When Do Consumers Infer Product Attribute Values?

Chris J. Hansen, Jeanneret & Associates, Inc.
George M. Zinkhan, University of Houston
ABSTRACT - Typically it is assumed that consumers evaluate products in terms of their ostensible attributes. An alternative model considers ostensible, accessed attribute values as cues which the consumer uses to make inferences about other, nonaccessed attribute values. A model is developed to explain when individual and task differences prompt inferences among attributes.
[ to cite ]:
Chris J. Hansen and George M. Zinkhan (1984) ,"When Do Consumers Infer Product Attribute Values?", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 187-192.

Advances in Consumer Research Volume 11, 1984      Pages 187-192

WHEN DO CONSUMERS INFER PRODUCT ATTRIBUTE VALUES?

Chris J. Hansen, Jeanneret & Associates, Inc.

George M. Zinkhan, University of Houston

ABSTRACT -

Typically it is assumed that consumers evaluate products in terms of their ostensible attributes. An alternative model considers ostensible, accessed attribute values as cues which the consumer uses to make inferences about other, nonaccessed attribute values. A model is developed to explain when individual and task differences prompt inferences among attributes.

INTRODUCTION

Customarily, it is assumed that consumers evaluate products in terms of their ostensible attributes. Consumers are presumed to search for attribute values either internally, in memory, or externally, in the environment. They process this information by a set of rules, and select an item for purchase. Theoretical models are formulated to relate the individual attribute evaluations to an overall evaluation of each product. All the proposed models typically assume that the consumer uses only the information presented or explicitly elicited.

However, McConnell (1968) has shown that consumers may infer the level of one attribute, quality, from the level of another attribute, price. Several studies followed, confirming that when price is the only independent variable, it is significantly related to quality (e.g., Peterson 1970). Huber and McCann (1982) have demonstrated that these inferred attribute values may weigh heavily in the choice of products. Purchase likelihood values vary greatly, depending upon whether or not inferences about attributes are prompted.

The research on price and quality has been considered generalizable to other types of product attributes and thus al alternative model has arisen. In this model ostensible, accessed attribute values serve as cues which the consumer uses to make inferences about other nonaccessed product attribute values (Huber and McCann 1982). These inferred attribute values are then combined with the accessed attribute values to arrive at an object choice.

Huber and McCann (1982) have suggested that a theory is needed to explain when individual and task differences prompt inferences among attributes. It is the purpose of this paper to explicate such a theory. Specifically, Huber and McCann induced inferential beliefs in the laboratory and found that these inferred beliefs had considerable influence on consumers' product evaluations and purchase intentions. From this perspective, the next logical step involves the investigation of inferential beliefs in a more naturalistic setting. Are attributes naturally inferred? The evidence cited above from the price/quality literature indicates that this is the case. When are such attribute inferences likely to be made? This last question is the focus of this paper.

If attributes are naturally inferred, then possible inferences should be considered in deciding what information to include in promotional campaigns. Leaving out certain information may lead to inferences about those missing attributes. In this sense, advertisers may be able to impart more information to consumers while explicitly stating less information in a promotional message. Thus, the study of attribute inferences could have important practical applications.

A starting point for theoretical development is the work of Huber and McCann (1982), who imply that there are three conditions which can lead to inference formation. The first is that there must be a high ecological correlation between the consumer's perception of the accessed attribute levels and his/her perception of the nonaccessed levels. Second, the nonaccessed attribute must be important so that an inference to its value has weight. And third, the context must be appropriate, though the necessary situational conditions are unspecified.

While Huber and McCann have suggested these conditions, they offer no rationale for them and urge that further theoretical development be undertaken in this area. Here, each of these conditions is evaluated, refined, and expanded. The cognitive processing which may underlie the observable conditions is elaborated. In this sense, a model is developed to describe the steps that might precede the formation of inferential beliefs. This model is represented in Figure 1.

In the sections which follow, each of the three conditions that precede attribute inferences is discussed. As these theoretical conditions are elaborated, it becomes clear that more than one observable variable may be necessary to capture the full flavor of each condition. Accordingly, an empirical model, which specifies six predictor variables, is developed and presented in Figure 2.

LINK BETWEEN ACCESSED AND NONACCESSED ATTRIBUTES

Consider the first theoretical requirement for attribute inference: a high correlation between accessed and nonaccessed attribute values. The association referred to here is within individuals. Relationships between attributes for groups of consumers are only relevant to the extent that they reflect relationships between attributes for individuals as well.

One appropriate method for determining the intra-individual association between attributes would be to have a group of subjects estimate the values of several brands with respect to two attributes. An ordinary Pearson-product-moment correlation could be computed for each subject to determine the degree of association between attributes perceived by each consumer. Subjects with high correlations (positive or negative) may infer one attribute from the other, but subjects with correlations near zero will not make such inferences.

Consider the possible mechanism for this association between attributes. Semantic information is thought to be stored in nodes or links (Bettman 1979). Nodes represent concepts and links are the relationships among the nodes. In the present theory, each attribute may correspond to a different node and when nodes are linked to one another inferences may be drawn from one attribute to another.

The strength of a link corresponds to how essential that link is to the meaning of the concept or attribute (Collins and Loftus 1975). In Nakanishi's (1974) model, contiguity of concepts or closeness of association is equivalent to the strength of a link. The stronger the link between two attributes, the more likely it is that an inference will be made. Operationally, the higher the correlation between levels of attributes, the more probable an inference. It is not necessary that the correlation be high, as Huber and McCann (1982) suggest. Other variables may compensate for a modest correlation between attributes; however, some minimum degree of contiguity is necessary. Empirical work needs to be done to determine how large the correlation coefficients between attributes must be in order for inferences to occur.

Patterns or chunks of information are stored in memory. Brand names, for example, may summarize chunks of attribute data (Jacoby, Szybillo, and Busato-Schach 1977). If the level of attribute b is inferred from the level of attribute a, a and b may be stored in the same chunk.

THE ATTRIBUTE GOAL AND HIGH INTERRUPT THRESHOLD

A second individual difference variable is hypothesized to be a necessary condition for a consumer to infer the level of one attribute from the level of another attribute. The inferred attribute must be important. That is, the attribute must be weighted highly when its levels are accessible.

Bettman (1979) theorizes that in the process of making choices, consumers form goal hierarchies. A consumer may have a goal such as, "looking for a soft drink with no sugar," "looking for information about washing machine;," "looking for milk with less than 1% milkfat," or "looking for a lightweight running shoe." Some of these goals may pertain to levels of product attributes. It is hypothesized that, given an attribute goal, the consumer has three options. The consumer may search either externally (in the environment) or internally (in memory) for attribute information, or may infer the value of one attribute from the value of another attribute.

There are two major influences on the selective aspects of attention: current goals and surprising, novel stimuli (Bettman 1979). If an attribute is weighted highly, the interrupt threshold may be set high; and in processing the goal, an inference might be made. Thus, a highly weighted attribute may allow for attribute inferences because a high interrupt threshold is maintained during the processing of the attribute goal.

How high must the attribute be weighted? High enough so that the attribute goal may be processed despite competing stimuli in the environment. Does processing of the attribute goal always lead to inference? Perhaps not; an external or internal search may take place instead of inference.

When environmental stimuli focus attention on an attribute, the goal hierarchy may change. Some goals may be dropped and a new attribute goal may be added which could lead to an attribute inference.

Huber and McCann's (1982) procedure included stimuli which focused attention on the reference attribute and the inferred attribute. In both of the conditions investigated, the subjects were asked to take into consideration what they knew about the two attributes (e.g., "price and taste"). In the "inference prompted" condition, subjects were asked to fill in their best guess of the missing attribute values. These instructions might be seen as meeting Huber and McCann's (1982) third requirement for inferences, an appropriate context; but an alternative explanation is possible. The instructions may have 1 e d t o the formation of an attribute goal. Under this explanation, processing of the goal was completed because distracting stimuli were not present. External and internal searches could not be successful, so inferences were made regarding the levels of the attributes.

The environment may focus the consumer's attention on an attribute in various ways: advertising, publicity, sales promotion, a salesperson's or a friend's comment, packaging, or by some other stimulus. One pattern may particularly increase the likelihood of an attribute goal being formed and an attribute inference being made. It is hypothesized that if salient information is frequently presented on an attribute but if in one particular instance this information is missing, then inferences are likely to be made. For example, if three out of five brands provide information about an attribute on the front of their packages and the other two brands do not provide any information with respect to the attribute, the consumer may infer the missing values. Consumers would be less likely to make inferences if the information was never provided or if it was provided in small print in an unobtrusive fashion. If the attribute information is always salient, inferences are also less probable.

Yates, Jagacinski, and Faber (1978) provide related research evidence in support of this hypothesis. In this study thirty-five students rated the satisfactoriness of two sets of 125 profiles of hypothetical university courses. One set of profiles was complete, containing specified levels of each course on each of four attribute dimensions. In the second set, 50 profiles reported the level of one of the four attributes to be "unknown." The other 75 profiles were complete, as in the first set.

Regression equations were formed for each of the two sets of 7; complete profiles to determine the relative importance of the four attributes in predicting satisfactoriness ratings. Regression weights for the dimension that was omitted in partial profiles tended to be larger in the set in which complete profiles were accompanied by partial profiles.

Students probably attended to the attribute more when information on it was sometimes omitted.

But whether attention increased or not, the "partial information" attributes were given more weight. Thus, providing salient information sometimes and leaving out information in other instances might be one way in which an attribute can come to be weighted highly and inferences may result.

In addition to information which may be provided, the consumer's own experience with the product may lead to the focusing of attention on particular attributes and weighting them highly. The consumer may have had experience with brands which have varied appreciably with respect to the nonaccessed attribute. And this experience may lead to the perception of important differences among brands with respect to this attribute.

For instance, many consumers may view cars as quite similar with regard to safety. Most drivers have not experienced an accident where serious injury has occurred, and an even smaller number have seen differences in the safety of cars responsible for differences in injuries. However, a driver who has experienced a collision between a large, domestic, relatively safe car and a small, foreign, relatively unsafe car is likely to attend to auto safety and weight this attribute highly.This consumer is likely to infer safety from size or whether the car is domestic or foreign.

This hypothesis is consistent with the speculations of Rabinowitz (1971) regarding the inference of quality from price. He surmises that such inferences are most likely to occur when the consumer perceives the differences in quality among alternative brands to be great. These perceptions of high variance among brands may be due to the varied experience of the consumer with the product.

The consumer's experience with brands that vary considerably may lead to inferences about the causes of outcomes. Bettman (1979) hypothesizes that as outcomes are interpreted, consumers actively form associations among attributes. An association may be formed between auto size and auto safety, for example. These processes could lead to cognitive restructuring involving the development of new chunks. For instance, domestic, large, and safe may all be stored as one chunk of auto information.

Goals are more likely to be formed regarding attributes which are particularly relevant to the benefits the consumer is seeking. Russo and Johnson (1979) hypothesize that consumers progress toward a buying decision by making increasingly higher level inferences. They begin with basic information, usually available from the environment. No inferences are required to process this information initially. The highest level of knowledge is the problem's solution, what to buy. It can only be inferred from information at the lower levels.

Thus, it is hypothesized that consumers will be more likely to make inferences when the accessible attributes are relatively low level attributes. For these purposes, the "level" is defined as the degree to which an attribute is associated with the relevant customer needs. Low level attributes may be tangible, and they may be easy to measure, but they are of less value to the consumer in the decision-making process. Higher level attributes are more directly related to the benefits the consumer is seeking.

An industrial buyer, for instance, may have information regarding the thickness of the steel used in various brands of large earth-moving equipment. However, this buyer would like to base the purchase decision, in part, on durability. E no ostensible information on the levels of this attribute is available, the buyer may infer values from the thickness of the steel.

Recall that the primary hypothesis of this section is that the inferred attribute is highly weighted. Just the opposite is the case for accessible attributes. Inferences are hypothesized to occur when accessible attributes are low level attribute; that is, when the accessible attributes have low weights.

VALUES OF THE ATTRIBUTE PERCEIVED TO BE VALID ARE NOT EASILY ACCESSED

The third major condition is that, in order for attribute inferences to occur, the consumer cannot easily access what s/he would perceive as valid values of the inferred nonaccessed attribute. The ease of obtaining information on the nonaccessed attribute is relative to the ease of obtaining information on the accessed attribute This level of difficulty must be considered in light of the motivation of the consumer. To "access" is to obtain the information by either external search, from the environment, or internal search, from memory. Accessing the attribute values is of little worth unless the consumer perceives these values to be valid. If the consumer does not have confidence in the information, s/he may infer values from other attributes s/he has more confidence in.

This hypothesis encompasses both individual differences, such as memory, as well as environmental differences, like the availability of information in the environment. Consider the environmental availability of information first.

The type of information sought in external search may be influenced by the nature of the environment; some types of information may be relatively more accessible than others. Quality may be inferred from price because price levels are relatively more accessible than quality levels. Rabinowitz (1971) conjectures that quality is inferred from price when it cannot be easily judged independently.

Similarly, Shapiro (1971) conjectures that quality is inferred from price, in part, because price is more easily measured. Price in most retail outlets is fixed, not subject to bargaining. Price is concrete and may, therefore, be viewed with a great deal of confidence. Most cues directly associated with quality are not so easily accessible or are not trusted.

These assertions may be generalized beyond the price-quality relationship to other pairs of attributes. Processing of attribute information is influenced by the form in which the information is presented. Information which is easy to process will tend to be used, while that which is difficult will be ignored (Russo, Krieser, and Miyashita 1975; Russo 1977).

Accessibility of information is particularly important when the environment presents very few cues. When consumers had only letters of the alphabet to distinguish quality among beers, a significant letter effect was found (Jacoby, Olson, and Haddock 1971). Similarly, subjects selected loaves of bread by the letters i, L, IN, and P. And despite the fact that the bread was identical, fifty percent of the subjects became loyal to one 'brand" or letter by the end of the 12-trial experiment.

Apparently people use stimuli systematically even when they are not actually associated with the qualities of interest (Dudycha and Naylor 1966). If a decision maker is placed in a situation were none of the available cues are of any value, then that decision maker may tend to pick out and use some of the cues as if they did have value (Blun and Naylor 1968).

A dearth of relevant information may exist in the store, as well as in experimental settings. A small number of attributes may be easily accessible in purchasing frozen fish sticks, for instance. The customer may have no information about the relative tastiness of the various brands z-d no information about their texture, their color, or the amount of chemical substances present. One of the few ostensible attributes may be price. And on the basis of price, the consumer may infer tastiness, or some other attribute.

Accessibility of information is also an important issue when the environment offers data on a large number of attributes. As the total amount of information increases, there first are increases in search activity; however, searching eventually decreases as too high an information load is experienced (Schroder, Driver, and Streufert 1967; Streufert, Suedfeld, and Driver 1965; Sieber and Lanzetta 1964). Excluding internal and external searches from consideration, the only method of determining attribute values is by inference. Thus, inferences may be more likely as the number of potentially accessible attributes increases

In addition to the research on searching, a considerable body of literature (e.g., Summers, Taliaferro, and Fletcher 1970) indicates that people use 2 small number of attributes to make judgments about objects, even though information may be presented on many attributes. This is evident from the regression equations which represent the judgments made. These equations usually have nontrivial weights for only a small number of the attributes available. These data 2_ c at odds with the judgment process subjects verbally report. Usually, subjects say their judgments were affected by more dimensions than the regression weights indicate.

One interpretation of the discrepancy between the verbal reports and the regression models is that subjects attend to the presented data values of only a few attributes and infer values for some of the remaining attributes. For example, car buyers may have a vast amount of data available to them. When questioned, a buyer may respond that he considered both gas mileage and acceleration (in addition to other attributes) before deciding which car to buy. In fact, the buyer may have only attended to the printed mileage values and made inferences about the levels of acceleration.

Thus a curvilinear relationship is hypothesized to exist between the quantity of attributes available and the likelihood of inferences being made. Inferences are more probable when the number of accessible attributes is high or low. A moderate amount of available attribute information is not likely to lead to inferences. This is in line with Berlyne's (1960) findings that whenever an evaluative variable, such as attribute inference, is plotted against a structural variable, such as amount of information, the result is a U-shaped function.

The availability of information on a large number of attributes is closely related to how easily the information itself is to process. Both influence the difficulty of the searching task (Bettman 1979). And consequently, both probably influence the likelihood of attribute inferences.

When consumers actively process incoming attribute information they generate cognitive responses such as: counterarguments, source derogations, and support arguments. These cognitive responses eventually lead to the acceptance or rejection of the message.

Even though an attribute value may be provided by an advertisement or by a salesperson, such information may be rejected. These sources could lack credibility with the consumer, and the potential customer may choose to infer the levels of these attributes from information on other attributes provided by sources valued more highly, such 25 friends, magazines, or news reports. The extern?l or internal availability of attribute information does not appreciably diminish the likelihood of the attribute values being inferred if the consumer rejects the information directly accessible

ELABORATION OF THE MODEL

Thus three major conditions must be met for an attribute inference to occur; these are represented in Figure 1. First, the accessed and nonaccessed attribute must be strongly linked in memory. Operationally, there must be a sizable correlation between the ostensible and the inferred attribute. Second, a nonaccessed attribute goal must be formed and a high interrupt threshold maintained. This is roughly equivalent to a high attribute weighting. Third, the consumer cannot easily access what are perceived as valid values of the attribute. Given these three conditions, an attribute inference is likely to occur.

Figure 1 provides a useful theoretical framework for thinking about attribute inferences. However, when the theoretical basis of the model is expanded, it becomes clear that there are additional variables that may warrant consideration.

Figure 2 presents a causal model which represents some of the empirical relationships which might be expected. One of the relationships is curvilinear with a moderate amount of information giving rise to attribute inference. Four relationships represented in Figure 2 are hypothesized to be positive and linear with high levels of product experience, chunking in memory, nonaccessed attribute importance weights, and competing promotional messages leading to the formation of inferential beliefs. The remaining relationship represented in Figure 2 is negative and linear; low accessed attribute importance weights are associated with attribute inference.

The observable conditions thought to influence inference formation are represented in Figure 2. Each of these six observable variables is related to one of three theoretical conditions. For example, the first observable condition relates to the first theoretical condition. The correlation between attributes reflects the theoretical linkage of the attributes in memory. The second through the fifth observable variables all correspond to the second theoretic-al condition. The importance of both the accessed and the nonaccessed attribute, the variety of the consumer's experience with the nonaccessed attribute, and the promotional messages regarding the nonaccessed attribute all may affect the formation and implementation of an attribute goal. Finally, the sixth observable condition, amount of information in the environment, affects the accessibility of what the consumer perceives as valid values of the attribute.

Each path in Figure 2, then, represents a hypothesis. The next step in the research process is to test these hypotheses empirically. Pretests are necessary to answer such questions as: what i low amount of information? what level o' attribute importance is too high? how is chunking in memory to be assessed? Rough measures for some of these constructs have already been alluded to here. For example, linked or chunked attributes can be identified through correlation coefficients or through examination of multidimensional scaling output. And, there is a tradition in the marketing literature (e.g., Jacoby et al. 1977) and in the aesthetic psychology literature (e.g., Berlyne 1960 1968) for measuring the optimal number of attributes and the complexity of environmental stimuli. Nonetheless, measurement of these variables is not a trivial issue.

SUMMARY

An important promotional decision is what information to include in a promotional message. To some extent all messages give incomplete information but what is omitted must be decided carefully. At present, little is known about consumers ' responses, which may, be determined by inferences prompted by their omission. These inferences will be a function of prior information in memory and the characteristics of the missing external information (Biehal 1983).

The foregoing represents an attempt to develop a model to explain when consumers infer a brand's performance on a missing attribute. Three theoretical conditions are postulated to lead to inferences. Six observable variables are thought to relate to the likelihood of inferences. It is hoped that consideration of the issues raised here will be of help in formulating and conducting additional research related to attribute inference.

FIGURE 1

ATTRIBUTE INFERENCE MODEL

FIGURE 2

CAUSAL MODEL OF ATTRIBUTE INFERENCES

REFERENCES

Berlyne, D. E. (1960), Conflict Arousal and Curiosity, New York: McGraw-Hill.

Berlyne, D. E. (1968), "The Motivational Significance of Collative Variables and Conflict," in Abelson et al. (eds.): 257-266.

Bettman, J. R. (1979), An Information Processing Theory of Consumer Choice, London: Addison-Wesley Publishing Company.

Biehal, G. (1983), "Some Implications of Information Processing for Consumer Behavior Research," Journal of Consumer Affairs.

Blum, M.L. and Naylor, J.C. (1968), Industrial Psychology, New York: Harper and Row.

Collins, A. M. and Loftus, E. F. (1975), "A Spreading-Activation Theory of Semantic Processing," Psychological Review, 83, 407-428.

Dudycha. L. and Naylor, J. C. (1966), "Characteristics of the Human Inference Process in Complex Choice Behavior Situations," Organizational Behavior and Human Performance, 1, 110-198.

Huber, J. and McCann, J. (1982), "The Impact of Inferential Beliefs on Product Evaluations," Journal of Marketing Research, 19, 324-333.

Jacoby, J., Olson, J.C., and Haddock, R.A. (1971), "Price, Brand Name, and Product Composition Characteristics as Determinants of Perceived Quality," Journal of Applied Psychology, 55 (6), 570-579.

Jacoby, J., Szybillo, G. J., and Busato-Schach, J. (1977), "Information Acquisition Behavior in Brand Choice Situations," Journal of Consumer Research, 3, 209-216.

McConnell, J. D. (1968), "Effect of Pricing on Perception of Product Quality," Journal of Applied Psychology, 52 (4), 331-334.

Nakanishi, M.. (1974), "Decision Net Models and Human Information Processing," in Buyer/Consumer Information Processing, G. D. Hughes and M. L. Ray, (eds.), Chapel Hill, N.C.: University of North Carolina Press, 75-88.

Peterson, R. A. (1970), "The Price-Perceived Quality Relationship: Experimental Evidence," Journal of Marketing Research, 7 (4), 525-528.

Rabinowitz, P. (1971), Formulating a Price Policy for Pot o' Gold Products. Master's Thesis, Unpublished. University of the Witwatersrand.

Russo, J. E. (1977), "The Value of Unit Price Information," Journal of Marketing Research, 14, 193-201.

Russo, J. E. and Johnson E. J. (1979), "What Do Consumers Know about Familiar Products?" Advances in Consumer Research. 7 417-423.

Russo, J. E., Krieser, G., and Miyashita, S. (1975), "An Effective Display of Unit Price Information," Journal of Marketing, 39, 11-19.

Shroder, J. .I., Driver, M. J. and Streufert, S. (1967), Human Information Processing, New York: Holt, Rinehart and Winston.

Shapiro, B. P. (1968), "The Psychology of Pricing," Harvard Business Review, 46, 14-25. 160.

Sieber, J. E. and Lanzetta, J. T. (1964), "Conflict and Conceptual Structure as Determinants of Decision-Making Behavior," Journal of Personality, 32, 622-641.

Streufert, S., Suedfeld, P., and Driver, M. J. (1965), "Conceptual Structure, Information Search and Information Utilization," Journal of Personality and Social Psychology, 2, 736-740.

Summers, D. A., Taliaferro, J. D., and Fletcher, D. J. (1970), "Subjective vs. Objective Description of Judgment Policy," Psychonomic Science, 18, 249-250.

Yates, J. F., Jagacinski, C. M., and Faber, M. D. (1978), "Evaluation of Partially Described Multiattribute Options," Organizational Behavior and Human Performance, 21, 240-251.

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