Professional Documents
Culture Documents
To cite this article: Peter J. Hills & Michael B. Lewis (2006) Reducing the own-race bias in face
recognition by shifting attention, The Quarterly Journal of Experimental Psychology, 59:6, 996-1002, DOI:
10.1080/17470210600654750
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)
contained in the publications on our platform. However, Taylor & Francis, our agents, and our
licensors make no representations or warranties whatsoever as to the accuracy, completeness, or
suitability for any purpose of the Content. Any opinions and views expressed in this publication
are the opinions and views of the authors, and are not the views of or endorsed by Taylor &
Francis. The accuracy of the Content should not be relied upon and should be independently
verified with primary sources of information. Taylor and Francis shall not be liable for any
losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities
whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial
or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or
distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use
can be found at http://www.tandfonline.com/page/terms-and-conditions
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY
2006, 59 (6), 996 – 1002
Short article
The own-race bias (ORB) in face recognition can be interpreted as a failure to generalize expert
perceptual encoding developed for own-race faces to other-race faces. Further, black participants
appear to use different features to describe faces from those used by white participants (Shepherd
& Deregowski, 1981). An experiment is reported where the size of the ORB was assessed using a stan-
dard face recognition procedure. Four groups were tested at two time intervals. One group received a
training regime involving learning to distinguish faces that varied only on their chin, cheeks, nose, and
mouth. Three control groups did not receive this training. The ORB, present prior to training, was
reduced after the critical perceptual training. It is concluded that the ORB is a consequence of a failure
of attention being directed to those features of other race faces that are useful for identification.
Own-race faces are better remembered than faces Meissner and Brigham (2001) people are 2.23
of another, less familiar race (e.g., Meissner & times more likely to recognize an own-race face
Brigham, 2001). This is the crux of the own-race than an other-race face in a recognition experi-
bias (ORB: Sporer, 2001) and has been implicated ment. There are many theories that have been
in many false identifications in eyewitness testi- developed to explain the ORB. These are summar-
mony (e.g., Leippe, 1995). It is a reliable effect ized in detail by Sporer (2001). From these the-
(Chance & Goldstein, 1996), which is arguably ories, procedures have been developed to reduce
consistent across racial groups (Bothwell, the ORB in participants.
Brigham, & Malpass, 1989; but see Anthony, One of the most frequently discussed methods
Copper, & Mullen, 1992, for a different result). for reducing the ORB is with experience (Galper,
According to a meta-analysis conducted by 1973; Slone, Brigham, & Meissner, 2000).
Correspondence should be addressed to Peter Hills, School of Psychology, Cardiff University, Park Place, Cardiff CF10 3AT,
UK. Email: HillsPJ@Cardiff.ac.uk
This research was supported by grant PTA– 030–2003– 00524 from the ESRC to Peter Hills. The authors would like to thank
Paul Hutchings for providing some of the photographs used for this experiment. Furthermore, the authors would like to thank
Rachael Elward, Mary Halsey, and Claire Jones for some of the data collection. Finally, the authors would like to thank Dan
Wright and an anonymous reviewer for detailed and thought-provoking comments on a previous draft of this paper.
Attempts have been made to reduce the ORB using together farther away from the mean. This makes
training regimes to increase contact with other-race them more difficult to distinguish from each
faces. Malpass, Lavigueur, and Weldon (1973) used other, but easy to differentiate from own-race faces.
a simple 1-hour visual training paradigm, whereby The face-space explanation suggests that other-
white participants were shown black and white race faces are not best described by the dimensions
faces and were given feedback for a four-alternative within face-space. Though these dimensions have
forced-choice recognition test. Training did not been specified precisely, it can be suggested
improve recognition accuracy scores, especially if that the reason that other-race faces are not appro-
participants were given shock feedback. priately stored on the face-space is that there are
Following on from these studies, Lavrakas, physiognomic differences between black and
Buri, and Mayzner (1976) conducted another white faces.
training study, in which white participants were Ellis, Deregowski, and Shepherd (1975) looked
given a concept-learning visual training task. The at the frequencies with which black and white
training faces were made from an Identi-Kit, and participants referred to various facial features.
the two concepts that participants had to learn Their results clearly showed a different pattern of
Downloaded by [Tufts University] at 11:22 31 October 2014
were either light eyes or a conjunction of dark features examined by black participants and
eyes and thick lips. Participants were given feed- white participants. Face outline, eye size, eye-
back for their judgement. Lavrakas et al. also brows, chin, and ears were significantly more fre-
tested whether the type of contact was relevant quently used by black participants as descriptors,
in the ORB and field dependence/independence while white participants used hair colour, hair
of the participants. Their results indicated that texture, and eye colour. Their data also show that
those white participants who had black friends black participants used lips, mouth, and nose
had a lower ORB than those who simply more frequently, especially when describing black
knew more black participants. Moreover, field- faces. Furthermore, black participants use a more
independent white participants were better able varied description, using more features than
to recognize black faces. Finally, the training on white participants. This may represent an import-
specific features did reduce the ORB, possibly ant differentiation between the face recognition
due to removing the deleterious effect of abilities of black and white participants or physiog-
remembering only skin-colour (Lavrakas et al.). nomic differences between black and white faces.
These results are consistent with a recent theo- How facial descriptions differ across black and
retical account of the ORB. Levin (2000) advocates white faces was looked at in detail by Shepherd
a feature selection process in the recognition of and Deregowski (1981). Using INDSCAL ana-
faces. By this, participants extract the most diag- lyses, physiognomic differences across black and
nostic visual feature. As such, in processing an white faces were discovered. Internal facial features,
own-race face the most diagnostic visual feature with round fat faces, thick lips, and broad noses are
could contain any individuating information. related to black physiognomy. This is contrasted
However, processing an other-race face will with features associated with hair, which are
involve a race feature as the most diagnostic. As related to white physiognomy. Hair features
such, an other-race face is processed according to explain roughly 85% of the variance in the
a prototype and is therefore not individuated, INDSCAL solution for white faces, but only 35%
thus making it more difficult to recognize. for black faces. The internal lower facial features
An alternative explanation based upon Valentine explain roughly 75% of the variance in the
(1991, expanded by Lewis, 2004), is that every face INDSCAL solution for black faces, but the same
is stored within a multidimensional face-space. The features only explain 35% of the variance for
face-space stores faces along relevant dimensions. white faces. Nevertheless, using a homogenous set
Other-race faces do not vary consistently along of faces, both black and white participants will use
these dimensions and as such are grouped close the same features to discriminate between them.
Taking these results with those of Lavrakas et al. Participants viewed the monitor at a distance of
(1976) indicates that training white participants to 60 cm.
look at specific features will indeed improve the rec- Two face sets were used to assess the ORB
ognition of black faces if those features are the most before and after training. These were the
diagnostic. The current experiment used a percep- MacBrain Face Stimulus Set1 and a face set
tual learning experimental procedure in which collected by Paul Hutchings at Cardiff University.
white participants were trained using white faces, Each face had two pictures taken of it, which
which can be discriminated upon only by lower were both neutral in expression and frontal views.
facial features (chin, nose, mouth, and cheeks). It One of these pictures was used in the presentation
was hypothesized that this training should sub- phase, while the other face was used in the recog-
sequently reduce the ORB. To control for greater nition phase of the recognition paradigm. This
exposure to faces, three control groups were was to avoid pictorial recognition. All stimuli had
included. One group received no training whatso- a resolution of 72 dpi and were 150 mm by 200 mm.
ever. A second group received noncritical-feature Three different training sets of faces were con-
training. In this condition, white participants were structed using a face composite computer package
Downloaded by [Tufts University] at 11:22 31 October 2014
given the same level of training, except that the FACESTM (InterquestTM ). This package uses 16
faces used for training differed only on features not features to make a face. Within each feature a
diagnostic to black faces. A final control group was choice of between 10 and 100 features can be
presented with the same faces as those in the made. For each set, 40 different target faces were
crucial feature-critical trained condition, except created. From these targets, 10 lookalikes were
that the task was extraneous to the face. In other created, which were identical in certain features
words, this control tested between simple presen- depending on condition.
tation of faces and actual attention paid to the faces. For the feature-critical faces, all the faces had
similar hair style and colour. The lookalikes only
differed according to “lower” facial features: the
METHOD chin, nose, mouth, and cheeks. They were identi-
cal to the targets according to “white” facial fea-
Participants tures. The coloured blob discrimination
A total of 124 Cardiff University undergraduates condition used these faces; however, each face
were randomly divided into one of four training had a coloured blob placed on it in a random
groups: feature critical, no training, noncritical place. The noncritical-feature faces were created
feature, and colour blob discrimination. All par- in the same way as the feature-critical faces,
ticipants had normal or corrected vision. The except that all the lookalikes were identical in
data from 2 participants were removed since they the lower regions: chin, nose, mouth, and cheeks.
had been raised in Africa, and they were replaced Only the eyes and hair styles differed. An
by additional participants. example of each of the faces is shown in Figure 1.
As an additional control, all participants filled
out the Modern Racism Scale (McConahay, 1986).
Apparatus and stimuli
All stimuli were presented on a high-resolution
Design
colour monitor, displayed using DirectRTTM
Research Software (EmpirisoftTM). Participant Participants’ ORB level was assessed before and
responses were recorded on a standard keyboard. after training. The ORB was tested in a simple
1
Development of the MacBrain Face Stimulus Set was overseen by Nim Tottenham and supported by the John D. and Catherine
T. MacArthur Foundation Research Network on Early Experience and Brain Development. Please contact Nim Tottenham at
tott0006@tc.umn.edu for more information concerning the stimulus set.
Procedure
This experiment involved three phases. The first
phase was the face recognition test, the second
was the training, and the third was the second
face recognition test.
The standard old/new recognition tests con-
sisted of participants being shown 20 faces (10
black and 10 white). Participants were asked to
assess the faces for attractiveness on a 0– 9 scale.
Each face was presented for 2 s. As a distractor
task, participants filled out a questionnaire,
which took 3 minutes. After the questionnaire,
the participants were given the recognition
phase, whereby they were presented 40 faces (20
Downloaded by [Tufts University] at 11:22 31 October 2014
feature does not accurately portray the perceptual precisely, what are the unique elements which
experience. What exactly it is about the elements help us recognize different faces. Clearly, the
of an other-race face that makes race the feature is ORB is transient enough to be reduced by refocus-
a question unanswered by Levin. ing of attention, but it is unlikely that an hour’s
An important aspect of this study is the use of training will be sufficient to reverse a lifetime’s
perceptual learning mechanisms, borrowed from experience. Nevertheless, it is clear that perceptual
the animal learning literature, in face recognition. learning will be able to assist in discovering the
Using such methods, researchers can be more dimensions of the face-space.
explicit in how they define constructs and dimen-
sions in face recognition, which is vital for research Original manuscript received 2 August 2005
to progress. Moreover, there is much that is Accepted revision received 15 February 2006
unknown about how face recognition develops.
Perceptual learning can be applied to develop-
mental aspects of face recognition. REFERENCES
One caveat of the present study must be
Downloaded by [Tufts University] at 11:22 31 October 2014
acknowledged. In this study, only white partici- Anthony, T., Copper, C., & Mullen, B. (1992). Cross-
pants were tested. The effects observed here racial facial identification: A social cognitive integ-
could be replicated across black participants who ration. Personality and Social Psychology Bulletin, 18,
296– 301.
show an ORB and therefore a lower ability to
Bothwell, R. K., Brigham, J. C., & Malpass, R. S.
recognize white faces than black faces. The
(1989). Cross-racial identification. Personality and
feature-critical condition in this case would be Social Psychology Bulletin, 15, 19 – 25
equivalent to the noncritical training used on Chance, J. E., & Goldstein, A. G. (1996). The other-
white participants. The effect described in this race effect and eyewitness identification. In
study may prove more difficult to obtain in black S. L. Sporer, R. S. Malpass, & G. Koehnken
participants, since black participants use a wider (Eds.), Psychological issues in eyewitness identification
range of features to describe faces (Ellis et al., (pp. 153– 176). Mahwah, NJ: Lawrence Erlbaum
1975), which may explain why black participants Associates, Inc.
are generally better at face recognition tasks than Ellis, H. D., Deregowski, J. B., & Shepherd, J. W.
are white participants (see Shepherd & (1975). Descriptions of White and Black faces by
White and Black subjects. International Journal of
Deregowski, 1981). As such, the perceptual learn-
Psychology, 10, 119– 123.
ing effect described here explains why black par-
Galper, R. E. (1973). “Functional race membership” and
ticipants show a lower ORB, since they use all recognition of faces. Perceptual and Motor Skills, 37,
dimensions to recognize faces, while white partici- 455– 462.
pants only use relevant dimensions. Hall, G. (1980). Exposure learning in animals.
One final point to make at this stage is the Psychological Bulletin, 88, 535– 550.
possibility that this simple training may have an Hills, P. J., Lewis, M. B., & Honey, R. C. (2005,
effect on stereotyping and prejudice of the partici- August). Stereotype priming in the recognition of unfa-
pants. Since the participants are showing less of an miliar faces. Paper presented at the BPS Cognitive
ORB, it may be that they are processing other-race Section Conference, Leeds, UK.
faces more appropriately, which may covertly Lavrakas, P. J., Buri, J. R., & Mayzner, M. S. (1976). A
perspective on the recognition of other-race faces.
reduce the stereotyping subsequently used. Social
Perception and Psychophysics, 20, 475– 481.
psychologists may wish to examine this hypothesis
Leippe, M. R. (1995). The case for expert testimony
as another method for reducing stereotyping. about eyewitness memory. Psychology, Public Policy,
In conclusion, the perceptual learning effect can and Law, 1, 909–959.
be used to explain many of the effects in face rec- Levin, D. T. (2000). Race as a visual feature: Using
ognition. It should become a useful tool in learn- visual search and perceptual discrimination tasks to
ing how humans learn about faces and, more understand face categories and the cross-race