February 15, 2013

Overselling research findings

The Herald has a story claiming that facial proportions indicate racism (in men).  Well, they have a headline claiming that. The story (and the research paper, even more explicitly) pretty much contradicts the headline, and says that facial proportions have nothing to do with racism but indicate whether men write magazine articles about express their racist views or hide them.

If you believe the story, the relationship is very strong

Looking at the photos from the first study, a new group of participants evaluated men with wider, shorter faces as more prejudiced, and they were able to accurately estimate the target’s self-reported prejudicial beliefs just by looking at an image of his face.

and to be fair to the journalist, that’s what the researchers said.  If you look at their actual results, it’s not what they found.

They found an average difference of 1.92 on a 6-point perceived-racism scale for men who differ by 1 unit on the facial proportion scale.  The full range of the facial proportion scale appears to only be about 0.7 units. The paper doesn’t tell us the actual distribution of the measurements, but according to another research paper I found on the internets, the standard deviation of this facial proportion scale is about 0.12.  That means two randomly chosen men would differ by about 0.17 units, and the relationship  would predict a difference in the 6-point perceived-racism scale of about 0.3 units.  The association with self-reported racism was about as strong, though I haven’t been able to find enough information to compute the predicted differences (it shouldn’t be this hard).

In my book, that’s not an “accurate estimate”.

 

 

avatar

Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »