The Representation and Recall of Message Arguments in Advertising: Test of a Schema-Based Model

James M. Hunt, Temple University
E. H. Bonfield, Rider College
Jerome B. Kernan, University of Cincinnati
ABSTRACT - The schema-copy-plus-tag (SCT) memory paradigm is tested in 8 2x2 (typical vs. atypical arguments: immediate vs. two-day recall) design in the context of print advertising. Since mixed results do not fully support the model, suggestions for its more effective testing are offered.
[ to cite ]:
James M. Hunt, E. H. Bonfield, and Jerome B. Kernan (1986) ,"The Representation and Recall of Message Arguments in Advertising: Test of a Schema-Based Model", in NA - Advances in Consumer Research Volume 13, eds. Richard J. Lutz, Provo, UT : Association for Consumer Research, Pages: 562-565.

Advances in Consumer Research Volume 13, 1986      Pages 562-565


James M. Hunt, Temple University

E. H. Bonfield, Rider College

Jerome B. Kernan, University of Cincinnati

[Completion of this research was made possible by a grant from the George A. Ramlose Foundation.]


The schema-copy-plus-tag (SCT) memory paradigm is tested in 8 2x2 (typical vs. atypical arguments: immediate vs. two-day recall) design in the context of print advertising. Since mixed results do not fully support the model, suggestions for its more effective testing are offered.


How advertising is represented in memory is of considerable interest to those in the field of marketing communication. Not only is this representation thought to have an immediate influence on advertising response, but it is also thought to have a longer-term influence on memory. The study undertaken here attempts to address this issue by testing a schema-based model of memory, in which message comprehension and representation are viewed as being largely under the control of pre-existing knowledge structure.

The model, based on Graesser and Nakamura's (1982) schema-copy-plus-tag (SCT) hypothesis, begins with the notion that message recipients engage in certain cognitive processes as a result of anticipating the receipt of a persuasive message (cf. Cialdini, Petty and Cacioppo 1981). Based on pre-message source, contextual, and thematic cues, these processes are thought to involve the retrieval of a generic schema that best relates to the communication issue. In that the schema is supposedly a stored representation of information typical of similar persuasive messages gained through experience, its presumed role is one of guiding message comprehension. In addition, it is thought to serve as an organized structure that facilitates message representation and storage. According to this line of reasoning, memory for ads is constructive and, thus, frequently incomplete and occasionally even highly distorted. In the encoding process, message arguments typical of the schema are thought to merge with similar arguments contained in the ad so as to produce a coherent, unified, expectation-confirming and knowledge-consistent representation of the message (see Alba and Hasher 1983). As a consequence, the memory trace makes no distinction between "typical arguments" that actually appeared in the message and those that did not (Schmidt and Sherman 1984). Arguments that are not typical of the schema are said to be represented separately from the schematic portion of the trace by a "unique tag." In essence, this tag distinguishes the item from schematically typical items within the trace.

Whether this model adequately depicts memory for ads is the focus of the study reported here. An experiment was conducted in which both immediate and delayed effects of advertising were assessed in the context of print media. Initial findings are reported in terms of both recall and recognition as well as related measures of attitude.


Although memory for complex events can be quite accurate, it is frequently found to be subject to distortions and omissions. these errors, of course, pose a problem for any theory of memory. One theoretical perspective that appears capable of accounting for such errors, however, is that of schema theory (see Alba and Hasher 1983). This perspective views memory of complex events as reconstructive, consisting of a generic schema together with certain abstracted event-related details thought to guide reconstruction. The term schema refers to the general knowledge structure a person possesses about a particular domain (Alba and Hasher, p. 203). According to this formulation, recall of a complex event is likely to contain little in the way of actual detail. Instead, it frequently produces such things as paraphrases (of prose passages), thematic intrusions, and interpretations and elaborations that are more in line with the generic schema than with the target event. Alba and Hasher (1983) give a more thorough account of this perspective in terms of four central encoding processes: selection, abstraction, interpretation, and integration.

A schema theory which asserts that all four processes occur would state that from any environmental event, only the information that is relevant and important to the currently activated schema will be encoded. Of the information selected, the semantic content of the message will be abstracted and the surface form will be lost. Further, the semantic content will then be interpreted in such a way as to be consistent with the schema. The information that remains will then be integrated with previously acquired, related information that was activated during the current encoding episode. The operation of one or more of these processes is likely to result in a representation that is less than totally accurate (p. 204).

As already pointed out, another hypothesized schema-related process is that of reconstruction. Occurring at retrieval, reconstruction is thought to combine selected event-related abstracted details with an activated schema to arrive at a probable recounting of the target event. One current and well articulated representation of schema theory is the schema-copy-plus-tag (SCT) model proposed by Graesser and Nakamura (1982), which attempts to explain memory for narrative descriptions of action sequences for which people possess prototypic schemata (essentially scripts). According to the SCT model, comprehension and storage of prose passages are carried out in terms of a generic schema that best fits the passage's main theme. Items that are most typical of the schema are "copied" directly from the schema into the memory representation. Thus, the memory trace may include some items that were not explicitly stated in the passage; there is no distinction between very typical statements that were actually in the passage and those that were not. In essence, typical message statements may actually be forgotten, i.e., typical information need not be stored, since it can always be derived from the prototypic script, or schema. On the other hand, statements that are atypical of the schema are thought to receive special attention in memory. Specifically, it is thought that such statements cannot be represented in the schematic portion of the trace. Instead, they are said to be linked to the representation by a "unique tag" by which they are distinguished from typical and other atypical items.

Schmidt and Sherman (1984) have extended this model to persuasive communication by opining a generic schema position - i.e., a generic schema containing typical arguments (as opposed to action sequences) related to a familiar communication issue. Thus, representation of a persuasive message is thought to entail the copying of typical schema-related arguments directly into the schematic portion of the trace and tagging atypical arguments as deviations from the schema.

If this formulation is correct, it has several implications for persuasive communication in general and advertising in particular. First, typical arguments should hold an advantage in retrievability because of their placement in the generic schema. Atypical items should suffer in that they are not strongly associated with the schema. In a general recall task, individuals are thought to recall message content by accessing the generic schema. This being so, typical items are more likely to be recalled than atypical items. Further, this difference should increase over time because of 8 faster decline in the retrievability of atypical, or tagged, information. A final effect has to do with intrusion errors. Since the memory trace for persuasive messages is thought to be undifferentiated in terms of typical arguments- i.e., whether they were actually contained in the message or not- there should be a greater likelihood of schema-based intrusions, and this should increase over time.

Although the constructivist position advocated by Schmidt and Sherman is based on a popular perspective of memory, it does not form an easy amalgam with other literature which seems to demonstrate that recall can be surprisingly accurate. It has been shown, for instance, that under certain recall conditions- instructions that stress accuracy or change subjects' retrieval strategies and situations that permit successive attempts st recalling the same material - schema-based recall errors can be eliminated (See Alba and Hasher; Hasher and Griffin 1978). Further, the Schmidt and Sherman formulation does not seem to mesh well with research showing that depth of message processing is enhanced by the use of unexpected atypical message arguments (Bunt and Kernan 1984). If this latter finding is correct, atypical arguments may become a more (not less) memorable part of the message representation. Because of these apparent -discrepancies, an experiment was conducted to test the schema-copy-plus-tag formulation in the context of print advertising. This involved the manipulation of argument typicality and recall interval over two levels (typical vs. atypical; immediate recall vs. two-day delayed recall). In accordance with schema theory, it was anticipated that subjects exposed to ads containing only typical arguments would exhibit better recall and more schema-based intrusions than subjects exposed to ads containing atypical information. Further, atypical items were expected to suffer more in terms of being recalled over time than typical items. Finally. information-based attitudinal data were expected to exhibit patterns that parallelled the hypothesized information decay.



Female and male undergraduate students (N=41) from two eastern universities participated in a study to evaluate a new student magazine. Ages ranged from 20 to 47; average age was 22. host of the subjects were either business majors (78%) or communication majors (15%). In addition, thirty-one subjects served in an argument-generation phase designed to ascertain typical and atypical arguments for the test advertisement.

Stimulus Material

For purposes of external validity, the experimental ad was embedded in a bogus test magazine supposedly designed for area college students. The ad promoted a new brand of ball-point pen, called the "Accupoint Pen," and consisted of picture and text. The picture showed a group of three business-like executives (two males and a female) examining 8 computer print-out. One person, who appeared to be the supervisor, had pen in hand. Next to this picture was a close-up of the pen. The text described these people as "brand managers."

Arguments used to construct the text of the ad were created by asking subjects in the generation phase to list typical and atypical arguments regarding the promotion of ball-point pens. Half the subjects in this phase were asked to list the arguments they thought would be typical of a normal pen sd. The other half was asked to list atypical arguments. This produced a set of fifteen general statements, eleven of which were designated typical, and four of which were designated atypical.

Two versions of the pen ad were constructed, a typical version and an atypical verSion. Each of these contained five of the eleven typical arguments. In addition, they both contained six filler statements. One argument was reserved for the typicality manipulation; this had to do with the pen's effect on the writer's handwriting. In the typical sd, the pen was represented as one that would "fit your handwriting," while in the atypical version it was described as "actually improving your handwriting." The textual location of this argument was identical in both versions--midway through the text.

Dependent Measures

Recall of the ad was obtained by asking subjects to write out as many of the statements from the Accupoint pen ad as they could remember. This task followed a short questionnaire dealing with subjects' overall impressions of the test magazine. It was felt that this questionnaire would serve as a distraction task for the immediate recall group and provide a realistic format for subjects whose recall was delayed.

Recognition was tested vis seven items (5 typical and 2 atypical). Two of the typical items involved arguments that actually had been presented; the remaining three pertained to unpresented arguments. Of the two atypical items, one related to an unpresented argument, while the other involved the manipulated argument. Thus, this second item represented material that was unpresented to typical subjects and presented to those in the atypical group. Subject rating of each statement was performed on the following six-point scale 1 = definitely not in the sd; 2 = fairly sure it was not in the ad, 3 = uncertain, but think it was not in the ad 4 = uncertain, but think it was in the sd; 5 = fairly sure it was in the ad; 6 = definitely was in the ad. The instructions indicated to subjects that some of the statements had not actually appeared in the Accupoint ad .

The final set of dependent measures consisted of two attitudinal statements. The first invoLved subjects' "attitude toward the Accupoint Pen ad;" the second involved "attitude toward the Accupoint Pen itself.^ Both items were rated on a 7-point bipolar scale, ranging from "very favorable" to "very unfavorable."


Because of certain physical constraints, subjects were selected and assigned to treatment cells on a class, rather than individual, basis--i.e., each of the four classes constituted one of the four treatment cells. Subjects were informed that the purpose of the experiment was to assess the concept of a new student magazine called Campus Week. This test vehicle consisted of a cover, calendar-of-events at various area campuses, a full-length feature article, and several ass dispersed throughout. General instructions requested subjects to read the entire magazine as though they had an interest in all of its content. No mention was made of a subsequent memory test. Subjects then received a magazine containing one of the two Accupoint ad versions. Eight minutes was the allotted reading time, after which subjects completed the questionnaire dealing with their general impressions. Following this, subjects in the immediate treatment group received a questionnaire containing the dependent measures. Subjects in the delayed group completed this latter questionnaire two days after their exposure to the stimulus material. At the outset of the questionnaire, both groups were told that because of the length of the magazine. they would be required to evaluate only a portion of it. No mention was made of which portion that would be until subjects opened the questionnaire. It was then that they learned they would be evaluating the Accupoint ad. It was felt that this procedure eliminated any presensitization error. To reduce possible communication between the immediate and delayed groups, subjects from different schools were used for these groups.



Recall data consisted of correctly-recalled items, as well as errors and additions. Errors were placed into one of three categories: (1) thematic intrusions-statements that were consistent with the ad's overall theme, but not actually presented; (2) schematic intrusions- statements that corresponded to typical statements from the generation. phase, but not actually presented; and (3) generalizations- statements that combined actual arguments into a more general assertion about the text. Additions included statements about the pen that were unrelated to the text. Scoring of protocols was conducted by having two judges independently classify the recalled items. The judges agreed on 962 of these classifications. Disagreements were resolved through discussion between judges.

Four separate ANOVAs were carred out on correct recall and error scores. As expected, a significant main effect of recall interval was found for correct recall (H = 2.00 and 0.72 for immediate and delayed intervals respectively), F(1,38) = 11.23, p < .002. However, no significant effect of typicality was found; nor was there a significant interaction effect. Of the three analyses involving error scores, only one was found to contain results that approached significance. That pertained to schematic intrusion scores. As might be expected, recall interval had a main effect (marginally significant), with means declining over the interval (M = .46 and M = .17), F(1,38) = 3.15, p < .08. Somewhat surprisingly, however, the typicality factor was found to have a main effect (again marginal) on the schematic intrusion scores in which the mean for typical subjects was .16 as compared with .48 for atypical subjects; F(1,38) = 3.83, p < .06. No significant interaction effect was found. Further analysis of thematic intrusion and generalization scores revealed no significant results.


Mean scores for both recall and recognition items are reported in Table 1. Items pertaining to recognition are reported in terms of their typicality and whether they were actually presented in the test ad. Results related to the manipulated typicality argument appear in item 5. Consistent with the manipulation, an ANOVA revealed a significant main effect of typicality for this item, F(1,38) = 4.91, p < .03. As expected, subjects who received an ad claiming the pen would "actually improve their handwriting" scored significantly higher (M = 4.00) than those who received the "fit your handwriting" version (M = 3.00). In addition, a marginal main effect of retention interval was found for this item (M = 3.92 and M = 3.06 for immediate and delayed intervals, respectively), F = 3.63, p < .06. No significant interaction effect was found.



Items 2 and 4 pertain to schema-typical items that actually appeared in the test ad. An ANOVA on item 2 revealed a significant main effect of typicality, F = 8.07, p < .007. Subjects exposed to the atypical ad version exhibited better recognition scores (M = 5.09) than those who received the typical version (M = 4.00). Analysis of the fourth item produced a marginally significant two-way interaction effect (F = 3.47, p < .07), with atypical/immediate recall subjects exhibiting better recognition than their typical counterparts. This ordering was reversed for delayed recall subjects. No other significant effects were found with regard to typical presented arguments.

Items 1, 3, and 6 involved schema-typical arguments that were unpresented. Of the three related ANOVAs, only one produced a main effect approaching significance. In responding to item 6, atypical subjects surprisingly scored higher (M = 3.17) than did typical subjects (M = 2.47), thus indicating a greater intrusion likelihood, or false recognition rate, on the part of the former group, F = 3.00, p < .09. An ANOVA pertaining to the first recognition item produced somewhat similar results. A significant two-way interaction effect (F = 4.57, p < .04) was found in which immediate recall subjects exposed to the atypical ad exhibited higher false recognition scores than typical ad subjects. This ordering was slightly reversed for delayed recall subjects. No significant results were found in conjunction with item 3.

Only one item, the seventh, was used to represent false atypical arguments. Analysis of the data pertaining to this item failed to produce any significant results.

Analysis of subjects' ability to discriminate between presented and unpresented information was performed by calculating the proportion of correct "yes" (ratings of 4, 5, and 6) and "no" (ratings of 1, 2, and 3) responses on the recognition test for atypical subjects. The mean percentage score for typical arguments was 0.62 for both immediate and delayed retention subjects. Respective meanS for atypical arguments were 0.77 and 0.75. Contrary to expectations, a two-way ANOVA on these scores produced no significant main or interaction effects.

Related attitudinal data also failed to exhibit significant differences between typicality groups or over the retention interval. These data appear in Table 1.


According to the results of this study, there is only modest evidence to support the proposition that the processing of print ads follows an SCT framework. It should be noted, however, that the experimental task represented 8 fair amount of external validity. Given the assignment of evaluating an entire magazine, subjects received no forewarning as to the exact experimental stimulus. Further, they were kept unaware that the study involved a memory test. This lack of focus raises the possibility that the experimental impact or effect size was slight, and this might explain why the SCT model received little support.

However, the study's design requirements seem to have been met. Subjects exposed to predetermined atypical argumentation, concerning the pen's effect on handwriting, apparently attended to and processed that information. Their reported recognition of the presence of that information in the ad was significantly greater than that of their typical counterparts. Thus there is little doubt that the atypical information was in some way represented in memory. The effects of that information, however, deviated from what would be expected from an SCT-based representational process. Specifically, subjects exposed to the highly typical at made fever (not more) schema-based intrusions than subjects exposed to an atypical ad stimulus. This finding prevailed for subjects' recall protocols and, to a lesser extent, their responses to recognition items-i.e., item 6, with a marginal main effect and item 1, with an interaction effect unfavorable to the SCT model. Further, the typical subjects exhibited no better memory performance in terms of recall and recognition accuracy than did atypical subjects. In fact, on one of the recognition items (the second item involving an actually presented schema-typical argument), the findings were opposite of SCT expectations. Typical subjects performed considerably worse than atypical subjects.

In addition to the recall and recognition items, analysis of subjects' discrimination scores do not mesh with an SCT prediction. According to the SCT model, subjects should be better able to discriminate between presented and unpresented atypical information than typical information. This prediction is derived directly from the SCT notion that memory for a prose passage is partitioned into two segments: a schematic portion and a portion related to tagged items. Since the schematic portion is thought to consist largely of typical items imported from previously formed knowledge structures, its contents are viewed as relatively indistinguishable in terms of whether they originated from previous knowledge or some target event. Thus, subjects should have more difficulty in discriminating between presented and unpresented typical information than discriminating between "tagged" information and unpresented atypical items - i.e., tagged items are thought to be represented in a separate portion of the trace. Accompanying this line of reasoning is a second SCT prediction having to do with the ability to discriminate between presented and unpresented information over time. Since tagged information is thought to decay more rapidly than typical items, the SCT model predicts that over time differences in ability to discriminate between various forms (presented/unpresented) of typical and atypical information should decrease. That is, as tagged atypical information is lost, it tends to merge with unpresented atypical items. Thus, atypical items should lose their advantage in discriminability. Neither of the above predictions were upheld by the data. Although a slight advantage in discriminating atypical items was noticed (0.76 > 0.62), this difference was far from being significant.

In sum, although many interesting findings emerged, a conservative conclusion must be that a schema-copy-plus-tag hypothesis does not account well for the data of this study. Precisely why this is so is not apparent, since no clear pattern of results emerged. In fact, the mixed findings make it difficult to offer alternative formulations. For instance, a depth-of-processing approach would predict better memory performance for subjects in the atypical groups. This did not occur, however. Another schema-based paradigm suggests that unless a schema is fully activated, incoming information is difficult to encode. As a consequence, memory performance suffers and reconstruction is more likely (see Alba and Hasher). It is possible that this occurred in the atypical group. That is, the presentation of atypical information in some way disarmed subjects, depressing schema activation. If this is so, these subjects would be more likely to exhibit schema-related reconstructions st retrieval time. Although there is some evidence of this occurring, the data do not square entirely with the notion of an encoding deficit. That is, if atypical information inhibited encoding, then these same subjects should have also exhibited poorer recall and recognition of presented items. This, however, did not occur.

It should also be noted that the SCT formulation might resist explication because of its failure, in reference "retrieval" predictions, to distinguish adequately between recognition and recall. As Krugman (1985) reminds us, these are measures of different kinds of memory, and to lump them together might be to mask the individual effects of either.

Finally, as suggested earlier, the unclear results likely have something to do with the efficacy of the study's manipulation. Although requiring subjects to read a lengthy test magazine may have improved external validity, it may simultaneously have suppressed experimental impact. If this is so, the schema-copy-plus-tag model may yet prove to be a viable representation of memory for ads. A more powerful test of SCT will be necessary, however, to determine whether the representation and recall of ads can be made to lie comfortably on this procrustean bet.


Alba, Joseph W. and Lynn Hasher (1983), "Is Memory Schematic?" Psychological Bulletin, 93 (March), 203-321.

Cialdini, Robert B., Richard E. Petty and John J. Cacioppo (1981), "Attitude and Attitude Change," in Annual Review of Psychology, 32, 357-404.

Graesser, Arthur C. and C. V. Nakamura (1982), Impact of a Schema on Comprehension and Memory, Psychology of Learning and Memory, 16, 60-109.

Hasher, Lynn and Mary Griffin (1978), "Reconstructive and Reproductive Processes in Memory," Journal of Experimental Psychology: Human Learning and Memory, 4 (July), 318-330.

Hunt, James M. and Jerome B. Kernan (1984), "The Role of Disconfirmed Expectancies in the Processing of Advertising Messages," Journal of Social Psychology, 124 (December). 227-236.

Krugman, Herbert E. (1985), "Point of View: Measuring Memory - An Industry Dilemma,- Journal of Advertising Research. 25 (August/September), 49-51.

Schmidt, Daniel F. and Richard C. Sherman (1984), "Memory for Persuasive Messages: A Test of a Schema-Copy-Plus-Tag Model," Journal of Personality and Social Psychology, 47 (July), 17-25.