Learning from mistakes: Objective and subjective error magnitudes predict the testing effect

Abstract The testing effect—where retrieval practice enhances memory more than restudy—remains mechanistically debated. We advance an error-driven learning (EDL) account, proposing that testing enhances memory by generating error signals that drive learning. Across two experiments using a color-wheel paradigm, we quantified retrieval errors on a continuous scale and examined three types of error signals: objective errors (OE), defined as the angular discrepancy between the correct color and the participant’s response; predicted errors (PredE), defined as participants’ prefeedback estimate of how inaccurate their response was; and perceived errors (PercE), defined as participants’ postfeedback estimate of the discrepancy between their response and the correct answer. Experiment 1 showed that larger OE during practice predicted greater memory updating. Experiment 2 further revealed that PredE and PercE uniquely predicted memory updating beyond OE. To formalize these dynamics, we developed a computational model (WRAP-E) that simulates distinct memory-strengthening roles for OE, PredE, and PercE across different stages of retrieval. Our findings highlight errors as not merely byproducts of failed recall, but as potent drivers of learning, both through their magnitude and metacognitive interpretation. This framework offers a unified account of the testing effect and carries implications for theories of memory, metacognition, and educational practice.