\\jciprod01\productn\H\HNR\28-1\HNR101.txt unknown Seq: 8 23-MAR-23 14:05
8
Harvard Negotiation Law Review
[Vol. 28:1
might be shown images of faces (say, Black and White faces) and
asked to match the faces with the words “good” or “bad,” or “pleasant”
or “unpleasant.” For a majority of White participants, it takes longer
to comply with the instruction to pair the Black face with the word
“good,” and it takes less time to pair the White face with the word
“good.” The IAT website has tests for race, gender, age, religion, ap-
pearance, and many other characteristics.
Use of the IAT has been controversial because of its limited “test-
retest reliability”: results for an individual can vary from one test ad-
ministration to the next.
24
Another area of imprecision is the “predic-
tive validity” of the IAT—the extent to which an unconscious bias
shown on the IAT correlates with explicit bias or biased behavior.
Meta-analyses of research on this question have shown that such a
correlation exists, but that it is only a partial correlation.
25
In other
words, while it is more likely that someone with unconscious bias will
behave in a biased manner or have explicit biases, researchers have
found some instances where people who have unconscious bias be-
have in a non-discriminatory manner.
26
However, we have not found
any research indicating that people with unconscious bias who be-
have in a non-discriminatory manner to one degree or another in a
controlled experiment can be counted on to behave that way
consistently.
Accordingly, the IAT is not yet sufficiently reliable to be used as
a screening device for hiring and similar purposes as applied to indi-
viduals. However, the test results in the
aggregate
are useful for
measuring the prevalence of unconscious associations in our soci-
ety.
27
And they are also useful as an awareness-raising tool in un-
conscious bias trainings because the experience of taking the test
brings to the level of conscious awareness the strength of our uncon-
scious associations. For example, one of the creators of the IAT, Prof.
Mahzarin Banaji, was “deeply embarrassed” by her first experience of
taking the IAT since, as a person of color, she did not think she could
24.
See
Gerald Guild,
The IAT: Questions of Reliability and Validity
, H
OW
D
O
Y
OU
T
HINK
? (Sept. 10, 2010), https://geraldguild.com/blog/2010/09/10/the-iat-ques-
tions-of-reliability-and-validity/,
archived at
https://perma.cc/47QU-4WYU (unac-
counted for variance suggests weak consistency).
25.
See
Anthony G. Greenwald et al.,
Understanding and Using the Implicit As-
sociation Test: III. Meta-Analysis of Predictive Validity
, 97 J. P
ERSONALITY
& S
OC
.
P
SYCH
. 17, 23 (2009).
26.
See id.
27. From time to time the IAT team publishes the aggregate results, such as
those reported in Tessa E.S. Charlesworth & Mahzarin R. Banaji,
Patterns of Implicit
and Explicit Attitudes: IV. Change and Stability from 2007 to 2020
, 33 P
SYCH
. S
CI
.
1347 (2022) (showing reduction in certain kinds of bias).