Published Ahead of Print on November 17, 2009, as 10.2105/AJPH.2009.159517
The latest version is at http://ajph.aphapublications.org/cgi/doi/10.2105/AJPH.2009.159517
RESEARCH AND PRACTICE
Combining Explicit and Implicit Measures of Racial
Discrimination in Health Research
Nancy Krieger, PhD, Dana Carney, PhD, Katie Lancaster, MS, Pamela D. Waterman, MPH, Anna Kosheleva, MS, and Mahzarin Banaji, PhD
A small but fast-growing body of public health
research is investigating the association between self-reported experiences of racial discrimination and population health.1–5 To date,
the strongest positive and generally linear associations have been observed for what are also the
most commonly studied outcomes: self-reported
psychological distress and self-reported health
behaviors (e.g., smoking, alcohol use, and other
drug use).1–5 By contrast, evidence for somatic
health outcomes, chiefly cardiovascular health,
has been more mixed, with studies variously
reporting linear, nonlinear, and no associations
between self-reported experiences of discrimination and health status.1–5
The largest published study on racial discrimination and risk of elevated blood pressure
reported a positive linear (dose–response) relationship among professional Blacks but a curvilinear association among working-class Blacks,
whose data revealed a J-shaped curve (i.e., blood
pressure was higher among respondents
reporting no discrimination than among those
reporting moderate discrimination and highest
among those reporting the most discrimination).6 This phenomenon of a linear relationship
among persons with more socioeconomic resources but a J-shaped curve among those with
fewer has been replicated in other studies on selfreported discrimination and health.7,8
One hypothesis that accounts for these results concerns what people are willing or able
to say.1,6 For persons with more socioeconomic
resources, a response of no discrimination may
more accurately capture the lack of personally
experienced discrimination, whereas among
persons with fewer socioeconomic resources, the
same response may reflect an accurate report of
no discrimination; a positive illusion, denial, or
internalized oppression that is not consciously
perceived; or a conscious decision not to report
the discrimination because it is uncomfortable or
dangerous to do so.1,6 The latter 2 scenarios
imply that a health risk may exist, via pathways
involving physiological and behavioral responses
Objectives. To improve measurement of discrimination for health research,
we sought to address the concern that explicit self-reports of racial discrimination may not capture unconscious cognition.
Methods. We used 2 assessment tools in our Web-based study: a new
application of the Implicit Association Test, a computer-based reaction-time test
that measures the strength of association between an individual’s self or group
and being a victim or perpetrator of racial discrimination, and a validated explicit
self-report measure of racial discrimination.
Results. Among the 442 US-born non-Hispanic Black participants, the explicit
and implicit measures, as hypothesized, were weakly correlated and tended to
be independently associated with risk of hypertension among persons with less
than a college degree. Adjustments for both measures eliminated the significantly greater risk for Blacks than for Whites (odds ratio = 1.4), reducing it to 1.1
(95% confidence interval = 0.7, 1.7).
Conclusions. Our results suggest that the scientific rigor of research on racism
and health will be improved by investigating how both unconscious and
conscious mental awareness of having experienced discrimination matter for
somatic and mental health. (Am J Public Health. Published online ahead of print
November 17, 2009: e1–e8. doi:10.2105/AJPH.2009.159517)
to racial discrimination as a psychosocial
stressor,1–5 with bodies revealing health effects of
exposures that can be subject to distortions in
perceptions, memory, and rationalizations.1,9 If
so, reliance solely on self-report measures of
racial discrimination, which has been standard to
date,1–5 may be problematic.1
The need to critically assess measurement of
exposure to discrimination is demonstrated by
what psychologists term person–group discrimination discrepancy (PGDD).10–15 It is welldocumented that people typically report more
discrimination for their group than for themselves personally, even though, statistically, the
group level cannot be high if every individual’s is
low. PGDD is postulated to arise from an
automatic protective mechanism,16 reflecting
a larger psychological tendency to view and
present oneself positively, even when such denial
may not be in one’s ultimate self-interest.17–19
During the past decade, social psychologists,
borrowing basic methods of cognitive psychology, have developed a new approach—the
Implicit Association Test (IAT), a computerbased reaction-time measure—to study
Published online ahead of print November 17, 2009 | American Journal of Public Health
phenomena for which self-report data might
not fully capture what people think and
feel.20–23 An outgrowth of several decades of
experimental research in the social, cognitive,
clinical, and neuropsychological sciences concerned with the general architecture of the
human mind, human learning, and unconscious
mental processes,20,24–28 the IAT is now one of
the most robust and widely employed measures
in social, cognitive, and even clinical psychology;
it is used to assess the ease with which the mind
makes associations. Building on the central finding that learning involves changes in neural
function of different neurons that are active at
the same time, the underlying cognitive principle
is that concepts in the mind that are more closely
associated with each other are more closely
linked. These associations can occur both for
conscious cognitive processes and for unconscious mental processes that lie beyond the reach
of introspective access, with evidence indicating
that implicit (unconscious) associations can form
about many different phenomena—the self, other
people, places, animals, memories, fantasies, inanimate objects, and nontangible ideas.20,27
Krieger et al. | Peer Reviewed | Research and Practice | e1
RESEARCH AND PRACTICE
For example, an IAT could measure how
much a person prefers flowers to bugs by
contrasting the time it takes to make associations
between the word pairs flowers and good and
bugs and bad and then comparing what happens
when participants are asked to pair flower with
bad and bugs with good. A difference in average
matching speed for opposite pairings determines the IAT score, a measure of strength of
association. Participants are typically aware that
they are making these connections but are
unable to control them because of the rapid
response times and the structure of the test. IAT
methods are well-described in the social psychology literature,21,22 and programming resources to develop IATs are available online.29
To address extant questions about measuring experiences of discrimination,1–5,30,31 we
employed a novel application of the IAT to assess
unconscious cognition about discrimination and
consider the implications of using both explicit
(self-report) and implicit measures for research
on racism and health. Our work builds on and
extends research that used the IAT to study
racial prejudice and stereotypes.32–35 Recent
health-related studies found that the IAT for
racial prejudice can predict physicians’ clinical
decisions.36,37 Focusing on the somatic health of
Black Americans, we hypothesized that the
explicit and implicit measures would be weakly
associated with each other (because of their
varying abilities to measure unconscious processes) and independently associated with risk of
hypertension (as modified by socioeconomic
position) and with the greater hypertension risk
among Blacks than Whites in the United States.
METHODS
We recruited our study population in 2007
and 2008 from the Project Implicit Web site,
the leading public Web site for online IAT
research.23 We restricted study participants to
US-born non-Hispanic persons aged 25 to 70
years who self-identified as being either (1) Black
or African American or (2) White, chosen
because these are the 2 groups most often
studied in research on racism and cardiovascular
health. We imposed the nativity and Hispanic
restrictions because evidence indicates that USborn and immigrant populations may differently
understand, experience, and respond to US racial
discrimination.1–4,30
In a Web-based study, it is not possible to
determine the response rate (because we cannot know the number of eligible persons who
viewed the Web site but did not take the test).
A total of 442 Black and 1018 White persons
reported that they met the study eligibility
criteria and participated in the study. Average
completion time for the survey was 15 minutes,
of which the IAT required 10 minutes.
Measures
Sociodemographic questions. We asked about
participants’ age, race/ethnicity, nativity, gender, and educational level. We also asked about
the educational level of their mother and
father, as a measure of childhood socioeconomic position.38,39
Explicit measures of exposure to racial
discrimination and response to unfair treatment.
We used the validated 9-item self-report Experiences of Discrimination (EOD) instrument
(available without charge at http://www.
hsph.harvard.edu/faculty/NancyKrieger.
html).6,40–42 This measure asks participants
whether they have ever ‘‘experienced discrimination, been prevented from doing something, or
been hassled or made to feel inferior in any of
the following situations on account of your race,
ethnicity or color,’’ and about the frequency of
occurrence within each situation. The 9 situations are at school; getting hired or getting a job;
at work; getting housing; getting medical care;
getting service in a store or restaurant; getting
credit, bank loans, or a mortgage; on the street
or in a public setting; or from the police or in
the courts. Each ‘‘yes’’ reply was scored as 1; the
total EOD situation score range was 0 to 9.
Previous research,6,40–42 including on blood
pressure, informed our categorization of the
EOD situation measure as 0= no discrimination,
1 to 2 =moderate discrimination, or 3–9 =high
discrimination.
Two single-item global measures assessed
how frequently (never, rarely, sometimes, or
often) the respondents felt they had been
discriminated against because of their ‘‘race,
ethnicity, or color’’ (personal) and how often
they felt that racial/ethnic groups who are not
White are discriminated against (group);
these questions were previously used in the
EOD validation study.41 Two items asked
about response to unfair treatment, with the
following categories: (1) ‘‘accept it as a fact of life’’
e2 | Research and Practice | Peer Reviewed | Krieger et al.
versus ‘‘try to do something about it,’’ and (2)
‘‘talk to other people about it’’ versus ‘‘keep it
to yourself.’’6,40–42 To control for how selfpresentation might affect the self-report
measures,41,42 we used a 5-item validated social
desirability scale.43
Implicit measures of exposure to racial
discrimination. As shown in Figure 1a, the IAT
was introduced by anchoring language explicitly addressing whether the participant
had been a target of discriminatory behavior.
It employed 2 sets of targets, (1) self (referred
to as IAT-person) and (2) group (IAT-group),
and 2 sets of attribute categorization terms: (1)
‘‘abuser,’’ ‘‘racist,’’ and ‘‘bigot,’’ and (2) ‘‘target,’’ ‘‘victim,’’ and ‘‘oppressed.’’ These attribute terms were derived from pilot studies we
conducted with Boston-based participants
who were college students, persons recruited
on the street, or members of a community
health center.44 As shown in Figure 1, the IAT’s
core contrast concerned how quickly participants
linked words or images that pertained to self
or to their group to words or images that
pertained to being a victim or perpetrator of
racial discrimination. The difference in speed
(in milliseconds) for the 2 associations produced
the raw IAT score, which we then normalized
by following a standard IAT protocol.22 A low
score indicated that a participant felt equally like
a victim and bigot, whereas a high score indicated the participant felt more like a victim
than a bigot.
Health data. We used questions on selfreported health status from the US National
Health Interview Survey45 for hypertension and
other health outcomes, such as smoking, diabetes, and psychological distress (measured with
the K6 scale: range = 0–24; scores ‡13 categorized as high psychological distress46,47). We
classified participants as being hypertensive if
they responded affirmatively to any of the
following questions: ‘‘Have you ever been told by
a doctor or other health professional that you
had hypertension, also called high blood pressure?’’ ‘‘Because of your high blood pressure/
hypertension, have you ever been told to take
prescription medicine?’’ ‘‘Are you now taking
prescribed medicine?’’ To address potential confounding, we obtained data on body mass index
(BMI; defined as weight in kilograms divided by
height in meters squared), computed from selfreported height and weight.
American Journal of Public Health | Published online ahead of print November 17, 2009
RESEARCH AND PRACTICE
FIGURE 1—Visual representation of the IAT test for being a target versus a perpetrator of racial discrimination showing (a) the target and
attribute categories employed in the 5 blocks of the IAT and (b) the contrast of associations that are easier versus harder to make for persons
who more strongly believe that Blacks are more discriminated against than Whites.
Published online ahead of print November 17, 2009 | American Journal of Public Health
Krieger et al. | Peer Reviewed | Research and Practice | e3
RESEARCH AND PRACTICE
Statistical Analyses
Our statistical analyses, performed in SAS,48
involved 3 steps. First, we described the distribution of the selected exposure and outcome
measures plus related covariates. To put these
data in context, we obtained US national data on
the corresponding variables.49,50 Second, we
quantified the correlations between the different
measures of racial discrimination and computed
the PGDD for both the explicit and implicit
measures. Third, we used logistic regression
models to examine risk of hypertension in relation to the explicit and implicit measures of
exposure to racial discrimination, with controls
for relevant covariates.
RESULTS
TABLE 1—Sociodemographic Characteristics, Explicit and Implicit Measures of Racial
Discrimination, and Health Status Among US-born Black Participants: 2007–2008
Black Participants
a
Age, y, mean (SD)
Women
Men
36 (9)
38 (10)
External Comparisons
Total Population
Women
Men
Education,b %
< High school
1.0
0.0
US = 19.5
US = 21.8
41.1
57.8
34.8
65.2
US = 62.4
US = 18.1
US = 64.7
US = 15.4
< Bachelor’s degree
68.6
72.1
‡ Bachelor’s degree
31.4
28.0
< Bachelor’s degree
77.0
68.4
‡ Bachelor’s degree
23.0
31.6
‡ High school
‡ Bachelor’s degree
Mother’s education, %
Father’s education, %
Explicit measures of racial discrimination
Table 1 shows the distribution of the sociodemographic, discrimination, and health characteristics of the 442 Black participants. The
average age of the participants ranged from
36 years (women) to 38 years (men), and they
were highly educated and had better health
than the US Black population overall; they also
were much less likely to report no experiences
of discrimination than were working-class and
less-educated samples of Black Americans.6,44,49,50 For example, 60% of study participants had a bachelor’s degree or higher
(versus 17% among the US Black population
aged 25 years and older in 200650), 23% were
hypertensive (versus 40% among the US Black
population aged 20–74 years in 2001–200449),
and only 3.6% reported never having experienced racial discrimination (versus 33%
among the Black participants in the United for
Health cohort of low-income, working-class
employees in the greater Boston area in 2003
to 200441,42).
The nonequivalence of the explicit and
implicit measures is also shown by our divergent PGDD results (Table 1). The explicit
measure produced the expected gap: participants reported more discrimination for their
group than for themselves personally (P < .001).
By contrast, the implicit measure produced no
such gap (women, P > .80; men, P > .28).
Moreover, as shown in Table 2, the explicit
EOD situation score and the 2 implicit measures, the IAT-person and the IAT-group, were,
as expected, weakly correlated (Pearson correlation coefficients < 0.17).
e4 | Research and Practice | Peer Reviewed | Krieger et al.
EOD situationsc
0, %
1–2, %
‡ 3, %
4.3
2.2
UH = 33 C = 20
UH = 32 C = 23 UH = 33 C = 16
18.8
15.9
UH = 22 C = 28
UH = 30 C = 29 UH = 18 C = 27
UH = 45 C = 52
UH = 37 C = 48 UH = 48 C = 57
77.0
81.9
4.5 (2.1)
4.2 (2.2)
Global-person (range = 1–4), mean (SD)
2.8 (0.6)
2.7 (0.6)
Global-group (range = 1–4), mean (SD)
3.7 (0.5)
3.6 (0.5)
Global PGDDd Wilcoxon
signed rank test (P value)
–12 327.5
(<.001)
–2758.5
(<.001)
Score, mean (SD)
Implicit measures of racial discrimination
IAT-person
0.37 (0.35) 0.33 (0.38)
IAT-group
0.40 (0.41) 0.40 (0.41)
IAT PGDDe paired t test (P value)
–0.25 (.802) –1.08 (.28)
Response to unfair treatmentc
Talk to others/take action
61.1
60.3
UH = 49 C = 75 UH = 42 C = 69
Talk to others/accept as fact of life
Keep to self/take action
31.5
1.7
23.5
5.2
UH = 27 C = 19 UH = 25 C = 18
UH = 8 C = 3
UH = 10 C = 5
UH = 12 C = 4
UH = 20 C = 8
US = 40
US = 38
US = 19
US = 25
keep to self/accept as fact of life
Social desirability
Hypertensionf
High psychological distressg
Current smokerh
Diabetesi
5.7
11.0
3.8 (0.6)
3.7 (0.6)
21.3
26.7
0.0
0.0
18.0
14.6
6.3
5.8
US = 4.2
US = 11.6
Note. C = Coronary Artery Risk Development in Young Adults (CARDIA) study; EOD = Experiences of Discrimination instrument;
IAT = Implicit Association Test; PGDD = person-group discrimination discrepancy; UH = United for Health study; US = US
population. Sample size for women was n = 304; for men, n = 138. Missing data < 1% for all variables other than smoking
(< 5%); all values derived from observed (nonmissing) data.
a
Range = 25–70 years
b
Comparison data are for US population, 2006 (aged ‡ 25 years).50
c
Comparison data are from United for Health study, 2003 to 2004,41,42 and CARDIA study, 1992 to 1993.6
d
Comparison data are from Web-based study internal comparison of PGDD (global-person minus global-group).
e
Comparison data are from Web-based study internal comparison of PGDD (IAT-person minus IAT-global).
f
Comparison data are for US population, 2001 to 2004 (aged 20–74 years, age-adjusted).49
g
Comparison data are for US population, 2005 to 2006 (aged ‡ 18 years, age-adjusted, not stratified by gender).49
h
Comparison data are for US population, 2006 (aged ‡ 25 years, age-adjusted).49
i
Comparison data are for US population, 2001 to 2004 (aged ‡ 20 years, age-adjusted, not stratified by gender).49
American Journal of Public Health | Published online ahead of print November 17, 2009
RESEARCH AND PRACTICE
TABLE 2—Correlations Between Explicit and Implicit Measures of Racial Discrimination Among US-born Male
and Female Black Participants, Overall and Stratified by Educational Level: 2007–2008
Women
IAT-Person,
Pearson Correlation
Coefficient (P Value)
IAT-Group,
Pearson Correlation
Coefficient (P Value)
Men
EOD,
Pearson Correlation
Coefficient (P Value)
IAT-Person,
Pearson Correlation
Coefficient (P Value)
IAT-Group,
Pearson Correlation
Coefficient (P Value)
EOD,
Pearson Correlation
Coefficient (P Value)
Total populationa
IAT-person
1.000
IAT-group
0.175 (.002)
1.000
1.000
EOD score
0.061 (.293)
0.152 (.008)
0.102 (.224)
1.000
–0.034 (.693)
1.000
0.189 (.120)
0.067 (.650)
1.000
0.075 (.381)
1.000
1.000
0.068 (.647)
1.000
Less than a Bachelor’s degreeb
IAT-person
1.000
IAT-group
EOD score
0.152 (.087)
0.063 (.481)
1.000
1.000
0.168 (.059)
Bachelor’s degree or higherc
IAT-person
1.000
IAT-group
0.210 (.006)
1.000
1.000
EOD score
0.056 (.466)
0.128 (.093)
0.054 (.612)
1.000
–0.106 (.322)
1.000
0.067 (.528)
1.000
Note. EOD = Experiences of Discrimination instrument; IAT = Implicit Association Test.
a
Women, n = 302; men, n = 183.
b
Women, n = 127; men, n = 48.
c
Women, n = 174; men, n = 90.
Table 3 first provides results for analyses,
stratified by educational level, that assess the
association between the explicit and implicit
measures with risk of hypertension among the
Black participants (crude and adjusted for
relevant covariates). It then shows the effect of
adjusting for these measures on Black–White
comparisons of hypertension risk. The first set
of these analyses revealed different associations by educational level. Among the Black
participants with less than a bachelor’s degree,
the EOD and IAT-person in the fully adjusted
model (model 4) tended to be independently
associated with hypertension. For the EOD,
the odds of having hypertension were 2.3
times as high among participants reporting no
discrimination as they were among those
reporting moderate discrimination (albeit with
wide confidence intervals [CIs]; 95% CI = 0.1,
39.7) and 4.1 times as high (95% CI = 0.8,
20.5) among persons reporting high as they
were among those reporting moderate discrimination, suggestive of a J-shaped curve.
The odds of being hypertensive likewise
tended to increase positively with the IATperson score (odds ration [OR] = 2.2; 95%
CI = 0.7, 6.8). By contrast, we found no
associations between risk of hypertension and
either the IAT or EOD score among participants with a bachelor’s degree or higher.
Finally, even within our healthy study population, the odds of being hypertensive were
1.4 times as high among the Black as they were
among the White participants in both the
bivariate analysis (model 1; 95% CI =1.1, 1.9)
and the analyses adjusted for sociodemographic factors and BMI (model 2; 95%
CI =1.0, 1.9).
Contributions from both conscious and unconscious awareness of racial discrimination to
these observed disparities were suggested by
analyses that compared the Black–White risk
after further taking into account the EOD and
IAT data. The distribution among the 1077
White participants of EOD scores was 48%
for 0; 36% for 1 or 2; and 16% for 3 or higher.
Among Whites, the IAT-person scores had
a mean 6SD = 0.11 60.35, and the IATgroup scores had a mean 6SD = 0.07 60.40.
Additional controls for the explicit and implicit
measures eliminated the excess risk among
Black respondents and reduced the OR to
a nonsignificant 1.1 (model 3; 95%
CI = 0.7, 1.7).
Published online ahead of print November 17, 2009 | American Journal of Public Health
DISCUSSION
Our study makes 2 useful contributions to
improving the science of studying racial discrimination and health. First, we demonstrated the feasibility of adapting the IAT for
population research on experiences of discrimination. Second, we provided evidence
that explicit and implicit measures of racial
discrimination are not only distinct but also
may both matter for health, albeit in ways that,
as predicted, vary by socioeconomic position.
Specifically, we found that among Black participants, regardless of educational level, the
explicit and implicit measures of racial discrimination were, as hypothesized, weakly
correlated, and the implicit measures did not
exhibit the personal–group discrimination
discrepancy of the explicit measures. Both the
explicit and implicit measures tended to be
independently associated with risk of hypertension among less-educated Black participants, and jointly controlling for both the
explicit and implicit measures eliminated the
observed significantly greater risk for hypertension among Black than among White respondents.
Krieger et al. | Peer Reviewed | Research and Practice | e5
RESEARCH AND PRACTICE
TABLE 3—Explicit and Implicit Measures of Racial Discrimination and Their Association With Risk of Hypertension
Among US-born Black Participants, Stratified by Educational Level and effect of discrimination
on Black versus White risk of hypertension: 2007–2008
Discrimination Measure
Model 1, OR (95% CI)
Model 2, OR (95% CI)
Model 3, OR (95% CI)
Model 4, OR (95% CI)
Less than a Bachelor’s degree (Blank participants only; n = 154)
EOD score
0
2.9 (0.2, 39.7)
2.1 (0.1, 34.9)
...
2.3 (0.1, 39.7)
1–2 (Ref)
>3
1.00
3.8 (0.9, 17.2)
1.00
4.6 (0.9, 23.0)
...
1.00
4.1 (0.8, 20.5)
IAT-person
2.1 (0.7, 5.9)
...
2.5 (0.8, 7.6)
2.2 (0.7, 6.8)
IAT-group
1.4 (0.6, 3.5)
...
1.0 (0.4, 2.6)
1.0 (0.4, 2.8)
1.2 (0.2, 9.3)
Bachelor’s degree or higher (Blank participants only; n = 243)
EOD score
1.5 (0.2, 8.8)
1.2 (0.2, 9.4)
...
1.00
1.00
...
1.00
‡3
IAT-person
1.1 (0.5, 2.4)
0.6 (0.3, 1.4)
1.2 (0.5, 2.7)
...
...
0.8 (0.3, 1.9)
1.2 (0.5, 2.9)
0.7 (0.3, 1.9)
IAT-group
0.7 (0.3, 1.5)
...
0.7 (0.3, 1.7)
0.7 (0.3, 1.7)
Blacks vs Whitesa
Effects of adjusting for implicit and explicit discrimination on Black versus White risk
1.4 (1.1, 1.96)
1.4 (1.0, 1.9)
1.1 (0.7, 1.7)
0
1–2 (Ref)
...
Note. OR = odds ratio; CI = confidence interval; EOD = Experiences of Discrimination instrument; IAT = Implicit Association Test. EOD scores are explicit measures, and IAT values are implicit measures.
Model 1 used bivariate regression. Models 2, 3, and 4 used multivariate regression and adjusted for gender, age, body mass index, social desirability, response to unfair treatment, mother’s
education, and father’s education.
a
For Blacks, n = 397; for Whites, n = 1018. For this comparison, model 1 used bivariate regression; model 2 used multivariate regression and adjusted for age, gender, body mass index, and
education (respondent’s, mother’s, and father’s); and model 3 used multivariate regressions and adjusted for model 2 factors plus implicit and explicit measures of racial discrimination, response
to unfair treatment, and social desirability.
Limitations
Our results are subject to several caveats.
First, the data were cross sectional, limiting
causal inference; the cross-sectional weak correlations between the explicit and implicit
measures nevertheless provided important evidence that they assessed different experiences.
Second, our Black participants had a much
higher level of education and better health
status than does the larger US Black population,
a reflection of who uses the Internet (lowerincome persons less than higher-income persons and Blacks and Hispanics less than
Whites, even at the same income level51). This
constrained socioeconomic heterogeneity in our
sample reduced the likelihood of observing
exposure–outcome associations and hence narrowed the range of variation in the study
covariates.42,52 It also contributed to a very low
prevalence of explicit self-reports of no discrimination, so that we could not test hypotheses
about whether IAT scores varied by sociodemographic characteristics or health status among
persons explicitly reporting no discrimination.
Third, we relied on self-report data for
hypertension. Validation studies, however,
show that the US National Health Interview
Survey45 questions we used yield estimates of
hypertension prevalence highly correlated with
those based on blood pressure measurement,
among both Black and White Americans.53 The
validity of our hypertension analyses was also
bolstered because we not only stratified by the
participant’s educational level but also controlled
for several key covariates (i.e., age, gender, and
BMI, as well as the educational level of the
participants’ mother and father, thereby capturing important aspects of both early life and adult
socioeconomic position, both of which are related to increased risk of adult hypertension54).
A further concern is whether the IAT accurately measures exposure to racial discrimination, as directed at the self (IAT-person) and
group (IAT-group). The test explicitly referred
to discriminatory behavior, used target terms
that explicitly referred to discrimination (e.g.,
‘‘victim of discrimination,’’ ‘‘feels effects of
prejudice’’). It empirically assessed whether
e6 | Research and Practice | Peer Reviewed | Krieger et al.
people were more likely to think that they
themselves and the specified group were more
likely to be the target than perpetrator of
discrimination, with the detected difference in
association times posited to reflect the existence of neural pathways shaped by previous
experience with racial discrimination. Whether
this previous experience pertains to actual
experience, as specified in the anchoring language, rather than a more general fear, however, remains unknown.
A related and now resolved controversy
concerned whether the IAT actually measures
people’s own attitudes rather than larger cultural attitudes.55 This criticism has been put to
the test in the empirical literature, with the
evidence indicating that the IAT in fact measures
people’s own attitudes. Specifically, a new comprehensive meta-analytic review of the IAT and
its correlates unequivocally found, especially in
the domain of racial prejudice and stereotyping,
that the IAT systematically predicts real, personal, important, and clinically significant outcomes.56
American Journal of Public Health | Published online ahead of print November 17, 2009
RESEARCH AND PRACTICE
Implications
The preliminary nature of our findings notwithstanding, we believe they have several
implications for research on racial discrimination and health. First, they suggest the need for
a robust research agenda investigating the
properties of IATs designed to measure different types of experiences of discrimination.
Topics warranting research include (1) whether
these IATs capture individuals’ actual experiences instead of fears of discrimination, including specific types of discrimination rather than
unfair treatment more generally; (2) their correlations (or lack thereof) with explicit measures of discrimination; and (3) their associations with diverse health outcomes. Such
research should also take into account that
mental awareness of racial discrimination is
only 1 component of how racism can harm
health, because of the evidence that discrimination can increase the risk of adverse health
behaviors (e.g., cigarette smoking, excessive
alcohol consumption, adverse drug use, and
harmful ‘‘comfort’’ eating) that can harm health
independent of any physiological stress response1–5 and can involve additional pathways,
such as social and economic deprivation; adverse
exposure to toxins, pollutants, and pathogens;
targeted marketing of licit and illicit harmful
substances (e.g., tobacco and alcohol); and inadequate and degrading medical care.1
Second, our findings confirm that the term
‘‘perceived discrimination’’ (frequently used in
the literature on racism and health2–5,31) should
not be conflated with self-reports of racial discrimination. Instead, as suggested by the weak
correlations between our IAT and EOD scores,
these constructs are distinct, not synonymous,
and should not be used interchangeably.1,6,41
Self-reports are what people are willing or able
to report, and this may not be identical to what
they consciously, let alone unconsciously, perceive.1,41,42
Third, our results underscore the importance of triangulating evidence derived from
the stories that bodies tell and what people
say.1,9,57 The more straightforward linear associations observed between self-reported experiences of racial discrimination and self-reported
mental health and health behaviors versus those
observed between self-reports and measured
somatic health status might, for example, reflect
(1) how the former associations depend on
shared mental processes underlying self-reports
and (2) how current conditions, regardless of
previous conditions, can affect current mental
state and health behaviors.58–60 By contrast, the
measurement of somatic health outcomes (many
of which, other than acute infections, involve
extended etiologic periods), is not affected by
self-report cognitive mechanisms and is therefore
able to detect nonlinear associations, if extant.
Other research has likewise found discordant
results for associations of psychosocial stressors
with self-reported versus measured health
status,61 with reasons for these discrepancies
a topic of active debate and investigation.62,63
Our study provides provocative evidence
that the scientific rigor of research on racism
and health will be improved by investigating
how both unconscious and conscious mental
awareness of having experienced discrimination matter for somatic and mental health.
Research on the validity and utility of jointly
using explicit and implicit measures of racial
discrimination—such as the EOD along with
the IATs we have developed—should be conducted in more representative and larger
study populations and in relation to a range
of mental health, behavioral, and somatic
health outcomes. j
About the Authors
Nancy Krieger, Pamela D. Waterman, and Anna Kosheleva
are with the Department of Society, Human Development,
and Health, Harvard School of Public Health, Boston, MA.
Dana Carney is with the Graduate School of Business,
Columbia University, New York, NY. Katie Lancaster is
with Radboud University Nijmegen, Netherlands. Mahzarin
Banaji is with the Department of Psychology, Harvard
University, Cambridge, MA.
Correspondence should be sent to Nancy Krieger, PhD,
Dept of Society, Human Development, and Health, Harvard
School of Public Health, Kresge 717, 677 Huntington Ave,
Boston, MA 02115 (e-mail: [email protected]).
Reprints can be ordered at http://www.ajph.org by clicking
the ‘‘Reprints/Eprints’’ link.
This article was accepted March 28, 2009.
Contributors
N. Krieger originated the study, oversaw data analysis,
and led writing of the article. D. Carney devised the IAT
and oversaw implementation of the Web-based study.
K. Lancaster technically implemented the study’s Webbased materials, managed the online data collection, and
assisted in preliminary data analysis. P. D. Waterman
and A. Kosheleva assisted with database management and
analyzed the data. M. Banaji guided development and
interpretation of the IAT measures. All authors helped to
conceptualize ideas, interpret findings, and review drafts
of the article.
Published online ahead of print November 17, 2009 | American Journal of Public Health
Acknowledgments
This study was funded by the Harvard/Robert Wood
Johnson Foundation Health & Society Program.
Human Participant Protection
This study was approved by the institutional review
board of Harvard University, and all participants gave
their informed consent.
References
1. Krieger N. Embodying inequality: a review of
concepts, measures, and methods for studying health
consequences of discrimination. Int J Health Services.
1999;29:295–352. (Republished and updated: Krieger
N. Discrimination and health. In: Berkman L, Kawachi I,
eds. Social Epidemiology. Oxford, UK: Oxford University
Press; 2000:36–75.)
2. Williams DR, Neighbors HW, Jackson JS. Racial/
ethnic discrimination and health: findings from community studies. Am J Public Health. 2003;93:200–208.
3. Paradies Y. A systematic review of empirical research on self-reported racism and health. Int J Epidemiol.
2006;35:888–901.
4. Mays VM, Cochran SD, Barnes NW. Race, racebased discrimination, and health outcomes among African Americans. Annu Rev Psychol. 2007;58:201–225.
5. Williams DR, Mohammed SA. Discrimination and
racial disparities in health: evidence and needed research.
J Behav Med. 2009;32:20–47.
6. Krieger N, Sidney S. Racial discrimination and blood
pressure: the CARDIA study of young Black and White
adults. Am J Public Health. 1996;86:1370–1378.
7. Caughy MO, O’Campo PJ, Muntaner C. Experiences
of racism among African American parents and the
mental health of their preschool-aged children. Am J
Public Health. 2004;94:2118–2124.
8. Huebner DM, Davis MC. Perceived antigay discrimination and physical health outcomes. Health Psychol.
2007;26:627–634.
9. Krieger N. Embodiment: a conceptual glossary for
epidemiology. J Epidemiol Community Health. 2005;
59:350–355.
10. Crosby FJ. Relative Deprivation and Working
Women. New York, NY: Oxford University Press; 1982.
11. Crosby F. The denial of personal discrimination. Am
Behav Sci. 1984;27:371–386.
12. Taylor DM, Wright SC, Moghaddam FM, Lalonde
RN. The personal/group discrimination discrepancy. Pers
Soc Psychol Bull. 1990;16:254–262.
13. Taylor DM, Wright SC, Porter LE. Dimensions of
perceived discrimination: the personal/group discrimination discrepancy. In: Zanna MP, Olson JM, eds. The
Psychology of Prejudice: The Ontario Symposium. Vol. 7.
Hillsdale, NJ: Erlbaum; 1994:223–255.
14. Quinn KA, Roese NJ, Pennington GL, Olson JM. The
personal/group discrimination discrepancy: the role of
informational complexity. Pers Soc Psychol Bull. 1999;
25:1430–1440.
15. Olson JM, Roesesc NJ, Meen J, Robertson DJ. The
preconditions and consequences of relative deprivation:
two field studies. J Appl Soc Psychol. 1995;25:944–964.
16. Hodson G, Esses VM. Distancing oneself from
negative attributes and the personal/group
Krieger et al. | Peer Reviewed | Research and Practice | e7
RESEARCH AND PRACTICE
discrimination discrepancy. J Exp Soc Psychol. 2002;38:
500–507.
17. Holtgraves T. Social desirability and self-reports:
testing models of socially desirable responding. Pers Soc
Psychol Bull. 2004;30:161–172.
18. Orne MT. On the social psychology of the psychological experiment: with particular reference to demand
characteristics and their implications. Am Psychol.
1962;17:776–783.
19. Rosenberg M. The conditions and consequences of
evaluation apprehension. In: Rosenthal R, Rosnow RL,
eds. Artifact in Behavioral Research. New York, NY:
Academic Press; 1969:279–349.
20. Greenwald AG, Banaji MR. Implicit social cognition:
attitudes, self-esteem, and stereotypes. Psychol Rev.
1995;102:4–27.
21. Greenwald AG, McGhee DE, Schwartz JL. Measuring individual differences in implicit social cognition: the
implicit association test. J Pers Soc Psychol. 1998;74:
1464–1480.
22. Greenwald AG, Nosek BA, Banaji MR. Understanding and using the Implicit Association Test: I. An
improved scoring algorithm. J Pers Soc Psychol. 2003;
85:197–216.
23. Project Implicit. Available at: https://implicit.
harvard.edu/implicit/research. Accessed March 11,
2009.
24. Hebb DO. The Organization of Behavior: A Neuropsychological Theory. New York, NY: Wiley; 1949.
25. Bliss TV, Lomo T. Long-lasting potentiation of
synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant
path. J Physiol. 1973;232:331–356.
26. Nisbett RE, Wilson TD. Telling more than we can
know: verbal reports on mental processes. Psychol Rev.
1977;84:231–259.
27. Fazio RH. How do attitudes guide behavior? In:
Sorrentino RM, Higgins ET, eds. Handbook of Motivation
and Cognition: Foundations of Social Behavior. New York,
NY: Guilford; 1986:204–243.
28. Stanley D, Phelps EA, Banaji MR. The neural basis
of implicit attitudes. Perspect Psychol Sci. In press.
29. Greenwald AG. Implicit association test materials.
Available at: http://faculty.washington.edu/agg/iat_
materials.htm. Accessed March 11, 2009.
30. Blank RM, Dabady M, Citro CF, eds. Measuring
Racial Discrimination. Washington, DC: National Academies Press; 2004.
31. Kressin NR, Raymond KL, Manze M. Perceptions of
race/ethnicity-based discrimination: a review of measures and evaluation of their usefulness for the health
care setting. J Health Care Poor Underserved. 2008;19:
697–730.
32. Fazio RH, Jackson JR, Dunton BC, Williams CJ.
Variability in automatic activation as an unobtrusive
measure of racial attitudes: a bona fide pipeline? J Pers
Soc Psychol. 1995;69:1013–1027.
33. Dovidio JF, Kawakami K, Johnson C, Johnson B,
Howard A. On the nature of prejudice: automatic and
controlled processes. J Exp Soc Psychol. 1997;33:510–
540.
34. Nosek BA, Smyth FL, Hansen JJ, et al. Pervasiveness
and correlates of implicit attitudes and stereotypes. Eur
Rev Soc Psychol. 2007;18:36–88.
35. Dunham Y, Baron AS, Banaji MR. From American
city to Japanese village: the omnipresence of implicit race
attitudes. Child Dev. 2006;77:1268–1281.
54. Lawlor DA, Smith GD. Early life determinants of
adult blood pressure. Curr Opin Nephrol Hypertens.
2005;14:259–264.
36. Green AR, Carney DR, Pallin DJ, et al. The presence
of implicit bias in physicians and its prediction of
thrombolysis decisions for Black and White patients.
J Gen Intern Med. 2007;22:1231–1238.
55. Olson MA, Fazio RH. Reducing the influence of
extra-personal associations on the Implicit Association
Test: personalizing the IAT. J Pers Soc Psychol. 2004;
86:653–667.
37. Sabin JA, Rivara FP, Greenwald AG. Physician
implicit attitudes and stereotypes about race and quality
of medical care. Med Care. 2008;46:678–685.
56. Greenwald AG, Poehlman TA, Uhlmann E, Banaji
MR. Understanding and using the Implicit Association
Test: III. Meta-analysis of predictive validity. J Pers Soc
Psychol. In press.
38. Krieger N, Williams D, Moss N. Measuring social
class in US public health research: concepts, methodologies and guidelines. Annu Rev Public Health. 1997;
18:341–378.
39. Shaw M, Galobardes B, Lawlor DA, Lynch J,
Wheeler B, Davey Smith G. The Handbook of Inequality
and Socioeconomic Position: Concepts and Measures.
Bristol, UK: Policy Press; 2007.
40. Krieger N. Racial and gender discrimination: risk
factors for high blood pressure? Soc Sci Med. 1990;
30:1273–1281.
41. Krieger N, Smith K, Naishadham D, Hartman C,
Barbeau EM. Experiences of discrimination: validity and
reliability of a self-report measure for population health
research on racism and health. Soc Sci Med. 2005;
61:1576–1596.
42. Krieger N, Chen JT, Waterman PD, et al. The
inverse hazard law: blood pressure, sexual harassment,
racial discrimination, workplace abuse and occupational
exposures in US low-income Black, White, and Latino
workers. Soc Sci Med. 2008;67:1970–1981.
43. Hays RD, Hayashi T, Stewart AT. A five-item
measure of socially desirable response set. Educ Psychol
Meas. 1989;49:629–636.
44. Carney DR, Krieger N, Banaji MR. Implicit measures
reveal evidence of personal discrimination. Self Ident. In
press.
57. Krieger N, Davey Smith G. Bodies count and body
counts: social epidemiology and embodying inequality.
Epidemiol Rev. 2004;26:92–103.
58. Kuh D, Hardy R, Rodgers B, Wadsworth ME.
Lifetime risk factors for women’s psychological distress in
midlife. Soc Sci Med. 2002;55:1957–1973.
59. Bar-Tal Y, Cohen-Mansfield J, Golander H. Which
stress matters? The examination of temporal aspects of
stress. J Psychol. 1998;132:569–576.
60. Cohen S, Kessler RC, Gordon LU, eds. Measuring
Stress: A Guide for Public Health and Social Scientists.
Oxford, UK: Oxford University Press; 1997.
61. Macleod J, Davey Smith G, Metcalfe C, Hart C. Is
subjective social status a more important determinant
of health than objective social status? Evidence from
a prospective observational study of Scottish men. Soc Sci
Med. 2005;61:1916–1929.
62. Adler N. When one’s main effect is another’s error:
material vs. psychosocial explanations of health disparities. A commentary on Macleod. Soc Sci Med. 2006;
63:846–850.
63. Macleod J, Davey Smith G, Metcalfe C, Hart C.
Subjective and objective status and health: a response to
Adler’s ‘‘When one’s main effect is another’s error.’’ Soc
Sci Med. 2006;63:851–857.
45. National Center for Health Statistics. National
Health Interview Survey. Available at: http://www.
cdc.gov/nchs/nhis.htm. Accessed March 11, 2009.
46. Kessler RC, Andrews G, Colpe LJ, et al. Short
screening scales to monitor population prevalences and
trends in non-specific psychological distress. Psychol Med.
2002;32:959–976.
47. Kessler RC, Barker PR, Colpe LJ, et al. Screening for
serious mental illness in the general population. Arch Gen
Psychiatry. 2003;60:184–189.
48. SAS Institute. SAS Language Reference, Version 8.
Cary, NC: SAS Institute; 2001.
49. Health, United States, 2007 With Chartbook on
Trends in the Health of Americans. Hyattsville, MD:
National Center for Health Statistics; 2007.
50. US Census Bureau. American Community Survey
(2006 data). Available at: http://www.census.gov/acs.
Accessed: July 9, 2008.
51. Mossberger K, Tolbert CJ, McNeal RS. Digital
Citizenship: The Internet, Society, and Participation. Cambridge, MA: MIT Press; 2008.
52. Krieger N. Why epidemiologists cannot afford to
ignore poverty. Epidemiology. 2007;18:658–663.
53. Vargas CM, Burt VL, Gillum RF, Pamuk ER. Validity
of self-reported hypertension in the National Health and
Nutrition Examination Survey III, 1988–1991. Prev
Med. 1997;26:678–685.
e8 | Research and Practice | Peer Reviewed | Krieger et al.
American Journal of Public Health | Published online ahead of print November 17, 2009