Key Memory Models

Why This Matters

Memory isn't just one thing. It's a collection of systems, processes, and structures that cognitive psychologists have spent decades trying to map. You're being tested on your ability to distinguish between these models and explain why each one matters for understanding how information gets encoded, stored, and retrieved.

The models here represent fundamentally different ways of conceptualizing memory: some focus on structural components (what memory is made of), others on processing dynamics (how information moves through the system), and still others on organizational principles (how knowledge gets structured and connected).

Don't just memorize the names and components of each model. Know what problem each model was designed to solve and how they relate to each other. Can you explain why Baddeley's working memory model was an improvement over the original multi-store model? Can you articulate the difference between structural and processing approaches to memory? These are the kinds of comparative questions that show up on exams, especially in FRQs asking you to evaluate or apply memory theories to real-world scenarios.

---

Structural Models: What Memory Is Made Of

These models propose that memory consists of distinct stores or systems, each with unique characteristics and functions. The key insight is that different types of information are handled by specialized components.

Atkinson-Shiffrin Model (Multi-Store Model)

This 1968 model was the first widely accepted framework for how memory is organized. It proposes a linear flow: information enters through the senses, gets selected for short-term processing, and may eventually make it into long-term storage.

• Three sequential stores form the backbone of the model. Sensory memory holds raw sensory input very briefly (iconic memory for vision lasts ~250–500 ms; echoic memory for sound lasts ~2–4 seconds). Short-term memory (STM) holds about 7 ± 2 items for roughly 15–30 seconds. Long-term memory (LTM) has essentially unlimited capacity and duration.
• Rehearsal is the critical mechanism for transferring information from STM to LTM. Without it, information decays or is displaced by new input.
• Attention acts as the gatekeeper between sensory and short-term memory, determining what gets processed further.

The model's main limitation is that it treats STM as a single, passive holding area. It can't easily explain why you can hold a phone number in mind while doing a spatial task like walking through a building. That limitation is exactly what motivated Baddeley's revision.

Baddeley's Working Memory Model

Baddeley and Hitch (1974) proposed this model to replace the idea of a single STM with a multi-component system that actively manipulates information, not just stores it.

• Four components handle different types of processing:
  • The phonological loop processes verbal and acoustic information (it's why you "hear" words in your head when reading).
  • The visuospatial sketchpad handles visual and spatial information (mental imagery, navigating a room).
  • The episodic buffer (added in 2000) integrates information from the other components and links it to long-term memory.
  • The central executive coordinates attention and manages cognitive resources across the other components. It explains why true multitasking is so difficult: the central executive has limited capacity.
• The model predicts that you can do two tasks simultaneously if they use different components (e.g., remembering a phone number via the phonological loop while mentally rotating an object via the visuospatial sketchpad), but performance drops when two tasks compete for the same component.

Tulving's Memory Systems

Tulving proposed that long-term memory isn't a single store but consists of at least three distinct systems, each with its own encoding and retrieval processes.

• Episodic memory stores personal experiences tied to a specific time and place ("my 10th birthday party").
• Semantic memory stores general knowledge and facts detached from personal context ("Paris is the capital of France").
• Procedural memory handles skills and habits, explaining why you can ride a bike without consciously remembering how you learned.
• Tulving's SPI framework describes how these systems interact: encoding is Serial (procedural → semantic → episodic), storage is Parallel (all three systems store simultaneously), and retrieval is Independent (you can access one system without needing the others). This independence explains why amnesic patients can learn new motor skills (procedural) even when they can't form new episodic memories.

---

Processing Models: How Encoding Determines Memory

These models shift focus from where information is stored to how it's processed. The depth and type of encoding matter more than time spent rehearsing.

Levels of Processing Model

Craik and Lockhart (1972) argued that memory isn't about moving information between stores. Instead, how well you remember something depends on how deeply you process it at encoding.

• Depth over duration: memory strength depends on how meaningfully information is processed, not how long it's held in STM.
• Shallow processing involves surface features. Structural processing (how a word looks) is the shallowest, followed by phonemic processing (how it sounds). Deep processing engages meaning and connections (semantic processing).
• Elaborative rehearsal, where you connect new information to existing knowledge, beats maintenance rehearsal (simple repetition) for long-term retention. Asking "how does this relate to what I already know?" is elaborative; repeating a definition over and over is maintenance.

A common criticism of this model is that "depth" is hard to define independently of memory performance, which makes the theory somewhat circular: deep processing leads to better memory, and we know processing was deep because memory was better.

Dual Coding Theory

Paivio's (1971) theory proposes that memory has two independent but connected coding systems: one for verbal information and one for visual/imaginal information.

• Verbal and visual encoding together produces stronger memory traces than either alone. This is why diagrams with labels, or flashcards with both a picture and a definition, tend to work well.
• Mental imagery serves as a second retrieval pathway. If you can't recall the word, you might recall the picture, and that image can cue the verbal information.
• Concrete concepts (e.g., "apple," "bicycle") are easier to remember than abstract ones (e.g., "justice," "entropy") because they're more easily visualized, giving them the advantage of dual codes.

---

Network Models: How Knowledge Is Organized

These models conceptualize memory as interconnected nodes rather than separate stores. Retrieval depends on activation patterns spreading through a web of associations.

Spreading Activation Theory

Collins and Loftus (1975) proposed that semantic memory is organized as a network of concept nodes connected by associative links. The strength of each link reflects how closely related two concepts are.

• Activating one concept primes related concepts. Thinking "doctor" makes "nurse" easier to retrieve because they're closely linked in the network. More distant concepts (e.g., "doctor" → "butter") receive less activation.
• Retrieval is a search process where activation spreads outward from a cue through connected nodes until the target is reached.
• Priming effects are the main experimental evidence for this model. In lexical decision tasks, people recognize "nurse" faster after seeing "doctor" than after seeing an unrelated word like "table."

Parallel Distributed Processing (PDP) Model

Rumelhart and McClelland's (1986) PDP model (also called connectionism) takes a very different approach from spreading activation. Instead of discrete concept nodes, memory is represented as patterns of activation distributed across many simple processing units.

• Information isn't stored in single locations but across networks of interconnected units. A "memory" is a particular pattern of activation across the network, not a thing sitting in one place.
• Simultaneous (parallel) processing allows the system to handle multiple inputs at once, unlike serial processing models.
• Learning strengthens connections between units that are co-activated, echoing the Hebbian principle: neurons that fire together wire together. Over many learning trials, the network adjusts its connection weights to produce the correct output patterns.
• Graceful degradation means the system can still function even when some units are damaged, because information is distributed. This explains why brain injuries typically cause partial memory loss rather than erasing entire memories cleanly.

---

Organizational Models: How Prior Knowledge Shapes Memory

These models explain how existing knowledge structures influence the encoding and retrieval of new information. What you already know determines what you can learn and remember.

Schema Theory

Bartlett (1932) introduced the idea of schemas, and the concept has been central to cognitive psychology ever since. A schema is an organized mental framework built from past experience that helps you interpret, encode, and retrieve new information.

• Top-down processing means schemas guide what you notice, what you encode, and what you later reconstruct during retrieval. You don't passively record events; you actively interpret them through existing frameworks.
• Schema-consistent information is generally remembered better because it fits neatly into your existing framework. But schemas can also cause memory distortions when you unconsciously "fill in" details that weren't actually present but are consistent with the schema. Bartlett's classic "War of the Ghosts" study showed that participants distorted an unfamiliar Native American folk tale to fit their own cultural schemas when recalling it.
• This is directly relevant to eyewitness testimony: a witness's schema for "robbery" might cause them to "remember" a weapon that wasn't actually present.

Episodic and Semantic Memory Distinction

This distinction (formalized by Tulving, 1972) categorizes the content of long-term declarative memory rather than explaining a processing mechanism.

• Episodic memory stores personal experiences with contextual details: what happened, where, and when. It involves autonoetic consciousness, the sense of mentally traveling back in time to re-experience an event.
• Semantic memory stores general knowledge abstracted from specific episodes: facts, concepts, word meanings. You know that Paris is the capital of France without needing to remember the specific moment you learned it.
• Bidirectional influence: repeated episodic experiences can become semantic knowledge (after visiting many restaurants, you develop a general "restaurant schema"), and semantic knowledge shapes how new episodes are encoded and remembered.

---

Quick Reference Table

---

Self-Check Questions

1. Both the Levels of Processing model and Dual Coding Theory emphasize encoding quality. What distinguishes their explanations for why some information is remembered better than others?

2. If a patient with brain damage can still recall general facts but cannot remember personal experiences, which model best explains this dissociation, and what specific memory systems are affected?

3. Compare and contrast the Atkinson-Shiffrin model with Baddeley's Working Memory model. What limitation of the original model did Baddeley's revision address?

4. A student uses the method of loci (imagining items placed in familiar locations) to memorize a list. Which two memory models best explain why this technique works?

5. An FRQ asks you to explain why eyewitness testimony can be unreliable. Which memory model would you use, and what specific mechanism would you describe?

6. How does the PDP model's concept of "graceful degradation" explain why brain damage typically causes partial memory loss rather than the complete erasure of specific memories?