The Brief Implicit Association Test

N. Sriram1 and Anthony G. Greenwald2

1
University of Virginia, Charlottesville
2
University of Washington, Seattle

Abstract. The Brief Implicit Association Test (BIAT) consists of two blocks of trials with the same four categories and stimulus-response
mappings as the standard IAT, but with 1/3 the number of trials. Unlike the standard IAT, the BIAT focuses the subject on just two of each block’s
four categories. Experiments 1 and 2 demonstrated that attitude BIATs had satisfactory validity when good (but not bad) was a focal category, and
that identity IATs had satisfactory validity when self (but not other) was a focal category. Experiment 2 also showed that a good-focal attitude
BIAT and a self-focal identity BIAT were psychometrically similar to standard IAT measures of the same constructs. Experiment 3 presented each
of six BIATs twice, showing that procedural variables had no more than minor influences on the resulting implicit measures. Experiment 4 further
demonstrated successful use of the BIAT to measure implicit stereotypes.

Keywords: Implicit Association Test, implicit measure, social cognition

In eleven years since its introduction, the Implicit Associa- focus of published research, designed variations on IAT pro-
tion Test (IAT; Greenwald, McGhee, & Schwartz, 1998) cedures (e.g., the Go/No-go Association Task of Nosek &
has been used in several hundred studies to provide mea- Banaji, 2001 and the single-category IATs of Bluemke &
sures of association strengths. The associations investigated Friese, 2008; Karpinski & Steinman, 2006) have had at least
have often corresponded to attitudes, identities, and stereo- the partial intent of managing the subject’s performance
types (Greenwald et al., 2002). Attitude IATs combine a con- strategy. The present research investigates a new modifica-
cept classification (e.g., Coke vs. Pepsi) with an attribute tion of the IAT – a Brief Implicit Association Test (BIAT)
classification representing positive versus negative valence – that uses simplified instructions and was designed with
(e.g., pleasant vs. unpleasant). Identity (or self-concept) the intention of reducing spontaneous variation in subject
IATs combine contrast of self versus other with a nominal strategy.
contrast (e.g., male vs. female; family vs. career; math vs. The IAT has four categories, each mapped onto one of
arts) or a trait contrast (e.g., strong vs. warm; large vs. the two responses. The BIAT instructs respondents to focus
small). Stereotype IATs combine social group categories on just two of the IAT’s four category–response mappings in
(e.g., male vs. female; Asian vs. Hispanic) with nominal each combined task. Prior to each combined task, subjects
or trait contrasts. are shown two category labels together with their exemplars
In combined task blocks of the IAT, subjects switch and are instructed (a) to ‘‘keep them in mind’’, (b) to
between classifying exemplars of one contrast (e.g., Pepsi respond to items from these two categories with a ‘‘focal’’
vs. Coke) and exemplars of the other contrast (e.g., pleasant response key, and (c) to respond to any other stimuli with
vs. unpleasant) on consecutive trials. In a {pleasant+ an alternative ‘‘nonfocal’’ response key. With two blocks
Pepsi}|{unpleasant+Coke} combined task, pleasant and and a total of fewer than 80 trials, the BIAT substantially
Pepsi are mapped to one response (e.g., left key) and unpleas- reduces administration time relative to the standard five-step
ant and Coke to the other response (e.g., right key). A second IAT procedure, which is often done with seven blocks of
combined-task block reverses the response mappings of one trials and typically involves 180 or more trials.
of the contrasts (e.g., {pleasant+Coke}|{unpleasant+Pepsi}). The four experiments in this report investigate properties
If Pepsi is more strongly associated with positive valence than of the BIAT. Experiment 1 provides initial assessments of
Coke, classification should be faster in the {pleasant+ the BIAT’s psychometric properties, unexpectedly finding
Pepsi}|{unpleasant+Coke} block than in the {pleasant+ that it matters which subset of the four category–response
Coke}|{unpleasant+Pepsi} block. mappings is selected for focus. Experiment 2 establishes
Various strategies can be used in performing the IAT’s convergence between BIAT and standard IAT measures of
combined tasks. One strategy is to prepare equally for all attitude and identity. Experiment 3 extends the BIAT to
four of the category–response mappings (e.g., pleasant–left; additional attitude and identity topics, and also to stereo-
Pepsi–left; unpleasant–right; Coke–right). Alternately, sub- types. Experiment 4 focuses on implicit stereotypes, includ-
jects can focus on just the two mappings associated with ing race, age, and gender stereotypes. Collectively, the four
(say) the left-side response, giving themselves an added experiments establish the ability of the BIAT to function
mental note to give the right-side response for ‘‘anything effectively in the range of domains in which standard IAT
else’’. Although such strategy variations have not been the measures have been successfully used.
Ó 2009 Hogrefe & Huber Publishers Experimental Psychology 2009; Vol. 56(4):283–294
DOI: 10.1027/1618-3169.56.4.283
284 Sriram & Greenwald: Brief Implicit Association Test

General Method The interval between the correct response on one trial and
presentation of the next stimulus was 400 ms.1
Subjects Response errors were signaled by a red ‘‘X’’, which
appeared centered below the stimulus and disappeared
Subjects were undergraduate students from the University of immediately when the correct response was made. The stud-
Washington Psychology Department’s undergraduate sub- ies were administered in individual subject stations using
ject pool, who provided data at desktop computers in indi- Inquisit Version 2.0 (2005) to control computer displays
vidual cubicles. and data recording. At the conclusion, subjects received
on-screen debriefing information.

Design of the BIAT Analysis Strategy

Each BIAT is composed of two combined-task blocks, each Measures of association strength based on IATs and BIATs
of which can be described by its two focal categories (e.g., a were computed using the D measure (Greenwald et al.,
block with pleasant and Pepsi focal might be followed by a 2003), which is an effect-size-like measure with possible
block with pleasant and Coke focal). One category is focal range of 2 to +2. D is computed as the difference between
in both combined tasks (pleasant in the example just given) mean latencies of the two BIAT blocks divided by the inclu-
and has a contrasting category that remains nonfocal in both sive (not pooled) standard deviation of latencies in the two
tasks (unpleasant in this example). Our convention for nam- blocks. This measure has been shown to have psychometric
ing BIATs lists the four categories, placing the category that properties superior to those of a wide variety of alternative
consistently remains nonfocal within parentheses. The strategies for using latencies from the IAT’s two tasks
implicit soft-drink BIAT measure in this example is named (Greenwald et al., 2003; Sriram, Nosek, & Greenwald,
Coke–Pepsi/pleasant–(unpleasant). As a further convention, 2007).2
the order of listing indicates interpretation of scores. High To estimate internal consistency of BIAT measures, split-
scores indicate greater strength of the association of the half reliabilities were computed by partitioning the trials in
first-listed category with the third. In the Coke–Pepsi/ each of the two blocks into two parallel subsets. For exam-
pleasant–(unpleasant) BIAT, scores above zero indicate that ple, in the 32-trial blocks used in Experiments 1 and 2, one
the Coke–pleasant association is stronger than the Pepsi– subset consisted of trials {1, 2, 5, 6, 9, 10, 13, 14, 19, 20, 23,
pleasant association. Table 1 describes a general schema 24, 27, 28, 31, 32} and the other subset consisted of the
for the structure of BIAT measures, illustrated for this remaining 16 trials. In the 20-trial blocks of Experiments
soft-drink attitude example. 3 and 4, one subset had trials {5, 6, 9, 10, 15, 16, 19, 20}
and the other subset was composed of trials {7, 8, 11, 12,
13, 14, 17, 18}; the practice, prefatory trials 1–4 were unan-
Procedure alyzed. D measures for each subset were computed from the
differences between mean latencies of the same subset in the
Prior to completing BIATs, subjects completed parallel self- two blocks, divided by the inclusive standard deviation of
report measures of strength of the corresponding associa- these latencies. Internal consistency was estimated as the
tions. The instructions for each BIAT block displayed all correlation between these split halves. Each BIAT was
exemplars for the upcoming block’s two focal categories administered twice, permitting computation of a test–retest
(see Figure 1). The two focal categories are typically distin- estimate of reliability.
guished from each other not only by category identity but Explicit attitude measures included (a) the difference
also by visual format (e.g., text vs. image, or distinct fonts between separately rated strengths of association of con-
if both are textual). Subjects required an average of about trasted concepts with positive or negative valence (e.g., dif-
10 s to process the BIAT block instructions. ference between the liking ratings for cola brands) and (b)
After the instruction page display (e.g., Figure 1), the single-item Likert-format measures of relative preference
lists of focal-category exemplars disappeared, but the focal between the contrasted concepts. Similar combinations based
category labels remained in view. On each BIAT trial, an on sets of three items were used to obtain measures of rela-
exemplar of one of the four categories appeared in center tive strength of associations of self with contrasted identities
screen. If the initial response to a stimulus was in error, sub- based on gender and ethnicity. As described by Greenwald
jects were obliged to give a second response, and latency et al. (2003), the correlations between implicit and parallel
was recorded to occurrence of the correct response. This cre- explicit measures served as the primary validity criteria for
ated a built-in error penalty, which is also a property of stan- attitude and identity BIATs. The stereotypes examined in
dard IAT measures (cf. Greenwald, Nosek, & Banaji, 2003). this research were ones that had been demonstrated to be

1
Variation in this interval (150–750 ms) was previously shown to have no significant impact on IAT measures (Greenwald et al., 1998).
2
Values absolutely larger than 1.5 are rarely observed with the D measure. Because standard deviations of D measures are typically smaller
than 0.5, means of D measures most often correspond to Cohen’s d values at least twice as large (see, e.g., the IAT Ds and their
corresponding Cohen’s ds in Table 5).
Experimental Psychology 2009; Vol. 56(4):283–294 Ó 2009 Hogrefe & Huber Publishers
 Sriram & Greenwald: Brief Implicit Association Test 285

Table 1. Structure of BIAT measure for associations with positive valencea

General instruction for nonfocal key:
No. of ‘‘Press for anything else’’ Instruction
Blockb trialsc Function [items are shown in brackets] for focal keyd
1 20 First combined task [Focal concept 2e or unpleasant words] Press for focal concept 1e or
pleasant words
2 20 Second combined task [Focal concept 1 or unpleasant words] Press for focal concept 2 or
pleasant words
3 20 Repeat Block 1 [alternately, repeat Block 2]
4 20 Repeat Block 2 [alternately, repeat Block 1]
a
Attitude measures with valence as attributes are likely to be the most frequent uses of BIAT measures. On the basis of present findings,
it is recommended that assigning the pleasant valence attribute to the focal key yields superior measures. As is true of the standard IAT,
the BIAT can be done with any concept used as focal in both of the first two blocks and, likewise, any suitable contrasting concept
nonfocal in both of these blocks.
b
Repetitions after the first two blocks are optional. Each pair of blocks produces a separate BIAT measure. Averaging these multiple
BIAT measures should increase measure reliability and sensitivity.
c
Trials 1–4 in each block may be limited to just the two concepts that alternate focal status in Blocks 1 and 2 (as in present Experiments
3 and 4), and then omitted from data analyses. This use of Trials 1–4 reinforces instructions about which of these two concepts is focal
in each block. Total numbers of trials in each block can be increased, which should increase reliability and sensitivity of measures.
d
Arbitrarily, the focal key was always on the right side in the present experiments. Properties of BIAT measures should be unaffected
by interchanging right and left key functions.
e
Focal concepts 1 and 2 are the two concepts that alternate focal status in Blocks 1 and 2. The order in which these concepts are made
focal is ordinarily counterbalanced.

societally pervasive in previous research (Nosek et al.,

2007). Partly because of limited individual-difference varia-
tion, implicit–explicit correlations are weaker for these than
for the attitudes and identities examined in the present re-
search. For these pervasive stereotypes, the main test of
validity of BIAT measures was their ability to detect the
{Pepsi} same stereotype that had been found in previous research
with standard IAT measures.
Data analyses used hierarchical multiple regressions. In
the first regression step, the D score, as criterion, was
regressed onto counterbalanced experimental design factors
{PLEASANT}
and their interactions. In the second step, the parallel self-
report measure was added as a predictor. In the third and
NICE, HEAVEN, HAPPY, PLEASURE final step, interactions of the self-report predictor with
design factors were added as predictors. For attitude and
identity measures, evidence for BIAT validity took the form
Two categories, and their items, are displayed above. of significant prediction of the IAT measure by the self-
Keep the two categories in your mind as you do the task. report measure in the second step. Evidence for validity
was strengthened if the self-report measure’s relationship
to the BIAT-measure criterion was not moderated by design
Press 'K' when an item matches EITHER category. factors in the third step.
Press 'D' for anything else.

If you make an ERROR you will see a RED X. Experiment 1

When this happens, make the CORRECT response to proceed.
Overview
Go FAST. A few errors are OK.
Experiment 1 was conducted shortly before the 2004 US
Press the Spacebar to begin the task. Presidential Election involving George W. Bush and John
F. Kerry. An attitude BIAT assessed implicit candidate pref-
Figure 1. Instruction screen for the {Pepsi+pleasant} erence and an identity BIAT measured the association
BIAT block. between self and gender. Previous research has shown
Ó 2009 Hogrefe & Huber Publishers Experimental Psychology 2009; Vol. 56(4):283–294
286 Sriram & Greenwald: Brief Implicit Association Test

substantial implicit–explicit correlations in these domains practiced the BIAT instructions using two nonsocial focal
(e.g., Aidman & Carroll, 2003; Greenwald et al., 2003; categories – curved (circle, oval, ring, ball) and bird (eagle,
Nosek, 2005; Rudman, Greenwald, & McGhee, 2001). swan, parrot, duck) – in a single 32-trial training block. Non-
These correlations, which were expected to be at least focal exemplars were drawn from angled (triangle, square,
moderate in size, were expected to provide a useful check block, pyramid) and mammal (elephant, bison, deer, cow).
on the validity of the BIAT format. After this training block, each of the four BIATs was admin-
Unlike standard IATs, each BIAT has up to four variants istered twice in immediate succession. Subjects received one
that differ on which of the four component categories is of 16 counterbalanced task sequences that varied the posi-
never focal in the two combined-task blocks. In the candi- tion of each BIAT within the overall sequence and the order
date attitude BIATs, two of these four variants were of combined blocks within each BIAT.
compared – Kerry–Bush/good–(bad) and Bush–Kerry/
bad–(good). In the first of these, the category bad was never
focal, and the measure was scored so that strong associations Results
of Kerry with good received high scores. The second was
scored in the same direction (association of Kerry with Correct responses to focal category items are called focal
bad received low scores) and the category good was never responses; those to nonfocal categories are nonfocal
focal. The two gender identity BIATs were identified as responses. Across the four BIATs, focal responses were fas-
female–male/self–(other) and male–female/other–(self), both ter (709 ms) than nonfocal responses (774 ms), t(39) = 9.6,
scored so that stronger associations of female with self than p = 10 11 and also had fewer errors, 8.5% versus 10.0%,
with other would receive high scores. t(39) = 2.9, p = .007. These findings were consistent with
the expectation that subjects would selectively attend to
the focal categories.
Stimuli
In the Kerry–Bush/good–(bad) and Bush–Kerry/bad–(good) Candidate Attitude BIATs
BIATs, four face images of each presidential candidate were
used as category exemplars. Exemplars for good were the Eight subjects had identical warmth ratings for both presi-
four words, happy, warm, love, and friend; exemplars for dential candidates. Of the remaining 32 subjects, nine
bad were angry, cold, hate, and enemy. Stronger associa- accorded Bush greater warmth and 23 did so for Kerry.
tions of Kerry with good than bad received high scores. A measure indicating preference for Kerry over Bush was
For the female–male/self–(other) and male–female/other– constructed by subtracting the Bush rating from the Kerry
(self) BIATs the categories were female (female, woman, rating (possible range: 9 to +9).
girl, she), male (male, man, boy, he), self (I, me, mine, self), In the Kerry–Bush/good–(bad) BIAT, Bush supporters
and other (they, them, their, other). Stronger associations of were faster on trials in the {good+Bush} block (658 ms)
female with self than with other received high scores. than on trials in the {good+Kerry} block (763 ms). Like-
wise, Kerry supporters were faster on trials in the {good+
Kerry} block (741 ms) than in the {good+Bush} block
(874 ms). The IAT effect (D measure) was satisfactory in
Design internal consistency, and the correlation between the self-
reported candidate attitude and the average of the two
Each subject completed eight BIATs, consisting of two rep- Kerry–Bush/good–(bad) BIATs was .76 (see Table 2).
etitions in immediate succession of each of the four distinct The hierarchical regression analysis of D measures for
64-trial BIATs. Order of combined tasks within BIATs was the Kerry–Bush/good–(bad) BIAT included in its first step
counterbalanced across subjects. For example, the Kerry– order (of combined-task blocks), half (first or second half
Bush/good–(bad) was either ordered as {good+Bush} fol- of the experiment), and their interaction. This first step
lowed by {good+Kerry} or as {good+Kerry} followed by revealed an order effect in which implicit preference for
{good+Bush}. Each block had 32 trials. For half the sub- Kerry was higher when the block with good and Bush focal
jects, the good and self versions preceded the other and preceded the block with good and Kerry focal, t(35) = 2.2,
bad versions and the remainder received the reverse order. p = .04). The expected strong effect of explicit (self-report)
preference emerged clearly in the second step, t(35) = 6.6,
p = 10 7, zero-order r = .76. The third step included as pre-
Subjects and Procedure dictors the multiplicative products of the explicit measure
with order of combined tasks, half of the experiment, and
Subjects (24 females and 16 males) were first asked to self- their interaction (cf. West, Aiken, & Krull, 1996). None of
identify as male or female, after which they provided these factors qualified the second step’s prediction of IAT-
warmth ratings on a 10-point scale (1 = very cold and effect D measures by explicit preference (ts < 1).
10 = very warm) toward each of George W. Bush, John F. For the Bush–Kerry/bad–(good) BIAT, mean latencies
Kerry, the Republican Party, and the Democratic Party. They on {bad+Kerry} and {bad+Bush} blocks were 712 ms
then reported political identity along a 5-point scale and 778 ms for Bush supporters, compared to 842 ms and
anchored by Democrat and Republican. Next, subjects 810 ms for Kerry supporters. Unexpectedly, neither the
Experimental Psychology 2009; Vol. 56(4):283–294 Ó 2009 Hogrefe & Huber Publishers
 Sriram & Greenwald: Brief Implicit Association Test 287

Table 2. Internal consistency and validity in Experiment 1 (N = 40)

Consistency and test–retest Implicit–explicit correlation

correlation Trial type Administration
a
BIAT a a1 a2 r12 All Focal Non 1 2
Kerry–Bush/good–(bad) .83 .82 .66 .49 .76 .75 .65 .70 .60
Bush–Kerry/bad–(good) .76 .79 .66 .17 .11 .02 .22 .05 .12
Female–male/self–(other) .94 .92 .83 .55 .70 .73 .60 .73 .47
Male–female/other–(self) .55 .62 .51 .07 .07 .14 .25 .06 .04
a
In naming BIATs (see text), the category named third is focal in both combined-task blocks; the category in parentheses is not focal in
either combined task.
Note. Internal consistencies a1 and a2 are for the two 64-trial BIAT measures that are intercorrelated (r12) and for their combination (a).
Validity is the correlation between corresponding implicit and explicit measures. Validity correlations are presented overall (All), across
focal and nonfocal trials (focal and non), and for first or second BIAT administrations (1 and 2). For N = 40, r values associated with
two-tailed a levels of .05 and .01 are .31 and .40, respectively.

test–retest correlation for this BIAT nor the implicit–explicit However, those in which bad or other were focal yielded
correlation was statistically significant. Also inconsistent measures that were psychometrically weak and did not cor-
with expectation, the Kerry–Bush/good–(bad) and Bush– relate in expected fashion with parallel self-report attitudes
Kerry/bad–(good) BIATs were not significantly intercorre- and self-reported sex, respectively.
lated, r = .18, p = .28.

Gender Identity BIATs

Implications for Theoretical Understanding
of IAT Measures
In the female–male/self–(other) BIAT dichotomous subject
gender served as the explicit measure and correlated In Rothermund and Wentura’s (2004) salience asymmetry
r = .70 with the average D from the two identical adminis- interpretation of IAT measures, negative valence (e.g.,
trations of the BIAT measure. Female subjects were faster bad) is a ‘‘figural’’ category that is cognitively salient in
on {self+female} (596 ms) than on {self+male} (727 ms), the context of positive valence (e.g., good). Similarly, other
while male subjects were faster on {self+male} (722 ms) is figural and salient in the context of self. Rothermund and
than {self+female} (785 ms). Internal consistencies of the Wentura used this theoretical interpretation to suggest that
female–male/self–(other) BIAT were satisfactory and the subjects easily give the same response when two salient or
test–retest correlation was r = .55, p = .0003. In the hierar- two nonsalient categories are assigned to the same response.
chical regression, the expected main effect of self-reported If the salience asymmetry interpretation is correct, then one
sex was evident in the second step, t(35) = 5.1, p = 10 5, would expect that BIATs that instruct focus on the categories
zero-order r = .70, and was not moderated by design factors assumed to be figural and salient would produce measures
in the third step. superior to BIATs that instruct focus on nonfigural catego-
In the male–female/other–(self) BIAT, mean latencies on ries. Although that expectation was clearly disconfirmed
{other+female} and {other+male} blocks were 722 and by Experiment 1’s findings, the results nevertheless sup-
750 ms, respectively, for male subjects and were 715 and ported the more general underlying idea that there is an
690 ms, respectively, for female subjects. Contrary to expec- important asymmetry in pairs such as good–bad and self–
tations, for this BIAT, both internal consistency and test– other.
retest correlation for the D measure were low (see Table 2). Proctor and Cho’s (2006) polarity correspondence the-
The D measure was uncorrelated with subject gender and ory resembles the salience asymmetry theory in supposing
was also unrelated to the female–male/self–(other) variant, that correspondence between categories on polarity (a term
r = .04. No significant effects emerged from the three-step encompassing salience, familiarity, and linguistic marked-
hierarchical regression. ness) underlies performance on IAT measures. Remarkably,
the definition of salience in Proctor and Cho’s polarity cor-
respondence is diametrically opposed to that in the salience
asymmetry theory – Proctor and Cho understand good and
Discussion self (rather than bad and other) to be salient. However, they
point out that this definition reversal is inconsequential
The choice of focal categories had striking and unexpected regarding interpretation of the IAT (Proctor & Cho,
effects on BIAT measures. Variants that used good or self as p. 433, footnote 4) because both theories predict faster
focal categories produced internally consistent and predic- responses when salient (or nonsalient) categories share a
tively valid implicit attitude and implicit identity measures. response than otherwise. These cognitive asymmetries are
Ó 2009 Hogrefe & Huber Publishers Experimental Psychology 2009; Vol. 56(4):283–294
288 Sriram & Greenwald: Brief Implicit Association Test

likely important in the BIAT; they are considered further in and self-focal variants of the BIAT had greater validity than
the General Discussion. the bad-focal and other-focal variants. Experiment 2’s pri-
mary objective was to compare standard IATs with the more
valid BIAT variants. To this end, although both variants were
Experiment 2 used, the good-focal and self-focal variants always preceded
the bad-focal and other-focal variants.
Overview
Experiment 2 assessed convergence between the BIAT mea-
Analysis
sures that were first used in Experiment 1 and corresponding As in Experiment 1, BIAT latencies were faster on focal than
standard seven-block IAT measures. Subjects (22 males and on nonfocal trials (726 ms vs. 777 ms, t(66) = 7.6, p =
45 females) provided data for both formats. 10 9) and were also more accurate (7.8% vs. 8.7% errors,
t(66) = 2.5, p = .016). For the standard IAT, mean latency
(703 ms) and mean error rate (7.8%) did not differ between
Design right and left response keys. Twelve subjects had identical
warmth ratings for both candidates, 13 subjects had higher
The standard IAT is a five-step, seven-block procedure first ratings for Bush, and 42 expressed greater warmth for Kerry.
reported by Farnham and Greenwald (1999; see also Relative warmth for Kerry over Bush provided the validity
Greenwald & Farnham, 2000), as a reduction of the consid- criterion for candidate attitude IATs. Subject gender provided
erably longer procedure used in the first IAT report by the validity criterion for gender identity IATs.
Greenwald et al. (1998). In Experiment 2’s standard IAT
procedure, the first two of these blocks had 16 trials each.
They were followed by two identical 32-trial combined-task Candidate Attitude IATs
blocks after which came a fifth 16-trial block that reversed
the concept classification. Two 32-trial combined-task In the standard attitude IAT and the Kerry–Bush/good–(bad)
blocks that incorporated the reversed concept classification BIAT variant, the patterns of means in the combined blocks
completed the sequence, for a total of 176 trials. The IAT for the pro-Bush and pro-Kerry subjects were similar to
measure obtained from the standard procedure was com- those reported in Experiment 1. Validity correlations of
puted from two latency contrasts, one based on Blocks 3 IAT measures with self-report measures were strongly posi-
and 6 and the other on Blocks 4 and 7 (Greenwald et al., tive for both the standard attitude IAT and the Kerry–Bush/
2003). The BIAT measures for Experiment 2 were identical good–(bad) BIAT (see Table 3). Replicating Experiment 1,
to those of Experiment 1, with two 32-trial combined tasks this validity correlation was considerably lower for the
each, administered twice in succession. Bush–Kerry/bad–(good) variant. The internal consistencies
were markedly higher for the standard attitude IAT and
the Kerry–Bush/good–(bad) BIAT than they were for the
Procedure Bush–Kerry/bad–(good) BIAT.
Hierarchical regressions for the three attitude IAT mea-
Experiment 2 used the same categories and exemplars as sures found no effects of design factors or their interactions
Experiment 1. The explicit scales for the political attitude in the first step. A strong effect of explicit preference
test were converted to a 9-point format (1 = ‘‘very cold’’ emerged in the second step for both the standard IAT,
and 9 = ‘‘very warm’’) that included a neutral point. The t(62) = 6.5, p = 10 7, zero-order r = .65, and the Kerry–
resulting difference measure (Kerry warmth minus Bush Bush/good–(bad) BIAT, t(62) = 6.8, r = .65, p = 10 8,
warmth) ranged from 8 to +8. Subjects were administered zero-order r = .63, Like Experiment 1, the effect of explicit
one of eight counterbalanced task sequences. The response preference was not significant in the hierarchical regression
keys ‘‘K’’ and ‘‘D’’ were used for the standard IATs, with of the Bush–Kerry/bad–(good) BIAT, t(62) = 1.6, r = .22,
‘‘K’’ consistently corresponding to good or self and ‘‘D’’ p = .13. The standard IAT correlated r = .65 (p = 10 8)
to bad or other. The BIATs used ‘‘P’’ and ‘‘Q’’, respectively, with the good-focal BIAT, but only r = .39 (p = .001) with
as the keys for focal and nonfocal responses. the bad-focal BIAT. The correlation between the two BIAT
After providing self-report measures, half the subjects did variants was r = .29, p = .02.
one each of the standard candidate attitude and gender IATs
followed by the four BIAT variants used in Experiment 1. Gender Identity IATs
Each BIAT was administered twice in succession. The
remainder completed the BIATs prior to the standard IATs. A positive correlation between subject gender and the IAT D
As in Experiment 1, a 32-trial {curved+bird} practice block measure would show that, as expected, male subjects were
preceded the BIATs. The order of combined-task blocks was faster when self and male shared a response and female
varied by counterbalancing but was the same for the brief and subjects were faster when self and female shared a response.
standard IATs done by the same subject. Within their half, the These correlations were strong for both the standard
standard attitude IATs preceded the standard identity IATs identity IAT and the female–male/self–(other) variant, but
(or vice versa). Experiment 1 had shown that the good-focal were weaker for the male–female/other–(self) variant
Experimental Psychology 2009; Vol. 56(4):283–294 Ó 2009 Hogrefe & Huber Publishers
 Sriram & Greenwald: Brief Implicit Association Test 289

Table 3. Internal consistency and validity in Experiment 2 (N = 67)

Consistency and test–retest Implicit–explicit correlation

correlation Trial type Administration
IAT or BIAT a a1 a2 r12 All Focal Non 1 2
Standard (seven-block) candidate attitude .90 .87 .85 .62 .65 – – .62 .55
Kerry–Bush/good–(bad) .77 .80 .71 .32 .63 .60 .59 .46 .58
Bush–Kerry/bad–(good) .66 .46 .64 .26 .24 .22 .23 .17 .22
Standard (seven-block) gender identity .93 .89 .89 .71 .76 – – .74 .67
Female–male/self–(other) .92 .85 .84 .68 .67 .67 .57 .68 .58
Male–female/other–(self) .67 .70 .50 .35 .46 .43 .38 .40 .35
Note. Internal consistencies a1 and a2 are for the two 64-trial BIAT measures that are intercorrelated (r12) and for their combination (a).
For the standard IAT, r12 is for the correlation between D measures based on Blocks 3 and 6 and ones based on Blocks 4 and 7 (see
text). Validity is the correlation between corresponding implicit and explicit measures. Validity correlations are presented overall (All),
across focal and nonfocal trials (focal and non), and administrations (1 and 2). For N = 67, r values associated with two-tailed a levels
of .05 and .01 are .24 and .31, respectively.

(see Table 3). Internal consistencies were higher for both the categories appeared on odd-numbered trials starting with
standard identity IAT and the self-focal BIAT than for the Trial 5. The preliminary four trials were intended to assure
other-focal BIAT. that key assignments for the categories that would switch
For the standard identity IAT, the effect of subject gender positions were effectively established prior to collection of
emerged in the second step of the regression, t(62) = 9.4, data from the trials (Trials 5–20) to be used for computing
p = 10 13, zero-order r = .76, as it did for the female– D measures. Four new measurement topics were added to
male/self–(other) BIAT, t(62) = 6.9, p = 10 8, zero-order the two investigated in Experiments 1 and 2. The total of
r = .67. The effect of gender in the second step was also sig- six BIATs included two attitude measures, two identity mea-
nificant, but weaker, for the male–female/other–(self) BIAT, sures, and two stereotype measures.
t(62) = 3.6, p = .001, zero-order r = .46. The standard IAT
correlated .68 (p = 10 9) with the self-focal gender-identity
BIAT and .43 (p = .0003) with the other-focal gender- Design
identity BIAT. The correlation between the two BIAT
variants was r = .56, p = 10 6. After a practice 20-trial block using nonsocial categories, all
subjects completed a set of six BIATs. Each consisted of
Discussion two 20-trial blocks and each was presented twice during the
session. For half the subjects, the two identical BIATs
As in Experiment 1, the choice of focal attribute categories appeared in immediate succession. For the remainder, the
affected psychometric properties of BIAT measures. The two repetitions were spaced so that the other five BIAT mea-
BIATs that used good and self as focal categories showed sures intervened. Half the subjects received the same category
strong convergence with their corresponding standard IATs. exemplars in both administrations and half received different
Compared to Experiment 1, the bad-focal and other-focal exemplars (for cola brands and political candidates, the vari-
BIATs had slightly improved properties but were neverthe- ety of available exemplars was limited and, therefore, exem-
less inferior to the BIAT variants that that used good and self plars were reused). The two attitude BIATs were Kerry–Bush/
as focal categories. In combination, Experiments 1 and 2 good–(bad) and Coke–Pepsi/pleasant–(unpleasant); the two
strongly suggest that the BIAT method is suitable for attitude identity BIATs were female–male/self–(other) and Asian–
measurement when good is a focal category (and, appar- American/self–(other); the two stereotype BIATs were
ently, only when good is a focal category) and for identity male–female/science–(arts) and African American–Euro-
measurement when self is a focal category. pean American/weapons–(gadgets). The order of tasks within
each BIAT was not varied, such that {good+Kerry}, {pleas-
ant+Coke}, {self+male}, {self+Asian}, {science+male},
and {weapons+African American} always appeared before
Experiment 3 the complementary block.

Overview
Materials
Experiment 3 tested a shorter version of the BIAT. It also
added an unanalyzed four-trial preface to each of its two trial Explicit Measures
blocks. These preliminary trials presented exemplars of the
two concepts that switched responses between blocks twice Subjects answered three standard-format questions on
each (e.g., Bush and Kerry). Exemplars of the two remaining 10-point scales for each of the six topics. The first two
Ó 2009 Hogrefe & Huber Publishers Experimental Psychology 2009; Vol. 56(4):283–294
290 Sriram & Greenwald: Brief Implicit Association Test

requested judgments of the degree to which each concept experiments, these hierarchical regressions revealed no
category was associated with the focal attribute category. effects that qualified the basic findings displayed in Table 4.
The third item was a measure of relative association in
which the two concept categories were used as scale
anchors. Explicit measures were scored by taking the differ- Discussion
ence between the ratings for the first two items and then
weighting that equally with the third item. The one excep- Using a BIAT procedure with only two blocks of 20 trials
tion to this standard self-report format was gender identity. each, Experiment 3 extended the evidence from Experi-
Subjects rated themselves on masculinity and femininity ments 1 and 2 for validity and psychometric soundness of
on 10-point scales and also reported their gender. The rating attitude and identity BIAT measures. The measures also
scales and stimulus sets are available in an online had acceptable internal consistency. The BIAT measures
supplement3. of attitude and identity showed expected strong correlations
with parallel self-report measures. This validity evidence did
not vary as a function of novelty (or lack thereof) of
exemplars in the second administration. Spacing between
Subjects and Procedure repetitions of identical BIATs had no significant effects on
test–retest or implicit–explicit correlations.
One hundred and forty-nine subjects (109 females and 40 Internal consistency and test–retest reliability of BIAT
males) participated. After the small/bird–(mammal) practice stereotype measures were somewhat lower than those for
BIAT, each of the six BIATs was administered twice, with the attitude and identity BIATs. The relatively low impli-
the two identity BIATs first, the two attitude BIATs next, cit–explicit correlations for the two stereotype BIATs are
and the two stereotype BIATs last. In both identity BIATs, not problematic. As previously noted, this is the pattern
self was focal. In the attitude BIATs either good or pleasant found for with standard IAT measures of stereotype (see
was focal. In the stereotype BIATs, science (rather than arts) Hofmann et al., 2005; Nosek, 2005; Nosek et al., 2007).
and weapons (rather than gadgets) were focal. Prior to the At the same time, the ability of the BIAT measures to detect
first administration of each BIAT, subjects answered its three the same implicit stereotypes found in previous studies with
associated self-report items. Subjects did the tasks in one of standard IAT measures was less than clearly established.
eight counterbalanced sequences that varied spacing Mean D scores for the weapons–race and gender-science
between BIAT repetitions, novelty of items across repeti- BIATs differed from zero in the expected directions, but
tions, and whether each measure was first administered rel- were relatively small, at .17 and .16, relative to standard
atively early or relatively late in the procedure. The IATs (e.g., mean Ds = .37 and .37 in Nosek et al., 2007).
experiment required approximately 30 min to complete.

Experiment 4
Results
Together, results of Experiments 1–3 appear to have clearly
Over all topics, and confirming observations in Experiments established the usefulness of BIATs for assessing implicit
1 and 2, mean latencies were generally faster on focal trials attitudes and identities – even when used in the shortened
(679 ms) than on nonfocal trials (749 ms), t(66) = 20.2, (40 trial) format of Experiment 3. Experiment 3’s uncertain
p = 10 44, and were also more accurate (5.5% vs. 7.6% appraisal of the success of BIATs for stereotype measures
errors, t = 9.1, p = 10 13). The implicit–explicit correlations led to Experiment 4, which focused entirely on BIAT stereo-
that support validity of BIAT measures were substantial for type measures. For Experiment 4, analyses were planned to
the two attitude BIATs and the two identity BIATs address questions for stereotypes of the form (illustrated
(see Table 4). For the two stereotype measures, the impli- here for the race–weapons IAT): Do the two BIAT variants,
cit–explicit correlations were positive, but small, which is Black–White/weapons–(gadgets) and White–Black/gadgets–
typical for standard IAT measures of stereotypes. (weapons) result in measures that (a) are internally consis-
In the hierarchical regressions that tested for procedural tent, (b) detect the modal stereotypic association with equal
influences on IAT D measures, spacing and novelty served sensitivity, and (c) are positively correlated with each other?
as design factors that, together with their interaction, were Experiment 4 used two BIAT variations in each of four ste-
entered in the first step. The explicit measure was entered reotype domains. The training task in Experiment 3 was pro-
in the second step and the third step added interactions of moted to a full component of Experiment 4. Associations
spacing and novelty with the explicit measure. The effect between size and type of animal were measured by bird–
of the explicit measure at its entry in the second step was mammal/small–(large) and mammal–bird/large–(small)
large for all attitude and identity IATs and also reached sta- BIAT variations. A disability–age stereotype was assessed
tistical significance in the gender-science stereotype IAT, but with young–old/able–(disabled) and old–young/disabled–
was absent for the weapons–race IAT. As in the preceding (abled) variations. Gender stereotypes pertaining to academic

3
The materials can be found at E-mail http://faculty.washington.edu/agg/Sriram&Greenwald_BIAT_supplement.pdf.
Experimental Psychology 2009; Vol. 56(4):283–294 Ó 2009 Hogrefe & Huber Publishers
 Sriram & Greenwald: Brief Implicit Association Test 291

Table 4. Internal consistency and validity in Experiment 3 (N = 149)

Consistency and test– Implicit–explicit correlation

retest correlation Trial type Administration
BIAT a a1 a2 r12 All Focal Non 1 2
Female–male/self–(other) .85 .79 .73 .67 .74 .74 .65 .69 .66
Asian–American/self–(other) .72 .76 .52 .39 .48 .48 .43 .49 .30
Kerry–Bush/good–(bad) .83 .79 .72 .56 .55 .49 .50 .46 .53
Coke–Pepsi/pleasant–(unpleasant) .78 .71 .74 .38 .57 .54 .53 .49 .46
African American–European American/weapons–(gadgets) .60 .54 .52 .20 .04 .05 .02 .10 .04
Male–female/science–(arts) .68 .68 .51 .24 .24 .25 .17 .16 .22
Note. Internal consistencies are for the two 32-trial BIAT measures (a1 and a2) that are intercorrelated (r12) and for their combination
(a). Validity is the correlation between corresponding implicit and explicit measures. Validity correlations are presented overall (All),
across focal and nonfocal trials (focal and non), and administrations (1 and 2). For N = 149, r values associated with two-tailed a levels
of .05 and .01 are .16 and .21, respectively.

disciplines were measured with male–female/math–(arts) and IAT D scores, along with Cohen’s d and internal consisten-
female–male/arts–(math) variants, and a race–weapons ste- cies for the eight BIATs. As in previous experiments, hierar-
reotype contrasted Black–White/weapons–(gadgets) with chical regressions involving counterbalanced procedure
White–Black/gadgets–(weapons) variants. All were scored variations did not qualify the findings presented in Table 5.
so that positive D scores would reflect the expected modal ste- The findings in Table 5 show that all eight BIAT variants
reotype. That is, responses to blocks in which the focal cate- successfully detected the expected modal stereotypic associ-
gories were {small+bird}, {large+mammal}, {disabled+ ation, with observed effect sizes ranging from moderate to
old}, {able+young}, {science+male}, {arts+female}, {weap- large. Cohen’s d values, which are presented for all of the
ons+African American}, and {gadgets+European American} measures in Table 5, ranged from 0.43 for gadgets/White–
were expected to be faster than those in the complementary Black to 1.8 for small/bird–(mammal). Internal consistencies
blocks for each of the eight BIAT variants. for the eight pairs of BIATs (combining the two repetitions
of each one) ranged from .63 to .78 and correlations
between the two administrations of the same BIAT variant
Materials were positive, averaging r = .29. Only one set of stereotype
BIATs displayed significant correlations with its set of paral-
The self-report measures for each topic followed the pattern lel self-report measures. These were the two that associated
used in Experiment 3. gender with academic domains (rs = .38 and .38,
p  .0003). All other implicit–explicit correlations were
numerically positive, but none differed significantly from
Subjects and Procedure zero.
Not shown in Table 5 are correlations between the two
Ninety undergraduates (59 females and 31 males) completed variations (which differed in focal categories) of each stereo-
the two BIAT variants for each of four stereotype domains. type BIAT. Positive correlations of at least moderate strength
Each of the eight BIATs was done twice in succession using would suggest that the two variations were capturing the
the 20-trial block structure introduced in Experiment 3, for a same stereotypic associations. This was found for two of
total of sixteen 40-trial BIATs. The two BIAT variants the four domains: Size–animal (r = .28, p = .008) and dis-
within each stereotype domain were consistently spaced ability–age (r = .36, p = .001). In those two domains, the
by interposing one BIAT variant from each of the three other corresponding explicit measures were also positively corre-
domains. Half of the subjects did all of the BIATs so that the lated with each other (rs = .44 and .55, respectively,
task embodying the expected stronger association (e.g., {dis- ps  .00002). In the other two domains, the correlations
abled+old}) preceded that with the alternative combination were weaker: Gender–academics (r = .14, p = .18) and
(e.g., {disabled+young}). Subjects were administered one race–weapons (r = .13, p = .24). In these latter two
of four task sequences that also counterbalanced the order domains, the corresponding explicit measures were also
of the two variations of each stereotype BIAT. weakly intercorrelated (rs = .12 and .11, respectively,
ps  .25).
Results and Discussion
Consistent with observations of Experiments 1–3, across all General Discussion
eight BIATs, focal trials had shorter latencies than nonfocal tri-
als (736 ms vs. 807 ms, t = 12.1, p = 10 19) and fewer errors The main features that distinguish the BIAT from standard
(8.5% vs. 11.8%, t = 6.9, p = 10 9). Table 5 presents mean IAT measures are (a) substantially fewer trials and (b) a task
Ó 2009 Hogrefe & Huber Publishers Experimental Psychology 2009; Vol. 56(4):283–294
292 Sriram & Greenwald: Brief Implicit Association Test

Table 5. Effect size, internal consistency, and validity in Experiment 4 (N = 90)

Consistency and test–retest
correlation
Stereotype BIAT variant Mean IAT D (Cohen’s d) a a1 a2 r12 Implicit–explicit correlation
Bird–mammal/small–(large) .71 (1.80) .78 .77 .65 .31 .16
Mammal–bird/large–(small) .65 (1.68) .75 .77 .50 .17 .12
Young–old/able–(disabled) .55 (1.62) .63 .66 .45 .27 .06
Old–young/disabled–(able) .58 (1.63) .72 .64 .63 .35 .11
Male–female/math–(arts) .33 (0.78) .74 .65 .68 .33 .38
Female–male/arts–(math) .37 (1.02) .74 .71 .56 .17 .38
Black–White/weapons–(gadgets) .30 (0.73) .74 .66 .62 .32 .18
White–Black/gadgets–(weapons) .17 (0.43) .68 .61 .38 .36 .06
Note. All BIAT measures were scored so that societally modal stereotypes would receive numerically positive scores. D is the IAT score
developed by Greenwald et al. (2003). Cohen’s d, in parentheses, is the mean D divided by its standard deviation. Internal consistencies
are for the two 32-trial BIAT measures (a1 and a2) that are intercorrelated (r12) and for their combination (a). For N = 90, r values
associated with two-tailed a levels of .05 and .01 are .21 and .27, respectively.

instruction to focus on just two of the four categories in each between BIAT administrations, and the novelty of stimuli
four-category test block. Subjects achieved the desired in the second presentation. Other than the ‘‘order effect’’,
focus, as indicated by responses to the two focal categories these procedural factors had no more than minor or inconsis-
being consistently faster and more accurate than responses tent influences on BIAT measures.
to nonfocal categories. In standard IAT measures, the associations tested first in
In previous research, the attitude and identity topics of sequence tend to appear stronger than those tested second.
Experiments 1–3 had produced substantial positive correla- This was first reported by Greenwald et al. (1998) and
tions between standard IAT measures and parallel self-report was subsequently reported in numerous studies (e.g., Klauer
measures. Explanation of this use of expected convergence & Mierke, 2005). This ‘‘order effect’’ has been speculatively
of IAT and self-report measures to establish the utility of attributed to associations being strengthened during perfor-
new IAT procedures appears in detail in Greenwald et al. mance of the first of the IAT’s two combined tasks. Weak-
(2003), pp. 199–200, 212). The BIAT measures of Experi- to-moderate order effects for the political attitude BIAT in
ments 1–3 indeed produced these positive correlations, Experiment 1 and for two of the stereotype BIATs in
although at slightly smaller magnitudes than for standard Experiment 4 were observed. However, these effects were
IAT measures, indicating that the BIAT functioned similarly reversed in direction relative to order effects observed in
to standard IAT measures. Experiment 1 also found two standard IAT measures – that is, the associations assessed
important exceptions to these large correlations and Experi- in the first combined task appeared weaker (not stronger)
ment 2 replicated these exceptions. Specifically, the than those assessed in the second combined task. This find-
expected correlations with parallel self-report measures ing may be a consequence of the limited practice preceding
occurred strongly only when positive valence was focal BIAT measures. This observation should be examined fur-
for attitude BIATs and when self was focal for identity ther in subsequent research.
BIATs. They did not appear when negative valence was fo- Comparisons of present findings with previous findings,
cal for an attitude BIAT or when other was focal for an iden- as well as the direct comparisons of BIATs and standard IATs
tity BIAT. built into present Experiment 2, indicated that the BIAT’s
For implicit stereotype measures, strong positive correla- reduction in trials produced relatively small decrements in
tions with parallel self-report measures are not expected. psychometric performance on test–retest and implicit–expli-
Therefore, evidence for usefulness of stereotype BIATs cit correlations. Such a result was perhaps anticipated by
was limited to observing whether BIAT results effectively Brendl, Markman, and Messner (2001), who proposed that
detected several implicit stereotypes that were known, from the IAT effect emerges, not at the level of single items,
previous research with standard IAT measures, to be but at the level of complete test blocks.
observed pervasively (cf. Nosek et al., 2007). Experiment
4 confirmed that the BIAT method effectively detected these
implicit stereotypes associated with age, race, and gender. Privileged Categories?
When good or pleasant was focal, attitude BIATs in the
Limited Influence of Procedural Variables present research produced findings similar to those obtained
previously with standard attitude IAT measures. Similarly,
Across Experiments 1–4, counterbalanced procedural fac- identity BIATs for which self was a focal category produced
tors included administration (first or second presentation findings similar to those observed in previous IAT research.
of a specific BIAT), order of blocks within BIATs, spacing These observations suggested that, compared to their
Experimental Psychology 2009; Vol. 56(4):283–294 Ó 2009 Hogrefe & Huber Publishers
 Sriram & Greenwald: Brief Implicit Association Test 293

complements, positive valence and self are, in some sense, and negative valence. Nevertheless, such an assumption is
privileged categories. quite plausible, in light of several scholarly treatments of
There was no corresponding evidence for privileged cat- the self that have described the self as drawing on memory
egories in the BIAT results for implicit stereotypes in Exper- structures that are considerably more complex than those
iments 3 and 4. Nevertheless, for two of the four stereotype that represent other persons (e.g., Greenwald, 1981;
topics of Experiment 4 (gender–academics and race– Kihlstrom & Cantor, 1984; Koffka, 1935).
weapons) variation of the BIAT’s focal categories revealed
asymmetries associated with the choice of focal categories.
Stereotype Asymmetry
Theoretical Interpretations In Experiment 4, correlations between pairs of stereotype
BIATs that involve naturally complementary pairs (i.e., able
Associative Focus complementary to disabled; large complementary to small)
were relatively high (age–ability, r = .36; animals–size,
Why do properties of BIAT measures vary with choice of r = .28). The two stereotype BIATs that showed weaker
focal categories? A possible theoretical explanation follows between-variant correlations were for gender–academics
directly from the BIAT instructions which may induce an (r = .14) and race–weapons (r = .13). For these two BIATs,
associative focus that allows the subject’s performance to the associated category pairs (math–arts and weapons–
be determined primarily by a single association. For exam- gadgets) appeared to be less naturally complementary. They
ple, when asked to focus on Pepsi and good, subjects’ asso- were also nominal categories, rather than adjectives.
ciations of Pepsi with positive valence may become more
accessible than other associations – viz., Pepsi with negative
valence and Coke with either positive or negative valence.
Several other researchers have aimed to achieve something Conclusions
resembling what we describe here as associative focus by
designing IAT-like procedures that are limited to three cate- The present findings imply that BIAT measures of implicit
gories – especially, the Go/No-go Association Test (Nosek attitudes should have a positive focal valence category and
& Banaji, 2001) and the Single Target and Single Category BIAT measures of implicit identities should use self as a
Association Tests (Bluemke & Friese, 2008; Karpinski & focal category. Stereotype measures based on the IAT may
Steinman, 2006). have distinct BIAT variants that differ as a function of the
category chosen as focal.
We close by stating some questions that await resolution.
Valence Asymmetry First, can spaced repetitions of a BIAT increase test–retest
reliability relative to the same number of trials used in a
To explain the observed superior psychometric properties of standard IAT format? Second, can multiple objects be placed
attitude BIATs that use positive valence as a focal category on a common evaluative scale by making pairwise compar-
requires not only the associative focus hypothesis, but also isons in individual attitude BIATs? Third, is it possible to
an assumption that the attitudes being measured in the present use the BIAT procedure to measure attitudes on a scale that
experiments are mentally represented more by positive than has a neutral-valence zero point? And, finally, are there
negative associations. The proposition that positive valence coherent BIAT measures of implicit attitude that emerge
is cognitively more prominent than negative valence derives from focusing on negative rather than positive valence?
(in the modern era) from Zajonc’s (1968) article on ‘‘attitudi-
nal effects of mere exposure’’, which documented the great-
er frequency of positive than negative valence in various
contexts, including lexicons. Support for this interpretation
has recently appeared in the proposal by Unkelbach, Fiedler, Acknowledgements
Bayer, Stegmüller, and Danner (2008) that, compared to
negatively valenced knowledge, positively valenced knowl- The University of Washington has applied for patent on the
edge is more densely structured in memory. Unkelbach BIAT method. The patent is managed by Project Implicit, a
et al.’s density hypothesis implies that associative structures nonprofit organization of which the second author is an offi-
involving positive valence should be both more prominent cer. Both the University of Washington and Project Implicit
and more cohesively structured than those involving nega- authorize free use of the BIAT method and published stimuli
tive valence (cf. Ashby, Isen, & Turken, 1999). for scholarly research, provided that reports of the research
clearly identify any modifications made to the BIAT and
appropriately cite the present article. Please contact Project
Self–Other Asymmetry Implicit (E-mail: feedback@projectimplicit.net) to request
a license for commercial or other nonscholarly use of the
There is no theory of self–other asymmetry parallel to BIAT. The authors thank Justin Storbeck and Yoav
Unkelbach et al.’s (2008) density hypothesis for positive Bar-Anan for comments on earlier drafts. Aarati Sriram,
Ó 2009 Hogrefe & Huber Publishers Experimental Psychology 2009; Vol. 56(4):283–294
294 Sriram & Greenwald: Brief Implicit Association Test

age 11, was the chief tester during the development of the Kihlstrom, J. F., & Cantor, N. (1984). Mental representations of the
BIAT. The final version of the BIAT was the one among a self. In L. Berkowitz (Ed.), Advances in experimental social
dozen or more variations that reliably identified her as psychology (Vol. 17, pp. 2–48). New York: Academic Press.
Klauer, K. C., & Mierke, J. (2005). Task-set inertia, attitude
female in the gender identity task. accessibility, and compatibility-order effects: New evidence
for a task-set switching account of the Implicit Association
Test effect. Personality and Social Psychology Bulletin, 31,
References 208–217.
Koffka, K. (1935). Principles of gestalt psychology. New York:
Aidman, E. V., & Carroll, S. M. (2003). Implicit individual Harcourt.
differences: Relationships between implicit self-esteem, Nosek, B. A. (2005). Moderators of the relationship between
gender identity, and gender attitudes. European Journal of implicit and explicit evaluation. Journal of Experimental
Personality, 17, 19–37. Psychology: General, 134, 565–584.
Ashby, F. G., Isen, A. M., & Turken, A. U. (1999). A neuropsy- Nosek, B. A., & Banaji, M. R. (2001). The Go/No-go
chological theory of positive affect and its influence on Association Task. Social Cognition, 19, 625–664.
cognition. Psychological Review, 106, 529–550. Nosek, B. A., Smyth, F. L., Hansen, J. J., Devos, T., Lindner,
Bluemke, M., & Friese, M. (2008). Reliability and validity of the N. M., Ranganath, K. A., et al. (2007). Pervasiveness and
Single-Target IAT (STIAT): Assessing automatic affect correlates of implicit attitudes and stereotypes. European
towards multiple attitude objects. European Journal of Review of Social Psychology, 18, 36–88.
Social Psychology, 38, 977–997. Proctor, R. W., & Cho, Y. S. (2006). Polarity correspondence:
Brendl, C. M., Markman, A. B., & Messner, C. (2001). How do A general principle for performance of speeded binary
indirect measures of evaluation work? Evaluating the infer- classification tasks. Psychological Bulletin, 3, 416–442.
ence of prejudice in the Implicit Association Test. Journal of Rothermund, K., & Wentura, D. (2004). Underlying processes in
Personality and Social Psychology, 81, 760–773. the Implicit Association Test: Dissociating salience from
Farnham, S. D., & Greenwald, A. G. (1999, June). In-group associations. Journal of Experimental Psychology: General,
favoritism = implicit self-esteem X in-group identification. 133, 139–165.
Paper presented at meetings of the American Psychological Rudman, L. A., Greenwald, A. G., & McGhee, D. E. (2001).
Society, Denver, CO. Implicit self-concept and evaluative implicit gender stereo-
Greenwald, A. G. (1981). Self and memory. In G. H. Bower types: Self and ingroup share desirable traits. Personality and
(Ed.), The psychology of learning and motivation (Vol. 15, Social Psychology Bulletin, 27, 1164–1178.
pp. 201–236). New York: Academic Press. Sriram, N., Nosek, B. A., & Greenwald, A. G. (2007). Scale
Greenwald, A. G., Banaji, M. R., Rudman, L. A., Farnham, invariant contrasts of response latency distributions. Unpub-
S. D., Nosek, B. A., & Mellott, D. S. (2002). A unified theory lished manuscript, University of Virginia.
of implicit attitudes, stereotypes, self-esteem, and self- Unkelbach, C., Fiedler, K., Bayer, M., Stegmüller, M., & Danner,
concept. Psychological Review, 109, 3–25. D. (2008). Why positive information is processed faster: The
Greenwald, A. G., & Farnham, S. D. (2000). Using the Implicit density hypothesis. Journal of Personality and Social
Association Test to measure self-esteem and self-concept. Psychology, 95, 36–49.
Journal of Personality and Social Psychology, 79, West, S. G., Aiken, L. S., & Krull, J. L. (1996). Experimental
1022–1038. personality designs: Analyzing categorical by continuous
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). variable interactions. Journal of Personality, 64, 1–48.
Measuring individual differences in implicit cognition: The Zajonc, R. B. (1968). Attitudinal effects of mere exposure.
Implicit Association Test. Journal of Personality and Social Journal of Personality and Social Psychology, 9, 1–27.
Psychology, 74, 1464–1480.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Received November 11, 2008
Understanding and using the Implicit Association Test: I. An Revision received January 26, 2009
improved scoring algorithm. Journal of Personality and Accepted January 31, 2009
Social Psychology, 85, 197–216.
Hofmann, W., Gawronski, B., Gschwendner, T., Le, H., & Anthony G. Greenwald
Schmitt, M. (2005). A meta-analysis on the correlation
between the Implicit Association Test and explicit self-report Department of Psychology
measures. Personality and Social Psychology Bulletin, 31, University of Washington
1369–1385. Seattle
Inquisit 2.0 [Computer software]. (2005). Seattle, WA: Milli- WA 98195
second Software LLC. USA
Karpinski, A., & Steinman, R. B. (2006). The Single Category Tel. +206 543 7227
Implicit Association Test as a measure of implicit social Fax +206 685 3157
cognition. Journal of Personality and Social Psychology, 91, E-mail agg@u.washington.edu
16–32.

Experimental Psychology 2009; Vol. 56(4):283–294 Ó 2009 Hogrefe & Huber Publishers