Is it? The actual discrimination can be quantified in a typical way. Attractiveness is only a bit trickier, but there's a high degree of agreement on attractive vs unattractive. Give a team of raters a group of people, present them pairs of people and ask them to select the more attractive one. Then you can create an average ordering, which you can map to whatever numerical range you want, and that score would be highly predictive of what future raters (and the real world) would find attractive.