Introduction
Recent research has shown a significant decrease of
expressed anti-Black attitudes (Martin, Carlson & Buskist, 2013). However,
this may not be because prejudice and stereotypes have decreased but because
they have changed their form and manifestation (Martin et al., 2013). In 2005,
a high school student named Kiri Davis replicated Clark and Clark’s (1939)
study, which had shown that most Black children preferred White-skinned dolls
over Black ones, and found similar results (Davis, 2005). This evidenced that
stereotypes and prejudice had not been overcome and could be
harmful to
modern society. However both Davis (2005)
and Clark and Clark (1939) measured explicit attitudes, which are attitudes of
which we are aware and that we express freely. The problem with explicit
attitude measurements is that participants may not want to express their true
attitudes or may be unaware of some of their attitudes. Thus, methods for
assessing implicit attitudes - attitudes that we are not aware of, are
developed to address these issues.
Furthermore, there has been growing interest in developing
implicit tests due to the limitations of existing methods for implicit attitude
exploration (O'Reilly, Roche, Ruiz, Tyndall & Gavin, 2012). For example,
the procedure of Watt, Keenan, Barnes, and Cairns (1991), which involved the
learning of a series of stimulus relations, was thought to be capable of
identifying implicit relations between stimuli in relation to social history.
This procedure was problematic because it required considerable amount of time
as well as focused attention and motivation (O'Reilly et al., 2012). Another
example is the widely used Implicit Association Test (IAT), which was developed
by
Greenwald,
McGhee and Schwartz (1998)
and
was
said to measure implicit attitudes. The IAT’s procedure paired pictures on the
screen with either stereotype consistent or stereotype inconsistent responses
and then compared the reaction times of participants (Nosek et al., 2007).
However, the test has been widely criticised. Amongst these criticisms were the
suggestions that the IAT was influenced by information from the stimuli (Govan,
& Williams, 2004) and that the statistical inferences from the IAT were
used inaccurately (Blanton & Jaccard, 2006).
A
recently
developed alternative
to the IAT is the Function Acquisition Speed Test (FAST), designed by O'Reilly
et al. (2012). It was based on the findings of Watt et al. (1991) that past
cultural learning could impair new learning. In contrast to the IAT, the FAST
uses combined accuracy and speed as a measure of implicit attitudes and gives
limited time for
responding.
This has
resolved the IAT’s problem which with calculation of response times (O'Reilly
et al., 2012). Moreover, feedback in the FAST is given both after correct and
incorrect responses whereas in the IAT it is given only after incorrect
responses that can function as punishment (O'Reilly et al., 2012). Another
difference is that the FAST measures learning rate whereas the IAT
measures
standardised
reaction time.
Irrespective of these methodological issues, studies on
stereotypes and prejudice have shown implicit and explicit preferences for
White/light-skin individuals. For example, Nosek, Banaji and Greenwald (2002)
gathered data from over 600,000 IATs via an Internet website with the aim to
measure attitudes towards different social groups. Their investigation showed a
general implicit preference for White over Black individuals. In relation to
this, the study of Nosek et al. (2007) supported these findings. This study
used a website to gather over 2.5 million datasets from IATs in order to
examine preferences and stereotypes on 17 topics. The results revealed that the
majority of the participants demonstrated both implicit and explicit
preferences for White/light-skin over Black/
dark-skin
individuals. However, it is unclear whether these findings are due to pro-White
or anti-Black bias.
The goal of this experiment is to investigate anti-Black
bias relative to both White faces and neutral stimuli in order to understand
whether
participants' learning is due to
pro-White bias or anti-Black bias. The study will use the recently developed
FAST
to measure participants’ implicit attitudes in two blocks –
Consistent and Inconsistent. In the Consistent block the “correct” answers will
refer to answers which are consistent with the expected stereotypes whereas in
the Inconsistent block the “correct” answers will refer to those inconsistent
with the expected stereotypes. Furthermore, the stimuli will be presented in
two conditions – Non-Relative and Relative. The Non-Relative condition will
measure the association between Black faces and negative words in relation to
neutral stimuli whereas the Relative condition will measure the associations
between Black faces - negative words and White faces - positive words in
relation to one another
.
Based on the
literature, our first hypothesis is that participants will learn faster in the
Consistent block than in the Inconsistent block. We base this hypothesis also
on the fact that the answers in the Consistent block are referred to as
“correct” if they are consistent with the expected stereotypes and the answers
in the Inconsistent block are referred to as “correct” if they are inconsistent
with the expected stereotypes. Our second hypothesis is that the difference in
the rates
of learning between blocks will be
higher in the Relative group than in the Non-Relative group. We base this claim
on the assumption that the Non-Relative condition measures the association
between Black faces and negative words in relation to neutral stimuli whereas
the Relative condition measures two associations (anti-Black and pro-White) in
relation to one another.
Methods
Participants
Participants were 153 Psychology undergraduate students.
There were 87 participants in the Non-Relative group and 66 participants in the
Relative group. They all participated in the study as part of a module
requirement. Participants were randomly assigned to one of the two groups.
Apparatus
The stimuli were presented on an iMac computer, with screen
size 21.5” and resolution 1920 x 1080 pixels. The FAST was designed and
delivered using
Livecode
Software. The
data were collated via Google Forms.
Materials
In this FAST experiment,
different stimuli were presented in two conditions – the Non-Relative and the
Relative condition. The Non-Relative condition included images of Black faces,
negative words, neutral objects and neutral words. It measured the association
between Black faces and negative words in relation to neutral stimuli and aimed
to investigate anti-Black stereotypes in isolation. The Relative condition
included images of Black faces, negative words, images of White faces and
positive words. It measured two associations in relation to one another and
aimed to investigate anti-Black stereotypes in relation to positive White
stereotypes.
The images of Black and White
faces were taken from Cunningham, Preacher and Banaji (2001). All of the
stimuli presented in the FAST can be viewed in Figure 1.
Figure
1
Stimuli presented in the experimental trials.
Design
The experiment had a mixed-factorial design. The independent
within-subjects variable was: block (Consistent or Inconsistent). The
between-subject variable was group (Non-Relative or Relative). The dependent
variable was the FAST slope – the rate of learning.
Procedure
Participants signed a consent form and started the
experiment. No practice trials were provided. The experimental trials began and
the FAST instructions were displayed with black letters in the centre of a
white screen. They stated that the participants’ task was to learn which button
to press when an image appeared on the screen. Participants had to press either
the Z or the M key and were asked to locate them on the keyboard. It was
pointed out that this part of the experiment would continue until the
participants had learned the task and could respond without error. They would learn
via feedback, which would inform them if they were correct or not. Participants
were given the opportunity to ask the researcher any questions they may have
had. They had to press any key to continue.
Stimuli started appearing one by one in the centre of the
white screen and participants had to sort them by pressing either Z or M.
Participants in each group completed two blocks – Consistent and Inconsistent.
The order of the blocks was randomised. Participants received feedback –
“Correct” or “Wrong” in red colour, after every sorting. In the Consistent
block, the answers were referred to as “correct” when participants sorted
together Black faces and negative words and either White faces and positive
words or neutral objects and neutral words depending on the condition. In the
Inconsistent block, the answers were referred to as “correct” when participants
sorted together Black faces with either positive or neutral
words
as well as negative words and either
neutral objects or White faces depending on the condition. If participants took
more than 3000 milliseconds to respond to a trial their response was recorded
as wrong. They could not correct their responses at any time.
Results
Scores for rates of learning (Slope) in the Consistent and
the Inconsistent block were calculated separately for each participant by
dividing the change in the number of correct responses
by
the change in the elapsed time. Faster learning of correct
responses was indicated by a higher slope. A score representing the difference between
the Consistent and the Inconsistent Slope was also calculated for each
participant. Faster learning in the Consistent block was indicated by a
positive difference whereas faster learning in the Inconsistent block was
indicated by a negative difference. A positive Slope Difference was referred to
as a FAST effect.
Mean scores were obtained for the Consistent Slope, the
Inconsistent Slope and the Slope Difference for the Non-Relative and the
Relative condition. The means and standard deviations in the Non-Relative and
Relative conditions are presented in Table 1. The effect size for the mean
difference between the Consistent and the Inconsistent Slope in the
Non-Relative condition was medium (d=0.48). The effect size for the mean
difference between the Consistent and the Inconsistent Slope in the Relative
condition was small (d=0.27).
Table
1
The means and standard deviations (in parentheses) in the
Non-Relative and in the Relative condition.
Consistent Slope
|
Inconsistent Slope
|
Slope Difference
|
|
Non-Relative
|
0.70
(
0.14
)
|
0.63
(
0.15
)
|
0.06
(
0.15
)
|
Relative
|
0.68
(
0.17
)
|
0.64
(
0.13
)
|
0.04
(
0.15
)
|
For both conditions, the paired samples t-tests showed that
there was
a
statistically significant
mean difference between the Consistent and the Inconsistent Slope,
t
(86)=3.90, p<.001, 2-tailed (for the
Non-Relative condition) and
t
(65)=2.26,
p=.027, 2-tailed (for the Relative condition). Participants learned faster in
the Consistent than in the Inconsistent block in both conditions. An
independent samples t-test showed that there was no statistically significant
mean Slope Difference between the Non-Relative and the Relative condition,
t
(151)=0.84, p=.401, 2-tailed. There was
no difference in the learning rate between the Relative and Non-Relative
groups.
Discussion
The study aimed to examine how much of the participants’
performance in the experiment was due to pro-White bias. The results showed
that participants learned faster when
the
“correct” responses were consistent with the expected stereotypes compared to
when they were inconsistent. This supported our first hypothesis that
participants would learn faster in the Consistent than in the Inconsistent
block. However, the results showed that the difference in learning rate between
blocks did not differ significantly across groups, which did not support our
second hypothesis that the learning rate difference would be higher in the
Relative group than in the Non-Relative. These findings indicated that
anti-Black attitudes were demonstrated, with no significant difference in
strength, relative to both White faces and neutral stimuli which suggested that
learning performance was not influenced by pro-White bias.
The results of this study supported the findings of Nosek,
Banaji et al. (2002) and Nosek et al. (2007) that there is a general preference
for White over Black individuals. However, the findings indicated that these
attitudes were not due to pro-White bias. The results further suggested that
positive-White associations do not contribute to
negative-Black
associations as the
negative-Black
associations also occurred relative to neutral images with no significant
change in strength. The present study also supported the results of Watt et al.
(1991) that past verbal or cultural learning could interfere with new learning.
The present study showed that participants took longer to learn a pairing of a
stimulus and response when the pairing was inconsistent with the widely-held
anti-Black stereotypes. This showed that the previously learned information
that dark-skin people are associated with negative stimuli interfered with the
new learning of associating dark-skin people with positive responses.
However, one limitation of this study which could account
for the lack of support for the second hypothesis is the difficulty of finding
strictly neutral stimuli. Thus, there is a possibility that the neutral stimuli
used in the Non-Relative condition may not have been “neutral” for the
participants. In some cultures, numbers have symbolic meanings. For example, 6
is thought to be the number of the devil and 7 is thought to be a sacred number
in some societies. Furthermore, the presented “neutral” objects may have been
associated with a particular event or sensation experienced by the
participants. A new study should be conducted in order to address this problem
by presenting the participants with the same experimental design and conditions
but with one difference – asking them to rate separately each stimulus
presented in the study on a scale ranging from “Very pleasant” to “Very
unpleasant”. This would reveal whether the neutral stimuli were truly
“neutral”.
Raising awareness about implicit anti-Black biases is the
first step towards reducing them. For example, our findings can be directly
applied in the educational setting as anecdotal evidence suggest that
anti-Black attitudes are pervasive in schools but are not always explicit.
School staff can use the provided knowledge to raise awareness amongst both
teachers and students about implicit anti-Black biases. This may reduce the
implicit anti-Black biases in schools due to the awareness of their existence.
The findings can also be applied to the police and legal systems where implicit
anti-Black attitudes can have terrible consequences. This may lead to creating
special programs for police officers and judges which will aim to reduce
implicit anti-Black bias. Another application would be in the field of the
media. Although there are a lot of explicit negative attitudes towards
dark-skin people in the media, some of the negative attitudes may be implicit.
Raising awareness about this problem amongst both the people who work in the
field and amongst the viewers will aid reducing implicit anti-Black
stereotypes.
In conclusion, the study showed that participants took
longer to learn pairings that were inconsistent with previously learnt racial
stereotypes. This suggests that pre-learnt stereotypes could interfere with
present learning. In addition, the findings showed that anti-Black bias
occurred relative to neutral images as well as relative to images of White-skin
individuals. This suggests that anti-Black bias was not enhanced by pro-White
bias. These findings are important because they provide knowledge that might
aid raising awareness about implicit anti-Black biases. This could help reduce
the anti-Black biases in society.