Knowledge Structures, Production Systems and Decision Strategies

Merrie Brucks, Carnegie-Mellon University
Andrew Mitchell, Carnegie-Mellon University
ABSTRACT - We suggest that in examining consumer decision making, more effort should be directed at understanding why consumers use a particular processing strategy instead of simply describing the resulting process. In addition, we suggest that more effort should be directed at understanding consumer decision making in "real world" situations, A conceptual model is presented which identifies the critical elements that affect the selection of a particular decision process. These elements include the task environment, problem perception, general goals and values, knowledge structures and production systems. Finally, knowledge structures and production systems are discussed in greater detail.
[ to cite ]:
Merrie Brucks and Andrew Mitchell (1981) ,"Knowledge Structures, Production Systems and Decision Strategies", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 750-757.

Advances in Consumer Research Volume 8, 1981      Pages 750-757

KNOWLEDGE STRUCTURES, PRODUCTION SYSTEMS AND DECISION STRATEGIES

Merrie Brucks, Carnegie-Mellon University

Andrew Mitchell, Carnegie-Mellon University

ABSTRACT -

We suggest that in examining consumer decision making, more effort should be directed at understanding why consumers use a particular processing strategy instead of simply describing the resulting process. In addition, we suggest that more effort should be directed at understanding consumer decision making in "real world" situations, A conceptual model is presented which identifies the critical elements that affect the selection of a particular decision process. These elements include the task environment, problem perception, general goals and values, knowledge structures and production systems. Finally, knowledge structures and production systems are discussed in greater detail.

INTRODUCTION

Over the last five or six years, considerable research has been directed at understanding the dynamics of consumer decision making. Most of this research has used a process tracing methodology, a brand/attribute information format and a choice task (e.g., Bettman and Zins 1977, Payne 1976, Green, Mitchell and Staelin 1978). Under this research paradigm, subjects are asked to select a particular brand from a set of brands that differ along a prescribed number of attributes where this information is generally presented in either a verbal (e.g., "very good on decay prevention") or a numerical form (e.g., "gets 35 miles per gallon"). To record the order in which subjects examine the information, protocols (e.g., Payne 1976), eye movements (e.g., Russo and Rosen 1976), or information monitoring techniques (e.g., Bettman and Kakkar 1977) are generally used.

This research began, in retrospect somewhat naively, with the belief that consumers use one of the many hypothesized information integration rules (e.g., linear compensatory) when making choices between brands and that the use of a process tracing methodology would reveal which decision rule an individual used. In addition, it was generally assumed that the use of a particular decision rule would be relatively stable either within subjects over choice tasks or between subjects within choice tasks. It was soon realized, however, that a process tracing methodology reveals information search strategies -- not information integration processes. Although a relationship probably exists between search strategies and information integration processes, the relationship will not be perfect since subjects are able to store some information about the brands in long term memory.

Even though information integration processes cannot be directly inferred from search strategies, the results of the studies in this area seem to indicate that consumers do not use one of the hypothesized information integration rules (Mitchell 1978, Batman and Park 1980). Although processing patterns can be identified (e.g., attribute vs. brand processing), these patterns tend to change frequently. These changing patterns have led Bettman to suggest that decision making under these conditions is frequently a constructive or a bottom up process (Bettman and Zins 1978). This means that the subjects are changing their processing strategy as they gather data during the task. The resulting processing strategies are, therefore, to a certain extent, data driven.

This research also indicates that consumers are very flexible in how they process information. Seemingly subtle changes in the information or its format can result in very different processing patterns (e.g., Bettman and Kakkar 1977). Finally, given a particular information set, there is considerable heterogeneity in how individuals process the information (e.g., Bettman and Jacoby 1976). In other words, each consumer seems to have his or her own idiosyncratic way of processing the information.

These results raise a number of interesting and interrelated problems concerning future research directions in this area. For instance, if individuals do not use one of the information integration rules, how do we define their processing strategies? If each individual has his or her own idiosyncratic way of processing information, how can we find generalizable principles that explain their behavior?

In this paper, we would like to present a preliminary discussion of these problems and a conceptual model that may prove useful in resolving them. The discussion and model represent our current thinking on these problems and we fully expect both to change as our research in the area evolves.

RESEARCH GOALS AND PROBLEMS

Although there are a number of possible goals for research on consumer decision making, we would like to suggest that the general goal for this research should be the prediction of how a given individual will make a decision in a particular situation and an understanding of why he or she used a particular processing strategy. We would also like to suggest that the general domain of phenomena studied in this area should be expanded beyond the information board paradigm. Since the task used in this paradigm appears to represent only a small percentage of the actual decision making tasks confronted by consumers, we believe that more effort should be directed at studying decision making in more "real world" environments.

In general, we believe that two important shifts need to occur in order to accomplish these goals. First, more emphasis needs to be directed at identifying the elements of the task and the individual that cause the between subject and task differences that we observe. This will require the development of a typology of task characteristics and procedures for measuring the individual level variables that affect the selection of a particular decision strategy. Second, more emphasis needs to be directed at understanding the cognitive activities (e.g., planning) that surround consumer decision making. In the past, we have been primarily concerned with describing how individuals processed or searched for information instead of investigating why they used a particular processing strategy. Both of these shifts require the integration of theories from cognitive psychology into conceptual models that provide an understanding of consumer decision making. We believe these shifts in emphasis will result in a better understanding of consumer decision making and may result in better models for predicting choice.

There are a number of problems, however, that need to be resolved before these goals can be accomplished. The first concerns the level of analysis. Process tracing procedures produce a series of dense observations preceding final choice. Bettman and Park (1980), for instance, have decomposed protocols taken in a choice task using the information board paradigm into a typology of 70 different phrases and found that their subjects used an average of 35 phrases in making a choice. The fundamental problem, then, is how to develop measures of decision strategies that are causally related to elements of the task and the individual. One type of measure might be obtained by aggregating the processes measures. Alternatively, measures of general problem solving strategies such as "mean-ends analysis" and "operator subgoaling" (Newell, 1980) or "divide and conquer" and "define and successively refine" (Aho, Hopcraft and Ullman 1974) might be used.

In the same vein, there are many different conceptual dimensions of the variables that are thought to influence decision strategies. A number of studies, for instance, indicate that knowledge about a particular domain will affect how individuals process information (e.g., Edell and Mitchell 1978, Marks and Olson 1981) and their search strategies (e.g., Bettman and Park 1980, Johnson and Russo 1981). There are, however, many possible conceptual dimensions that one might use to identify differences in a domain of knowledge between two individuals. In addition, there are many possible conceptual dimensions that might be used to differentiate between two different tasks. In summary, the first problem involves the question of how to develop a typology of decision strategies and obtain measures of the task and individual so that causal relationships between the two may be identified.

The second, and related problem, concerns the indication that many decision processes are constructive or bottom up processes. How can we hope to predict how an individual will make a decision when he or she doesn't know how the decision will be made when he or she begins the task? This problem, of course, is related to the first. If, for instance, we use the proper level of aggregation for the dependent variable, we may be able to predict which decision strategy an individual will use in a given task based on specific elements of the task and the individual. This, of course, would be essentially a static analysis of a dynamic process. Alternatively, as discussed later in this paper, we may be able to develop procedures for understanding and predicting a constructive process.

This suggests that there are at least two different approaches for achieving the goals outlined earlier. The first involves the development of a typology of decision strategies which describes the process the individual used in making a decision. The second involves the prediction of the evolving dynamic process when it is constructive. At this point its difficult to assess which approach will prove more useful. Consequently, it would appear that fruitful research may proceed at both levels.

Finally, we would like to note that the consumer choice situation is essentially an ill-structured problem. Simon (1978) suggests that these problems differ from well-structured problems in three important ways. First, the criterion for determining if the goal has been achieved is poorly defined. Second, the problem instructions do not contain all the information required to solve the problem. Finally, the set of possible alternative moves at each stage of problem solving is not well defined. These differences, of course, increase the difficulty of understanding consumer decision making. Since there is no well defined goal to the problem, for instance, we need to understand how consumers decide when they have enough information about the alternatives to make a choice.

In the next section we present a simple conceptual model which we believe will be useful for studying decision strategies using either a static or a dynamic approach. This model is heavily influenced by previous research examining problem solving in well-structured domains (e.g., Greens 1978, Simon 1978). The critical individual level variables of this model are briefly discussed and then examples of how the model may be used to understand decision processes in both information-board situations and more "real world" situations are outlined.

CONCEPTUAL MODEL

A conceptual model of the elements affecting the decision making process is presented in Figure 1. This model is a relatively simple one since little research has been directed at understanding the cognitive processes involved in decision making in ill-structured task environments (Simon 1978). We expect, however, that as research progresses in this area we will have a better understanding of these elements and cognitive processes.

FIGURE 1

CONCEPTUAL MODEL

Within this model, there are three different sets of critical elements that affect the decision making process. The elements of the first set exist in long term memory, are unique to each individual and affect all decision making processes. These are structural variables which include general goals or values, knowledge structures and production systems. The second set define the particular task and third are process variables which evolve during the decision process. These include, problem perception, information integration, and if the task involves the acquisition of external information, encoding processes. These specific elements will now be discussed in greater detail and then, two examples will be used to illustrate the model.

Elements of the Model

General Values and Goals.  An individual's general values and goals are behavioral end states that the individual wants to achieve. These may include states such as professional success, wealth, and attractive physical appearance. These general values or goals will affect an individual's general approach to a purchase decision and the attributes that are used in evaluating the alternative brands. For instance, in purchasing clothes, an individual who is concerned with having an attractive physical appearance will probably be concerned with the style of the cloches while someone who does not highly value this general goal may be primarily concerned with price.

Knowledge Structures and Production Systems.  The remaining two elements of the first set are knowledge structures and production systems. These two elements represent the distinction that is frequently made between declarative knowledge and procedural knowledge (e.g., Anderson 1980). Declarative knowledge is our knowledge about concepts, objects or events. It is generally believed that this information is organized into packets of information or schemata (e.g., Rumelhart and Ortony 1977, Schank 1980). We may, for instance, have an organized set of information about fuel efficient automobiles, in general, and the Volkswagen Rabbit, in particular. We may also have stored organized information about events, such as visiting an automobile dealer to look for a new car, which are called scripts (Schank and Abelson 1977).

Production systems are representations of cognitive skills such as typing letters or solving geometry problems. The basic element of a production system is a condition-action statement. This statement allows the system to take a particular action when s particular condition is recognized. The particular condition may be an external stimulus such as a green traffic light or an internal state such as the lack of knowledge about a particular automobile. Similarly, the action may be a particular behavior such as stepping on the gas pedal of an automobile or a mental act such as adding two numbers together.

Problem Perception.  Problem perception is the individual's internal representation of the problem. Considerable research on problem solving indicates that this internal representation will affect the strategy that an individual uses to solve the problem (e.g., Newell and Simon 1972, Greens 1978, Einhorn and Hogarth 1981). Tversky and Kahneman (1981) and Simon and Hayes (1976) have shown that individuals may use different strategies for solving the same problem when the problem is presented differently (i.e., problem isomorphism).

A number of researchers have used different terms for this general notion. These include problem space (Newell and Simon 1972), decision frame (Tversky and Kahneman 1981), and problem frame (Wright and Rip 1980). We have selected a different term because we believe that our conceptualization differs from the definitions applied to these other terms. We currently view problem perception as containing two factors. The first are the general characteristics of the problem. These characteristics may include, for instance, the amount of time before a decision is required, the specific goals and subgoals for the problem, and the possible consequences of making a poor decision. The second factor, which we call the workspace, is the current state of the system with respect to the decision process. The current state of the system might contain, for instance, the number of automobile dealers visited and the amount of knowledge about the various alternative brands scored in long term memory.

This conceptualization differs from the definition of a decision frame as "the decision maker's conception of the acts, outcomes and contingencies associated with a particular choice" (Tversky and Kahneman 1981), or problem frames which Wright and Rip (1980) view as the information integration rules used to integrate information in making a decision. We find the former to be somewhat limiting in terms of the problems we wish to consider and we view the latter as a production system. Our notion of problem perception is probably closest to the concept of a problem space which was developed from the study of well-structured problem (Newell and Simon 1972). This concept contains two basic factors (a) a conceptual understanding of the problem consisting of the initial state, the goal scale and the path constraints and (b) the different possible states of the space and the operators (Newell 1980). However, since our model is developed for ill-structured as opposed to well-structured problems, elements such as path constraints and operators may not be critical elements of problem perception. Finally, it should be noted that the cognitive limitations of humans imply that the cognitive elements of problem perception must necessary be few in number (five to seven), unless some of this information is transferred to long term memory.

Encoding and Information Integration Processes.  Encoding processes represent the transformation of external data into on internal symbolic code. These processes are influenced by knowledge structures and include both inferences and evaluative processing (i.e., counterarguing). For instance, an individual may infer that a particular automobile gets good gas mileage even though this information was not available to that individual. The information integration process involves the integration of encoded information from either external sources or from memory to form an evaluation or to make a choice.

Decision Making Examples

We will now discuss the conceptual model with respect to two different decision tasks. The first involves the standard information board task, while the second involves a more "real world" task. This latter example uses the task and model proposed by Hayes-Roth and Hayes-Roth (1979).

In the standard information board task, the subject is presented with an array of brands that vary along a number of different attributes. The information presented to the subject in the task (e.g., the product class and the brand names) and the relevant knowledge the individual has stored in long term memory (e.g., knowledge about the product category) result in a cognitive representation of the problem (i.e., problem perception). This cognitive representation for evaluating the alternative products may include the critical attributes that need to be considered. We have found, for instance, differences in how individuals process nutritional information when this information is associated with different foods (Brucks, Mitchell and Staelin 1980). This cognitive representation of the problem then triggers s particular set of productions that are used to start processing the information. In other words, the perception of the problem represents the condition portion of a production and the actual processing of information (e.g., encoding and information integration) represents the action portion. These productions may be at the level of the typology of phrases developed by Bettman and Park (1980) or they may be combinations of these phrases.

After a particular production has been executed the workspace is updated. This updated workspace, along with the subject's original understanding of the problem, then triggers a new production which then processes more of the information. These cycles continue until a choice is made.

What is critical, then, within this framework is understanding how an individual cognitively represents the problem and how the individual updates this representation at different points in time during the decision making process.

In more "real world" situations; the system becomes somewhat more complex; however, the structure remains the same. As mentioned previously, we use the task and the model proposed by Hayes-Ruth and Hayes-Ruth (1979) to demonstrate this. The task involved having subjects develop a plan for completing a number of errands in an afternoon. The subjects were given the errands (e.g., meet a friend for lunch, pick up medicine for dog at the vet) and a map of a city with all the critical locations marked (e.g., veterinarian). The problem was designed so that it was very difficult to complete all the errands in a single afternoon.

The model developed for understanding the cognitive processes that the individual uses in developing the plan contains two basic elements: a number of different "specialists" and a "blackboard". The "specialists" are essentially production systems and their decisions are recorded on the "blackboard". The "blackboard" performs the function of recording the current state of the system with regard to the planning process. It is, therefore, similar to our notion of the workspace in problem perception. These different states trigger "specialists" which make decisions with regard to the planning process which, in turn, change the state of the system. For instance, knowledge that the individual is at a certain location in the town at a particular point in the planning process may cause a particular "specialist" to then select the errand that is closest to the current location. The process continues until a plan is formulated.

This, of course, is s simplified description of the model. In the actual model, the "blackboard" consists of five different planes called the plea, plan-abstractions knowledge-base, executive and meta-plan. These five planes also contained different levels. An artificial intelligence program was developed based on this model, which provided a good representation of different subjects' planning processes.

This description is provided to indicate how our model might be used to understand actual "real world" decision processes. An obvious extension would be to understand how the formulated plan would change if it were actually implemented. An individual, for instance, might use more time completing a particular errand than expected (e.g., long time waiting for the vet) and then would have to revise the plan. The model, however, can also be used to describe the cognitive activity involved in revising the plan. With some adjustments this model might also be used to examine the strategy or plan that an individual uses in purchasing an automobile or a refrigerator. We are currently examining how the Hayes-Roth and Hayes-Roth model might be revised to study this task.

In summary, we are suggesting that in studying consumer decision making, more effort should be directed at understanding the cognitive activities surrounding the decision process. In other words, we should be more concerned with understanding why a consumer used a particular strategy instead of simply recording the actual process. We have presented a relatively simple conceptual model that might be used to provide a better understanding of these processes. In this model, the critical elements are problem perception, generalized goals or values, knowledge structures and production systems. In the next two sections we examine the latter two elements in greater depth.

KNOWLEDGE STRUCTURES

Knowledge structures contain our generalized knowledge about the external environment. Models of this generalized knowledge are generally conceptualized as network models, where the nodes of the network represent concepts and the links between nodes represent the relationship between concepts (e.g., Anderson 1976). As discussed in the previous section, it is currently believed that human knowledge is organized into packets of information called schemata (Rumelhart and Ortony 1977). Individuals, for instance, may have a schema for fuel efficient automobiles that contains their generalized knowledge about these objects. Consequently, when we are presented with information about a fuel efficient automobile we activate this schema to process, encode, evaluate, and interpret the information. Ortony (1980) has recently suggested a dual layered theory of memory. The first layer contains our organized knowledge about specific concepts (e.g., schemata), while the second layer contains the relationships between schemata. Both of these layers are conceptualized as networks.

It is generally believed that our generalized knowledge will affect hay we make decisions (e.g., Mitchell 1978, Olson 1978, Bettman 1979). Only recently, however, have researchers begun to actually examine how knowledge affects decision making (Bettman and Park 1980, Johnson and Russo 1981). Most of these studies, however, use only a single measure of an individual's knowledge of a particular domain. Frequently, this measure is based on the subjects' previous purchase behavior with respect to the product. Bettman and Park, 1980, for instance, use a subject's self reported experience with the product class as a measure of knowledge. Given the complexity of possible knowledge within a domain, multiple measures of knowledge should provide a better understanding of any related cognitive processes than a single measure. Objective measures of knowledge structures would also appear to be better than measures of previous behavior with respect to the product or self report measures. A teenager, for instance, who has never owned an automobile, may be more knowledgeable about different aspects of automobiles than an adult who has owned three automobiles.

Since there are many different possible conceptual dimensions of knowledge structures, we must identify which dimensions will provide the best explanatory power to explain the phenomena of interest. Kanwar, Olson and Sims (1981) have moved in this direction recently by proposing three different measures of cognitive structure --dimensionality, abstraction and articulation. These measures are obtained through the use of free elicitation (Olson and Muderrisoglu 1979) and a modification of the repertory grid (Kelly 1955). The first measure, dimensionality, is the number of concepts activated by elicitation procedures within a particular content domain. The second, abstraction, refers to the level of abstractiveness that the related concepts within a domain are encoded. The third measure, articulation, is the number of levels that the individual uses to discriminate between objects within the domain.

The measures proposed by Kanwar, Olson and Sims (1981) are structural measures of an individual's knowledge within a particular domain. They essentially measure how information is organized within a domain. Alternative measures might be based on the content of knowledge within a particular domain. Although structural measures and content measures may be related, there may he important differences between these two types of measures. Kanwar, Olson and Sims (1981), for instance, found that a single measure of the content of knowledge within a particular domain (i.e., nutrition) was orthogonal to chair structural measures.

Currently, we believe that content measures of a domain may provide a better explanation of the resulting cognitive processes within that domain then structural measures. In other words, we believe that what you know is more important than how moth you know about a particular domain. It seems clear, however, that multiple measures are required to determine the extent of an individuals knowledge within a domain.

Some work on developing typologies of the content of knowledge within a domain has been done by educational psychologists (e.g., Bloom et al. 1956). We have recently developed a similar typology for classifying the knowledge that a consumer may have about a particular product category or purchase decision. The different types of knowledge within this typology are: (1) terminology (2) specific facts (3) relationships (4) criteria for evaluation and (5) procedural information.

Terminology refers to knowledge of terms used within a particular domain. These might include knowledge of the meaning of U.S. RDA percentages within the nutritional domain or the meaning of electronic fuel injection within the automotive domain. Specific facts are knowledge about objects within the domain. Examples of these are "apples do not contain much vitamin C" in the nutritional domain and "a Datsun B-210 gets 35 miles per gallon" in the automotive domain.

Causal relationships are knowledge about how different attributes of the objects within a domain affect the object's performance. Within the nutritional domain, for example, knowledge that if you don't get enough vitamin C you will get scurvy or within the automotive domain, knowledge that small four cylinder automobiles get good gas mileage are examples of this type of knowledge. The criteria for evaluation measure refers to knowledge that is used in evaluating objects within the domain. Examples of this measure might involve "cutoff' values for the different attributes. In the nutritional domain, a criteria for evaluation might be the knowledge that any fruit that has over 20% of the U.S. RDA of Vitamin A is "good" on that dimension or any small four cylinder automobile that gets over 40 miles per gallon is "good" on that dimension.

The final measure is procedures. This is knowledge concerning how the individual should behave with respect to objects within a domain. For instance, knowledge that you should have a green vegetable every day or that you should change the oil in your car every 5,000 miles are examples of procedures for the nutritional and the automotive domains, respectively.

In summary, an individual's knowledge about a particular domain will have a critical effect on how he or she processes related information. In order to understand the relationship between knowledge and processing activities, we need to develop procedures for measuring the different conceptual dimensions of knowledge.

PRODUCTION SYSTEMS

As mentioned previously, production systems are a representation of procedural knowledge or cognitive skill. Production systems have been used to represent cognitive skills in memory scanning (Newell 1973), language comprehension (Anderson, Kline and Lewis 1977) and solving geometry problems (Neves and Anderson 1981).

The basic element of a production system is a condition-action statement. If an individual recognizes a particular condition (e.g., a green traffic light), he or she executes a particular action (e.g., walk across the intersection). A production system contains four basic elements (Rychener and Newell 1978): (1) a production memory, (2) a working memory (3) a recognize-act cycle and (4) conflict resolution principles. The production memory contains the condition-act statements while working memory contains information about the present state of the system. This is the knowledge that triggers the condition element of the condition-act statements. Consequently, working memory is similar to our notion of the workspace in problem perception. Conflict resolution principles are evoked whenever the current state of the system may trigger more than one condition-act statement. Finally, the recognize-act cycle executes a particular condition-act statement which, in turn, changes the state of the system. In general, network models have been used to represent procedural information in memory (e.g., Anderson 1976, Norman, et. al. 1975, Rumelhart and Ortony 1977).

FIGURE 2

RELATIONSHIP BETWEEN THE TIME REQUIRED TO COMPLETE THE TASK AND THE NUMBER OF TIMES ITS BEEN REPEATED

Learning obviously occurs within production systems. In many tasks, a power law type relationship has been found between the number of times a particular task has been executed and the amount of time that it takes to complete the task (Figure 2). These tasks include perceptual motor skills, perception, motor behavior, elementary decisions and problem solving (Newell and Rosenbloom 1981). In a pretest of an experiment, we found the same relationship between the number of times an individual evaluated different brands based on the same prescribed attributes and the amount of time that it takes him or her to make the evaluation.

Anderson (1980) has suggested that three basic stages occur in learning a cognitive skill. The first is a cognitive stage where the individual learns the steps that one must go through in performing the task. For example, in learning to drive a standard shift automobile, the individual learns that to shift gears the clutch must first be depressed, then the shift lever is moved from one position to the other, and finally the clutch must be released. At this stage, the information is stored in declarative form. The second stage is the associative stage. Here, the individual uses the declarative knowledge to execute the different steps required to perform the task. This causes a production system for the different steps to be formed. At this point in the learning process individuals generally need to devote all of their cognitive resources (e.g., attention) to executing the task. As they execute the task a large number of times, less and less attention is required. Finally, at the third stage, the autonomous stage, individuals can execute the task with little or no conscious attention --the task has become automatic. A number of studies have examined automatic processes in both perception and reading (e.g., LaBerge and Samuels 1974, Schneider and Shiffren 1977) and research is currently being directed at understanding the factors affecting the development of automatic processes (Shiffren and Dumais 1981).

Although the stages identified by Anderson (1980) provide insights into the learning of a cognitive skill, they do not provide an explanation of why the amount of time to execute a task seems to follow a power law. Two explanations for this phenomena have been presented. The first, by Newell and Rosenbloom (1981) uses a chunking hypothesis. Under this hypotheses, individuals learn to react to a pattern of stimuli in the environment instead of individual stimuli. For instance, in a task where an individual must give a particular response to a particular sequence of lights, the individual may at first process the status of each light sequentially or serially. With practice, however, the individual may react to the entire sequence of lights. Evidence for this explanation comes from research examining novice and expert chess players. This research, for instance, indicates that expert chess players react to the general pattern of the pieces on the chess board (Chase and Simon 1973). Newell and Rosenbloom (1981) have shown mathematically that this form of chunking will result in power law type relationship between the time required to execute the task and the number of times the task has been repeated.

A second explanation comes from Neves and Anderson (1981). They hypothesize two mechanisms that may yield the power law type relationship. The first is proceduralization which occurs at the early stages in learning a cognitive skill. With this mechanism, declarative information that is required in executing a production in integrated into the production system. This, of course, reduces the need to retrieve the required information in executing the production which, in turn, reduces the amount of time required to execute the task. The second mechanism is composition which may occur at both early and late stages of learning. This mechanism combines two or more productions into a single production. Consequently, instead of a single condition activating a production, multiple conditions activate the production. This is similar, then, to the chunking notion of Newell and Rosenbloom (1981). Neves and Anderson (1981) have used simulations to demonstrate that these two mechanisms can result in the power law type relationship between the number of times a task is executed and the time required to complete the task.

It seems clear that understanding an individual's procedural knowledge is critical in understanding consumer decision making. To date, however, little research has been directed at understanding the development of the cognitive skills involved in consumer decision making or the measurement of these skills. We suspect that more research will be directed in this area in the near future.

CONCLUSION

In this paper we have suggested that the goal of future research on consumer decision making should center on predicting what decision strategy an individual will use in a particular situation. In addition, we have suggested that more research should be directed at understanding consumer decision making in "real world" situations. The achievement of these goals require the identification of the critical elements of the situation and the individual which cause the selection of a particular decision strategy. Once these elements have been identified and measured, they may be used to predict either global measures of the decision strategy or the evolution of the actual decision strategy when it is a constructive process. The former is essentially a static analysis while the latter provides an understanding of the dynamics of decision making.

These approaches require more emphasis on why a particular decision strategy was selected as opposed to simply a description of the resulting strategy. They also require a better understanding of the cognitive activity involved in decision making.

Whichever approach is used, and we believe that both will prove fruitful, effort must be directed at identifying and measuring the critical elements of the task and the individual that will affect the selection of a particular decision strategy. In this paper, we represent a simple conceptual model that identifies these elements. They include the task environment, problem perception, general goals or values, knowledge structures and production systems. If we want to predict the selection of a particular decision strategy, future research effort will need to be directed at obtaining a better conceptualization of these elements and procedures for measuring the critical dimensions of these elements.

REFERENCES

Aho, A. J., Hopcraft, E. and Ullman, J. D. (1974), The Design and Analysis of Computer Algorithms, Reading, Mass.: Addison-Wesley.

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

Anderson, J. R. (1976), "Language, Memory and Thought," Hillsdale, NJ: Lawrence Erlbaum Associates.

Anderson, J. R., Kline, P. J. and Beasley, C. M. Jr. (1980), "Complex Learning Processes," in R. E. Snow, P. A. Federico and W. E. Montague, eds., Aptitude, Learning, and Instruction: A Cognitive Process Analysis, Hillsdale, NJ: Lawrence Erlbaum Associates.

Anderson, J. R., Kline, P. J. and Beasley, C. M. (1979), "A General Learning Theory and its Application to Schema Abstraction," in G. H. Bower, ed., The Psychology of Learning and Motivation, Vol. 13, New York: Academic Press.

Anderson, J. R., Kline, P. J. and Lewis, C. (1977), "A Production System Model for Language Processing," in P. Carpenter and M. Just, eds., Cognitive Processes in Comprehension, Hillsdale, NJ: Lawrence Erlbaum Associates, 271-311.

Bettman, J. R. (1980), An Information Processing Theory of Consumer Choice, Reading, Mass.: Addison-Wesley.

Bettman, J. R. and Jacoby, J. (1976), "Patterns of Processing in Consumer Information Acquisition," in B. B. Anderson, ed., Advances in Consumer Research, Vol. 3, Chicago: Association for Consumer Research, 315-320.

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

Bettman, J. R. and Park, C. W. (1980), "Effect of Prior Knowledge and Experience and Phase of the Choice Process on Consumer Decision Processes: A Protocol Analysis," Journal of Consumer Research, 7, 234-248.

Bertram, J. R. and Park, C. W. (1980), "Implications of a Constructive View of Choice for Analysis of Protocol Data: A Coding scheme for Elements of Choice Processing," in J. C. Olson, ed., Advances in Consumer Research, Vol. 7, Ann Arbor: Association for Consumer Research, 148-153.

Bettman, J. R. and Zins, M. A. (1977), "Constructive Processes in Consumer Choice," Journal of Consumer Research, 4, 75-85.

Bloom, B. S., Englehart, M.D., Furst, E. J., Hill, W. H. and Krathwohl, D. R. (1956), Taxonomy of Educational Objectives: The Classification of Educational Goals, Handbook I: Cognitive Domain, New York: David McKay Company, Inc.

Brucks, M., Mitchell. A. A. and Staelin, R. (1980), "The Disclosure of Nutritional Information: An Information Processing Approach," Working Paper, Graduate School of Industrial Administration, Carnegie-Mellon University, Pittsburgh, PA 15213.

Chase, W. G. and Simon, H. A. (1973), "Perception In Chess," Cognitive Psychology, 4, 55-81.

Edell, J. and Mitchell, A.(1978), "An Information Processing Approach to Cognitive Responses," in S. C, Jain, ed., Research Frontiers in Marketing: Dialogues and Directions, Chicago: American Marketing Association, 178-183.

Einhorn, H. J. and Hogarth, R. M. (1981), "Behavioral Decision Theory: Processes of Judgment and Choice," Annual Review of Psychology, 32.

Green, R., Mitchell, A. A. and Staelin, R. (1977), "Longitudinal Decision Studies Using a Process Approach: Some Results from a Preliminary Experiment," in B. A. Greenberg and D. A. Bellenger, eds., Contemporary Marketing Thought, Chicago. American Marketing Association, 461-466.

Greens, J. G. (1978), "Nature of Problem-Solving Abilities," in W. K. Estes, ed., Handbook of Learning and Cognitive Processes, Vol. 5, Hillsdale, NJ: Lawrence Erlbaum Associates, 239-270.

Hayes-Roth, B. and Hayes-Roth, F. (1979), "A Cognitive Model of Planning," Cognitive Science, 3, 275-310.

Hayes-Roth, F., Waterman, D. A. and Lariat, D. B. (1978), "Principles of Pattern-Directed Inference Systems," in D. A. Waterman and F. Hayes-Roth, eds., Pattern Directed Inference Systems, New York: Academic Press, 577-601.

Johnson, E. and Russo. J. E. (1981), "Product Familiarity and the Learning of New Information," in K. Monroe, ed., Advances in Consumer Research, Vol. 8, Ann Arbor: Association for Consumer Research.

Kanwar, R., Olson, J.C. and Sims, L. S. (1981), "Toward Conceptualizing and Measuring Cognitive Structures," in K. Monroe, ed., Advances in Consumer Research, Vol. 8, Ann Arbor: Association for Consumer Research.

Kelly, G. A. (1955), The Psychology of Personal Constructs, Vols, 1 and 2, New York: Norton.

LaBerge, D. and Samuels, S. J. (1974), "Toward a Theory of Automatic Information Processing in Reading," Cognitive Psychology, 6, 293-323.

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

Mitchell, A. A. (1978), "An Information Processing View of Consumer Behavior," in S. C. Jain, ed., Research Frontiers in Marketing; Dialogues and Directions, Chicago: American Marketing Association, 188-197.

Neves, D. M, and Anderson , J. R. (1981), "Knowledge Compilation: Mechanisms for the Automatization of Cognitive Skills," in J, R. Anderson, ed., Cognitive Skills and Their Acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Newell, A. (1980), "Reasoning, Problem Solving, and Decision Processing: The Problem Space as a Fundamental Category," in R. S. Nickerson, ed., Attention and Performance VIII, Hillsdale, NJ: Lawrence Erlbaum Associates, 693-718.

Newell, A. (1973), "Production Systems: Models of Control Structures," in W. G. Chase, ed., Visual Information Processing, New York: Academic Press, 463-526.

Newell, A. and Rosenbloom, P. W. (1981), "Mechanisms of Skill Acquisition and the Law of Practice," in J. R. Anderson, ed., Cognitive Skills and Their Acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Newell, A. and Simon, H. (1972), Human Problem Solving, Englewood Cliffs, NJ: Prentice Hall.

Norman, D. A., Rumelhart, D. E. and the LNR Research Group (1975), Explorations in Cognition, San Francisco: W. H. Freeman and Company.

Olson, J. C. (1978), "Theories of Information Encoding and Storage: Implications for Research," in Andrew A. Mitchell, ed., The Effect of Information on Consumer and Market Behavior, Chicago: American Marketing Association, 49-60.

Olson, J. C. and Muderrisoglu, A. (1979), "The Stability of Responses Obtained by Free Elicitation: Implications for Measuring Attribute Salience and Memory Structure," in W. Wilkie, ed., Advances for Consumer Research, Vol. 6, Ann Arbor, MI: Association for Consumer Research, 269-275.

Ortony, A. (1978), "Remembering, Understanding, and Representation," Cognitive Science, 2, 53-69.

Palmer, S. E. (1980), "Fundamental Aspects of Cognitive Representation," in E. Rosch and B. B. Lloyd, eds., Cognition and Categorization, Hillsdale, NJ, Lawrence Erlbaum Associates, 262-303.

Payne, J. W. (1976), "Task Complexity and Contingent Processing in Decision Making: An Information Search and Protocol Analysis," Organizational Behavior and Human Performance, 16, 366-387.

Rumelhart, D. E. and Ortony, A.(1978), "The Representation of Knowledge in Memory," in R. C. Anderson, R. J. Spiro and W. E. Montague, eds., Schooling and the Acquisition of Knowledge, Hillsdale, NJ: Lawrence Erlbaum Associates, 37-54.

Russo, J. E. and Rosen, L. D. (1975), "An Eye Fixation Analysis of Multi-Alternative Choice," Memory and Cognition, 3, 267-276.

Rychener, M. D. and Newell, A.(1978), "An Instructable Production System: Basic Design Issues," in D. A. Waterman and F. Hayes-Roth, eds., Pattern-Directed Inference Systems, New York: Academic Press, 135-154.

Shank, R. C. (1980), "Language and Memory," Cognitive Science, 4, 243-284.

Shank, R. C. and Abelson, R. (1977), Scripts, Plans, Goals and Understanding, Hillsdale, NJ: Lawrence Erlbaum Associates.

Schneider, W. and Shiffren, R. M. (1977), "Controlled and Automatic Information Processing: I. Detection, Search and Attention," Psychological Review, 84, 1-66.

Shiffren, R. M. and Dumais, S.T. (1981), "The Development of Automatism," in J. R. Anderson, ed., Cognitive Skills and Their Acquisition, Hillsdale, New Jersey: Lawrence Erlbaum Associates.

Shiffren, R. M. and Schneider, W. (1977), "Controlled and Automatic Human I Information Processing: II Perceptual Learning, Automatic Attending and a General Theory," Psychological Review, 84, 492-512.

Simon, H. A. (1978), "Information Processing Theory of Human Problem Solving," in W. K. Estes, ed., Handbook of Learning and Cognitive Processes, Vol. 5, Hillsdale, New Jersey: Lawrence Erlbaum Associates, 271-296.

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

Tversky, A. and Kahneman, D.(1981), "The Framing of Decisions and the Rationality of Choice," Science.

Wright, P. and Rip, P. D. (1980), "Product Class Advertising Effects on First-Time Buyers' Decision Strategies," Journal of Consumer Research, 7, 176-188.

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