Own-Race Faces Capture Attention Faster than Other-Race Faces: Evidence from Response Time and the N2pc

Studies have shown that people are better at recognizing human faces from their own-race than from other-races, an effect often termed the Own-Race Advantage. The current study investigates whether there is an Own-Race Advantage in attention and its neural correlates. Participants were asked to search for a human face among animal faces. Experiment 1 showed a classic Own-Race Advantage in response time both for Chinese and Black South African participants. Using event-related potentials (ERPs), Experiment 2 showed a similar Own-Race Advantage in response time for both upright faces and inverted faces. Moreover, the latency of N2pc for own-race faces was earlier than that for other-race faces. These results suggested that own-race faces capture attention more efficiently than other-race faces.


Introduction
Extensive research has provided evidence that people process own-race faces better than otherrace faces (see [1] for a meta-analytic review). The Own-Race Advantage (ORA) effect (also known as the own-race effect, own-race bias, or in-group advantage, see [2]) has been traditionally found in face recognition tasks [1], and recently was found in a configural processing task [3] as well as in a feature processing task [4]. Some researchers are concerned about whether there is an ORA in attention, but the evidence is not consistent and the underlying mechanisms are not clear.
Some research shows that own-race faces capture attention more efficiently than other-race faces. For example, Hodsoll et al. [5] investigated whether the preferential allocation of attention to infant faces was influenced by the race of the faces and of the perceivers, by using a dotprobe task. Their data showed that participants responded faster to infant faces that appeared in the same location as own-race faces than other-race faces. Golby et al. [6] asked participants undergoing functional MRI to view own-race faces, other-race faces and objects, and to press a human face detection task (e.g., to search or identify a human face among animal faces) does not need racial categorization or individuation processing; asking participants to detect human faces among animal faces is an effective way to get at the unique effect of attention. Experiment 1 employed a visual search paradigm as previously used by Levin [10] except that we instructed participants to judge whether human faces were present in a particular set of stimuli, instead of performing a race-searching task. If there is an own-race advantage in attentional capture, we expected to observe faster detection of own-race faces than other-race faces when these were placed among animal faces. Experiment 2 was designed to explore the neural correlates of ORA in a human face detection task.

Method
Participants. Twenty-one Chinese participants (9 males, 12 females between the ages of 19 and 32 years, M = 23 years, SD = 3 years) from Sun Yat-Sen University, and twenty-two Black African participants (11 males, 11 females between the ages of 18 and 28 years, M = 22 years, SD = 4 years) from the University of Cape Town were paid to participate in this experiment. All participants had normal or corrected-to-normal vision. The study was approved by the Human Research Ethics Committee of Department of Psychology in Sun Yat-Sen University and that in University of Cape Town. The participants all gave their written informed consent before taking part in the experiment.
Materials. Photographs of three African American male faces [21] and three Chinese male faces collected in China were used as human faces. They were all grey-scale, front-view, digitally sectioned head shots with neutral facial expressions, with no glasses, beard, ears, or hair. Six animal faces served as distractors: three dogs and three cats. All these photos were digitally sectioned in the same way as the human faces. All faces were matched in luminance and contrast with Photoshop software; each subtended a visual angle of 1.7°×2.1°(see Fig 1a). The individual in Fig 1 has given written photo release form (as outlined in the PLOS consent form) to publish these case details.
Design. A 2 (Race of participants: Black or Chinese) × 2 (Race of human faces: Black or Chinese) × 3 (Set size: 2, 4, 6) mixed design was employed, with the Race of participants as a between-subjects variable.
Procedure. Participants were individually tested in a dimly lit room. The experiment was developed and executed with E-Prime (http://www.pstnet.com/eprime.cfm). All stimuli were presented at a viewing distance of 70 cm. Each trial began with a centrally presented fixation cross for 500 ms. Participants were asked to maintain gaze on the fixation cross, which was followed immediately by the search array, which remained on the screen until participants made responses. Then the next trial began. Half of the trials contained a target, and the other half of the trials contained no target face. In the target-present trial, one target human face (Black or Chinese) and 1, 3 or 5 animal faces (dogs' and cats' faces) were symmetrically positioned in one of the six positions around the central fixation point, with a radius of 5°. The number of animal faces was 1, 3 or 5, and the target face was equally likely to appear in the six possible positions. In the target-absent trials, no human faces were present, and there were 2, 4 or 6 animal faces on the screen. In each trial the participants were asked to search for the human face quickly and accurately, and respond using the keyboard by pressing the "f" key if a human face was present and the "j" key if no human face was present. We conjecture that such a human face detection task does not need racial categorization or individuation processing. Therefore, we used it to get at the unique effect of attention.
Participants completed nine practice trials before the formal test. The formal test included four blocks. In each block, half of the trials were target-present trials and half were targetabsent trials; in the target-present trials, Black faces and Chinese faces served as targets with equal probability. All conditions were randomized. Each block contained 108 trials, and there were 432 formal trials in total.

Results
Response time. Only correct human face responses were included in the analysis. Trials with response times (RTs) shorter than 100ms were discarded. Trials with RTs greater than 3 standard deviations from the mean of each condition were discarded (about 1.5% of all trials). Mean RTs for each participant were submitted to a mixed analysis of variance (ANOVA) with Race of faces (Black vs. Chinese) and Set size (2, 4, 6 items) as within-subjects variables and Race of participants (Black vs. Chinese) as a between-subject variable (see Fig 2).
There was a significant main effect of Race of faces

Accuracy
The accuracy for all participants was above 80% (see Table 1  We found a significant own-race advantage in a human face search task, which was inconsistent with the detection preference towards other-race faces in the race searching task reported by Levin [10]. The present ORA result was consistent with the own-race advantage in a task which requires individualization [18][19][20]. This provides a possibility that the own-race advantage in the current study comes from individual processing rather than attention. This possibility was further tested in Experiment 2.
A limitation of Experiment 1 is the small set of human face stimuli. As a consequence, it is very likely that idiosyncratic characteristics of the images or the depicted persons, and not race per se, drive the ORA effect. This should not be a very significant factor though, because the ORA was observed for both Chinese participants and Black participants, which indicates that the observed ORA effect was not related to the stimuli per se, but to the interaction of stimulus and participants' race.

Experiment 2
Experiment 2 was designed to explore the neural correlates of ORA in a human face detection task and to explore the underlying mechanism by recording N2pc (N2-posterior-contralateral), N170, and N250 waves.
First, we recorded N2pc to test whether the ORA in RT in Experiment 1 was related to attention. The N2pc wave, an attention-related ERP component, is used to measure the attentional processing of faces in the present study. The N2pc wave is typically elicited at posterior scalp sites contralateral to the position of a task-relevant visual stimulus between 200 and 300 ms after stimulus onset, as previously observed in experiments investigating the allocation of attention in visual search tasks [22][23][24]. The N2pc latency is correlated with rapid shifts of attention [25] [26]. If ORA in RT is related to the ability to capture attention (i.e. own-race faces capture attention more efficiently than other-race faces), searching for own-race faces would produce an earlier N2pc latency than searching for other-race faces.
Second, we recorded N170 to test whether the ORA in RT in Experiment 1 is related to categorization processing or configural processing. The N170 is a negative potential recorded around 170 ms after stimulus onset at posterior temporal sites, larger for faces compared with objects, and is believed to reflect face-specific configural processing [27][28][29] and has been used to measure race category experience (see [30] for a review). Studies using the categorization tasks found that upright own-race faces elicited larger [13] or lower [31] N170 amplitude than upright other-race faces. Although both studies used race categorization tasks, the former asked participants to indicate whether the face was Black or White, the latter required participants to identify whether a face belonged to their own race or not. However, studies using raceirrelevant tasks [32] [33] did not observe a race difference on N170 amplitude (see Wiese [34]  for a review on how N170 varies with task demands). Therefore, if ORA in RT in Experiment 1 is related to race categorization, we would expect to observe a race difference in N170. A configural processing advantage of the own-race faces could also contribute to the ORA effect [3] in the human search task. Inversion can impair the configural encoding of faces [35] [ 36], and is reflected in a latency delay and an amplitude increase of the N170 [27][28][29] (i.e., inversion effect). Therefore, Experiment 2 further examined the possible differences between races on configural processing of faces by adopting upright faces and inverted faces as stimuli. If the ORA in RT is related to configural processing, we would observe an inversion effect for own-race faces but not for other-race faces in RT and in N170.
Finally, we recorded N250 to testify whether the ORA in RT in Experiment 1 is related to individuation processing. N250 is a bilateral component that shows an ongoing negativity that peaks around 250 ms after stimulus onset, which measures the subordinate-level categorization or individuation (see [37] for a review). Several studies have found that other-race faces elicit more negative N250 than own-race faces [38] [39]. Compared with basic-level training, subordinate-level individuation training for other-race faces elicited an increased response in N250 component [40]. If ORA in RT is related to individuation processing, we would expect to observe a race difference in the N250.

Method
Participants. Eighteen Chinese participants were paid to participate in this experiment. Two of them were excluded for eye movement artifacts or excessive alpha activity, leaving 16 participants (6 males, 10 females) between the ages of 19 and 27 years (M = 22 years, SD = 2.09), with normal or corrected-to-normal vision. One participant was left-handed, and others were right-handed. The study was approved by the Human Research Ethics Committee of Department of Psychology in Sun Yat-Sen University. The participants all gave their written informed consent before taking part in the experiment.
Materials. The experiment materials were exactly the same as those used in Experiment 1 except the following two differences: First, only four faces in a search array were used in the present experiment, given that the difference in detection time between Chinese faces and Black faces was large enough to show itself for a set size of four faces in Experiment 1. Second, inverted versions of human faces and animal faces were included in the present experiment to investigate whether there was configural processing involved.
Procedure. Participants were seated 70 cm from the screen in a dimly lit room. The experiment was run in E-Prime (http://www.pstnet.com/eprime.cfm). Each trial (see Fig 1b) began with a centrally presented fixation point for 400-800 ms, followed immediately by a search array, which remained on screen for 1500 ms, containing four faces at a distance of 4°from central fixation. The inter-trial interval (ISI) jittered from 400-600 ms. In each trial participants were asked to search for the human face among animal faces as quickly and accurately as possible. If a human face was present, participants were required to press the number key "8". If no human face was present, no response was needed. Participants were also instructed to keep fixating on the center of the screen during the trials.
Participants finished 16 practice trials before the formal test. The formal experiment was divided into 12 blocks, with half of the blocks for upright faces and the other half for inverted faces. The sequence of conditions was counterbalanced across participants. The target face was present in half of the trials; own-race faces and other-race faces were designated as the target with equal probability. Each block contained 96 trials. There were 1152 trials in total and 144 trials in each experimental condition. Participants were allowed to take a rest between blocks.
EEG recording and data preprocessing. Electroencephalogram (EEG) signals were recorded from 64-channel Ag/AgCl electrodes (ANT, Netherlands) mounted into an elastic cap according to the International 10-20 System. These electrodes were recorded using a right mastoid reference electrode, and the signals were re-referenced offline. Vertical eye movements were monitored (Electrooculogram, EOG) with electrodes above and below the left eye against the reference. Horizontal eye movements were recorded with two electrodes placed at the outer canthi of each eye. Impedance for each electrode was kept below 5 kO. For the N170, the average signal from all electrodes ('average reference') was used as reference. For the N2pc wave, all signals were digitally re-referenced offline to the average of the left and right mastoids. All recordings were amplified with a band pass of 0.01-100 Hz. Data were sampled at 512 Hz.
EEG analysis was performed with the assistance of the EEGLAB 6.0 toolbox. Raw data were band-pass filtered between 0.01 and 40 Hz. EEG was then segmented into epochs from 100 ms before the stimulus onset to 400 ms post-stimulus. Baseline correction was performed using the first 100 ms of the epoch. Trials with artifacts exceeding ±60μV in the horizontal and vertical EOG channels were eliminated. Trials that showed amplitudes exceeding ±100 μV for other channels were also eliminated. About 20-25% of trials were excluded on average due to artifacts. We also examined the averaged horizontal EOG waveforms after artifact rejection to ensure that small eye movements did not contaminate the average ERP waveforms. Incorrect responses and response time (RT) that were three standard deviations away from the mean in each experimental condition, which counted for about 1.7% of all trials, were excluded from final analysis. Thereafter, grand-averaged waveforms were calculated separately for all experimental condition combinations: Target race (Black, Chinese), Orientation (upright, inverted).
Mean amplitudes for N170 were measured at electrodes O1, O2, PO7, PO8, for the time window 180-230ms. Mean amplitudes for N250 were measured at these electrodes for the time window 260#x2013;320ms. The peak latency of N2pc was measured at the same four electrodes.

Results
Response time. A repeated measures analysis of variance (ANOVA) with Race (Black, Chinese) and Orientation (upright, inverted) as within-subjects factors was calculated on mean response times. As in Experiment 1, the main effect of Race was significant, F (1, 15) = 8.07, p <.05, η p 2 = .35. That is, Chinese participants searched human faces faster when the faces were Chinese (M = 638.67ms, SE = 30.66ms) than when they were Black (M = 652.80ms, SE = 28.50ms). Orientation produced a significant main effect, F (1, 15) = 37.36, p < .0001, η p 2 = .71, with longer RT for inverted faces than for upright faces. The Race × Orientation interaction was not significant, F (1, 15) = 2.11, p = .17, η p 2 = .12, indicating that upright and inverted faces showed the same own-race advantage pattern (shown in Table 2). Accuracy. The overall accuracy for each participant was above 90%. A repeated measures analysis of variance (ANOVA) was conducted. Only a main effect of Orientation was observed, Although the four-way interaction among Race, Orientation, Hemisphere and Electrode was significant, F (1, 15) = 6.07, p < .05, η p 2 = .29, the analysis of separate Hemisphere or Electrode did not show any significant interaction of Race and Orientation (ps >.31). Other main effects and interactions did not reach significance (ps > .15). N250. The four-way ANOVA for the mean amplitudes of N250 (see Fig 3) revealed a significant main effect of Hemisphere, F (1, 15) = 6.66, p < .05, η p 2 = .31, indicating that a reliable N250 effect was more negative at left hemisphere than right hemisphere; and that there was a significant interaction of Hemisphere and Electrode, F (1, 15) = 7.56, p < .05, η p 2 = .34. A simple effect analysis showed a significantly more negative effect for PO7 than PO8 (p = .01), but a marginally significant more negative effect for O1 than O2 (p = .06). Other main effects and interactions did not reach significance (ps > .14).
N2pc. The N2pc wave was measured as the deflection recorded, contralateral to the visual field where the target appeared. Specifically, the N2pc amplitude is larger at electrode sites contralateral to the position of the target compared to ipsilateral sites. To quantify the peak latency of the N2pc wave, the averaged waveform in each condition recorded at the ipsilateral electrode sites was subtracted from the corresponding waveform recorded at the contralateral electrode sites at the occipital electrodes (i.e. when the target appeared in left visual field, PO8-PO7, O2-O1; when target appeared in right visual field, PO7-PO8, O1-O2). The N2pc wave can be seen as a more negative (i.e. less positive) voltage beginning at approximately 300 ms poststimulus. The grand average ERP waveform is shown in Fig 4. Within-subjects analysis of variance (ANOVA) with factors of Race (Black, Chinese), Orientation (upright, inverted) and Electrode (O1/O2, PO7/PO8) was employed for the N2pc latency. The ANOVA revealed a significant main effect of Race, F (1, 15) = 5.42, p < .05, η p 2 = .27, indicating that the latency for Chinese faces (Mean = 328.09ms, SE = 5.84ms) was earlier than for Black faces (Mean = 335.48ms, SE = 5.22ms). The main effect of Orientation was N2pc waves compared with inverted faces. These results confirmed that own-race faces captured attention more quickly than other-race faces, and upright faces attracted attention more efficiently than inverted faces. The Race × Orientation interaction was not significant, F (1, 15) = 2.80, p = .12, η p 2 = .16. These ERP results were consistent with the behavioral results (see Fig 4).
Correlation between RTs and N2pc latencies. We computed Pearson correlations for the differences of Chinese and Black faces in response time (RT: Chinese-Black) with those in N2pc peak latency (N2pc latency: Chinese-Black) separately for the upright faces and the inverted faces. For upright faces, the difference in detection was marginally significantly correlated with the difference in the N2pc latency for electrode O1/O2, r(16) = .49, p = .06, and for PO7/PO8, r(16) = .45, p = .08. For inverted faces, their correlation was also marginally significant for electrode O1/O2, r(16) = .47, p = .07, but did not reach clear or marginal significance for PO7/PO8, r(16) = .13, p = .63. The tendency of this correlation implies that the larger the ORA in detection time, the earlier N2pc waves were evoked by own-race faces than other-race faces (see Fig 5). This result suggests that faster detection of own-race faces is related to faster attention being directed to them.

Discussion
Experiment 2 replicated the ORA in RT. We attribute this ORA to attentional capture, which was confirmed by the earlier onset of N2pc waves for own-race faces than for other-race faces, and was further supported by the marginally significantly positive correlation between RTs and N2pc latencies of ORA size.
This Own-Race Advantage in behavior was not due to more configural processing for ownrace faces than for other-race faces for two reasons. First, we did not observe a larger inversion effect for own-race faces than that for other-race faces in RT; N170 inversion effects were not observed for either own-or other-race faces. Second, own-race faces and other-race faces showed no difference in N170, which was regarded as an indicator of configural processing. This result was in line with previous research [32] [33] that used a race-irrelevant task, but inconsistent with other research [13] [31] that used a race categorization task; this suggests that the ORA in behavior is not related to race category processing. Moreover, this ORA in behavior is not due to more individual processing for own-race faces than for other-race faces, because no race difference was observed in N250.
We found an opposite (marginally significant) inversion effect for the N170, inconsistent with previous studies [27][28][29]. A possible reason is that human faces were presented intermixed Own-Race Faces Capture Attention with animal faces in the present study, since animal faces cannot elicit N170 amplitude inversion effects [41].
Although some previous studies have already found that own-race faces evoke a larger N2 wave, they used a categorization task [12][13][14] or an individual personality judgment task [13]. In addition, the posterior N2 is sensitive to different task-relevant visual features (a discrimination task), while the N2pc has been associated with the selection of the targets and the suppression of irrelevant distractors (a detection task) [42].

General Discussion
In the present study, we found behavioral and ERP evidence that own-race and other-race faces are not processed equally even when no racial categorization or individuation is needed. The target human faces were detected faster when they were own-race faces than when they were other-race faces. More importantly, own-race faces elicited an earlier N2pc peak latency than other-race faces, which was correlated with faster detection. Those results indicate attentional superiority for own-race faces, consistent with previous results [5] [6], but inconsistent with the findings using a change blindness paradigm [8] [9] or a race-searching task [10] [11]. As we argued in the introduction, the discrepancy between these studies is due to the different processing levels required by the tasks. There is an other-race advantage in race category processing [10] [17] [18] and an own-race advantage in individuation processing [1] [18][19][20]. Therefore, the present study provides important evidence that there is an own-race advantage in the detection and differentiation of human/non-human faces. The own-race advantage at the human/non-human level of face processing implies a preference for own-race faces in early perceptual processing. Specifically, an earlier N2pc in the present study was elicited in response to own-race faces than other-race faces among animal faces, demonstrating that the own-race faces can capture the attention more successful than otherrace faces in a visual search array. This pattern of results could be explained from several aspects: Firstly, this advantage may be shaped by experience with own-race faces. That is, it may be due to the familiarity of stimuli. Previous studies show that target familiarity speeds up the allocation of visual-spatial attention [43] [44]. For example, Christie and Klein [43] used an attention allocation task to indicate whether the target position changed upward or downward, and their results showed better performance for word targets than non-word targets. Their findings demonstrate that familiar items may rapidly attract attention. Similarly, Chanon and Hopfinger [44] recorded participants' eye movements when they viewed a scene with an old encoded target in it or the same scene with a new target in it. They found that old objects were fixated on significantly sooner than new objects, suggesting that attention could be exogenously drawn to the location of familiar items. Secondly, from an evolutionary perspective, because most interactions take place with in-group members, people are likely to be motivated to allocate attentional resources to own-race faces prior to maximize the expected utility (see [45] for a utilitarian theory account). Own-race faces can be regarded as high-valued stimuli. They have priority to be processed because valued stimuli can capture attention [46].
A limitation of the present study is that we did not explore whether the ORA in attention can explain the ORA in the traditional face memory task. However, recently researchers indeed found that participants spend more time on own-group faces than other-group faces during a self-paced face learning phase, and the individual differences in attention to own-group members during learning mediated superior memory for own-group members [47]. Divided attention during a face learning phase indeed impairs memory for own-race faces more than for other-race faces [48].
Another limitation of the present study is that we did not explore how attention interacts with some other processes in cross-race face processing, such as categorization or individualization. We suggest that attention varies with task demands: Categorization processes need more attention for other-race faces, and individual processing needs more attention for ownrace faces. Further research is needed to resolve this issue.
In summary, the present study found an ORA in a human searching task with faster detection time, a shorter latency of N2pc for own-race faces than other-race faces. These results indicate that we have better performance for own-race faces than other-race faces in the early processing stage. Moreover, the present results provide electrophysiological evidence that ownrace faces attract attention more efficiently than other-race faces.