I have calculated D-prime measurements for for a memory performance task. Subjects viewed images that were either old or new and had to indicate their response via a button press. Thus, their response was either correct or wrong. From the subjects responses, I calculated their Hit rates (correct/total old) and FA rates (incorrect/total new) and calculated the d-prime values (z(H)-z(FA)) in R.

Now, when I look at my d-prime scores, they seem to be normally distributed around zero:

Normally distributed D-prime

From what I've gathered from the literature, this should not be the case. D-prime values should be above zero and negative values could represent sampling errors or response confusion e.g. Stanislaw & Todorov (1999) https://www.researchgate.net/file.PostFileLoader.html?id=536ce28fd5a3f2b41b8b46c5&assetKey=AS%3A273596423835649%401442241878760

or in this interpretation: http://www.csic.cornell.edu/201/signal_detection/#negative_d_prime.

Now, I'm wondering if I can assume that in some cases my subjects were just confused about which button represented which response? For example, those samples around -3 really seem to indicate a large sensitivity to the signal, however in complete reverse. Could I use absolute values of these extreme negative d-prime samples to still include them in my sample and further interpretation?

EDIT: I had another look at the data, it seems like this effect does not stem from the errors subjects have made. The max error percentage was 60%, with an average of 42%. Perhaps it's the way I calculated the z-scores? I used the scale function in R, but I don't see why that would be a problem?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.