Memory for diverse faces in a racially attentive context

The Cost of Racial Salience on Face Memory: How the Cross-Race Effect is Moderated by Racial Ambiguity and the Race of the Perceiver and the Perceived

The other race effect (ORE) in part describes how people are poorer at identifying faces of other races compared to own-race faces. While well-established with face memory, more recent studies have begun to demonstrate its presence in face matching tasks, with minimal memory requirements. However, several of these studies failed to compare both races of faces and participants in order to fully test the predictions of the ORE. Here, we utilized images of both Black and White individuals, and Black and White participants, as well as tasks measuring perceptions of face matching and similarity. In addition, human judgements were directly compared with computer algorithms. First, we found only partial support for an ORE in face matching. Second, a deep convolutional neural network (residual network with 29 layers) performed exceptionally well with both races. The DCNN’s representations were strongly associated with human perceptions. Taken together, we found that the ORE was not robust or compelling in our human data, and was absent in the computer algorithms we tested. We discuss our results in the context of ORE literature, and the importance of state-of-the-art algorithms.

The cross-race effect (CRE) describes the finding that same-race faces are recognized more accurately than cross-race faces. According to social-cognitive theories of the CRE, processes of categorization and individuation at encoding account for differential recognition of same- and cross-race faces. Recent face memory research has suggested that similar but distinct categorization and individuation processes also occur postencoding, at recognition. Using a divided-attention paradigm, in Experiments 1A and 1B we tested and confirmed the hypothesis that distinct postencoding categorization and individuation processes occur during the recognition of same- and cross-race faces. Specifically, postencoding configural divided-attention tasks impaired recognition accuracy more for same-race than for cross-race faces; on the other hand, for White (but not Black) participants, postencoding featural divided-attention tasks impaired recognition accuracy more for cross-race than for same-race faces. A social categorization paradigm used in Experiments 2A and 2B tested the hypothesis that the postencoding in-group or out-group social orientation to faces affects categorization and individuation processes during the recognition of same-race and cross-race faces. Postencoding out-group orientation to faces resulted in categorization for White but not for Black participants. This was evidenced by White participants' impaired recognition accuracy for same-race but not for cross-race out-group faces. Postencoding in-group orientation to faces had no effect on recognition accuracy for either same-race or cross-race faces. The results of Experiments 2A and 2B suggest that this social orientation facilitates White but not Black participants' individuation and categorization processes at recognition. Models of recognition memory for same-race and cross-race faces need to account for processing differences that occur at both encoding and recognition.

It is well known that adults' face recognition is characterized by an 'other-race effect' (ORE; see Meissner & Brigham, 2001), but few studies have investigated this ORE during the development of the face processing system. Here we examined the role of experience with other-race faces during childhood by testing a group of 6- to 14-year-old Asian children adopted between 2 and 26 months in Caucasian families living in Western Europe, as well as a group of age-matched Caucasian children. The latter group showed a strong ORE in favour of own-race faces that was stable from 6 to 14 years of age. The adopted participants did not show a significant reversal of the ORE, unlike a recently reported study (Sangrigoli et al., 2005), but rather comparable results with Asian and Caucasian faces. Their pattern of performance was neither influenced by their age of adoption, nor by the amount of experience they accumulated during childhood with other-race faces. These results indicate that the balance of performance with Asian and Caucasian faces can be modulated, but not completely reversed, in children whose exposure to own- and other-race faces changes drastically during the period of maturation of the face recognition system, depending on the length of exposure to the new face race. Overall, experience appears to be crucial during childhood to shape the face recognition system towards the most predominant morphologies of faces present in the environment.

Many studies investigating first impressions from faces employ stimulus sets that comprise only White faces. It is argued that participants lack the necessary perceptual expertise to provide reliable trait evaluations when viewing faces from ethnicities that differ from their own. In combination with a reliance on White and WEIRD participants, this concern has contributed to the widespread use of White face stimuli in this literature. The present study sought to determine whether concerns about the use of so-called other-race faces are justified by assessing the test-retest reliability of trait judgments made about same- and other-race faces. In two experiments conducted on 400 British participants, we find that White British participants made reliable trait judgments about Black faces, and Black British participants made reliable trait judgments about White faces. It is important that future work be conducted to determine how widely these results generalize. Considering our findings, however, we suggest (a) that the default assumption in future first impressions research should be that participants-particularly those recruited from diverse communities-are able to form reliable first impressions of other-race faces and (b) that faces of color be included in stimulus sets whenever possible. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Face perception is critical to social interaction, yet people consistently recognize own-race faces more accurately than other-race faces, a robust phenomenon known as the other-race effect (ORE). Although facial variability affects recognition, how naturalistic variations modulate the other-race effect remains unclear. To address this, we compared recognition performance on two standardized memory tests that differ in facial variability: the Cambridge Face Memory Test (CFMT), which includes controlled variability (systematic changes in lighting, and pose), and the Variability Face Memory Test (VFMT), which incorporates high naturalistic variability (e.g., lighting, poses, expression, viewing distance, hairstyle, and appearance). Sixty East Asian young adults (18-25 years) with limited cross-race contact completed both tests using own-race (Asian) and other-race (White) face sets. Both tests assess between-person discrimination and within-person recognition of identities across variable images. Results revealed a significantly stronger ORE under high variability (VFMT), where recognition of other-race faces was substantially impaired relative to own-race faces. In contrast, the ORE was minimal under low variability (CFMT). These findings suggest that naturalistic facial variation disproportionately disrupts other-race face recognition, particularly in within-person recognition tasks. This supports theoretical accounts of compressed face-space representations for other-race faces, in which limited experience reduces tolerance for natural variability. Overall, these results highlight that real-world facial variability amplifies the ORE, emphasizing the role of perceptual experience with facial variability in cross-racial face recognition under natural conditions.

ABSTRACTAdults show impaired recognition of other-race compared to own-race faces. This other-race effect (ORE) is suggested to be the result of asymmetrical perceptual experience with own- and other-race faces during development. However, it is unclear whether the impact of experience on adults’ ORE differs across development, and whether experience during adulthood can exert similar effects as experience during development. To investigate these questions, we tested face recognition in White adults, East Asian (EA) adults born and raised in Canada, and EA adults who immigrated to Canada at different ages from infancy to adulthood. When recognizing upright faces, White adults and EA immigrants demonstrated a reliable ORE, whereas EA adults born in Canada showed no ORE. These effects were not present when recognizing inverted faces. Notably, age of arrival positively predicts the magnitude of the ORE. Our study highlights the influence of early experience on the ORE and suggests that the ORE appears relatively unmalleable during adolescence and adulthood.

Summary Two recognition memory experiments using photographs of black, white, and Japanese faces as stimuli were performed. In Experiment I, white Ss showed most accurate recognition memory for white faces, next most accurate recognition for black faces, and least accurate memory for Japanese faces. In Experiment II, both white and black Ss were tested with all three groups of faces. Again, it was found that whites did better on white faces than on black faces and did least well on Japanese faces. Blacks, in contrast, did best on black faces, next best on white faces, and also did least well on Japanese faces. This interaction of race of 5 with race of pictured face points to differential prior experience with various kinds of faces as the basis for these differences in memory performance. More generally, these findings support the applicability of the concept of schema to the processes by which faces are discriminated, processed, stored, and remembered.

Numerous studies have repeatedly demonstrated a face advantage, showing that faces are processed more efficiently and faster compared to other stimuli in our environment. This long-standing result has also been explored in terms of high- and low-level visual properties of faces, revealing it to be unlikely that the advantageous processing could be explained on the basis of low-level feature differences such as luminance, contrast, or spatial frequency. In this study, we explored saccadic programming in relation to the own-race bias, a phenomenon describing superior performance to recognise own-race faces compared to other-race faces. Using an anti-saccade paradigm, twenty Caucasian and twenty Chinese participants were presented with images of Western Caucasian and East Asian faces, all controlled for low-level visual features. Participants were given a cue instructing them to either saccade toward the face stimulus (pro-saccade) or away from the image (anti-saccade). We found that Chinese participants produced significantly higher anti-saccade error rates for Asian compared to other-race faces, while Caucasians revealed prolonged saccadic reaction times for correctly performed anti-saccades when presented with Caucasian but not other-race faces. The own-race bias was thus demonstrated in an anti-saccade task, suggesting an involuntary saccadic bias towards own race faces.

Contextual variation in automatic evaluative bias to racially-ambiguous faces

Past research has shown that infants raised in a racially homogenous environment demonstrate an advantage for processing own-race faces at the individual level in comparison to other-race faces (Lewkowicz, 2014; Mauer & Werker, 2014; Quinn, Lee, & Pascalis, 2018). This advantage for processing own-race faces develops across infancy and has been referred to as the own-race bias (e.g., Anzures, Quinn, Pascalis, Slater, Tanaka, & Lee, 2013). The processes that lead to development of the own-race bias are complex and not fully understood. Recent work suggests that by 10 months of age, differences in own- and other-race face processing may be related to attentional processes (Roth & Reynolds, 2022). The current study utilized event-related potentials (ERP) to examine neural correlates of infant attention during initial exposure to own- or other-race faces. Level of attentional engagement was also compared across early to late trials to determine whether infants demonstrated increased or decreased attention to a face with repeated exposure depending on race. Twenty-three infants were tested at 10 months of age. Only infants raised in a racially homogenous environment (based on parental report) were included in the dataset. Participants were familiarized with 20 repeated 1000 ms presentations of a single exemplar of either an own-race face (n = 13) or an other-race face (n = 10). Following familiarization, presentations of the familiar face were interspersed with presentations of novel faces until the infant was off task. EEG was recorded throughout testing using the EGI Geodesic EEG System (GES 400) 128-channel system with a 1000 Hz sampling rate. An offline bandpass filter was applied from 0.10 to 30.00 Hz. The analysis of the EEG focused on the Nc ERP component associated with level of attentional engagement in infancy (Reynolds & Richards, 2005). Mean amplitude of the Nc component was analyzed at midline-frontal electrodes from 345-650 ms following stimulus onset using a 200 ms prestimulus baseline period. Participants contributed a minimum of 8 artifact-free trials per condition to the ERP averages. Infants familiarized with an other-race face demonstrated greater amplitude Nc in comparison to infants familiarized with an own-race face (F(1,21) = 4.65, p = 0.037, ηp2 = .191). Interestingly, infants showed increased Nc amplitude to other-race faces following familiarization and decreased Nc amplitude to own-race faces following familiarization. These findings indicate that 10-month-old infants demonstrate greater attentional engagement during initial exposure to other-race faces than during initial exposure to own-race faces. This greater amplitude Nc could reflect a novelty effect for other-race faces for infants raised in racially homogenous environments. The increase in Nc amplitude following familiarization with an other-race face may be indicative of continued attentional engagement due to incomplete processing of the face. Given the potential for early developing perceptual biases to contribute to subsequent development of social biases and prejudice (Xiao et al., 2018, 2019), these findings provide important insight into potential mechanisms associated with the own-race bias in face processing in infancy.

It is well known that same-race faces are better recognized than other-race faces, a phenomenon often reported in the literature as the other-race face effect (e.g. for a review Meissner & Brigham, 2001). However, the theoretical explanation of this effect is still debated. It has been suggested that it could be the result of a reduced ability to encode configural information on other-race faces (Rhodes et al., 1989). Here we adapted a paradigm developed by Tanaka and Farah (1993), in which it was shown that facial features are recognized better when presented in whole faces than isolated. This whole/part advantage has been taken as evidence that faces are processed holistically, a type of configural processing (Maurer et al., 2002). Subjects performed a delayed matching task in which a face was first presented for 500 ms, followed by two simultaneously presented faces or two isolated face features. The task was to match one of the items to the first one, the distractor being different from the target by only one feature. We tested 20 Caucasian and 20 Asian subjects who demonstrated an other-race effect as measured in an old/new recognition task. The three-way interaction between race of subjects, face race and condition (whole faces or isolated features) was significant. Caucasian subjects showed a larger whole/part advantage for their own-race faces than for other-race faces, but Asian subjects showed an equally large whole/part advantage for both race of faces. These results suggest that we process same-race faces more holistically than other-race faces. This effect may be related to visual experience, since our Asian subjects had been living among Caucasian faces for about a year on average. Interestingly, there was no correlation between the magnitude of the other-race effect and the holistic processing for other-race faces, suggesting that the other-race effect may not be related to a differential holistic processing between same-race and other-race faces.

The other-race effect (ORE) is the disadvantage of recognizing faces of another race than one's own. While its prevalence is behaviorally well documented, the representational basis of ORE remains unclear. This study employs StyleGAN2, a deep learning technique for generating photorealistic images to uncover face representations and to investigate ORE's representational basis. To this end, we collected pairwise visual similarity ratings with same- and other-race faces across East Asian and White participants exhibiting robust levels of ORE. Leveraging the significant overlap in representational similarity between the GAN's latent space and perceptual representations in human participants, we designed an image reconstruction approach aiming to reveal internal face representations from behavioral similarity data. This methodology yielded hyper-realistic depictions of face percepts, with reconstruction accuracy well above chance, as well as an accuracy advantage for same-race over other-race reconstructions, which mirrored ORE in both populations. Further, a comparison of reconstructions across participant race revealed a novel age bias, with other-race face reconstructions appearing younger than their same-race counterpart. Thus, our work proposes a new approach to exploiting the utility of GANs in image reconstruction and provides new avenues in the study of ORE.

Qualitatively similar processing for own- and other-race faces: Evidence from efficiency and equivalent input noise

6- to 10-year-old children do not show race-based orienting biases to faces during an online attention capture task