We got some pretty weird results in a memory recognition task. Subjects are asked to indicate if the word presented has been learned (old) or not (new). And we have 4 different conditions during previous words encoding period. And all 4 conditions have been randomized for each subject during both encoding and retrieval period. The weird thing is, 4 out of 7 subjects got negative d'prime in one or two conditions. For example, one subject get d'prime of 0.69/0.66/0.69/-0.15 in four separate conditions, and another get -0.12/0.28/0.58/0.31. I am really confused about what does the negative d'prime mean. And do I need to exclude those subjects who have negative d'prime?