Dear all,
thank you very much for the help.
Concerning my last e-mail about negative d' values, questions were made about the fact that my identification task has 3 labels for the subjects to select when a vowel sound drawn from a continuum is presented (vowels with F1 varying from 250 to 650 Hz; the labels are "u", "o" and "O"("O" is similar to the "o" in "off").
I think d' is appropriate since there is almost no overlap between the categories in the extremes of the continuum - "u" and "O". I have imagined two solutions:
1. take only the d's obtained from the probabilities of responding with the label corresponding to the category in the middle of the continuum - "o".
2. simply split the results in two at an arbitrary point in the middle of the '"o" category' (where we have a region of "100%" (AND THAT IS ANOTHER QUESTION TO WITCH I HAVE SOME ALTERNATIVES BUT NO CLEAR CRITERION TO CHOOSE ONE! - up to now the 0,01-0,99 range is O.K. for my purposes). That sounds a little arbitrary but I dont think that it is a great evil...but there is still a question to be made: is it better either to discard the extremely rare cases of overlap between "u" and "O" or calculate d' keeping the data as they are?
At the end, the differences between the alternatives are very small.
To be honest, really arbitrary is my choice cause I was not able to decide whats the greater of two evils.
What do you think?
Daniel.