I would like to use d'prime as a standardized measure to compare the outcome of different psychophysical studies. Some studies use 'same-different' paradigms (i.e., are these 2 stimuli same or different?), for which d' has been calculated simply as: d' = z(H) - z(F).
However, other studies are simple identification tasks in which participants must memorise a label for a novel stimulus (e.g., A, B, C, or D) and then label that stimiulus correctly at test (each stimulus is presented independently, aka 'one interval'). For these experiments, I cannot use d' = z(H) - z(F) because the studies do not provide a Hit Rate or False Alarm - only the raw % correct (i.e., ppts identified the images correctly for 92% of trials). Is there a straight-forward way to calculate d'prime for these studies that provides a fair comparison to same-different studies?
Many thanks.
Ryan
Even for same-different paradigms, you can't calculate d' based on a paper that only reports accuracy; if a paper says people were 90% correct, you don't know how much of that 90% correct was hits and how much was correct nos, and you don't know how much of the 10% incorrect was false alarms and how much was misses.
For a four-alternative identification design, I'm not sure if there is any equivalent to d'. There's not a clear "miss" case in such an example (what counts as a hit, miss, false alarm, etc., depends on which of the four categories you are looking at). There may be other signal detection measures available but I'm not aware of them.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics