I ran an experiment in which people are presented with a list of target items, and then get tested on their memory for those items. For the test, they are presented with the target items, but also with an identical number of non-target items.
Their responses to the non-target items are called "correct rejections" when the participant said the item was not a target, and they are called "false alarms" when the participant responded it was a target.
I used a glmer model in which I coded correct rejections as 1 and false alarms at 0, before learning that the convention is to actually report and plot the false alarms. To analyse false alarms (or the probability of a response being a false alarm) I need to code them as 1, but I already have the coefficients and everything from the models I run with them being coded as 0.
Would it be ok if I use the same results but just change the signs? And if not, why?
