Marek Nieznański
Affiliation: Cardinal Stefan Wyszynski University in Warsaw
Category: Psychology
Keywords: source memory, memory distortions, judgment biases, memory models
Date: Thursday 4th of September
Time: 15:30
Location: GSSR Plenary Hall (268)
View the full session: Memory
Deep memory distortions have been defined by Brainerd (2021) as a family of memory biases that emerge by considering multiple episodic memories for a set of individual items. These are relations among memory judgments that violate basic properties of reality as expressed in the rules and axioms of classical logic or probability theory. Following the research of Brainerd and his collaborators on this topic, which has been developed for more than a decade (e.g., Brainerd et al., 2010, 2014; Brainerd & Reyna, 2008; Nakamura & Brainerd, 2017), our first aim was to investigate, in parallel, the disjunction fallacy and the conjunction illusion, which underlie one of the main deep distortions in source memory, namely the overdistribution. The second aim was to introduce conjunction probes into the Conjoint Process-Dissociation (CPD) model (Brainerd et al., 2012), and test validity of different variants of this multinomial model. Additionally, we considered the question of whether overdistribution is a real memory effect, as postulated by Brainerd (2021), or whether it is due to a specific response bias, as suggested by Kellen et al. (2014). In this endeavor, we conducted three experiments in which several independent variables expected to affect overdistribution in source memory were manipulated. Episodic overdistribution can be defined as a phenomenon in which an item that was encountered during the study phase of a memory task in only one source (context) is attributed during the memory test to that source and to others as well, depending on which source is specified in the test instruction. As a result, some single-presented items are treated as if they came from multiple sources (Brainerd & Reyna, 2008). This phenomenon was interpreted from the perspective of dual-process models of memory, which postulate that memory judgments are based on recollection and familiarity. According to Brainerd and Reyna (2008, Brainerd et al., 2010), when familiarity succeeds but recollection fails, an item’s familiarity is attributed indiscriminately to plausible sources provided at memory test, this means that episodic classifications are “description-dependent”. Due to this familiarity-without-recollection, human memory judgments exhibit disjunction fallacy; that is, the summed probabilities of accepting mutually exclusive real events exceeds the probability of accepting the disjunction of those same events; for two events A and B, P(A) + P(B) > P(A or B). The relation between memory judgments is subadditive when it should be additive (Brainerd, Reyna, & Aydin, 2010). From the axiom of additivity in classical probability theory, we can conclude that whenever P(A) and P(B) are subadditive with respect to P(A or B), the probability of conjunction of these events, P(A and B), must exceed zero, in particular, probabilities should be additive – so that: P(A) + P(B) = P(A or B) – P (A and B). A convenient experimental procedure used in studies on disjunction fallacies in episodic memory is a conjoint recognition (CR) paradigm (Brainerd et al., 1999). In this paradigm, for source memory tasks, three types of test items (Source A items, Source B items, and unrelated distractors) are factorially combined with three types of recognition test instructions (probes), two of them nondisjunctive: “Was it presented in Source A?”, “Was it presented in Source B?” and one disjunctive: “Was it presented in Source A or Source B?”. In our research, we extended this paradigm by adding conjunctive probes: “Was it presented in Source A and Source B?”. By this extension, in addition to disjunction fallacy, we were able to simultaneously assess the conjunction illusion, which occurs whenever people accept that items occurred in more than one source though they actually never did. We conducted three experiments, which tested the influence of several manipulations on overdistribution in source memory judgments. In Experiments 1 and 2, we manipulated subjects’ expectations concerning possibility of sources overlap. We expected that such manipulation should influence at least guessing on conjunctive probes, however, such an effect was not confirmed. In Experiment 2, we tested the effect of the normative word frequency. We found higher level of familiarity-without-recollection for high-frequency words than low-frequency words. Finally, in Experiment 3, we compared source memory for concrete and abstract words and we observed higher response bias in disjunctive probes for abstract than concrete words. In all experiments, we found significant conjunction illusions in every condition used, similarly, disjunction fallacies were also present for most conditions, except for concrete words in Experiment 3. In each experiment, we also tested the CPD model for conjoint recognition paradigm with added conjunctive probes. We tested the original version of this model (Brainerd et al., 2012) assuming one-high threshold (1HT) in item recognition as well as a two-high threshold (2HT) version proposed by Kellen et al. (2014), and a new proposition of an extended model which includes an additional parameter representing guessing for familiar but unrecollected items. In all three experiments, we confirmed good fit of theses models to data. We also found advantage of the 2HT CPD model over the 1HT version of this model in terms of better compromise between model fit and complexity. The extended model proved the best information criteria values, however, it also encountered important problems in parameter specification. Due to these problems, the question on the role of guessing for recognized items was left open. Our research confirmed that conjunction illusion is significant and robust in source memory judgments and it is usually accompanied by disjunction fallacy. However, when subjects’ memory judgments for nondisjunctive, disjunctive, and conjunctive probes are analyzed jointly, they obey the additive rule of classical probability under the assumption that sources of items can overlap, which was in fact not true for any test items but could be assumed by the subjects.