The Role of Knowledge on Problem Solving and Consumer Choice Behavior

Michelene T. H. Chi, University of Pittsburgh
ABSTRACT - This paper outlines major developments in cognitive psychology during the last two decades in the study of the role of knowledge in problem solving. Parallel developments in consumer research will be pointed out when appropriate. Suggestions about how one might study consumer choice decisions will also be provided.
[ to cite ]:
Michelene T. H. Chi (1983) ,"The Role of Knowledge on Problem Solving and Consumer Choice Behavior", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 569-571.

Advances in Consumer Research Volume 10, 1983      Pages 569-571


Michelene T. H. Chi, University of Pittsburgh


This paper outlines major developments in cognitive psychology during the last two decades in the study of the role of knowledge in problem solving. Parallel developments in consumer research will be pointed out when appropriate. Suggestions about how one might study consumer choice decisions will also be provided.

In Search of Heuristics

In the sixties, particularly with the advent of computer science and the notion of computer simulation put forth by Newell, Shaw, and Simon (1958), research in problem solving concentrated almost entirely on the discovery of powerful heuristics that can minimize the problem space that has to be searched to arrive at a solution. In order to make this problem a manageable one to study, research in the late sixties and early seventies concentrated on puzzle-type (or MOVE) problems that have few operators and a finite number of states. Solution consists of searching for a path (the application of a sequence of operators) that can change the problem from its initial (unresolved) state to the final (or solved) goal state. One example of a MOVE problem is the missionaries and cannibals across a river under the constraint that cannibals can never outnumber missionaries in the boat or on either side of the river.

Thus, in some ways, a MOVE problem is not unlike the focus in consumer research on externally provided information of brand-attributes and discovering how consumers faced with such information make a decision (Johnson & Russo 1978).

One of the main findings of research on solving MOVE problems is that human subjects do appear to resort to the use of heuristics such as means/ends analysis (for missionaries and cannibals), or the formation of subgoals, or working backward.

Although research on MOVE problems flourished well into the seventies (see Atwood & Polson 1976, for the water jug problem; Greeno 1974, for hobbits and orcs problem; and Simon 1975, for the Tower of Hanoi problem), the limitation of their findings soon became obvious. Two major events, occurring at about the same time, question the utility of pursuing the investigation of powerful heuristics. First, in cognitive psychology, the pioneering research of de Groot (1966) and subsequent replication and elaboration by Chase and Simon (1973) showed that what differentiated a knowledgeable (or expert) person's problem solution from a novice's was not the power of the heuristic employed by the experts. In the chess research, the expert players do not seem to search any deeper than the novice players. The experts are simply better at pursuing the "better" path in their search for a good move. Similarly, in medical diagnosis research, both experts and novices tend to use the same kind of generate-and-test heuristic. The specialists do not entertain hypotheses any differently from the medical students. They simply entertain a more accurate hypothesis (Elstein, Shulman, & Sprafka 1978). Likewise, expert and novice bridge players don't plan their plays any differently (Charness 1979). Hence, what differentiates an expert from a novice is not the use of a more powerful heuristic, but simply that they pursue the better path to solution.

The second event that minimizes the role of powerful heuristics in problem solving occurs in artificial intelligence (A.I.) research. It became apparent that as the search space becomes unmanageably large (as in the case of chess or other knowledge-rich domains), the heuristics uncovered in the MOVE problems were not sufficiently powerful to reduce the search space. For example, it is not feasible to use means/ends analysis for finding a good move in chess. This forced A.I. researchers to turn to psychological research to obtain clues about what are the skills that experts' have at minimizing search. Basically, the cognitive research up to this point showed not only that experts and novices use heuristics in pretty much the same way, but more in the visual display or in the information presented. For example, chess experts can "see" bigger patterns and more reflects an internal organization in semantic memory that must be explored.

The Structure of Content Knowledge

The uncovering of the limitation of an expert's search heuristic sets the stage for the next batch of studies, which looked at the organization of the knowledge base. The Chase and Simon (1973) research on chess uncovered the size of experts' and novices' chunks. Rietman (1976) expanded this chunk notion to include embedding or overlapping chunks. In bridge, experts tend to chunk hands by suit whereas novices tend to chunk hands by the high cards (Charness 1979). For programming control words, experts' recall reveal a hierarchical tree structure whereas novices' organization of the control words reveal a linear organization. For scrambled lines of program codes, experts' recall will reorder the lines back into their original programs, whereas novices tend to group lines together that have the same syntactic structures (Adelson 1981). And in physics, experts organize the problems according to the underlying physics principles that lead to their solution whereas novices organize them according to the physical entities in the problem (Chi, Feltovich, & Glaser 1981).

The emerging theme of this research is basically that experts organize their knowledge on the basis of the meaning (whether that is the functional relation among the components, such as the attack-defense relations in chess, or the principles underlying the problems, as in the case of physics, or organizing the programming control words and lines of programming codes by their functions rather than their syntax), whereas novices organize their knowledge according to superficial surface features of the presented information (such as the syntax of the program codes, and the entities in the physics problems). In sum, there are basically three conclusions drawn from this research on the structure of experts' and novices' content knowledge. First, experts are able to "see" different things in the presented stimuli than the novices. Second, that the experts "see" tends to be related to the meaning or the functional relations (the "deep level" analysis) whereas novices tend to retain "surface level" information. And third, the experts' superior knowledge structure presumably allows them to manifest better performance on memory tasks.

Interaction of Knowledge Base and Processing

Given that it is clear that experts "see" more than the novices, and they also have more integrated and meaning based structures. it is still not clear how this might facilitate their problem solving processes. Before tackling that question, let us first consider whether there is evidence for the interaction between having the appropriate content knowledge and processing. In the next section, we will consider how the experts' knowledge structure might actually effect their solution processes.

That processing information is somehow effected by the knowledge base is quite evident in a number of studies. In the classical chess findings, again, it is clear that experts can recall a greater number of chess pieces only when the chessboard is a meaningful one; that is, taken from a regular game. When the pieces are randomized, experts' recall are not any better than novices' recall. In logical reasoning, it has been found that people in general have difficulties making logical deductions. But if it is presented in a meaningful context, they will be able to do so quite easily. For example, the "if p then q" type of logical inferences often cause problems for people, but if it is presented in the context of "every time I go to Manchester, I travel by train," then people will not have difficulty making that type of deduction (Wason & Johnson-Laird 1972). More recently, Chase and Ericsson (1982) have trained a single subject to extend his digit span from 7 to 80 digits, by building hierarchical retrieval structures. However, this retrieval structure is only useful for retrieving digits because the subject has a large store of meaningful digit strings in the form of running times. Thus, to increase his digit span, this subject recodes the digits into running times, and at the same time, builds up a hierarchical retrieval structure for encoding and unpacking the digits. Hence, without the familiar patterns of running times, such a retrieval structure would not have been facilitating for either processing the information or for retrieving the information. Finally, in consumer research, Johnson and Russo (1981) also suggested that experienced consumers develop more efficient decision procedures. For example, they would know to ignore the attribute of cruising range in purchasing a car, since this information can be derived from other information known about the car.

In general, it seems that experts, in their domain of expertise, perform remarkably well, in all kinds of tasks. And we know that this is not because they use sophisticated heuristics. But how, exactly, does their greater knowledge base facilitate their problem solving, is still unclear. For example, even though chess experts are able to recall a greater number of chess pieces, thus reflecting larger chunk structures in memory, how does having these large chunks facilitate the selection or a good move? The next section considers two theoretical explanations.

How Structured Knowledge Facilitates Processing

There are two theoretical formats that can be used to postulate how structured knowledge facilitates the use of a particular processing sequence. If we adopt the formalism of a production system (Newell & Simon 1979), then one way to think about this is that the patterns that one "sees" in the problem (such as the patterns on the chessboard, or the cues in the problem statement) match the "conditions" of the productions. Productions are condition-action pairings. The conditions of a production must be met before the actions associated with the conditions can be triggered. Thus, when the appropriate patterns in the environment are seen, they match the conditions of certain production, which then triggers the appropriate actions. These actions need not immediately be the solution procedures. An important point to note is that they may often be the intermediate knowledge states that can lead to the subsequent matching of other conditions which, in turn, can lead to triggering of actions that are more directly related to the solutions. Nevertheless, such a view would be consistent with the observation that experts seem to consider the "best" solution quite "automatically," without much search. For example, the chess expert only considers the best path for a good move, and the medical specialist only entertains the most plausible hypothesis for a diagnosis.

Such an analysis allows for several predictions of how novices may be deficient in finding a solution. Novices may lack certain productions altogether. Or more likely, they may not possess productions with as well-specified conditions as the experts. Or alternatively, they may lack the intermediate productions. That is, their conditions specify explicit final actions, as in the case of novice physics problem solvers. Novice solvers generate explicit equations as a direct response to the surface entities in the problem statements (Chi, Feltovich, & Glaser 1981).

There has not been extensive research which attempts to show the existence of direct links between the cues in the environment and the associated actions. The few studies to be cited below do suggest that this may be true. In chess, for example, the experts can recall a sequence of moves better than the novices, suggesting that the sequence of moves is stored as a meaningful chain of actions (Chase & Simon 1973). In baseball research, knowledgeable individuals are more likely than less knowledgeable individuals to generate appropriate actions for a given baseball game state that are oriented toward the goals of the game (Chiesi, Spilich, & Voss 1979). More recently, Chase (in press) has attempted to show the existence of this direct linking between external environment and explicit procedures in cab drivers. When expert and novice cab drivers are asked in the laboratory to generate a route from one destination to another, experts would sometimes fail to mention a particular street or turn that has to be made. But when the described routes are traveled in the field, the experts almost never fail to make the turn on the appropriate street, even if it was not mentioned in the laboratory. This is particularly -true if they recognize a shorter route in the field which they did not think of in the laboratory. This is interpreted by Chase to mean that the cues in the environment can trigger direct actions when the conditions are met.

The important point to note in this framework is that it is not the case that novices may lack certain procedures (or action, such as solving certain equations in physics) nor is it that they are unable to identify the relevant cues (such as identifying the key and important words in a physics problem statement); but rather, it is the linking of the appropriate conditions with the appropriate actions that the novices have to build up in the knowledge base.

An alternative but compatible theoretical view is to think of the knowledge structure as organized into "packets of units" which have been called schemata (Rumelhart 1981). A schema is simply an organized body or knowledge which incorporates information about how this knowledge can be used. Hence, a schema is a representation of the generic concepts in memory. It has variables which can be seen as attributes describing the concepts. The variables can be filled with prototypical information if none is provided in the environment. Another way to think of the variables is that they are the knowledge associated with a particular concept. It is the activation of this associated knowledge that allows inferences to be drawn and additional knowledge to be generated, thus facilitating information encoding such as comprehension. Also, the activation of a particular schema will guide processing because the schema will seek information that can fill its variables. Thus, by the nature of a schema, different procedures are activated, which is analogous to the notion of conditions triggering assoCiated actions in a productions system.

Applications to Consumer Choice

It seems logical to propose that consumers' decisions are based on prior knowledge about the product, and selection is based on whichever brand best fits one's knowledge of the product. If this view is correct, it would suggest that consumer research should be oriented towards the assessment of the knowledge structure that a particular consumer or a prototypical consumer has, and see whether this knowledge structure can predict his/her selection strategies. For example, a schema framework can be used to assess the knowledge structures of the consumers in the Johnson and Russo (1981) study. We might find, for example, that the experts' schema for a subcompact car is very elaborate, with numerous variables (or attributes). Some variables are deemed more important than others, and are thus associated with greater strength to the produce. Hence, the experts' schema may possess several salient variables which dominate their schema. If this is the case, then it would explain why experts under a judgment instruction task, would recall more statements about the presented brand-attribute display matrix than novices and intermediates. They simply have more variables associated with their automobile schema, and these can be instantiated by the attributes presented in the information display, allowing for better recall. However, their elaborated schema may also explain why the experts' recall is poorer than the intermediates', under the instruct: on of a choice task. One interpretation is that the experts' schema is so well-specified, that the attribute values presented were not compatible with the salient variables in their own schema, because it is these salient ones that would dominate their choice decision. This interpretation is also consistent with the proportion of statements that experts make which referred to their preferred choice. The fact that the experts, more frequently than others, refer to their preferred choice in their protocols, suggests that the preferred model is the one that is the most compatible with their existing schema of the product.

In sum, both consumer researchers and advertisers can probably profit in understanding choice behavior by investing efforts in assessing the prototypical schema of the product. Consumer's selection of a product probably depends on the brand that has the best match between the attributes of the product and the variables in the existing schema.


Adelson, B. (1981) "Problem solving and the development of abstract categories in programming languages," Memory and Cognition, 9, 422-433.

Atwood, M. E. and Polson, P. G. (1976) "A process model for water jug problems," Cognitive Psychology, 8, 191-216.

Charness, N. (1979) "Components of skill in bridge," Canadian Journal of Psychology, 33, 1-50.

Chase, W. G. (in press) "Spatial representations of taxi drivers," In D. R. Rogers and J. A. Sloboda (Eds.), Acquisition of symbolic skills, New York: Plenum Press.

Chase, W. G. and Ericsson, R. A. (1982) "Skilled memory," In J. R. Anderson (Ed.), Cognitive skills and their acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Chase, W. G. and Simon, H. (1973) "The mind's eye in chess," In W. Chase (Ed.), Visual information processing, New York Academic Press.

Chi, M. T. H., Feltovich, P. J. and Glaser, R. (1981) "Categorization and representation of physics problems by experts and novices," Cognitive Science, 5, 121-152.

Chiese, H., Spilich, G. J. and Voss, J. F. (1979) "Acquisition of domain-related information in relation to high and low domain knowledge," Journal of Verbal Learning and Verbal Behavior, 18, 257-273.

de Groot, A. (1966) "Perception and memory vs. thought: Some old ideas and recent findings," In B. Keinnuntz (Ed.), Problem solving, New York: Wiley.

Elstein, A. S., Shulman, L. S. and Sprafka, S. A. (1978) Medical problem solving, Cambridge: Harvard University Press.

Greeno, J. G. (1974) "Hobbits and orcs: Acquisition of a sequential concept," Cognitive Psychology, 6, 270-292.

Johnson, E. J. and Russo, J. E. (1978) "The organization of product information in memory identified by recall times," In H. Keith Hunt (Ed.), Advances in Consumer Research, Vol. V, Ann Arbor: Association for Consumer Research.

Newell, A., Shaw, J. C. and Simon, H. A. (1958) "Chess-playing programs and the problem of complexity," IBM Journal of Research and Development, 2, 230-335.

Reed, S. K. and Simon, H. A. (1976) "Modeling strategy shifts in a problem solving task," Cognitive Psychology, 8. 86-97.

Reitman, J. (1976) "Skilled perception in Go: Deducing memory structures from inter-response times," Cognitive Psychology, *, 336-356.

Rumelhart, D. E. (1981)'Schemata: The building blocks of cognition," In R. Spiro, B. Bruce and W. Brewer (Eds.), Theoretical issues in reading comprehension, Hillsdale, NJ: Lawrence Erlbaum Associates.

Simon, H. A. (1975) "The functional equivalence of problem solving skills," Cognitive Psychology, 7, 268-288.

Wason, P. C. and Johnson-Laird, P. N. (1972) Psychology of reasoning: Structure and content, Cambridge: Harvard University Press.