Perception advance online publication
doi:10.1068/p6136
Two faces of the other-race effect: Recognition
and categorisation of Caucasian and Chinese faces
Liezhong Ge, Hongchuan Zhang½, Zhe Wang, Paul C Quinn#, Olivier PascalisÀ,
David KellyÁ, Alan Slater^, Jie Tian &, Kang Lee^ô
Zhejiang Sci-Tech University, P. R. China; ½ University of California, San Diego, USA; # University
of Delaware, USA; À University of Sheffield, UK; Á University of Glasgow, Scotland, UK; ^ University of
Exeter, UK; & Chinese Academy of Sciences, Beijing, P. R. China; ^ Institute of Child Study, University
of Toronto, 45 Walmer Road, Toronto, Ontario M5R 2X2, Canada; e-mail: kang.lee@utoronto.ca
Received 24 July 2008, in revised form 5 February 2009; published online 29 June 2009
Abstract. The other-race effect is a collection of phenomena whereby faces of one's own race
are processed differently from those of other races. Previous studies have revealed a paradoxical
mirror pattern of an own-race advantage in face recognition and an other-race advantage in
race-based categorisation. With a well-controlled design, we compared recognition and categorisation of own-race and other-race faces in both Caucasian and Chinese participants. Compared
with own-race faces, other-race faces were less accurately and more slowly recognised, whereas
they were more rapidly categorised by race. The mirror pattern was confirmed by a unique negative correlation between the two effects in terms of reaction time with a hierarchical regression
analysis. This finding suggests an antagonistic interaction between the processing of face identity
and that of face category, and a common underlying processing mechanism.
1 Introduction
The other-race effect is a collection of phenomena whereby faces of one's own race
are processed differently from those of other races. One such occurrence is the ownrace recognition advantage whereby own-race faces are recognised more accurately
and faster than other-race faces (Bothwell et al 1989; Brigham and Malpass 1985;
Chiroro and Valentine 1995; Valentine 1991). The effect of race on face recognition is
robust in that it occurs across different racial groups (Bothwell et al 1989; Rhodes
et al 1989; Shepherd and Deregowski 1981), age groups (Chance et al 1982; Pezdek et al
2003; Sangrigoli and de Schonen 2004a, 2004b), and in both laboratory and field settings
(Brigham et al 1982; Cross et al 1971). It has also been confirmed by several meta-analytic
studies (Bothwell et al 1989; Meissner and Brigham 2001).
Overshadowed by the vast literature on the own-race recognition advantage is a
paradoxical other-race categorisation advantage. When participants are asked to categorise faces by their race, they respond faster to other-race faces than to own-race faces
(Caldara et al 2004; Levin 1996, 2000; Valentine and Endo 1992). This other-race
categorisation advantage has also been demonstrated to be robust with various face
stimuli across different racial groups, by using either a race-based categorisation task
or just a simple visual-search task. It should be noted, however, that this effect can
be modulated by task, stimulus, and participant characteristics, such as simultaneous
race ^ sex categorisation (eg Stroessner 1996), face distinctiveness (eg Valentine and
Endo 1992), and participant racial stereotype tendencies (eg Zarate and Smith 1990).
Despite this apparent mirror pattern in response latency for own-race and otherrace face processing in different tasks, in most studies the two effects have been
investigated separately, with little concurrent examination of the paradoxical phenomena. To recognise a face among others, one must rely on identity-specific facial
information; whereas to tell whether a face is Caucasian or Chinese, one must rely on
ô Author to whom all correspondence should be addressed.
2
L Ge, H Zhang, Z Wang, and coauthors
information specific to race. According to the classic face-recognition model (Bruce
and Young 1986; Burton et al 1990), race-specific information, as well as other face
attributes, like gender, age, or facial expression and gaze direction, are visually derived
semantic codes. These modes are believed to operate independently of the identityspecific information. Earlier behavioural, neuroimaging, and neuropyschology studies
seem to support this hypothesis (Bruce 1986; Etcoff 1984; Haxby et al 2000; Humphreys
et al 1993; Young et al 1985, 1993).
In contrast, recent evidence has come to suggest that identity-specific and categoryspecific codes may interact with each other and be processed with the same cognitive
mechanisms (Bruyer et al 2004; Ganel and Goshen-Gottstein 2002; Le Gal and Bruce
2002). Following this point of view, the mirror pattern of the two face-race effects
(recognition and categorisation) reflects the trade-off and competition between processing individual identity and categorical facial information of faces from own and other
races. It is crucial to note that neuroimaging studies have found a significant difference
between own-race and other-race faces in activation in the FFA (fusiform face area)ö
an area sensitive to processing of facial identity (Golby et al 2001). However, the results
were only evident in a recognition task that relied heavily on identity-specific codes. Thus,
it is currently unknown whether similar results would be found with a categorisation task
that requires the processing of race-specific codes.
To date, this possibility has been examined by only two behavioural studies with
mixed results. In one study, Levin (1996) divided Caucasian participants into two groups:
a deficit group showing the own-race recognition advantage in accuracy and a non-deficit
group not showing the effect. Both groups categorised computer-distorted faces as
Caucasian or Black. They showed no difference in the other-race categorisation advantage
in terms of reaction time, although there was a general other-race categorisation advantage in reaction time when the data for the two groups were combined. This finding
suggested that whether or not individuals had an own-race recognition advantage did
not predict whether they also had an other-race categorisation advantage. In contrast,
Levin (2000) again divided participants into two groups and asked them to search for a
Caucasian or Black face among other-race face distractors. This time, the deficit group
showed a greater other-race categorisation advantage relative to the non-deficit group as
measured by search time. Nevertheless, no significant differences were found in the same
study between the two groups in terms of the other-race categorisation advantage when
a go/no-go task was used (participants were asked to respond to a target face category
but not to a non-target face category).
However, these equivocal outcomes might be due to the different methodologies
used in Levin's studies. For example, the stimuli used for testing the own-race recognition advantage were individual faces, but those used to assess the other-race categorisation
advantage were morphed or average faces. Also, whereas the recognition paradigm
remained identical for various experiments, the method for testing categorisation
varied. In one case, a race-based categorisation task was used (Levin 1996), and in
another instance, a go/no-go or visual search paradigm was employed (Levin 2000).
Furthermore, accuracy was used to measure the own-race recognition advantage
and reaction time was used to measure the other-race categorisation advantage. This
mismatch in dependent measures might have also contributed, to a certain extent, to
the inconsistent findings. Perhaps, more importantly, participants were students from
a major US university that is ethnically diverse, and these participants might have had
extensive exposure to various other-race faces (indeed, although the other-race recognition
advantage is highly robust, a significant proportion of participants in both studies did not
show it at all). Thus, any close relationships between the own-race recognition advantage
and the other-race categorisation advantage might have been obscured by these factors.
Other-race effects of recognition and categorisation
3
In the present study, to directly examine the interrelation between the paradoxical
own-race and other-race face effects, we recruited participants who had near-zero direct
contact with other-race individuals in the UK and China where over 91% and 99% of
the population are either Caucasian or Chinese, respectively. The participants completed a recognition task in which they were required to recognise previously seen
Chinese and Caucasian faces and a race-based categorisation task in which they were
asked to judge the race of Chinese and Caucasian faces. The two tasks were structured
such that the face stimuli were randomly assigned to each task, with the same number
of stimuli, the same timing parameters, and the same manual response. It should be
noted that the other-race categorisation advantage has been found only with the latency
measures and categorisation accuracy has been typically near ceiling. In contrast, the
own-race recognition advantage has been obtained in terms of both latency and accuracy.
To further ensure comparability between the two tasks, we specifically focused on participants' reaction times when recognising and categorising own-race and other-race faces.
To ascertain whether dependent measure mismatches contributed to the inconsistent
findings in the previous studies, we also examined the relationship between participants'
recognition accuracy and their categorisation reaction times.
On the basis of the existing evidence, we expected to observe both the own-race
recognition advantage and the other-race categorisation advantage among Chinese and
Caucasian participants, with the former in terms of both accuracy and response latency,
and the latter in terms of response latency. More importantly, if the face-identity and
race-information processing is carried out independently (Bruce and Young 1986), the
two effects should not correlate with each other for both Chinese and Caucasian participants. Alternatively, if a common underlying mechanism is at work, a significant negative
correlation should be observed between the two effects. By a common-mechanism
account, identifying own-race faces would compete with categorising them; when processing capacity remains constant, increased proficiency at recognising own-race faces must
be compensated with a decrease in categorising the same faces.
2 Method
2.1 Participants
Thirty-two Han Chinese students (sixteen females) from Zhejiang Sci-Tech University,
and thirty-five Caucasian students (twenty females) from Sheffield University, participated in the present study. Participants reported no direct contact with other-race
individuals. In Sheffield, the population consisted of 91.2% Caucasians and 1% Chinese.
In Hangzhou, 99.9% of the population is Han Chinese.
2.2 Stimuli
64 Caucasian and 64 Chinese young-adult upright faces with neutral expression (half
male and half female) were used. All faces were full-colour, high-quality photographic
images taken frontally at a fixed position, digitised in 24-bit colours with a resolution
of 6406480 pixels.
2.3 Design and procedure
The experiment was a 262 factorial design, with face race (Caucasian versus Chinese)
and task type (recognition by identityöhereafter the recognition tasköversus categorisation by raceöhereafter the categorisation task) as within-subjects factors. The
64 faces from either race were divided into two lists with the same number of female
and male faces. For each participant, the faces from one of the two lists were used for the
recognition task and those from the other list were used for the categorisation task.
In other words, the same participant did not see the same faces in the recognition and
categorisation tasks. The assignment of the two lists to the two tasks and the task order
was counterbalanced between participants.
4
L Ge, H Zhang, Z Wang, and coauthors
For the recognition task, participants first passively viewed and were asked to
remember 16 Caucasian and 16 Chinese faces presented in random order, repeated
for three times to enhance memorisation (a centrally located fixation cross-hair was
presented for 500 ms between the same stimuli). After this phase, those learned faces
were randomly mixed with another 32 unlearned faces (16 from each of the two race
categories) for recognition. Participants were asked to press either `1' or `2' on the number
pad to indicate whether the face had been previously seen or not.
For the categorisation task, participants were asked to press the same keys as in the
recognition task to indicate whether the face was a Caucasian or Chinese face, which
was shown only once. The key assignment was counterbalanced between participants
in both tasks.
Participants sat in a dimly lit quiet room and saw the faces from a visual angle of
12.4 deg in height and 16.4 deg in width. Faces were presented with E-Prime (Psychology
Software Testing, Pittsburgh, PA) via a PC computer. In the recognition task, the faces
were presented for 2 s per face in the learning phase. In the recognition phase and the
categorisation task, the faces were presented for up to 5 s depending on the key press.
Participants were asked to respond as fast and as accurately as possible. Before each
study or test face was presented, participants were asked to look at a centrally located
fixation cross-hair with a random variable interstimulus interval between 500 and
1000 ms.
3 Results
Preliminary analyses showed that the effects of participant gender and face gender were
not significant. Thus, the two factors were excluded from further analyses. We first
performed an omnibus ANOVA to examine the two face-race effects (for recognition
and categorisation) in the two participant groups (Caucasian and Chinese), on both
the reaction time and accuracy. Then a hierarchical regression analysis was conducted
to evaluate the relationship between the two face-race effects.
3.1 Reaction time
Figure 1 shows the accuracy and reaction time results for Chinese and Caucasian faces
from both the recognition and categorisation tasks. The reaction time for each participant for each face type in each condition was obtained by averaging the latencies of
the trials in which the participant gave correct responses. Trials with reaction times
above two standard deviations were excluded. The ANOVA was performed with task
type (categorisation versus recognition) and face race (Caucasian versus Chinese) as
within-subjects factors, and participant race as between-subjects factor on the reaction
time data.
A significant main effect of task type was found (F1, 65 14:86, p 5 0:001,
Z 2 0:19). Participants were faster in the categorisation task (mean, M 906:85, standard deviation, SD 205:69) than in the recognition task (M 1028:71,
SD 284:44). There was also a significant effect of face race (F1, 65 8:41, p 5 0:01,
Z 2 0:11), and of participant race (F1, 65 4:97, p 5 0:05, Z 2 0:07). The Caucasian
faces were responded to faster (M 952:43, SD 217:73) than were the Chinese faces
(M 983:13, SD 214:67); and Caucasian participants were faster (M 910:07;
SD 211:61) than Chinese participants (M 1025:49, SD 211:61).
There was a significant interaction between task type and participant race
(F1, 65 29:61, p 5 0:001, Z 2 0:31). Caucasian participants were faster in the categorisation task (M 763:14, SD 284:15) than in the recognition task (M 1057:00,
SD 205:49), whereas Chinese participants performed similarly fast in both tasks
(categorisation: M 1050:56, SD 284:15; recognition: M 1000:42, SD 205:49).
The interaction between face race and participant race was not significant.
Other-race effects of recognition and categorisation
5
Correct reaction time=ms
1200
1000
800
600
400
200
0
Accuracy
(a)
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
Caucasian faces Chinese faces
Stimuli
d0
(b)
(c)
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
Participant
Caucasian
Chinese
Caucasian faces
Chinese faces
Stimuli
Figure 1. (a) Reaction time, (b) accuracy, and (c) d 0 for Caucasian and Chinese faces in Caucasian
(solid line) and Chinese (dotted line) participants in the recognition (left panels) and categorisation
(right panels) tasks. Standard error bars are shown.
The crucial three-way interaction between task type, face race, and participant race
was significant (see figure 1a for means and standard errors) (F1, 65 28:59, p 5 0:001,
Z 2 0:31). Paired t-tests showed that Caucasian participants recognised Caucasian faces
faster than Chinese faces (t34 ÿ4:99, p 5 0:001), and categorised Chinese faces faster
than Caucasian faces (t34 2:68, p 5 0:05), whereas Chinese participants recognised
Chinese faces faster than Caucasian faces (t31 2:25, p 5 0:05), and categorised Caucasian
faces faster than Chinese faces (t31 3:32, p 5 0:01). This mirror pattern between the
own-race recognition advantage and the other-race categorisation advantage (figure 1a)
indicates that, as expected, own-race faces and other-race faces were processed differently
in the two tasks, for both Caucasian and Chinese participants.
3.2 Accuracy
An ANOVA with task type (categorisation versus recognition) and face race (Caucasian
versus Chinese) as within-subjects factors, and participant race as a between-subjects
factor, was performed on the accuracy data as measured by percentage of correct
responses (figure 1b). The main effect of task type was significant (F1, 65 122:58,
6
L Ge, H Zhang, Z Wang, and coauthors
p 5 0:001, Z 2 0:65). Participants were more accurate in the categorisation task
(M 0:97, SD 0:07) than in the recognition task (M 0:88, SD 0:03). The main
effect of the participant race was also significant (F1, 65 7:00, p 5 0:05, Z 2 0:10).
Caucasian participants (M 0:93, SD 0:04) were more accurate than Chinese participants (M 0:91, SD 0:03). The main effect of the face race, however, was not significant
(Caucasian faces: M 0:92, SD 0:04; Chinese faces: M 0:92, SD 0:05). There was
a significant interaction between face race and participant race (F1, 65 9:60, p 5 0:01,
Z 2 0:13).
Moreover, this interaction was further modulated by the task type, inducing a significant three-way interaction (see figure 1b for means and standard errors) (F1, 65 13:15,
p 5 0:01, Z 2 0:17). Paired t-tests showed that Caucasian participants were more
accurate with the Caucasian faces than the Chinese faces (t34 3:69, p 5 0:01), and
Chinese participants were more accurate with the Chinese faces than the Caucasian
faces (t31 ÿ2:31, p 5 0:05) in the recognition task. This result indicates an own-race
recognition advantage in accuracy for both Caucasian and Chinese participants (figure 1b). However, both groups of participants failed to show a significant other-race
categorisation advantage in accuracy in the categorisation task. This is consistent with
previous reports that also failed to show the other-race categorisation advantage in
terms of accuracy, likely as a result of the ease of the categorisation task.
To rule out the possibility that response bias might have unduly affected response
accuracy, we performed a signal detection analysis based on the data from the recognition task. For the three occasions when false alarm (FA) rates were 0, we followed
the conventional correction formulae whereby FA 1=(2N) when the observed FA 0
and N maximum number of false alarms.
An ANOVA with face race (Caucasian versus Chinese) as a within-subjects factor
and participant race as a between-subjects factor was performed on the discriminability measure (d 0 ) for the recognition task. A significant interaction was found for d 0
in the recognition task (figure 1c) (F1, 65 18:657, p 5 0:001, Z 2 0:23; Caucasian
participants on Caucasian faces: mean d 0 3:58, SD 0:62; Caucasian participants
on Chinese faces: mean d 0 2:97, SD 0:74; Chinese participants on Caucasian faces:
mean d 0 3:17, SD 0:79; Chinese participants on Chinese faces: mean d 0 3:33,
SD 0:62). Paired t-tests showed a significant advantage for own-race faces for both
groups (t34 3:33, p 5 0:01 for Caucasian participants; t31 ÿ2:80, p 5 0:01 for Chinese
participants).
However, another ANOVA with face race (Caucasian versus Chinese) as a withinsubjects factor and participant race as a between-subjects factor was performed on the
response bias measure (criterion c) for the recognition task showed neither a significant
main effect nor a significant interaction of face race and participant race (face race:
F1, 64 1:87, ns; participant race: F1, 64 0:28, ns; interaction: F1, 64 0:69, ns). The means
and standard deviations of criterion c for each condition and group are as follows: Caucasian participants on Caucasian facesömean criterion c 0:12, SD 0:29; Caucasian
participants on Chinese faces ömean criterion c 0:15, SD 0:37; Chinese participants
on Caucasian facesömean criterion c 0:05; SD 0:37; Chinese participants on
Chinese facesömean criterion c 0:14, SD 0:37). Thus, the own-race recognition
advantage findings based on the accuracy data were not due to response biases.
3.3 Relationship between the two face-race effects
The above results revealed the anticipated mirror pattern between the two face-race
effects in terms of group mean response latency. To examine whether this mirror pattern
also existed at the individual participant level, we computed the size of the own-race
recognition advantage in reaction time. This measure was obtained by subtracting each
participant's reaction time for the own-race faces from that for the other-race faces in
Other-race effects of recognition and categorisation
7
the recognition task. The difference scores greater than 2 standard deviations were
excluded from the subsequent analysis (four scores in total). Similarly, we computed
the size of the own-race recognition advantage in discriminability by subtracting each
participant's d 0 for recognising the other-race faces from that for recognising the
own-race faces. Further, the size of the other-race categorisation advantage in reaction
time was obtained by subtracting the reaction time of the own-race faces from that of
the other-race faces in the categorisation task to generate a different score for each
participant.
A correlation analysis collapsed across the participant race revealed a significant
negative correlation between the size of the own-race recognition advantage in reaction
time and the size of the other-race categorisation advantage in reaction time (see
figure 2; r63 ÿ0:50, p 5 0:0001), consistent with the mirror pattern found in the
ANOVA analysis. However, the size of the own-race recognition advantage in discriminability and the size of the other-race categorisation effect in reaction time was not
significantly correlated (r63 0:15, ns).
Size of the own-race recognition advantage=ms
200
150
y ÿ0:20666 ÿ 40:103
R 2 0:078
100
50
0
ÿ300
ÿ200
0
ÿ100
100
200
300
400
500
ÿ50
ÿ100
ÿ150
ÿ200
ÿ250
ÿ300
Size of the other-race categorisation advantage=ms
Figure 2. The correlation between the sizes of the own-race recognition advantage in reaction time
(x axis) and the other-race categorisation advantage in reaction time (y axis).
However, the significant relationship found between the size of the own-race recognition advantage in reaction time and the size of the other-race categorisation effect
in reaction time could originate from general factors such as processing speed and
participant race. To rule out these possibilities, two hierarchical regression analyses
were performed so as to examine the unique relationship between the two face-race
effects after these extraneous factors were controlled.
For the first regression analysis, we used the size of the own-race recognition
advantage in reaction time as the dependent measure. At the first step, the participant
race factor was entered into this model to account for any difference between Chinese
and Caucasian participants because a significant effect of race was found in terms
of participant race (see the ANOVA results above). Also, the participant's recognition
reaction time was entered into the model because it was found to be significantly
related to the size of the own-race recognition advantage in reaction time (r63 ÿ0:36,
p 5 0:001). This variable was thus entered into the regression analysis first to partial
out any possible contribution of the overall processing speed in face recognition.
8
L Ge, H Zhang, Z Wang, and coauthors
Table 1. Summary of the hierarchical regression analysis of the relation between the own-race
recognition advantage and the other-race categorisation advantage.
Step
B
Measures
With own-race recognition advantage in reaction
1
Participant race
Correct recognition reaction time
2
Other-race categorisation advantage
3
Other-race categorisation advantage
6participant race
Other-race categorisation advantage
6correct recognition reaction time
Confidence
interval
(upper/lower
bound)
t
2
Rchange
time as dependent variable
0.30***
86.02
ÿ0.14
ÿ0.25
0.89
0.001
41.29/130.75
ÿ0.26/ÿ0:02
ÿ0.42/ÿ0:08
ÿ1.50/3.29
ÿ0.001/0.002
3.85***
ÿ2:37 *
ÿ2.96**
0.72
0.94
Overall R 2
With own-race recognition advantage in discriminability as dependent variable
1
Participant race
0.71
0.34/1.09
3.81***
Correct recognition reaction time
0.00
ÿ0.001/0.001
ÿ0.62
2
Other-race categorisation advantage
0.00
ÿ0.001/0.002
ÿ0.63
3
Other-race categorisation advantage
0.002 ÿ0.02/0.02
0.72
6participant race
Other-race categorisation advantage
0.00
0.00/0.00
0.24
6correct recognition reaction time
Overall R 2
0.09*
0.02
0.41***
0.22**
0.01
0.01
0.23***
Note: *** p 5 0:001; ** p 5 0:01; * p 5 0:05.
2
The model was significant (see table 1; Rchange
0:30, F2, 60, change 12:94, p 5 0:0001).
Consistent with the above ANOVA finding, Caucasian participants had a greater ownrace recognition advantage in reaction time than Chinese participants. Also, the faster
the correct recognition reaction time, the greater the own-race recognition advantage
in reaction time.
Second, the critical factor, the size of the other-race categorisation advantage
in reaction time, was entered into the model. This second step was also significant
2
(Rchange
0:09, F1, 59, change 8:76, p 5 0:01; table 1). Thus, after partialling out the effect
of the participant race and reaction time in face recognition, the size of the own-race
recognition advantage was still significantly related to the size of the other-race categorisation advantage in terms of response latency. The faster one recognised own-race
faces than other-race faces, the slower one categorised own-race faces relative to
other-race faces. At the third step, we further tested the effect of the interaction of
the size of the other-race categorisation advantage with the participant race and the
reaction time in face recognition on the size of the own-race recognition advantage.
The interaction terms were not significant (table 1).
For the second regression analysis, we used the size of the own-race recognition
advantage in discriminability as the dependent variable, which was obtained by subtracting the d 0 for recognising the other-race faces from that for recognising the
own race faces. The same predictors used in the first regression analysis were used in
this second regression analysis with the predictors and their interaction terms entered
into the analysis in three hierarchical steps. The crucial second and third steps were not
significant, suggesting that the own-race recognition advantage in discriminability was
not significantly related to the other-race categorisation advantage in reaction time (table 1).
Other-race effects of recognition and categorisation
9
4 Discussion
In the present study, we concurrently examined the mirror pattern of the own-race
recognition advantage and the other-race categorisation advantage. As expected, Caucasian participants recognised Caucasian faces more accurately and faster than Chinese
faces but categorised Chinese faces faster than Caucasian faces. This mirror pattern
was completely replicated with Chinese participants who recognised Chinese faces more
accurately and faster than Caucasian faces but categorised Caucasian faces faster than
Chinese faces.
This finding is the first in the literature to obtain a clear mirror pattern of the ownrace and other-race face recognition and categorisation effects in terms of the reaction
time. The fact that Chinese participants' responses were entirely opposite to Caucasian
participants' responses when recognising and categorising the same face stimuli rules
out the possibility that our findings were due to the specifics of face stimuli used. Further,
because our recognition and categorisation tasks had highly similar tasks structure and
demand, we can confidently attribute the mirror pattern of the own-race and other-race
effects to processing differences involved in recognising and categorising own-race and
other-race faces.
More importantly, our correlational analyses revealed that the own-race recognition advantage in reaction time was significantly related to the other-race categorisation
advantage in reaction time. Further, the hierarchical regression analyses confirmed
that this relation was still significant after partialling out the effects of such factors as
participants' recognition reaction time and race. In particular, the faster participants
recognised their own-race faces relative to other-race faces, the slower they categorised
their own-race faces. This result suggests that the own-race recognition advantage is
closely related to the other-race categorisation advantage in terms of processing speed.
An implication is that expertise at efficiently recognising one's own-race faces is not a
cost-free accomplishment. It is achieved at the cost of efficient categorisation of such
faces. This finding is in line with the hypothesis that the processes of recognition and
categorisation may have a common underlying processing mechanism. Recognising
faces with which one has a high level of expertise interferes with categorising them.
However, the exact nature of this interference is unclear. The lack of significant
correlation between the own-race recognition advantage in discriminability and the
other-race categorisation advantage in reaction time suggests that the interference may
be rooted in efficiency rather than accuracy. In other words, the increased expertise
at recognising own-race faces may negatively affect the speed by which we categorise
such faces but not whether we can categorise the faces correctly. This possibility
explains in part the inconsistent findings by Levin (1996, 2000) who typically used
accuracy or discriminability to measure the own-race recognition advantage but latency
to measure the other-race categorisation advantage. However, it is premature to accept
this efficiency hypothesis. This is because we, like many other previous researchers,
were unable to obtain the other-race categorisation advantage in discriminability owing
to participants' near-ceiling performance in the categorisation task. Thus, at this point,
the current finding regarding the interaction between the cross-race face categorisation
and recognition effects should be interpreted with caution owing to such limitations.
Future studies need to increase the difficulty level of the categorisation task (eg reducing
the viewing time) to avoid ceiling performance. In this way, one can ascertain whether the
mirror pattern of the two cross-race effects exists in terms of not only response latency
but also discriminability.
Another point is also worth noting. It is possible that the mirror pattern of the ownrace and other-race effects observed here is a manifestation of a broader phenomenon.
It has been found that people recognise faces of their own age and gender better than faces
of other ages or gender (Anastasi and Rhodes 2005, 2006; Wright and Sladden 2003).
10
L Ge, H Zhang, Z Wang, and coauthors
It has been recently found that when faces were assigned into arbitrary in-group and
out-group categories, participants recognised in-group faces better than the out-group
faces, similar to the own-race recognition advantage (Bernstein et al 2007). Although
concurrent categorisation studies have yet to be conducted with regard to categorising
own-age and other-age, gender or arbitrary in-group and out-group faces, it seems
reasonable to conjecture that a mirror pattern of the recognition and categorisation
effects will be observed beyond cross-race face processing.
There have been some suggestions why increased experience and expertise with
processing one category of faces should affect detrimentally the categorisation of faces
of this category. One influential theoretical framework (Valentine 1991) suggests that
faces are represented in a multidimensional space where each dimension represents a
type of perceptually relevant face information. It has been hypothesised that the distances between face representations are tuned by experience. Compared with own-race
faces, the distances between other-race faces are shorter and hence form a higher
density (Byatt and Rhodes 2004). This higher density interferes with individual identification which is needed in the recognition task because each other-race face can be
confused with its neighbours. However, this higher density can benefit categorisation at
the group level because of the increased activation of the other-race faces as a group.
Another possibility proposed by Levin (1996, 2000) is that categorisation may not
necessarily take place earlier than individuation. The sequence of categorisation and
individuation depends on one's processing expertise. Levin (1996, 2000) argued that when
individuals process a category of faces with which they have limited experience, categorical information is encoded first, followed by individuating information. In contrast,
when individuals process a category of faces for which they have expertise, individuating
information is automatically encoded first, followed by categorical information. This
hypothesis predicts that the response latency in recognition of own-race faces should
be faster than categorisation of the same faces, and recognition of other-race faces
should be slower than categorisation of the same faces.
A third alternative hypothesis is that individuals devote differential processing
resources (eg attention) to a face's categorical and individuating information depending
on whether the face is in-group and familiar versus out-group and unfamiliar (Sporer
2001). When encountering unfamiliar out-group faces, individuals may devote more
resources to categorical information than individuating information. In contrast, when
encountering familiar in-group faces, individuals may devote more resources to individuating information. Indeed, recent studies showed that the improvement in recognition
of other-race faces can be achieved by directing participants to attend to the individuating
information of other-race faces (Hills and Lewis 2006; Hugenberg et al 2007).
Specifically designed studies are needed to test these different hypotheses directly.
For example, it has been found that own-race faces are represented in a more densely
distributed fashion than other-race faces (Byatt and Rhodes 2004). To confirm the
multidimensional hypothesis, one needs to assess whether such difference in distribution density indeed has differential consequences for recognition and categorisation of
own-race and other-race faces (see Jaquet et al 2008 for possible methods to test such
possibilities). To test Levin's hypothesis, one needs to design tasks especially sensitive
to the sequences of processing of face race and identity information. Sporer's hypothesis
on the other hand requires obtaining information about resource allocation during the
processing of own-race and other-race faces. Evidence to support and disconfirm
these hypotheses not only will shed light on the exact relation between other-race recognition and categorisation but also on the nature of cognitive mechanisms underlying face
identity and category processing.
In summary, the present study used identical face stimuli and task demand and
structure and revealed that both Chinese and Caucasian participants showed the same
Other-race effects of recognition and categorisation
11
mirror pattern of the own-race recognition advantage and other-race categorisation
advantage in terms of reaction time but not in terms of discriminability. Further, after
partialling out the effects of various factors, the speed at which individuals categorised
other-race faces as opposed to own-race faces was significantly related to the speed at
which they individuated own-race faces as opposed to other-race faces. This significant
negative correlation suggests an antagonistic relationship between face individuating and
categorisation in reaction time: increased efficiency at individuating faces may come
at the cost of efficiency at categorising the same faces. This antagonistic relationship
may be a general face processing phenomenon and reflects competition between categorisation and individuation of faces with which individuals have different levels of expertise.
Acknowledgment. The study was supported by grants from the Natural Science Foundations of
China (30528027 -28 and 30770710), NSERC, and NIH (R01 HD46526).
References
Anastasi J S, Rhodes M G, 2005 ``An own-age bias in face recognition for children and older adults''
Psychonomic Bulletin & Review 12 1043 ^ 1047
Anastasi J S, Rhodes M G, 2006 ``Evidence for an own-age bias in face recognition'' North American
Journal of Psychology 8 237 ^ 252
Bernstein M J, Young S G, Hugenberg K, 2007 ``The cross-category effect: Mere social categorization is sufficient to elicit an own-group bias in face recognition'' Psychological Science 18
706 ^ 712
Bothwell R K, Brigham J C, Malpass R S, 1989 ``Cross-racial identification'' Personality and Social
Psychology Bulletin 15 19 ^ 25
Brigham J C, Maas A, Snyder L D, Spaulding K, 1982 ``Accuracy of eyewitness identifications
in a field setting'' Journal of Personality and Social Psychology 42 673 ^ 681
Brigham J C, Malpass R S, 1985 ``The role of experience and contact in the recognition of faces
of own- and other-race persons'' Journal of Social Issues 41 139 ^ 155
Bruce V, 1986 ``Influences of familiarity on the processing of faces'' Perception 15 387 ^ 397
Bruce V, Young A, 1986 ``Understanding face recognition'' British Journal of Psychology 77 305 ^ 327
Bruyer R, Leclere S, Quinet P, 2004 ``Ethnic categorisation of faces is not independent of face
identity'' Perception 33 169 ^ 179
Burton A M, Bruce V, Johnston R A, 1990 ``Understanding face recognition with an interactive
activation model'' British Journal of Psychology 81 361 ^ 380
Byatt G, Rhodes G, 2004 ``Identification of own-race and other-race faces: Implications for the
representation of race in face space'' Psychonomic Bulletin & Review 11 735 ^ 741
Caldara R, Rossion B, Bovet P, Hauert C A, 2004 ``Event-related potentials and time course of
the `other-race' face classification advantage'' NeuroReport 15 905 ^ 910
Chance J E, Turner A L, Goldstein A G, 1982 ``Development of differential recognition for ownrace and other-race faces'' Journal of Psychology 112 29 ^ 37
Chiroro P, Valentine T, 1995 ``An investigation of the contact hypothesis of the own-race bias in
face recognition'' Quarterly Journal of Experimental Psychology, Section A: Human Experimental Psychology 48 879 ^ 894
Cross J F, Cross J, Daly J, 1971 ``Sex, race, age, and beauty as factors in recognition of faces''
Perception & Psychophysics 10 393 ^ 396
Etcoff N L, 1984 ``Selective attention to facial identity and facial emotion'' Neuropsychologia 22
281 ^ 295
Ganel T, Goshen-Gottstein Y, 2002 ``Perceptual integrality of sex and identity of faces: Further
evidence for the single-route hypothesis'' Journal of Experimental Psychology: Human Perception
and Performance 28 854 ^ 867
Golby A J, Gabrieli J D E, Chiao J Y, Eberhardt J L, 2001 ``Differential responses in the fusiform
region to same-race and other-race faces'' Nature Neuroscience 4 845 ^ 850
Haxby J V, Hoffman E A, Gobbini M I, 2000 ``The distributed human neural system for face
perception'' Trends in Cognitive Sciences 4 223 ^ 233
Hills P J, Lewis M B, 2006 ``Reducing the own-race bias in face recognition by shifting attention''
Quarterly Journal of Experimental Psychology 59 996 ^ 1002
Hugenberg K, Miller J, Claypool H M, 2007 ``Categorization and individuation in the cross-race
recognition deficit: Toward a solution to an insidious problem'' Journal of Experimental Social
Psychology 43 334 ^ 340
12
L Ge, H Zhang, Z Wang, and coauthors
Humphreys G W, Donnelly N, Riddoch M J, 1993 ``Expression is computed separately from facial
identity, and it is computed separately for moving and static faces: Neuropsychological evidence''
Neuropsychologia 31 173 ^ 181
Jaquet E, Rhodes G, Hayward W G, 2008 ``Race contingent aftereffects suggest distinct perceptual
norms for different race faces'' Visual Cognition 16 734 ^ 753
Le Gal P M, Bruce V, 2002 ``Evaluating the independence of sex and expression in judgments
of faces'' Perception & Psychophysics 64 230 ^ 243
Levin D T, 1996 ``Classifying faces by race: The structure of face categories'' Journal of Experimental Psychology: Learning, Memory, and Cognition 22 1364 ^ 1382
Levin D T, 2000 ``Race as a visual feature: Using visual search and perceptual discrimination tasks
to understand face categories and the cross-race recognition deficit'' Journal of Experimental
Psychology: General 129 559 ^ 574
Meissner C A, Brigham J C 2001 ``Thirty years of investigating the own-race bias in memory
for faces öA meta-analytic review'' Psychology Public Policy and Law 7 3 ^ 35
Pezdek K, Blandon-Gitlin L, Moore C, 2003 ``Children's face recognition memory: More evidence
for the cross-rate effect'' Journal of Applied Psychology 88 760 ^ 763
Rhodes G, Brake S, Taylor K, Tan S, 1989 ``Expertise and configural coding in face recognition''
British Journal of Psychology 80 313 ^ 331
Sangrigoli S, Schonen S de, 2004a ``Effect of visual experience on face processing: a developmental
study of inversion and non-native effects'' Developmental Science 7 74 ^ 87
Sangrigoli S, Schonen S de, 2004b ``Recognition of own-race and other-race faces by three-monthold infants'' Journal of Child Psychology and Psychiatry 45 1219 ^ 1227
Shepherd J W, Deregowski J B, 1981 ``Races and facesö a comparison of the responses of Africans
and Europeans to faces of the same and different races'' British Journal of Social Psychology
20 125 ^ 133
Sporer S L, 2001 ``Recognizing faces of other ethnic groupsöAn integration of theories'' Psychology
Public Policy and Law 7 36 ^ 97
Stroessner S J, 1996 ``Social categorization by race or sex: Effects of perceived non-normalcy on
response times'' Social Cognition 14 247 ^ 276
Valentine T, 1991 ``A unified account of the effects of distinctiveness, inversion, and race in face
recognition'' Quarterly Journal of Experimental Psychology, Section A: Human Experimental
Psychology 43 161 ^ 204
Valentine T, Endo M, 1992 ``Towards an exemplar model of face processingöthe effects of race and
distinctiveness'' Quarterly Journal of Experimental Psychology, Section A: Human Experimental
Psychology 44 671 ^ 703
Wright D B, Sladden B, 2003 ``An own gender bias and the importance of hair in face recognition''
Acta Psychologica 114 101 ^ 114
Young A W, Hay D C, McWeeny K H, Flude B M, Ellis A W, 1985 ``Matching familiar and
unfamiliar faces on internal and external features'' Perception 14 737 ^ 746
Young A W, Newcombe F, Dehaan E H F, Small M, Hay D C, 1993 ``Face perception after brain
injury öSelective impairments affecting identity and expression'' Brain 116 941 ^ 959
Zarate M A, Smith E R, 1990 ``Person categorization and stereotyping'' Social Cognition 8 161 ^ 185
ß 2009 a Pion publication
ISSN 0301-0066 (print)
ISSN 1468-4233 (electronic)
www.perceptionweb.com
This article is an advance online publication. It will not change in content under normal circumstances
but will be given full volume, issue, and page numbers in the final PDF version, which will be made
available shortly before production of the printed version.
Conditions of use. This article may be downloaded from the Perception website for personal research
by members of subscribing organisations. Authors are entitled to distribute their own article (in printed
form or by e-mail) to up to 50 people. This PDF may not be placed on any website (or other online
distribution system) without permission of the publisher.