Robust Effects of Signed Prediction Error on Declarative Memory

Humans and other animals continuously predict potential rewards that may follow their decisions. When their predictions are erroneous, they subsequently experience reward prediction errors (RPEs), which influence both procedural and declarative memory. It remains unclear why declarative memory sometimes increases with signed RPEs (SRPE) and sometimes with unsigned RPEs (URPE), posing a central question about whether our brain prioritizes “better-than-expected” or “different-than-expected” information. However, recent studies suggest that large negative RPEs are critical for eliciting URPE-related neural signatures. Moreover, some previous studies reporting the SRPE effect had systematically larger positive than negative RPEs, an asymmetry that can make an URPE effect look like an SRPE one. Given this, we implemented two improved experimental designs incorporating negatively biased or symmetric RPE manipulations. Across five experiments with a total of 379 participants, the SRPE effect persisted in Experiments 1-4; however, after overnight consolidation in Experiment 5, a shift toward a URPE effect emerged. This finding provides strong evidence for the robust SRPE effect, while also identifying overnight consolidation as a potential boundary condition for a shift between SRPE and URPE.