Disentangling storage from retrieval in associative long-term memory : aging- and working memory-related differences and measurement issues

Preview

PDF
DissStreitberger_postdisputation.pdf - Published
Download (3MB)

URN:	urn:nbn:de:bsz:180-madoc-725236
Document Type:	Doctoral dissertation
Year of publication:	2026
Place of publication:	Mannheim
University:	Universität Mannheim
Evaluator:	Erdfelder, Edgar
Publication language:	English
Institution:	School of Social Sciences > Kognitive Psychologie mit Schwerp. Kognitives Altern (Kuhlmann 2015-)
Subject:	150 Psychology
Keywords (English):	associative long-term memory , storage , retrieval , multinomial modeling
Abstract:	For an association to be accurately recalled from long-term memory, it must be successfully stored and retrieved (e.g., Shiffrin & Atkinson, 1969). If it cannot be recalled, it may be due to a storage failure and/or a retrieval failure. Because we can only observe successful or unsuccessful recall/recognition, and not the underlying latent processes, multinomial modeling has been developed to disentangle them based on categorical data (Erdfelder et al., 2009; Riefer & Batchelder, 1988). Two multinomial models, along with their corresponding materials and paradigms, were consistently used throughout this dissertation. The free-then-cued-recall model (Rouder & Batchelder, 1998) is based on a paradigm in which weakly associated pair words are presented together during study, followed by a free and then a cued recall test. On the other hand, the pair-clustering model (Batchelder & Riefer, 1980) is based on a paradigm where pair words with preexisting semantic associations are presented separately during study, followed by a free recall only. A relevant application area for separating storage from retrieval in associative memory using these two models is aging. Older adults show a disproportionately bigger impairment in associative than in item memory (Naveh-Benjamin, 2000; Old & Naveh-Benjamin, 2008), but it is unclear whether this deficit primarily relies on storage (Chalfonte & Johnson, 1996) or retrieval problems (Burke & Light, 1981). In Manuscript 1 (Streitberger & Kuhlmann, 2026a), we therefore investigated the locus of this aging-related associative deficit and found that older adults had difficulties with both storage and retrieval of associations. This, however, only applied to weak associations (i.e., free-then-cued-recall paradigm; Rouder & Batchelder, 1998), that is, when participants could not rely on existing semantic knowledge for support. In turn, storage, and possibly also retrieval, for preexisting strong associations (i.e., pair-clustering paradigm; Batchelder & Riefer, 1980) did not decrease with age. A plausible explanation is that semantic knowledge increases with age, and older adults rely on it for compensation. To more generally understand why groups or individuals might systematically differ in storage or retrieval processing, we examined potential mechanisms. A range of studies show that working memory capacity predicts long-term memory performance, regardless of paradigm or analysis type (for a review, see Unsworth, 2019). However, it is rather unclear whether working memory benefits long-term memory due to more effective storage or retrieval - a question that we tackled in Manuscript 2 (Streitberger et al., 2024). Focusing on younger adults only, we replicated the correlation between working and long-term memory, but found this relationship to be differentially driven by storage and retrieval depending on the type of encoding. When associative encoding was inherent to the paradigm (i.e., free-then-cued-recall paradigm; Rouder & Batchelder, 1998), participants with higher working memory capacity showed more effective storage, but not more effective retrieval. Furthermore, when associative encoding was not inherent to the paradigm (i.e., pair-clustering paradigm; Batchelder & Riefer, 1980), working memory correlated with storage and retrieval only when associative encoding was instructed and not otherwise. For the two different application areas, both of our studies showed that the results depended on the material, paradigm, and the multinomial model. As detailed in the manuscripts, this may very well be due to material-specific (e.g., semantic relatedness; Bein et al., 2015) or paradigm-specific (e.g., simultaneous versus successive presentation; Moscovitch, 1992) reasons. Precisely because such paradigm-specific factors influence the results, it is psychometrically informative to examine whether the storage and retrieval parameters converge across models. We therefore investigated the parameters’ construct validity in Manuscript 3 (Streitberger & Kuhlmann, 2026b). As a precondition, we needed estimates of reliability of the individual parameters and found evidence of test-retest reliability for almost all of them, with the exception of the retrieval parameter of the pair-clustering model. We then estimated construct validity of storage across models and of retrieval across models, and found that they, in fact, represent different processes. The paradigms, therefore, seem to elicit different processes in associative memory, reflected in the models’ parameters. It is, therefore, not surprising that the effects of aging (Manuscript 1; Streitberger & Kuhlmann, 2026a) and working memory (Manuscript 2; Streitberger et al., 2024) on storage and retrieval depend on the material, paradigm, and model used.