The apparent attraction of average faces

By Gordon Rugg

In a previous article, we looked at what happens when you take two concepts that are normally viewed as opposites, and instead treat them as two separate concepts. We used the example of what happens when you treat liking and disliking as two separate concepts, and ask people to rate items in relation to both liking and disliking.

The result is that people are willing and able to do so. The image below shows types of response that we’ve seen in real data.

Item A has been rated low both for liking and for disliking; it’s just boring, with little to be said for or against it.

Item B has been rated high both for liking and for disliking; it produces strong but ambivalent feelings. An example that we saw involved university departmental websites, where some were strongly liked because they signalled high quality, but simultaneously strongly disliked because that same signal of high quality was viewed as implying unforgivingly high expectations.

Item C has been rated high like/low dislike by some participants, and low like/high dislike by others. This is informally known in the UK as the Marmite effect, where people either love something or hate it, with few people in between.

This approach of uncoupling apparent opposites is well established in some fields, but isn’t yet widely known outside them. We’ve been using it for a while in software evaluation, where it’s invaluable for improving software mockups before committing to the final design. We’ve also blogged about ways of using it to represent expressive and instrumental behaviour; handedness; and gender roles, going back to the literature where we first encountered it, in Bem’s work on androgyny (Bem, 1974).

The advantages of using this approach are clear when you see examples. In the next section, we’ll look at the background theory on which it works. We’ll then apply it to an apparently paradoxical finding about facial attractiveness, to show how the underlying issues can be swiftly and easily teased apart via this representation.

Background theory

In terms of underlying theory, what we’re doing in these cases is asking whether a pair of concepts are not the end points of a scale running from a negative to a positive value. Instead, we’re asking whether they might be better treated as two separate concepts, with each given its own scale.

Two situations where this approach is useful are when:

  • We’re dealing with portmanteau terms
  • We’re dealing with privatives

Portmanteau terms in this context are terms which lump together several different concepts, and treat them all as part of the same “bag” (hence the name portmanteau, a type of bag). The word “good” is often used as a portmanteau which subsumes assorted different attributes. Often, those attributes point in the same direction (for example, cheap and fast and reliable are usually all desirable qualities for a product or service). Sometimes, though, one or more of the terms within a portmanteau will point in the opposite direction to the others, so a product may be perceived as being good overall, but bad with regard to some of its attributes. In this context,using a pair of scales allows us to tease out the attitudes across a batch of several attributes.

A privative term in this context is a single attribute which fits on a scale which is anchored at one end with an absence, as opposed to a negative. For example, a reservoir and a bank balance can both have a positive value for their contents (litres of water in the reservoir, or money in a bank balance). However, the concept of a reservoir having a negative amount of water in it is meaningless, since it can’t go beyond empty, whereas the concept of a bank balance having a negative value is meaningful, since you can have an overdraft.

This distinction often causes problems for people trying to learn subjects such as physics. A classic example is that in physics, hot and cold are not treated as opposites; instead, cold is a low value for heat. Absolute zero in physics is the complete absence of heat; the concept of a value below this is by definition meaningless.

We’ll draw on both the concept of portmanteau terms and the concept of privatives in the examples and discussions that follow.

Aesthetics and threat

In the next example, we’ll apply this approach to beauty/ugliness, and show how it is a simple way of explaining an apparent paradox.

First, some background. There’s been a lot of research into perceptions of attractiveness in photographically averaged faces.

Photographically averaged faces are created by superimposing multiple photos of different people to create a single composite image, as in the example below. (I’ll resist the temptation to go into detail about issues involved in using categories such as “Mexican” as group identifiers.) This approach was originally developed by Galton in the 1870s, and has been extensively used since then, partly because it lends itself well to formal experimental designs, and partly because the development of cheap computing power made it easy to use in terms of logistics (hence the number of key papers on the topic that were published in the 1990s). Used under academic fair use terms

Research within this approach has consistently found that averaged faces are viewed as attractive, and that individual faces similar to the average are perceived as attractive (e.g. Komori et al, 2009(a); Komori et al, 2009(b); Trujillo et al, 2014).

However, research using this approach has also found that the averaged faces tend to be viewed as more attractive than any of the individual faces which went into the averaged photo. (Originally reported by Galton, and replicated by e.g. Langlois et al, 1994, and Rhodes et al, 1999).

This doesn’t initially appear to make sense, because by definition averages should not be higher than all of the component scores that go into the average, regardless of whether we’re talking about mean averages, medians or modes. Enquist et al (2002) give a good overview of the literature and of this apparent paradox.

Similarly, an article title by Alley & Cunningham (1991) neatly sums up another classic apparent paradox of averaged faces: “Averaged faces are attractive, but very attractive faces are not average”.

So what’s going on here?

There has been considerable discussion of this in the literature, typically in relation to evolutionary factors and in relation to cognitive factors, as in the articles cited above. Much of this is now well understood within the facial attractiveness research field, but not widely known outside it. I’ll return to both these themes in another article.

The figure below shows how these apparent paradoxes can be neatly explained via simply representing beauty and ugliness as separate concepts. The figure shows scores for three original photos and for an averaged photo from the three originals. For clarity and simplicity, it uses fictitious data to illustrate the underlying concepts. It also uses beauty and ugliness as the concepts involved, rather than attractiveness, for reasons that will become apparent in the discussion below.



Original photos 1, 2 and 3 (in stars) all have higher scores for beauty than does the averaged photo (A, in a circle).

However, the original photos all have medium values for ugliness. These will typically be due to minor facial blemishes.

In the averaged photo, those blemishes will be lost as a result of the averaging process, so the averaged photo will have a much lower score for ugliness than the individual photos.

In the image above, the results are easy to see: The averaged photo has a lower rating for beauty than the original photos, but it also has a lower value for ugliness.

So, if these results are compressed into a single scale, where beauty and ugliness are crunched into a single value for attractiveness, then the higher scores for ugliness in the original photos will drag down the higher scores for beauty in the original photos. The final outcome will depend on the values and the way of calculating the tradeoff between beauty and ugliness, but it’s perfectly possible to end up with a situation where an averaged photo comes out on a single scale as being more attractive overall than any of the original photos.


So, representing beauty and ugliness as two separate factors can make sense of an apparent paradox about perceptions of attractiveness. This has numerous practical implications, in fields ranging from reconstructive plastic surgery to market research.

This particular finding is well established in the literature on perceived facial attractiveness, but isn’t so well known elsewhere.

In the bigger picture, this raises broader questions, such as which other variables need to be included if we’re looking at attractiveness, and why there’s a lot of research into attractiveness, particularly attractiveness in women, but comparatively little research into ugliness.

There’s also the even bigger question of which other fields are treating core concepts as opposites when they might better be treated as separate.

The next article in this series will look at these questions.

Notes, references and links

You’re welcome to use Hyde & Rugg copyleft images for any non-commercial purpose, including lectures, provided that you state that they’re copyleft Hyde & Rugg.

There’s more about the theory behind this article in my latest book: Blind Spot, by Gordon Rugg with Joseph D’Agnese

You might also find our website useful:

Related articles:

Overviews of the articles on this blog:


Alley, T.R. & Cunningham, M.R. (1991), Averaged Faces Are Attractive, but Very Attractive Faces Are Not Average. Psychological Science, 2(2), pp 123-125.

Bem, S. L. (1974), The measurement of psychological androgyny. Journal of consulting and clinical psychology, 42, pp 155-162.

Enquist, M., Ghirlanda, S., Lundqvist, D. & Wachtmeister, C-A. (2002). An ethological theory of attractiveness. In Rhodes, G. & Zebrowitz, L. (eds) (2002), Advances in Visual Cognition, Vol. 1: Facial Attractiveness, Ablex, Westport, CT.

Komori, M., Kawamura, S. & Ishihara, S. (2009a). Effect of averageness and sexual dimorphism on the judgment of facial attractiveness. Vision Research, 49, pp 862-869.

Komori, M., Kawamura, S. & Ishihara, S. (2009b). Averageness or symmetry: Which is more important for facial attractiveness? Acta Psychologica, 131(2), pp 136-142.

Langlois, J. H., Roggman, L. A., & Musselman, L. (1994). What is average and what is not average about attractive faces? Psychological Science, 5, pp 214-220.

Rhodes, G., Sumich, A. & Byatt, G. (1999). Are average facial configurations attractive only because of their symmetry? Psychological Science, 10, pp 52-58.

Trujillo, L.T., Jankowitsch, J.M., & Langlois, J.H. (2014). Beauty is in the ease of the beholding: A neurophysiological test of the averageness theory of facial attractiveness. Cognitive, Affective, & Behavioral Neuroscience, 14, 3, pp 1061-1076.


1 thought on “The apparent attraction of average faces

  1. Pingback: Beauty, novelty and threat | hyde and rugg

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.