People are notoriously awful at recognizing faces from other races. It's a human foible often explained by the notion that we have more experience looking at members of our own race and thus acquire "perceptual expertise" for characteristics of our own kind.
One influential version of that hypothesis argues that the so-called cross-race recognition deficit can be modeled by assuming that faces of other races are more psychologically similar than are faces of one's own race. But Daniel Levin, PhD, a cognitive psychologist at Kent State University, has been unsatisfied with that argument.
"The perceptual expertise position is pretty intuitive, and it makes sense," he says. "But I'm arguing that it's not really the case. The problem is not that we can't code the details of cross-race faces--it's that we don't."
Instead, he says, people place inordinate emphasis on race categories--whether someone is white, black or Asian--ignoring information that would help them recognize people as individuals. In recent research, Levin has shown that people can, in fact, perceive fine differences among faces of people from other races--as long as they're using those differences to make race classifications.
For example, Levin explains, "When a white person looks at another white person's nose, they're likely to think to themselves, 'That's John's nose.' When they look at a black person's nose, they're likely to think, 'That's a black nose.'"
The results are important, Levin maintains, because they help explain the long-standing question of why people are poor at recognizing the faces of people who belong to other racial groups. Such an understanding could be useful in a variety of settings, including training police and others in the justice system to identify faces more accurately.
The work, published in this month's Journal of Experimental Psychology: General (Vol. 129, No. 4), proposes an explanation "that's at a different level," says Tom Busey, PhD, an Indiana University psychologist who studies face recognition. Although Busey doesn't believe Levin's findings warrant completely discarding perceptual expertise models, he suggests the work can be viewed as helping explain how the cross-race recognition deficit is created.
Levin hypothesized that when people see cross-race faces, they code race-specifying information at the expense of individuating information--something they don't do when they see same-race faces.
Looking for such an asymmetry, in his first experiment, Levin compared how well people recognize faces of other races with how readily they locate these faces in a visual search task. He first made two morphed "average" faces, one derived from 16 black faces and one from 16 white faces. Next, the individual features--eyes, nose and mouth--of each average were inserted within a morphed average of all 32 faces. This method provided Levin with one black and one white face that differed only in the internal facial features that specified their races.
Using these faces, Levin tested 25 participants' ability to locate a black face amid a sea of white faces, and vice versa. For half of the experimental trials, the participants--most of whom were white, but a few of whom were Asian--were instructed to press one key if a black face appeared among a number of white faces, and to press a second key if no black face appeared. In the remaining trials, the target face was white. Across the two search conditions, half of the trials contained a target face, and half did not.
Next, participants completed a recognition test. They were shown yearbook photos of 16 white and 16 black males. Then they were shown another set of photos and asked to indicate whether they recognized any of the faces from the previous set.
As Levin expected, on the recognition test, participants were better at recognizing white faces than they were at recognizing black faces. Paradoxically, he found that participants who performed most poorly on the recognition task were most likely, in the search task, to locate a black face among other white faces more quickly than they identified a white face among black faces.
The results indicate, says Levin, that "participants who are poor at recognizing black faces appear to code 'blackness' as a visual feature, while they may not code 'whiteness' at all."
That finding sets the stage for Levin's further hypothesizing that race is a specific visual feature that people use in perceiving faces, says MaryAnn Baenninger, PhD, a face recognition and spatial ability researcher at the College of New Jersey. "There's some salient characteristic or set of characteristics that are associated with the cross-race effect, and he shows that pretty convincingly here," she says.
Perceiving subtle variations
To test the notion that people are able to perceive subtle differences among faces of people from other races, Levin next explored how readily people distinguish among cross-race faces versus own-race faces in making race classifications. Using the two average black and white faces from the earlier experiments, he created a continuum of faces that ran from black at one end to white at the other. Thirteen participants viewed pairs of faces that differed by 20 percent along the black-white continuum. For half the trials, participants judged which of the two faces was most similar to the face at the black end point face. For the other half, they judged which was most similar to the face at the white end.
He found that participants were more often accurate when discriminating between two faces at the black end of the continuum than they were for faces at the white end of the continuum. That finding demonstrates, Levin explains, that people possess the perceptual expertise to detect minute differences among cross-race faces.
A final experiment corroborated those results. As before, for faces on a black-white continuum, participants were better at discriminating between subtly different black faces than they were for subtly different white faces. But on a different continuum that had black faces at both end points, making it impossible for faces to be distinguished based on race, participants did not show such skill at discriminating between faces. That suggests that the extent to which the subtle variations convey race information, as opposed to individuating information, is an important part of the discrimination task, Levin argues.
The final experiment also showed that participants who were poorest at recognizing black faces on the recognition task made the most accurate race-based discriminations between black faces on the discrimination task.
Together, the findings are bad news for the similarity hypothesis, Levin believes.
"If you're arguing that the reason people are rotten at recognizing cross-race faces is because their brains can't code them accurately, then you should never observe a situation where people who are rotten at recognizing cross-race faces are paradoxically accurate at discriminating among subtle variants of those faces," he says. Instead, he observes, people's perceptual skills do extend to cross-race faces. Levin's findings are congruent with a few studies that have taught people to recognize cross-race faces more effectively, even with only brief training, he says. That's encouraging because it provides further evidence that the cross-race recognition deficit is not caused by a dearth of perceptual experience.
"If you really had to reorganize your whole face recognition system and build up a lifetime's worth of experience in order to recognize cross-race faces effectively, you'd think it would take a long time," Levin argues. "But it really doesn't seem to."
Dissent and synthesis
Brian Mullen, PhD, a Syracuse University social psychologist who several years ago published data supporting similar ideas for thinking about the cross-race recognition deficit, argues that Levin's analysis disregards a central finding in his own research. Black research participants do have difficulty in recognizing cross-race faces, Mullen says, but not as much as white participants do.
"These patterns are not always universal and symmetrical," he says. "It is an all-too-common mistake for theorizing and research on intergroup perceptions to be based solely on the responses of majority participants to minority targets."
Tim Valentine, PhD, of Goldsmiths College, University of London, also challenges Levin's interpretation. In order for Levin to support his claim that people more quickly classify other-race faces according to their race than they classify own-race faces, he says, "it is necessary to show that an effect for one race of participants reverses for the other race--for example, that black participants classify white faces faster than black faces. Levin has never shown this crossover that is critical for his hypothesis."
Levin disagrees, however, that showing such a reversal is critical. His argument, he emphasizes, depends only on having found that people who are poorest at recognizing cross-race faces are in fact best at discriminating between them on the basis of race.
And Levin concurs with Mullen that members of minority groups are likely to respond differently than are members of majority groups. Indeed, he points out, his report discusses previous research that suggests that minority group members tend to code not only people of other races at the category level, but also do so for people of their own race.
Ultimately, suggests Alice O'Toole, PhD, a psychologist at the University of Texas at Dallas who also studies face recognition, Levin's new findings may be compatible with perceptual expertise and similarity hypotheses.
"I see less division in the ideas than he does," O'Toole says. "One consequence of the perceptual problems that we may have with other-race faces could simply be that race is just a much more salient aspect of our encoding of faces of other races than it is of faces of our own race. I think the hypotheses are compatible, but Levin's idea is at more of a social level of analysis."
Levin acknowledges, "The problem with the [perceptual expertise] models is not really that they're wrong, per se. Rather, it's a problem of focus. They're focused on this sort of reductivist analysis of similarity, when they really ought to be focused on trying to figure out why people use the features they use."
This article is part of the Monitor's "Science Watch" series, which reports news from APA's journals.