Levels of processing: what Craik and Lockhart actually proposed in 1972

Levels of processing describes how far you analyse information at encoding. You notice its appearance, its sound, or its meaning. Fergus Craik and Robert Lockhart proposed in 1972 that a memory trace lasts longer when this analysis goes deeper. Deeper means closer to meaning.

Most summaries name three levels and give one example. This article shows the experiments behind the claim with their numbers. This article also gives you one question that assigns any study task to a level. This article also explains the criticism that followed the proposal from the start.

• The claim: what you process by meaning, you keep longer. Depth means degree of semantic analysis.
• Craik and Tulving tested it in 1975 with three kinds of questions about words. The questions concerned print, rhyme, and meaning.
• In one experiment, students later recognised 15% of words from a print question and 81% from a sentence question. Both figures concern words with yes answers.
• Time on task did not explain the gap. A hard but shallow task took longer and still lost.
• Two limits stay: depth has no independent measure, and the best encoding depends on the later test.
• Relating material to yourself helped most of all in later studies. It is a later finding, not one of the original levels.

The three levels in one table

Craik and Lockhart described a continuum in 1972, not three fixed boxes. The three levels below are the three kinds of questions Craik and Endel Tulving used in 1975. The table shows them with example questions from that paper. The questions and words were English.

The German standard lexicon lists the approach under its English name. The Dorsch lexicon of psychology gives Verarbeitungsebenen and Theorie der Verarbeitungstiefe as German labels. It defines the idea as retention differences from shallow sensory analysis to degrees of semantic elaboration.

The one question that sorts any task

Exam tasks often describe what a person does with material and ask for the level. Use one question: what did the person have to attend to in order to solve the task? Work through four steps.

1. 1.Find the task the person performs on the material.
2. 2.If the person only checks appearance, such as letters or print, the processing is structural and shallow.
3. 3.If the person checks sound, such as a rhyme, the processing is phonemic.
4. 4.If the person must use meaning, such as a category or a sentence, the processing is semantic and deep.

Watch for one trap. A hard task is not deep by itself. The third row of the next table shows a difficult task that stays shallow. All seven tasks come from the studies behind this article.

What Craik and Lockhart proposed in 1972

In the 1960s, leading memory models explained retention with stores. Information moved from a sensory store to a short-term store and then to a long-term store. Craik and Lockhart of the University of Toronto questioned this view in 1972. They did not ask in which store information sits. They asked what the person does with it.

Their proposal has three parts.

• A memory trace is a by-product of perception. When you analyse a stimulus, a trace of that analysis remains.
• Analysis starts with simple features such as lines, brightness, or pitch. Later steps detect patterns and extract meaning. Greater depth means a greater degree of semantic or cognitive analysis.
• How long a trace lasts depends on depth of analysis. Deeper analyses produce traces that are more elaborate, longer lasting, and stronger.

The levels are not fixed steps. Craik and Lockhart described processing as a continuum. They even wrote that spread of encoding might describe the later steps more accurately than depth. They kept the word depth because it conveyed the idea well.

Type I and Type II rehearsal

Craik and Lockhart separated two ways of repeating information. In Type I processing, the person keeps information active at the same level, for example by inner speech. They called this primary memory. The information stays available, but by their account no durable trace forms. In Type II processing, the person analyses the information more deeply. Only this second kind should improve memory.

One year later, Craik and Michael Watkins gave the two kinds the names maintenance rehearsal and elaborative rehearsal. The first kind holds information in place. The second kind connects it with meaning. Their result read: maintenance rehearsal does not improve memory performance. This claim met objections, as the criticism section shows.

The 1972 evidence

Craik and Lockhart drew mainly on incidental learning studies. In such studies, subjects work on words and do not know that a memory test follows. The task sets how they process the words.

• Tresselt and Mayzner gave three groups different tasks in 1960. One group crossed out vowels, one copied the words, and one rated how far each word was an instance of the concept economic. The third group recalled four times as many words as the first group and twice as many as the second.
• In studies by Hyde, Jenkins and colleagues, people with a meaning task recalled as much as a group that tried to learn on purpose. Both groups did clearly better than a group that only searched for certain letters or estimated the number of letters.

Craik and Lockhart drew one result from this: the intent to learn helps only when it leads to more effective processing.

A framework, not a theory

Craik and Lockhart wrote: Our approach does not constitute a theory of memory. It provides a conceptual framework within which memory research might proceed. They called their own position speculative and far from complete. They had studied encoding only, not retrieval. The label theory of depth of processing stays in use, but the authors spoke of a framework.

Levels of processing vs the multi-store model

The Dorsch lexicon calls levels of processing a term first set against the multi-store hypothesis. The table shows where the two views differ. The multi-store model of Atkinson and Shiffrin dates to 1968. Craik and Lockhart dates to 1972.

Atkinson and Shiffrin wrote in 1968 that any information in the short-term store moves to some degree into the long-term store while it stays there. By that claim, long rehearsal should always help. Craik and Lockhart denied exactly this point.

What Craik and Tulving showed in 1975

Craik and Endel Tulving tested the approach in 1975 in ten experiments. Tulving also drew the distinction between episodic memory and semantic memory. The 1975 title speaks of retention of words in episodic memory.

The subjects took the studies for research on perception and reaction time. For each word, they answered one yes or no question. In most experiments, they did not expect a memory test.

Experiment 2: 15% against 81%

In Experiment 2, 24 students worked through 60 questions. Each level had 10 yes questions and 10 no questions. They heard the question, the word appeared 2 seconds later for 200 milliseconds, and they pressed a key for yes or no. All words were common concrete nouns with five letters. Then the subjects received a list of 180 words: the 60 shown words and 120 new words. They marked every word they had seen.

Meaning questions took longer than print questions. The print question asked whether the word was printed in capital or small letters. For yes words, recognition rose from 15% after the print question to 81% after the sentence question. The authors called this more than a fivefold increase, in the same people in the same experiment. For no words, it rose less steeply, from 19% to 49%.

Time or depth?

Deep questions took longer. Perhaps time explained the better retention and not depth. Experiment 5 tested this with a hard but shallow task. The subjects decided whether a word matched a consonant and vowel pattern. This decision took 1.7 seconds, the sentence question 0.85 seconds. Still, the subjects recognised the words from the sentence question better.

Craik and Tulving drew this result: the kind of task sets memory performance, not the time. In their words, a minimal semantic analysis helps memory more than an extensive structural analysis.

Intent and reward

One of the two authors called himself more sceptical by nature than the other. He suspected that the strict rules of the studies were not needed. Experiment 9 therefore broke several of those rules. Twelve subjects sat as a group in a course on learning and memory, and they knew about the memory test. A new word came every 6 seconds: it stayed visible for 1 second, then 5 seconds remained for the answer. The pattern matched Experiment 2. A repeat with 12 other people confirmed it.

In Experiment 10, the subjects received money for each recognised word: 1, 3, or 6 cents, according to the kind of question. The reward had no effect whatever. Even well paid words from the print question did not reach the values of rhyme or category words. The subjects reported that the category words were simply easier to recognise.

Why yes words stuck better

For rhyme and meaning questions, the subjects recognised yes words better than no words, although both answers came equally fast. Craik and Tulving explained this with Schulman congruity. When question and word fit together, they form one unit, and the trace grows more elaborate. A no word lacks this unit.

Craik and Tulving therefore held elaboration and spread for perhaps better terms than depth. Their result: what sets recall is neither the intent to learn nor effort, time, or amount of rehearsal. It is the kind of processing.

What the criticism says

In the Craik and Tulving studies, the core finding survived all controls by the authors: meaning questions led to better retention than print or sound questions. Debate concerns how much the term depth explains, and whether shallow processing truly achieves so little.

• Risk of circularity: Craik and Tulving wrote in 1975 that the approach needs an independent index of depth. Without one, any well remembered event is too easily labelled deeply processed. Time served as their first index, but Experiment 5 showed that time is not depth.
• Three problems after Baddeley: Alan Baddeley judged in 1978 that the approach was not proving productive. It lacked an independent measure of depth. Within sound and meaning, no further levels could be set. Supporting terms such as congruity and elaboration could not be measured independently either.
• Three claims that may be false after Baddeley: rote repetition of sound still produces learning. Shallow processing may leave durable traces. A fixed chain of ever deeper stages is unlikely. Baddeley proposed to drop the levels claim.
• Transfer-appropriate processing: Morris, Bransford and Franks had words judged by meaning or by rhyme in 1977. After a review by Ekuni, Vaz and Bueno, the meaning task won only when the test asked for meaning. When the test asked for rhyme recognition, the rhyme task won. The best processing is the one that fits the demands of the test.

Craik and Lockhart answered in 1990 in a retrospective on their approach. They described how research since 1972 had refined depth, for example with elaboration and distinctiveness of encoding. Their result: many original ideas had survived, and as a research framework the approach had been largely successful. They also wrote that the 1972 paper had caused many misunderstandings about what they meant. In 2002, Craik returned to common objections, including the measure of depth and the link of encoding and retrieval. After the Ekuni review, Craik treated transfer-appropriate processing as a complement, not a refutation.

Self-reference: deeper than meaning?

Rogers, Kuiper and Kirker used the same method with trait adjectives in 1977. Their subjects rated the words in four tasks: structural, phonemic, semantic, and with reference to their own person. In two experiments, they recalled the words from the self-reference task best. The authors called self-reference a rich and powerful encoding process.

A meta-analysis by Symons and Johnson in 1997 confirmed this self-reference effect. Self-reference beat semantic encoding and beat reference to another person. Against reference to another person, the gain was smaller. The authors explain the effect with elaboration and organisation: the self is a well developed construct in frequent use. Self-reference is not one of the levels from 1972 and 1975. It is a later finding with the same method.

What this means for learning psychology terms

New material fades fast without practice, as the forgetting curve shows. Levels of processing shows which kind of work with a new term leaves a stronger trace. The studies used word lists, not psychology terms. The steps below transfer their findings to learning.

1. 1.Process meaning, not only form. Ask which category the term belongs to and in which sentence it fits. These questions match the deep tasks at Craik and Tulving.
2. 2.Connect the term with yourself. Find one example from your own day. This matches the task that won at Rogers and colleagues.
3. 3.Practise in the form in which you face a test. If the test asks for a definition, practise defining. This follows from transfer-appropriate processing.
4. 4.Do not only repeat a definition silently. After Craik and Lockhart, this holds the information briefly. Test yourself instead. Our guide to learning psychology terms explains why retrieval beats a second reading. Spaced repetition spreads that retrieval over time.

Psychology World is a dictionary app with more than 18,000 psychology terms, and it supports this kind of practice. Look up levels of processing and related terms such as elaboration or incidental learning. Save the terms you find in your own list. Practise the list as flashcards, as multiple choice, or by typing. The multiple choice mode needs a list with at least four words. The home page shows how the learning loop works and what each study mode does.