Picture-Word Consistency and the Elaborative Processing of

MICHAEL J. HOUSTON, TERRY L. CHILDERS, and SUSAN E. HECKLER*

Many advertisers believe the pictorial and verbal components of an ad should

convey the same meaning. Based on theoretical and empirical evidence from a va-

riety of areas, three experiments were conducted that show superior recall for ads

in which the picture and copy convey discrepant information about product attri-

butes when the picture and brand name are linked interactively. An elaborative

processing explanation for the effect is supported by the finding that this superiority

diminishes if consumers have less opportunity to process the ad and form associative

linkages in memory.

Picture-Word Consistency and the ElaborativeProcessing of Advertisements

Virtually all forms of marketing communications relyon a combination of verbal and nonverbal elements toconvey the intended message. Within the store, manu-facturer point-of-purchase displays, packages, and re-tailer on-site merchandising materials often feature col-orful graphic and visual elements along with verballyoriented elements to attract consumer attention and con-vey a message about a product. Outside the store, news-paper and magazine advertisements as well as televisioncommercials often emphasize their message through thesame form of visually oriented communications, whereasradio uses music and sound effects. Despite this prom-inent use of nonverbal elements as message cues, sur-prisingly little research has examined the effects of non-verbal cues on consumer information processing. Evenless research has investigated the effective integration ofverbal and nonverbal elements in the design of market-ing communications. Recent studies on the effects ofpictorial material in print advertisements generally hasexamined the impact of pictures added to verbal materialin comparison with verbal material alone. Because of the

* Michael J. Houston is Research Professor and Chairman of Mar-keting and Terry L. Childers is Associate Professor of Marketing,University of Minnesota. Susan E. Heckler is Assistant Professor ofMarketing, University of Michigan.

The research was supported by funds provided to the first authorfrom the McManus-Bascom Professorship at the University of Wis-consin and by a BER grant from the School of Management, Uni-versity of Minnesota, to the second author.

prominence of marketing communications containing bothverbal and visual material, the determination of the mosteffective integration of visual and verbal cues warrantsmore attention.

We report research that examines altemative combi-nations of pictorial and verbal material in print adver-tisements. Specifically, our interest is in consumer mem-ory for messages in which the semantic content of pictorialmaterial is consistent with or discrepant from the verbalcopy. After summarizing previous research on the ef-fects of pictorial material in advertisements, we use the-ory and research from a variety of areas as a frameworkfor formulating hypotheses. We then test the hypothesesand discuss the implications of the findings.

PREVIOUS RESEARCH

Research on the inclusion of pictures in advertise-ments is of two broad categories. In one the effects ofpictorial messages on memory are tested and in the otherthe impact of pictures on consumer attitudinal responsesis studied. Each type of effect is worthy of study—at-titudes because of their anticipated link to purchase be-havior and memory because of the impact of internal in-formation throughout the consumer decision process.Memory effects also can be a basis for inferring the na-ture of consumer information-processing activities. (Lynchand Srull 1982 offer a useful perspective on memory asa basis for understanding consumer information pro-cessing.) In our study we use memory as a basis for un-derstanding consumer processing of pictures and wordsthat differ in meaning within an advertisement. There-

359

Journal of Marketing ResearchVol. XXIV (November 1987), 359-69

360 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1987

fore, we restrict our literature review to research on thememory effects of pictures.'

Research in cognitive psychology provides substantialevidence that in a wide variety of memory tasks picturesare remembered better than words (Alesandrini 1982;Paivio 1971). Though less plentiful, research in a con-sumer context confirms this finding. Using photographstaken from advertisements, Shepard (1967) found thatpreviously seen pictures were recognized accurately whenpresented alongside novel pictures to a greater extent thanverbal material tested in the same way.

More recent research has examined the conditionsleading to picture superiority in an attempt to gain anunderstanding of the picture superiority effect. Using ad-vertisements taken from the Yellow Pages of telephonebooks, Lutz and Lutz (1977) compared interactive pic-tures, noninteractive pictures, and their verbal counter-parts in terms of their impact on brand name recall. In-teractive pictures were those in which the brand nameand product class both were included in a pictorial for-mat (e.g., a picture for Rocket Messenger Service show-ing a courier atop a rocket in flight). Noninteractive pic-tures consisted of either the brand or the product class(but not both) in a visual format. The verbal-only con-dition consisted of the brand-product class verbal refer-ents without an accompanying picture. Ony in the caseof interactive pictures was the learning of pictures su-perior to the learning of words in a test of brand namerecall.

Childers and Houston (1984) examined immediate anddelayed brand name recall for interactive pictures versusbrand-product class verbal referents when processed atsensory versus semantic levels. They found picture su-periority in immediate recall when stimuli were pro-cessed at a sensory level but no superiority when stimuliwere processed at a semantic level. However, pictureselicited better recall in the delayed measure under bothlevels of processing. Though she did not use interactivepictures, Kisielius (1982) also found that verbal infor-mation accompanied by a pictorial representation of theverbal content was recalled more readily than the verbalinformation alone.

Edell and Staelin (1983) compared framed pictures,unframed pictures, and exclusively verbal content in termsof subjects' recall of brand-relevant items and aspects ofthe messages they saw (e.g., layout, tone). Framed pic-tures were messages that included verbal material equiv-alent in content to the picture. Unframed pictures weremessages that consisted only of the picture. Verbal mes-sages consisted of the verbal descriptions. Superior re-call occurred for the framed pictures over both unframedpictures and verbal-only messages.

Collectively the studies cited provide compelling evi-

' Research on attitudinal effects is reported by Mitchell and Olson(1981), Mitchell (1983), Percy and Rossiter (1983), Edell and Staelin(1983), and Kisielius and Stemthal (1984).

dence that consumers process pictorial material in ad-vertisements differently than verbal material, as revealedby the superior memory effects of pictures. However,the differential effects of picture-word similarity and dis-similarity warrant further attention. Verbal material se-mantically different from pictorial content affords an op-portunity to convey additional information about a product.For example, a picture can convey information about oneproduct attribute while the verbal copy discusses anotherattribute. The examination of potential positive and det-rimental consequences of such messages in comparisonwith those of messages with picture-word overlap is thepurpose of our study.

THEORETICAL FOUNDATIONS

The effects of presenting similar versus discrepant in-formation within a message, whether in a picture-wordformat or an entirely verbal format, have not receivedfocused attention in marketing research. Consequently,adherence to a single theory for purposes of hypothesisgeneration is not appropriate. Theoretical principles andempirical findings from several theoretical domains areavailable as a basis for anticipating the memory effectsof alternative combinations of picture-word content inadvertisements. These domains include the specific areaof person memory and judgment within social cognition,text comprehension, and schema theory. Collectively,they allow the development of hypotheses about overallmemory and the specific content of memory for ads withalternative picture-word combinations.

Perhaps the most extensive model with potential rel-evance to the issues of interest is the associative storageand retrieval model proposed by Hastie (1980) in the studyof person memory. He hypothesized that information in-congruent with a prior expectancy about a person is re-membered better than information congruent with thatexpectancy. The essence of this explanation is that in-formation incongruent with a prior expectation (i.e., in-formation in direct confiict with the expectation) is moredifficult to comprehend and, when received, will be heldin working memory longer than congruent information.In other words, incongruent information stimulates moreelaborate internal processing. The result is a greaternumber of associative pathways linking the incongruentinformation to existing knowledge. These additionalpathways make the incongruent information more re-trievable from memory, thereby enabling greater recall.Hastie and Kumar (1979), Hastie (1980), Srull (1981),and Lui and Brewer (1983) all report support for thismodel.

A considerable amount of evidence contradicts thepredictions of the Hastie model, asserting instead thatindividuals tend to remember more readily informationthat is congruent with a prior expectancy. Srull (1981)summarizes this research and resolves the apparent dis-crepancy by suggesting that the nature of the memorytask infiuences whether congruent or incongruent infor-mation is remembered best. In a recall task incongruent

PICTURE-WORD CONSISTENCY 361

infonnation is remembered best because the greaternumber of associative paths enhance retrieval. However,in a recognition task the retrieval process is bypassed.When congruent infonnation is presented, it is linked toexpectancies at the prototypical level, thus allowing eas-ier recognition. Srull (1981) further points out that vir-tually all of the studies finding greater memory for con-gruent information involved a recognition task. Hepresents a series of experiments that conform to the pre-dictions of the Hastie model and show the memory su-periority of incongruent information is eliminated in rec-ognition tasks (Srull 1981).

In the area of text comprehension, O'Brien and Myers(1985) developed and tested a proposition similar to thatof the Hastie model. Specifically, they examined the rel-ative memory effects of text difficult or easy to com-prehend. Ease of comprehension was manipulated byimbedding an expectation and subsequent predictable orunpredictable information within the same text material.Predictable information following the established con-text of the material represented relatively easier com-prehension. Unpredictable information represented a moredifficult level of comprehension. O'Brien and Myers hy-pothesized that under more difficult levels of compre-hension subjects would engage in more elaborative pro-cessing by attempting to integrate the unpredictablematerial into preceding portions of the text. The resultwould be improved memory and recall under conditionsof more difficult comprehension. O'Brien and Myersfound evidence supporting this contention.

Finally, research in the area of schema theory providesinsights into the issues we explore. Recent research in aconsumer behavior context has examined informationprocessing effects when consumers receive informationeither congruent or incongruent with an activated schema.Meyers-Levy and Tybout (1986) tested the effects ofgiving consumers product attribute infonnation that waseither entirely congruent or partially incongruent with anactivated schema. In the context of soft drinks, partialschema incongruity was achieved by including descrip-tive information on an attribute ("all natural") not nor-mally associated with soft drinks. Though the researchprimarily considered evaluative reactions to the product,a portion of it examined the hypothesis that schema in-consistency stimulated more effortful processing and im-proved memory over schema consistency. The resultssupported the hypothesis.

Research in the context of salesperson effects on con-sumer information processing (Sujan, Bettman, Sujan1986) has used schema theory as a basis for anticipatingconsumer responses to salespeople whose characteristicsare either consistent with or discrepant from the consum-er's salesperson schema. When consumers encounter asalesperson who matches the typical salesperson, theyresort to simple heuristics to form impressions and en-gage in information processing during the sales encoun-ter. An encounter with an atypical salesperson triggersgreater information processing because consumers de-

pend more on product-specific infonnation presented bythe salesperson to form impressions. A series of findingsgenerally confirmed these expectations.

The studies cited reflect a diversity of theoretical do-mains and research contexts. Common throughout them,though, is a fundamental result: when individuals haveexpectations, either created within the research proce-dures or schema-based, and they receive infonnation thatdeviates from the expectations, they engage in a moreelaborative form of information processing than wheninfonnation meeting their expectations is encountered.Furthermore, the studies by O'Brien and Myers (1985),Meyers-Levy and Tybout (1986), and Sujan, Bettman,and Sujan (1986) suggest the deviation from expecta-tions need not to be as strong as that implied by the Has-tie model to have this effect. Apparentiy the discrepantnature of the information is sufficient to prompt elabo-ration and it need not be in direct conflict with the ex-pectation.

The preceding discussion conveys the theoreticalframework used to anticipate differences in the basicprocess by which consumers encode ads containing al-temative combinations of picture-word content. Thisframework identifies two key conditions for predictingmemory effects of within-message variations in picture-word content.

1. An expectation must be established as a basis for deem-ing subsequent information consistent or discrepant.

2. Adequate opportunity for processing must be present toallow elaborative encoding and the formation of memorypaths when discrepant infonnation is encountered.

Given the presence of these two conditions, recall ofthe content of messages containing discrepant informa-tion should exceed that of messages containing consis-tent infonnation. An important point to emphasize hereis that these conditions suggest a format different fromthat of a message merely presenting two pieces of in-formation. It is not the quantity of infonnation that ac-counts for the anticipated superiority of messages withdiscrepant content. Rather, it is the creation of an ex-pectation followed by infonnation that deviates from theexpectation. If two pieces of information are presentedin a message but one does not create an expectation aboutthe other, such a message will not be processed as ex-tensively as a message meeting the two conditions. Theissue is how to integrate the elements of a message (i.e.,pictures and copy) to establish these conditions.

The nature of pictorial stimuli suggests that picturescan be used effectively to imbed expectations within amessage. For example, Holbrook and Moore (1981) ar-gue that, in general, pictures promote a more holistic andintegrative form of processing than do words. However,an interactive picture portraying the brand name andproduct class may be a specific pictorial device partic-ularly effective in establishing expectations within amessage. The greater memorability of interactive overnoninteractive pictures has been demonstrated by Lutz

362 JOURNAL OF AAARKETING RESEARCH, NOVEMBER 1987

and Lutz (1977). Alesandrini (1982) further suggests thatinteractive pictures can be effective strategies for com-municating a product name and a prime benefit (speedof delivery in the preceding example). More importantto our research is her suggestion that an interactive pic-ture can serve as an advance organizer. An advance or-ganizer cues the receiver to the content or importance ofthe subsequent messages, that is, it creates an expecta-tion of what is to follow. Implicit in this role of advanceorganizer is the assumption that recipients of picture-wordmessages attend to the pictorial component first. Indirectsupport for this premise comes from traditional adver-tising design principles, which advocate the use of pic-tures in ads for their "attention-getting" ability and thebelief that artwork is usually the first part of the ad tobe examined (Bolen 1984). Also, perception researchusing eye movements has found that fixations are influ-enced by pattern variability or irregularity in shapes orforms (Haber and Hershenson 1980; Mackworth andMorendi 1967). The disdnctiveness of the lines and curvesof the pictorial vocabulary in comparison with the verbalvocabulary is another factor underlying the picture su-periority effect (Nelson, Reed, and Walling 1976). Ad-ditionally, Mitchell (1983) reports that when subjectsevaluated a primarily visual advertisement, concurrentprotocols indicated the visual information was used toform a verbal understanding of the advertised brand. Thesefmdings support the premise that the pictorial componentof the ad is likely to be examined initially and has thecapacity to establish an expectation for the verbal partof the ad.

HYPOTHESESThe nature of interactive pictures serves as a basis for

integrating pictures and copy within a message, eitherconsistently or discrepantly, and hypothesizing the ef-fects. Interactive pictures can be used as an advance or-ganizer to imbed an expectation within the advertise-ment. Verbal copy that is either consistent with ordiscrepant from the picture can follow the picture. Com-parisons between interactive consistent and discrepant adsenable us to investigate the processing strategy invokedand its effects on memory. Elaborative processing of thediscrepant copy should establish a more extensive as-sociative network in memory, enhancing retrieval andsubsequent recall. Comparisons with correspondingmessages in which the same visual material is associatedwith different brand names (i.e., noninteractive) and ac-companied by the same copy can be examined to elim-inate a quantity-of-information explanation of the supe-rior memory for discrepant messages. Such an explanationwould not rely on the elaborative processing mechanismelicited by the expectations embedded in the interactiveinconsistent ads. In this context the following hy-potheses are offered.

H]! Total recall of the copy of ads containing interactivepictures is superior to that of ads containing nonin-teractive pictures.

H2: Total recall of the copy of ads containing interactivepictures accompanied by discrepant verbal copy issuperior to that of ads with interactive pictures ac-companied by consistent verbal copy. For ads con-taining noninteractive pictures, no difference in re-call occurs between consistent and discrepantmessages.

Support of H, would reaffirm the superiority of mes-sages with interactive pictures over those with noninter-active pictures. Findings consistent with H2 would bemuch more compelling for the purposes of our research.They would establish the effectiveness of picture-worddiscrepancies in an advertisement and the role of inter-active pictures in forming the expectation on which theconsistency or discrepancy of the copy would be based.

The content of advertising copy typically entails moreinformation than simply the consistent or discrepant in-formation. Advertising copy includes brand names,product classes, attribute information, and other infor-mation sucb as phone numbers and addresses. Conse-quently our key theoretical premise should be tested bymeans of a measure that captures the entire memory net-work resulting from an ad—not just memory for con-sistent or discrepant information. If discrepant infor-mation stimulates more elaborative processing of copycontent and copy content contains information beyondjust the discrepant information itself, the entire copy willbe processed more elaboratively. Thus, the entire mem-ory network for the ad copy should be different for mes-sages with discrepant verbal content and H2 should betested by using total recall of the content of ad copy.

In advertising research tbe specific content of memoryfor an ad is of interest in addition to overall memory.However, our hypothesized elaborative processingmechanism pertains to the latter and offers limited guid-ance for predicting the specific content of the memorynetwork. To arrive at hypotheses about the specific con-tent of memory, we return to O'Brien and Myers' (1985)research in a context that conforms more closely to thatin advertising research. They imbedded an expectationand subsequent expected or unexpected information witbinthe same text material and hypothesized an elaborateprocess for coping witb unexpected portions of the text.Of interest here is their prediction about tbe specific na-ture of tbe elaboration: wben unexpected material is con-fronted, subjects will attempt to integrate the unexpectedmaterial with preceding portions of the text. Wben ex-pected material is confronted, no such process occurs.Thus, text material containing unexpected informationwill yield improved memory for the unexpected infor-mation and for material preceding it in the text. O'Brienand Myers (1985) report fmdings tbat support their pre-dictions. We now have a basis for hypothesizing the spe-cific content of the memory network for consistent adsversus discrepant ads.

H3: A product attribute contained in discrepant verbal copyof interactive ads is recalled better than a product


attribute contained in consistent verbal copy of in-teractive ads.

This hypothesis reflects the fact that it is the product at-tribute that is manipulated as either consistent with ordiscrepant from the attribute conveyed by the picture.

Two additional hypotheses can be developed by usingthe O'Brien and Myers (1985) findings.

H4: Recall memory for verbal information preceding dis-crepant information in an interactive ad is superiorto recall memory for corresponding information inconsistent interactive ads.

H5: Recall memory for information following verballypresented attribute information does not differ be-tween consistent and discrepant interactive ads.

Noninteractive ads are not expected to produce the pat-tem of effects suggested in H3 through H5 for interactiveads because the former fail to establish expectations aboutthe content of verbal material.

The prediction in H3 is not expected to apply to bothrecall and recognition measures of memory. Srull (1981)indicates that the prototypical nature of consistent infor-mation leads to its easier recognition. This notion is con-sistent with a distinction commonly made between recalland recognition tasks, that is, recognition does not in-volve a retrieval process but requires only the discrim-ination of presented items from previous items (Lynchand Srull 1982). The notion that a recognition task wouldnot benefit from the increased associative linkages formedin the processing of discrepant infonnation, but wouldbenefit from the prototypical nature of processing con-sistent information, leads to the following hypothesis.

Hg: Recognition memory for consistent information ininteractive ads equates to that for discrepant infor-mation in interactive ads.

The key to the anticipated superior recall effects ofdiscrepant messages is that the discrepant verbal materialprompts elaborative processing, that is, copy material isheld in working memory for a longer time. If less elab-orative processing occurs, the superiority of discrepantmessage recall over consistent message recall should bereduced or eliminated. One approach for reducing theamount of processing is to limit the opportunity to en-gage in elaborative processing by reducing subjects' timeof exposure to a message. The implications for adver-tising are important because different exposure condi-tions (e.g., print vs. broadcast media) vary in availableexposure time and therefore in the extent to which theyallow elaborative processing of messages. Accordingly,the following general hypothesis (restricted to the use ofinteractive pictures) is suggested.

H7 As the time of exposure to messages is reduced, thesuperiority of the memory effects of discrepant mes-sages over those of consistent messages diminishes.

Three experiments were conducted to investigate the

hypotheses. Experiments 1 and 2 address H, through Hgand experiment 3 addresses H7.

EXPERIMENT 1

Design

Two independent variables were manipulated in a 2 x2 factorial design. The first factor represented the inter-active versus noninteractive nature of the picture part ofthe ads. Interactive pictures are those that pictorially rep-resent the brand name and product class, whereas non-interactive pictures bear no relationship to the brand nameand may or may not represent the product class. In ourstudy interactive pictures were designed also to conveya brand attribute (e.g., durability) to establish a basis forconsistent or discrepant verbal material to follow. Forthe noninteractive treatment condition the same pictureswere used but were associated with unrelated brand names.

The second independent variable represents the con-sistency factor. Pictures were combined with either con-sistent verbal content (i.e., copy that described the sameattribute portrayed in the picture) or discrepant verbalmaterial (i.e., copy that described an attribute differentfrom the one represented in the picture). Hence if a pic-ture conveyed durability, consistent copy discussed du-rability whereas discrepant copy discussed a different at-tribute (e.g., value).

Message Development

Ten professionally produced pictures (black-and-whiteline and shaded drawings) were used. These pictures andtheir interactive brand names were selected from a setdeveloped for use in a previous study, at which time pre-tests had established that each picture adequately con-veyed its intended attribute (Childers and Houston 1984).Thus, for example, BuUett Trucking Company was por-trayed by a picture of bullet-shaped trucks with a manriding a bullet on top of the trucks. Pretests had estab-lished that this picture conveyed the attribute of "on-timeservice." For the present research we developed a non-interactive brand name and consistent and discrepantverbal copy to accompany the picture.

In generating brand names for the noninteractive treat-ment, we avoided names high in mnemonic value (e.g.,bizarre names, rhymes, alliterations). In developing thecopy part of a message, we used the same number ofsentences and roughly equivalent number of words forcorresponding consistent and discrepant copy. Across the10 ads, the number of words ranged from 30 to 40. Theverbal part of the ad always appeared under the pictureand occupied 20 to 30% of the space. The brand nameand product class always were mentioned near the be-ginning of the copy. All copy material other than theattribute infonnation was identical across treatment con-ditions.

A pretest was conducted to ensure that consistency wasbeing manipulated effectively. Pretest subjects were as-signed randomly to either consistent or discrepant mes-

364 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1987

Product class

Auto tiresExpress mailPest controlFencesTruckingAuctioneerPizzaAuto partsPlumbingElectrician

Interactivebrand name

RhinoRocketArrowSupermanBulletOwlOn-the-ninDollarTool-readyAirborne

TableKEY COMPONENTS OF

Noninteractivebrand name

BellCapitolGlobalOakVikingRoseStarBallBearDolphin

1EXPERIMENTAL ADS

Attributeportrayed

in picture andconsistent copy

Durable/toughSpeedy deliveryEffectiveStrengthOn-time serviceKnowledgeable/wiseQuick deliveryInexpensiveAlways on callQuick service

Attributeportrayed in

discrepant copy

Good valuePride in serviceLong-time community presenceCustomized fencesSafe transitLow costDelicious tasteFriendlyReliable workReasonable cost

sages and asked to rate the extent to which the picturesand words for each message conveyed the same mean-ing. The group receiving the consistent messages ratedpicture-word overlap significantly higher (p < .01) thandid the group exposed to the discrepant messages. Also,each group rated its messages in the appropriate intendeddirection. Table 1 provides details of each of the 10 mes-sages and their treatment variations.^

Procedures

Subjects for experiment 1 were 112 undergraduate stu-dents enrolled in a basic marketing course at a majormidwestem university. They were assigned randomly toone of the four treatment conditions, resulting in cellsample sizes of 28 subjects. The experiment was con-ducted in a group setting in which subjects were toldthey would be evaluating a set of advertisements earlyin the development stage.

All test messages and measures were contained inbooklets randomly assigned to subjects. Ads were placedin random order in each booklet. However, the tasksperformed by subjects were always in the same order.Subjects first were given an example and practice ad notcontained in the set of 10 test messages.

Subjects were exposed to each ad for 15 seconds andthen told to tum the page. They were given 45 secondsto complete a series of semantic differential items aboutthe previously seen brand and ad. This exercise servedas a filler task between exposures to the series of ads.^The same sequence was repeated until subjects had viewedand evaluated the 10 messages assigned to them

Next subjects were given a 3-minute distractor task

^Ideally, a pretest establishing the equal memorability of verbal copyin the absence of pictures would have been appropriate. However, thefact that the words were identical (except for the attribute) suggestsit is not unreasonable to assume equivalence.

'These items also served as global measures of affective reactionsto the ad and brand. Analyses showed no significant effects on thesemeasures.

consisting of a social desirability measure and a set ofdemographic questions. Then they were allowed 10 min-utes to complete written protocols in which they weretold to write down everything they could remember aboutthe advertisements they had viewed. Separate boxes wereprovided for each ad. Subjects could recall ads in anyorder. After the written protocols, subjects were giventwo minutes to perform an aided recall task. Each of the10 product classes contained in the messages was listedwith a blank to the left. Subjects were instructed to fillin the brand names.

Measures

To obtain a total recall score and analyze effects onthe recall of various components of copy, we developedfour measures of memory for specific copy elements ofeach ad from the unaided written protocols.

1. Product class recall—correct mention of the product classrepresented in the ad.

2. Brand name recall—correct mention of the brand nameof the advertised product or service.

3. Copy attribute recall—correct reference to the productattribute conveyed in the copy.

4. Additional copy recall—correct mention of specific copyinformation beyond brand, product class, and copy at-tribute. This information included such items as the pres-ence of phone numbers, location, or how to contact theadvertiser. It always followed the mention of the copyattribute in each ad.

Two graduate students who did not know the purposesof the research served as coders for the measures. Theywere given coding format sheets outlining the key ele-ments of ad copy. For each memory protocol providedby a subject, the coders indicated whether each elementwas recalled correcdy. The two coders demonstrated anintercoder reliability of .91 across all coded pieces ofinformation. Any discrepancies were discussed and re-solved by the coders with the researchers before analysisof the fmdings.

The redundancy of product attribute information in thepicture and copy of consistent ads presents difficulties


in computing total copy recall scores and making com-parisons between treatment variations on copy attributerecall. The difficulty lies in ascertaining, for subjects ex-posed to consistent ads who correctly mentioned theproduct attribute, whether the source of the informationwas the picture or the copy. This situation precludes ex-clusive focus on memory for the verbal part of each ad.Consequently, certain modifications in the computationof total copy recall and the analyses of total copy recalland copy attribute recall were necessary in experiment1. Experiment 2 was conducted in part to eliminate thisdifficulty.

For the initial analysis, a total copy recall score (TCRl)that excluded copy attribute recall was computed. Com-ponent recall scores on product class, brand name, andadditional copy information were totaled for each subjectin each of the four cells across the 10 ads. Such a mea-sure captures any variations in overall memory networksresulting from the different treatments and controls forredundancies. Second, a total copy recall score that in-cluded copy attribute recall (TCR2) was computed sim-ilarly for subjects exposed to discrepant ads. For eachtotal copy recall measure, summed scores across the 10ads were used in the analyses.

Analyses also were performed to determine whethercomponent recall scores varied in the same ways. Again,for copy attribute recall, comparisons were restricted tothe two discrepant cells. In addition, a measure of brandname recall based on the aided recall measure was de-veloped. For purposes of analysis each of the componentrecall measures was represented for each subject by thesum of the correct responses across the 10 ads.

Results

Experiment 1 allows complete tests of Hi, H2, H,, andH5 and a partial test of H3. Treatment means for eachdependent variable are reported in Table 2. The follow-ing discussion summarizes the results of the plannedcontrasts relevant to each hypothesis.

Analysis of variance on TCRl shows strong supportfor Hi. Interactive ads elicit recall significantly superiorto that of noninteractive ads [F = 23.34, /> < .001).

Table 2EXPERIMENT 1 FINDINGS

Treatment means

Dependentvariable

Interactive Noninteractive

Consistent Discrepant Consistent Discrepant

TCRlUnaided product

classUnaided brandAdditional copyAided brandTCR2Copy attribute

11.66

6.633.441.593.59——

13.34

7.294.611.435.39

13.75.43

8.46

6.281.141.04

.86——

9.68

6.361.042.291.36

10.681.00

Furthermore, the comparison between the two discrepantcells on TCR2 reveals significant superiority for inter-active ads it = 2.70, p < .005). These findings replicateprevious empirical support for interactive ads but do notaddress the fundamental issue of the research—the ef-fects of discrepant pictures and words.

The appropriate series of one-way contrasts (with p =.05) provides strong support for H2. For TCRl a signif-icant difference (r = 1.66) in the hypothesized directionis found between the means for interactive discrepant andinteractive consistent ads but not between consistent anddiscrepant ads within the noninteractive cell (r = 1.22).Furthermore, the mean value in the interactive discrep-ant cell significantly exceeds it = 3.63) the mean in thenoninteractive discrepant cell. This pattern of resultssuggests it is the discrepancy from the expectation cre-ated by the interactive picture that accounts for the su-periority of such ads. The null effects in the noninter-active cell and the superiority of the interactive discrepantcell over the noninteractive discrepant cell eliminate thequantity-of-information issue as a competing explana-tion. A unique encoding process appears to occur forinteractive discrepant cells.

H3 pertains to recall of the copy attribute and predictsit will be superior when the copy is discrepant from thepicture. Because of the inability to ascertain the sourceof memory for the product attribute in consistent ads,treatment comparisons on copy attribute recall are re-stricted to the discrepant cells. Partial support for H3 wouldbe indicated by superior recall in the interactive version.However, copy attribute recall is found to be signifi-cantly greater in the noninteractive version (/ = 2.19, p< .05).

H4 is based on the expected specific nature of the pro-cess by which discrepant ads are encoded. Research byO'Brien and Myers (1985) suggests that when an un-predictable item is encountered the message recipient at-tempts to integrate it with previously processed infor-mation. In our research brand name and product classinformation immediately preceded unpredictable infor-mation. Recall data on brand name and product classwere examined to determine whether component recallscores conform to the H, prediction. The same series ofcontrasts used to test H2 was used to test H4.

Summing unaided product class and brand name recallforms a single measure of memory for information pre-ceding the discrepant copy that can be used to test H,.Performing the series of contrasts on this measure yieldssupport for H4. The two interactive cells differ signifi-cantly it = 3.05), the noninteractive cells do not (f =.03), and the discrepant cells do (t = 4.50). For productclass recall alone, the pattern of means conforms to pre-diction. Recall is greatest for interactive discrepant ads.However, it is not significantly greater than recall of in-teractive consistent ads (r = 1.47). The other two con-trasts fit expectations. Null effects are found within thenoninteractive cell (f = .18) and the interactive discrep-ant cell mean significandy exceeds its noninteractive


counterpart (t = 2.07). For both unaided and aided brandname recall, the pattern of results of the contrasts fitsexpectations. For unaided brand recall the interactive cellsdiffer significantly (t = 2.39), the noninteractive cellsdo not (t = .20), and the discrepant cells do (t = 1.29).The corresponding f-values for aided brand recall are 4.00,1.11, and 8.96, respectively.

All other information in the copy appeared after men-tion of the copy attribute. As predicted in H5, recall ofthis information is not significantly greater (t = .38) forinteractive discrepant ads than for interactive consistentads.

Discussion

The results of experiment 1 provide generally strongsupport for the superior memory effects of interactivediscrepant ads. The superior overall memory network as-sociated with such ads lends support to the key theoret-ical premise of the research. The interactive picture cre-ates an expectation about the content of the verbalmaterial. When the verbal material presents discrepantinformation, a more elaborative form of processing oc-curs, leading to enhanced memory.

Evidence supporting the hypothesized specific natureof the elaboration is not as strong. Though recall of in-formation preceding and following discrepant copy in-formation conforms to predictions, recall of the specificpiece of discrepant information does not. This result doesnot detract from our theoretical basis for anticipating im-proved memory for discrepant ads. It does, however, raisedoubts about whether the specific nature of the elabo-ration conforms to that suggested by O'Brien and Myers(1985). For example, the specific content of recall maybe explained more by retrieval processes than encodingprocesses. Because of the fictitious nature of the brandsused in the research and the more extensive memory net-work for interactive discrepant ads (as shown by totalrecall measures), retrieval efforts may have been di-rected more at accessing labeling information (brand andproduct class) for previously unknown brands. If re-trieval cues had been provided to help subjects structurethe retrieval effort, the specific content of memory couldbe attributed more directly to encoding processes. Theadditional cues provided by a structured task would dif-ferentiate between recall effects attributable to a failureto generate a cue category (a retrieval deficit), such as"the attribute(s) conveyed in the ad," and those due toa failure to encode the information in memory at initialexposure (a storage deficit). Furthermore, a structuredrecall task would enable us to ascertain more readilywhether the memory source for consistent product attri-bute infonnation was the picture or the verbal material.Experiment 2 was designed to examine these issues.

EXPERIMENT 2

A second experiment was conducted to assess moredirectly the specific nature of the elaborative encodingprocess used when discrepant material is presented. Data

collected through this experiment enabled us to replicatetests of H| and H2, examine further the processes rep-resented by H3 through H5, and test Hg, which states thatconsistent product attribute information as well as dis-crepant information will be recognized.

Design and Procedures

The design of experiment 2 was identical to the 2 x2 factorial design of experiment 1. The same sets of adswere used for the treatments. A total of 116 student sub-jects were assigned randomly to the treatments, resultingin equal cell sizes of 29 subjects.

All procedures were identical to those in experiment1 with the exception of the method used to collect recalldata and the addition of a recognition task. In place ofa completely unaided recall task, a structured recall taskwas administered. This task differed from the one in thefirst experiment in that category cues were provided forthe major elements of the ad (brand name, product class,copy attribute, and picture attribute), as well as for acategory associated with other less pertinent informationsuch as phone numbers. Subjects were given 10 pageson which to indicate separately for each ad all the in-formation they could recall within each of these cate-gories. The aided brand name task followed this task, asin experiment 1. The final task was a recognition test forthe product attribute. Each product class was specified.A set of five product attributes was given for each prod-uct class, one of which was the product attribute actuallydiscussed in the copy. Subjects indicated which of theset of attributes was presented in the ad they had seen.

Results

The results of experiment 2 are summarized in Table3, which gives treatment means. The effects on totalmemory correspond closely to those observed in exper-iment 1. TCRl and TCR2 in the interactive discrepantcell significantly exceed those in both the interactiveconsistent (t = 4.02, 1.66, respectively) and noninter-active discrepant (t = 5.55, 5.80, respectively)cells. Nulleffects on TCRl (t = .25) are found in the noninteractive


Treatment means

Dependentvariable

Interactive Noninteractive

Consistent Discrepant Consistent Discrepant

TCR2TCRlProduct classBrand nameCopy attributeAdditional copyAided brand nameCopy attribute

recogniton

13.319.555.762.963.76

.833.83

8.19

14.9712.937.074.142.031.725.17

2.54

12.728.486.62

.864.241.001.03

7.80

9.178.276.34

.72

.901.21.83

2.72


cell. Though a significant effect is seen on this compar-ison for TCR2, the direction of the difference is oppositewhat would be damaging to the theoretical premise ofthe research.

The overall pattern of effects on component recall scoresdoes not support the hypothesized specific nature of theencoding process. Though memory for product class andbrand name conforms to expectations, memory for othercomponents of the copy does not. Contrary to H3, recallof the copy attribute is superior in consistent ads (t =4.81). Recall of copy content appearing after the dis-crepant copy attribute is superior under interactive con-ditions (t = 4.81). Thus, the specific pattern of resultssuggested by H3 through H5 is not confirmed. The en-coding process suggested by O'Brien and Myers (1985)is not evident.

The second experiment included a copy attribute rec-ognition test. Recognition of the consistent copy attri-bute is superior to 5iat of the discrepant copy attribute(/ = 10.66) under interactive conditions. Thus, the ex-pected pattern of difference in recall and recognition forspecific consistent and discrepant information is not ob-served. Superior recognition and recall of consistentproduct attributes are found.

Though experiment 2 does not confirm the specificencoding process hypothesized, it does support the elab-orative processing explanation of the general encodingprocess for ads that present discrepant verbal informa-tion. The third experiment was designed to test a specificproposition underlying the model.

EXPERIMENT 3

The results of experiments 1 and 2 support elaborationas the explanation of the superior effects of combininginteractive pictures with discrepant verbal copy. With thistheoretical framework as a guide to understanding howconsumers process these types of messages, we can be-gin to identify and test the conditions under which theresults of experiments 1 and 2 can be expected to holdor not hold. H7 specifies one such condition: when theopportunity to engage in elaborative processing is re-stricted, the superior memory effects of interactive dis-crepant ads should diminish. Experiment 3 tested thishypothesis.

Design and Procedures

Experiments 1 and 2 established the general superi-ority of messages with interactive pictures over those withnoninteractive pictures. They also provided evidence thatinteractive pictures can be used to embed expectationswithin a message. Therefore, experiment 3 included onlymessages with interactive pictures in a 2 x 2 factorialdesign. The consistency factor was manipulated in thesame way as in experiments 1 and 2 with the same mes-sages. The second factor was exposure time. One levelof exposure time was 15 seconds, the length of exposuretime used in experiments 1 and 2. The reduced level ofexposure time was 10 seconds. A pretest revealed that

10 seconds was the minimum time necessary to processthe picture and orally read through the copy once.

A new pool of 80 student subjects was assigned ran-domly to the four treatments. Except for the 10-secondexposure time as a manipulation, the procedures of ex-periment 3 were identical to those of experiment 1 (i.e.,the key recall task was unstructured). The same mea-sures were used as dependent variables.

Results

Data analysis procedures again consisted of plannedcontrasts to test the general hypothesis. As before, thekey variable for hypothesis testing was TCRl. Compo-nent recall scores also were examined to determine theextent to which they conformed to previously observedpatterns. Furthermore, two-group comparisons involvingthe discrepant cells were performed on TCR2 and on copyattribute recall. Table 4 summarizes the mean values foreach dependent variable. The specific contrasts neces-sary to support H7 should show memory for discrepantads significantly in excess of memory for consistent adsunder 15-second exposures, but not under 10-second ex-posures.

The key finding related to the theoretical purposes ofexperiment 3 is for TCRl. The planned contrasts showsuperiority for discrepant ads under 15-second exposuretimes (f = 2.40), but not under 10-second exposure times(t = 1.31). Thus, the results on the central variable sup-port H7. This pattern carries over to several componentrecall measures. Discrepant ads are superior under 15-second exposures for product class recall (t = 1.71), un-aided brand recall {t = 2.98), and aided brand recall (r= 2.00), but not under 10-second exposures (t = 1.10,1.36, and .96, respectively).

Because only discrepant cells are examined in the two-group comparison of TCR2, the results of this test areless relevant. The comparison does show a significantlyhigher level of recall for messages in the 15- than in the10-second exposure time, and hence may confirm thatreducing exposure time did restrict elaborative process-ing. A similar result is found for the specific recall ofthe copy attribute.

Two important points emerge from experiment 3. First,


Dependentvariable

TCRlUnaided product

classUnaided brandAdditional copyAided brandTCR2Copy attribute

15-second

Consistent

9.25

5.702.501.054.20——

exposure

Discrepant

12.00

6.554.151.405.35

13.501.50

10-second exposure

Consistent

8.50

5.451.701.353.85——

Discrepant

10.00

6.002.451.054.40

10.70.70


we find additional evidence supporting the superiormemory effects of discrepant messages of the type in-vestigated. Second, based on the underlying theory guid-ing our research, a condition in which this result weak-ens is identified. When the opportunity for elaborativeprocessing of verbal material is limited by a reduced ex-posure time, the superiority of messages with discrepantverbal content diminishes.

CONCLUSIONS AND IMPUCATIONS

Our findings provide evidence contradictory to a com-mon belief that the pictorial and verbal.components ofan advertisement should convey the same message (cf.Schultz 1981). Semantically discrepant pictures and wordswere combined in a way that yielded recall superior tothat observed for semantically consistent pictures andwords. Specifically, by using an interactive picture toconvey one attribute and establish an expectation incombination with copy discussing a different attribute,we elicited better recall of copy material.

Three separate experiments in which this finding wasobtained provide support for the theoretical basis of thehypothesized explanation. The essence of the explana-tion is that an interactive picture establishes an expec-tation about the content of verbal material. When dis-crepant verbal information is encountered, moreelaborative processing occurs and results in a more ex-tensive memory network for the ad. Experiments 1 and2 show that the mere quantity of information containedin such ads is not a plausible alternative explanation. Ex-periment 3 shows that elaborative processing at encodingis central to understanding the observed effects. Whenthe opportunity for elaborative processing is restricted byreduced exposure times, the superiority of discrepantmessages diminishes.

Experiment 2 was conducted to examine further thespecific nature of the elaborative processing effort. Wehypothesized that subjects' attempts to integrate the dis-crepant information into previously seen material wouldlead to superior memory for information preceding thediscrepant copy but equivalent memory for informationfollowing it. The findings do not support this hypothesis.Such ads yielded superior memory for information bothpreceding and following the discrepant content. Thus,these ads appear to result in an elaboration that spreadsthroughout the ad. For advertising purposes, this effectis more desirable than the one hypothesized. Further-more, what appears sufficient to prompt this elaborationis the establishment of an expectation followed by dis-crepant information that need not be in sharp contrast tothe expectation. We used product attributes different from,but not in conflict with, the expectation. This findingsuggests advertisers have more latitude in creating mes-sages to achieve the observed effects than is implied bysocial cognition research.

The final major finding pertains to memory for thespecific verbal element (the product attribute) that makes

the ad consistent or discrepant. We hypothesized that re-call of the discrepant attribute would exceed recall of thepredictable attribute, but that recognition of the consis-tent attribute would be equivalent. Experiment 2 showsboth recall and recognition to be greater for consistentthan for discrepant product attributes. The recall findingsare counter to O'Brien and Myers' (1985) observationsand may be due to the redundancy of the information inconsistent ads. Another possible explanation is that thedual-code nature of the consistent information makes itmore readily available in memory (Paivio 1971). Fi-nally, Mandler (1982) offers insights into how slight in-congruities may depress memory for the incongruity.

Poorer recall of the discrepant copy attribute in inter-active ads does not detract from their tactical superiority.Such ads promote superior overall learning of messagecontent. Remembering attribute information withoutlinking it to a specific brand would be of little benefit.Brand name recall for other types of messages includedin the study is substantially lower. Hence, these ads maybe particularly useful during the introductory phase fora brand, when creating consumer awareness is a primarymarketing objective. An important consideration in theuse of discrepant content is identified in the final ex-periment—advertisers should ascertain whether the ex-posure conditions that allow elaborative processing and/or the individual factors leading to it (e.g., high involve-ment or individual differences in motivation) are pres-ent.

The practical limitations of our study suggest possibledirections for future research. For example, we inves-tigated alternative combinations of pictures and wordsand used pictures of a specific type—interactive pic-tures—to establish expectations by conveying brand at-tribute information. However, many companies cannotuse interactive pictures because their brand names arenot suited to such a portrayal. Future research shouldexamine other potential means for establishing expec-tations from which discrepant information can follow.The possibilities include other types of pictures, othernonverbal stimuli (e.g., music), and even verbal mate-rial.

REFERENCES

Alesandrini, Kathryn Lutz (1982), "Strategies that InfluenceMemory for Advertising Communications," in InformationProcessing Research in Advertising, R. J. Harris, ed. Hills-dale, NJ: Lawrence Erlbaum Associates, 65-81.

Bolen, William H. (1984), Advertising, 2nd ed. New York:John Wiley & Sons, Inc.

Childers, Terry L. and Michael J. Houston (1984), "Condi-tions for a Picture-Superiority Effect on Consumer tA&m-ory," Journal of Consumer Research, 11 (September), 643—54.

Edell, Julie A. and Richard Staelin (1983), "The InformationProcessing of Pictures in Print Advertisements," Journal ofConsumer Research, 10 (June), 145-61.


Haber, Ralph Norman and Maurice Hershenson (1980), ThePsychology of Visual Perception. New York: Holt, Rinehartand Winston.

Hastie, Reid (1980), "Memory for Behavioral Information thatConfirms or Contradicts a Personality Impression," in Per-son Memory: The Cognitive Basis of Social Perception, R.Hastie et al., eds. Hillsdale, NJ: Lawrence Erlbaum Asso-ciates, 155-77.

and P. A. Kumar (1979), "Person Memory: Per-sonality Traits as Organizing Principles in Memory for Be-haviors," Journal of Personality arid Social Psychology, 37(January), 25-38.

Holbrook, Morris B. and William L. Moore (1981), "FeatureInteractions in Consumer Judgments of Verbal Versus Pic-torial Presentations," Journal of Consumer Research, 8(June), 103-13.

Kisielius, Jolita (1982), "Detecting and Explaining VividnessEffects in Attitudinal Judgements," unpublished doctoraldissertation. Northwestern University.

and Brian Stemthal (1984), "Detecting and ExplainingVividness Effects in Attitudinal Judgments," Journal ofMarketing Research, 21 (February), 54-64.

Lui, Layton and M. B. Brewer (1983), "Recognition Accu-racy as Evidence of Category-Consistency Effects in PersonMemory," Social Cognition (Summer), 89-107.

Lutz, Kathy A. and Richard J. Lutz 0977), "Effects of In-teractive Imagery on Learning: Application to Advertising,"Journal of Applied Psychology, '62 (November), 493-8.

Lynch, J. G. and T. K. Srull (1982), "Memory and Atten-tional Factors in Consumer Choice: Concepts and ResearchMethods," Journal of Consumer Research, 9 (March), 18-37.

Mackworth, N. H. and A. J. Morendi (1967), "The Gaze Se-lects Informative Details Within Pictures," Perception andPsychophysics, 2, 547-52.

Mandler, George (1982), "The Structure of Value: Accountingfor Taste," in Affect and Cognition: The 17th Annual Car-negie Symposium, M. S. Clark and S. T. Fiske, eds. Hills-dale, NJ: Lawrence Erlbaum Associates.

Meyers-Levy, Joan and Alice M. Tybout (1986), "SchemaCongruity as a Basis for Natural Object Evaluation," work-ing paper. Department of Marketing, Northwestern Univer-sity (September).

Mitchell, Andrew A. (1983), "The Effects of Visual and Emo-tional Advertising: An Information-Processing Approach,"in Advertising and Consumer Psychology, L. Percy and A.Woodside, eds. Lexington, MA: Lexington Books, 197-217.

and Jerry C. Olson (1981), "Are Product Attribute Be-liefs the Only Mediator of Advertising Effects on Brand At-titudes?" Journal of Marketing Research, 18 (August) , 3 1 8 -32.

Nelson, Douglas L., Valerie S. Reed, and John R. Walling(1976), "The Pictorial Superiority Effect," Journal of Ex-perimental Psychology: Human Learning and Memory, 2 (5),523-8.

O'Brien, Edward J. and Jerome L. Myers (1985), "WhenComprehension Difficulty Improves Memory for Text,"Journal of Experimental Psychology: Learning, Memory, andCognition, 11 (February), 12-21.

Paivio, Allan (1971), Imagery and Verbal Processes. New York:Holt, Rinehart and Winston.

Percy, Larry and John R. Rossiter (1983), "Mediating Effectsof Visual and Verbal Elements in Print Advertising uponBelief, Attitude, and Intention Responses, in Advertising andConsumer Psychology, L. Percy and A. Woodside, eds.Lexington, MA: Lexington Books, 171-96.

Schultz, Don E. (1981), Essentials of Advertising Strategy.Chicago: Crain Books. "•

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

Srull, Thomas K. (1981), "Person Memory: Some Tests ofAssociative Storage and Retrieval Models," Journal of Ex-perimental Psychology: Human Learning and Memory, 1(November), 440-63.

Sujan, Mita, James R. Bettman, and Harish Sujan (1986),"Effects of Consumer Expectations on Information Process-ing in Selling Encounters," 23 (November), 346-52.

Documents

Picture-Word Consistency and the Elaborative Processing of