Measurement of Memory

views updated

MEASUREMENT OF MEMORY

There are many techniques for measuring memory. Some methods are highly specialized, either for the purpose of addressing a particular research question or for diagnosing a specific memory dysfunction, but most measurement procedures are variants of the basic techniques described in this entry.

Measures of Recall

Immediate Serial Recall: Memory Span

In the most commonly used version of this task, a randomly ordered sequence of digits (e.g., 4-7-8-2-5-9) is read once to the participant, who is required to repeat them in the same order. The resulting measure is known as the digit span, defined as the number of digits that can be repeated in the correct order without error. An immediate difficulty is that an individual's performance may fluctuate slightly from one occasion to the next, six digits being correctly repeated on one occasion, for example, and seven on the next. The formal measurement of span is therefore usually taken as the number of digits that can be correctly recalled on 50 percent of occasions. The digit span test is found in such standard test batteries as the Wechsler Adult Intelligence Scale. Span can also be measured using items other than digits: letters or words, for example. However, it should be noted that the span measure obtained may vary with the kinds of items employed. Thus word span is typically somewhat shorter than digit span, and word span itself will vary depending on various features of the words, such as their length and familiarity.

Free Recall

In free recall, a list of items is presented and the task is to recall them in any order. The measure of memory is the number (or proportion) of these items recalled. If people are allowed to begin recalling immediately after the presentation of the last item, this simple measure is subject to a strong recency effect: The last few items in the list will be recalled very well (and usually recalled first). There is also a weaker primacy effect: The first few items in the list will be better recalled than those in the middle positions. In some applications these serial position effects are eliminated by inserting "buffer" items at the beginning and end of the list that are not counted as part of the recall measure.

Cued Recall

In cued recall, the task is to recall each item in response to a cue provided by the tester. This cue may have been presented along with the item at the time of study (an intra-list cue) or be an item not studied before (an extra-list cue). For example, a participant may study word pairs such as dog-tail and when tested be given the intralist cue dog and asked to recall the word with which it was paired. Alternatively, he or she may study a list of single words (including tail but not dog) and when tested be given the extra-list cue dog as a potential aid to recall. A special case of extra-list cuing is category cuing. If the list contained items from several different categories (fruits and animals), cued recall could take the form of providing the participant with the category labels as cues.

A potential difficulty in measuring cued recall is guessing. This problem arises when the items to be recalled have strong prior associations with the cue. For example, suppose the word orange was studied and at recall the extra-list cue fruit is provided. The participant may have forgotten the item orange but offer it merely because it is a plausible guess to the cue fruit. The usual solution to this difficulty is to obtain a baseline measure of how often the item is wrongly recalled in response to the cue when it was not on the study list. In this example, the control measure for the word orange would be its recall rate to the cue fruit when orange was not part of the study list.

Measures of Recognition Memory

Forced-Choice Recognition

In this procedure memory is measured by presenting each of the previously studied items (the "old" items) with one or more new items or "lures" and instructing the participant to choose which of these items is old. The measure is then the number or proportion of items correctly identified as old. There are two difficulties with this measure. Guessing poses an obvious problem because in the case of a two-alternative forced choice, someone who remembered nothing at all could guess correctly half the time. Increasing the number of lures reduces the expected rate of correct guessing but does not eliminate the problem entirely. Researchers have proposed various methods of "correcting for guessing," but they are often not used, partly because the problem of guessing is not as serious as it may seem. Normally one is interested not in the absolute number of items recognized but in the comparison of recognition rates across different occasions or conditions. If the influence of guessing is the same across these conditions, then the differences in the recognition rates between conditions will provide an adequate comparative measure.

A second potential source of difficulty is the nature of the lures. A recognition test can be made more or less difficult by altering the degree of similarity between the correct ("old") items and the lures. Thus errors are more likely if the lure is a synonym of the old item than if it is an unrelated word chosen at random.

Single-Item (Yes/No) Recognition Tests

In what may seem the simplest form of a recognition test, participants are shown each test item in turn and asked to respond "yes" if they have seen it before (an old item) and "no" if they have not (a new item). The test list contains a mixture of old and new items. A possible measure of memory would be the proportion of items correctly identified as old, a measure referred to as the hit rate. However, this measure has a serious shortcoming: It will be influenced by the participant's criterion for saying "yes." Adopting a lax criterion (that is, saying "old" even if the item is only faintly familiar) can yield a high hit rate, usually at the expense of mistakenly saying "old" to a large number of new items. Such errors are termed false positives or false alarms. Thus, from the participant's point of view, responding in a yes/no recognition test becomes a trade-off between hits and false alarms. The difficulty from a measurement point of view is that different participants may adopt different criteria, so that comparing hit rates alone can be very misleading.

Clearly what is needed is some way of adjusting the hit rate to take into account criterion differences reflected in the false alarm rate. One method is to take as a score the difference between hits and false alarms, a procedure that has little theoretical justification but offers a simple, and in many circumstances adequate, measure. A more sophisticated method is to make use of a model known as the theory of signal detection, a decision model taken from psychophysics. The model yields a measure, termed d (d-prime), that is independent of the participant's criterion and can be interpreted as a measure of the ability to discriminate difference in subjective familiarity between the old and the new items.

Other Measures of Memory

Measures of Prospective Memory

Prospective memory is remembering to perform an action at a future point in time: to follow an instruction to buy milk on the way home from work, for example. In order to measure prospective memory it is necessary to designate the point at which the future act of remembering is to be performed. In this regard it is possible to distinguish between time-based and event-based tasks. In time-based prospective remembering, an action must be performed at a specified time in the future, such as remembering to turn off the oven in ten minutes. In event-based prospective remembering, an action must be performed contingent on the occurrence of some other event, such as remembering to follow an instruction to lock the door on the way out.

Measures of Implicit Memory

Implicit measures are obtained through observing performance on a task that indirectly reveals the influence of past experience. A common example of an implicit memory test is word-fragment completion. Suppose a participant has seen a list that includes the word assassin. Some time later the participant is given a fragment completion test consisting of some but not all of the letters of the word (e.g., a—a—in) and asked to find the word that could be formed by filling in the blanks. No reference is made to the prior list. The essential aspect of an implicit memory test is that participants receive no instructions to remember items from the prior list, nor are they informed of the list's possible relevance. They may therefore be quite unaware that their performance is being influenced by a past experience. The measure of implicit memory is the improvement in performance on the word-fragment completion task. Improvement is measured relative to word-fragment performance obtained under control conditions in which the solution word (assassin) was not on the prior study list. Among the other tasks that can be used to measure implicit memory is stem completion. In this task participants are given the initial letters of a word and asked to add letters to complete a word. The major difficulty with implicit measures is to ensure that the participant is not making use of explicit memory strategies. One way to achieve this is to prevent participants from becoming aware of the relationship between the initial study and subsequent test phases of the procedure. Another is to instruct participants that they should avoid conscious strategies by responding with the first thing that comes to mind.

The Relationship among the Different Measures

The various measurement procedures described in this entry are not merely alternative ways of estimating a single "true" quantity that could be thought of as the "amount remembered." In this regard the measurement of memory is very different from the measurement of physical quantities such as length or weight. Various techniques for measuring the distance between two points should yield the same result, and any variation should be regarded as reflecting errors of measurement. The situation is quite different in the case of memory measurement. For example, in evaluating memory for an event, a recognition test and a recall test should not be thought of as alternative methods of measuring a single ideal quantity (the "amount" remembered) but as different measures of memory performance on two distinct, although possibly related, tasks. Much research suggests that under many circumstances different measures of memory are quite dissociated or uncorrelated, reflecting the fact that different measurement procedures tap different aspects of the memory system.

See also:IMPLICIT MEMORY; MEMORY SPAN