Memory

The Multi-Store Model of Memory

Atkinson and Shiffrin (1968)

The multi-store model (MSM) proposes that memory consists of three structurally and functionally distinct stores: sensory memory, short-term memory (STM), and long-term memory (LTM). Information flows through these stores in a linear sequence, with attention, rehearsal, and retrieval as the key processes.

Sensory Memory

Information from the environment is detected by sensory receptors and held briefly in sensory memory. Each sensory modality has its own store:

• Iconic store (visual): holds visual information for approximately $0.5$ seconds. Sperling (1960) demonstrated its capacity: participants could recall approximately $4$--$5$ items from a $3 \times 4$ grid flashed for $50\;\mathrm{ms}$, but when cued to recall a specific row immediately after the display, they could recall approximately $3$--$4$ items from that row, suggesting the full display was briefly stored.
• Echoic store (auditory): holds auditory information for approximately $3$--$4$ seconds.

Capacity: high (all sensory input is briefly stored)

Duration: very brief (milliseconds to a few seconds)

Encoding: modality-specific (visual or auditory)

Information that is attended to passes from sensory memory to STM; the rest is lost through decay.

Short-Term Memory (STM)

Capacity: George Miller (1956) proposed that STM can hold approximately $7 \pm 2$ items (the "magical number seven"). Jacobs (1887) found an average digit span of $9.3$ items and an average letter span of $7.3$ items. Capacity can be increased through chunking — grouping individual items into meaningful units (e.g., "BBC" as one chunk rather than three letters).

Duration: Peterson and Peterson (1959) found that STM duration is approximately $18$--$30$ seconds when rehearsal is prevented. Participants recalled trigrams (three-letter nonsense syllables) with high accuracy after $3$ seconds but accuracy declined to approximately $10\%$ after $18$ seconds.

Encoding: primarily acoustic (sound-based). Baddeley (1966) found that participants made fewer errors when recalling acoustically similar words in STM after an interval, but more errors when recalling semantically similar words, suggesting STM relies on acoustic coding.

Long-Term Memory (LTM)

Capacity: theoretically unlimited. There is no known upper limit to the amount of information that can be stored in LTM.

Duration: potentially a lifetime. Bahrick et al. (1975) tested participants' memory for Spanish learned in school up to $50$ years earlier and found that recall declined for the first $3$--$6$ years but then levelled off, with participants retaining some knowledge even after $50$ years.

Encoding: primarily semantic (meaning-based). Baddeley (1966) found that participants made more errors when recalling semantically similar words (e.g., "big, huge, large") in LTM after a $20$-minute delay, compared to acoustically similar words, suggesting LTM relies on semantic coding.

Processes

• Attention: selects information from sensory memory for transfer to STM
• Rehearsal: maintenance rehearsal (repeating information to keep it in STM) and elaborative rehearsal (linking new information to existing knowledge in LTM) transfer information from STM to LTM
• Retrieval: accessing stored information from LTM. Retrieval failure (rather than storage failure) is often the cause of forgetting

Evaluation of the MSM

Strengths:

• Provides a clear, testable framework that has generated extensive research
• Supported by evidence for distinct STM and LTM stores (e.g., brain damage cases: Clive Wearing has intact STM but impaired LTM; KF has impaired STM but intact LTM)
• Case studies of brain-injured patients provide strong evidence for the structural distinction between STM and LTM

Limitations:

• Oversimplified: the model treats STM as a unitary store, but research shows it has multiple components (Baddeley and Hitch's working memory model)
• Rehearsal is not the only route to LTM: information can enter LTM without rehearsal (e.g., deeply processed information, flashbulb memories)
• The model is passive and mechanistic: it describes storage and transfer but does not account for how information is actively processed or manipulated
• The linear flow of information does not reflect the complexity of real memory (e.g., information can flow from LTM back to STM)

The Working Memory Model

Baddeley and Hitch (1974)

The working memory model (WMM) proposes that STM is not a unitary store but an active, multi-component system for temporarily storing and manipulating information.

Components

Central executive: the supervisory system that allocates attention to tasks, coordinates the slave systems, and integrates information. It has limited capacity and is responsible for higher-order cognitive processes (planning, problem-solving, reasoning). The central executive is modality-free and does not store information itself.

Phonological loop: processes auditory and verbal information. Comprises two sub-components:

• Phonological store (inner ear): holds auditory information for approximately $2$ seconds
• Articulatory rehearsal process (inner voice): rehearses verbal information subvocally, refreshing the contents of the phonological store and maintaining information beyond $2$ seconds

The word length effect (words that take longer to articulate are recalled less well) supports the existence of the articulatory rehearsal process.

Visuospatial sketchpad: processes visual and spatial information. Stores and manipulates mental images and spatial layouts. Evidence: participants perform better on spatial tasks when using the visuospatial sketchpad than when simultaneously performing a verbal task.

Episodic buffer (added by Baddeley, 2000): a limited-capacity temporary store that integrates information from the phonological loop, visuospatial sketchpad, and LTM into a coherent episodic representation. It serves as the interface between working memory and LTM and accounts for the fact that people can integrate verbal and visual information.

Evidence for the WMM

• Dual-task performance: participants can perform a verbal task and a spatial task simultaneously with minimal interference, supporting the existence of separate subsystems. However, performing two verbal tasks simultaneously causes significant interference, supporting the single phonological loop.
• Brain imaging: fMRI studies show that different brain areas are activated during verbal tasks (left hemisphere, Broca's area) and spatial tasks (right hemisphere, parietal and occipital areas).
• Case studies: KF (Shallice and Warrington, 1970) had impaired verbal STM but intact visual STM, supporting the separation of the phonological loop and visuospatial sketchpad.

Evaluation of the WMM

Strengths:

• Accounts for findings that the MSM cannot explain (e.g., dual-task performance, brain damage affecting one component but not others)
• Supported by a substantial body of experimental and neuropsychological evidence
• More realistic than the MSM: working memory is active and involves processing, not just storage

Limitations:

• The central executive is poorly specified: its exact functions and capacity are unclear, and it is difficult to study empirically
• The episodic buffer is a relatively recent addition introduced to address the model's inability to explain the integration of information from different modalities — it may be an ad hoc fix
• The model does not explain how information is transferred to LTM or how forgetting occurs within working memory

Types of Long-Term Memory

Episodic Memory

Episodic memory stores personal experiences and events, including their temporal and spatial context ("what, where, when"). Examples: your first day at school, what you had for breakfast yesterday. Tulving (1972, 1983) distinguished episodic memory from semantic memory.

Semantic Memory

Semantic memory stores general knowledge, facts, concepts, and their meanings, independent of personal experience. Examples: knowing that Paris is the capital of France, understanding the meaning of "democracy," knowing that water freezes at $0^{\circ}\mathrm{C}$.

Procedural Memory

Procedural memory stores skills, routines, and motor actions that can be performed automatically without conscious awareness. Examples: riding a bicycle, typing, playing a musical instrument. Procedural memory is highly resistant to forgetting and is often preserved in amnesic patients (e.g., Clive Wearing could still play the piano despite severe anterograde amnesia).

Evidence for the Distinction

• Brain imaging: different brain regions are associated with different types of LTM. The hippocampus is critical for episodic memory; the temporal cortex for semantic memory; the basal ganglia and cerebellum for procedural memory.
• Amnesia case studies: HM (Scoville and Milner, 1957) had intact procedural memory (learned mirror-drawing tasks) but impaired episodic and semantic memory, supporting the distinction between procedural and declarative (episodic + semantic) memory.
• Clive Wearing: severe anterograde and retrograde amnesia (cannot form new episodic memories or recall most past ones) but intact procedural memory (can play music and conduct a choir).

Explanations of Forgetting

Interference

Interference occurs when other information competes with the target memory, making retrieval more difficult.

Proactive interference (PI): older memories interfere with the retrieval of newer memories. For example, calling your new partner by your ex-partner's name. Underwood (1957) found that participants who had previously learned many word lists made more errors when learning a new list.

Retroactive interference (RI): newer memories interfere with the retrieval of older memories. For example, changing your password and being unable to recall the old one. McGeoch and McDonald (1931) found that participants who learned a second list of synonyms recalled fewer items from the first list than those who learned a second list of unrelated words.

Interference is strongest when the competing memories are similar in content.

Retrieval Failure

Cue-dependent forgetting (Tulving, 1975): information is stored in LTM but cannot be retrieved because the appropriate cues are absent. According to Tulving's encoding specificity principle, memory is improved when the conditions at retrieval match the conditions at encoding.

• Context-dependent forgetting: failure occurs because the environmental context at retrieval differs from the context at encoding. Godden and Baddeley (1975) had divers learn word lists on land or underwater. Recall was approximately $50\%$ better when the learning and recall contexts matched.
• State-dependent forgetting: failure occurs because the internal physiological or psychological state at retrieval differs from the state at encoding. Carter and Cassaday (1998) found that participants who learned information while experiencing antihistamine-induced drowsiness recalled more when tested in the same drowsy state.

Eyewitness Testimony

Factors Affecting the Accuracy of EWT

Anxiety: the weapon focus effect (Loftus et al., 1987) demonstrates that the presence of a weapon draws attention away from the perpetrator's face, reducing identification accuracy. In Loftus's study, participants who viewed a scenario involving a weapon (a man holding a cheque covered in blood) were less accurate in identifying the man than those who viewed a scenario without a weapon (a man holding a pen).

Moderate anxiety may improve accuracy for central details but impair accuracy for peripheral details (Christianson, 1992). High anxiety impairs accuracy overall due to the narrowing of attention and the impact of stress on memory consolidation.

Age: children are generally less accurate eyewitnesses than adults. They are more susceptible to leading questions, have poorer recall of details, and are more likely to identify someone from a lineup even when the culprit is absent (Parker and Carranza, 1989). Elderly witnesses also show reduced accuracy compared to younger adults.

Leading questions: Loftus and Palmer (1974) found that the wording of questions could alter memory. Participants who were asked "About how fast were the cars going when they smashed into each other?" gave higher speed estimates ($40.8\;\mathrm{mph}$) than those asked "when they hit each other" ($31.8\;\mathrm{mph}$). A week later, participants in the "smashed" condition were more likely to (falsely) report seeing broken glass ($32\%$ vs. $14\%$).

This is explained by the post-event misinformation effect: information encountered after an event (e.g., leading questions, media reports, conversations with other witnesses) can be incorporated into the original memory, altering the memory trace.

The Cognitive Interview

Fisher and Geiselman (1992) developed the cognitive interview (CI) to improve the accuracy and completeness of eyewitness testimony. It is based on psychological principles of memory retrieval.

Four main techniques:

1. Report everything: the witness is encouraged to report every detail they can remember, even if it seems trivial or irrelevant. This increases the amount of information retrieved.
2. Reinstate the context: the witness mentally recreates the physical and psychological context of the event (weather, sounds, emotions, surroundings) to provide effective retrieval cues.
3. Change the order: the witness recalls the events in a different chronological order (e.g., from the end to the beginning). This disrupts reliance on schema-based expectations and may retrieve details that would not emerge in a standard forward recall.
4. Change perspective: the witness recalls the event from the perspective of other people who were present (e.g., the victim, another witness). This provides additional retrieval cues.

Evaluation: the CI has been shown to increase the quantity of correct information recalled by $25$--$45\%$ compared to standard police interviews, without a significant increase in errors. However, it is more time-consuming to conduct and requires trained interviewers.

Common Pitfalls

• Confusing STM duration and capacity. STM duration is $18$--$30$ seconds (Peterson and Peterson); capacity is $7 \pm 2$ items (Miller).
• Confounding the MSM with the WMM. The MSM proposes a unitary STM; the WMM proposes a multi-component working memory with separate subsystems.
• Confusing proactive and retroactive interference. Proactive = old interferes with new; retroactive = new interferes with old.
• Stating that leading questions "change" the memory. The more precise explanation is that they introduce post-event misinformation that is integrated into the memory trace (the misinformation effect).
• Confusing episodic and semantic memory. Episodic = personal events with context; semantic = general knowledge without personal context.

Practice Problems

Problem 1: MSM Evaluation

Outline and evaluate the multi-store model of memory (12 marks).

The multi-store model (Atkinson and Shiffrin, 1968) proposes that memory consists of three stores: sensory memory, short-term memory, and long-term memory. Information flows through these stores via attention, rehearsal, and retrieval. Sensory memory has a large capacity but very short duration (milliseconds). STM has a limited capacity ($7 \pm 2$) and short duration ($18$--$30$ seconds). LTM has unlimited capacity and potentially permanent duration.

One strength is that the model is supported by evidence from brain-damaged patients. Clive Wearing suffered damage to his hippocampus from herpes simplex encephalitis, resulting in severe anterograde and retrograde amnesia. He cannot form new memories or recall most of his past, yet his STM remains intact (he can hold a conversation). This supports the idea that STM and LTM are separate stores.

A second strength is that each store has distinct characteristics supported by research. Peterson and Peterson demonstrated STM's brief duration; Sperling demonstrated sensory memory's large capacity and brief duration; Bahrick et al. demonstrated LTM's very long duration.

One limitation is that the model is oversimplified. It treats STM as a unitary store, but Baddeley and Hitch's working memory model demonstrates that STM has multiple components (phonological loop, visuospatial sketchpad). Case study KF had impaired verbal STM but intact visual STM, which the MSM cannot explain.

A second limitation is that the model overemphasises rehearsal as the mechanism for transferring information to LTM. Craik and Lockhart's levels of processing theory demonstrated that deeper, more meaningful processing (semantic processing) leads to better recall than shallow rehearsal, even without rehearsal. Information can enter LTM without maintenance rehearsal.

Overall, the MSM provides a useful foundational framework but fails to account for the complexity of memory as demonstrated by subsequent research.

Problem 2: WMM Application

Explain how the working memory model accounts for a student's ability to listen to a lecture while taking notes, but difficulty in listening to a lecture while simultaneously discussing a different topic with a neighbour.

The working memory model explains this pattern through its multi-component structure.

Listening to a lecture while taking notes involves the central executive coordinating two separate slave systems: the phonological loop processes the spoken lecture (auditory-verbal information), while the visuospatial sketchpad processes the spatial layout of the notes being written. Because different slave systems handle the two tasks, minimal interference occurs, and the student can perform both simultaneously.

Discussing a different topic with a neighbour while listening to a lecture creates interference because both tasks require the phonological loop. The lecture and the conversation are both verbal-auditory inputs competing for the same limited-capacity subsystem. The phonological loop cannot simultaneously process two streams of verbal information, so the student experiences difficulty and one or both tasks suffer.

This illustrates a key prediction of the WMM: dual-task performance is easier when tasks draw on different slave systems than when they compete for the same subsystem.

Problem 3: Leading Questions

A witness to a car accident is asked by a police officer: "How fast was the car going when it ran the red light?" The car did not actually run a red light. Explain how this question might affect the witness's memory.

This question is a leading question because it presupposes that the car ran a red light (a false premise). According to Loftus and Palmer's (1974) research on the post-event misinformation effect, the phrasing of questions can alter a witness's memory of an event.

The question introduces post-event misinformation — information that was not part of the original experience but is presented after the event. When the witness processes this question, the false information ("ran the red light") may be integrated into their memory of the event through one of two mechanisms:

1. Response bias: the witness adjusts their response to fit the question's presupposition without actually changing their memory (they may give a higher speed estimate because the question implies recklessness).

2. Memory alteration: the misinformation is incorporated into the original memory trace, replacing or altering the genuine memory. If asked later, the witness may genuinely recall the car running a red light, even though this did not happen.

This has serious implications for the legal system: police and lawyers must be careful to use neutral, non-leading questions to avoid contaminating eyewitness testimony.

Problem 4: Retrieval Failure

A student revises in a quiet library but takes the exam in a noisy sports hall. Using cue-dependent forgetting theory, explain why the student may underperform and suggest how this could be addressed.

According to Tulving's encoding specificity principle, retrieval is most effective when the cues available at the time of recall match the cues present at the time of encoding. The discrepancy between the quiet library (encoding context) and the noisy sports hall (retrieval context) creates a context-dependent forgetting effect.

At encoding, the student's memory traces are associated with contextual cues: the quiet atmosphere, visual surroundings, lighting, and the absence of external noise. At retrieval, these cues are absent, and different cues (noise, crowds, different visual environment) are present. The mismatch reduces the accessibility of the stored information.

Practical solutions:

1. Context reinstatement: the student could mentally recreate the library environment during the exam (imagining the quiet, the desk, the book layout), providing internal retrieval cues that match encoding conditions.

2. Context similarity: the student could practise revision in conditions that more closely match the exam environment (e.g., practising in a noisy room or a large hall) to encode the information with more exam-relevant cues.

3. Multiple context learning: studying the same material in different environments creates multiple sets of contextual cues, reducing dependence on any single context.

Problem 5: Cognitive Interview Evaluation

Evaluate the cognitive interview as a technique for improving eyewitness testimony.

Strengths:

1. Effectiveness: Fisher and Geiselman's research found that the CI increases the amount of correct information recalled by $25$--$45\%$ compared to standard police interviews. This is a substantial improvement with practical significance for criminal investigations.

2. Theoretical basis: the CI is grounded in well-established psychological principles of memory retrieval (cue-dependent forgetting, encoding specificity, context reinstatement), giving it a scientific foundation.

3. No significant increase in errors: despite producing more information, the CI does not significantly increase the number of inaccurate details recalled, suggesting that the additional information is genuinely retrieved rather than fabricated.

Limitations:

1. Time-consuming: the CI takes longer to conduct than standard interviews (approximately $2$--$3$ times longer), which is impractical for police forces with limited resources and large caseloads.

2. Training requirements: police officers require training to use the CI effectively. Without proper training, officers may not implement the techniques correctly, reducing their effectiveness. Training is costly and time-intensive.

3. Not universally effective: the CI is less effective with child witnesses and witnesses with cognitive impairments, who may struggle with the cognitive demands of reinstating context and changing perspective.

4. Increased confidence without increased accuracy: some research suggests that the CI increases witness confidence in their testimony, but confidence is a poor predictor of accuracy. This could create problems in court if juries are influenced by confident but inaccurate testimony.