Research Report - Psychological Sciences - Purdue University

PSYCHOLOGICAL SCIENCE

Research Report THE LOSS OF POSITIONAL CERTAINTY IN LONG-TERM MEMORY James S. Nairne Purdue University Abstract—Long-term memory for sequential position was examined following retention intervals that ranged from 30 sec to 24 h. The Estes perturbation model (1972) is shown to provide a reasonable qualitative fit of the dynamics of forgetting, even though the model was designed to account for the phenomena of immediate memory. Similarities among the forgetting processes ofshortand long-term memory are considered. In studies of immediate memory for order, the ability to position an item correctly in a short list declines rapidly with the passage of time. After only a few seconds of distraction, subjects' response distributions resemble generalization gradients anchored around the true serial positions. Response likelihood peaks at the correct position and declines gradually as distance from the true value increases (Healy, 1974; Jahnke, Davis, & Bower, 1989; Lee & Estes, 1977, 1981; Shiffrin & Cook, 1978). Estes (1972) has characterized these functions, or uncertainty gradients, with equations of the following form. The probability that an item will be remembered as occupying a particular interior position / at time n + 1 is: (e/2)X,- _ (1)

where 6 is the probability that the subject's position memory will undergo a perturbation and adopt a neighboring position value. The net result is that subjects are likely initially to place items in nearby positions when incorrect, but over time the position functions flatten and distant placements become more likely. For items occupying the endAddress correspondence to James S. Nairne, Department of Psychology, Purdue University, 1364 Psychology Building, Room 3152, West Lafayette, IN 47907-1364, E-mail: NAIRNElffiBRAZIL.PSYCH.PURDUE.EDU. VOL. 3, NO. 3, MAY 1992

points of the position dimension (first and last in the list), the equations are slightly different: for example.

tion should show a characteristic form, and the uncertainty curves should resemble generalization gradients that flatten with the passage of time. METHOD

Uncertainty about the endpoint values increases more slowly because response position cannot perturb beyond the dimension boundaries. In this case, better performance is found for the primacy and recency items and the serial position curve is generally bow-shaped. In the initial application to immediate memory, perturbations in the mnemonic representations of position were argued to arise from the rehearsal-based recodings that are likely to occur throughout the retention interval (Cunningham, Healy, & Williams, 1984; Estes, 1972; Lee & Estes, 1977, 1981). However, recent work from this laboratory (Nairne, 1990, 1991) has shown that the perturbation equations can apply to the loss of position information in general, even under conditions in which rehearsal-based recodings are unlikely to occur. Subjects in these experiments were asked to make pleasantness ratings about five words in each of five lists. After a 10-min distraction period, a surprise reconstruction task was administered that required subjects to place the rated items in their original positions of occurrence. The resulting serial position curves and positional uncertainty gradients were fit quite well by the perturbation model, without any changes in its basic assumptions, even though the retention was clearly long-term. The current report continues this application by examining the loss of positional certainty across a variety of retention intervals, ranging from 30 sec to 24 h. In addition to providing basic data on the forgetting of position information over a time-course of hours, the intent was again to make at least qualitative comparisons with the predictions of the perturbation model. The forgetting func-

Subjects The subjects were 144 undergraduates who participated for credit in an introductory psychology course. The experimental sessions were conducted in small groups. Materials and Design The experiment employed 25 words, arranged sequentially in five lists of five words. The stimulus items were medium- and high-frequency nouns, four to six letters in length, taken from the Paivio, Yuille, and Madigan (1968) norms. Assignment of the words to lists and positions within lists was determined randomly by sampling without replacement from the overall pool of 25 words. All subjects received the same words presented in the same presentation orders. Retention interval was manipulated between subjects; each retention group contained the same number of participants (A^ = 24). Procedure Subjects were asked to make pleasantness ratings on a scale ranging from I (unpleasant) to 3 (pleasant) for each of the 25 words as it was presented. The words were presented aloud on a tape recorder and 2.5 sec separated the onset of each item. Each list of five words began with the word "ready" and ended with a 5-sec blank interval. Subjects were not informed about the subsequent retention test, nor were they given any information about why the words were grouped in lists. They were simply led to believe that they were participat-

Copyright © 1992 American Psychological Society

199


Positional Uncertainty ing in a speeded ruting task. Subjects wioli: their ratings on a response sheet containing five rows of five bkmks; one row was designated lor each list. Following the rating task, subjects in five of the retention conditions were excused and told to return at a designated time for further word ratings. The five retention groups were excused for 2, 4, 6. 8, or 24 h. respectively. Subjects in the sixth retention condition were asked to count backwards from 100, by writing on the back of their rating sheet, for 30 sec. At the point of test, subjects in all conditions were handed a new response sheet that once again contained five rows of five blanks. Above each row of blanks on the test sheet, however, the five rated words from a list were typed, but in a new random order. Everyone was told that each grouping of five words contained items that had been presented together in a list; the task was to reconstruct the original order of presentation for each list. As I have argued elsewhere (Nairne, 1990, 1991), reconstruction tasks of this type can be viewed as pure tests of position, or possibly order, memory because the critical items are made available at the time of test.

RESULTS AND DISCUSSION The predictions of the perturbation model were based on iterative applications of Equations 1 and 2. Although there are more sophisticated versions of this model (Healy, Fendrich, Cunningham, & Till, 1987; Lee & Estes, 1981). the earlier long-term data were fit quit well by the simple version (Nairne, 1991), so it seemed appropriate to continue its use here. The perturbation rate, e, was set at 0.12 (based on inspection of the data), and the equations were applied five times to fit the 30-sec group and, thereafter, five times for every 2 h of retention interval. Thus, the equations were applied 10, 15, 20, 25, and 65 times for the 2-, 4-, 6-, 8-, and 24-h delay conditions.

mance and represent the proportions correct averaged across the five serial positions. The open circles show the predictions of the perturbation model and also were calculated by averaging performance across the five serial positions. The forgetting curve shows a characteristic, negatively accelerated form: Greater amounts of loss are seen early in the function and performance reached a level close to asymptote at about 8 h of delay. Of main interest, the function is fit quite well by the model except for the 24-h condition, where subjects performed significantly better than predicted.

The Uncertainty Gradients A more detailed display of how performance varied as a function of serial position is shown in Figure 2. The three rows of panels—depicting only the 30sec, 4-h, and 24-h retention conditions for ease of comparison—show how position responses were distributed across serial position for each of the five presented positions; for example, the first panel in the first row shows how often subjects in the shortest delay condition placed the first item in a hst in each of the five possible response positions. The open circles in each panel display the predictions of the perturbation model, based on the same parameter values used in Figure I. Consistent with earlier work (Nairne, 1991), the serial position curves were bow-shaped, and the response distributions resembled generalization gradients anchored around the true presented positions. Subjects were

0

The Forgetting Curve The data of primary interest, reconstruction performance as a function of retention interval, are shown in Figure 1. The filled circles display subject perfor200

2

4

6

a

10

12

H

IS

ia

20 ; 2

more likely to position an item correctly if that item occurred first or last in a list (e.g., the first and last panel in a row); when an error occurred, subjects were likely to place the item incorrectly in a nearby position. Moreover, as predicted, with the passage of time the gradients flattened and distant error placements became more likely. Qualitatively, it is clear that the perturbation model captures the essential trends in the data quite well. Knowledge about temporal serial position is not allor-none, but is better characterized as a distribution of positional uncertainty that changes systematically with delay. In this respect the present data mirror the patterns of immediate memory, and suggest that similar mnemonic processes may be operating in both the short- and long-term cases. More problematic for the perturbation model are the subject data at the longest retention interval (24 h). The simple version of the model, represented by Equations 1 and 2, predicts that performance should approach chance at long delays. Clearly, subjects in the present experiment were performing better than expected after 24 h, and seem unlikely to approach chance performance in any reasonable time frame. It is interesting to note that the perturbation model also underpredicts performance at long distractor intervals in immediate memory. Healy et al. (1987) found significantly better subject performance than expected at long distractor intervals (e.g., 60 sec), and proposed another parameter to reflect more "permanent" retention of position. Although the psychological significance of such a parameter is unclear in the present instance, it seems likely that a modified version of the model, along the lines proposed by Healy et al. (1987), would provide a better fit. A close examination of the uncertainty gradients indicates that performance after 24 h was especially good in the primacy portions of the list (relative to the model predictions), and this finding may ultimately prove to be of some importance.

24

Retention Interval (Hours)

The "Dating" Curves Fig. 1. Overall proportion correct reconstruction performance, as a function of retention interval, for subjects and model.

Another way to represent the data of Figure 2 is to plot the expected values of the position distributions for the subjects VOL. 3, NO. 3, MAY 1992


James S. Nairne 30 Seconds

12

12

3 4 5

3 4 5

12

3 4 5

12 3 4 5

12 3 4 5

4 Hours

12

12

3 4 5

3 4 5

12

12 3 4 5

12 3 4 5

3 4 5

24 Hours 1.0-

1,0-

1,0-

1.0-

LO-

0,8-

0,8-

0,B-

0.8-

OS-

0,6-

0,6-

0.6-

o.e-

0,6-

0,4-

0,4-

0,4-

0,4-

0,4-

0.2-

0,2-

0,2-

0,2-

0,2-

0.0

0,0

0,0

0.0

0,0

12

12 3 4 5

3 4 5

12

3 4 5

12 3 4 5

DISCUSSION AND CONCLUSIONS

12

3 4 5

Serlat Posllloti

Fig. 2. The positional uncertainty gradients for each of the five serial positions after 30 sec, 4h, and 24 h of delay. Each row of panels shows, consecutively, positions one through five. The closed circles represent the actual data; the open circles are the predictions of the model. and model. These data, for the same three retention intervals, are shown in Figure 3, The left panel shows the subject data; the right panel shows the perturbation model predictions. In each case, the points can be compared to idealized performance where position re-

sponses match the actual presentation positions. Analyzing performance in this way is instructive because forward and backward "telescoping" effects, which are common in the assignment of dates to naturalistic events (Huttenlocher, Hedges, & Bradburn, 1990; Rubin & Modei

Data

c o a.

tn 0) CC

2-

n o

1

2

3

Serial Position

4

1

2

3

4

Serial Position

Fig. 3. The expected values of the position response distributions for each serial position. The solid lines in the left and right panels show idealized performance. VOL. 3, NO. 3, MAY 1992

Baddeley, 1989), are displayed clearly. When assigning a date (position) to an item that occurred early in a sequence, subjects (and model) tend to place the item at a later position than its actual occurrence; in other words, it is brought to a more recent point in time, mimicking the way that a distant object is brought closer into view by a telescope. The same process operates in the opposite direction for items that occurred late in a sequence. Subjects tend to place these items, on average, at earlier positions in the list than those where they actually occurred. Although robust, neither of these tendencies is particularly mysterious; they arise as a consequence of the list boundaries that prevent items from drifting to positions outside the list presentation window.

The present experiment provides new data on the forgetting of order, or position information, over a time course of hours. The forgetting curve showed a characteristic form, and the serial position curves and uncertainty gradients followed the qualitative patterns commonly found in studies of immediate retention. An immediate memory model, the perturbation model of Estes (1972), predicted the empirical trends quite well overall, except for the longest retention interval. Because the dynamic properties of forgetting were explained adequately by the model, it appears that the previous fits by Nairne (1991), who sampled only one retention interval, cannot be attributed simply to an accessing of the fixed remnants of initial short-term memory processing. Still, although the descriptive aspects of short- and long-term forgetting appear similar, it remains to be seen whether the perturbation model can adequately explain both cases within a single application. For example, is there a single rule for the application of 9 that will fit forgetting functions over seconds, minutes, and hours? To answer this question convincingly, data will need to be collected that allow for comparable comparisons of order retention across the different time scales. At present, the procedures used in delayed and immediate retention 201


Positional Uncertainty arc different, making direct comparisons ditlicult. In immediate memory experiments ol the type studied by Healy cl al. (1987) and Eslcs (1972), subjects expected the order test and each list was lollowcd by an interfering distractor task. The current procedure used incidental learning and. with the exception of the 30-sec retention group, no immediate distractor task. It is reasonable to assume that a variety of factors, including intention and potential interference, might atTect the likelihood that the perturbation process will occur. Finally, these results provide support for some recent models dealing with the naturalistic retention of occurrence information (Huttenlocher et al., 1990; Rubin & Baddeley, 1989). In these cases, the pertinent data have been collected from the dating of prior events (e.g., colloquium topics, movies seen), with retention usually being assessed after weeks or months. Despite the fact that very little control could be exercised over the items, the sequences of occurrence, or the exact retention intervals, these researchers have typically found that uncertainty about the date of occur-

rence (as measured by the absolute error) increases regularly with elapsed time, Vhe currenl data confirm these trends under the controlled conditions of the laboratory, and provide systematic analyses of response distributions as a function of sequential position. Acknowledgments—I would like to thank Cynthia Aguilar and Leo Woessner for heip in collecting the data.

REFERENCES Cunningham, T,F,. Healy. A,F., & Wiiliams, D.M, (1984), Effects of repetition on short-term retention of order information. Journal of Experimental Psychology: Learning. Memory, and Cognition.'10. 575-597, Estes. W.K,, (1972), An associative basis for coding and organization in memory. In A,W, Meiton & E, Martin (Eds,), Coding processes in human memory (pp. 161-190), Washington, DC: Winston. Healy, A,F, (1974), Separating item from order information in short-term memory. Journal of Verbal Learning & Verbal Behavior. 13. 644— 655, Healy, A,F,, Fendrich, D,W,, Cunningham, T,F., & Tili. R,E, (1987), Effects of cuing on shortterm retention of order information. Journal of Experimental Psychology: Learning. Memory, and Cognition. 13. 413-425,

Hutteniocher. J., Hedges, L., & Bradburn. N. (1990). Reports of elapsed time: Bounding and rounding processes in estimation. Journal of Experimental Psychology: Learning. Memory, and Cognition. 16. i96~2l3. Jahnke. J . C . Davis. S.T,. & Bower. R,E. 0989). Position and order information in recognition memory. Journal of Experimental Psychology: Learning. Memory, and Cognition, 15. 859867. Lee, C,L..& Estes, W,K. (1977). Order and position in primary memory for letter strings. Journal of Verbal Learning &. Verbal Behavior. 16. 395-118, Lee. C L , , & Estes. W.K. (1981), Item and order information in short-term memory: Evidence for multilevel perturbation processes. Journal of Experimental Psychology. Human Learning and Memory. 7. 149-169. Nairne. J.S, (1990). Similarity and long-term memory for order. Journal of Memory and Language. 29. 733-746. Nairne. J,S, (1991), Positional uncertainty in longterm memory. Memory & Cognition 19. 332340, Paivio. A.. Yuille. J C . & Madigan. S.A. (1968). Concreteness. imagery, and meaningfulness values for 925 nouns. Journal of Experimental Psychology Monograph. 76 (1, Pt. 2), Rubin, D C . & Baddeley. A.D. (1989). Telescoping is not time compression: A model of the dating of autobiographical events. Memory &. Cognition. 17. 653-661. Shiffrin, R.M.. & Cook. J.R. (1978), A model for short-term item and order retention. Journal of Verbal Learning and Verbal Behavior. 17. 189-218, (RECEIVED

4/17/91;

REVISION ACCEPTED

7/8/91)

American Psychological Society The American Psychological Society was founded in 1988 as an independent, multipurpose organization to advance the discipline of psychology, to preserve the scientific base of psychology, to promote public understanding of psychological science and its applications, to enhance the quality of graduate education, and to encourage the "giving away" of psychology in the public interest. All members of the American Psychological Society receive Psychological Science, the new review publication. Current Directions in Psychological Science, and the APS Observer as part of their annual membership dues, which are $95.00 per year through 1992. For membership information and applications contact the American Psychological Society, Suite 1100, 1010 Vermont Avenue, NW, Washington, DC 20005-4907, Telephone: 202-783-2077; Fax 202-7832083; Bitnet: APSAGK(a^UMUC or APS2(aUMUC.

202

VOL. 3, NO. 3, MAY 1992