Effects of Sleep Deprivation on Short-Term Recognition Memory

Transcription

1 Journal of Experimental Psychology Human Learning and Memory 1975, Vol. 104, No. 2, Effects of Sleep Deprivation on Short-Term Recognition Memory Donald J. Polzella University of Dayton A probe-recognition short-term memory paradigm was used to inquire into the precise effects of sleep deprivation on human memory. It was found that recognition performance, as measured by d', was generally impaired for each subject after 24 hr of sleep deprivation. While d' was shown to decrease exponentially as the number of items intervening between the target and the probe increased, this decay rate was not affected by sleep loss. In addition there was confirmation of a previously observed increase in the positive skewness of reaction times after wakefulness. The data were consistent with the hypothesis that sleep deprivation increases the occurrence of lapses, periods of lowered reactive capacity, which prevent the encoding of items in short-term memory. In order to account for the performance decrements that frequently follow acute sleep loss Williams, Lubin, and Goodnow (1959) proposed and tested the lapse hypothesis. Lapses are identified as involuntary, intermittent periods of lowered reactive capacity that increase in frequency and duration as a function of the degree of wakefulness. Depending upon the nature of the experimental task, a lapse has particular behavioral manifestations. Thus, in subject-paced reaction time tasks, the distribution of reaction times became increasingly positively skewed with increased sleep deprivation. On the other hand, in experimenter-paced vigilance tasks lapses led to impaired accuracy. Of immediate interest are the results of an informationlearning task that required subjects to mem- This report is based on a dissertation submitted by the author to the Horace H. Rackham School of Graduate Studies, University of Michigan, in partial fullfilment of the requirements for the PhD degree. This research was supported by the Advanced Research Projects Agency, Department of Defense, and monitored by the Air Force Office of Scientific Research under Contract F C with the Human Performance Center, Department of Psychology, University of Michigan. The author is indebted to the members of his dissertation committee for their critical advice: E. Martin, W. F. Kincaid, R. G. Pachella, R. W. Pew, and M. E. Grunzke. Requests for reprints should be sent to D. J. Polzella, Department of Psychology, University of Dayton, Dayton, Ohio orize the answers to a list of questions read at a fixed rate. Testing commenced 10 min after presentation; subjects were asked to supply the correct answer to each question as it was read. Experimental subjects began to show a recall decrement at about 28 hr of sleep loss, and their performance continued to decline as the task was administered at successive stages of deprivation. The investigators attributed the memory impairment to attentional failures, assuming that when a lapse occurred during the presentation of an item, subjects would fail to register that item for later recall. In a more elaborate test of the lapse hypothesis Williams, Gieseking, and Lubin (1966) conducted two additional experiments. Experiment 1 involved the free recall of word lists under conditions that assured sensory registration by requiring subjects to write down each word as it was presented. Again the effects of only one night's sleep loss proved detrimental to immediate memory. In Experiment 2 experimental and control subjects scrutinized pictures for 10 sec and after 24 hr were asked to recognize those pictures among distractors. Once again there was a decrement due to wakefulness. Pictures seen under sleep deprivation were not as easily recognized as pictures seen under normal conditions. Equally interesting was the fact that pictures

2 SLEEP LOSS AND MEMORY 195 seen under normal conditions but tested after 24 hr of sleep loss showed no decrement. What, then, is the cause of the memory impairment with acute sleep loss? In the language of human information processing (cf. Lindsay & Norman, 1972), the Williams et al. (1966) finding that sleep loss affects memory even when to-be-remembered material is attended to would appear to rule out a failure in sensory information storage; moreover, since sleep loss had no significant effect on material already learned, we may also rule out a failure in retrieval from longterm memory. Therefore, impairment probably stems from problems encountered (lapses?), during short-term memory encoding. An existing short-term memory paradigm (Wickelgren & Norman, 1966) was chosen as the vehicle to test this hypothesis. The paradigm involves the serial presentation of some number of items followed by a single item, designated a probe, to which recognition judgment is made. There are at least two advantages to this method. First, sources of interference can be effectively controlled. Thus the probe may be a repeat of an item presented at any serial position in the study list. The number of items presented before the to-be-repeated item can be specified as a potential source of proactive inhibition (PI), and the number of items presented between the to-be-remembered item and the probe can be specified as a potential source of retroactive inhibition (RI). A second advantage to this paradigm is that it has been successfully modeled (Wickelgren & Norman, 1966). Thus the strength of an item in memory, as measured by the sensitivity parameter (d'~) given by the theory of signal detectability, appears to decay exponentially as a function of the amount of RI acting on that item at the time it is being probed. This function may be described by the following general equation, d' = of' + A, where a is designated the acquisition parameter, i.e., the strength of an item in shortterm memory at the time it is presented, < is designated the decay parameter, i.e., the rate at which the strength of the item decays, A, is designated the long-term memory parameter, i.e., the strength of an item in longterm memory (reflected in the asymptote of the decay function), and x refers to the number of items intervening between presentation and probe, i.e., the amount of RI. The choice of the probe-recognition paradigm, then, was made with the assumption that its flexibility and precision would permit a functional analysis of forgetting with sleep loss. For example, if a decrease in the acquisition parameter, a, were observed, it could be inferred that the transfer of information from sensory information storage to short-term memory was impaired. A decrease in cj> would signal difficulty encountered while information was being held in short-term memory, and a decrease in X would indicate that the strength of items in long-term memory was attenuated. METHOD Subjects. The subjects were five males aged 21 to 32. Their sleep habits were typical with normal bedtimes ranging from 10:30 p.m. to midnight and normal rising times ranging from 6:00 a.m. to 8:00 a.m. Four of the subjects were paid $70 for participating in this study. The fifth subject was the author. Stimuli. Two types of stimuli comprised the experimental material: pairs of letters and pairs of digits. Letter pairs were generated from a sample of eight consonants, B, F, H, K, N, Q, v, and x. Digit pairs were generated using the digits 2 through 9 inclusive. This format produced 56 distinct letter stimuli and 56 distinct digit stimuli (excluding those with repeated 1 symbols). Apparatus. All experimental events were visually presented on a cathode-ray tube controlled by a PDP-1 computer. The subject sat alone in a small room; two fans provided auditory homogeneity. Procedure. In general, a trial consisted of the serial presentation of 1-13 stimuli, followed by a probe stimulus that remained visible until the subject indicated whether he recognized the probe as having been presented during that trial. The subject then indicated how confident he was in his judgment. The subject's judgment, confidence rating, and response latency (the time elapsing from the presentation of the probe until the initial judgment) were all recorded. More specifically, each trial began with the presentation of a 1.27 X 1.27 cm visual mask that appeared in the center of the cathode-ray tube. This mask remained visible for 250 msec. Following a blank interval of 250 msec, the mask reap-

3 196 DONALD J. POLZELLA peared for another 250 msec. This pattern was continued for 2 sec (250 msec blank, 250 msec mask, etc.) and was followed by the presentation of the first stimulus for 250 msec in the center of the cathode-ray tube. All stimuli were presented at the rate of 2/sec: Each lasted 250 msec and was followed by a mask of the same duration. The initial 2-sec blank-mask sequence was included so that the subject could anticipate the relatively fast rate of presentation. The mask following each stimulus was included in order to obliterate all traces of the stimulus after 250 msec. Subjects were given no explicit rehearsal instructions. The last stimulus in the study sequences was followed by a random pattern of lines that extended over the entire cathode-ray tube. This pattern was visible for 500 msec and preceded the presentation of a probe stimulus, which remained visible until the subject pressed the yes button with his right index finger (indicating that he recognized the probe as having been presented during the immediately preceding study sequence) or the no button with his left index finger (indicating that he did not recognize the probe). Subjects were instructed to press the appropriate button as soon as they were able to decide. After this recognition judgment the words ENTER CONFIDENCE appeared on the cathode-ray tube, signaling the subject to rate his decision by pressing one of the three buttons labeled 1 (unsure), 2 (sure), and 3 (very sure). After the confidence rating was entered the words CORRECT or INCORRECT appeared on the tube for 2 sec as feedback. After this, a period of 5 sec elapsed before the reappearance of the blank-mask pulse indicating the beginning of the next trial. A series of 160 trials constituted one experimental session, and each session lasted approximately 30 min. In addition, a random series of 10 practice trials was presented before each session began in order to eliminate warm-up effects. The entire experiment included seven rested sessions and two sleep-deprived sessions covering a total of 9 consecutive weekdays beginning on Monday of Week 1 and ending on Thursday of Week 2. Sleep-deprived sessions occurred on Days 4 and 8. Each subject was tested at the same time on the 9 days at 45-min intervals, with the first subject beginning his session at 7:30 a.m. and the last at 10:30 a.m. Sleep-deprived sessions. On Wednesday of Week 1 and Tuesday of Week 2 all subjects reported to the laboratory at approximately 11:00 p.m. and remained together and awake until the experimental sessions began on the following morning. No formal structure was planned for the wakefulness periods, and the subjects engaged in a variety of activities including listening to the radio, playing cards, and watching television. Those subjects who normally drank coffee were permitted to do so; however, they were limited to two cups and were not permitted to drink coffee within 2 hr of the experimental session. At no time did any subject express the desire to drink beyond this quota. All subjects ate a light breakfast at approximately 6:00 a.m. Design. An experimental session consisted of 160 trials. On half the trials the probe stimulus appeared within the preceding sequence (repeat) ; on the remaining half the probe stimulus did not appear within that sequence (nonrepeat). Orthogonal to this, 80 trials consisted of digit-pair stimuli and 80 trials consisted of letter-pair stimuli. These were alternated to reduce intertrial interference due to stimulus similarity. Four levels of PI (0, 1, 2, or 4 stimuli) and five levels of RI (0, 1, 2, 4, or 8 stimuli) were varied orthogonally on the 80 repeat trials. The level of PI is the number of stimuli that were presented before the to-be-remembered item. Analogously, the level of RI is the number of stimuli that were presented between the to-be-remembered item and the probe. Thus, the length of a study sequence could vary between 1 (PI = 0, RI = 0) and 13 (PI = 4, RI = 8) stimuli. Each PI-RI combination occurred four times in an experimental session, twice with digit-pair stimuli and twice with letter-pair stimuli. The 80 nonrepeat trials were matched with these 80 repeat trials with respect to the length of the study sequence. The sequence of repeat and nonrepeat trials was randomly determined, with the constraint that every PI-RI combination plus an equal number of nonrepeat trials of appropriate length appear once with digit-pair stimuli and once with letter-pair stimuli within the first 80 trials of the experimental session. This had the effect of dividing each session into two halves, each half matched with respect to the number of repeat and nonrepeat trials and with respect to the particular PI-RI combinations presented. RESULTS J The dependent measure of primary focus is the sensitivity parameter, d', which was computed by using the single point on the memory operating characteristic (MOC) corresponding to the yes-no criterion. While it was not possible to use the method of confidence judgments to generate the entire MOC (three subjects failed to vary their ratings), the two methods of determining d' generally yield similar estimates (Wickelgren & Norman, 1966). The appropriateness of pooling the raw data over experimental sessions and over stimulus types was investigated in two analyses of variance that compared the subjects' performance on both types of stimuli across the rested and sleep-deprived sessions, re- 1 The raw data for each subject may be obtained from the author.

4 SLEEP LOSS AND MEMORY 197 spectively. Because certain data were not available (one subject missed the first session, and the data from the entire second session and from 83 trials of the fourth session for two other subjects were not stored because of equipment problems), a procedure for estimating missing observations discussed by Winer (1962, p. 282) was incorporated in these analyses. No significant difference was found in d' across the seven rested sessions or between the two sleep-deprived sessions, F<1 and F(l, 4) = 1.32, respectively. For both the rested and sleep-deprived sessions d' was significantly lower with letter stimuli, F(l, 4) =21.04, p <.01 and F(l, 4) = 11.21, p <.05, respectively, but there were no significant interactions. In view of these results, pooling the data for additional analyses did not seem inappropriate. Effect of sleep deprivation on d'. Table 1 lists the overall d' for each subject under rested and sleep-deprived conditions, and all five subjects show the predicted decrement with sleep loss. A test proposed by Gourevitch and Galanter (1967) was used to analyze these differences. The test is designed to assess whether such differences reflect inherent binomial variation in the frequencies observed (i.e., hits, misses, correct rejections, false alarms) or whether they are due to statistically significant variation. For sufficiently large samples the statistic G is normally distributed and may be assigned a significance level in a table of normal curve areas. With the raw data pooled over subjects, d' under rested conditions was 1.69 and under sleep-deprived conditions This difference was found to be highly significant (G = 5.50, p <.001), indicating that sleep loss detrimentally affected subjects' sensitivity. 2 Individual subject analyses were also performed and the obtained significance levels (two-tailed) are noted in Table 1. Sleep deprivation and RI. It will be recalled that Wickelgren and Norman (1966) 2 Although the use of pooled data to estimate d' should generally be avoided, its effect is to increase the signal-noise variance (see Lockhart & Murdock, 1970). In the present case this results in a conservative test of the null hypothesis. Subject Pooled TABLE 1 EFFECT OF SLEEP DEPRIVATION ON d' Rested Sleepdeprived P <.14 <.34 <.002 <.08 <.08 <.001 modeled the probe-recognition paradigm and showed that the strength of an item placed in memory, as measured by d', decreases exponentially as a function of RI. Thus the appropriate raw data were pooled over subjects to yield group estimates of d' as a function of RI and experimental condition. Exponential curves were fitted to these group estimates according to the general procedure outlined by Lewis (1960). It was found that the best-fitting function for both the rested and sleep-deprived data was of the general form d' = a^ + A.. The following equations were obtained: Rested: d' = (3.04).64*+.58. Sleep deprived: d' = (2.34).71" These are smooth functions plotted in Figure 1. It appears that sleep loss had a detrimental effect on A, the asymptotic level of d', and on a, the acquisition parameter. On the other hand, <, the decay rate parameter, was little affected. According to the Gourevitch and Galanter (1967) test the effect of sleep loss was significantly manifested at all levels of RI (p <.05). Sleep deprivation and PI. The appropriate raw data were pooled over subjects to yield group estimates of d' as a function of PI and experimental condition. These estimates are plotted in Figure 2. As was the case for RI, the effect of sleep loss was significantly manifested at each level of PI (p <.05). In addition, the apparent inverse relationship between d' and PI was significant, F(3, 12) = 15.88, p <.01. Sleep deprivation and reaction time. The median reaction time (RT) was computed

5 198 DONALD J. POLZELLA Rested: d'=(3.04) O o Sleep-Deprived: d'=(a.34).7l x +.22 FIGURE 1. Values of d' as a function of retroactive interference (RI) under rested and sleepdeprived conditions. in milliseconds for each subject for hits, misses, correct rejections, and false alarms under both rested and sleep-deprived conditions. A two-way (Experimental Event (e.g., hit, miss, etc.) X Experimental Condition) repeated measures analysis of variance indicated that there was a significant difference in RT depending on the event that occurred, F(3, 12) = 85.43, p <.01; however, sleep deprivation had no significant effect on RT, F(l, 4) = 1.79, nor was there a significant interaction (F < 1). As was noted above, Williams et al. (1959) have observed that sleep deprivation may affect the degree of skewness of RTs but have little effect on the median. This effect would be due to the increased incidence of lapses, which are signaled by relatively slow Rested o- o Sleep-Deprived FIGURE 2. Values of d' as a function of proactive interference (PI) under rested and sleepdeprived conditions. RTs on a subject-paced task. Thus a frequently used measure of skewness, gi, which is obtained from the second and third moments about the mean, was computed for each RT distribution (i.e., hits, misses, etc.) for each subject. A two-way (Experimental Event X Experimental Condition) repeated measures analysis of variance on these ffts indicated that RT was significantly more positively skewed with sleep loss than with rest, F = 7.95, p <.05. No other effects were significant. Lapses have been operationally identified by Bills (1931), Bjerner (1949), Lisper and Kjellberg (1972), and Williams et al. (1959) as occurring when the RT is at least twice as long as the average RT during baseline. I judged this criterion inappropriate here, since its utility is dependent upon the average RT being equal under rested and sleepdeprived conditions. Therefore in the present study a lapse was operationally defined as any RT that was at least twice as long as the median RT for a particular distribution. In accordance with this definition Figure 3 shows the mean percentage of lapses that occurred under rest and sleep deprivation for each RT distribution. A two-way (Experimental Event X Experimental Condition) repeated measures analysis of variance on the transformed data (arc sine) confirmed what inspection of Figure 3 and the analysis of skewness suggest: A significantly greater number of lapses occurred under conditions of sleep loss than under rested conditions, F(l, 4) = 9.61, p <.05. There were no other significant effects. It will be recalled that the lapse hypothesis not only predicts an increased incidence of lapses following sleep loss but also attributes performance decrements to these lapses. We may thus ask whether lapses in the present study, identified by the specified increase in RT, signaled a deterioration in recognition performance as well. Figure 4 is addressed to this problem and shows the mean probability that a given recognition response was correct (hit or correct rejection) as a. function of the RT of that response. Referring to the figure, "Fast" are those RTs that constituted the lower third of the upper half of

6 SLEEP LOSS AND MEMORY 199 the RT distribution for each subject, "Medium" are those RTs that constituted the middle third of the upper half, and "Slow" are those RTs that constituted the middle third of the upper half. Thus, the abscissa of Figure 4 may be loosely interpreted as delineating successive blocks of RTs, each containing a greater number of "true" lapses than the preceding one. A two-factor (RT X Experimental Condition) repeated measures analysis of variance on the transformed (arc sine) percentage data revealed a significant effect of RT on recognition, F(2, 8) = 29.82, p <.01, a significant detrimental effect of wakefulness, F(l, 4) = 13.38, p <.05, and no significant interaction (F<1). Thus, recognition memory was increasingly impaired as the frequency of lapses increased. This was true for both experimental conditions. DISCUSSION The results of this study were generally consistent with the lapse hypothesis. Thus, analyses of the RT data showed that there was a greater incidence of lapses after sleep loss. Moreover, these lapses were accompanied by decrements in memory. Lapses probably occurred during the presentation of study lists as well, although this was not verified directly. Indeed, probes were presented at the rate of only about S/min, and subjects spent the bulk of their time attending to the experimenter-paced study lists where lapses would have prevented the encoding of stimuli. The results relating sleep loss, RI, and PI were also consistent with the lapse hypothesis. Viewing the exponential decay of d' as a function of RI, the most important observation was that sleep deprivation had little if any effect on decay rate. Rather, its deleterious effects were similarly manifested at every retention interval. This was reflected mathematically by decreases in a and \, the acquisition and long-term memory parameters, respectively. Presumably, lapses prevented the encoding of stimuli in short-term memory, which, in turn, prevented their encoding into long-term memory. This explanation is also consistent with the observa I I 10 9 LU 7 u I 0 RESTED >-cn ujcn oz cos UJO -J(K KP <a SLEEP-DEPRIVED tion that the effects of sleep loss on d' were similarly manifested at each level of PI. At least one problem remains: Invoking the lapse hypothesis to explain short-term memory impairment after sleep loss appears «0.8 * ^0.7 o u 0.6 FIGURE 3. The mean percentage of lapses that occurred under rested and sleep-deprived conditions. 0.5 Rested o- o Sleep-Deprived 0.4- _L J_ I "Fast" "Medium" "Slow" REACTION-TIME FIGURE 4. The probability of a correct response as a function of reaction time under rested and sleep-deprived conditions.

7 200 DONALD J. POLZELLA to be inconsistent with the earlier finding of Williams et al. (1966) that there will be memory impairment even when lapses are not apparent. Nevertheless, all of the existing data may be accommodated. Thus, in their experiment, to-be-remembered stimuli were presented at a 10-sec rate. Although initial encoding was verified (subjects wrote down each item as it was presented), lapses could have occurred during the interstimulus intervals, where they would affect rehearsal efficiency. This would result in differential recall because the probability of recalling an item is an increasing function of the amount of rehearsal it receives (see Rundus & Atkinson, 1970). REFERENCES Bills, A. G. Blocking: A new principle in mental fatigue. American Journal of Psychology, 1931, 43, Bjerner, B. Alpha depression and lowered pulse rate during delayed actions in a serial reaction test: A study in sleep deprivation. Ada Physiologica Scandinavica, 1949, 19, Suppl. 65, Gourevitch, V., & Galanter, E. A significance test for one parameter isosensitivity functions. Psychometrika, 1967, 32, Lewis, D. Quantitative methods of psychology. New York: McGraw-Hill, Lindsay, P. H., & Norman, D. A. Human information processing. New York: Academic Press, Lisper, H., & Kjellberg, A. Effects of 24-hour sleep deprivation on rate of decrement in a 10- minute auditory reaction time task. Journal of Experimental Psychology, 1972, 96, Lockhart, R. S., & Murdock, B. B. Memory and the theory of signal detection. Psychological Bulletin, 1970, 74, Rundus, D., & Atkinson, R. C. Rehearsal processes in free recall: A procedure for direct observation. Journal of Verbal Learning and Verbal Behavior, 1970, 9, Wickelgren, W. A., & Norman, D. A. Strength models and serial position in short-term recognition memory. Journal of Mathematical Psychology, 1966, 3, Williams, H. L., Gieseking, C. G., & Lubin, A. Some effects of sleep loss on memory. Perceptual and Motor Skills, 1966, 23, Williams, H. L., Lubin, A., & Goodnow, J. J. Impaired performance with acute sleep loss. Psychological Monographs, 1959, 73(14, Whole No. 484). Winer, B. J. Statistical principles in experimental design. New York: McGraw-Hill, (Received June 4, 1974; revision received October 21, 1974)