A novel approach for modelling subsequent memory reports by separating decidedness, recognition and confidence

In episodic memory studies, memory performance is often assessed by presenting participants with previously seen old items and previously unseen new items in order to assess both, subjects' hit rates and false alarms. Sometimes, subsequent memory responses are collected as recognition-confidence ratings, e.g. on a scale ranging from "definitely new" over "undecided" to "definitely old". Despite higher informational value, this is problematic, as it conflates recognition (i.e. one rather than the other end of the scale) with confidence (i.e. outer rather than inner area of the scale). Here, we describe a computational model which separates three cognitive processes of episodic memory retrieval from reported recognition-confidence ratings, namely (i) decidedness, i.e. the tendency to give a neutral or non-neutral response; (ii) recognition, i.e. the ability to label previously seen items as "old"; and (iii) confidence, i.e. the act of labeling an item as "definitely" rather than "probably" old or new. We empirically validate our model by recovering memory-related differences between young and older adults (N = 106 + 153 = 259), especially the over-confidence phenomenon in older adults as well as superior memory performance and meta-cognitive accuracy for young adults. Interestingly, we find that decidedness does not vary between old vs. new items and young vs. older subjects, thus constituting a third process distinct from recognition and confidence.