A REM Model of Collaborative Recognition Memory

Recognition memory is often studied as an individual process, yet many recognition decisions are made collaboratively. Collaborative recognition requires individuals to communicate and combine separate, imperfect memory evidence to reach a joint decision. Here, we develop a computational model of collaborative recognition memory (CO-REM) grounded in the well-established Retrieving Effectively from Memory (REM) model of individual recognition. We compare collaborative decision rules based on categorical old/new decisions, confidence judgments, and the direct integration of memory evidence. The simulations show that groups can outperform even their best individual member, but the benefits depend on differences in individual recognition ability and how memory evidence is communicated and combined. Confidence sharing can produce collective gains or losses depending on how individuals map their memory evidence onto confidence judgments. The model also predicts greater collaborative benefits for recognizing previously studied items (targets) than for correctly rejecting new items (lures), consistent with some previous empirical findings. Although the predicted performance differences are small, CO-REM provides a framework for identifying the conditions under which collaboration improves recognition and for guiding future empirical investigations.