Gender Recognition: How We Perceive Inverted Faces

Abstract
Gender perception is a natural human ability that has been researched numerous times (Bruce et al., 1993). The purpose of this study was to determine if participants could determine the gender of inverted faces with some crucial facial information missing. It was hypothesized that participants would be more accurate and confident when determining the gender of a female face as opposed to a male face; an upper face as opposed to a lower face; and a whole face as opposed to either part of the face. Participants were 28 Chapman University students between the ages of 18 and 30. Each participant was shown a series of 48 randomized photographs of inverted faces shown as a whole face, upper face, or lower face on a computer screen online, then asked to determine gender of the face and rate their confidence level of their choice. We found that our hypothesis was mostly supported by the fact that participants were more accurate and confident with whole faces than upper or lower faces. Refuting the hypothesis, however, participants were more accurate and confident with male faces than female faces. Perhaps this is because people rely on external features such as hair and clothing when determining gender, as well as on whole face information than partial face information. Perceiving gender is something that humans do automatically and we cannot stop, even if we try to. When we are unable to determine the gender of another person, it radically changes our behavior. Gender perception is a critical piece of how we interact with others.

Gender Perceptions
The ability to quickly and accurately recognize gender is a natural human quality that most people do not think much about-until it's no longer easy. Many researchers have created studies based around how people perceive gender. Sometimes, the familiarity of the face is what is being measured. Other times, it is where a person looks at a face to determine gender. Or like our study, how accurately one can determine gender of an inverted face.
In a study researching how average people determine the gender of ambiguous faces by Bruce, Burton, Hanna, Healey, Mason, et al. (1993), it was hypothesized that if less information is given by the pictures, then guesses will be less accurate. The participants were shown a face and asked to determine the gender of it. Each of the pictures of the faces was altered in combinations by different levels. There was a natural format and a laser format, a 'normal' size head or a 'scaled' standard size. As far as viewpoints, there was a ¾ view, and also a profile view. Along with all of that, there were four levels of a mask: no mask, or a small rectangular box concealing the nose, eyes, or chin. The researchers measured participants' accuracy and response time. It was found that participants were generally more accurate when shown a 'natural' picture format instead of a laser picture format. When shown a picture in laser format with a ¾ view, people did quite well with accurately determining the gender of the face in the picture by using 3-dimensional information such as contours of the face and certain shaded areas. It was also found that participants were more accurate when more information was available, i.e. no mask on the face. These authors' hypothesis and experiment design prompted our study, where faces will be shown to our participants that are sometimes covered, and sometimes not (Bruce et al., 1993).

Another study about perceptions of gender by Campbell, Benson, Wallace, Doesbergh, and Coleman (1999) focuses mainly on how participants rely on the look of eyebrows to determine gender. They hypothesized that if the distance between eyebrow and eyelid is small, then participants will guess the face is male. Conversely, if the eyebrow to eyelid distance is larger, then participants will guess female. The variables of this study were the distance between the eyebrow and the eyelid, the gender of the face shown, and the position of the brows on the faces shown to participants. The participants were shown faces with eyes straight ahead, with eyebrows raised, or with eyebrows lowered, then asked to determine the gender of each face. Reaction time and accuracy were measured. The researchers found their hypotheses to be correct. Participants determined models with short eyebrow to eyelid distance to be male, and models with larger eyebrow to eyelid distance to be female. This ties into our experiment since eyebrows are sometimes covered in the faces shown to our participants, and eyebrows seem to be an important factor in determining gender.

A study conducted by Tanaka and Farah (1993) is very connected to our study in the sense that some of the faces shown to their participants are inverted, which was done in our experiment. Their study is also concerned with whole faces as opposed to parts of faces. It was hypothesized that if faces were inverted, then participants would rely on the whole face to determine gender. Also, if the faces were not inverted, then participants would rely on pieces of the face to determine gender. The variables were an inverted face or an upright face, and whole or partial information given. The accuracy of the participants was measured. The researchers found that people were more accurate when determining the gender of whole faces, as opposed to parts of faces (Tanaka & Farah, 1993).

An additional study, performed by McKone and Peh (2006), also used inverted and non-inverted faces. In one experiment, 30 inverted faces and 30 non-inverted faces were shown to participants. The dependent variable was the accuracy of the participants when determining the face as 'familiar' or 'unfamiliar'. It was hypothesized that if a face was upright, then participants will determine that they have seen it before. Participants determined that they had seen many of the upright faces before, and not many of the inverted faces, and their hypothesis was supported (McKone & Peh, 2006). This study may be used to our advantage in the sense that people generally have a harder time recognizing inverted faces than upright faces.

Metzger (2006) conducted an experiment testing the ability to recall distinctive and typical faces at intervals of 3, 6, and 12 weeks. It was hypothesized that if a distinctive face was shown, then performance and confidence will be higher in the earlier trials of the experiment, but as time wears on, performance and confidence for a distinctive face will begin to become more similar to that of typical faces. This experiment is relevant to our study, as they both are concerned with confidence level as well as performance. The variables that were changed were the faces shown to participants and whether they were distinctive or typical, and also the amount of time that each participant was tested on their recall: immediately, after 3 weeks, after 6 weeks, or after 12 weeks. One of the variables being measured was whether the participant regarded the face shown at their recall time as old or new. The other variable being measured was how confident the participant considered their decision (from 'guessing' to 'very confident'). Although the hypothesis was supported, participants' memory for distinctive faces deteriorated to that of typical faces after a longer amount of time, participants' confidence level for judgment of distinctive faces remained higher than that of typical faces, even after 12 weeks (Metzger, 2006).

Also using the familiar versus unfamiliar face method, Stacey, Walker, and Underwood (2005) created an experiment around showing faces to participants: either familiar (famous/recognizable) or unfamiliar faces. They tested to see where people look when gauging whether or not a face is familiar-the internal region or the external region of the face. It was operationally defined within the article that the "internal region of a face" includes the eyes, nose, and mouth, and the "external region of a face" includes every other part of the face. It was hypothesized that if a face was more familiar, then more time would be spent looking at the internal region, as opposed to an unfamiliar face. It was found that people's eye movements fell on the internal region of the face about 91% of the time. This is very connected to our study in that it shows how heavily people rely on eyes, nose and mouth when looking at a person's face. This study could possibly be telling us that our participants will be more accurate on the whole face pictures, because people generally look at a combination of eyes, nose and mouth.

It is hypothesized with our study, that if the face shown is a female, then it is more likely that the participant will be more accurate and confident. Also, if the whole face is shown as opposed to just upper or just lower face, there will be more correct responses and more confidence. It is further predicted that participants will be more accurate and more confident when the upper face is shown instead of the lower face.
Method
Participants

All participants were college students and psychology majors from Chapman. 20 participants were female and 8 were male. Each participant was between the ages of 18 and 30 years old, and each gave informed consent to participate in the experiment.
Research Design

One independent variable was the gender of the model. The other independent variable was the amount of information's shown: whole face, upper face, or lower face. Accuracy and confidence was measured.
Materials

Black and white photographs were taken of 8 Caucasian male models and 8 Caucasian female models at each of the levels of information given. For the whole face condition, hair and neck were covered with a black drape. For upper face condition, hair and the face from the nose down was covered with a black drape. For the lower face condition, neck, hair, and the face from the nose up was covered with a black drape. A total of 48 photographs were taken. Eye color was not available in any of the photos and neither was any other gender-identifying information such as makeup or facial hair. The experiment was conducted online and presented to participants on Dell computers with 14" monitors in a quiet room. Participants used a computer mouse to make their responses.
Procedure

The experimenter gathered 3-4 participants at a time and led them to a separate room with computers. Each participant sat at a computer and listened to instructions given verbally by the experimenter. The experiment was conducted completely online. Participants read the informed consent, and clicked "I Agree" with their mouse to indicate that they agreed with the terms in the informed consent. Participants were then prompted by the computer to type in their first name and press 'enter' to continue. Each was then given an individual user ID number and then asked by the experimenter to write that number down on a blank sheet of paper. Participants then pressed the 'enter' key on the keyboard to continue. A map was shown and students were asked to click on the region in which their school was located.

They were then prompted to select their university from a drop-down menu. After this, participants were asked identify their gender and age. After this, each was given instructions on the computer screen. The experiment then began. The online experiment displayed to each participant a randomized series of the 48 photographs taken prior to the experiment, with each of the faces inverted. Participants were asked to determine if each face was a male or a female, and click their desired choice on the computer screen. After making their selection, they were shown a scale that began at "Least Confident" and ended at "Most Confident" with an arrow positioned in the center of the scale. They had to click and drag the arrow to a position on the scale that corresponded with how confident they were about their choice. After asserting their confidence level, each participant was shown another face in the series and asked to do the same procedure over again for all 48 photos.

At the conclusion of the experiment, participants were thanked for their participation, and then asked if they would like to send their data. A unique quality of this experiment is the fact that all data is archived online where only the experimenter has access to it. After clicking "Send Data," participants were shown a box of raw data, and then debriefed. Each participant was then shown a box of his or her own correct choices in the experiment, divided up by information given (full view, eyes only, and mouth only), and then further by gender. Each participant was thanked by the experimenter and left.

Results
We examined accuracy and confidence levels of gender perception across male and female inverted targets for full face, upper face, and lower face by multiple analyses of variance (MANOVA). It was hypothesized that would be more accurate and confident when determining the gender of a female face than a male face; and more accurate and confident when shown a whole face as opposed to parts of a face. Please see Table 1 for the means and standard deviations for accuracy and Table 2 for the means and standard deviations for confidence ratings.

The main effect of amount of information for accuracy was significant, F (2, 27) = 36.89, p < .001. Participants were more accurate for full faces (.95) compared to upper faces (.91) and lower faces (.78). This means more information supported more accuracy. The main effect of amount of information for confidence was significant, F (2, 27) = 66.92, p < .001. Participants were more accurate for full face (78.1) compared to upper face (66.23) and lower faces (60.38). This means more information supported more confidence. There were no main effects for gender for accuracy or confidence. Accuracy also depended upon the interaction between gender and level of facial information, F (2,27) = 2.94, p = .53 (see Figure 1). Participants were more accurate for male whole faces (.97) and female whole faces (.93) than for male upper faces (.93) or female upper faces (.88). Participants were least accurate for lower faces when the target was male (.75) relative to lower faces when the target was female (.80). Participants were more accurate with male whole and upper faces than we were for female whole and upper faces. However, participants were more accurate with female lower faces than male lower faces.
Discussion

The purpose of this study was to try to understand how people are so god at gender perception and which part of the face we look at to determine this. Gender perception studies are widespread, however, few incorporate inversion as well as different levels of information. It was predicted that if the face shown is a female, then it is more likely that the participant will be accurate and more confident. It is also predicted that if the whole face is shown, there will be more correct responses and more confidence; also more confidence and accuracy when the upper face is shown in comparison to the lower face being shown. All in all, we found that participants were, as hypothesized, more accurate and more confident when determining the gender of full faces than for that of lower faces or upper faces. We also found that participants were more accurate and confident with upper faces as opposed to lower faces. Refuting part of the hypothesis, our data showed that participants were more accurate with male upper and whole faces than female upper and whole faces. These results show that more information is definitely a large predictor in gender perception, and that more information is gleaned from the upper face than the lower face. It also means that it is generally easier to determine the gender of a male than a female when the face is inverted.

Our results echo the results shown by Bruce, et al. and by Tanaka & Farah, both in which participants were more accurate in determining gender of faces when more of the face was shown. Both of their experiments dealt with showing the whole face and covering parts of the face as ours did. All of our results show that people definitely rely on more information to determine the gender of faces. The supported hypothesis from Campbell, et al. was also apparent in our results in which participants were more accurate when eyebrow information was given than when mouth-only information was given. Participants in our study were vastly more accurate with determining the gender of upper faces than lower faces. Perhaps eyebrows are as important as Campbell, et al. hypothesized.
The fact that our participants were more accurate when the whole face was shown than when only parts of the face were shown ties in to Stacey, Walker, & Underwood's supported hypothesis that people generally rely on all parts of a person's face when determining familiarity of a face. Even though the faces shown to our participants were inverted, they still relied more on whole face information than partial face information. Our results were not too connected to the studies performed by McKone & Peh or Metzger, as these studies were more focused on remembering a face than gender identification.

As with any experiment, there were some flaws apparent to the researchers after the data was collected. If this experiment were to be executed again, the confidence scale should have numeric values, instead of just being a bar that participants click and drag to their desired confidence level. Having the numeric values operationally defines confidence a little better than a blank scale. An equal number of male and female participants would also make the data more accurate than having 8 male and 20 female, as we did. There was also the obvious flaw of the data sometimes not sending when participants clicked 'send' at the end of the experiment. Some participants were also uncomfortable with the black drape surrounding the faces of the models and this is something that can be changed in future research.

Further research is definitely suggested in this field. Having time as a dependent variable would be an interesting twist to the experiment, showing how long it took each participant to make a decision on gender perception. Also, having the models be of more diverse backgrounds would be a worthy change to the experiment, seeing as how not everyone we encounter in day-to-day life is of Caucasian background.

Studying how people determine gender is a very important subject. It is something we do from infants, almost automatically. We change our behavior depending on whether a person is of same gender or opposite gender than us. Our gender perception is a very important part of how we interact with other people, and our interactions can be extremely challenged when we cannot accurately determine the gender of a person.

References
Bruce, V., Burton, A.M., Hanna, E., Healey, P., Mason, O., et al. (1993). Sex discrimination:
how do we tell the difference between male and female faces? Perception, 22. 131-152.
Campbell, R., Benson, P.J., Wallace, S.B., Doesbergh, S., & Coleman, M. (1999). More

about brows: How poses that change brow position affect perceptions of gender.
Perception, 28. 489-504.
McKone, E., & Peh, Y.X. (2006). Memory conjunction errors for realistic faces are consistent

with configural processing. Psychonomic Bulletin & Review, 13(1), 106-112.
Metzger, M.M. (2006). Face distinctiveness and delayed testing: Differential effects on

performance and confidence. The Journal of General Psychology, 133(2), 209-217.
Stacey, P.C., Walker, S., & Underwood, J.D.M. (2005). Face processing and familiarity:

Evidence from eye-movement data. British Journal of Psychology, 96(4), 407-423.
Tanaka, J.W., & Farah, M.J. (1993). Parts and wholes in face recognition. The

Quarterly Journal of Experimental Psychology, 46A (2). 225-245.