Understanding the multi-store model

STM uses predominantly acoustic coding — information is represented in a sound-based format, even when input is visual. This is why we make errors on acoustically similar letters (B, C, D, G) in short-term recall tasks. Conrad (1964) demonstrated this by showing participants were more likely to confuse letters that sound similar than letters that look similar. LTM uses predominantly semantic coding — information is represented in terms of its meaning rather than its sound. Baddeley (1966) confirmed this: participants who had recalled a list made far more semantic confusions in long-term recall than in short-term recall. This difference in coding format was one of the key pieces of evidence that STM and LTM are genuinely distinct stores rather than one continuous memory system.

Several converging lines of evidence supported the model: - **Serial position effect**: the primacy effect (better recall of early list items) reflects transfer to LTM through rehearsal; the recency effect (better recall of later items) reflects items still held in STM. Glanzer and Cunitz (1966) showed that an unfilled delay before recall abolished the recency effect but left the primacy effect intact — consistent with STM having decayed and LTM being unaffected. - **H.M. case study**: a patient (Henry Molaison) with bilateral hippocampal removal had intact STM but severely impaired new LTM formation (Scoville & Milner, 1957), suggesting functionally independent stores. The KF case (Shallice & Warrington, 1970) showed the opposite dissociation — severely impaired STM with intact LTM — from a left-hemisphere lesion unrelated to the hippocampus, providing further evidence that STM and LTM are separate systems. - **Peterson and Peterson (1959)**: STM duration of ~15–30 seconds without rehearsal, consistent with a time-limited temporary store. - **Conrad (1964) and Baddeley (1966)**: different coding formats in STM (acoustic) and LTM (semantic) imply separate systems.

Several findings challenged the model: - **Oversimplification of STM**: the model treats STM as a single, passive store. Baddeley and Hitch (1974) showed it is better described as a multicomponent working memory system that actively processes information. - **Rehearsal is not sufficient for LTM transfer**: Craik and Lockhart (1972) showed that depth of processing — not amount of rehearsal — determines long-term retention. Maintenance rehearsal (mere repetition) produces poor long-term memory even after extensive practice. - **Oversimplification of LTM**: the model does not distinguish between episodic, semantic, and procedural LTM — all of which have different properties and rely on different brain systems. - **Passivity**: the model implies memory is a one-way pipeline, when in reality LTM knowledge actively shapes what is encoded into and retrieved from STM.

The multi-store model (Atkinson & Shiffrin, 1968) treats STM as a single, unitary store that holds information temporarily before it either decays or is transferred to LTM through rehearsal. Its STM component is passive — a waiting room for information. Baddeley and Hitch (1974) replaced this with the working memory model, arguing that what the multi-store model called "STM" is actually a multicomponent system with a central executive (attentional control), a phonological loop (verbal/acoustic information), a visuospatial sketchpad (visual/spatial information), and later an episodic buffer (multisensory integration). Working memory is active — it does not just hold information but manipulates it for ongoing tasks like reading, arithmetic, and reasoning. The working memory model better explains dual-task performance, individual differences in cognitive ability, and the complexity of everyday cognitive tasks.