Converting AUCs to d'

[mtrevin5]

Hi,
I have been trying to convert the AUCs generated from the MRMC variance analyses to d' values using d'=√2z(AUC). However, those d's aren't equivalent to d′s calculated from d'=z(H)−z(FA). I had assumed they would be the same but would you have any insights as to why they aren't?

Thank you,
Melissa

[brandon-gallas]

Hi Melissa,
Your notation is pretty sparse, but my first reaction is that the relationship only holds if the scores are continuous, Gaussian, and have the same variance across the diseased and non-diseased groups.

I hope that helps.
Brandon