Social
Psychology, 40, 543–550.
Neumann, R., & Seibt, B. (2001). The structure of
prejudice: Associative strength as a determinant
of stereotype endorsement. European Journal of
Social Psychology, 31, 609–620.
Norman, D. A., & Shallice, T. (1986). Attention
to action. Willed and automatic control of
behavior. In R. J. Davidson, G. E. Schwartz, &
D. Shapiro (Eds.), Consciousness and self
regulation: Advances in research (pp. 1–18).
New York: Plenum.
Nosek, B. A. (2005). Moderators of the relationship
between implicit and explicit evaluation.
Journal of Experimental Psychology: General, 134,
565–584.
Pettigrew, T. F., & Meertens, R. W. (1995). Subtle
and blatant prejudice in Western Europe.
European Journal of Social Psychology, 25, 57–75.
Poehlman, T. A., Uhlmann, E., Greenwald, A. G., &
Banaji, M. R. (2006). Understanding and using
the Implicit Association Test: 3. Meta-analysis of
predictive validity. Unpublished manuscript.
Reisberg, D. (1983). General mental resources and
perceptual judgments. Journal of Experimental
Psychology: Human Perception and Performance, 9,
966–979.
Roediger, H. L., & McDermott, K. B. (1995).
Creating false memories: Remembering words
not presented in lists. Journal of Experimental
Psychology: Learning, Memory, & Cognition, 21,
803–814.
Smith, E. C., & DeCoster, J. (2000). Dual-process
models in social and cognitive psychology:
Conceptual integration and links to underlying
memory systems. Personality and Social Psychology
Review, 4, 108–131.
Strack, F., & Deutsch, R. (2004). Refl ective and
impulsive determinants of social behavior.
Personality and Social Psychology Review, 8,
220–247.
Wagner, U., van Dick, R., & Zick, A. (2001).
Sozialpsychologische Analysen und Erklärungen
von Fremdenfeindlichkeit in Deutschland
[A social psychological analysis and explanation
of hostility toward foreigners in Germany].
Zeitschrift für Sozialpsychologie, 32, 59–79.
Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000).
A model of dual attitudes. Psychological Review,
107, 101–126.
Zick, A. (1997). Vorurteile und Rassismus. Eine
sozialpsychologische Analyse [Prejudice and racism:
A social psychological analysis]. Münster:
Waxmann.
Paper received 6 September 2006; revised version accepted
7 June 2007.
87
Hofmann et al. control resources and interracial interaction
Appendix
IAT attribute stimuli (English translation in parentheses)
Study 1
Positive words: Negative words:
ALLEGRIA (cheerfulness) SCHIFO (disgust)
AMORE (love) OMICIDIO (murder)
FELICITÀ (happiness) DISASTRO (disaster)
LEALTÀ (sincerity) ODIO (hate)
ONESTÀ (honesty) ANTIPATIA (antipathy)
PACE (peace) RABBIA (annoyance)
PARADISO (paradise) BOMBA (bomb)
SIMPATIA (sympathy) TERREMOTO (earthquake)
Study 2
Positive words: Negative words:
HEITERKEIT (cheerfulness) ÄRGER (anger)
LIEBE (love) ELEND (misery)
FREUDE (happiness) ANGST (anxiety)
GESUNDHEIT (health) UNGLÜCK (bad luck)
GLÜCK (luck) VERRAT (treason)
FRIEDEN (peace) STREIT (quarrel)
PARADIES (paradise) KRANKHEIT (illness)
SPASS (fun) PANIK (panic)
Biographical notes
wilhelm hofmann is an assistant professor at
the University of Würzburg. His main research
interests are in the domains of social cognition
and self-regulation. Specifi cally, he studies
the relationship between implicit and explicit
attitudes and the conditions under which
implicit vs. explicit attitudes infl uence self-
regulatory behavior.
tobias gschwendner is currently a PhD student
at the University of Koblenz-Landau. His
research interests include attitude/trait-behavior
consistency, person × situation interactions, as
well as social attitudes and prejudiced behavior.
luigi castelli is an associate professor at the
University of Padova. His main areas of research
include stereotyping and social perception,
particularly stereotype activation and the
perception of in-group members who use
stereotypes.
manfred schmitt is a full professor of psychology
at the University of Koblenz-Landau. His main
research interests are in the domains of social
justice, social emotions, altruism, and attitude-
behavior consistency.
