of Draft March 21, 2006
Page 1
Chapter 8. MEASURING MEMORY
The first successful measurement of human memory, accomplished by Hermann Ebbinghaus in 1885, was a breakthrough achievement in psychological research: it showed that psychologists could study phenomena that seem inextricably bound up in subjective experience. In remembering a previous experience, people are often focused inward on the memory, rather than outward on the presentation of a stimulus (as is characteristic of measurements of sensitivity or of mental processing time). As a result, memory seems for many people to be a completely subjective experience. In order to measure memory, Ebbinghaus objectified it by asking What does a memory allow us to do? and then measured products of our use of memory. The success of this approach has led to its adoption as a general strategy for measuring a full array of apparently subjective psychological characteristics, from personality, attitudes, beliefs, and intelligence to perceptions, dreams, and feelings. Ebbinghaus described his approach in a slim monograph titled Uber das Gedachtnis (translated from the German as On Memory), and his analysis still serves as a model of Hermann Ebbinghaus (1850-1909) careful psychological experimentation. Remarkably, the data he reported all came from testing his own memory; however, his methods were readily applied to investigating the memory of others, and similar results were found. In the years since Ebbinghaus solved the problem of measuring memory, psychologists have developed a much better understanding of how memory works. Contemporary theories of memory now make precise, quantitative predictions about a wide variety of characteristics of memory-based performance. Moreover, the advent of brain imaging techniques, which allow visualization of activity in the brain as people carry out memory tasks, has renewed efforts to localize memory in particular parts of the brain. Indeed, brain stimulation studies have occasionally suggested that some memories can be elicited by stimulating a particular section of brain tissue.
Draft of March 21, 2006
Page 2
This chapter provides an overview of the measurement of memory, beginning with the work of Ebbinghaus and continuing to a short review of contemporary measures.
Ebbinghaus approach to memory. Ebbinghaus approach to memory began with his focus on what memory allows us to do, rather than on what memory allows us to experience. Having some memory of a prior event, even if indistinct, allows us to learn faster and to respond more quickly and more accurately. Having a memory also alters the way we react to events that have a connection to that prior experience (as illustrated by our rating someone to be familiar to us if we have met them previously at a party). Studying memory experimentally requires 3 components: a set of stimulus materials, a procedure for presenting these materials so they can be remembered, and a measure of memory. For his research, Ebbinghaus developed each of these components himself, inventing the CVC syllable as a type of stimulus material, serial list learning as a procedure, and the savings score as a measure of memory. Briefly mentioned in chapter 4 on graphing, each is worth describing in more detail. The CVC syllable is a 3 letter combination in which the initial and terminal letters are consonants and the middle letter is a vowel (including y), such as byq, caj, dor, and dog, cat and rat. That some CVC combinations are words underscores an important fact about Ebbinghaus invention that is often overlooked: the primary significance of the CVC for Ebbinghaus was that it allowed him to create stimulus items in a mechanical fashion (by creating all possible combinations of the initial consonant set, the vowel set and the terminal consonant set) so that he could then create stimulus lists that were meaningless. Serial list learning is a procedure which requires memorizing a list of items presented in a fixed order. Anyone who has to memorize lines for a play or movie, or passages of poetry for recitation, is engaged in serial list learning. From Ebbinghaus perspective, the critical ingredient in serial list learning is new learning, which involves learning the succession from one item to the next, regardless of how familiar or unfamiliar individual items are. As a basis for deciding how much time to spend on a list, Ebbinghaus studied each list until he achieved one perfect recitation of it, which is an example of learning to a criterion (as contrasted with learning over a fixed
Draft of March 21, 2006 number of repetitions).
Page 3
The savings score compares the time needed to re-learn a given list to the time needed to learn that list initially, on the idea that any residual memory of the initial presentations will be apparent in a savings in the time needed to re-learn the list. The formula for calculating savings is shown in Equation 8-1: Eq. 8-1 Percent Savings = ((TOL - TRL) / TOL) x 100% (where TOL = Time for Original Learning and TRL = Time for Re-Learning). This formula expresses the amount of time saved during re-learning, compared to the time for original learning, as a percentage of the time for original learning. For example, if original learning required 120 s and the re-learning required 40 s, the savings would be (120 - 40) / 120 x100% , which becomes 80/120 x 100% or .667 x 100% = 66.7%, meaning that re-learning took place with a savings of 67% of the time needed for original learning. The special importance of the savings measure for Ebbinghaus is that it provided him a way to measure his memory without having to cue himself about which list to recollect for any given test. Some of Ebbinghaus results (previously presented in Table 6-2 and Figure 6-8) are presented in Table 8-2 and Figure 8-2:
Table 8-2. Ebbinghaus data on retention.
a
Retention Interval (hrs)
Mean Percentage Savings rounded from 8.75 hrs
0 100
.3 58.2
1 44.2
9a 35.8
24 33.8
Figure 8-2. Ebbinghaus forgetting function
Draft of March 21, 2006
Page 4
After Ebbinghaus. For the generation of American psychologists that followed Ebbinghaus, the study of memory was overshadowed by the emergence of studies of learning and conditioning. This emergence began with Thorndikes analysis of animal intelligence and his formulation of the Law of Effect as a general explanatory principle of the learning process, and it was followed by reports of Pavlovs discovery of principles of classical conditioning (both topics are discussed further in Chapter 11). With a few notable exceptions, like Sir Frederic Bartletts Remembering, published in the 1930s, studies of memory did not attract much research interest until information-processing models of cognition appeared in the 1950s and 1960s. These models, based on an analogy between human cognition and the processing of information by digital electronic computers (Simon & Newell, 1972), highlighted the importance of human memory because of the critical role of memory in computer operations. The increased attention to human memory resulted in a major re-conceptualization of it. Rather than being viewed as a single faculty, memory was described as consisting of multiple components, each with distinctive characteristics and requiring different tools for its investigation. The modal model of memory of the 1960s and 1970s, for instance, included 3 types of memory: iconic memory, short-term memory and long term memory, each with a different representational format, storage capacity, and persistence. This new conceptualization shifted the focus of research from memory for an entire list to memory for the individual items in the list; and from memory following mastery of a list to memory following only one or two repetitions. With these shifts in focus, researchers began to use meaningful items, such as words and identifiable pictures, rather than CVC syllables that had to be learned before they could be remembered as units. Using meaningful materials as stimuli also permitted researchers to rely on recall as a measure of memory. Thus, contemporary studies of memory tend to use stimulus materials, procedures and memory tests different from those invented by Ebbinghaus. The varied nature of newer approaches is exemplified in the research of Endel Tulving, a pioneering investigator who has highlighted important distinctions among different forms of memory and has developed methods to measure the different forms. For example, he has distinguished among anoetic (non-knowing, or procedural, memory, which is evident in changed
Draft of March 21, 2006 reactions to stimuli, based on past experience with them), noetic, (knowing, or semantic memory, which is evident in the facts, general information and other knowledge one can express) and autonoetic (selfknowing, or episodic memory, which is evident in the remembrance of specific experiences, including memory for the time and context of the experience). Among the methods Tulving has proposed to measure each of these forms of memory are priming methods to measure procedural memory (e.g., Schacter & Tulving, 1995) and rememberknow discrimination methods to measure separately semantic and episodic memory. The next sections will describe examples of each of these methods.
Page 5
Endel Tulving (19xx - )
Measures of memory: Priming . Priming refers to enhanced performance on a task as a consequence of prior exposure to the task or at least to some parts of it. Usually, the task itself can be done in the absence of this prior exposure, but it can be done more quickly, more accurately, or more predictably following the exposure. Tulving (e.g., Tulving, Shacter & Stark 1982) introduced the word fragment completion task as one method to obtain priming effects. For this task, a participant is given a word fragment in which some of the letters of the word are replaced by blanks, as in a _ _ a _ _ i n , with the task being to fill in the blanks with letters to make a word (in this case, assassin). In the study by Tulving and his colleagues, participants were asked to study a list of 96 words for a subsequent memory test (the type of test was left unspecified). Testing was carried out either 1 hour or 1 week later, and the test consisted of 48 word fragments, of which half were based on words that were studied (old) and half were based on words that were non-studied (new). Tulving et al. found that, on average, their subjects completed 56% of the fragments based on studied words but only 42% of the fragments based on non-studied words. The difference of 14% (56 - 42 = 14) is an example of a priming effect. Tulving and his colleagues have suggested (e.g., Tulving & Shacter, 1995) that priming effects might be a direct measure of procedural memory, because they are found even for people who suffer from anterograde amnesia, which is a kind of mental impairment caused by disease
Draft of March 21, 2006
Page 6
or accident in which a person is unable to remember new information for more than a few tens of seconds. Many people with anterograde amnesia show priming effects that are more or less the same as those for people without this impairment. The fact that someone can show evidence of memory (by showing priming effects) but have no experience of remembering the events that produced the priming is an example of a dissociation between one aspect of memory and another, and dissociations are frequently used as evidence for the separation of components of memory. Subsequent research has shown that, in general, priming is not a pure measure of procedural memory, because priming can occur when people remember their previous experiences as well as when they cannot. However, an interesting technique invented by another researcher, Larry Jacoby, does seem to provide a way to extract a measure of procedural memory from priming (and other memory tasks). This technique, called the process dissociation procedure, will be described later.
Measures of memory: Remembering, knowing the past, recall and recognition. A priming score, like the savings score developed by Hermann Ebbinghaus, is an implicit memory measure that reveals the persistence of memory without requiring an explicit declaration that a current task is related to a past experience. The ability to say I remember this is a hallmark of explicit memory measures. Such measures are very familiar to students, who encounter them in the form of different types of exam questions, all designed to determine how much information students remember from past experiences the in class. Explicit memory measures commonly occur in one of two forms, either as a recall task (as with essay, short answer, or fill-in-the-blank questions, that all require reproducing past information) or as a recognition task (as with multiple choice or matching questions, that all require identifying whether information on the test was presented in the course). In recall tasks, people are asked to reproduce information from memory. Because adults are well practiced in speaking and writing (or typing) words, laboratory studies of memory often present words as stimuli and ask research participants either to recite or to write down all the
Draft of March 21, 2006
Page 7
words they remember during test periods. Recall can be carried out to comply with different kinds of instructions, but three types of instruction are commonly used, creating three types of recall test: free recall, serial recall, and cued recall. In free recall, participants are asked to
recall as many studied items as possible, without regard to the order in which items were presented; for example, a responder might write the last word presented, then the first word, then a word from the middle of the list, and so on. In serial recall, responders recall items in the order in which they were presented. In cued recall, responders are given explicit cues and are asked to recall items that go with the cue. In recognition tasks, people are asked to indicate whether or not some given information is remembered from a prior experience. Because recognition tasks present a test item, rather than ask for a re-production of an item as in the recall task, laboratory studies of recognition include a wider variety of materials than do studies of recall. Verbal materials are very common, because they are easy to present for testing as well as for studying, but music, pictures, and other kinds of stimulus materials are found much more often in recognition studies than in recall studies. Additionally, recognition can be tested in a wider range of subjects, including animals and young children. Recognition tasks are classified in terms of the way subjects respond: yes/no recognition tasks ask subjects to indicate for each single item whether it is old (previously presented in the experiment), or new (not previously presented in the experiment), whereas forced choice recognition asks subjects to select which item, from a set of 2 or more items, was presented previously in the experiment. Regardless of whether explicit memory is measured with a recall task or a recognition task, the basis for responding is linked sometimes to a clear and detailed memory of a particular moment in ones past and at other times only to general knowledge or to a nonspecific feeling of familiarity. Tulving (1985) showed that people could categorize judgments based on memory as either a remember (R) judgment, based on memory of a specific moment, or a know (K) judgment, based on a feeling of familiarity or general knowledge of what had happened in the past. He suggested that these judgments also provided a way to distinguish between semantic and episodic memory, using Know judgments to index semantic memory and Remember judgments to index episodic memory. The utility of the Know/Remember paradigm has been
Draft of March 21, 2006 demonstrated repeatedly in contemporary research.
Page 8
Measures of Memory: General features. Psychologists have been measuring memory for more than a century, and this research has clearly established several characteristic features of memory-based performance: the bowed serial position curve, the spacing effect, the list length effect, and the levels of processing effect. Bowed serial position curve. When people are presented with information in a list format, their memory for individual items in the list depends on the items position within the list, or serial position. In free recall, a task in which people recall information from a list in any order they choose, the characteristic result of asking for recall immediately after presentation of the information is a bowed serial position curve, illustrated in Figure 8-3 for a 15-item list. Information at the beginning of the list (with serial positions 1, 2, 3, etc.) and information at the end of the list (serial positions 13, 14, 15) is recalled most consistently, whereas information from the middle sections of the list (serial positions 5 to 11, say) is recalled less well.
Figure 8-3. Bowed serial position curve for 15-item list. The first item in the list occupies Serial Position 1 and the last item Serial Position 15.
Draft of March 21, 2006
Page 9
Measures of Memory: Recognition. When we recognize someone, we experience a feeling of familiarity which may be accompanied by emotional responses ranging from joy to dread, and we may also be reminded of specific moments from our past experiences. Although this complex mixture of subjective experiences is not readily measurable, the fact of recognition allows us to discriminate between what is familiar and what is novel. Thus, if we apply to recognition memory the strategy for objectification that Ebbinghaus applied to memory for lists of items and ask what recognition memory can be used for, we are led to focus on how well people can distinguish between information that has been previously encountered and information that is new in some way. The ability to distinguish between old and new can be measured in several ways. The simplest way is to present a set of items for study then test with a mixture of items that were not previously studied (New) and items that were studied ( Old). Better recognition will be shown by higher levels of recognition responses to Old items coupled with lower levels to New items; conversely, poorer recognition will be shown by fewer recognition responses to Old items as well as more responses to New items. It is important to note that the critical ingredient for measuring recognition consists of the combination of responding Old to old items and New to new items (simply responding Old to every test item leads to high levels of errors on the new items, and responding New to every test item leads to high levels of errors on old items). The measurement problem for recognition memory is very similar to that discussed in Chapter 4 in regard to measuring sensitivity, and the problem gives rise to a similar solution. As in the measurement of sensitivity, we can create a 2 x 2 matrix, with rows defined on the basis of the nature of the event presented and columns defined on the basis of the response given. In this case, however, the presented events are either new items or old items, and the responses are either New or Old. Table 8-3 shows the resulting organization. The rows of the table represent the two possible stimulus conditions of New Item or Old Item, the columns of the table represent the two possible responses given by the subject of New or Old, and the cells of the matrix represent the 4 possible combinations of stimulus condition and response. Consider the top row, which is for trials on which new items are presented. If on such trials the subject responds correctly that
Draft of March 21, 2006
Page 10
the item is New, the response is designated a Correct Rejection. If the subject incorrectly responds Old, the error is designated a False Alarm. Next consider the second row, which is for trials on which old items are Table 8-3 Test Condition New Item Old Item Response New Correct Rejection Miss Old False Alarm Hit presented. If the...