Abstract
Beauty is described as one of the goals of humankind, alongside truth and goodness. Values are the basis for decision making. In decision theory, a rational decision maker's preferences will be transitive along the dimension of expected value. Is beauty a value? Are our beauty preferences transitive? If so, then beauty ratings must be one-dimensional. That is what we test here. Participants are presented with two images simultaneously, one to the left of the central fixation and the other to the right. Then they are asked to choose which one is more beautiful, and rate by how much on a scale from 1-9 (very little to very much). Participants are presented 6-14 images, which are randomly selected from a 45-image subset of the OASIS image database. Independent studies provide normative beauty and valence ratings for the OASIS images. We picked our 45 images to fall into three categories (low, medium, and high) according to beauty and valence ratings. In random order, participants are presented with all possible image pairings, and each pair is presented twice. The SD of repeat testing ranges from 1.4 to 2.6. We assessed how well a one-dimensional beauty model fits each participant's difference ratings. We find that the root mean square error (RMSE) of this one-dimensional model (1.3 to 2.0) is not significantly bigger than expected from the test-retest SD. Thus, we find that beauty is one-dimensional, at least for OASIS images.
Meeting abstract presented at VSS 2018