What you lookin’ at?

The eyes are the primary social signal. It’s the eyes we spend most of the time looking (“To See, Act” [Hack #15]). Even when the other person is talking, we look most at the eyes, not the mouth. We use them to signal turn-taking in conversation, to read emotions from, like fear…and we use them to work out what another person is looking at.

It’s this – gaze perception – that I’ve been getting interested in. How accurately can we tell where someone is looking? How accurately can we tell if someone is looking at us, or not? I’ve been looking out for some actual figures here, basic parameters on how small a difference we can detect in where someone is looking, either when they are looking at us, or at someone else.

Obviously, to be able to answer this question with actual parameters would have all sorts of implications. For, say, the design & manipulation of pictures showing people looking at things, for VR interfaces and, also, I guess it might give a better idea of when someone can tell i’m looking at them, and when they just can’t know I am for sure. You know, just as a sort of side benefit…

One of the great things about science, is that if you have a question you can usually be pretty sure that someone, somewhere, has asked the same question and done some experiments to find out the answers. If I’m able to be curious about something it’s pretty much a dead cert that someone else already did. So – I figure – time to dip into the back catalogue of psychological research.

And here it is. Argyle & Cook’s (1976) Gaze and Mutual Gaze. Notable not just for the fact that it’s a book about experimental psychology that quotes Sartre, but it also contains summaries of experiments from the sixties looking into just the questions i’m interested in. Here’s my summary of what I found:

First, some ball-park figures, from two experiments:

Imagine two people, 80 cm apart. The ‘sender’ looks at one of seven points near the ‘receivers’ face. The points are about 2.4 cm apart and two of them are the reciever’s eyes. The receiver’s job is to judge where the other person is looking. By chance you’d get this right about 14% of the time (1 in 7). Actually, people get it right about 35%. At 2 m people perform at chance. [full details in ref 2]. Moral: At normal conversational distances, people might be able to tell where on your face they are looking, but more likely they’ll get it wrong (so if you don’t want to look people in the eyes look at the bridge of their nose and unless they’re very close they won’t be able to tell that you’re not looking them in the eye.

Now imagine two people, one being looked at, one looking, at either 1.5 m or 3 m apart. A third person, next to the ‘looked at’ (let’s called them the ‘observer’) is trying to tell where the looker is looking. It’s a bit like sitting on a train next to your friend and trying to work out who someone a few rows in front is looking at. I won’t go into the details of the points the looker had to focus on, but they were either on the person, or up to about two persons widths either side of them (so imagine five people sitting in a row). At 3 meters the observer only correctly judged the lookers gaze about 50% of the time, which isn’t much above chance. And most of these judgements were found to be based on the direction of the lookers head, not the direction their eyes were pointing. At 1.5 m the observer was right for about 65% of the times the looker’s gaze was straight at the looked at. [full details in ref 3]

So maybe it seems like you can pretty much get away with looking at whoever you want at around 3 m, and they’ll be none the wiser. But, before jumping to conclusions, some other results of the experiments

Eye directed gaze is overestimated, for someone looking at the face. In other words, if you look at someone’s face they are likely to assume you are looking at their eyes.

For gaze targetted near the face, face directed gaze is overestimated. (In other words, if someone is looking roughly at you, you are likely to assume they are looking directly at you).

Head angle biases judgement of gaze direction. We tend to assume that people are looking mostly the way their head points. Even if we can see the eyes, head angle affects where we think they are looking. Gibson & Pick (1963) found that a 30 degree turn of the head shifted judgements of where someone was looking by an average of 3 degrees in that direction

As distance between looker and lookee increases, head angle become more and more influential on the judgement of the looker’s gaze

Horizontal discrimination is better than vertical discrimination. We’re more accurate judging where someone is looking left-to-right than up-to-down.

So, it seems, we can’t very accurately detect gaze (at least at distance beyond the normal conversational), but we have biases to assume that if it’s possible, someone probably is looking at us.

If you’ve got this far, I hope you’ve enjoyed the journey along the back-shores of psychology research (before electronic abstract indexing!). I couldn’t find any information on depth-of-gaze perception (i.e. can we tell at what distance from their face someone is focusing?) so if anyone knows any leads on this send them along.

1. Argyle, M., & Cook, M. (1976). Gaze and Mutual Gaze. New York: Cambridge University Press.

2. Kruger & Huckstedt (1969), cited in Argyle & Cook (1976).

3. Von Cranach and Ellgring (1973), cited in Argyle & Cook (1976).

4. Gibson, J.J. & Pick, A.D. (1963). Perceptions of another person’s looking behavior. American Journal of Psychology, 76, 386-94