Appalling research

I stumbled upon a summary of an article from The Lancet dating from 1998 entitled "Optimum body-mass index and maximum sexual attractiveness" (you can download the article itself in pdf here). The research purports to demonstrate that the Body Mass Index is of greater importance to the sexual attractiveness of the female body than the waist to hip ratio. At first I was incredulous that someone would actually write such an article, but reading over the summary and article itself only made me more disgusted with this research.

The authors had 40 male undergraduates rate the attractiveness of 50 female bodies (faces were not shown) on a scale of 1 to 7 and estimated a polynomial regression equation on the results. BMI enters the equation as a main effect, squared, and cubed to account for its predicted nonlinear effect. The summary above reproduces their figure demonstrating the fit of their predicted values to means of the actual ratings.

I can't believe this kind of thing gets published. First, they use ordinary least squares regression on ordered, non-zero data. They should be using a generalized linear model like ordered probit, otherwise their coefficients will produce nonsensical outcomes for certain values. Second, the pattern is clearly more complicated than the polynomial function they've specified. I would have at least suggested a fractional polynomial, if not a spline model. This type of sloppy work is highly offensive.

Fortunately, I found a response to the article clearly illustrating several of the faults of this research. Further studies in this area can be done much more cheaply due to subsequent research discovering that chickens and humans have very similar judgments of human attractiveness (see the article, "Chickens Prefer Beautiful Humans" here). This research deservedly won the Ig Nobel prize for Interdisiplinary Research in 2003.