The Gaze Experiment: A study of gazing habits and biases with respect to different facial features.
AbstractCan we tell which part of our face a person is gazing at when they are gazing at us? The experiment in question gives us an idea to where and why we think a person is looking at that location. There were 27 students split into pairs. Within the pair there was an experimenter and a participant. The experimenter would look at different parts of the participant's face and ask the participant to tell them where they were looking at. Our results showed us that the participants were most accurate at recognising when the eyes, hair, right and left ears were being looked at with correct responses ranging between 46.3%-51.85%. And least accurate when the nose, mouth and chin were being looked at with correct responses ranging between 22.59%-25.19%. Percentage incorrect responses was highest for eyes at 30.1%. The eyes, hair and ears were the easiest to be recognised because those features are the furthest away from other facial parts. The eyes have the highest incorrect response percentage because most people assume it is their eyes people are looking at during social interaction so if the participant was unsure where the experimenter was looking at, the guess would have likely been the eyes.

IntroductionNon-verbal communication between humans was first scientifically investigated in 1872 where Charles Darwin wrote “The Expression of Emotions in Man and Animals”. Since then many scientists have investigated the topic. There are many different non-verbal channels of communication. Michael Argyle (1988, 1994) suggested that facial expressions, gaze, voice, gesture, posture amongst others were channels of non-verbal communication.
One of the most widely studied channels are gazes. A similar experiment to ours was performed in 1963 by Gibson & Pick. They tested whether a participant could distinguish whether the experimenter was gazing at their eyes. They tested this by looking at the eyes of the participant and just outside their face and asking them to determine whether they were being stared in the eyes. They had another independent variable that we do not use in our experiment, the experimenter would stare at seven points on a horizontal line 10cm apart with the centre point being on the nose of the participant. They found that the participant could correctly distinguish whether their eye were being stared at 84% of the time. However when the experimenter wasn’t facing the participant head on, they found their success rate drop. The experiment showed that both head direction and pupil direction are important in determining where a person is gazing.
After Gibson & Pick, Von Cranach & Ellgring (1973) also performed a similar experiment to ours. They increased the number of variables that would affect a person's judgement about where they were being stared at, eg. head direction, gaze duration, number of points on the face. The testing of different points on the face is similar to our experiment. They found that the participants made the most errors perceiving that their eyes were being looked at when they were not. This is understandable as the eyes are where we would assume people who look at us are looking at. It daily human interaction, when peoples' gazes are fixed upon our faces we assume they are looking at our eyes.
As a result of Von Cranach & Ellgring's experiment we can predict that we would find that the eyes will have the highest number of incorrect responses, representing the fact that people assume they are being looked at in the eye because it is a social norm. From the combination of the two above experiments we can also predict that facial extremities like the chin, ears and hair will have higher correct responses as opposed to the nose and mouth. Gibson & Pick showed that when looking slightly outside the face, past the ears or hair or chin, we can tell that the experimenter is not looking at our eyes. And with Von Cranach & Ellgring's experiment we can assume that points within the face (nose and mouth) will be harder to for the participant to recognise.
MethodParticipants
There were 27 participants, five male and 22 female aged between 18-20 years old. The participants were drawn from students studying Introduction to Psychological Research at Durham University. Seven of the participants wear glasses, four wear contacts lenses and the remainder have uncorrected vision.
Apparatus
Each experimenter was given a gaze record sheet. The experimenter would had to note the gender and any visual aids the participant may have had. The gaze record sheet had a grid with T (target) and R (response) heading the columns. The rows would have different targets the experimenter had to look at and were filled by letters, each letter donated different target: E= eyes, N= nose, M= mouth, C=chin, H=hair, RE=right ear, LE=left ear. The response column would then be filled with the letter of which facial feature the participant was looking at.
Design
The experiment was a within-subjects design. The independent variable was the seven different facial targets the experimenter would look at. The two dependent variables are the percentage of correct responses and the percentage of incorrect responses or response bias of each facial target. The experimenter would random where they chose to start on the column of targets to get rid of fatigue effect and practice effect.
Procedure
The subjects will pair up, one taking the role of experimenter and the other participant. Any visual aids should be worn to give the subjects their best vision. The pair should sit facing each other 1.5m away from each other. The experimenter should place the record sheet in front of them and cover it so the participant cannot see it and fill out the participant's gender and visual aids (if any). The experimenter will then choose a random point on the column to start.
Once the experimenter has chosen, he/she should look to the ceiling and then lower their gaze to rest upon the facial target that the experimenter has randomly chose to start with. The participant should know to responds within five seconds, and if at the end of the five seconds the participant is still unsure he/she should guess. The participant will then say where they think the experimenter is looking at. The experimenter will then note down where the participant said in the response column next to the target the experimenter was looking at on the gaze record sheet using the letters that donate the facial features mentioned earlier. Continue with each target on the sheet until all have been completed. No feedback should be given during the trial. The experiment should be repeated with the participant and experimenter roles reversed.
ResultsWe totalled up the number of correct and incorrect responses for each facial part. In table 1 we take the total number of correct responses for each facial part and divide it by 270 (the number of people (27) multiplied by the number of times the target was presented (10)) and multiply it all by 100 to obtain a percentage. The percentage figures show us what percentage of the time do the participants provide a correct answer to when the facial part is gazed at. We can see that the hair, ears and eyes have the highest percentage of correct responses and the nose, mouth and chin have the lowest.
Table 1. Percentage of Correct Responses Percentage Correct Response | H | RE | E | LE | N | M | C | 51.85 | 51.11 | 46.30 | 48.89 | 24.07 | 25.19 | 22.59 |

Table 2 is slightly different, it represents the percentage of incorrect responses expressed in percentage of errors. We can see that the eyes, nose and mouth have relatively higher incorrect response rates. The ears, hair and chin have extremely low incorrect response rates relatively. The figures were calculated by dividing the total number of incorrect responses for each facial target by the total number of incorrect responses over all participants and targets, multiplied by 100.
Table 2. Percentage of Incorrect Responses Percentage Incorrect Response | H | RE | E | LE | N | M | C | 7.15 | 4.48 | 30.1 | 5.51 | 27.7 | 18 | 7.1 |

The diagram show us the mean response of correct answers for each facial part. We can see that most participants managed to recognise that their hair, ears and eyes were being looked at half the time. Whereas the bars for the nose, mouth and chin are about the half the height of the other facial parts showing us that participants were less sure of when those facial parts were being looked at, averaging slightly above two correct guesses each.

The next graph shows us a different pattern. The graph above clearly shows two distinct groups that the participants could identify at different competencies. The graph below shows us again two groups that the participants could identify at different competencies. The difference now is that the eyes, nose and mouth fall into one group and the ears, chin and hair fall into a separate group. The eyes, nose and mouth were identified incorrectly 2-3 times more than the ears, chin and hair facial parts.

DiscussionFrom our correct responses results we can infer that we can distinguish between the top facial features better than the bottom half of our face. This could be down to the fact that we can recognize horizontal eye movement better than vertical, or simply because there are fewer facial targets on a horizontal axis than the vertical therefore giving us a higher chance to guessing a correct target when we recognise a horizontal eye movement. There is another explanation why the results show a higher correct response rate for the upper half of our face. The facial targets on the upper half of our face are also further apart than the bottom half of our face which might explain why there is a higher correct response rate.
The incorrect responses results tells a different story. The facial extremities (hair, chin and ears) incorrect responses are 2-3 times lower than that of the facial targets within the face (nose, eyes and mouth). This relation could be explained by the fact that we stare people in the eyes by nature during social conversations so maybe we use it as a reference point. With the eyes, nose and mouth being a close group, the small pupil movements of the observer may not be noticed. Also these three facial targets are within the face where we may recognise being stared at with no difference between the targets. The facial targets on the extremities of the face boarder on being off the face and maybe that is why we can tell better. We can tell when someone is gazing just off our face pretty well from Gibson & Pick's research and this supports ours.
The target with the highest incorrect responses was the eyes. This is understandable as we would guess that it was the eyes that was being looked at when someone is looking at our facial area because it is a social norm to be looking at the eyes. This finding supports Von Cranach & Ellgring's experiment that eyes will have the highest incorrect responses. The only difference from our predictions is that we would assume the chin would have a higher correct response rate being on the edge of our face. I think this could be that the chin, mouth and nose are closely spaced and the chin unlike the other facial targets on the edge of the face has two targets between it- nose and mouth.
Our experiment was not entirely accurate. We didn’t measure exactly 1.5m between the experimenter and participant. Also if one of the pair was significantly shorter/taller than the other, the distance between the faces will have increased. On top of the increased distance we would have been adding an angled view of the other's face which would influence the result as shown in Gibson & Picks's results. There were a lot of targets that had to be viewed in a row, this meant that both/either or experimenter and participant would move their heads, shift positions in their seats, etc.
If we were to perform this experiment again, to control the extraneous variables of head movement, body movement, height and distance difference we can use a two head frames 1.5m apart where both the experimenter and participant will rest their heads. This will control the extraneous variables in our experiment. However by controlling these variables that occur naturally in real life social interactions we may actually be making the results less reliable. In real life people shift and move constantly when looking at each other and taking this away could make our experiment less ecologically valid.
ReferencesArgyle, M. (1994). The Psychology of Interpersonal Behaviour.

Gibson & Pick (1963). Perception of another person's looking behaviour. American Journal of Psychology, 76, 386-394.
Langton, S.R., Watt, R.J., & Bruce, V. (2000). Do the eyes have it? Cues to the direction of social attention. Trends in Cognitive Sciences, 4 (2), 50- 59.
Von Cranach, M., & Ellgring, J.H. (1973). In M. Von Cranach & I Vine (Eds.). Social communication and movement. London: Academic Press.