Abstract
Suppose that you are told that two children — whom you have never seen — are very similar in appearance. Do you now imagine that they are more likely to be siblings? More likely to be close in age? More likely to be the same gender?
What objective information about age, gender, and genetic relatedness is conveyed by a rating of similarity?
We asked observers to rate the facial similarity of pairs of children depicted in photographs. Observers were not instructed to consider the ages, genders, or possible relatedness of the children: they were told only to rate facial similarity. We estimated the Shannon information (“bits”) about relatedness, age difference, and gender difference contained in these ratings of facial similarity.
The stimuli were 48 pairs of color photographs of children (3 to 12 years of age) from two provinces of northern Italy. Each photograph depicted the child from the shoulders to the top of the head with his or her face clearly visible. Half of the 96 children were male, and half female. In half of the pairs, the children were siblings, in the other half, the children were not related. Age difference, gender difference, and relatedness were counterbalanced across the pairs. Thirty observers rated the similarity of the 48 pairs on a scale of 0 to 10.
The maximum possible Shannon information that can be conveyed concerning a binary choice is 1 bit. The estimated Shannon information conveyed by a similarity rating was 0.171 bits (relatedness), 0.091 (gender difference), and 0.015 (age difference). The first two values are significantly greater than 0 (p < 0.0001 for each), while the third is not (p > 0.05). We conclude that the observers incorporate information about relatedness and gender difference into their ratings, but include little information relevant to age difference. In particular, the observers discounted the large physical changes in appearance that occur as children grow in forming a judgment of facial similarity.
Supported by NIH/NEI grant EY08266 and HFSP grant RG0109/1999-B.