Re: identification task...negative d' values.

[Pierre Divenyi <pdivenyi@xxxxxxxxx>]

Dan,

I grew up on Green's and Swets' signal detection book, as introduced 
by the phenomenal teacher Jim Egan was. I remember an episode that 
made me turn my back to negative d' presentations of any data. I had 
just stumbled upon an interesting phenomenon: a systematic 
underestimation of an unfilled time interval marked by two sine 
bursts frequencies f1 and f2, when compared to an interval marked by 
bursts of f1 only. While the results were robust when reached using 
the method of adjustments, I felt as being on soft ground as long as 
I did not have a discriminability measure as well. So, I ran a 2AFC 
experiment (without feedback, to allow the subjects to respond to 
what they heard rather than to what they were told to hear) and 
obtained a d' psychometric function extending to values below zero, 
as a function of time difference. When I presented those data at an 
Acoustical Society meeting, Dave Green asked what I really meant by 
negative d' , commenting that d' was a measure of differential 
sensitivity which, whether positive or negative, marked 
discriminability, and that a negative d' only meant that the subject 
mislabeled the stimuli.
My interpretation of Daniel's results is similar: his subjects 
identified the stimuli consistently incorrectly, meaning that they 
could hear the difference between them but put them in the wrong bins.

Pierre