Learning Context and the Development of Product Category Perceptions

Eloise Coupey, Duke University
Kent Nakamoto, University of Arizona
ABSTRACT - Recent studies of consumer decision making suggest that consumer perceptions of product categories play an important role in brand evaluation. The present research examines the influence of the task context (preference vs. perception) in which category learning takes place on the emergence of category perceptions. We propose that because the task biases the consumer's use of product information the task also biases category perceptions. Thus the learning context is expected to affect subsequent product judgments and preferences.
[ to cite ]:
Eloise Coupey and Kent Nakamoto (1988) ,"Learning Context and the Development of Product Category Perceptions", in NA - Advances in Consumer Research Volume 15, eds. Micheal J. Houston, Provo, UT : Association for Consumer Research, Pages: 77-82.

Advances in Consumer Research Volume 15, 1988      Pages 77-82

LEARNING CONTEXT AND THE DEVELOPMENT OF PRODUCT CATEGORY PERCEPTIONS

Eloise Coupey, Duke University

Kent Nakamoto, University of Arizona

ABSTRACT -

Recent studies of consumer decision making suggest that consumer perceptions of product categories play an important role in brand evaluation. The present research examines the influence of the task context (preference vs. perception) in which category learning takes place on the emergence of category perceptions. We propose that because the task biases the consumer's use of product information the task also biases category perceptions. Thus the learning context is expected to affect subsequent product judgments and preferences.

INTRODUCTION

The organization of product knowledge in memory -has long been considered a critical factor in consumer decision making. Recent studies have drawn on a variety of organizing principles for consumer knowledge, including schemas, scripts, explicit rules, and categories (Marks and Olson 1981; Sujan 1985; Wright 1986). These structures enable the consumer to create meaningful representations of information that will enable him to simplify, to impose order, and to control his decision processes. In the present study, we focus on product categories as a means by which consumers might structure product knowledge. Sujan (1985) has suggested that the use of such categories enables the consumer to rapidly evaluate a product. Once a novel item is classified as an example of a known category, the affect associated with the category can be associated with it.

A number of studies have begun to examine the use of categorical or schematic knowledge in consumer judgment; however, these studies all assume preexisting category perceptions. The development of consumers' category perceptions through product experience has received much less attention. We wish to begin to explore this issue, under the assumption that consumers learn about product categories through exposure to and experience with specific items or brands in a product class. This category knowledge can then be applied by the consumer in subsequent judgment tasks.

We argue that various environmental factors influence the way consumers encode product information in memory. As a result, a consumer's perceptions of the category, and consequently the brands assigned to that category, are biased. This paper will examine learning goals and the distribution of product features across brands in the category as factors that might vary in the consumer's exposure to a group of products, and which might influence the encoding and retrieval of product class knowledge in memory.

THE STRUCTURE AND REPRESENTATION OF PRODUCT CATEGORIES

Two different but interrelated frameworks for describing category learning and structure have emerged in recent studies: exemplar models and feature-based models (Estes 1986). Both frameworks adopt two principles of categorization (Rosch 1977): (1) a need for cognitive economy, getting the most valuable information for the least cognitive effort, and (2) a need to assume that the world as perceived is inherently structured -- not just a random collection of attributes.

In exemplar models categorization is accomplished by encoding all specific instances of item occurrence in long term memory. Subsequent category decisions are made by consulting the stored array of exemplars and comparing them to the newly presented item. The item is then assigned to a category on the basis of its similarity to instances of that class (relative to other classes). In contrast, proponents of the feature-based approach to categorization assume that information about specific instances or exemplars is lost. Instead, only information about the relative relations between features is stored for future use. In the present study, dependent measures are taken that are intended to capture the encoding strategy employed by consumers, whether exemplar or feature-based. The information derived from these measures will also provide an opportunity to explore the possibility that different category perceptions are formed depending on the use of different encoding strategies, and whether such use is dictated by the task context.

Studies of categorization have focused primarily on common taxonomic categories, collections of exemplars or featural configurations which are related by some degree of similarity of features. In contrast, recent work by Barsalou (1985) has shown the need for consideration of what he terms "goal-derived" categories or purpose-directed groups of exemplars, such as "things to consider when buying a car," and "what to eat on a diet." Barsalou's research implies that people form category representations differently, depending upon the expected use for the stored information. This means that it is possible to delineate two separate types of memory structures for a product, one based on the consumer's general knowledge of a product class, a "perceptual" category (similar to the common taxonomic category), and another type of representation created in order to make a choice -- a goal-derived category. (For purposes of simplicity and clarity, we will assume that choice is based on preference.)

Barsalou notes that common taxonomic categories can be characterized by the central tendency (median or mode) of the frequency distribution of features or attribute levels over instances. The likelihood that an item is perceived as a member of a category is related to the proximity of its features to these modal values.

This conclusion does not generalize well to goal-derived categories. Members of a goal-derived class may share few features, and items in the category may be complements rather than substitutes for one another. In addition, the goal often dictates searching for extreme rather than typical product features (e.g., zero calories for diet foods rather than average number of calories per serving). The differences in the structure and use of the two types of categories lead Barsalou to conclude that while central tendency predicts the structure of common taxonomic categories, the frequency with which one encounters an object as a member of a category (frequency of instantiation) and the proximity of the object's features to ideal levels play analogous roles in goal-derived categories.

It would appear, then, that one important determinant of the representation of category-level information in memory is the learning goal. Two such goalsCchoice and judgmentChave been considered by Biehal and Chakravarti (1982, 1983). In their view, choice implies a two-step process in which the consumer first compares products in an offered set and then picks one. Judgment refers to the overall evaluative process by which the consumer examines the available information and then organizes it into a category representation.

Logically, it would appear that the consumer will have higher overall retention for items encountered in judgment than in choice, because choosing implies picking one, so that inferior brands can be ignored. This view of the memory processes operating in categorization also implies that the consumer will have higher retention for the items chosen. Consistent with these suppositions, Biehal and Chakravarti (1982) showed that subjects had higher recall for product attribute information in a "directed learning" (judgment) condition. Only for the chosen alternative was recall comparable in the choice condition.

Factors That Influence Encoding and Retrieval

In general terms, the manner in which people process information is determined largely by the amount and type of information available to them, and by the purpose for which the information is to be used. In the present study we focus on learning goal as a determinant of the type of information processing pursued. It is important to note that we are not concerned here with learning motivation, which appears to have little effect on retrieval performance (Craik and Tulving 1975). Rather, the learning goals are expected to affect the types of processes that are carried out at encoding. Although motivation may drive these processes, differing retention depends on differing processes, not the motivation itself (Anderson, 1980).

Biehal and Chakravarti (1982) found that subjects processing information to make a choice used primarily attribute-level information, whereas subjects processing to make a judgment stored information by brand. Brand processing seems to operate as a convenient memory heuristic, as it enables consumers to "chunk" information for faster, more complete storage. In contrast, processing information to make a choice requires the consumer to analyze and compare as many facets of the product with others of its class as possible, hence the attribute-level processing. Biehal and Chakravarti focused on brand vs. attribute -based processing strategies, and examined recall of brand/attribute data.

In the present study we are interested in the representation of more general category characteristics, so we will examine perceptions of overall category characteristics as a function of learning goal as well as recall of product information. In addition, we consider two intrinsic properties of the categoryCnumber of exemplars and skewness of the distribution of features over the set of exemplars. We expect that different learning goals will bias the consumer's sensitivity to changes in these objective characteristics.

To summarize, we suggest that preference and perceptual judgment tasks will lead to distinctly different types of category perceptions. Structuring processes, notably encoding and inter-item organization, are influenced by the purpose for which the category information is to be used. Although many of the factors which influence the structuring processes are similar for both types of categories, the degree and manner in which they are employed may differ depending on the type of category to be formed. This leads to differential storage and consequently differential availability of information.

EXPERIMENT

In this experiment the effects of three different learning goals on category structure are examined. The goals are preference, evaluation of inter-brand similarity, and directed learning of the brand data for recall. We suggest that subjects who learn about a category in the context of choosing a preferred alternative will form a different category representation than subjects who learn about the category in the context of similarity evaluation or in anticipation of a recall task. Because in preference subjects focus on better brands and features:

H1a: Preference goal subjects will be able to recall fewer brands than will similarity and directed learning subjects;

H1b: Preference goal subjects will provide poorer estimates of inferior attribute values than similarity and directed learning subjects;

H1c: Preference goal subjects will show more ability to recall and describe an ideal brand than will similarity and directed learning subjects.

The second factor manipulated in this experiment is the skewness of the distribution of presented exemplars. We manipulated the distributions of one feature so that subjects saw either a left-skewed, right-skewed, or symmetric distribution. This was done in order to ascertain the degree to which subjects are sensitive to the objective shape of the distribution, and how the learning v goal affects this sensitivity. Flannagan, Fried and Holyoak (1986) have found that people seem to have a prior expectation that these distributions will be unimodal and symmetric. As noted earlier, however, a goal of preference would be expected to bias information search toward superior values. With the manipulation of skew, then, we expect that subjects who learn about a category by assessing similarity or in preparing for a recall task will attempt to organize the presented information as a "normal" distribution, whereas preference goal subjects will utilize a non-normal distribution of "best to worst" ranking without regard to the most frequently occurring levels. Thus,

H2a: Perceptions of the range of attribute levels for preference goal subjects will show less sensitivity to manipulations of skewness than subjects in the similarity and directed learning conditions unless the skewness increases the frequency of brands at the desirable end of the range:

H2b: Subjects in the similarity and directed learning conditions will bias modal estimates toward a perceived "central tendency," the center of the range;

H2c: In general, similarity and directed learning subjects will provide more accurate estimates of modes and modal frequencies than preference goal subjects.

The final factor considered in the following study is the number of exemplars presented. Two competing effects are expected to operate. First, increasing the number of exemplars should give stronger impressions of the range of attribute values associated with the category. On the other hand, a subject's ability to discriminate exemplars in memory should decline. In view of these competing effects, this factor is treated as exploratory. No formal hypotheses are presented with respect to its interaction with goal and skew in influencing category structure.

Method

One hundred students were paid for participation in the two-part study. Each participant was first exposed to a set of descriptions of microwave ovens. Each description (brand) was given a fictitious name, and was characterized by weight, power, length of warranty and noise level. Subjects were told that all other characteristics of the ovens were the same for all brands.

Brands were presented in sixteen pairs, and this exposure constituted the learning period. The learning goal was manipulated by assigning three different tasks, one per subject, during the learning period. These tasks were:

1) to study the brands carefully, learning as much about each brand as possible (the directed learning condition);

2) to rate the similarity of each pair of brands on a nine-point scale (the similarity condition); or

3) to state their preference for one brand in the pair over the other (the preference condition).

While the number of brand pairs was fixed for all subjects, the number of different brands presented varied. Either 10 or 16 brands were presented. Thus, for the sixteen brands condition, the subject saw each brand twice. In the ten brand condition, all brands were presented at least three times, and two were presented a fourth time.

In addition to the learning task and number of brands presented, the distribution of the brands on the descriptive attributes was manipulated. Each of the four descriptive attributes assumed five levels over the set of brands. Power level varied from 712 to 1502 watts in roughly 200 watt increments. Weight varied for 24 to 91 pounds in roughly 15 pound increments. Noise level varied from very quiet to very noisy, and warranty length from 0 to 24 months in six month increments. For weight, noise level and length of warranty, the distribution of brands of the five attributes was fixed for all subjects. The distribution was symmetric, and from lowest to highest level the frequency of occurrence of each level was 1,2,4,2,1 for the ten brand set and 2,3,6,3,2 for the sixteen brand set. For the power attribute, the same symmetric distributions as before were used for one-third of the subjects. The other subjects saw either a left- or right-skewed distribution. In the left-skew condition, the frequency of occurrence of the five power levels was 1,1,2,2,4 or 2,2,3,3,6. In the right-skew condition the frequencies were reversed.

Following the learning period, an unrelated distraction task was assigned. Then, subjects were asked a series of questions about their perceptions and preferences regarding the brands of microwave ovens they had seen previously. Subjects were asked first to list all brand names they could recall. Next, they were asked to specify the highest and lowest levels of the four attributes that had appeared in any of the product descriptions. These data also allowed us to assess perceptions of range, calculated as the difference between the maximum and minimum attribute levels. Subjects then specified the modal value of each of the four features (the value that appeared most frequently), and provided an estimate of the perceived frequency of its appearance. Finally, subjects were asked to name their ideal brand, and to list the attributes of that brand.

Results

Our first hypothesis, that subjects will form different category representations depending on their learning goal, was examined through brand recall, perception of range and modal values, and recall of an ideal brand. First, the number of actual brand names recalled from the learning period was calculated. An analysis of variance on correct brand name recall with number, skew and goal as factors was performed. The main effect of goal was significant (F2 83=17.23, p<.01), with preference goal subjects producing the greatest number of correct responses and similarity goal subjects the fewest. There is also a significant interaction of goal and skew for correct brand responses (F4,83=2.86, p<.05). Recall for preference improves as the frequency of more desirable product descriptions (higher power) increase. No other effects or interactions were significant.

TABLE 1

MEAN NUMBER OF BRANDS CORRECTLY RECALLED

These results support the hypothesis that learning goal affects the structure of the category being formed, as preference subjects appear to be more concerned with remembering and organizing the more highly desirable items. These results also suggest that similarity subjects are occupied with attribute-based comparisons of similarity, and, contrary to expectation, are unable to simultaneously process and encode attribute information and brand names.

An analysis of variance with number, skew and goal as factors on the numbers of incorrect brand names listed (names that had not been presented during the learning period) is consistent with the previous results. The effect of goal is significant (F2.83=4.95, p<.01); preference subjects list, overall, the fewest number of wrong brand names. Similarity subjects have almost equal amounts of correct and incorrect names, which indicates that are unable to discriminate between what they feel are correct and what are actually incorrect product names.

An effect of skew that approaches significance for incorrect responses (F2,83=2,57, p<.10) indicates that the number of incorrect responses generated increases as the skew increases the number of desirable brands. This suggests the possibility that, regardless of the learning goal, subjects generally anchor on the more desirable items, at least with regard to products. No other effects were significant.

TABLE 2

MEAN NUMBER OF BRANDS INCORRECTLY RECALLED

With respect to category characteristics, we expected that preference would focus attention away from -inferior levels, but not from superior ones. For the three interval level attributes -- power, weight, and warranty length, we examined the difference between the reported minimum value and the actual value. An analysis of variance with number, skew and goal as factors for minimum weight resulted in a significant effect of goal (F2,72=2.97, p<.05). Preference subjects were less accurate, as predicted, but so were similarity subjects. However, there was evidence of confusion on the part of the subjects as to whether greater weight in a microwave oven was more desirable than less weight.

Results for minimum warranty provide stronger support for H1b. An ANOVA with all factors found a significant interaction of number and goal (F2,71=3.70, p<.05). This result indicates that preference subjects were less able to accurately recall the minimum warranty period than either directed or similarity goal subjects in the ten brand condition. The presentation of additional exemplars in the sixteen brand condition appears to aid recall of these category characteristics for similarity and preference goal subjects while detrimentally affecting directed learning goal subjects. The same pattern obtains for power, but because the analysis involves the manipulation of skew for this attribute, the analysis is reported below. Thus, the differing abilities of subjects to accurately recall minimum attribute values appears to be determined by the learning goal, which affects the type of information encoded hence influencing the type of category representation formed.

Subjects were also asked to state an ideal brand: the one brand of all those presented that they would most like to own. If the subject provided a brand name that had been presented, it was counted as a correct response. Partial names and not-presented names were considered incorrect. The frequency of correct versus incorrect responses was examined as a function of learning goal. As expected, the frequency of correct responses was higher for preference subjects than for similarity goal subjects, but it was unexpectedly higher for directed learning subjects as well (X2=17.9, p<.001). Again, we conclude that the learning goal manipulation leads preference subjects to form category representations that focus more on the desirable items than subjects who evaluate similarity.

Our second hypothesis concerns the effects that differences in skewness of groups of exemplars would have on subjects' tendencies to organize product information into normal or non-normal distributions. We looked at estimates of mode, modal frequency and range values to assess sensitivity to skew as a function of learning goal.

Subjects' perceptions of the modal value for power provided a manipulation check for skew. An ANOVA for all possible effects and interactions of number, skew, and goal revealed a significant effect of skew (F2,73=6.97, p<.001). As expected, subjects are affected by skew and in general are oversensitive, so that subjects presented with normally distributed sets are best able to state the mode, while right-skew subjects (numerous undesirable brands) underestimate it, and left-skew subjects (numerous desirable brands) overestimate it.

Skew was expected to influence perceptions of extreme values in the-category, as measured by the reported minimum and maximum attribute levels and the implied range calculated from these values. For power, an ANOVA by number, skew and goal finds only an effect of. skew on the range between lowest and highest possible power levels (F2,83=7.30, p<.001), such that subjects in the right-skew condition were most nearly able to accurately recall the range of presented power levels. Subjects in the left-skew condition gave inflated range estimates, while subjects in the symmetric condition underestimated the range.

One possible explanation for these results is that subjects have two natural reference points for considering category information: the mean and he best end. However, the low (worst) end is salient only in the right-skew condition when the frequency of brands at that end is high. Subjects in the left-skew condition, anchoring on the best levels, have hazy perceptions of the lower levels, seeing them as very undesirable and as a result, very far away. This leads to inflated estimates of range. This interpretation is further supported by a multivariate analysis of variance performed on the estimates of maximum and minimum power levels with number, skew and goal as factors, in which the effect of skew was significant (F4,138=3.71, P<.01). For the maximum levels reported, the accuracy of perception follows the skew: accuracy improves as the modal value improves. For the minimum level, subjects in the normally distributed condition are best able to state the correct value. Right-skew condition subjects, faced with an abundance of less desirable items, have a pessimistic view of the minimum and maximum, recalling both as lower than they actually were. Left-skew condition subjects, also focusing on the more desirable levels, have a less accurate perception of the low end, seeing it as very different, and hence underestimating it.

One further finding supportive of this biased focus can be seen in the modal values reported for weight, an attribute for which skew was not manipulated. Subjects appeared to "halo" off power, so that weight showed effects of skew. Specifically, for the right-skew distribution accuracy of reported mode improved as number of exemplars increased, but for the left-skew and symmetric conditions the accuracy of perception decreased as number increased. Also, as in the case of minimum power estimates, subjects exhibited a tendency to assign conservative estimates of mode in the left-skew condition for weight and to some extent for noise (p<.10). An analysis of variance was also performed on subjects' estimates of the frequencies with which the mode occurred, again with number, skew and goal as factors. For weight, a significant effect of number by skew F2,79=3.98, p<.05 supports the preceding findings of conservatism and haloing. As mentioned earlier, we included the manipulation of the number of presented exemplars to observe the effects of the amount of information to be learned. In the MANOVA on maximum and minimum estimates of power, the number by goal interaction was significant (F4,138=2.40, p<.05). It appears that the accuracy of perception improves for similarity and preference goal subjects as the number of presented exemplars increases, but that the opposite is true for directed goal subjects. These results indicate that subjects who learn the product information without a well-defined goal or strategy, such as the directed learning subjects, are at a disadvantage as the number of exemplars increases; they become overloaded with -unorganized information.

TABLE 3

MEAN REPORTED POWER MAXIMA AND MINIMA

The idea of information overload is supported by a MANOVA on perceptions of modal levels for power, weight and noise, with number, skew and goal as factors. The effect of skew was significant (F4,144=3.39, p<.01) as was the effect of the number and goal interaction (F4,144=2.61, p<.05). As the number of presented exemplars increases, the accuracy of similarity goal subjects decreases, and the accuracy of preference goal subjects increases.

TABLE 4

MEAN PERCEPTIONS OF MODAL POWER LEVEL

DISCUSSION

We have seen that learning goal and skew separately and jointly have implications for the organization and completeness of encoded product information. In general, it appears that directed learning subjects, without a strategy for learning, opt for a sketchy perusal of attributes, placing greater emphasis on a brand-based organization. Similarity goal subjects take the opposite tack; they focus primarily on attributes, showing little inclination or ability to link the attributes as features of a particular brand. As a result, they are less prone to exhibit haloing, as they consider each attribute separately.

Thus, in processing product information by attributes, subjects construct categories that include more complete range information, including more accurate perceptions of maxima and minima. At the same time, it does not appear that similarity subjects distill summary representations, such as preferred brands, from the attribute-level information.

Falling into the continuum of brand-to-attribute processing are the preference goal subjects. Although they are cognizant of the importance of the manipulated attribute, power, they do not rely upon attributes as the basis for category structuring. Instead, they appear to use the attribute information to make a preference judgment at the moment of comparison and then encode the preferred brand into the category. While this method does enable them to isolate one ideal brand, it also hinders their ability to recall its relative merits by attribute.

Preference goal subjects create categories that are designed to serve one purpose: to make the optimal choice. As such, the recall of lower range values is poor; these values were presumably eliminated as unimportant because they were undesirable. In addition, the recall of other attributes is distorted due to the tendency of preference goal subjects to halo off the most important attribute. This suggests that they do not attend to all attributes equally, even at the time of initial presentation and encoding. Instead, they opt for a heuristic approach of relying heavily on the most important attribute.

The amount of information to be learned also interacts with skew and goal to affect category perceptions. When category perceptions were strong, increased numbers of exemplars may improve accuracy of category perceptions. The cost is increasing confusion regarding the particular brands. In the absence of strong category-level perceptions, increasing the number of exemplars only served to confuse perceptions of category characteristics.

Several limitations and suggestions for future research should be noted. First, this study was confined to the initial phase of category perception formation. One direction in which to extend the results of this study would be to examine the influence of extant structures on future category decisions. While the directed learning subjects in this study appeared to focus on brands, they developed weak category perceptions. Thus, it seems that brand-oriented processing need not be linked to the use of category-level information. Examining the effects of these differing structures might help to delineate the roles of brand and category data.

In addition, an exploration of the effects of order of presentation and amount of inter-item variance of exemplars in category learning is suggested by the work of Elio and Anderson (1984). This extension would necessitate the inclusion of a transfer phase (in which novel instances are presented for consideration as members of the category) to examine the effect of order and variance on the "confirming' sets of exemplars.

CONCLUSION

It has been proposed that category perceptions influence consumer decision processes. The current study begins to investigate the prior question of how category perceptions arise in the first place. Specifically, we have demonstrated that global category perceptions are prone to numerous biases as a function of the learning environment. An examination of these biases not only provides insight on how consumers deal with novel product information, but also suggests ways in which product category perceptions might be invoked in decision processes.

REFERENCES

Anderson, J., (1980), Cognitive Psychology and Its Implications San Francisco: W. H. Freeman and Company.

Barsalou, L. (1985), "Ideals, Central-Tendency, and Frequency of Instantiation as Determinants of Graded Structure in Categories." Journal of Experimental Psychology: Learning, Memory, and Cognition, 11 (October), 629-654.

Biehal, G. and D. Chakravarti (1982), "Information-Presentation Format and Learning Goals as Determinants of Consumers' Memory Retrieval and Choice Processes," Journal of Consumer Research, 8 (March), 431-441.

Craik, F. and E. Tulving (1975), "Depth of Processing and the Retention of Words in Episodic Memory," Journal of Experimental Psychology: General, 104, 268-294.

Elio, R. and J.R. Anderson (1984), "The effects of information order and learning mode on schema abstraction," Memory and Cognition, 12 (January), 20-30.

Estes, W.K. (1986), "Memory Storage and Retrieval Processes in Category Learning," Journal of Experimental Psychology: General, 115, 155-174.

Flannagan, M., L Fried, and K. Holyoak (1986), "Distributional Expectations and the Induction of Category Structure," Journal of Experimental Psychology: Learning, Memory, and Cognition, 12 (April), 241-256.

Johnson, E. and J.E. Russo (1984), "Product Familiarity and Learning New Information," Journal of Consumer Research, 11 (June), 542-550.

Marks, L.J. and J. Olson (1981), "Toward a Cognitive Structure Conceptualization of Product Familiarity," in Advances n Consumer Research, Vol. 8, ed. K. B. Monroe, Ann Arbor, MI: Association for Consumer Research.

Rosch, E. (19707), "Principles of Categorization," in Cognition and Categorization, eds. E. Rosch and B. Lloyd, Lawrence Erlbaum Assoc., Publishers; Hillsdale, N.J.

Sujan, M. (1985), "Consumer Knowledge: Effects on Evaluation Strategies Mediating Consumer Judgments," Journal of Consumer Research, 12 (June), 31-46.

Wright, P. (1986), "Schemer Schema: Consumers' Intuitive Theories About Marketers' Influence Tactics," in Advances in Consumer Research, Vol.13, ed. R. Lutz, Gainesville, FL; Association for Consumer Research.

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