Journal of Experimental Psychology:Human Learning and Memory1977, Vol. 3, No. 4, 386-396

Long-Term Memory for Pictures

Jean M. Handler and Gary H. RitcheyUniversity of California, San Diego (La Jolla)

The concept of a scene schema was used to predict the kinds of informationthat will be remembered from complex pictures over relatively long periodsof time. Recognition of eight types of transformations on both organizedand unorganized pictures was tested either immediately following presenta-tion or at intervals of a day, a week, or 4 months. Certain kinds of informa-tion in organized pictures remained virtually intact over a 4-month period,whereas other kinds disappeared completely. It was concluded that a sceneschema contains an inventory of objects in a scene and their locationsrelative to each other, but it does not include descriptive information aboutthe appearance of the objects or the overall spatial composition of the scene.

People remember places and faces for along time. Bahrick, Bahrick, and Witt-linger (1975) have shown that people canrecognize pictures of high school classmateswhom they have not seen for many years.Even when subjects are briefly exposed toa large number of new pictures, recognitionrates tend to remain very high for severaldays (Nickerson, 1968; Shepard, 1967;Standing, Conezio, & Haber, 1970) andremain above chance for up to ayear (Fajnstzejn-Pollack, 1973; Nickerson,1968). However, these studies have notcontrolled the relationship of distractorsto targets and so cannot specify the basisfor such good performance. Just what dopeople remember about a picture, the namesof the objects it contains, the relationshipsamong the objects, their physical appear-ance? We know little about these aspectsof picture memory.

In a recent study, Mandler and Parker(1976) explored retention of several specific

This research was supported in part by NationalInstitute of Mental Health Grants MH-24492 andMH-15828. The second author was supported by aNational Science Foundation predoctoral fellowship.We would like to thank George Campbell forstimulus preparation and members of the JMM labfor help throughout.

Requests for reprints should be sent to Jean M.Mandler, Department of Psychology C-009, Uni-versity of California at San Diego, La Jolla, Cali-fornia 92093.

kinds of information. They found that ifpictures formed "real-world" scenes, spatialrelationships among items in the pictureswere retained very well over a week's time.If, however, the pictures consisted ofhaphazard collections of the same items,retention of spatial relationships after aweek was close to chance. The physicalappearance of the items, although not aswell encoded in the first place, showedlittle loss over a week for either kind ofpicture. These data suggest that not onlydo different kinds of information decay atdifferent rates but that they are alsoaffected by the degree to which the picturesare organized, that is, whether or not thepictures activate familiar schemata.

When we speak of the organization of apicture, we are referring to the extent towhich familiar objects are related to eachother in expected ways, not to any intrinsicstructure in the stimulus itself. For example,Goldstein and Chance (1971) found thatpictures of faces were better retained over a2-week period than either snowflakes orinkblots. Although snowflakes are highlysymmetric configurations, and thus in somesense well organized, people obviously havenot incorporated many of their details intoa schema that can be used to facilitaterecognition.

Schemata of faces or scenes are builtup through experience in the world. Justas we develop a set of expectations about

386

MEMORY FOR PICTURES 387

events and their sequences in daily life,we develop expectations about objects andtheir relationships in the visual realm. Theorganized street scene in Figure 1 is anexample. You have never seen the picturebefore, but the objects in it have a rela-tively high probability of occurring to-gether and the spatial relationships amongthem are also highly probable in terms ofyour past experience. The same objectsin the unorganized version are not arrangedin previously seen ways and may not evenbe identified as a functional class. Detailsof the physical appearance of the objectsin both scenes are less probable and may beless important to recognition of the scene.

The notion of schemata controllingmemory is an old one; what a schemaconsists of is less well understood. Wewould like to be able to specify moreexactly the kinds of information containedin a scene schema so that we can predictwhich aspects of pictures people willremember and which they will forget. Inprevious research (Mandler & Johnson,1976; Mandler & Parker, 1976; Mandler& Stein, 1974) we have investigated fourtypes of information:

1. Inventory information, specifyingwhat objects a picture contains.

2. Descriptive information, specifyingthe figurative detail of the objects in theinventory (i.e., what the objects look like).

3. Spatial location information, specify-ing where the objects are located, includingtheir relations to other objects, such as"left of," "facing," and so forth.

4. Spatial composition information, speci-fying areas of filled versus empty space inthe overall composition of the picture,without regard to the nature of the itemsfilling spaces.

The taxonomy does not include actionsor reference to what is inferred to behappening in a picture. Such inferences arepresumably also controlled by real-worldschemata, by our knowledge of humanmotivation and expected sequences ofevents, but the present taxonomy empha-sizes the more static, figurative charac-teristics of scenes.

in n[ i" * :«i /35,!™.~— ~~ '!*•"*'*[] |f —*. A-

ir

Figure 1. An example of an organized and un-organized version of a picture.

Our use of the term "information" inpictures differs from that used by someother investigators in this area. Loftus andBell (1975), for example, define the in-formative details of a picture in the senseof information theory. More precisely, theydefine an area of detail as informative "tothe extent that it has a low a priori proba-bility of being there given the rest of thepicture and the subject's past history. Thus,for example, in a picture depicting a farmscene, a tractor would be an uninformativedetail, whereas an octopus would be aninformative detail" (p. 104).

In the information-theoretic sense, weare investigating "uninformative" details,since we are studying familiar, expectedobjects in pictures. In our use of the term,an object is informative if it activates aschema of a particular type of scene; thus,a tractor is more likely to contribute tothe activation of a farm schema than is anoctopus. We assume that both encodingand retention of some kinds of informationare enhanced by activation of such aschema.

Loftus and Bell (1975) make a furtherdistinction between informative details andgeneral information in a picture. At thevery brief exposure times they studied,general information may be equivalent tospatial composition information. They sug-gest that general information, that is, thefamiliarity of a picture, accrues continu-ously as a function of exposure time, butthat finding an informative detail increasesfamiliarity by a quantum jump. Althoughthis may be a sensible account of encodingduring a brief exposure time, a distinctionbetween gradually accruing informationand acquisition of details is less useful whenlonger study times are used. We assume

388 JEAN M. HANDLER AND GARY H. RITCHEY

that general information consists of thesum of details that subjects encode andthat "details" are denned by the four typesof information we have listed.

Mandler and Johnson (1976) studiedrecognition of the four types of informationby using five types of distractor. A typechange, in which an object is replaced by aconceptually different object, was used toassess recognition of inventory information.A token change, in which an object isreplaced by another object of the sameconceptual class, but differing in details ofits appearance, was used to assess recogni-tion of descriptive information. A rearrange-ment, in which two objects of similar sizeand shape are interchanged, was used toassess spatial location information. A movechange, in which an object is moved slightly,but not enough to affect basic left-rightrelations among objects, was used to assessspatial composition information. A deletion,in which an object is removed from thepicture, was also used to assess spatialcomposition, although this change affectsinventory information as well. In thatexperiment, speeded responses were re-quired and it was assumed that spatialcomposition would be the primary basison which a deletion would be detected.Parker (Note 1) has shown that deletionsare detected faster than type changes,providing additional evidence for thisassumption.

Mandler and Johnson (1976) foundthat immediate recognition depended onthe type of information that was variedin a given distractor and also on whetherthe pictures were organized or unorganizedscenes. Spatial location information wasbetter recognized in organized scenes, andspatial composition information was betterin unorganized scenes. Neither inventorynor descriptive information was affectedby organization.

The present experiment added three newtransformations. An addition, in which anew object is added to the picture, shouldtest the same kind of information asdeletion and therefore should be affectedby the organization variable in the sameway. The other new transformations were

an orientation change, in which an objectis reversed in its left-right orientation, anda size change, in which an object is madelarger or smaller. In a previous study,Mandler and Parker (1976) treated bothsize and orientation as descriptive charac-teristics of objects, because recognition ofeach object was tested individually, out ofthe context of the rest of the picture. Inthe present experiment, both types oftransformation took place in context andthus affected the relationships amongobjects. This type of relational informationhad previously been included in the cate-gory of spatial location information, whichwas defined to include relational informa-tion such as "facing." A more descriptiveterm for this category would be spatialrelation information, and we will use thisterm hereafter. Orientation and locationof objects relative to each other clearlybelong to this category, but relative sizeof objects is less clear-cut, since a changein size affects the spatial composition of apicture in addition to changing spatialrelations.

Retention of the four kinds of informa-tion over a period of 4 months was studied,using both organized and unorganizedpictures. It was expected that some kindsof information would be retained longerthan others and furthermore that theorganization of the pictures would differ-entially affect retention of the variouskinds of information. Just as informationrepresenting the meaning of sentences isbetter retained than information havingto do with their surface structure (e.g.,Graesser & Mandler, 1975; Kolers &Ostry, 1974; Sachs, 1974), so we wouldexpect information conveying the centralmeaning of a picture to be better retainedthan surface details. Our understandingof the way in which visual information isintegrated is not yet sufficient to makedetailed comparisons of types of informa-tion in pictures with syntactic, physical,and phonological aspects of verbal material.Nevertheless, the spatial composition of apicture (representing areas of filled andempty space without regard to content)and the descriptive details of objects seem

MEMORY FOR PICTURES 389

less central to the meaning of a picturethan inventory and spatial relationinformation.

Inventory and spatial relation informa-tion should not only be longer retained butshould also be affected in a major way bythe extent to which the pictures areorganized. Inventory information may beencoded and retained for brief periods,regardless of the structure of the picture,yet not be available at a later time unlessorganized in a stable fashion. Mandlerand Johnson (1976) suggested that ascene schema may not contain preciseinventory information because they foundno effects of organization on recognitionof inventory information in an immediatetest. However, inventory information maybe successfully encoded from an un-organized picture, perhaps as a list ofitems, yet not remembered over the longrun because the list itself is not organizedin the form of a schema. Therefore, after4 months we would expect type transforma-tions to be better recognized in organizedpictures. Similarly, spatial relations, astested by rearrangement and orientationchanges, should also prove superior inorganized pictures. The effect of sizechanges is less clear-cut for the reasonsdiscussed earlier.

Recognition of token changes, represent-ing descriptive detail, should be unaffectedby scene schemata; thus, performance onthis transformation would not be expectedto differ for the two kinds of picture. Sceneschemata deemphasize spatial compositioninformation in organized pictures. However,if this kind of information is difficult toretain, moves should be poorly recognizedfor both kinds of picture after a longretention interval. Deletion and additionhave an ambiguous status. To the extentthat they are detected on the basis ofspatial composition information their long-term retention should also be poor; if theyare detected on the basis of a change in theinventory, they should be better retainedin organized pictures. A comparison ofimmediate and long-term recognition ofthese two transformations should clarifythe basis on which they are recognized.

Method

Subjects

Ninety-six undergraduates enrolled in introduc-tory psychology classes at the University of Cali-fornia, San Diego, participated in the experimentfor class credit. Twenty-four subjects were assignedto each of four retention intervals; half of each groupviewed organized pictures and half viewed un-organized pictures.

Stimuli

The set of organized target pictures consisted ofeight 35-mm slides of black and white line drawingsof naturalistic scenes. These pictures were similar tothose used in Mandler and Johnson (1976) exceptthat each contained approximately six objectsinstead of eight. As a result, the pictures werenoticeably less dense in appearance. The extrablank spaces created by reducing the number ofobjects allowed an object to be added withoutcrowding and also allowed move changes of some-what greater magnitude than previously used. Fourpictures were outdoor scenes and four indoor;within this division, two of each type includedseveral people, and the other two were primarilyinanimate in character.

The set of unorganized pictures was generated byremoving the perspective line from each organizedpicture, inverting the picture, and then rotatingeach object back to its normal upright position.Then four of the six objects were interchanged tocreate essentially a zero correlation between hori-zontal and vertical coordinates of objects in the twoversions. This method of disorganizing picturesdisrupted meaningful relations among objects whilemaintaining approximately the same density ofobjects in both versions of a given picture. Anexample of an organized and unorganized versionof a picture is shown in Figure 1.

Eight transformations were generated for eachpicture to create distractors for the recognitiontests: (a) addition—an object compatible with themeaning of the scene was added to the picture; (b)deletion—an object was removed from the picture;(c) type change—an object was replaced by a con-ceptually different object of the same size and shape;(d) token change—an object was replaced by anotherobject of the same size, shape, and conceptual class,but which differed in details of appearance; (e)rearrangement—two objects of approximately thesame size and shape were interchanged, with re-arrangements made in the horizontal plane so thatapparent size would not be altered in the organizedscenes; (f) orientation change—an object was re-versed so that it faced in the opposite direction;(g) move—an object was moved in the horizontalplane by a distance equal to its width (half of themoves were toward the center of the picture andhalf toward the periphery); (h) size change—an

390 JEAN M. HANDLER AND GARY H. RITCHEY

WfraT

Figure 2, Examples of the eight transformations.(The top left picture shows an addition, the topright a deletion. In the second row, a type changeis on the left, a rearrangement on the right. In thethird row, an orientation change is on the left, atoken change on the right. In the bottom row, amove change is on the left, a size change on the right.)

object was increased by 67% in area or decreasedby 40% (so that a 67% increase would be requiredto match its original size). Half of the size changeswere increases, half decreases.

An example of each type of transformation for anorganized scene is shown in Figure 2. Transforma-tions were chosen so that they did not violatereal-world relationships or change the meaning ofany scene. No object in a picture was used in morethan two transformations. Transformations wereassigned approximately equally often to large andsmall objects, to each location, and to animate orinanimate objects. The same objects were trans-formed for the organized and unorganized versionsof a given picture.

Apparatus

A Kodak random-access slide projector with aLafayette shutter was used to present 42 X 63 cmimages on a screen positioned 1.81 m in front of thesubject, with the image slightly above eye level.

Exposure duration was controlled by a Huntertimer.

Design

The between-subjects factors were organizationof the pictures (two levels) and retention interval(four levels: immediate test, a day, a week, andapproximately 4 months). The 4-month intervalvaried by a few days depending on success in reach-ing subjects. Each cell of the design contained 12subjects. Each subject saw 8 targets and was testedon recognition of all 8 transformations. A completerecognition test of 8 pictures with 8 transformationson each consists of 64 targets and 64 distractors ifprobability of old and new stimuli is to be kept at.50. To reduce learning effects during the course ofthe recognition test, each subject saw only half ofthe recognition items (each target four times andeach type of transformation four times). Thus, ittook 2 subjects to generate a complete set of within-

Table 1Proportion Correct on Targets andTransformations for Organized (0) andUnorganized (U) Pictures at Four RetentionIntervals

Target &trans-

formation

TargetsOU

AdditionOU

DeletionOU

Type changeOU

Rearrangement0U

Orientationchange

OU

Token changeOU

Move0U

Size changeOU

Retention interval

Imme-diate

.77

.76

.88

.94

.85

.92

.94

.94

.90

.81

.73

.63

.67

.54

.54

.58

.44

.33

Day

.81

.74

.88

.90

.81

.85

.96

.85

.94

.67

.65

.52

.63

.69

.42

.54

.31

.29

Week

.77

.67

.83

.98

.77

.77

.92

.92

.85

.60

.56

.40

.58

.56

.31

.50

.19

.40

4-Month

.62

.59

.81

.77

.88

.73

.90

.71

.83

.48

.52

.52

.46

.54

.56

.58

.42

.40

MEMORY FOR PICTURES 391

I 1 1 1 1 1ADDITION DELETION

~i 1 1 1 1 1 1 1 1 1 1 r

TYPE REARRANGEMENT ORIENTATION TOKEN

—i 1 1 1 r

MOVE SIZE

ORGANIZEDUNORGANIZED

3.0

2.5

2.0

1.0

5 -

V-5

I-t-D W 4M I+D W 4M I+D W 4M I+D W 4M I+D W 4M I+D W 4M I+D W 4M I+D W 4M

RETENTION INTERVAL

Figure 3. The d' scores for recognition of eight transformations on organized and unorganizedpictures at various retention intervals. (I + D represents immediate and day tests combined.W = week and 4M = four months.)

subjects data points. Subjects were assigned topairs and their data combined for purposes ofanalysis, in effect reducing the number in each cellof the design from 12 to 6 subjects.

Six orders of presentation of the eight targetswere used, one for each pair of subjects in a cell.For any given order of presentation, the recognitiontest was arranged so that seven different picturesalways occurred between successive versions of agiven picture. The various types of transformationswere scattered throughout the recognition test sothat equal numbers of each type appeared in eachquarter of testing. No more than three transforma-tions or three targets appeared in sequence.

Procedure

Subjects were tested individually. They were toldthat they were to view a series of eight slides for10 sec each and to study each slide as completelyand carefully as possible. Subjects in the day, week,and 4-month conditions then left, with no furtherinstructions until they returned after the appropriateinterval. At the start of the recognition test, subjectswere told that they would see a series of slides, halfof which would be exactly the same as the originalslides they had seen and half different. To informthem of the types of transformations they would see,a practice set was then given, consisting of a targetpicture that they had not seen before followed by amixed series of repetitions of the target and eachtype of transformation. Subjects were required torespond "same" or "different," and their responseswere either confirmed or corrected.

Following the practice session, subjects were givena response sheet and told to record an "S" for sameor "D" for different for each of the 64 test slidesand to circle a confidence rating from one (low) tothree (high) for each of their responses. Subjectswere instructed to call out "same" or "different"as soon as they had decided and then to record theiranswers on the response sheet. The 64 test slideswere then presented. Each slide was terminated assoon as the subject had responded or after a maxi-mum of 8 sec; the interval between slides wasapproximately 5 sec. In the immediate test condi-tion, the interval between presentation of targetsand onset of the recognition test was approximatelyS min. After the recognition test subjects weredebriefed.

Results

To facilitate comparison with otherstudies, the mean proportion correct scoreson target pictures and the eight types oftransformations for the various conditionshave been provided in Table 1. However,since hit rates and false-alarm rates variedas a function of retention interval andorganization of the pictures, all statisticalanalyses used d' scores. The rejectionregion was set at p < .05. Breakdowns ofinteractions used analyses of simple maineffects, or in the case of retention interval,pairwise individual comparisons.

392 JEAN M. HANDLER AND GARY H. RITCHEY

The d' scores of interest are presentedin Figure 3. An analysis of variance wasused to assess the effects on recognitionof retention interval, organization of thepictures, and transformations. There wasa large main effect of retention interval,F(3, 40) = 19.97, MSe = .695. A break-down of this effect showed that there wasno significant difference in performancebetween the immediate and day tests.Accuracy dropped significantly from a dayto a week and from a week to 4 months.The immediate and day tests have beencombined in Figure 3 for simplification ofpresentation.

Performance varied markedly on organ-ized and unorganized pictures, F(l, 40)= 10.56, MSe = .695, and also as a functionof type of transformation, F(7, 280)= 70.06, MSe = .352. In addition, therewas an interaction between these twovariables, F(7, 280) = 6.01, MSe = .352.Analysis of the simple main effect of organi-zation for each transformation indicatedthat only type, rearrangement, and orien-tation were better recognized in organizedpictures.1

The most interesting question concernedthe temporal effects of organization onrecognition of the various transformations.Figure 3 illustrates the significant three-wayinteraction of retention interval, organiza-tion, and transformations, F(2l, 280)= 1.64, MSe = .352. The chief finding isthe differential loss over time of variouskinds of information in the two types ofpicture. In organized pictures, there wasno significant loss over 4 months in therecognition of addition, deletion, and typechanges; that is, subjects were almost asaccurate in detecting these changes after 4months as at immediate test. In unorgan-ized pictures, however, recognition of thesetransformations declined significantly. Atimmediate test, the differences on thesetransformations between organized and un-organized pictures were not significant,although deletion was marginally superiorin unorganized pictures (p < .06); by 4months, recognition of these transforma-tions was markedly poorer than in organ-ized pictures. The crossover effect for

addition and deletion suggests that spatialcomposition information was of use inrecognition at the early stages of testingbut that this type of information was goneby 4 months.

Concerning spatial relation information,there was only a marginal decline in re-arrangement over 4 months in organizedpictures, but a highly significant decline inunorganized pictures; after 4 months,recognition of this transformation in un-organized pictures was essentially at chance.In general, orientation information was lesswell retained in both types of picture thanwas location. However, it differed in thetwo types of picture, reaching chancelevels of performance in the week test forunorganized pictures, but not until 4months for organized pictures.

Size changes were poorly recognized atall retention intervals for both types ofpicture, as was the measure of spatialcomposition information, move. Recogni-tion of descriptive information, measuredby token changes, did not differ for thetwo kinds of picture; performance for bothwas essentially at chance after 4 months.2

Confidence judgments closely paralleledaccuracy. There was a large effect oftransformations on confidence, F(1, 280)= 16.00, MSe = .139, and the rank orderof confidence correlated 1.0 with rank orderof accuracy on the eight transformations.

1 As found in previous studies in this series, theeffects of these variables and their interaction weresignificant when pictures instead of subjects wereused as a random variable, indicating generality ofthe findings across the stimulus set.

2 The difficulties in interpretation of this kind ofinteraction, due to the wide range of d' scores,should be noted. In particular, poorly recognizedtransformations suffer from a floor effect as a func-tion of retention interval. When only the four bestrecognized transformations were used in the analysis(all involving inventory and spatial-relation in-formation) there was a significant interaction be-tween organization and retention interval, F(3, 40)= 5.44, MSe = .59, with no significant variationdue to type of transformation. Because of the rangeof scores being considered, we have stressed onlywhether various transformations showed a changein performance over 4 months and the point atwhich performance reached chance levels.

MEMORY FOR PICTURES 393

Confidence dropped as retention intervalincreased, ^(3, 40) = 5.32, MS, = .078;however, the drop was significant only atthe 4-month interval. Subjects were alsomore confident of their judgments fororganized pictures, .F(l,40) = 4.32. In aprevious experiment, Mandler and Johnson(1976) had suggested that subjects had agreater subjective feeling of knowing organ-ized pictures well, and the higher confidencejudgments support that suggestion.

There was a large interaction betweenconfidence in "same" and "different"judgments and old and new stimuli; that is,confidence varied for correct rejections,hits, false alarms, and misses, F(l, 40)= 153.36, MSe = .027. The differences inconfidence were ordered in the same wayas reaction time to respond on the fourtypes of trials (cf. Mandler & Johnson,1976). Overall, subjects detect transfor-mations most rapidly and are most confi-dent of correct rejections. The time toreach a decision of "same" takes longer,and whether the trial is a hit or a falsealarm, subjects are less confident of theirdecision. The decision to say "different"to a target comes after the longest searchand is associated with the lowest level ofconfidence.

The confidence judgments were used todetermine receiver operating characteristiccurves for the four retention intervals. Thedata produced excellent straight-line fits(r2 = .97 or higher in all four cases),indicating normal variance for each of thedistributions. The slopes were all greaterthan 1.0 (1.59, 1.73, 1.87, and 1.25 for thefour intervals, respectively), indicatinggreater variance in the distractor distri-bution than in the target distribution.Although such a finding is relatively rarein the signal detection literature, it seemsto be a predictable result of a "noise"distribution that consists of both similarand dissimilar distractors.

Discussion

Long-term memory for complex scenesis indeed very good, but it cannot berepresented as a gradually fading copy.

Certain kinds of information remain virtu-ally intact over a 4-month period, whereasother kinds disappear completely. We willsummarize the findings in terms of the fourtypes of information we have identifiedand relate them to the predictions madeabout the influence of scene schemata onmemory.

Spatial Relation Information

Previous work in this series had shownthat relative locations of objects werebetter encoded and retained in organizedpictures. This result was confirmed andextended in the present study and amplifiedby inclusion of orientation changes.Memory for both location and orientationof objects was superior in organized pic-tures, although orientation was less wellrecognized than locations. It should bestressed that the rearrangement and orien-tation transformations did not alter themeaning or likelihood of the pictures; thus,recognition did not depend on violation ofreal-world information. The location andorientation of a car and a truck relative toeach other (as long as they are not engagedin meaningful interactions, such as a chaseor a crash) seem insignificant in terms ofthe schema of a street scene. Yet relativelocations are remarkably well retained overa period of 4 months. Orientation, however,although temporarily supported by a sceneschema, is not retained over the long run.It is of interest that the important aspectof orientation is its relational nature.Mandler and Parker (1976) did not finddifferences in retention of orientation overa week's time in organized and unorganizedpictures. However, they tested recognitionof each object presented alone. Whentested in isolation, orientation of an objectis a descriptive characteristic; when testedin context, it affects spatial relations.

Comparable statements might be madeabout recognition of size changes in andout of context. However, even though thesize changes were larger in this experimentthan in the Mandler and Parker (1976)study, they were not well recognized. Evenat immediate testing, a 67% increase in

394 JEAN M. MANDLER AND GARY H. RITCHEY

size of one object in the context of fiveothers tended not to be recognized, al-though it is a perceptually discriminatechange.8 Our unavailing efforts to achievememorable size transformations suggestthat relative size is usually poorly processed,and unless changes are large enough toviolate real-world relationships among ob-jects they will not be detected even whenpresented in the context of a real-worldscene.4

Descriptive Information

The present data confirm the earlierfindings that information about the figura-tive detail of items is not influenced byorganization nor is it well retained overtime. Thus, the greater accuracy of spatialrelation information in organized picturesis not attained at the expense of processingthe details of individual objects; this typeof information is unimportant in bothkinds of picture.

Inventory Information

The primary measure of this informationwas the type transformation. In general,type changes were better recognized inorganized pictures. Although not signifi-cantly better at immediate testing (as alsofound by Mandler and Johnson, 1976),inventory information was better retainedover time in organized pictures, and at 4months it remained the best remembered ofall the information tested. It is possiblethat an effect of organization even atimmediate test would be found if morecomplex pictures were studied.

Spatial Composition Information

As measured by the move transforma-tion, this information was poorly recog-nized even at immediate testing. Thisfinding was somewhat surprising, since thechanges were larger than those used byMandler and Johnson (1976). However,the pictures were less dense than those pre-viously used, since there were only six ob-jects in the space formerly containing eight;

the larger expanses of blank space mayhave reduced the importance of spatial-composition information. Percent correctscores on move were significantly higher forunorganized than for organized pictures,similar to the previous findings; however,the d' differences were not significant inthe present study because of the lower hitrates for unorganized pictures. This kindof discrepancy can occur when percentcorrect scores are below chance. In thecontext of more salient transformations,a move (or size) change that is not differ-entiated from the targets will be correctat approximately 1 — hit rate and thereforecan average below 50% correct. However,d' scores should not range much below zerounless schematic distortion occurs, that is,if a particular distractor is preferred over atarget. Since the d' scores did not rangesignificantly below zero, the most likelyhypothesis is that differences on the movetransformation were masked by floor effectsin the d' analyses.

Spatial composition is also tested byaddition and deletion, although compli-cated by the fact that these transformationsmeasure inventory information as well. Itis of interest that addition and deletiondiffered from the type change as a functionof organization. Aside from move, theywere the only transformations with higherscores in unorganized pictures at shortretention intervals. At 4 months, recogni-tion of these changes, along with type, waspoorer for unorganized pictures. The mostplausible explanation is that recognition ofaddition and deletion can be mediatedeither by spatial composition or inventoryinformation. At immediate test, the spatialcomposition information is available, but

8 In a matching-to-sample test, subjects alwaysdetect these size changes, but they find them only74% of the time within the 8-sec interval used in therecognition test. Parker (Note 1) found that changesof this magnitude are detected in approximately 2sec in a recognition test when a single picture hasbeen intensively studied.

4 Mandler and Stein (1974), however, found thatchildren recognized 50% size changes significantlybetter in organized pictures.

MEMORY FOR PICTURES 395

as it drops out, these changes can only bedetected on the basis of the inventory ofobjects. Since inventory information isbetter retained in organized pictures,recognition of addition and deletion inunorganized pictures suffers over the longrun.

The exact role of spatial compositioninformation in picture recognition needsfurther clarification. Mandler and Johnson(1976) found marked differences in recog-nition of deletions and moves compared toother transformations, and Parker (Note 1)has shown that deletions are much morerapidly detected than other types oftransformations. Just how long spatialcomposition information is retained is stillunknown, but the present data indicatethat it is not available in long-termstorage.

Our conclusion from this summary ofthe temporal effects of organization onrecognition is that a scene schema containsan inventory of objects and their locationsrelative to each other. It does not includedescriptive information about the objectsor the overall spatial composition of thescene. It contains some information aboutrelative orientations of the objects, butthis information is not stable. The dataalso indicate that over long retentionintervals recognition comes to depend moreheavily on a scene schema than it doesimmediately after encoding. As suggestedby Mandler and Johnson (1977), verbalmaterials, such as stories, can be encodedand temporarily maintained even if theyfit no familiar schema, but over time theirretrieval becomes schema dependent. Simi-larly, in the visual realm, after 4 monthscertain kinds of information, such as thephysical' appearance of objects, can nolonger be retrieved. Other kinds of infor-mation, such as spatial location, will onlybe retrieved if they match a familiarschema.

In this investigation of scene schematawe have focused on the visual representa-tion of scenes, that is, the kinds of thingsyou would expect to see if you looked at ascene, rather than the general knowledgeyou have about properties of the world.

It may be difficult to differentiate visualfrom semantic knowledge since the two areso intertwined, yet we would suggest thatreal-world schemata come in various forms,each containing different kinds of informa-tion useful for negotiating the environment.A visual schema may emphasize differentkinds of information than a verbal schemaor description of a scene.

The location of objects may be lessimportant in a schematic description of aroom than in its visual representation. Inthe latter case, locations of objects arenecessary to provide a frame of referencefor possible actions. It seems reasonablethat the kinds of information retained orlost from visual input over the long runreflect their relative utility for action.Spatial composition, informing us of thepresence or absence of objects, is momen-tarily important because it provides theperipheral context allowing us to remainoriented in space. Orientation of objects,in turn, is of considerable importance at themoment of direct interaction with them,such as sitting down on a chair. Whetheror not a room contains expected kinds offurniture, and where the furniture islocated, however, seems more crucial toa long-term visual representation if itis to be used to guide anticipated futureinteractions.

These speculations rest on the assump-tion that people encode drawings of scenesmuch as they do real scenes and that thetypes of information that are importantto understanding and memory are the samefor both. Although our knowledge of theinformation that people gather together toform a schema of a scene is still limited,the consistency of the data found in theseveral studies in this series providesconvergent evidence for the reality ofschema-controlled differences in encodingand retention of visual information. Theextent to which a visual schema can bedifferentiated from our general knowledgeabout scenes has yet to be determined.

Reference Note

1. Parker, R. E. Picture processing during recogni-tion. Manuscript submitted for publication, 1977.

396 JEAN M. MANDLER AND GARY H. RITCHEY

References

Bahrick, H. P., Bahrick, P. O., & Wittlinger, R. P.Fifty years of memory for names and faces: Across-sectional approach. Journal of ExperimentalPsychology: General, 1975, 104, 54-75.

Fajnstzejn-Pollack, G. A developmental study ofdecay rate in long-term memory. Journal ofExperimental Child Psychology, 1973, 16, 225-235.

Goldstein, A. G., & Chance, J. E. Visual recognitionmemory for complex configurations. Perception &Psychophysics, 1971, 9, 237-241.

Graesser, A. II, & Mandler, G. Recognition memoryfor the meaning and surface structure of sentences.Journal of Experimental Psychology: HumanLearning and Memory, 1975, 104, 238-248.

Kolers, P. A., & Ostry, D. J. Time course of loss ofinformation regarding pattern analyzing opera-tions. Journal of Verbal Learning and VerbalBehavior, 1974, 13, 599-612.

Loftus, G. R., & Bell, S. M. Two types of informa-tion in picture memory. Journal of ExperimentalPsychology: Human Learning and Memory, 1975,104, 103-113.

Mandler, J. M., & Johnson, N. S. Some of thethousand words a picture is worth. Journal ofExperimental Psychology: Human Learning andMemory, 1976, 2, 529-540.

Mandler, J. M., & Johnson, N. S. Remembrance ofthings parsed: Story structure and recall. Cogni-tive Psychology, 1977, 9, 111-151.

Mandler, J. M., & Parker, R. E. Memory fordescriptive and spatial information in complexpictures. Journal of Experimental Psychology:Human Learning and Memory, 1976, 2, 38-48.

Mandler, J. M., & Stein, N. L. Recall and recogni-tion of pictures by children as a function of organi-zation and distractor similarity. Journal of Experi-mental Psychology, 1974, 102, 657-669.

Nickerson, R. S. A note on long-term recognitionmemory for pictorial material. PsychonomicScience, 1968, 11, 58.

Sachs, J. S. Memory in reading and listening todiscourse. Memory & Cognition, 1974, 2, 95-100.

Shepard, R. N. Recognition memory for words,sentences, and pictures. Journal of Verbal Learningand Verbal Behavior, 1967, 6, 156-163.

Standing, L., Conezio, J., & Haber, R. N. Percep-tion and memory for pictures: Single-trial learningof 2500 visual stimuli. Psychonomic Science, 1970,IP, 73-74.

Received September 21, 1976Revision received December 29, 1976 •