The Effect of Negative Valence on False Memory Formation in the Deese–Roediger–McDermott Paradigm: A Preregistered Meta-Analysis and Preregistered Replication

Participants in the Deese–Roediger–McDermott (DRM) paradigm learn lists of words (e.g., bed, tired) associated with a nonpresented lure (i.e., sleep). In subsequent memory tests, individuals tend to report the nonlearned lures, that is, exhibiting false memories. Priorly, the DRM task has been criticized for not capturing the aversive nature of (clinically and forensically relevant) real-life memories. To obtain a robust estimate of the influence of negative versus neutral word lists on the DRM effect, we conducted both a preregistered meta-analysis (k_recall = 49, n_recall = 2,209, k_recognition = 75, n_recognition = 3,008, k_responsebias = 31, n_responsebias = 1,128) and replication (n_final = 278) predicting increased false memories for negative valence in recall and recognition. For recall, we found significant frequentist evidence in the meta-analysis for a reversed valence effect (d = −0.18, i.e., reduced false memories for negative content vs. neutral), whereas the replication displayed null results (d = 0.03). For recognition, both the meta-analysis (d = 0.23) and replication (d = 0.35) showed that negative valence (vs. neutral) increased false memories. However, this effect may be confounded by shifts in response tendencies as controlling for response bias nullified the valence effect in our meta-analysis (dmeta = 0.05), and we found evidence for differential response bias in our replication (dreplica = 0.39). Hence, the effect of valence on false memory reports in the DRM may not represent a systematic difference in emotional information but instead depend on how memory is tested, and be partly attributable to differential response tendencies.