Journal of Experimental Psychology: Human Perception and Performance

Copyright © 1985 by the American PsychologicalAssociation VOL. 11, NO. 1 FEBRUARY 1985

H o w M u c h Is an Icon Worth?
Geoffrey R. Lofius and Carrie A. Johnson
University of Washington We report a new technique for assessing the amount of information extracted from the icon t that follows a briefly presented picture. The problem of how to measure such information was formulated in terms of how much physical exposure of a picture an icon is worth. Consider two types of stimulus presentations, each with a base duration of d ms. The first is a d-ms picture followed by an icon, and the second is a d + a-ms picture not followed by an icon. How large does a have to be so that equivalent amounts of information are extracted in the two cases? To answer this question, we showed people pictures and later tested their memory for the pictures. We found that the physical exposure duration needed to reach a particular level of performance was approximately 100 ms longer when an icon was not permitted versus when the icon was permitted. This value was independent of the base duration and the luminance of the picture. Moreover, the same value was obtained using three different kinds of memory test and four different sets of pictures. We conclude that an icon is worth approximately 100 ms of additional physical exposure duration. A reasonable explanation for this robust equivalence between icon and stimulus is that the same encoding processes are responsible for extracting information from the icon and from the physical stimulus. Therefore, any variable that affects these encoding processes must affect extraction of information from the icon and the physical stimulus in an identical manner. This prediction was confirmed for one such variable, picture luminance.

Arthur P. Shimamura
University of California, San Diego

When a visual stimulus is presented and then followed by darkness, information can be extracted from the iconic image (icon) that follows (Neisser, 1967; Sperling, 1960).
This research was supported by National Science Foundation Grant BNS82-09617 to G. Loftus. We thank Brian Williamson, Lori Sumi, and Beth Vosper for running the experiments. Vincent DiLollo, Ralph Haber, Harold Hawkins, Helene Intraub, Pardo Mustillo, Tom Nelson, John Palmer, and an anonymous reviewer provided helpful advice on earlier versions of this article. We are especiallyindebted to Ellen Markman for her extensiveand insightfulcritique of the manuscript. Requests for reprints should be sent to Geoffrey R. Loftus, Department of Psychology, University of Washington, Seattle, Washington98195.

W h e n one views such a stimulus, there is n o subjective dividing line that separates the offset o f the physical stimulus from the onset of the icon. Indeed, naive observers t h i n k that a n icon is a n extension o f the physical stimulus. They speculate, for example, that the projector b u l b or the cathode-ray tube phosphor that displays the stimulus must extinguish relatively slowly after it is t u r n e d The term icon has been used in two different ways. Sometimesit refersto an imagethat is formed immediately at the time of stimulus onset and continues atter stimulus offset. Other times it refers only to the image that follows stimulus offset. In this article, it will be used in the latter

Copyright 1985 by the American Psychological Association, Inc. 0096-1523/85/$00.75

T h e control conditions were ancillary to the m a j o r Stimulus Luminances (Millilamberts) Stimulus Adapting field Projector on. Procedure. it will show some p e r f o r m a n c e level that we t e r m p(d. Intraub .i) = p(d + a. n o m a s k was presented. Because a n icon is p e r m i t t e d with a delayed mask (Loftus & G i n n . ?). I n the delayed-mask conditions. They were randomly placed into two slide trays of 72 slides per tray. Within each mask condition there were six levels of exposure duration. The noise mask consisted of a jumble of straight and curved. To derive a m e a s u r e o f the icon's worth. one by one. Experiment 1 A n old~new recognition procedure was used in E x p e r i m e n t 1. Timing of all stimuliwas controlledby Gerbrandstachistoscopicshutters with rise and fall times of approximately 1 ms. I n each o f four experiments. 1984) b u t n o t with a n i m m e d i a t e mask (Sperling.89 purpose of the e x p e r i m e n t a n d were i n c l u d e d to resolve a n a p p a r e n t discrepancy i n the literature. then it will show a p e r f o r m a n c e level that we t e r m p(d + a.07 38. landscapes. Subsequently.300 ms. and no mask. for a total of 18 experimental conditions. T h e worth of the icon m a y n o w be characterized as the value o f a required so that p(d. no slide Fixation spot Pattern mask Bright background Black markings Gray markings Luminance 0. All relevant luminances are shown in Table 1.m a s k control condition. Each picture was followed either by a n i m m e d i a t e noise m a s k or by a noise mask that was delayed 300 ms. the m a s k i m m e d i a t e l y followed the picture's offset. were presented. if the picture is followed by a m s o f additional physical exposure d u r a t i o n b u t n o icon. I m m e d i a t e l y following the study phase was a test phase in which the target pictures. Experiment 1). ?). Suppose a picture is presented for a base d u r a t i o n o f d ms. We assess here the viability o f this hypothesis. target pictures were presented. I n the control conditions. the m a s k was delayed by 300 m s following the picture's offset. This powerful subjective experience led us to hypothesize that the i n f o r m a t i o n extracted from the icon might be parsimoniously characterized i n t e r m s o f the i n f o r m a t i o n that could be extracted from a physical extension o f the stimulus. again one by one.19 2. An experimental session consisted of a study phase followed by a test phase using the pictures in the first slide tray and then another study and test phase using the pictures in the second slide tray.43 0.m a s k control condition. I n the i m m e d i a t e . Two additional Kodak standard projectors were used to display the mask and a dim fix~ttionpoint that occurred just prior to each target. The relevant luminances of target and mask were such that when the mask was displayed concurrently with a target picture. complex. They were run in 36 groups of 5 subjects per group. The design was not completely factorial. m e m o r y for the pictures was tested. Turvey. Two i n d e p e n d e n t variables were comb i n e d in the study phase. The first was exposure d u r a t i o n .38 25. The . in a n old~new recognition m e m o r y test. 1973).m a s k condition did n o t differ significantly from performance in a n o . 1963. found that performance in a n i m m e d i a t e . prepared as 35-mm slides. which varied from 62 to 1. AND A. the target could not be seen at all.57 2.and no-mask conditions. For each tray. I n a n initial study phase.m a s k conditions. If the picture is followed by a n icon. r a n d o m l y i n t e r m i n g l e d with distractor pictures. t). One hundred eighty Universityof Washington undergraduates participated for course credit.2 G. C. a n d the second was the presence or absence o f a noise m a s k following the picture's offset. Design. The target pictures were displayed by a Kodak random access carousel projector and subtended a visual angle that ranged from 15° to 22° horizontal and from 10° to 15° vertical. There were three levels of mask condition: immediate mask. black and gray lines on a white background. The targets were 144 naturalistic color pictures. i n contrast. we reason as follows. Loftus a n d G i n n (1984) f o u n d that picture m e m o r y p e r f o r m a n c e in a n i m m e d i ate-mask condition was poorer t h a n perform a n c e in a n o . delayed mask.(1980. then w h e n it is later tested. depicting seascapes. LOFTUS. and cityscapes. JOHNSON. for inspection. because the exposure durations in the immediate-mask conditions were longer than those in the delayed. 36 pictures were presented at the time of study. Stimuli. Apparatus. depending on where the subject sat. naturalistic pictures were presented for varying exposure durations. as indicated in Table 2. The same mask was used in all four experiments. Method Subjects. p e r f o r m a n c e differences between these two conditions reflect i n f o r m a t i o n extracted from the icon. Alternatively. SHIMAMURA Table 1 off.

Qualitatively. First.050 Delayed mask 62 270 320 350 620 1. thus the delayed mask must disrupt some process other than extraction of information from the icon.5~ /Z ~'~Detoyed . which reflects information extracted from the icon. The test order was different for the two slide trays. t ~. 1978.~ '~ ~ 0. the subject was asked to respond yes or no corresponding to whether he or she judged the picture to have occurred during the study phase. in turn. We are primarily interested in the difference between the immediate. Experiment 1) found no difference between a no-mask and an immediate mask condition.165. . there was only a single false-alarm rate. . performance in the no-mask control condition was higher than performance in the delayed-mask condition. which. who found an effect of mask delay 0. For each test picture. The difference between the delayed-mask and no-mask conditions replicates a finding reported by Loftus and Ginn (1984) and indicates that a noise mask disrupts memory performance even at 300 ms following stimulus offset.050 Immediate mask 320 525 575 620 880 1. . .9 No M a s k ~ 1 ~ .and delayedmask conditions. Each data point is based on 720 observations. At the time of test.HOW MUCH IS AN ICON WORTH? Table 2 3 Exposure Durations (ms)for Experiment 1 Masking condition No mask 62 270 320 35O 620 1.7 ~ I00 msec Shift t 0. I000 (msec) . all hit rates are directly comparable to one another. all 72 pictures in the slide tray were shown in a random order for 6 s apiece.. 500 Exposure Durotion I .~__ Results Because all study conditions were intermixed within the study phase. 0. the difference between the immediate. Second. . . As is evident. Mosk Mosk ~ . . . The sequence of events on each study trial was as follows. but for each tray the order was identical over all 36 groups in the experiment. in agreement with past experiments (e. Potter & Levy. . . The bottom panel shows the delayed-mask and immediatemask conditions only. Figure l (top panel) shows hit rate as a function of exposure duration for each of the three mask conditions. 1975. Intraub (1980. Thus. First. ~ 0. Lofius & the right. There are several noteworthy aspects of these data.9 --. a 300-ms tone signaled the subjects to fixate a dim spot that concurrently appeared at the center o f the viewing field.. Each picture occurred once in each of the 18 study conditions over the 18 groups for which it appeared as a target. An intertrial interval of 8 s was followed by the warning tone signaling the start of the next trial. Figure l.. our results replicate those of Hulme and Merikle (1976).and no-mask curves is smaller at long exposure durations. 1969). and the delayed-mask curve has been displaced 100 ms. however. This point will be elaborated below.5 t 0 .) .300 18 study conditions occurred in random order with the restriction that each condition occur once during the first t8 study trials and once during the second 18 study trials. . control conditions). We did. (The top panel shows all mask conditions on the same axes. . Experiment 1: Recognition memory (hit probability) as a function of stimulus exposure duration. followed by either a 300-ms mask (in the immediatemask conditions) or 300 ms o f darkness followed by a 300-ms mask (in the delayed-mask conditions) or simply darkness (in the no-mask. which was 0.. I.g. . was higher than performance in the immediate-mask condition. . A picture was then presented. performance in all three mask conditions increased with exposure duration. Each of the 144 pictures appeared as a target for 18 of the 36 groups and as a distractor for the remaining 18 groups. . .-=. . Loftus & Kallman. . Lofius and Ginn present evidence that the picture's icon has vanished by 300 ms. ' . where absolute performance level is at about the same as it was in Intraub's experiment.

Experiment 2 In Experiment 2. I00. and weddings were prepared as 35-ram slides. Second. Apparatus. Having an icon following a 270-ms picture is therefore equivalent (at least in terms of subsequent recognition memory performance) to having an additional 370 . nameable details. which are very strong in recognition memory experiments. First. Sixty new pictures of cityscapes. 200. if a picture is not permitted an icon (immediatemask conditions). There were two levels of mask condition: immediate. we wanted to tap a different subset of memorial information than that required in a recognition test. a great deal of statistical power is needed. memory was tested in a paradigm described by Intraub (1980.69. it coincides quite well with the immediate-mask curve. One explanation for the results of Experiment l is that an icon is equivalent to an additional .and delayed-mask curves in the top panel of Figure l. 100 ms of physical exposure duration at least in terms of that subset. First. then an icon should be worth 100 ms of additional physical exposure duration no matter what subset of the memory representation is tapped. then the resulting hit rate is roughly 0.and 300-ms delayed-mask. Design and procedure.270 = 100 ms of what we term additional physical exposure duration.g. it is worth. Third. As in Experiment I. however. As noted earlier.4 G. test-position effects. SHIMAMURA on recognition memory that diminished with increasing target exposure durations. amount of extracted information) that is a monotone transformation of hit rate. The primary criterion for inclusion of a given picture was that it contain a variety of identifiable. do not come into play when each picture is tested immediately after it is presented. of 270 ms. a picture presented for a base duration.) These methodological concerns are not trivial: Because the present data analysis technique requires a very accurate assessment of the shapes of different curves. JOHNSON. we can assess the effect on it of base stimulus duration by performing a lateral translation of one of these curves relative to the other. home interiors. LOFTUS. Because an icon's worth is represented graphically by the horizontal difference between the immediate. for extracting information used in a recognition test. Each of the 60 Discussion Viewing a picture leads to a memory representation that can be conceptualized as a set of information about the picture. Six exposure durations---50. They were run in 12 groups of 4 to 9 subjects per group. number of reported details is both more sensitive than yes-no recognition and is simpler from a design standpoint (in that one does not have to counterbalance for target/distractor. 350. Note that these results must logically hold not only for the particular dependent variable--hit rate--that we have used in this experiment but also for any other dependent variable (e. and 500 ms---were factorially combined with mask condition to produce 12 experimental conditions. 150. C. the t w o principal independent variables were exposure duration and mask condition. Seventy-three University of Washington undergraduates participated for course credit. as just indicated. AND A. this value of 100 ms is independent of the picture's base exposure duration. The apparatus was the same as in Experiment I... However. our goal is to obtain a quantitative characterization of the icon in terms of how much additional physical exposure duration. it requires approximately 370 ms of exposure duration to achieve this same hit rate.g. d') or any hypothetical construct (e. The major results of Experiment l can thus be summarized as follows. Consider. Second. If this second explanation is correct. If such a picture is followed by an icon (delayed-mask conditions). having an icon is equivalent to having approximately 100 ms of additional physical exposure duration. Experiment 4) and by Loftus and Ginn (1984): Subjects were asked to report as many details as they could from a picture that they had just seen. The bottom panel of Figure l shows that when the delayedmask curve is shifted 100 ms to the right. for example. Method Subjects. d. Experiments 2 through 4 assessed this possibility by testing pictures in different ways that presumably tap different subsets of memorial information. A stronger explanation is that the memory representation that accrues from a d-ms picture followed by an icon is identical in all respects to the memory representation that accrues from a d + 100-ms picture without an icon. This is the icon's worth. Some subset of that information is used as the basis for responding in an old/new recognition test. Stimuli. This dependent variable was chosen for three reasons. landscapes.

5 I I I I I I 0 I00 200 300 400 500 0 I00 200 300 400 500 Exposure Duration (msec) ExposureDura.5 (D 105 msec Shift / / J X / t " I I I mediate Mask -- 1. E x p e r i m e n t 4).m a s k conditions seems at odds with Intraub's result.5 details). f o u n d n o significant difference between a n o . In the fight panel. the curves in Figure 2 converge at long exposure durations. A n i n f o r m a l survey o f the data indicated n o qualitative differences i n the sorts o f details that were reported as a f u n c t i o n o f the various e x p e r i m e n t a l conditions. p e r f o r m a n c e increased with increasing exposure d u r a t i o n .HOW MUCH IS AN ICON WORTH? pictures occurred once in each of these 12 conditions over the 12 groups of subjects. The present finding of a substantial difference between the delayed-mask a n d i m m e d i a t e . They were instructed. Each data point is based on 365 observations. Finally. the 60 pictures were presented one by one. " a person in the m i d d l e " or "a boat in the u p p e r left") rather than mere m e n t i o n o f some physical characteristic. Each test trial consisted of the followingsequence of events." Followingthe 20 s was the warning tone for the subsequent trial. For example. within each block of 12 trials. In an experimental session.g. subjects had 20 s to write down as many details as they rememberedfrom the target. Experiment 2: Mean number of details reported as a function of stimulus exposure duration. Followingthe experiment proper. However. the response "two people" should count as two details. with separate curves for the two mask conditions. If d u r a t i o n had been long e n o u g h to p r o d u c e the absolute p e r f o r m a n c e level o b t a i n e d by I n t r a u b (approximately 4.5 / ~" SL°ndord. T h e responses almost i n v a r i a b l y consisted o f the n a m e s o f objects (e.Error / 0. subjects were told to use their own judgment.a n d i m m e diate-mask conditions would p r o b a b l y have been m i n i m a l . Next came the target picture.5-swarning tone signaled the subjects to look at the fixationpoint.5 I I I I I I 1 I I I Delayed x ~ Mosk x ~ " ~ " 2. there were very few cases in which responses were ambiguous. 5 Results and Discussion Subjects had n o difficulty carrying out the response task. (The left panel shows both mask conditions on the same axis. Figure 2 (left panel) shows the m e a n n u m ber o f reported details as a f u n c t i o n o f exposure duration.. In such cases. t h e n the difference between the delayed. "Write down as many details as you can so that a person looking at your list would be able to reproduce the picture as accurately as possible. a n d p e r f o r m a n c e i n the delayed-mask conditions was superior to perf o r m a n c e i n the i m m e d i a t e .) . In practice. each condition occur once. the delayed-maskcurve is displaced to the right by 105 ms.tion (msec) Figure 2. using a similar paradigm. a 1. They were shown 60 pictures again. N o n e x i s t e n t details were written down in less t h a n 5% o f the trials.m a s k condition. I n t r a u b (1980. The 12 experimental conditions occurred randomly over the 60 trials with the restriction that. As i n E x p e r i m e n t l. subjects scored their own data. First. and for each picture they wrote down the number of correct details that they had originally listed.m a s k conditions.m a s k a n d a n i m m e d i a t e . The icon as a source o f i n f o r m a t i o n was assessed using the same t e c h n i q u e as in Ex- 3. They were told that "a detail" should correspond to a single object listed from the picture.

Results and Discussion The m i d d l e left panel o f Figure 3 shows m e a n rating as a f u n c t i o n o f exposure d u r a tion. T h e rationale for this d e p e n d e n t variable was m u c h the same as the rationale for u s i n g a new d e p e n d e n t variable i n E x p e r i m e n t 2. I n the m i d d l e . Design and procedure. signaling that the just-presented picture should now be rated. The 10 experimental conditions occurred in random order with the restriction that within each block of 20 trials. the experimenter explained to the subjects what an old~newrecognition test was.s u b j e c t i v e rating o f how well a picture would be r e m e m b e r e d . The apparatus was the same used in Experiments 1 and 2 except that (a) all timing and response collection was controlled by an Apple II computer. T h e fight panel o f Figure 2 shows that w h e n the delayed-mask curve is shifted 105 m s to the right. It p e r m i t t e d m u c h m o r e efficient data collection t h a n was possible i n Experim e n t s 1 a n d 2. Experiment 3 E x p e r i m e n t 3 was a replication o f Experim e n t 2 except that a t h i r d d e p e n d e n t varia b l e . the c o n c l u s i o n that the icon is worth 100 m s b e c o m e s m o r e tentative for long exposure times (e. each condition occur twice. On each trial. a beep was sounded. i m m e d i a t e . therefore. After all subjects had entered their responses. Subjects saw a series of 160 target pictures and rated each one.and delayed-mask conditions across equal performance ranges. as shown in Table 3.. with separate curves for the i m m e d i a t e a n d delayed-mask conditions. followedby the signaling tone for the next trial.w a s used to probe the m e m o r y representations o f the pictures. it coincides perfectly with the i m m e d i a t e . a l-s tone and a simultaneous fixation point signaledthe start of a trial. 2 in each of the 10 conditions. the use o f subjective ratings is methodologically very simple. the following series of events occurred. First. First. along with our expectation that an icon would be worth approximately 100 ms.. was used in Experiment 3. C. Second. and the other 160 were used in the actual experiment. Next. A new set of 240 pictures. on a scale from I to 5. If any subject wanted more practice. AND A. Apparatus. drawn from the same pool as the Experiment 1 pictures.. 300 m s or more). Pictures were followed by either an immediate or a 300-ms delayed mask. I n Experim e n t 2. subjects saw but did not rate 40 practice pictures--4 in each of the 10 experimental conditions. then all subjects saw and rated 20 more practice pictures. They were run in 10 groups of from 4 to 8 subjects per group. and (b) all projection apparatus was enclosed in a soundproof box. She then explained that their task would be to rate each of a series of pictures. They then saw and rated 20 more practice pictures. This m e a n s that at longer exposure durations. for a total of 10 conditions. Stimuli. there was a ½-sdelay. the perform a n c e curves are negatively accelerating. Followingthese instructions. T h a t is. the experiment proper began. SHIMAMURA Table 3 p e r i m e n t 1. the assessment o f horizontal differences between curves becomes increasingly less reliable. we wanted to probe the m e m o r y representation using a variety o f different depend e n t variables in order to test the hypothesis that the entire m e m o r y representation following a d-ms.6 G. Note that each immediate-mask duration is 100 ms greater than the corresponding delayed-mask duration. Exposure Durations (ms) for Experiment 3 Mas~ngcondition Delayed mask 30 60 90 120 150 Immedia~mask 130 160 190 220 250 Method Subjects.m a s k presentation. When all subjects were comfortable with the task. One s after the offset of the mask. a c o n s t a n t a m o u n t o f vertical noise (e. Subjective rating is simply a d e p e n d e n t variable that is different from the previous two. the a m o u n t indicated by the s t a n d a r d error) implies a n increasingly greater a m o u n t o f horizontal noise as the curves b e c o m e flatter.g. as to how well they would later be able to recognize it. For this reason. Each of the 160 test pictures was shown in each of the 10 conditions across the 10 groups of subjects. a n d again this value is i n d e p e n d e n t o f the base exposure d u r a t i o n o f the picture. There were five values of exposure duration within each of the two mask conditions. Each subject entered a rating from 1 (definitely would not remember the picture) to 5 (definitely would remember the picture) into a response box on his or her desk. Sixty-fiveUniversityof Washington undergraduates participated for course credit. I n b o t h E x p e r i m e n t s 1 a n d 2. JOHNSON. Eighty of these pictures were used for practice. This arrangement reflected our desire to compare immediate.m a s k curve. LOFTUS. At the start of a session. the best estimate o f the equivalent additional exposure d u r a t i o n cont r i b u t e d by the icon is 105 ms. the target picture was displayed. delayed-mask p r e s e n t a t i o n is equivalent to that following a d + 100 ms.g.

008 observations.. (The middle panels show unconditional data. the top panels show data from the 60 highest rated pictures. 1. [ Standard error I I _ _ ~ I I 0 100 200 300 0 Exposure Duration 100 (rnsec) 200 300 Figure 3. Subjects reported that their rating o f an individual picture was determined more by the idiosyncratic characteristics o f the picture than by the experimental condition in which the picture was presented. Each middle-paneldata point is based on 1. These data are shown in the bottom and top panels o f Figure 3.0. .. 2. Experiment 4 Experiments 1-3 suggest that the information extracted from an icon is equivalent to the information extracted from a 100-ms extension of the physical stimulus. c a~ 0 #f o / 2. In the right-hand panels. .0 -1" r HIGH-RATED = PICTURES ~ ~ Shift i 3. the horizontal separation between the immediate. Means for the 10 experimental con- 7 ditions were computed separately for the lowest rated 60 pictures and the highest rated 60 pictures..HOW MUCH IS AN ICON WORTH? right panel. For both high. the delayed-mask curve has been shifted to the right by 110 ms and overlaps almost perfectly with the immediate-mask curve.) . . There was indeed substantial interpicture variation.0 I [ Standard error _j_ i i i i 3.and low-rated pictures. . the worth o f a picture's icon is not affected by the picture's rated recognizability. Evidently..and lower-paneldata point is based on 378 observations. we computed a mean rating over all 10 groups for each o f the 160 pictures.47. . Experiment 3: Mean recognizability ratings as functions of stimulus exposure duration. went to an overexposed shopping center.and delayed-mask curves is 110 ms. the delayed-mask curves are displaced to the right by 110 ms.0 - J 2.5 / Imrned!0te Mask t Standard error LOW-RATED PICTURES llOshiftmsec / j 3. One hy- 4. .o J I10 msec i ~. The lowest rating. Each top...23. Accordingly. The highest rating (awarded to a picture o f a sunset over sailboats) was 4. and the lower panels show data from the 60 lowest rated pictures. .5 i ~ Delayed j ~ '. .

information is extracted more efficiently from the bright pictures than from the dim pictures. with the restriction that each condition be shown twice in each block of 40 trials. did not affect extraction of information from the icon.8 G. If this hypothesis is correct. One variable that affects information extraction from the stimulus is luminance. This finding replicates those of Loftus (1982. The logic of Experiment 4 was this. The design was the same as that of Experiment 3 except that target pictures were shown at one of two luminance levels. Suppose. information extracted from the icon. also under computer control. then such independence would not occur unless by coincidence. to be similarly affected by an identical luminance manipulation. Alternatively. while affecting extraction of information from the stimulus.and delayed-mask conditions as shown in Table 4. Bright pictures were shown at the same luminance level used in Experiments 1-3. in press). then information extraction from the icon will be similarly less efficient. This resulted in a total of 20 conditions. They were run in 20 groups of 4 to 8 subjects per group. though unaffected by luminance in an absolute sense. respectively. C.0 log units using a neutral-density filter. Dim pictures were attenuated by 2. that luminance. but in a very specific way: The information extracted from the icon of a dim picture must still be equivalent to the information extracted from an additional 100 ms of the dim picture itself. with stimilus luminance as an additional factor. the worth of an icon is identical--110 ms--for bright and dim pictures. poorer in a low-luminance (dim) condition. in press) has shown that highluminance pictures are remembered better than low-luminance pictures in both a recognition and a detail-recall test. First. Discussion The effect of luminance on extraction of information from the icon has been examined previously in a Sperling (1960) partial-report . reflecting less efficient extraction of information from the physical stimulus. the worth of an icon must be independent of stimulus luminance. In other words. AND A. as measured by recognition and detail-recall tests. and the icon was found to be worth 100 ms of additional physical exposure. the right panels show the delayed-mask curves shifted horizontally so as to provide the best overlap with the immediate-mask curves. then any variable affecting information extraction from the physical stimulus must similarly affect information extraction from the icon. Experiment 4 was essentially a replication of Experiment 3. There were five levels of stimulus duration within the immediate. Second. In this event. If the same encoding mechanisms continue to operate during the icon. Loftus (1982. Method Subjects. as expected. Results Mean rating as a function of exposure duration is shown in Figure 4. Design and procedure. The data are clear. As in Figure 3. The same stimuli used in Experiment 3 were used in Experiment 4. for example. LOFTUS. SHIMAMURA Table 4 pothesis to explain this result is that the encoding mechanisms that extract information from the icon are the same as those that extract information from the physical stimulus. Stimuli. at least as indicated by the subjective ratings. if the encoding mechanisms operating during the icon are different from those operating during the physical stimulus. Ninety-six University of Washington undergraduates participated in exchange for course credit. who found memory performance. which were presented in a random order over the 160 trials of the experiment. The same apparatus used in Experiment 3 was used in Experiment 4. was used to vary target luminance on a trial-to-trial basis. A filter wheel. Suppose that memory of an immediate-masked picture is. Exposure Durations (ms)for Experiment 4 Masking condition Delayed mask 30 60 90 120 150 Immediate mask 140 170 200 230 260 Apparatus. The procedure was identical to that used in Experiment 3. Data for dim and bright pictures are shown in the top and bottom panels. would increase for dim pictures when measured by additional physical stimulus duration. JOHNSON. The pictures of Experiments 1-3 were shown in a high-luminance (bright) condition. Luminance level was factorially combined with both mask delay and stimulus duration.

-/ I o/?m~eedsi~te /o I I ~ A I BRIGHT PICTURES 1 I _ /"x/x / X Delayed Mask /x D. Adelson and Jonides (1980) found very little effect of luminance. less information is extracted from the icon of a dim picture than from the icon of a bright picture. In the right-hand panels. in agreement with Keele and Chase (1967) and Long and Beaton (1982).. In any event. Experiment 4: Mean recognizability ratings as functions o f stimulus exposure duration. (Top panel shows data for dim pictures. Each data point is based on 768 observations. the delayed-mask curves are displaced to the right by 110 ms. and bottom panel shows data for bright pictures. whereas Keele and Chase (1967) and Long and Beaton (1982) both found better performance with higher luminances. The differences are probably attributable to different luminance levels used in the different experiments. in an absolute sense.HOW MUCH IS AN ICON WORTH? 9 paradigm.o I10 msec Immediate Mask ~L Standard error 2 n A I I I l I00 200 3 0 0 Exposure Duration I00 200 (reset) 300 Figure 4. but DIM PICTURES I I I 3 Delayed Mask x~'x/X f I10 msec Shift x ~ 2 E "4-tD rr c t3 . the present results indicate. that.) . Both are worth I l0 ms of additional physical exposure duration.

105. respectively. These considerations provide a tentative explanation for the apparent discrepancies between the results of our Experiments l and 2 and the results of Intraub's (1980) Experiments l and 4. most information-processing models assume that the rate at which information is extracted from a stimulus declines over the time during which the stimulus remains present (e.m u s t decline with stimulus duration. Conditions leading to high performance are exactly those under which the icon would be expected to make a relatively small contribution to the total amount of information extracted from the picture. and two luminance levels. the amount of information acquired from the last hundred ms of the stimulus itself---and likewise. of course. be coincidental. 100 m s . and 4. 2. l0 s--then one seems to acquire very little information during the last 100 ms of the presentation. 105-. very little information seems to be acquired from the icon of a 10s picture. AND A. If a picture is presented for a very short time--say. an icon produced the same increase in performance as did an additional 105 ms of physical exposure duration. This notion accords well with a common-sense view of amount of information. Loftus & Kallman. Likewise. for example. "strength of sensory residual" (see Hawkins & Shulman.g. JOHNSON. following base durations that ranged from 50 ms to 350 ms. Luminance affects both. and 110 ms in Experiments 1. imply that the absolute amount of information extacted from a picture's icon is independent of the picture's base duration. a picture is presented for a long timemsay. According to these models. General Discussion An icon was found to be worth 100. Figure 2. Massaro. 3. and thus longer lasting. In each experiment the value of the icon. This finding was most clearly illustrated in Experiment 2 (cf. Indeed. Like Hulme and Merikle (1976). in a more general sense. with display conditions leading to a high absolute performance level. The present results indicate that. whereas a large amount of information seems to be acquired from the icon of. Intraub found that. The experimental procedures that we used do not permit an easy assessment of whether the 100-. we found that the differences between masked and unmasked conditions were greatest at the short exposure durations where the contribution of the icon to total extracted information would be expected to be largest. If. Hawkins and Shulman (1979) attribute the Keele and Chase results to an effect of luminance on "Type II persistence"--higher luminance produces an overall greater. information extraction from the icon is tied directly to information extraction from the physical stimulus. We will proceed on that premise and tentatively conclude that two different physical stimuli--a d-ms picture with an icon . 1979.. 1970.. PsychologicalEquivalence The estimated worth of an icon differed by only 10 ms over three dependent variables. Rumelhart. Kowler & Sperling. therefore. say. In Experiment 2. in contrast. 1969). Figure 1). But it can be parsimoniously explained by assuming that the same encoding mechanisms operate during stimulus and icon.l0 G. the amount o f information extracted from the i c o n . was independent of the base duration of the picture. 1982. SHIMAMURA less information is extracted from 110 extra ms of a dim picture than from 110 extra ms of a bright picture. 110. the difference between masked and unmasked pictures was small and nonsignificant. In the present experiments. measured in terms of additional physical exposure duration. This finding does not. of course. Intuition suggests that such a small difference in times can be ignored. independent of luminance level. 1980. a 50-ms picture. This regularity could. A striking regularity emerges: The icon is worth 110 ms of additional physical exposure. LOFTUS.. left panel).t h e n one seems to acquire a large amount of information from it. and 110-ms values found in the four experiments are statistically different from one another. 1979. we measured the difference between masked and unmasked pictures over ranges of base exposure durations that produced wide variations in performance. Whatever variable affects these encoding mechanisms must therefore affect stimulus and icon in identical ways. C. Krumhansl. four sets of pictures.

The results o f identical to the physical stimulus throughout Experiment 4 provided evidence that the its existence and then abruptly vanished. a d-ms picture followed by an icon and a d + 100-ms picture It is important to distinguish between an without an icon produce m e m o r y represen.when measured in terms o f the physical ically different but psychologically equivalent stimulus that it follows. within a second or two Icon Worth Versus Icon Duration following stimulus offset. In this sense. It to disruption by a noise mass. then encoding processes that extract information worth and duration would be the same. In this situation. would mean physical stimuli psychologically equivalent? that the icon must always be worth 100 ms An extreme example of stimuli that are phys.. Within result from these two different stimuli--or at this framework. by temporal integration the degree that (a) attempts to disconflrm it (e.g. But these two kinds of stimuli may have suggested that picture perception inbe equivalent at the level of memory repre. The most extreme hypothesis permitted by our findings is that. an is found in the study of color vision. Given ulus. If the icon remained value of an icon's a representation of the picture that can be tulate that the memory representations that used in a subsequent memory test. by impairing conceptual processes. susceptible equivalent throughout the system. like any hypothesis. extract information from the physical stim. that the icon does fade--that it becomes If this mechanism. 1980. 1970a. as meahypothesis.HOW MUCH IS AN ICON WORTH? 11 and a d + 100-ms picture with no i c o n . picture from a d + 100-ms. as it affects information extraction from the To what degree are these two different physical stimulus. 1967).A typical estimate o f the icon's duration is .icon's worth.. in turn. and an icon's duration. This experiments. for example. to account for it. and a delayed mask tions required for our three performance exerts its influence on memory performance tests--are equivalent at a fairly early stage. The use of different (see Coltheart. The present situation is not as extreme. suggested by the results progressively less valuable as a source of of Experiment 4. as measured in the present tations that are identical in all respects.tion from the icon in exactly the same way lead to the same memory performance. We obtained similar results for the processes. This. 1967). Rather. delayed-mask operate past 300 ms following stimulus offset.perceptual system. continue to is easy to distinguish a d-ms. contrast.are subject to attentional strategies. fail and (b) evidence is found for a mechanism synchrony judgment (e.g. immediate-mask Potter (1976) and Loftus and Ginn (1984) picture. wherein icon's worth would be a constant of the lights with entirely different spectral compo. The exact picture sets and different ways of tapping the relationship between worth and duration is memory representation all produced the same not immediately evident. 1980. Eriksen & Collins. which tapped relatively short. Efron.volves (at least) two qualitatively different sentation. sitions can be perceived as identical. periments 2-4. The present experiments 1970b.g. Sperling.completed by 300 ms. Conceptual processes and the detail recall and rating tests of Ex.however. It is thus reasonable to pos. is indeed correct.operate on the output of perceptual processes. An from the icon are the same as those that icon is generally not assumed to act this way. then any information over the course o f its existence-variable (e. and result term memory. for a review). DiLollo.. and other paradigms satisfy both criteria.. composition) must affect information extrac. perceptual processes are least the subsets of the memory representa. cation of the picture. information as to which of the Perceptual and Conceptual Processes physically different stimuli has been presented is lost at the receptor level--the first stage in The superiority of the no-mask over the the visual system--so the stimuli must be delayed-mask conditions in Experiment 1 indicates that encoding processes. is credible to sured. it is assumed to fade. Perceptual processes operate on a delayed recognition test in Experiment 1. picture or its icon and result in the identifiwhich tapped relatively long-term memory. spatial frequency the icon's worth must be less than its duration.

& Speding. however. R. 74.. W. JOHNSON.. 25. 30. Abrupt changes in visual stimulation enhance processing of form and location information. G. t980. R. Journal of Experimental Psychology. Pola. AND A.. E (1967). G. Intraub. W. M. (1970a). R. 511-523. Journal of Experimental Psychology: General. R. 31-38.. Short-term visual storage. Two definitions of persistence in visual perception. G. The details of this model. We believe. Pola. G. Two types of information in picture memory. that the two phenomena can be integrated within a single model that assumes (a) perceptual processes consist of the extraction of information from the picture or its icon. (1980). Perceptual and conceptual masking of pictures. Picture memory: Data and methodology. (1982). M. C. rate would be relatively lower when the picture ended and thus would take a relatively shorter time to fall to the same criterion. We have replicated these effects with complex pictures (Loftus & Shimamura. M. 37-55. & Ginn. 295-303. 27. (in press). The relationship between the duration of a stimulus and the duration of a perception. Memory and Cognition. At first glance. Perception & Psychophysics. 75-97. (1975). . R. hence it would take a long time for rate to fall to some criterion. and (c) the icon remains phenomenologically present until the information-extraction rate falls below some criterion. Processing time and memory for pictures. (1980). & Jonides. Canadian Journal of Psycholog£. New York: Academic Press. E. under some circumstances. Neuropsychologia. 109. hence more information would be extracted overall. 50-ms) picture. duration and energy. Efron. 1970a.12 G. According to this model. V. & Bell. Transient stimulation does not aid visual search: Implications for the role of saccades. 435-441. & Collins. Following a short-duration (e. M. (1974). DiLollo. (b) the rate of information extraction declines as a function of the amount of information already extracted. Picture perception: Effects of luminance on available information and informationextraction rate. The effect of luminance on persistence is. J. following a long-duration (e. (1980). L. Temporal integration in visual memory. 27. & Merikle. Loftus. Canadian Journal of Psychology. Coltheart. the dependence of persistence on exposure duration and luminance seems inconsistent with the present finding that the icon as a basis for information extraction is independent of these variables. 2. Perception & Psychophysics. R. J. Presentation rate and the representation of briefly glimpsed pictures in memory. Perception & Psychophysics. Journal of Experimen- tal Psychology: Learning." Human Perception and Performance. Direct estimates of the apparent duration of a flash.. & Chase. Bowen. Visual persistence: Effects of flash luminance. Haber & Standing. Iconic memory and visible persistence. because the relevant information would be extracted faster from a bright than from a dim picture. 14. W. will be described in a subsequent article. 8. Journal of Experimental Psychology. & Standing.. 1-10. information extraction rate would still be high at the time the stimulus ended and the icon began. Krumhansl. (1980).. L (1979). 104. G. E. In C. H. E M. (1970b). the bright-picture rate would drop below the dim-picture rate. (1982). 1-12. 486-493. 1985). Conversely. Handbook of research methods in human memory and cognition (pp. and its fit to the data. 216-229. R. R.. Efron. the rate would drop more quickly.. J. Journal of Experimental Psychology. Vision Research. Puff (Ed. S. It has been shown that the subjective duration-the persistence--of an icon decreases with increasing stimulus duration and stimulus luminance (Bowen." Human Learning and Memory. the rate would remain above criterion longer for a dim than for a bright picture. L. H.. Eventually. Indeed Coltheart (1980) cites this kind of inconsistency as evidence that visual persistence and information extraction result from different mechanisms. (1984). (1976).). 6. Perception & Psychophysics. C.. & Matin. DiLollo.. L. Eriksen. C. 32. The estimate of 1 0 0 ms for the icon's worth is thus reasonable. 6. Hulme. (1980). Journal of Experimental Psychology: Human Learning and Memory. SHIMAMURA on the order of 250 ms. 24. & Matin. W. 103-113. 1970). Perception & Psychophysics. 383-385. Effect of stimulus duration on perceptual onset and offset latencies. thus. (1970). 0967). R. & Shulman. 8. somewhat more complex. Kowler. 348-350. Some temporal characteristics of visual pattern perception. 257285). S. 476-484. t974. The psychophysics of visual storage. R. 300-ms) picture. according to the model. Loflus. Efron. Keele. Haber. H. References Adelson. However. G. Loftus.g. LOFTUS. Journal of Experimental Psychology: General Loflus. 183-228. N. 231-234. stimulus duration would affect persistence as follows. 1970b. Perception & Psychophysics.g.. A bright picture would start with a higher information-extraction rate than would a dim faicture. G. 10. Hawkins.

(1970). died on July 19. Cognitive psycholog3z New York: Appleton-Century-CroRs. Rumelhart. Received March 30. & Levy.Jouma/ of Mathematical Psychology. R m #107 25 Cross C a m p u s R o a d Storrs.. Journal of Experimental Psychology: Human Learning and Memory.. G. 7. & Beaton. Massaro. M. Potter. 1984 • Editor for Psychological Bulletin N a m e d : Search for New Editor Continues David Zeaman. 1984. C. H. U. A model for visual memory tasks. & Shimamura. Perceptual processesand forgetting in memory tasks. J. M. Journal of Experimental Psychology: Human Learning and Memory. (1982). Short-term conceptual memory for pictures. G. T. prepare a statement o f one page or less in support o f each nomination. 2. (1979). Psychological Review. To nominate candidates. Picture perception: Information extraction and visualpersistence Manuscript in preparation. (1969).. Turvey. Neisser. House. 10-15. Connecticut 06268 APA's Publications and C o m m u n i c a t i o n s Board is continuing its recently opened search for a new editor. 1-52. Effective immediately. A multicomponent theory of the perceptionof briefly exposed visual displays. 5. (1960). 191-218. A. Journal of Experimental Psychology. M. (1963). J. E. W. D. Psychological Review. I. Sperling.HOW MUCH IS AN 1CON WORTH? Loftus. On peripheral and central processes in vision: Inferences from an information-processing analysis of masking with patterned stimuli. 13 Potter. (1969). L y m a n Porter. Human Factors. and one o f the journal's associate editors. Journal of Experimental Psychology: Human Perception and Performance. R. M. P. Sperlin~ G. Betty J. G. . 509-522. Editor Psychological Bulletin D e p a r t m e n t o f Psychology U-20. 1984 Revision received June 28. 197-211. (1976). (1985). Psychological Monographs. Encoding and use of detail information in picture recognition. 77. editor o f Psychological Bulletin. (1973). Long. 1985 to the chair o f the search committee: Barbara Strudler Wallston Box 512 Peabody Vanderbilt University Nashville. T h e term o f editorship is from 1987 t h r o u g h 1992. 81. 80. D. (1967). R. Paul Mussen. 5. 19-31. & Kallman. The information available in brief visual presentations. Candidates for the journal editorship must be m e m b e r s o f APA and should be available to start receiving manuscripts in early 1986 to prepare for issues published in 1987. 8. Z e a m a n ' s colleague at the University o f Connecticut. 1-29. Wilbert McKeachie. 557-567. E. G. R. Recognition memory for a rapid sequence of pictures. House. Tennessee 37203 The other members o f the search committee are Elizabeth Lofius. 383-391.. Loftus. will complete David Z e a m a n ' s term and serve as editor t h r o u g h 1986. authors should submit manuscripts to: Betty J. 74. C. The case for peripheral persistence: Effects of target and background luminance on a partial-report task. and Lee Sechrest. Submit nominations no later than February 1..