Chief Science Correspondent
Science and fashion crossed paths last week, when a New Jersey mom successfully petitioned Lands' End to bridge the gender gap in science-themed apparel. Now, researchers from the University of North Carolina claim to have solved the mystery of society's fashion choices (whether or not they involve children's science novelty t-shirts).
The study, published by the PLOS ONE on July 17, seeks to determine an "empirical approach to fashionableness," using a theoretical approach known as the "Goldilocks Principle."
A term derived from the well-known children's classic, the Goldilocks Principle states that moderate values are preferable over extremes within a spectrum. The term has been applied to a wide range of fields including psychology, astronomy, and American politics.
While this conclusion might make logical sense to some participants, I believe the statistical analysis behind the claim may be critically flawed.
The team reported that moderately matching outfits, both for men and women, tend to be preferable to overly matching or clashing outfits. Participants were shown a variety of four-color color palate combinations, and then asked to choose the ones they found most aesthetically pleasing.
The study reports a significant quadratic trend for both men's and women's fashion. Essentially, they graphed the participant's responses on an x-y axis, plotting a palate's perceived fashionableness versus its degree of coordination/matching. They claim these points fall along a bell-shaped curve; in essence, a moderately coordinated outfit results in the highest level of approval, and the extremes result in the lowest level of approval.
"These data suggest a simple answer to the question 'what to wear?,'" the team writes. "Select a color combination that is neither completely uniform, nor completely different."
The team reported that the men's and women's quadratic trends carry R-squared values of 0.28 and 0.44, respectively. They call these trends significant; and technically they are correct, as their p-values were both sufficiently low (<0.001).
However, if you've taken a basic algebra course, you may remember that an R-squared value determines the strength of the trend, showing how closely the points match a line or, in this case, a curve. An R-squared value of 1.0 would be considered perfect (But also essentially impossible. A more practical example of a "perfect" R-squared value from a research setting might look something like 0.99975).
I would not expect the team to generate a set of data with an R-squared value of 0.99975 from a human survey trial, especially one with only 239 participants. In fact, if they had reported a value that high I would have probably been suspicious.
However, the values of 0.28 and 0.44 do not convince me that this model actually holds a high degree of merit. R-squared values can be arbitrary, as there is not necessarily a standard "cut-off" value among mathematicians. The analysis of R-square is simply much more complex than that. However, the values suggest that fewer than half of their data points actually conform to the trend that they claim as their result.
The low p-value may indicate that this model sufficiently matches these data points to the curve. However, it likely will not, despite what the paper claims, serve as a sufficient predictor for large-scale fashion preferences.
The researchers may have found a trend, but as of now it appears to be an extremely weak one. Without a larger data set, I'm not yet convinced.