Constructing a More Difficult Recognition Test For Television Commercial Scenes

ABSTRACT - One of the disadvantages to using recognition as a measure of memory is that scores are often indiscriminantly high because the task is too easy. This paper investigates varying the length of time of exposure during the test, the length of the mask about the test stimulus, the number of exposures to the stimulus and the length of the decay period as variables that might reduce the ceiling effects often associated-with recognition. Data show that controlling test scene length, its interaction with mask length, and decay period can increase the difficulty of the recognition task.

Citation:

Michael L. Rothschild, Lisa Qualheim, Brian Deith, and Yong J. Hyun (1990) ,"Constructing a More Difficult Recognition Test For Television Commercial Scenes", in NA - Advances in Consumer Research Volume 17, eds. Marvin E. Goldberg, Gerald Gorn, and Richard W. Pollay, Provo, UT : Association for Consumer Research, Pages: 785-791.

Advances in Consumer Research Volume 17, 1990      Pages 785-791

CONSTRUCTING A MORE DIFFICULT RECOGNITION TEST FOR TELEVISION COMMERCIAL SCENES

Michael L. Rothschild, University of Wisconsin

Lisa Qualheim, University of Wisconsin

Brian Deith, University of Wisconsin

Yong J. Hyun, University of Wisconsin

ABSTRACT -

One of the disadvantages to using recognition as a measure of memory is that scores are often indiscriminantly high because the task is too easy. This paper investigates varying the length of time of exposure during the test, the length of the mask about the test stimulus, the number of exposures to the stimulus and the length of the decay period as variables that might reduce the ceiling effects often associated-with recognition. Data show that controlling test scene length, its interaction with mask length, and decay period can increase the difficulty of the recognition task.

INTRODUCTION

Measuring memory of advertising is of interest to both academic and industry researchers. While recall has in the past been the most commonly used dependent variable in testing responses to television commercials, recognition has become more viable in recent years. It has been argued that recall may mask the amount of actual memory that exists, and that recognition is more appropriate to the level of learning necessary for in-store brand choice decisions (Bettman 1979).

While recognition measures may be more effective and appropriate in certain situations, one important problem remains. Recognition measures tend to produce ceiling effects, or indiscriminately high scores, especially for pictorial stimuli (Shepard 1967; Haber 1970), and even more so for dynamic pictorial stimuli (Goldstein, et. al. 1982) such as movies (their stimuli) or television commercials. These results make it difficult to use recognition to test memory for visual components of television commercials. The purpose of this paper is to investigate a method which will reduce these ceiling effects by making the task more difficult.

To date, a number of ways of avoiding ceiling effects have been developed. For example, (1) the memory decay period can be increased (Shepard 1967; Singh, Rothschild and Churchill 1988), (2) the number of distractor items in a forced choice recognition test can be increased (Singh and Rothschild 1983), or (3) the similarity of the distractors to the target items can be manipulated (Dole and Baddeley 1962). This paper will discuss manipulating the length of exposure to visual recognition test items as another method of avoiding ceiling effects. Using short test item exposure times, if effective, will have two advantages over other methods. First, less subject time will be required than for designs allowing either relatively long exposure to test items or long delays between exposure to commercials and test taking; this will also reduce subject fatigue. Second, the need to invent distractor items of comparable difficulty will be reduced compared to designs employing multiple distractor items.

LITERATURE REVIEW

This section is organized into two parts which describe (1) the Search of Associative Memory (SAMFmodel and (2) factors affecting recognition test scores for pictorial stimuli.

The Search of Associative Memory (SAM)

There are several dominant models of recall and recognition memory in the psychological literature. Dissatisfaction with the validity of any current models led Gillund and Shiffrin (1984) to develop this model for both recall and recognition. Recall is assumed to involve a rather lengthy search of long-term memory plus a fast, often implicit, > recognition-like process that allows a subject to I decide that an item recovered from the sampled I image is appropriate. Since only the recognition aspects of the model are important to this paper, the recall segments will not be explained further.

In the SAM model, recognition consists of a complex direct access process in which many memory images are contacted by a retrieval cue. It is assumed that in a "yes/no" recognition test the subject probes memory with two cues, the context cue and the tested item. Context is used in a broad sense here to include subject factors such as mood and alertness, as well as environmental factors such as time of day, other objects in the testing room, lighting, etc.

The number of images in long term memory contacted by the context and test item cues determines the familiarity of the tested item. If the familiarity is greater than some subjective criterion, the subject replies "yes", otherwise s/he replies "no". In "forced-choice" and "batch" recognition tests subjects generate familiarity values for all test items and choose the items with highest familiarities.

According to Gillund and Shiffrin the strength of long-term memory and hence the familiarities of items on a recognition test are functions of three factors: (1) rehearsal and coding processes that take place during the study of, or exposure to, the to-be-remembered material, (2) pre-experimental associations or memories for all test items, and (3) the match, or similarity, between the cue encodings at study and test.

Given these factors, Gillund and Shiffrin use Signal Detection Theory (SDT) to explain subjects' performance in recognition tests. In SDT, the familiarity values for items are assumed to fall into two overlapping, roughly "normal" distributions. The distribution of old, or previously studied items, should be higher on the familiarity scale than the distractor items, although the tails of the distributions will probably overlap. Familiarity values for distractors in a recognition test will be influenced by pre-experimental memory for the distractors (factor 2 above), as well as similarity between distractors and target test items (factor 3 above). Familiarity values for target (previously studied) test items will be influenced by these two factors plus the amount of processing subjects perform during exposure to the study material (factor 1 above). Subjects are assumed to place a criterion or cut-off value near the intersection of the two distributions. Criterion values are assumed to be subjective and variable based on subjects' evaluations of the familiarity of the distractors as a group and the targets as a group. Test items evoking a familiarity value greater than the criterion elicit a "yes" response and other test items elicit a "no" response.

In the context of advertising, the SAM model predicts that using recognition to test advertising effectiveness can lead to indiscriminately high test scores. In general recognition scores will be highest when target and distractor familiarity distributions are farthest apart. Picture memory may be stronger and more enduring than verbal memory in this model if pictures tend to be more novel or to provide more information than verbal stimuli. At the same time, test item novelty causes pre-experimental memory for test items to be lower. Thus stimulus richness and novelty tend to push the distractor and target distributions apart.

High scores can be avoided by increasing the similarity of distractor and target test items, hence moving the two familiarity distributions closer together. One example concerns subjects' knowledge of an upcoming test. If subjects are unaware of an upcoming test, then they are less likely to try to remember the "study" material, their encoding during "study" may not be very effective, and the target and distractor test items should appear equally unfamiliar.

Factors Which Affect Recognition Test Scores

There are a wide variety of methods and variables that impact upon recognition test scores. One of the most common concerns the length of time between study and test. For example, Shepard (1967) presented subjects with 612 pictures to examine at their own pace. Subjects were then given two-item forced choice recognition tests after no delay, a two hour delay, a three day delay, a seven day delay or a 120 day delay. The average recognition accuracy scores ranged from 99.7 to 57.7%. Shepard explained these results in a Signal Detection Theory framework. In a two-item forced-choice recognition test, subjects presumably choose whichever item seems more familiar and call it "old". Over time the familiarity of the truly "old" items fades, leading to more errors.

A study by Singh and Rothschild (1983) showed that manipulating the number of distractor in a forced choice recognition test and manipulating the number of exposures to study material also could affect recognition memory. In their study, subject viewed segments of local news programs interspersed with commercials. Commercials were repeated 1, 2, or 4 times (a within subjects factor) and subjects received either a 5- or 9-alternative visual recognition test (between subjects factor). Significant main effects on recognition were found for repetition and for number of distractors.

Decreasing length of exposure to test items may also diminish subject performance. A study 1 Franken and Rowland (1974) found just such an effect for recognition memory of pictures. During study, 150 color sides were shown to subjects for seconds each. Immediately thereafter a "yes/no" recognition test was given consisting of 15 slides randomly selected from the 150, plus 15 new slides. Eight exposure durations ranging from 120 to 5000 milliseconds (msec) were used. The corresponding recognition accuracies obtained ranged from 77.54 to 92.7%. At test exposure levels of only 120 msecs, subjects performed quite well; scores reach an asymptote by 500 msecs.

When considering using extremely brief test item exposures, the question of masking needs to considered. Studies show that visual sensory memory or iconic store can last about 250 msecs. (Sperling, 1960). [The iconic store or visual sensory memory holds visual information in an unidentified form for a brief time before higher mental processes recognize the information. Its function is to hold information during eye blinks so that sight seems to be continuous. A mask is a non-test stimulus that can immediately precede or follow the test stimulus and has the capability to disrupt or enhance the iconic storage process depending upon its length and placement. Operationally, the mask often appears as a random pattern.] A visual mask following stimulus presentation will erase or interfere with images in the icon, hence shortening the time subjects effectively perceive the stimulus. Because iconic memory fades and because individuals differ with respect to rate of fading (Sperling 1960), using a mask following stimulus presentation should give better control over the effective length of the stimulus.

Forward masking is also of interest when stimulus presentation is very brief. A mask preceding the stimulus will have the effect of focusing subjects' attention on the screen. This i especially important in studies using variable presentation rates, relatively long interstimulus intervals, or any presentation device other than a tachistoscope. The length of the forward mask is not critical as long as it is more than 100 msecs. Di Lollo (1980) has shown that at lengths of less than 100 msecs forward masks interfere with perception of stimuli so that the mask and the stimuli appear to overlap; as a result subjects see both at the same time. This work also implies that if a stimulus interval is less than 100 msecs, a mask following it may be perceptually integrated with the stimulus. If the following mask is long enough to persist beyond the mask-stimulus integration, subjects may be better able to cognitively separate the two and more accurately judge the familiarity of the stimulus.

A study by Dole and Baddeley (1962) demonstrated that the similarity of the distractors to the target items also affects recognition test scores. They concluded that when memory is imperfect, only some of the relevant characteristics of the study material can be remembered. If the distractors and the target items on a recognition test are very similar, they will both be similar to the imperfect memory and subjects will be unable to tell which they have seen before. While the Dole and Baddeley study used numbers as stimuli, the results can be expected to transfer to studies using pictorial or verbal stimuli.

The studies reviewed above provide a number of methods of decreasing recognition test scores. While their authors did not use the Gillund and Shiffrin SAM model for recognition memory to interpret the results, they fit into the SAM model rather well. In the following section four hypotheses are developed concerning methods of reducing recognition test scores based on the SAM model of recognition.

HYPOTHESES

As suggested by Gillund and Shiffrin's model for recognition memory, test scores can be affected by manipulating the familiarity of distractors to target items in a recognition test. This can be accomplished by manipulating the processing that takes place during study of the to-be-remembered material, manipulating pre-experimental memory for test items, or manipulating the match between cue encodings at study and at test. In this section four methods discussed above are developed into hypotheses.

Manipulating the match between cue encodings at study and at test can be accomplished by manipulating the length of exposure during the test period. This may affect the match between the subject's internal representation of the test item and its corresponding memory from previous study. While such a notion may seem obvious, it has not been tested in advertising and rarely tested in psychology; it needs to be tested for quite brief time periods in order to be useful to advertising tests of recognition. Therefore:

H1: Increasing the test length of exposure to the recognition stimulus has a positive effect on recognition memory when dealing with brief time periods of exposure.

Forward masks of less than 100 msecs may be perceptually integrated with the test item. Following masks may be perceptually integrated with test items of less than 100 msecs. However if the following mask is longer than 100 msecs minus the length of the test item, subjects may have an easier time of cognitive!y separating the two. Therefore:

H2: Increasing the length of the forward mask (up to 100 msecs) and the following mask (up to the length of the test item minus 100 msecs) will have a positive effect on recognition memory.

The match between cue encodings at study and at test can also be manipulated by varying the delay between study and test. This has been shown often in the past but needs to be observed within the present context of brief text exposures. Therefore:

H3: Increasing delay between study and test has a negative effect on recognition memory.

Multiple exposures study material should create a stronger memory for that material and a "richer" context encoding. This also has been shown in the past but needs to be observed within the present context. Therefore:

H4: Increasing the number of exposures to study material has a positive effect on recognition memory.

Of these four hypotheses, the first two are concerned with issues that have not been tested in the advertising and consumer research literature, while the latter two will serve as replications and extensions of previous work. The first two represent the major thrust of this study.

METHOD

The experimental design was a 3 (scene length of 30, 60 or 90 msecs) by 3 (scene plus mask length of 180, 300 or 420 msecs) by 2 ( 15 or 30 minute delay between presentation of study material and recognition test) by 2 (1 or 2 exposures to the study materials) split plot design. The first two factors, length of scene and length of recognition task, were within subjects factors, while length of decay period and number of exposures were between subjects factors.

Subjects

Twenty undergraduate students participated in the study. Subjects were assigned to one of three conditions such that there were 10 subjects in one condition (one exposure, 15 minute delay) and 5 subjects in each of the other two conditions (two exposures, 15 minute delay, and 2 exposures, 30 minute delay). All subjects were exposed to all levels of scene length and scene plus mask length.

Procedure

Subjects were told they were part of a control group for a study measuring EEG (brain wave) responses to television programming, and were asked to watch a series of video tapes. To enhance the cover story, subjects were asked to sit as still and quietly as possible while viewing, as movement caused muscle artifact noise in EEG patterns and we wanted them to act in the same way as the actual EEG subjects. Subjects were allowed to move about as much as they liked between video tapes; they participated in the study one at a time. In actuality, the study was a pretest for a later EEG study; the issue of the recognition test discussed herein was the sole purpose for the pretest.

Subjects were not told specifically what they would be viewing, nor was there any hint that memory measures might follow. Subjects sat in a small room (12' by 13') about 6 feet from a 13" Sony Trinitron color television. They stayed in their seats for the duration of their participation except for a short 2-3 minute break between viewing of the final tape and the beginning of the testing period.

All subjects began by viewing a 10 minute acclimation tape. They then viewed a stimulus tape consisting of 12 different 30 second commercials. The commercials were separated by 30 seconds of blank tape. A tape recorded voice asked subjects to close their eyes 5 seconds after each commercial ended and to open their eyes again 5 seconds before the next commercial began. At this point, some subjects saw the set of commercials a second time. After viewing the commercials, all subjects viewed a 7 minute distractor tape, followed by a short 2-3 minute break.

Some subjects were then given an unaided recall test for the commercials which served as a distractor and 15 minute decay period.. All subjects were given the recognition tests. Finally, subjects were debriefed and asked specifically not to discuss their participation with other students.

In sum, subjects were randomly assigned to one of three groups. Group 1 had two exposures to the messages during initial viewing and participated in the recall test; group 2 also had two exposures but no recall test, while group 3 had one exposure and no recall test. In all other respects, all subjects participated equivalently.

Stimulus Materials

Acclimation tape: This 10 minute tape consisted of 3 short segments of college sporting events. The purpose of the tape was simply to let subjects get comfortable with their surroundings.

Commercial tape: Twelve thirty-second commercials intuitively judged to evoke a variety of positive, neutral and negative emotions were selected. None of the selected commercials had ever been aired in the test city and none of the brands represented in the commercials were marketed in the test city.

Distractor tape: A 7 minute tape of a man making a vase on a potter's wheel was shown to all subjects after final viewing of stimulus commercials to allow memory to decay.

Measurements

All subjects judged two practice and thirty-six actual pictorial test items as presented on video tape. Each item consisted of 1, 2 or 3 frames (33, 67 or 100 msecs) from a commercial preceded and followed by a random dot pattern mask. Half of the items were scenes from the test commercials and half were foils from another group of similar, previously unseen, untested commercials. The items were imbedded in one of three scene plus mask lengths (200, 333, and 466 msecs).

In constructing the-recognition test video tape, the masks were laid down first for the scene plus mask length period, and then the test scenes were recorded over and in the middle of each mask. As a result, the length of the scene and the length of the scene plus mask were controlled, but the length of the forward and following masks took on the value of the time remaining after the other values were controlled. While not necessarily optimal, this construction was dictated by editing equipment capabilities. The test items were recorded at 10 second intervals and a voice announced the item number of each test item. The ten seconds was ample time for subjects to view the stimulus and record the answer. Subjects recorded their responses on computer readable answer sheets based on a 5 point scale ranging from "definitely saw before" to "definitely did not see before."

RESULTS

Table 1 shows how each of the independent variables in the study affected recognition. Answers from the 5 point scale were summarized into binary responses with all midpoint "not sure" responses counted as incorrect. The value in each cell of Table 1 corresponds to the percentage of correct responses. Note that a proportion of 40% would indicate random guessing (2 values of the 5 point scale); overall, subjects performed significantly better than chance (D4.50; p<.001). Table 2 shows analysis of variance results.

Effect of Scene Length on Memory

There was a significant main effect for scene length (33, 67, or 100 msecs) on scene recognition (F=38.96t p<.001). As Table 1 shows, there also were significant differences between correct recognition scores for scene lengths of 33 (.26 recognition) and 67 (.53) msecs (t=6.15; p<.001), between scene lengths of 67 (.53) and 100 msecs (.65) (t=2.62; p<.005), and between scene lengths of 33 and 100 msecs (t=8.87; p<.001).

There was a significant interaction between length of scene and length of scene plus mask (F=10.32; p<.001) so that recognition scores were highest when the scene was relatively long but the scene plus mask was relatively short. When the scene was short, the length of the scene plus mask did not matter. There was no significant interaction between scene length and number of exposures (F =.09; ns). In calculating interactions, length of delay was suppressed due to the incomplete nature of the design.

TABLE 1

RECOGNITION SCORES AS A FUNCTION OF INDEPENDENT VARIABLES

Effect of Length of Delay

There was a significant main effect for length of delay (F = 9.01, p < .003). Subjects with a short delay were correct 51% of the time while subjects in the long delay were correct only 41% of the time. Interactions between length of delay and other variables were suppressed due to empty cells in the design. Length of delay period (15 versus 30 minutes) was determined by the absence or presence of the recall test.

Effects of Number of Exposures and Scene Plus Mask Length

There was a marginal main effect due to number of exposures (F=2.78; p<.096), and no effect due to scene plus mask length (F=.05; ns). There was no significant interaction between number of exposures and scene plus mask length (F=1.30; ns), nor was there a significant three way interaction between number of exposures, scene length and scene plus mask length (F=.80; ns).

DISCUSSION

The purpose of this study was to continue the investigation of the potential viability of recognition as a response variable in the study of memory for advertising. One of the key inhibitors to the widespread use of recognition has been the general tendency toward high scores during its use. It has been shown in other studies that the length of the decay period, the number of repetitions of the stimuli, and the similarity between the distractors and the test stimuli all can have an impact on lessening ceiling effects. In the current study, the variables of concern were primarily the length of the stimulus during the test and the impact of a surrounding mask.

By showing the test stimulus for brief periods of time, recognition scores, in the aggregate, were kept lower. In earlier work, a decay period as short as that used herein, coupled with an exposure period of as little as one half second would have led to levels of recognition above 90% and little variance across stimuli. It seems clear that the extremely brief exposure periods used in the present study have accomplished the desired goals.

Presumably the extremely short test exposure periods did not allow subjects to create clear internal representations of the test scenes. Consequently, a memory search using the imprecise internal representation of the test stimulus was less likely to contact memories of the studied material. When subjects' perceptions of the test stimuli were vague, they may have tried to fill in the missing parts, or cognitively create a clearer image using whatever memory they had of the studied material and any pre-experimental pictorial images.

TABLE 2

ANALYSIS OF VARIANCE

The length of the scene plus the mask was not found to significantly affect recognition memory although there was a significant interaction between scene and scene plus mask length. A main effect due to mask and scene length may have been obscured by nonindependence in that the preceding and following masks were the same length for each test item. The interaction between scene and scene plus mask also could imply that forward and following mask lengths may affect recognition of different length scenes differently. This issue needs to be addressed more extensively in future work.

The major focus of this study was to shed light on the potential for seeing differences in memory as a function of the length of a test stimulus. In addition, by examining decay and repetition effects under the conditions of brief test stimuli, further insights could be gained.

Length of delay was found to affect recognition scores. As hypothesized, the length of delay between study and test had a negative effect on recognition. Even a fifteen minute difference was found to significantly affect recognition memory for the pictorial stimuli. The delay allowed time for subjects' memories to fade, and also progressively altered the context. As the difference between study context and test context increased, the memory test showed poorer performance.

Number of exposures was found to only marginally affect recognition memory. Perhaps this was due to the phenomena of particularly strong memory for pictorial stimuli even after one exposure. It could be that the second exposure did not significantly increase memory for the visual aspects of the commercials. Even though memory after one or two exposures may have been quite high, recognition ceiling effects were avoided by using extremely brief exposure to test items. This would lead to the recommendation that one repetition be used (again to simplify things for the subject) when recognition testing involves extremely brief exposures to the test stimuli.

A test using a 30 minute decay with a 33 msec test would be incredibly difficult, and, as in the present case, would lead to mostly "don't know" responses. More realistic would be a 15 minute decay with a 67 msec test. This level of difficulty would allow subjects to respond at a low, but better than chance rate, and would also minimize the amount of time that a subject would need to be detained.

There were several potential design flaws in this study, yet the key finding concerning the value of extremely short test stimuli still emerged as significant. The potential flaws include the following: (1) Small number of subjects. While there were only 20 subjects in the study, there were 720 observations, given the within subjects nature of the design. (2) Incomplete design. The incomplete factorial design made it impossible to test for all the interactions; while that test will need to be repeated, the main effects still clearly remain. (3) Confound. The length of the decay period is confounded with the presence of the recall test; that is, in the 30 minute decay, there was a recall test which was absent in the 15 minute decay condition. This should not be a concern because other work (Darley and Murdock 1971; Postman, Jenkins and Postman 1948) has shown that a preceding recall test does not bias the results of a following recognition test. In the present case this is even less of an issue since the recall test was a verbal test of the entire message, while the recognition test was a pictorial test of scenes. In addition, the recognition scores after the recall test are lower than those without this test. The results, therefore, are consistent with he decay literature; the recall test of the whole seemed to act as a distractor for the parts.

The potential value of using these data is twofold: (1) to keep recognition scores from being so high that there is not enough variance to discriminate between memory for different stimuli, and (2) to allow the recognition test to be done more quickly and without the lags generally necessary to allow for the decay of memory.

Furthermore, as the need for more precise advertising tests becomes more prevalent, researchers are beginning to test component parts of messages rather than entire messages. As this trend increases, the issues discussed herein will become more relevant. Showing frames that represent scenes of commercials will allow advertisers to learn what is happening at various times in the commercial, and to then adjust the script to link well remembered scenes with key message points, or to rework those scenes that are remembered poorly.

REFERENCES

Bettman, J.R. (1979), "Memory Factors in Consumer Choice: A Review," Journal of Marketing, 43 (2), 37-53.

Di Lollo, V. (1980), "Temporal Integration in Visual Memory," Journal of Experimental Psychology: General, 109 (1), 75-97.

Darley, C.F. and B.F. Murdock, jr. (1971), "Effects of Prior Free Recall Testing on Final Recall and Recognition," Journal of Experimental Psychology, 91, 66-73.

Dole, A.B. a"Alternatives in Testing Recognition Memory," Nature, 196 (Oct. 6), 93-94.

Franken, R.E. and G.L. Rowland (1974), "Visual Memory: Accuracy of Decisions of Familiar-Novel for Tachistoscopically Presented Stimuli," Perceptual and Motor Skills, 38, 1039-1043.

Gillund, G. and R.M. Shiffrin (1984), "A Retrieval Model for Both Recognition and Recall," Psychological Review, 91 (1), 1 -67.

Goldstein, A.G., J.E. Chance, M. Hoisington, and K. Buescher (1982), "Recognition Memory for Pictures: Dynamic vs. Static Stimuli," Bulletin of the Psychonomic Society, 20, 1, 3740.

Haber, R.N. (1970), "How We Remember What We See," Scientific American, 222 (May), 104-112.

Postman, L., W.O. Jenkins, and D.L. Postman (1948), "An Experimental Comparison of Active Recall and Recognition," American Journal of Psychology, 61, 511-519.

Shepard, R.N. (1967), "Recognition Memory for Words, Sentences, and Pictures," Journal of Verbal Learning and Verbal Behavior, 6 (February), 156-63.

Singh, S.N. and M.L. Rothschild (1983a), "Recognition as a Measure of Learning from Television Commercials," Journal of-Marketing Research, 20 (August), 235-248.

SinghS.N. and M.L. Rothschild (1983b), "The Effect of Recall on Recognition: An Empirical Investigation of Consecutive Learning Measures," Advances in Consumer Research, Vol. 10, Richard P. Bagozzi and Alice M. Tybout, eds. Ann Arbor, MI: Association for Consumer Research, 271 -276.

Singh, S.N., M.L. Rothschild, and G.A. Churchill (1988), "Recognition Versus Recall as Measures of Television Commercial Forgetting," Journal of Marketing Research, 25 (February), 72-80.

Sperling, G. (1960), "The Information Available in Brief Visual Presentations," Psychological Monographs, 74 (11, Whole No. 498).

----------------------------------------