Lesson 6 of 11 · 18 min

Testing levels of processing

Analyse Hyde and Jenkins (1973) and Craik and Tulving (1975), interpret levels of processing data, and evaluate whether depth of processing is a valid explanation of memory.

What you will be able to do

  • Describe the design and findings of Hyde and Jenkins (1973) and Craik and Tulving (1975).
  • Identify the variables, design and controls of a levels of processing experiment.
  • Calculate and compare percentages recalled for structural, phonemic and semantic conditions.
  • Evaluate the validity of depth of processing as an explanation of memory.

Saying that deeper processing gives better memory is easy. Testing it is harder: you need a way to control how deeply a person processes each word, without them deciding to learn it some other way. The solution, used in every study in this lesson, is the orienting task: a question about each word that forces one level of processing, followed by a memory test the participants did not expect.

History

Hyde and Jenkins (1973)

Aim. To find out whether the kind of processing done on words, and the intention to learn them, affect recall.

Procedure. Participants were presented with a list of words and did one orienting task with each, such as rating how pleasant the word was (semantic), checking whether it contained the letters E or G (structural), or other tasks about the word's form and grammar. Some groups were told they would be asked to recall the words (intentional learning); others were not (incidental learning). Then everyone was asked to recall as many words as they could.

Findings. Participants who rated pleasantness recalled far more words than those who checked letters. Being told to learn made little difference: an incidental group that processed meaning recalled about as much as an intentional group doing the same task.

Conclusion. What determines recall is the depth of processing at encoding, not the intention to remember.

History

Craik and Tulving (1975)

Aim. To test the levels of processing model by controlling the level of processing of each word.

Procedure. Words were flashed one at a time (for 200 milliseconds in Experiment 2, with 24 students). Before each, participants were asked a question that required structural processing (Is the word in capital letters?), phonemic processing (Does the word rhyme with WEIGHT?) or semantic processing (Would the word fit the sentence "He met a ___ in the street"?). Half the answers were "yes" and half "no". Participants then took an unexpected recognition test, picking out the words they had seen from among new ones.

Findings. Recognition was poorest for words given structural questions, better for phonemic, and best for semantic. In Experiment 2, recognition of words given a "yes" answer rose from 15% after case questions to 81% after sentence questions, more than five times higher. Words given "no" rose less steeply, from 19% to 49%, so at the deeper levels "yes" words were remembered much better.

Conclusion. Deeper processing produced better memory, as the model predicts. But the "yes" advantage, and later experiments in the same paper, suggested depth alone was not the whole story.

Interactive

The three levels and their results

How well words processed at each level were recognised in Craik and Tulving (1975).

Reading the design. In Craik and Tulving's experiment each participant answered questions at all three levels, so it is a repeated measures design.

  • IV: the level of processing required by the orienting question (structural, phonemic or semantic).
  • DV: memory for the words, operationalised as the proportion (or percentage) of words correctly recognised.
  • Controls: the same exposure time for every word; words at each level matched for length and frequency; question types mixed in a random order so no level was always first or last; an unexpected test so participants did not choose their own strategy.

Hyde and Jenkins used independent groups: each group did one orienting task, so no one's processing of one task could carry over into another.

Why is it important that participants in a levels of processing experiment do not know their memory will be tested?

Choose one answer

A class replication. A class of 20 students each read 30 words with an orienting question before each: 10 structural, 10 phonemic and 10 semantic, in a random order. After a 5-minute filler task they wrote down every word they could recall. The table gives the class totals.

Words recalled by a class of 20 students (10 words per level per student; illustrative class data).
Level of processingWords presented (total)Words recalled (total)
Structural20034
Phonemic20062
Semantic200128

Calculate the percentage of semantically processed words that the class recalled.

in %

Worked example

Analyse a repeated measures levels of processing experiment

  • Complex unfamiliar
  • Technology active

Eight students each processed 10 words structurally (Is it in capitals?) and 10 semantically (Is it a kind of food?), mixed in a random order, then did an unexpected free recall test. Words recalled (out of 10):

Student12345678
Structural23124231
Semantic76857696

A paired t-test gave t(7)=8.42t(7) = 8.42, p<0.001p < 0.001.

(a) Calculate the mean recall in each condition. (b) Justify the use of a paired t-test. (c) Draw a conclusion about the hypothesis that semantic processing produces better recall than structural processing. (d) Evaluate whether these data show that depth of processing causes better memory.

4 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Mean = total ÷ number of participants.

  2. Repeated measures or matched participants → paired t-test; independent groups → unpaired.

  3. Compare p with 0.05 and state the decision about the hypothesis, limited to the sample.

Watch out for

  • Using an unpaired test for repeated measures data.

  • Saying the result "proves" the levels of processing model.

Evaluating the validity of depth of processing. Craik and Tulving's own experiments, and the critics who followed, raised four problems.

  1. Circularity. There is no independent measure of depth. If deep processing is whatever produces good memory, and good memory is explained by deep processing, the argument goes in a circle (Baddeley, 1978).
  2. Time and effort. Semantic questions may take longer to answer, so deeper might just mean longer. Craik and Tulving tested this with a shallow task that took longer than a semantic one (deciding the pattern of consonants and vowels in a word): memory was still poorer after the shallow task, so time alone does not explain the effect.
  3. Elaboration, not just depth. At the deeper levels, words that got a "yes" answer were remembered far better than "no" words (81% against 49% after sentence questions), and words in a more complex sentence frame were remembered better than in a simple one. How richly a word is linked to other information (elaboration) matters as well as its level.
  4. The test matters. When the memory test asks which words rhymed with a cue, words processed phonemically are recalled better than words processed semantically (Morris, Bransford & Franks, 1977). Memory is best when the processing at encoding suits the retrieval task, not always when it is "deepest".

Note

Open to interpretation

The same pattern of results (semantic > phonemic > structural) can be explained by depth, by elaboration, by distinctiveness or by the match between encoding and the test. This is a good example of the syllabus point that memory investigations cannot always give a definitive answer: the data are clear, but what they show about memory is still debated.

Explain why the levels of processing model has been criticised as circular.

Write it out as you would in an exam. Put maths between dollar signs, like $x^2 + 1$. Worth 1 mark.

0 / 4000

Summary

  • Orienting tasks control the level of processing; an unannounced test (incidental learning) stops participants using their own strategies.

  • Hyde and Jenkins (1973): rating pleasantness (semantic) gave much better recall than checking letters (structural); the intention to learn made little difference.

  • Craik and Tulving (1975): recognition was best after semantic questions, then phonemic, then structural ("yes" words: 15% after case questions, 81% after sentence questions); at the deeper levels "yes" words beat "no" words.

  • Depth is hard to measure independently (circularity); time alone does not explain the effect; elaboration and the match with the test also matter.

  • Report recall as a percentage of words presented, and compare conditions with values from the data.

Finished the lesson?

Try the 3 quick checks above and the lesson completes itself, or mark it complete now.

Lesson 6 of 110% read

Save your progress with a free account