The Organization of Product Information in Memory Identified By Recall Times

The 2 x 2 Latin square divided the subjects into two groups (with 10 subjects in each group). Because 8 recall times (4 attribute cues and 4 brand cues) were obtained from both products and from each of the 20 subjects, a total of 320 observations formed the data base.

Subjects

Twenty-four undergraduates at Carnegie-Mellon University participated in the experiment. They were either volunteers (2) or were fulfilling a course requirement for introductory psychology (22). Two subjects were eliminated because they made more than three errors on the sixteen possible trials. Additionally, two subjects took an excessive amount of time to learn the information booklet for the first product. They were not asked to finish the experiment and their data were discarded. Thus, a total of twenty subjects provided data for the experiment.

When the response time data were analyzed, the means of 19 subjects ranged between 8.8 sec and 16.4 sec. The remaining subject's mean was 33.4 sec. Because this subject's responses were so unique and because they would thoroughly disrupt any pattern of results that held over the other 19 subjects, he was dropped from the study. One of the two subjects who failed to pass the error criterion had participated in the same condition as the dropped subject. This former subject was returned to the sample to maintain ten subjects per group.

Procedure

Subjects were told that they were participating in a study examining consumers' knowledge of various products. They were then asked to learn the contents of the first booklet. Instructions emphasized that the material in the booklet had to be remembered accurately. Subjects were also asked not to switch between pages of the booklet but rather to look at the book one page at a time. After they had informed the experimenter that they thought they had learned the booklet, subjects were administered a criterion test. Subjects were asked to name the value of four different product-attribute combinations. If a subject could not successfully name the four values, he was given the booklet and told to try again. Subjects who passed the criterion test went on to the cued speeded recall task. The total time until the subject passed the test was recorded.

In the recall task subjects were presented either an attribute or a brand name. If the probe was an attribute name, they were told to recall all of the appropriate brand names and their values. For a brand name probe, they recalled the various attributes and values for that brand. The recall time was measured (in msec) from the onset of the probe to the end of the list. For each booklet the subject received four attribute probes and four brand probes. Order was randomized across subjects.

After completion of the first recall test, subjects were given the second booklet, this time with the opposite organization. The procedure for criterion testing and cued speeded recall was repeated.

Finally, subjects were presented an empty matrix for the first product class. The four attributes and the four brands were given but the sixteen values were missing. The subjects were asked to recall these values, and to write them down, in whatever order they wished. The output order was recorded. This was done for the second product as well. Then the subjects were debriefed and released.

The written free recall measure is not necessarily an independent measure of memory structure. Inherent in the cued memory task is recall by both brand and attribute. This task demand could cause reorganization which could well affect subsequent free recall. Also the time between learning the information and recall varies and the structure as therefore measured may be different than that measured with the cued recall. High convergence between these two recall methods may well wait until they are tried under a different experimental procedure.

RESULTS

Encoding Time and Error Rate

A preliminary analysis tested for differences in the time necessary to learn the product information. An analysis of variance showed that learning times did not differ between the types of input organization, the different products or the two groups of subjects. Also, no interactions among these factors were significant. This precludes the possibility that differences in recall time could be due to differences in how well the lists were initially learned.

In the same way, differences in recall time could be confounded with differences in error rates across conditions. If speed is traded off for accuracy, then a difference in recall time can be due to different error rates as well as to the effect of an independent variable. Error rate was defined as the proportion of cued recalls for which at least one of the four required values (belonging to a single brand or a single attribute) was incorrect. The overall error rate was 7.8% (25 of 320). Using individual X²-tests for the equality of two proportions, no significant differences in error rate were found between levels of input organization, product category, or cue type. Taken together, these analyses imply that differences in error rates or thoroughness of learning can be ruled out as explanations for observed differences in recall speed.

Recall Times

Differences in recall times were tested by an analysis of variance. An examination of the residuals of the appropriate model indicated that a log transformation was needed to stabilize the variance over different levels of recall time. The results of the analysis of variance on the transformed recall times are reported in Table 1. The total degrees of freedom, 294, reflect the removal of the 25 recall trials on which an error occurred. Note that there are no replications in the design. The error term was estimated from the Group x Probe and Subject x Probe interactions. If these terms include any systematic efforts, the resulting F tests will be conservative.

TABLE 1

ANALYSIS OF VARIANCE OF RECALL TIMES

The critical prediction concerned the A x C interaction. This interaction captures the effect of storage organization (as induced from input organization) on the cued recall. It was predicted that recall would be faster when the probe type matched the structure stored in memory. Attribute probes would result in faster recall with attribute-organized information, and brand probes would be faster for brand-organized information. The "incompatible" probes would lead to slower recall.

As predicted, the A x C interaction is statistically significant and, furthermore, is the largest effect (in terms of mean squares) in the model. Before pursuing this and related effects, the subject and cue differences are noted. The presence of significant subject differences is typical in chronometric analysis. They are potentially attributable to many factors including motivation and strategy differences. There was no difference between groups, indicating that subject differences were not systematically related to group. Note that in this design the A x B interaction is confounded with group. Therefore, the lack of a significant group effect validates the assumption of zero A x B interaction. The only other possibility is that both effects were of equal magnitude and opposite direction, thus canceling each other out. Individual cue effects (Factor D) were also significant. This result was explored with a series of a posteriori contrasts. These tests revealed only one significant difference: for cooking oil the cue "cost" was faster than the cue "fat count." In general, the differences among cues exhibited no systematic relation, and they are of little intrinsic interest. The observed differences may be due to any articulation time or to irrelevant factors affecting memory, such as word frequency or concreteness.

To communicate the results involving the major independent variables, means over subjects and cues were computed. These values are plotted in Figure 2. The two pairs of crossing lines represent the A x C interaction, separately for the two product categories. Using a priori contrasts, both interactions were found to be significant. For air conditioners F(1, 203) = 47.30, p < .001; for cooking oils the interaction was less dramatic, F(1, 203) = 5.93, p < .05. Both the contrasts and an inspection of Figure 2 reveal that, though the predicted effect of compatibility between cue type and input organization is present for both product categories, the size of this effect differs across products.

FIGURE 2

MEAN RECALL TIMES

Returning to Table 1, it can be seen that both Factors B (product category) and C (cue type) show significant effects. As shown in Figure 2, cooking oil information was recalled faster than air conditioner information, holding all other factors constant. One possible explanation for this difference is articulation time. Though speaking times were approximately equalized across all brands and sizes within a product category, there were differences between categories. The total number of syllables over the eight brand or attribute names is 27 for air conditioners and 16 for cooking oil. Thus, the articulation time would be expected to be longer for air conditioners, as confirmed by the means in Figure 2. The other possibility is that cooking oil information is genuinely easier to recall. However, both products were learned as quickly and both yielded equal error rates during recall. These results suggest that the difference in recall times is not due to memory retrieval but rather to an output factor like articulation.

Differences between cue types are also apparent in Figure 2. The mean recall times were 12.97 sec for attribute probes and 13.70 sec for brand probes. This difference, though relatively small (2.7%), is intriguing. Although one can only speculate given the present data, this effect might be related to a more general preference for attribute information. Factor A, input organization, exhibited no significant effect. The equality of recall times over brand and attribute inputs is compatible with the absence of differences in learning time or error rate.

The B x C and G x C interactions are probably attributable to differences in large A x C interaction for each product combined with the nature of the design. As noted earlier, any A x B interaction is confounded with G. Thus, G x C is confounded with A x B x C, and the G x C interaction may well be explained by the product differences (Factor B) in the A x C interaction that are so apparent in Figure 2. In a similar way, the B x C interaction, which is smaller, seems to have no natural interpretation. It may be related to product differences or to the large A x C effect combined with the confounding of Factors A and B within groups. In sum, the G x C and the B x C interactions show that product category affects the strength and nature of the effect of display format on memory structure. However, these differences are small when compared to the main A x C interaction, accounting for less than a fourth of its variance.

Finally, the interactions with subjects (S x A, S x C and S x A x C) should be acknowledged as statistically significant. Since these effects are not large and reflect only individual differences, they are not particularly meaningful. For completeness, it is noted that the error term is composed of the G x D and S x D interactions.

In summary, the pattern of recall times indicates a strong effect of congruence between input organization and cue type. Attribute probes led to faster recall when the input organization was attribute-based, and similarly for brand cues with brand-organized information. This result confirms the main prediction of the study. In addition, marked product differences were found when the input x probe (A x C) interaction was examined separately for each product.

Results for Individual Subjects

Chronometric studies are usually based on data analyses that are complete within individual subjects. This requires the collection of many data points to enable reliable estimates of mean response time and statistical tests with adequate power. Collecting this large amount of data becomes more costly as the task becomes more cognitively complex. In such cases, it is not unusual to find a subject participating for 20 or more hours in one experiment (e.g., Dosher, 1976). In the present study only 16 response times are available for each subject. Thus, although statistical tests of within-subject differences can be performed, some real differences may be obscured by variability.

The 16 recall times were classified by input organization and cue type (Factors A and C), creating four groups with four observations each. For each input organization, there are two mean response times, one for attribute cues and one for brand cues. The difference between the latter two means can be used to measure congruence between memory structure and input organization. This congruence will be termed input bound-ness. If the product information was learned in an attribute format, the mean for attribute probes is subtracted from the mean for brand probes. The larger this difference, the greater the input boundness. These then can be thought of as estimates of the degree to which a subject's internal representation is congruent with the external source. For each subject two such measures exist, one for attribute-organized input and one for brand-organized input. By summing these two measures, individual estimates of input boundness are obtained.

Using the estimate of MSE obtained from the analysis of variance, a t-test was performed for each subject. The significance level for this t-test (one-tailed) was set at .10 to favor the detection of differences in spite of the small number of data points. Using this criterion, 8 of 20 subjects exhibited significant input boundness. Thus, the main prediction and finding of this study, the congruence between input organization and memory structure, may be characteristic of less than half the subjects.

The reduced number of subjects showing an effect of input boundness could be due, in part, to the reduced statistical power of the individual tests. Additionally, the individual estimates of input boundness are confounded with product category. The observed differences across products may weaken comparability between mean response times for a single subject. Finally, it must be remembered that typical chronometric analyses of individual subjects require much larger sample sizes than available here.

Written Free Recall

After both sets of product information had been learned and probed via the cued speeded recall task, a written free recall was collected. For each product class, an empty 4 x 4 matrix was shown. The row and column labels were provided, and subjects were required to write in the 16 missing cell values. Each such output sequence consists of 15 intercell transitions.

Each of these transitions can occur within a column, within a row, or both to a different row and a different column. Since the booklets arranged information by column, a column transition was judged to be congruent with the original display. Row transitions could only occur if the memory structure is dissimilar to the original display. Therefore, each transition was coded as congruent (column) or incongruent (row or other) with the original input organization. At the end of a column or row, when all entries are filled, no more transitions within that file are possible. Thus, transitions from filled files could not be used to measure congruity with input, and they were excluded from the analysis. This left a data base of 12 transitions for each protocol. The overall proportion of congruent transitions is .72 (344 of 480). This value indicates a strong tendency for the written free recall to be congruent with the input organization. This confirms the major finding from the recall time data, the significant tendency for memory structure to be determined by input organization.

The proportion of congruent transitions were partitioned by the two possible independent variables, input organization and probe type. These data are reported in Table 2. The striking aspect of the pattern of proportions is their uniformity except for the brand-cooking oil cell. The other three proportions are all quite high, indicating input boundness. But when cooking oil information had been learned in a brand-organized format, considerable reorganization occurred. That is, subjects tended to reorganize from a brand-based input to an attribute-based memory structure. Again, this result confirms the findings from the recall time data.

TABLE 2

PROPORTION OF FREE RECALL TRANSITION CONGRUENT WITH INPUT ORGANIZATION

At this point it is worth recalling the problems associated with the written free recall data that were discussed in the method section. The problematical nature of this recall technique, as it was implemented in this experiment, implies that all findings must be tempered with qualification. It is comforting that the free recall data averaged over subjects agree with the results based on the cued speeded recall task. However, a stricter test of the convergent validity between these two recall techniques can be based on the data of individual subjects. The proportion of congruent transitions was computed for each subject and correlated with the Bound statistic that is based on recall times and discussed earlier. Both are measures of input boundness. The correlation over the 20 subjects was .33, a value significantly greater than zero (p = .02) but not very high. It should be concluded that the free recall transitions are not as reliable as they should be--or as they could be with a different experimental procedure. However, when averaged over subjects, they probably present a satisfactorily accurate picture of memory organization.

DISCUSSION

Summary of Results

The main finding was the congruence between input organization and memory structure. Averaged over the twenty subjects, both cued recall times and the output order of the written free recall supported this conclusion. Individual subject analysis of recall times suggested that 10 of the 20 subjects were significantly input bound. Input boundness was sensitive to type of product, with air conditioners inducing relatively consistent input boundness and cooking oil exhibiting more reorganization of the memory structure.

An important question is whether people prefer attribute-based or brand-based memory structures. A very preliminary result of the present study is that attribute structures were preferred. This result is as intriguing as it is tentative.

Generality

The findings of this study, although clearly supporting the view that memory structure is input bound, should be considered only in light of product differences. There was much greater reorganization (less input boundness) for cooking oil than for air conditioners. Possible factors that can influence the input boundness of a product class need to be isolated. It is interesting to note, for example, that both the brand names and the attribute names were more familiar for cooking oil. Also, at least one of its attributes (polyunsaturated) fat count is health-related and apt to be especially salient. Clearly, a wider range of products will have to be examined before any relation between reorganization potential and product type can be established.

Because the subjects in this experiment were undergraduates, the findings may not generalize to representative shoppers. Bettman and Kakkar (1977) argue that such shoppers would be more brand conscious than undergraduates. A shopper's experience within a particular product category may contribute to brand consciousness or, alternatively, to the ability to reorganize by attributes. Questions of this nature can only be answered by studies designed to test the generality of the present findings over a much wider sample of products and subjects.

The Chronometric Technique

One of the positive results of this study is the demonstration of the usefulness of chronometric analysis. This is especially true for the study of the representation of product information in memory. The present use of chronometric analysis to identify the memory structure of product information parallels the recent recognition of the importance of this topic (e.g., Olson, 1977 and Bettman, 1978). However, chronometric analysis can be applied to a wide range of other consumer research problems. See Gardner, Mitchell and Russo (1978) for a discussion of this point and an application of this technique to low involvement advertising. As a final methodological note, the reader should be reminded that the chronometric technique used here is among the least sophisticated. More powerful methods are available and, hopefully, will be brought into the study of consumer behavior. However, even the simple chronometric techniques may be adequate given the size of the effects observed here which are considerably larger than those reported in most chronometric studies.

Implications for Policy

Memory structure for product information, and its relation to input format on the one hand and decision rules on the other, has important policy implications. We give one example. To the extent consumers are input bound, their memory structures will reflect the organization of product information in the environment. As just pointed out, this organization is largely brand-based (advertising, etc.). This is all well and good if consumers make better purchase decisions using brand-based strategies. But if attribute-based strategies are superior, and a growing body of evidence suggests this (e.g., Russo and Dosher, 1976), then brand-organized storage contributes to less accurate and more effortful purchase decisions. We may then find a conflict between superior decision rules and the organization of currently available product information.

This suggests a potential need for public policy intervention. More importantly, it suggests that any policy of information provision may want to alter the current format of product information. If the ultimate policy goal is to improve consumer decisions through information disclosure, the format of the information should be designed for compatibility with the preferred decision rules.

Future Research

A major direction for future research is to examine this relation between memory structures and decision rules. Does the structure of stored product information determine the decision rule(s) that can be employed? Or, conversely, does a preferred decision rule lead to the reorganization of product information into a compatible memory structure? The relative strength of influence between memory structure and decision rules is an important question in consumer research.

A related issue concerns the conditions surrounding the acquisition of knowledge about products. Consumers may intentionally retain information with the conscious goal of making a purchase decision. Knowledge can also be retained, possibly incidentally, for more genera], future use (Greeno. 1976).

A second topic for future research is the influence of different external organizations of product information on memory structure. What is the preferred memory organization if a complete brand x attribute array is available? Suppose the externally available information is very unstructured, possibly in a discursive format or only accessible in a random order. Will shoppers attempt to organize an unstructured external display into a consistently preferred internal structure? Finally, what if product information is available as it typically is in the market place: from several sources (advertising, use experience, word of mouth, published ratings, etc.) that are not naturally comparable and that may include inconsistencies? Further, most real-world product information is brand-centered. Does this imply that memory structures are usually brand-organized whether or not attribute-based structures ere preferred?

In the real world, memory of product information is the link between information acquisition and the purchase decision. This topic is a major understudied area of consumer behavior. There are many important research questions associated with it, only a few of which have been posed above. By actively studying memory of product information, it may be possible to close a major gap between current laboratory research and the real world consumer setting.

REFERENCES

John R. Anderson, Language, Memory and Thought (Hills-dale, NJ: Lawrence Erlbaum, 1976).

James R. Bettman, An Information Processing Theory of Consumer Choice (Reading, MA: Addison-Wesley, 1978, forthcoming).

James R. Bettman and Pradeep Kakkar, "Effects of Information Presentation Format on Consumer Information Acquisition Strategies," Journal of Consumer Research, 3(March, 1977), 233-240.

William G. Chase, "Elementary Information Processes," in W. K. Estes, ed., Handbook of Learning and Cognitive Processes, Vol. 5 (Hillsdale, NJ: Lawrence Erlbaum Associates, in press).

Barbara Anne Dosher, "The Retrieval of Sentences from Memory: A Speed-Accuracy Study," Cognitive Psychology, 8(July, 1976), 291-310.

James G. Greeno, "Indefinite Goals and Well-Structured Problems," Psychological Review, 83(November, 1976), 479-491

Meryl Gardner, Andrew A. Mitchell and J. Edward Russo, "Chronometric Analysis: An Introduction and an Application to Low Involvement Perception of Advertisement," in H. Keith Hunt (ed.), Advances in Consumer Research, Vol. 5, Association for Consumer Research, 1978.

Roger E. Kirk, Experimental Design: Procedures for the Behavioral Sciences (Belmont, CA: Brooks/Cole, 1968).

Donald A. Norman, Memory and Attention: An Introduction to Human Information Processing, Second Edition (New York: Wiley, 1976).

Jerry C. Olson, "Theories of Information Encoding and Storage: Implications for Marketing and Public Policy," presented at the Conference on the Effect of Information on Consumer and Market Behavior (Carnegie-Mellon University, May, 1977).

John W. Payne, "Heuristic Search Processes in Decision Making," in Beverlee B. Anderson, Ed., Advances in Consumer Research, Vol. 3, Association for Consumer Research, 1976.

Herbert A. Simon and John R. Hayes, "The Understanding Process: Problem Isomorphs," Cognitive Psychology, 8(April, 1976), 165-190.

Herbert A. Simon and Alan Newell, "Thinking Processes," in David H. Krantz, R. Duncan Luce, Richard C. Atkinson, and Patrick Suppes, Eds., Contemporary Developments in Mathematical Psychology, Vol. 1, (San Francisco: Freeman, 1974).

Saul Sternberg, "Memory Scanning: Mental Processes Revealed by Reaction Time Experiments," American Scientist, 57(1969), 421-457.

Endel Tulving and Gordon H. Bower, "The Logic of Memory Representations," in Gordon H. Bower, Ed., The Psychology of Learning and Motivation, Vol. 8 (New York: Academic Press, 1974).

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