Posted: May 16, 2012 9:05 pm
There's also a consideration that with the small number of participants, potentially small number of assignments to the "gay face" category that taking the mean proportion A'f (the A' considering only those female faces that were guessed to be gay) will lead to excessive susceptibility to noise. I suppose it's possible that a weighted mean was used (which would appear to be the statistically valid thing to do) but they don't state that's what they did.
ETA: Direct link to the paper: http://www.plosone.org/article/info%3Ad ... ne.0036671
ETA: Direct link to the paper: http://www.plosone.org/article/info%3Ad ... ne.0036671