10 Lab 10. The Generation Effect: Don’t Let Someone Else Do Your Work

Classroom demonstration

Procedure

Depending on which lab section you are enrolled in, your lab instructor might have given you information about this experiment the week before this lab is discussed in class. To be prepared, all students should please bring four sheets of paper, labeled “List 1, “List 2,” etc.

All students will receive additional instructions at the start of the lab period.

STOP READING AT THIS POINT AND PARTICIPATE IN THE EXPERIMENT. PLEASE DO NOT READ ANY FURTHER UNTIL THE EXPERIMENT IS OVER.

Introduction

When studying for an exam, students often produce things like study guides or outlines. These can be very helpful in preparing for the test. Sometimes, these well-meaning students pass out their study guides to fellow classmates, in hopes that their friends may be able to benefit from their work. Unfortunately, this seldom yields the intended benefit. Why doesn’t studying someone else’s study guide help? The answer lies at the heart of this lab– the generation effect.

Thought Experiment

Try this with two of your friends. Tell one of them you are collecting data for a psychology experiment, and as part of it, you need to have them make a list of 10 cars made by General Motors (Chevy, Pontiac, and Buick). Have this first friend come up with a list while you time them. After you have the list of 10 cars, give that list to your second friend, and let them study it for as much time as it took the first person to generate it. In a day or so, give both friends a surprise recall test. Ask them to try to recall all the words on the list. Which of your friends do you expect to recall more items?

Similar to the effect we see with study guides, some students specifically look for used books in which the previous owner has done extensive highlighting or outlining in the book. They figure that person has already done the “hard part” for them. They are likely to be disappointed with their grade in the course.

Many students come to my office and ask me how to study for the exam. I tell them the kinds of things you are used to hearing: read the book several times, take notes, make outlines, highlight while you read. Students usually find it frustrating when I tell them that they shouldn’t bother studying the outlines they just worked so hard on generating (well, at least not exclusively). Why would I give them such advice? Because the important part, with respect to learning the material, was not in having the lists or outlines, but in making them.

Slamecka and Graf (1978) performed the first experiment investigating the generation effect. They had two learning conditions in their experiment. Some subjects read lists of paired-associate words, like DOG-CAT. Later, they were given the word DOG as a cue, and were asked to come up with the paired-associate.

In a second condition, subjects were given the first word in the pair, and the first letter of the second word, like DOG-C__. Their task was to generate the appropriate associate. Like the first group of subjects, they were later given DOG as a cue, and asked to recall CAT (or whatever word they had generated). Somewhat to Slamecka and Graf’s surprise, the subjects in the second condition–those who generated the associates rather than studying them–performed better on the cued recall test.

The exact cause of the generation effect is still being debated. Some researchers provide an explanation similar to a levels-of-processing theory: generating words (rather than simply reading them) promotes deeper processing, and deeper processing leads to better memory. Others favor an explanation that says the generation directs more attention to the task, and this increased attention leads to better encoding. It’s also likely the case that generating items is somewhat more interesting than reading them, and as you well know, there is a strong relationship between interest and memory.

Using the Generation Effect to Test for Differences in Memory Systems and Memory Processes

We have just seen that the generation effect improves memory. However, this is not always true. Performance on some types of memory tests, called indirect tests of memory, or tests of implicit memory24, does not improve when the items are generated. Instead, performance on these tests is enhanced when items are simply read in isolation, rather than generated in response to a cue. This difference might not sound important, but it truly is. A great deal of current work in memory theory hinges upon explanations of why the generation effect is reliably obtained in tests of recall and recognition, but not on tests of implicit memory.

Direct and Indirect Tests of Memory

Direct Tests of Memory

In Lab 1 we talked about how most dependent variables measure either speed or accuracy. For accuracy measures, the two most common tests are recall tests and recognition tests. In fact, almost every test you take in college is one of these two types. Essay tests measure your ability to recall information, multiple choice tests measure your ability to recognize information. These are often referred to as direct tests of memory.

Indirect Tests of Memory

In the past decade or so, though, new tests of memory have been used, called indirect tests or tests of implicit memory. They are so called because we have found that your performance on these tests does not depend on your conscious awareness of having seen the information before. In other words, your performance might be enhanced even though you are not aware of it.

This might sound odd, but hold on. Freud realized long ago that there are often unconscious effects of memory on behavior. He proposed that not every memory is available for conscious introspection, yet those memories can still influence our lives greatly. Don’t believe it? Consider this: How important are the first 5 years of your life, in terms of shaping what the rest of your life is like? Without a doubt, the first five years are the most important. Yet, most of us have only a few fleeting memories of this entire 5-year period!

Is it fair to say, then, that we are completely amnesiac for this period? Not really. A better way to say it would be that we have few conscious memories of this period. We might have a fear of heights that we don’t understand, yet it is a real fear. Sometimes these fears can be traced to learning experiences in childhood, where we had a frightening experience involving heights (we might have had an Uncle Buck, for example, who liked to fling us in the air whenever he saw us, but seldom caught us on the way down). Watson’s classic case of Little Albert and his fear of the white rat (and later other white or furry things) is a prime example.

The onset of conscious memories in children is a fascinating topic in and of itself (see Bruce, Dolan, & Phillips Grant, 2000; Fivush & Schwarzmueller, 1998; Usher & Neisser, 1993, for thorough discussions). But for present purposes, we are more interested in seeing how these indirect tests of memory can reflect learning and memory when direct tests like recall and recognition cannot.

Before discussing how the generation effect (or lack thereof) can offer insight into these questions, let’s review a few kinds of indirect tests of memory.

1. Repetition Priming Effects are simply quicker (or more accurate) responses to later presentations of items. Let’s take an example where subjects are shown a picture of an object and asked to name it. If the subject can name the object in 900 msec the first time the picture is shown, the same subject will usually name the same object more quickly if it is shown a second time. What makes this an indirect test of memory? The repetition priming persists even when an individual has no conscious recollection of having seen the item before. Even if the person denies having seen the object in a prior study, the object will be named more quickly the second time. This is how H. M. can show improvement on certain tasks over time, though he will report having never performed the task before.

2. Enhanced Fragment Completion tests are another way to measure implicit memory. If I ask a group of subjects to try to solve word fragments (like those seen on Wheel of Fortune):

G R _ S _

F _ V _ R

_ S S _ S S _ N

I might find that, on average, people can solve 35% of these fragments. However, if prior to the fragment completion test, I have them study a list of items which include the words GRASS, FEVER and ASSASSIN, I find that they do even better–they might get 60% of them instead of 35%. Again, this is true whether or not they have a conscious memory of studying the items previously.

3. Biased Homophone Spelling tests are a third way to measure implicit memory. These tests make use of homophones–words that sound alike but have different spelling, like “night” and “knight,” “horse” and hoarse,” and “bow” and “beau.” In most of these homophones, there is one dominant spelling of the word. If I read these words aloud, and asked people to spell them, I’d almost always get the spellings: “NIGHT”, “HORSE” and “BOW.” However, if prior to the spelling test I bias the less dominant meaning in some way–for example, I might tell you a story which has dragons and castles and jousts and kings, and then ask you to spell “NIGHT” (I’d have to give you the word in spoken format, of course). In this case, I can make it much more likely that you’ll spell it “KNIGHT,” and that’s true even if you have no conscious memory of that story25.

4. Speeded Perceptual Identification; tests involve rapidly presented letter strings, which subjects are asked to identify. These items are presented so quickly that subjects usually identify relatively few items. For example, I might flash the word “building” on the computer screen so quickly that only 10% of my subjects can correctly identify it. However, if they have seen the word previously, they are much more likely to identify the word–maybe 50% correct instead of 10%. Once again, this is true whether or not there is any conscious recollection of having seen the word previously.

We are going to study the generation effect in the context of this last test, and the work of Larry Jacoby and colleagues (Jacoby & Dallas, 1981). Jacoby and Dallas had subjects learn words under one of three encoding conditions. Some subjects read words in isolation, called the “Read Only” encoding condition. In our example, they were shown the word CAT in isolation. A second group of subjects was shown the same target word, CAT, but that word was presented with a semantically associated word, “DOG.” Thus, subjects saw DOG->CAT, but were told that the second word CAT was the only word they would be asked to remember. This encoding condition was called the “Read-Associate” encoding condition, since subjects read the word in the presence of a semantic associate. A final group was given the semantic associate DOG, and was asked to generate the appropriate associate, when they were given the first letter, DOG -> C__. This was called the “Generate Condition.” These encoding conditions are reviewed in Figure 10.1.

Read-Only Encoding Condition:

___ -> CAT

Read-Associate Encoding Condition:

DOG -> CAT

Generate Encoding Condition:

DOG -> C__

Figure 10.1 Demonstration of the Encoding Conditions used in Jacoby & Dallas (1981). (The Target Word “CAT” is shown in boldface.)

Jacoby and Dallas used two different types of memory tests, one a direct memory test (recognition) and the other an indirect memory test (speeded perceptual identification). In the recognition test subjects were given target words (like “CAT”) mixed in with some distracter words (like “BOOK”) and asked to identify those that they recognized from the encoding trials. In the speeded perceptual identification test, subjects were presented words very quickly, and asked to identify them. Some were the target items from the encoding test (“CAT”) and others were new items (“GLASS”). The degree to which subjects identified the target words better than the new words is called the “priming effect.” Larger priming effects mean that subjects were more likely to identify the target words in the perceptual identification test.

The direct test of memory, the recognition test, displayed the standard generation effect, and is shown in Figure 10.2. Items in the generate condition were better recognized.

Figure 10.2 Performance on Recognition Tests as a function of the Encoding Condition, from Jacoby & Dallas (1981)

Why should the encoding conditions produce such different results? Why should the Read-Only Condition lead to best performance in a speeded perceptual identification test? Though the matter is still being debated, Roediger (Roediger, 1990; Roediger, Gallo, & Geraci, 2002), among others, maintains that it has to do with the properties of something known as Transfer-Appropriate Processing (TAP). I like to view TAP as being a combination of the Encoding Specificity Principle (Lab 9) and the Levels of Processing (also discussed briefly in Lab 7?). The LOP framework, basically says that memory will be best when information is processed deeply, or meaningfully. The Encoding Specificity Principle says that performance will be best when encoding conditions match retrieval conditions. TAP combines the two in the following way: it says that “depth” of encoding really should be determined relative to the retrieval task. Sometimes, a rhyming encoding condition (usually thought of as an intermediate “depth”, when the retrieval test is recall or recognition) might lead to the best performance. Not surprisingly, this would happen if the later test also involves rhyming. When might an encoding condition emphasizing perceptual processing lead to best test performance? When the test also involves perceptual processing.

Roediger’s TAP explanation works as follows: the “Read-Only” encoding condition emphasizes perceptual encoding–because no semantic associates are present, the subjects attend primarily to the perceptual features. If the retrieval task also involves perceptual processing–as it does in a speeded perceptual identification test–then memory will benefit from a “Read-Only” encoding process.

However, if the retrieval task demands conceptual, semantic processing, as in a recall or recognition test, the “Generate” encoding condition will lead to best test performance. Intermediate encoding conditions, like the “Read-Associate” condition, lead to intermediate performance on each task, since task involves both perceptual and conceptual encoding.

Though it is well beyond the scope of this lab, Roediger (1990) combines these principles in a very powerful and convincing alternative explanation to Tulving’s Multiple Memory System model. Who is right? Ask me in a decade or so (and even then, I probably won’t give you a straight answer).

The title of this lab encourages you to “Do you own work.” This is true, to the extent that the later retrieval task will demand conceptual, semantic retrieval. If you ever have a final exam where you are given word fragments to complete, you might want to study differently26…

Whatever the underlying cause of the generation effect, it is a real and strong phenomenon. Though it is tempting to read someone else’s notes from class instead of attending and taking them yourself, or to read their summary of a chapter rather than producing your own, resist that temptation.

Questions for Lab 10

1. What are the independent and dependent variables in this experiment? What things did we control?

2. Graph your results (individual and group data). Did we obtain a generation effect?

3. Why do you think generation leads to better memory?

4. In some experiments (i. e., Jacoby & Witherspoon, 1982), it has been shown that reading (rather than generating) leads to better memory. This is true only when the task is what is known as a speeded perceptual identification task. In these kinds of tasks, a word is flashed to a subject on a computer screen. The word is flashed so fast that much of the time, the subject cannot tell what word was just presented. In these kinds of tests, Jacoby and Witherspoon found that subjects who read the words did better than those who generated them. Explain how this be reconciled with the standard generation effect. (Hint: look back to the encoding specificity lab).

5. Why did this experiment use high-probability word pairs, like DOG-CAT and BANK-MONEY? Would you expect similar results with less-frequent word pairs? What would be some problems with using less frequent pairs?

6. Under what kinds of conditions might the Read-Associate encoding condition lead to the highest performance?

7. Some students claim to study differently for multiple-choice tests than they do for essay tests. Given that you have now studied about Encoding Specificity (Lab 9) and Generation Effects (Lab 10), what would you tell your fellow students about studying this way?

Data Sheet for Lab 10

The Generation Effect

Name:

Individual Condition:Read OR Generate

(Circle One)

Report Mean Percent Recalled:

Note:

Since the order of presentation of the list types was varied between subjects, the order in which the list types are listed on this table may not be the order in which you received the presentation of the list.

Graphs for Lab 10

The Generation Effect

Individual Data

Individual Condition:Read OR Generate

(Circle One)

NAME:

The Generation Effect

Group Data

NAME: