The three features of each store

Every memory store is described by three things: how information is coded (the format it is stored in), how much it can hold (capacity) and how long it lasts (duration).

• Coding — Baddeley (1966). Participants recalled word lists that were acoustically similar (cat, cab, can), acoustically dissimilar, semantically similar (great, large, big) or semantically dissimilar. Recalling immediately (STM), they did worst on acoustically similar words. Recalling after 20 minutes (LTM), they did worst on semantically similar words. Conclusion: STM codes largely acoustically, LTM largely semantically.
• Capacity — Jacobs (1887) digit span: mean 9.3 digits and 7.3 letters. Miller (1956): the span of immediate memory is about 7 ± 2 items, extended by chunking.
• Duration — Peterson and Peterson (1959): participants recalled nonsense trigrams after counting backwards (to prevent rehearsal). After 3 seconds recall was about 80%; after 18 seconds it had fallen to about 3%. STM duration is therefore roughly 18–30 seconds without rehearsal. Bahrick et al. (1975) tested 392 Americans on school yearbook photos: photo recognition was about 90% accurate within 15 years of graduating and about 70% after 48 years — LTM can last a lifetime.

Evaluation: Baddeley's word lists are artificial — they have no personal meaning, so they may not reflect everyday memory. Peterson and Peterson's nonsense trigrams have the same problem. Miller may have over-estimated capacity: Cowan (2001) reviewed the evidence and concluded STM is closer to 4 ± 1 chunks. Bahrick's study, by contrast, has high external validity because it used real, meaningful memories.

Which store, which code?

Atkinson and Shiffrin's multi-store model (1968)

• Sensory register — one store per sense (iconic = visual, echoic = acoustic). Very high capacity but duration under half a second. Material passes on only if you pay attention to it.
• Short-term memory — coded acoustically, capacity about 5–9 items, duration about 18–30 seconds. Maintenance rehearsal keeps material in STM and, if prolonged, transfers it to LTM.
• Long-term memory — coded semantically, potentially unlimited capacity and lifetime duration. Recall works by transferring material back into STM (retrieval).

Evaluation: supported by Baddeley's coding studies and by case studies such as HM, who could not form new long-term memories after hippocampal surgery but had a working STM — showing the stores are separate. But the model says STM is a single, unitary store, and this is wrong: patient KF (Shallice and Warrington, 1970) had very poor STM for auditory digits but much better recall for visually presented digits, implying at least two STM components. The model also over-simplifies rehearsal: Craik and Watkins showed it is the type of rehearsal (elaborative, meaningful processing) rather than the amount that determines long-term storage.

Baddeley and Hitch's working memory model (1974)

WMM replaces the MSM's single STM with an active, multi-component system:

• Central executive — the supervisory 'attentional' component. It allocates processing resources to the slave systems. Very limited capacity; it stores nothing itself.
• Phonological loop — deals with auditory information and preserves word order. Subdivided into the phonological store (the 'inner ear' — words you hear) and the articulatory process (the 'inner voice' — maintenance rehearsal). Its capacity is about 2 seconds of speech.
• Visuo-spatial sketchpad — the 'inner eye', storing visual and spatial information; capacity about 3–4 objects (Baddeley). Logie (1995) divided it into the visual cache (form and colour) and the inner scribe (arrangement of objects).
• Episodic buffer — added by Baddeley (2000). A temporary, limited-capacity store (about 4 chunks) that integrates information from the other components into a single record and links working memory to LTM.

Dual-task evidence: people can do a verbal and a visual task simultaneously with little loss of performance, but two tasks that use the same slave system interfere badly — because that component's capacity is exhausted.

Evaluation: KF supports the separate slave systems — his auditory (phonological) recall was impaired while visual recall was fine. The word-length effect shows the phonological loop's 2-second limit: people recall more short words than long words, and the effect disappears under articulatory suppression (repeating 'the the the'), because the articulatory process is occupied. The weakest link is the central executive: it is vague and unfalsifiable, and researchers suspect it is not one thing but several separable functions.

Which component?

Interference — proactive and retroactive

Forgetting from LTM occurs because two sets of information compete. Interference is worse when the memories are similar.

• Proactive interference (PI) — old memories disrupt new ones (your old phone number stops you learning the new one).
• Retroactive interference (RI) — new memories disrupt old ones (this year's French vocabulary blocks last year's Spanish).

McGeoch and McDonald (1931) had participants learn a list of 10 words to perfect recall, then learn a second list. Recall of the first list was worst when the second list was synonyms of the original (about 12% accuracy) and best when it was numbers or when participants rested. So similarity of material is the key variable.

Baddeley and Hitch (1977) asked rugby players to recall the names of teams they had played that season. Recall depended not on how long ago the game was but on the number of games played in between — real-world retroactive interference.

Evaluation: the laboratory studies are highly controlled but artificial: real learning is rarely as similar, or as close together in time, as two word lists learned in one session. Interference also seems only to cause temporary loss — Tulving and Psotka (1971) found that giving participants cued recall restored accuracy to about 70%, whatever the number of interfering lists, suggesting the memory was still there and the true problem was accessibility, not availability.

Retrieval failure and cue-dependent forgetting

Tulving's encoding specificity principle (ESP): a cue helps recall only if it is present at encoding and present at retrieval. If cues differ, or are absent, you get retrieval failure — the memory is available but not accessible.

• Context-dependent forgetting — Godden and Baddeley (1975). Deep-sea divers learned word lists on land or underwater and recalled in the same or a different environment. Recall was about 40% lower in the non-matching conditions: the external environment is a cue.
• State-dependent forgetting — Carter and Cassaday (1998). Participants learned and recalled material on or off a mildly sedating antihistamine. When the internal state at learning and recall did not match, recall was significantly worse: your bodily/emotional state is a cue.

Evaluation: retrieval failure has strong real-world application — it is why the cognitive interview asks witnesses to mentally reinstate the context of a crime, and why revising in conditions similar to the exam hall helps. Baddeley himself, however, argued that context effects are actually quite weak in real life: contexts have to be very different indeed (land vs underwater) before an effect shows. There is also a problem of circularity in the ESP — it is difficult to test independently whether a cue really was encoded.

Which explanation?

Leading questions and post-event discussion

Loftus and Palmer (1974) — leading questions. 45 students watched film clips of car crashes and were asked "About how fast were the cars going when they ___ each other?" Only the verb changed.

• Mean speed estimate for "contacted": 31.8 mph. For "smashed": 40.8 mph.
• In a second experiment, a week later, participants were asked whether they had seen broken glass (there was none). 16 of 50 in the 'smashed' condition said yes, compared with 7 of 50 in the 'hit' condition.

Two possible mechanisms: a response bias (the wording nudges the answer but the memory is unchanged) or substitution (the wording actually alters the stored memory — the broken-glass finding supports this).

Gabbert et al. (2003) — post-event discussion. Pairs of participants watched the same crime from different angles, so each saw details the other could not. After discussing it, 71% of witnesses recalled aspects of the event they had not actually seen; the figure in a control group with no discussion was 0%. This is memory conformity — witnesses go along with each other, through NSI or ISI.

Evaluation: the research has huge practical value — Loftus argues eyewitness testimony should be treated with great caution in court, and it has shaped police interview practice. But film clips carry far less emotional impact than a real crime, and Foster et al. (1994) found participants were more accurate when they believed they were watching a real robbery and their responses would influence a trial. Lab studies may therefore over-state the unreliability of real witnesses.

Which effect?

Anxiety — does it help or harm accuracy?

Anxiety has a negative effect — Johnson and Scott (1976), 'weapon focus'. Participants in a waiting room overheard either a casual conversation, after which a man walked out holding a pen with grease on his hands, or a heated argument, after which a man emerged holding a bloodied paper knife. Later, 49% of the low-anxiety group correctly identified the man from 50 photos, compared with 33% of the high-anxiety group. The explanation is that anxiety narrows attention onto the weapon and away from the face.

Anxiety has a positive effect — Yuille and Cutshall (1986). A real shooting outside a gun shop in Vancouver, in which the thief was killed. 13 of the 21 witnesses were re-interviewed five months later. Their accounts remained strikingly accurate, and they resisted two leading questions inserted by the researchers. Witnesses who had reported the highest stress at the time were the most accurate (about 88% accurate, compared with about 75% for the less-stressed group).

The resolution — the Yerkes-Dodson inverted U. Deffenbacher argued that as arousal rises, accuracy improves up to an optimum and then declines at very high levels. This explains both sets of findings.

Evaluation: Johnson and Scott may not have tested anxiety at all — Pickel (1998) showed that identification was poorest when the object held was unusual (a raw chicken or a handgun in a hairdressing salon), suggesting the effect is caused by surprise, not fear. Field studies like Yuille and Cutshall have high ecological validity but poor control: witnesses had discussed the event and read media coverage in the interim. The inverted-U itself is criticised as overly simplistic, since it treats anxiety as purely physiological and ignores the cognitive component.

The cognitive interview (Fisher and Geiselman, 1992)

Built directly on memory theory — especially the encoding specificity principle — the CI has four techniques:

• Report everything — include every detail, even apparently trivial ones. Small details may trigger other memories, and the witness cannot know what matters.
• Reinstate the context — mentally return to the scene: the weather, your mood, what you could smell. This provides contextual cues (Godden and Baddeley).
• Reverse the order — recall events in a different chronological order. This prevents recall being driven by expectations/schema about how such events usually unfold, and makes lying harder.
• Change perspective — describe the event from another person's viewpoint, again to disrupt the effect of schema.

The enhanced CI adds social elements: build rapport, minimise distractions, let the witness control the flow of the interview, avoid leading questions.

Evaluation: Köhnken et al. (1999) meta-analysed 55 studies and found the CI produced an average 81% increase in correct information compared with the standard police interview — but also an increase in incorrect details, so the CI raises quantity at some cost to accuracy. It is also time-consuming and requires special training, which is why many police forces use only selected elements (typically 'report everything' and 'reinstate the context'), making the research base hard to compare with practice.

Which CI technique?

WMM capacity

Proactive or retroactive?

Filing the evidence

Tap a card, then the part of the topic it belongs to. Being able to place a study instantly is what separates a top-band essay from a vague one.

Component and job

Tap an item on the left, then its partner on the right.

The big ideas to know

You have covered the whole of AQA 4.1.2 Memory. Press Finish to see your score.