Featured Paper from
Essential Science IndicatorsSM
According to the fifth bimonthly update of
Essential Science Indicators from
Reuters, the paper ranked at #4 among Psychiatry
& Psychology papers is "Measuring individual
differences in implicit cognition: The Implicit
Association Test" (Greenwald AG, McGhee DE, Schwartz
JLK, Journal of Personality and Social Psychology
74: 1464-80, June 1998), with 927
citations. In the Web of
Science®, this paper shows 1,025
The paper's lead author, Dr. Anthony Greenwald, is a Professor in the
Department of Psychology and Adjunct Professor of Marketing and
International Business at the University of Washington in Seattle.
In the interview below,
ScienceWatch.com talks with Dr. Greenwald about
this paper and its impact on the research
What factors prompted you to do this study?
In 1995, Mahzarin Banaji and I published a theoretical paper (also highly
cited, approaching 900 citations) that described an emerging field of
research that we named "implicit social cognition." The paper synthesized
an intriguing subset of work of the previous decade or so on
long-established social psychological topics of attitudes, stereotypes, and
self-esteem. Our new approach was to show that attitudes, stereotypes, and
self-esteem could operate outside of awareness (i.e., "implicitly").
In the paper (Greenwald AG, Banaji MR, " Implicit social cognition:
Attitudes, self-esteem, and stereotypes," Psychological Review,
102: 4-27, 1995; available on my
web page, we lamented the lack of measures that
could assess individual differences in this implicit aspect of social
cognition. In our research, then supported by National Science
Foundation, we were trying to develop such measures.
Would you sum up this paper's methods and findings for our
"Research of the last decade using
the IAT has yielded the conclusion that many
people possess a substantial collection of
'hidden' attitudes and
The procedure that became the IAT was one of several methods that were
being tried out in my lab at the University of Washington.
I programmed a computer to present a series of names of flowers and
insects. The task was to classify each as rapidly as possible. Flower names
required a key press with left hand, insect names a key press with right
hand. The next task was similar, but with a series of pleasant words (left
key) and unpleasant words (right key). I was the first subject, and I found
both of these tasks to be quite easy. Next I combined the two—left
key for either pleasant words or flower names, and right key for either
unpleasant words or insect names. I remember thinking that this "combined"
task was surprisingly easy. I didn’t understand why it should be so
easy to keep track of response assignments for four different categories.
The final task of the procedure was a combined task that differed from the
previous one just by switching the key assignments for flower and insect
names. This new task required left-key responses for either pleasant words
or insect names, and right key responses for either unpleasant words or
flower names. Immediately, I could see that this was a difficult task.
After completing it (slowly), my first thought was that if I just repeated
it a few times I would get good at it. An even greater surprise was the
discovery that practice didn't help.
It appeared that this new procedure was doing what we wanted for a measure
of associations corresponding to attitudes (in this case, attitudes toward
flowers and insects). What was happening was that, when assignment of
concepts to keys was consistent with established associations (i.e.,
pleasant and flower on the same key) performance could be fast. If the
assignments were at odds with established associations (flower and
unpleasant on the same key) performance would be slow. The
performance speed variations could work nicely as a measure of the
strengths of these evaluative (attitude) associations.
It was a relatively short step to construct a test in which the concepts of
Black and White race (represented by well known names of Black and White
celebrities) replaced flower names and insect names. I was again the first
subject at this task. I was both distressed and elated to discover that my
performance was much faster when White names and pleasant words had the
same response. The distress was because I was not expecting to find that I
had a stronger association of White than Black with pleasant. The elation
was because I could immediately see that this task was producing strong
enough effects so that it might work as a measure of the strengths of the
associations corresponding to race attitudes.
The co-authors of that first IAT article were Debbie McGhee and Jordan
Schwartz, both of whom were graduate students who helped in collecting the
data from the first experiments in 1994 and 1995, still a few years before
the name "Implicit Association Test" was given to the method.
How was the paper received by the research community when it was
I can't think of an occasion in which I have seen so many of my
psychologist colleagues pick up a new research method and start to make use
of it. Wide adoption of the method was helped because it required no more
equipment than a standard desktop computer. I am sure that part of the
interest others had in using the method was that many thought the task must
be producing such strong results because of something other than its
sensitivity to association strengths. Quite a few researchers conducted
experiments with the aim of showing that something else was artifactually
causing the strong effects observed with the method. At the same time, even
more saw the method’s potential to be used to measure associations on
a wide variety of socially relevant associations.
How does the IAT compare with other methods?
The IAT has some features in common with other "reaction time" measures
that have been used as indicators of automatic aspects of human cognition.
It also has the advantage of producing stronger effects than other tasks,
which provides the key to it being useful in measuring individual
differences. The IAT has properties somewhat like standard clinic
blood-pressure measures. It is subject to multiple influences, but
nevertheless has enough sensitivity and validity to be useful. Other
available techniques are less well validated and produce weaker effects,
leaving them less useful in detecting individual differences.
Ever since the IAT was developed, other researchers have been trying to
develop superior measures. I have been trying to do that too (without
success beyond producing a few modifications in the IAT that have increased
its research usefulness). Nevertheless, it is reasonable to expect that
some of those efforts will ultimately succeed.
Where have you taken this research since the publication of this
There are two main new developments. One is a meta-analysis of 10 years of
research on "predictive validity" of IAT measures—the ability to
predict interesting behavior. The meta-analysis is now in
press in Journal of Personality and Social Psychology, the
leading international journal of social psychology (Greenwald AG, Poehlman
TA, Uhlmann E, Banaji MR, "Understanding and using the Implicit Association
Test: III. Meta-analysis of predictive validity," Journal of
Personality and Social Psychology 2009.
"I can't think of an occasion in
which I have seen so many of my psychologist
colleagues pick up a new research method and
start to make use of it."
The other is a brief version of the IAT (in
press), which will make it easier to use IAT in
research conducted on the Internet and in surveys, for which the time
required by the original form of the measure was excessive (Sriram N,
Greenwald AG, "The Brief Implicit Association Test," Experimental
What would you like the "take-away lesson" about your work to
Research of the last decade using the IAT has yielded the conclusion that
many people possess a substantial collection of "hidden" attitudes and
stereotypes (i.e., associations measured by the IAT that don't match
self-reported measures of the same constructs—the difference
typically taking the form of the IAT measure showing stronger negative
attitudes and stereotypes). Research using the IAT has also established
that these hidden associations predicted discriminatory behavior (which is
presumably often unintended).
Many would like to know how to block effects of these associations on
behavior. Ideally, some simple procedure could eliminate or modify the
associations. But no such magic bullet has yet been found. For the present,
the best method of avoiding unintended discrimination resulting from hidden
biases is to be aware of possessing the hidden biases and to disrupt their
undesired effects on behavior.
Self-administration of the IAT is a good method of
assessing one’s own hidden biases. Disrupting effects of hidden
biases on behavior requires understanding of how various situations
allow the possibility of unintentional discrimination. Providing such
understanding is a reasonable and desirable goal of diversity
Anthony G. Greenwald, Ph.D.
Professor, Department of Psychology
Adjunct Professor, Marketing and International Business
University of Washington
Seattle, WA, USA
KEYWORDS: IMPLICIT COGNITION, IMPLICIT ASSOCIATION TEST,
AUTOMATIC ATTITUDE ACTIVATION, STEREOTYPES, SELF, VARIABILITY,
PREJUDICE, WORDS, IMPLICIT SOCIAL COGNITION, UNINTENDED