A Computational Process Model of Evaluation Based on the Cognitive Structuring of Episodic Knowledge

ABSTRACT - An overview of the architecture of a computational process model of evaluation and judgment is presented. The model consists of three components: declarative memory, working memory and production systems. Declarative memory contains both episodic and semantic knowledge and a goal structure. The production systems form generalized knowledge from episodic knowledge and define a number of different processes used in forming evaluations and judgments.

Citation:

Terrence R. Smith, Andrew A. Mitchell, and Robert Meyer (1982) ,"A Computational Process Model of Evaluation Based on the Cognitive Structuring of Episodic Knowledge", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 136-143.

Advances in Consumer Research Volume 9, 1982      Pages 136-143

A COMPUTATIONAL PROCESS MODEL OF EVALUATION BASED ON THE COGNITIVE STRUCTURING OF EPISODIC KNOWLEDGE

Terrence R. Smith, University of California, Santa Barbara

Andrew A. Mitchell, Carnegie-Mellon University

Robert Meyer, Carnegie-Mellon University

[This research was partially supported by a grant from the Department of Transportation.]

ABSTRACT -

An overview of the architecture of a computational process model of evaluation and judgment is presented. The model consists of three components: declarative memory, working memory and production systems. Declarative memory contains both episodic and semantic knowledge and a goal structure. The production systems form generalized knowledge from episodic knowledge and define a number of different processes used in forming evaluations and judgments.

INTRODUCTION

In consumer behavior, as in most other behavioral sciences, it is of critical importance to understand how humans make evaluations and judgments. Although a number of different theoretical and methodological approaches have been used to study evaluation and judgment, it is safe to say that we do not possess a clear understanding of how they are formed. We know, for instance, that human judgments are frequently subject to bias (e.g., Kahneman and Tversky, 1973) and context effects (e.g., Tversky and Kahneman, 1981), however, we do not understand the reasons for such bias and context effects.

In order to understand how humans make evaluations and judgments, we believe that it is important to understand the cognitive processes that underlie the formation of evaluation and judgment. Since these cognitive processes and their interactions are likely to be complex, we believe that it is necessary to construct a computational process (CP) model of them (e.g., Mitchell and Smith, 1982). In constructing SUCh a CP model, there are several elements that must be included in the model that appear critical to the evaluation process. First, the evaluation process is clearly dependent on the content and structure of an individual's knowledge. In particular, we believe that the knowledge structure should contain both episodic knowledge as well as general or semantic knowledge, since we believe that episodic knowledge both underlies many cases of evaluation and judgement and provides a basis for semantic knowledge. For example, the availability heuristic (Kahneman and Tversky, 1973) suggests that individuals may search for specific instances of events when comparing how frequently two objects or events occur. In addition, evaluations based on episodic and semantic information may differ (e.g. Tybout, Calder and Sternthal, 1981).

Second, it is important to specify the processes by WhiCh an individual's knowledge structure is accessed, modified or structured during the processes of evaluation and judgement. In particular, evaluations may be viewed as resulting from processes by which a knowledge structure is searched for the most appropriate information given the current context. Third, we believe that the CP model must represent the affective feelings of the individual. In most situations, we believe that affective feelings or emotion play an important role in forming evaluations or judgments. Finally, we believe that the model must also represent the formation of expectations and inferences. In evaluating an object, for example, we believe that individuals form expectations about the object and its performance and that these expectations guide much of the evaluative processing that occurs.

We are currently developing a CP model of evaluation and judgmental processes. In this paper. we present a basic overview of the model. Our modelling strategy is to explain the results of a number of empirical studies using our current knowledge concerning memory and cognitive processes. The empirical results and theories of cognitive psychology provide the constraints on our model. We realize that these constraints do not completely specify the model and that additional empirical studies are necessary. We hopeS however, that these constraints will allow us to make considerable progress in formulating the model. We emphasize that this paper represents an initial formulation of the model structure. We view the building of the CP model as taking a number of years with many different model-test-modify cycles. Hence, future versions of the model may be quite different from the model presented here.

In the remainder of the paper, we first discuss the empirical constraints and the theoretical basis for our CP model. We then present an overview of our model of evaluation.

Empirical Constraints

Our initial empirical constraints derive from three general areas of research. The first area of research is attitude theory, and more specifically the attitude theory of Fishbein (e.g., Fishbein and Ajzen, 1975) and self perception theory (e.g., Bem, 1972). According to the theory proposed by Fishbein, attitudes have an informational basis. In particular, Fishbein assumes that individuals have a number of concepts associated with the attitudinal object. These concepts are generally assumed to be verbal generalization formed about the attitudinal object. The individual's attitude toward an object is the product of the strength of the associations between the attitudinal object and the concept and the evaluation associated with the concept summed over all the salient concepts or associations. In addition, attitude change can only occur by changing the strength of these associations, the evaluation of the concepts or the salience of the concepts.

In self-perception theory, it is assumed that individuals do not base their attitudes on generalized beliefs concerning an object, but rather search their memory for instances of their behavior toward object. For instance, when individuals are asked whether they like rye bread, they search their memory to determine whether they have purchased rye bread. If they find SUCh instances and cannot attribute an external cause to the instance, they will probably reply that they like rye bread, while if they cannot find such instances, they will reply in the negative. The main difference between Fishbein's attitude theory and self-perception theory seems to be that Fishbein's theory assumes that attitudes are based on generalized knowledge while self perception theory assumes that attitudes are based on episodic knowledge. It ShOUld be noted, however, that Fishbein has never stated specifically that his theory is based only on generalized knowledge. Hence his theory may be interpreted to include episodic information (e.g., see Mitchell and Olson, 1981).

A partial resolution to these conflicting approaches may be based on the research of Chaiken and Baldwin (1981). They found that when subjects exhibited a weak relationship between their attitudes toward a concept and the cognitive base of this attitude, they were more influenced by the recall of their past behaviors concerning the concept than subjects WhO exhibited a strong relationship between the affective component and the cognitive component. These results suggest that when individuals have generalized knowledge about an object and when this information is integrated in a tightly knit knowledge structure, attitudes may be primarily based on generalized knowledge. When they lack such generalized knowledge or when such Knowledge is not well integrated, attitudes may be primarily based on the recall of their behavior with respect to the object.

The second area of research is behavioral decision theory which has examined judgmental biases. This research indicates that individuals use a number of heuristics in forming judgments and that these judgments are subject to a number of specific biases. Among the biases are availability, concreteness, and represenativeness (Kahneman and Tversky, 1973; Tversky, 1973; Nesbitt and Ross, 1980). The availability heuriStic occurs when individuals make judgments about the frequency of the occurrence of an event. With these types of judgments it is suggested that individuals search their memory for instances of that event and judgments based on the recall of these events. Since the recall Of certain types of events may be biased, the resulting judgments will also be biased.

When individuals rely on both abstract information and information about specific events, the concreteness heuristic suggests that they will tend to emphasize the latter. For example, knowledge that a neighbor had a bad experience with a particular brand of automobile may be weighted more heavily in evaluating that brand of automobile than the results of a survey of a large number of owners indicating a low repair rate. The empirical evidence for these effects, however, are not consistent (Taylor and Thompson, 1982).

Finally, with representativeness, judgments are based on how close the fit is between the instance and previous instances or generalizations of a class of objects. In these situations, individuals frequently ignore base rate information. The representativeness bias often occurs when subjects are given a description of an individual and are asked to judge whether that individual is a member of a particular class of individuals (e.g., engineer or professor of Chinese history). Under these conditions, individuals behave as though they only consider the match between the target individual and the stereotypes of the two classes. They tend to ignore the fact that one class of individual occurs with greater frequency in the population.

The third area of research is person memory. Recent research in this area has examined the recall of events involving a particular person (e.g., Hastie and Kumar, 1970; Hastie, 1981; Srull, 1981). In this research, subjects are provided with information about a target individual and are then given a series of descriptions of the target individual's behavior which are either congruent or incongruent with the original information. The findings indicate that subjects have better recall for the incongruent information. Hastie (1981) has proposed an associative network model of memory to explain these results. In the model, the behavior of the target individual is linked to a subject node for that individual, while links are formed between the different behavioral manifestations. More links. however, are formed for the incongruent behavior because the individuals spend more time trying to interpret incongruent behavior.

The significance of the three sets of empirically based constraints lies both in the behavior that the CP model should exhibit and in the architecture of the CP model. In particular, the model should exhibit behavior that depends on access to episodic knowledge in some contexts and more general or semantic knowledge in other contexts. The output of the CP model should also exhibit the biases characterizing heuristic judgment, as well as having a tendency to recall incongruent rather than congruent behavior. Hence the architecture of the CP model should be designed to allow the elicitation of such behavior. In particular, this involves considerations of how an individuals knowledge structure is organized and now it is accessed in contexts requiring evaluations and/or judgments.

Theoretical Basis

While the theoretical basis for the model is provided by a large body of research in both artificial intelligence and cognitive psychology, three specific areas of research have a special significance. The first area of research concerns network models of semantic knowledge (e.g., Anderson, 1976). Within these models, generalized knowledge is organized and stored in memory in the form of propositions. The links between the concepts represent the relationships between concepts in the propositions and the strengths of these relationships (Figure 1). The retrieval of information from the semantic network occurs through spreading activation. One or more nodes in the network are activated and activation spreads through the network according to the strength of the associations in the linkages. When a node achieves a sufficient level of activation, it too becomes activated. The stronger the associations between two nodes, the more likely that one node will be activated when the other one is activated.

FIGURE 1

ASSOCIATIVE NETWORK MODEL OF MEMORY

The probability of recalling a particular proposition is a function of the strength of association between the subject of the proposition and the object of the proposition and the strength of the associations linked to that subject from other propositions (e.g., Anderson, 1976). Consequently, the other propositions linked to the same subject interfere with the retrieval of a particular propositional link to that subject. Recently, it has been suggested that propositions with the same general topic may be linked together in subdivisions of the structure so that interference occurs only within a specific topic pertaining to that subject (Smith, 1981).

Activating two nodes in a memory structure may result in the activation of a different set of nodes than activating only one of the nodes. This may occur if the strength of the association between one of the activated nodes and other nodes is relatively weak and the other activated nodes linked to these other nodes. If only one node is activated, activation may not spread to the second node because the link is relatively weak. However, when both nodes are activated, the node will receive activation from both nodes which may push the activation over the threshold level.

A second area of research is exemplified in the work of Schank and his colleagues on memory organization. In their earlier research on "understanding" (Schank and Abelson, 1977), they emphasized the importance of the episodic basis of human knowledge structures. In more recent research (Schank, 1981a, 1981b), they have emphasized the importance of two distinct knowledge structures: Memory Organizational packets (MOPs) and Thematic Organizational packets (TOPs). MOPs contain knowledge about events and objects, while TOPs contain generalized knowledge about abstract concepts (e.g., justice). The general knowledge about events and objects contained in MOPs guide our everyday interactions with people and objects and are used to form our expectations. For instance, when we go to a store to purchase clothing, we have a general knowledge of what happens in such situations, and this knowledge forms the basis for our expectations. When a particular event does not conform to our expectations, however, it becomes linked to the appropriate MOP. In this respect, the model is somewhat similar to the one proposed by Hastie (1981). Recently, Abelson (1981) has suggested that emotions occur primarily when a particular event does not meet our expectations. A similar proposal was made Dy Simon (1967) a number of years ago.

The final area of research is represented by the recent work of Bower and Cohen (1982) in which emotion is represented as nodes in a semantic network model of memory. The model was developed to explain the results of a series of experiments in which it was found that mood can both facilitate and impede learning. Subjects WhO learned word pairs were able to recall more associations when they were in the same mood at both learning and recall (Bower, 1981). In the model proposed by Bower and Cohen (1982), emotion can be treated as a node in memory, although the resulting link will not be defined by a particular relationship (Figure 2). When an individual is in a happy mood and given the first word of a learned associated pair of words, the activation spreads from the provided word and from the emotional node WhiCh increases the amount of activation that spreads to the target word.

FIGURE 2

ASSOCIATIVE NETWORK MODEL WITH EVALUATIVE NODES

The significance of this theoretical research lies in the computational structures and processes that it suggests as a basis for our CP model of evaluation. In particular, the research indicates that both episodic and semantic knowledge structures may be important in models of cognitive process, while activation spread is an important mechanism for accessing such structures. Furthermore, the work of Bower and Cohen suggests how affect may be incorporated into both episodic and semantic structures.

Domain of Applicability

There are two basic evaluation and judgment processes that we would like to model. The first is the formation of judgments and evaluations that are based only on internal knowledge. Examples of these judgments are "How many deaths are due to automobile accidents annually in the United States?" and "what do you think of the new Chevrolet Cavalier?" In both of these instances, individuals must rely on information stored in long term memory in making these judgments. The important questions in modeling these processes are what information is activated and what cognitive processes are used in forming these judgments or evaluations.

The second type of evaluation and judgment involves the presentation of new information. This information may be about a new object, such as a new brand of beer, or it may be new information about a known object. In the former case, the information about the new object is evaluated using previously stored information while in the latter case the new information is integrated with the previous information. If an evaluation had previously been formed, then thiS evaluation may be changed based on the interpretation of the new information.

Assumptions of the Model

There are a number of assumptions that we have used in our initial formulation of the model. Host of these assumptions are derived from models and theories from cognitive psychology and artificial intelligence. These assumptions are:

1. The basic unit of the model is an episode which is the representation of an event experienced by the individual.

2. Generalized information may be extracted from episodes and this information is represented by propositions.

3. Controlled processing is required to generate this information from episodes.

4. Recall of information from long-term memory occurs through spreading activation.

5. Affect and evaluation are represented by nodes in memory. Each affective and evaluative node has a valence associated with it.

6. Affective nodes are only linked to episodes. Evaluations are linked to structures of generalized knowledge. Affective nodes may be linked in memory.

7. Information decays from declarative memory. Episodic knowledge decays at a faster rate than semantic Knowledge.

8. There are a number of goal states that the individual would like to achieve or avoid. An evaluation is connected to each goal state.

Evaluation Processes

Currently we are considering four general processes involved in evaluation. The first involves pattern matching (Hayes-Roth, 1978). Here information about a new object triggers a particular structure in memory that has an affective node attached to it and the value of this node is then transferred to the new object. Mitchell (1980, 1982), for instance, has suggested that these processes may explain the results of the Mitchell and Olson (1981) experiment. Here subjects were shown advertisements for new brands of facial tissue which contained positively evaluated photographs. The results of the study indicated that products featured in advertisements with positively evaluated photographs had more favorable attitudes than could be predicted from the beliefs that were formed.

The second process involves the activation of the knowledge (either generalized or episodic knowledge) associated with a particular brand. This activation process activates the affective or evaluative nodes associated With the object and these evaluations are combined to form an evaluation of the object. Note that this process occurs only when the evaluation is based on internal information and there are evaluative or affective nodes attached to each association with the object.

The third process occurs when the individual acquires new information about an object and must form an evaluation of this information in order to evaluate the object. Initially, the individual may make inferences about the object and generate counterarguments and support arguments. Then, if the information is accepted, we assume that the individual attempts to link the new information to a particular goal state. For instance, if the individual learns that a new brand of toothpaste has fluoride, he or she might link this information to a goal state of preventing tooth decay. Since this goal state has a positive evaluation, the individual would evaluate the knowledge that the new brand of toothpaste has fluoride positively. In some cases, these processes may he based on the activation of knowledge structures for classes of objects (Fiske, 1982).

The final process is mental simulation. If an individual has difficulty linking information about a new object to goals, he or She may produce a mental simulation of using the object. For instance, an individual may mentally simulate driving a particular automobile, wearing a particular coat or cooking a meal in a particular kitchen. This may help the individual determine if the object will be useful in obtaining a particular goal state.

Overview of Knowledge Representation

As discussed previously, we hypothesize that episodic and semantic knowledge occur in the same knowledge structure; that much of our semantic knowledge is derived from episodic knowledge by inferential processes; and that semantic knowledge guides the acquisition of further episodic knowledge. We exemplify such structures in Figure 3, in which there is a node for Budweiser beer and a number of generalized concepts linked to the central node. There are also a number of episodes linked to the central node and related nodes at the semantic level. These episodes include product usage experience, advertisements and other events associated with Budweiser.

In the model, the individual experiences a number of events that involve the object (e.g., product usage). While experiencing these events, the individual may form linkages between the event and the object and may extract generalized knowledge from these events which then gets stored at a nigher level, although this does not necessarily happen. For example, an individual may drink Budweiser Deer at a party, but because of the excitement of the party, day not link the consumption experience to Budweiser. Instead, the event may be linked to an entirely different node (e.g., fun party node). Even if the linkage is made, the individual may or may not extract generalizable knowledge from the event.

Based on this model, we would expect to find different individuals possessing very different knowledge structures concerning a particular object. Some individuals might have structures containing both episodic and generalized knowledge (Figure 3); others may only have linkages only to episodic knowledge (Figure 4); while still others may have very little episodic knowledge linked to an object (Figure 5)

FIGURE 3

KNOWLEDGE STRUCTURE FOR BUDWEISER BEER

FIGURE 4

KNOWLEDGE STRUCTURE CONTAINING ONLY EPISODIC KNOWLEDGE

FIGURE 5

KNOWLEDGE STRUCTURE CONTAINING ONLY GENERALIZED KNOWLEDGE

After Bower and Cohen (1982), we represent affect with nodes in the knowledge structure. These nodes may take on positive, neutral or negative values. Each episode in memory has an associated affective node, which becomes activated when the episode is recalled. In addition, these episodes may also be connected to the object and to the concepts associated to the object. In Figure 6, for example, there are affective nodes connected to each of the episodes linked to Budweiser beer, there is an evaluative node connected directly to Budweiser beer and two of the generalized concepts linked to the Budweiser node (e.g., tastes good and malty taste). The affective node attached to the representation of the television commercial might be conceptualized as the attitude toward the advertisement (Mitchell and Olson. 1981).

FIGURE 6

KNOWLEDGE STRUCTURE WITH EVALUATIVE AND AFFECTIVE NOTES

An individual who possesses little generalized knowledge about an object and is asked to evaluate it might activate the episodes linked to the object which would in turn activate the affective nodes associated with the episodes. An individual having both generalized knowledge and episodic knowledge might form an evaluation based on both types of knowledge. We hypothesize that the relative impact of each type of information will depend on the strength of the associative links. In some cases, a particular episode may be strongly linked to the central node (e.g., a bad usage experience) and this episode will be activated whenever the central node is activated. In this case we would expect the evaluation connected with this episode to have a strong impact on the resulting evaluation. In other cases, no episode will be strongly linked to the central node and the evaluation will be based almost entirely on the generalized knowledge. Finally, if an evaluation of the object is stored at the central node, this evaluation may be used.

Components of the Model

The model consists of three basic components a declarative memory, working memory and systems (e.g. Brucks and Mitchell, 1981) and is shown in Figure 7. The declarative memory contains both episodic and semantic knowledge linked in a network structure. In addition, declarative memory contains a structure. Working memory includes both activated information from declarative memory and information that is attended to from the environment. The production systems consist of a series of condition-act statements where the condition portion of the statement matches on the contents of working memory.

Our current conceptualization of the model contains the following Production systems:

* GENR: Uses current goals to create generalized knowledge in the form of propositions from activated episodes.

* ADD: Uses current goals to add new episodes and generalized knowledge from working memory to declarative memory.

* ACSG: Uses the contents on working memory to activate goal structures and makes these structures available in working memory.

* ACTN: Uses the contents of working memory to activate episodic and semantic knowledge in declarative memory and makes this knowledge available in working memory.

* LINK: Uses current goals to link generalized knowledge from working memory to a goal.

* INT: Interprets information in working memory and generates inferences, counterarguments and support arguments using knowledge from declarative memory and current goals.

* EVAL: Integrates activated affective and evaluative nodes to form an overall evaluation of a concept or forms a judgment.

* SIM: Uses the contents of working memory to simulate an episode based on current goals and the task.

* ATT: Attends to information in the environment based on the task and current goal structures and makes this information available in working memory.

* OUT: Outputs the judgment or evaluation.

FIGURE 7

CONCEPTUAL MODEL

With these production systems and our conceptualization of declarative memory, we believe that we will be able to represent the cognitive processes involved in evaluation and judgment. For instance, when asked to make an evaluation of an object using only the information in declarative memory, ACTG would activate the appropriate goal structures and ACTN would activate the appropriate knowledge structures and evaluative or affective nodes and make this information available in working memory. Based on the goals and the information in working memory, INT, GENR or LINK may be used to inferences or link generalized knowledge to goals to form an evaluation of that knowledge. Finally, EVAL would take the affective and evaluative information in working memory and form an evaluation of the object. Further experiments will be required to specify these production systems and the different types of knowledge structures that are required.

SUMMARY

In this paper we have discussed the basic structure of a CP model of evaluation and judgment. The model has been developed to provide an understanding of the mental processes involved in forming judgments and evaluations based on information stored in long term memory, information about a new object and new information about a known object. The model contains three components: declarative memory, working memory and systems. Declarative memory contains the goal states of the individual and episodic and generalized knowledge. Affective and evaluative nodes may be attached to each of these structures.

As discussed in the introduction, this structure represents over initial thinking on the model. We fully expect that it will undergo many changes after a more thorough review of the literature and after some initial experiments to test aspects of the model.

It is our hope that the construction of this model will provide a better understanding of the cognitive processes underlying judgment and evaluation and will result in the identification of general principles Which underlie these processes. We also hope that the model will provide a better understanding of the evolution of knowledge structure over time and provide insights into when and why current models of judgment and evaluation (e.g. conjoint measurement) may fail to provide good predictions. For instance, we would predict that evaluations based primarily on episodic knowledge would be relatively unstable over time since episodic knowledge decays at a faster rate than semantic knowledge and the activation of episodic knowledge is probably more context dependent.

REFERENCES

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

Bem, D.J. (1972) "Self-Perception Theory." in L. Berkowitz, ed., Advances in ExPerimental Social Psychology, Vol. 6, New York: Academic Press.

Bower, G.R. (1981) "Mood and Memory." American Psychologist, 36, 129-148.

Bower, G.H. and P.R. Cohen (1982) "Emotional Influences in Memory and Thinking." In M.S. Clark and S.T. Fiske, Affect and Cognition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Brucks, M. and A.A. Mitchell (1981) "Knowledge Structures, Production Systems and Decision Strategies.- In K. Monroe, ed., Advances in Consumer Research, Vol. 8, Ann Arbor, MI: Association for Consumer Research.

Chaiken, S. and X.W. Baldwin (1981) "Affective-Cognitive Consistency and the Effect of Salient Behavioral Information on the Self-Perception of Attitudes," Journal of Personality and Social Psychology, 41, 1-12.

Fishbein, M. (1967) "A Consideration of Beliefs and Their Role in Attitude Measurement." In M. Fishbein, ed ., Readings in Attitude Theory and Measurement, New york: John Wiley and Sons. 257-266.

Fishbein, M. and I. Ajzen (1975) Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research. Reading, MA: Addison-Wesley.

Fiske, S.T. (1982) "Schema-Triggered Affect: Applications to Social Perception." In M.S. Clark and S.T. Fiske, eds., Affect and Cognition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Hastie, R. (1981) "Schematic Principles in Human Memory." In E.T. Higgens, et. al., eds. The Ontario symposium on Personality and Social Psychology. Hillsdale, NJ: Lawrence Erlbaum Associates.

Hastie, R. (1980) "Memory for Information which Confirms or Contradicts a General Impression." In R. Hastie, et. al., eds ., Person Memory: The Cognitive Basis of Social Perception. Hillsdale. NJ: Lawrence Erlbaum Associates.

Hastie, R. and P. Kumar (1979) "Person Memory: Personality Traits as Organizing Principles in Memory for Behaviors," Journal of Personality and Social Psychology, 37, 25-38.

Hayes-Roth, F. (1978) "The Role of Partial and Best Matches in Knowledge Systems." In D.A. Waterman and F. Hayes-Roth, eds., Pattern-Directed Inference Systems, New York: Academic Press, 557-576.

Kahneman, D. and A. Tversky (1973) "On the Psychology of Prediction," Psychological Review, 80, 237-351.

Mitchell, A. (1982) "Cognitive Processes Initiated by Exposure to Advertising." In R. Harris, Information Processing Research on Advertising, Hillsdale. NJ: Lawrence Erlbaum Associates.

Mitchell, A. (1980) "The Use of an Information Processing Approach to Understand Advertising Effects." In J.C. Olson, ed., Advances in Consumer Research, Vol. 7, Ann Arbor, MI: Association for Consumer Research.

Mitchell, A. and T. Smith (1982) "The Applicability of Computational Process Models of Representing Consumer Behavior." In A. Mitchell, ed., Advances in Consumer Research, Vol. 9, Ann Arbor, MI: Association for Consumer Research.

Mitchell, A. and J.C. Olson (1981) "Are Product Attribute Beliefs the Only Mediator of Advertising Effects on Brand Attitude?" Journal of Marketing Research, 18, 318-332.

Nesbitt, R. and L. Ross (1980) Human Inference: Strategies and Shortcomings of Social Judgment. Englewood Cliffs, NJ: Prentice Hall.

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

Norman, D.A. (1981) "The Categorization of Action Slips," Psychological Review, 88, 1-15.

Schank, R. (1981a) Dynamic Memory: A Theory of Reminding in ComPuters and PeoPle. Unpublished Manuscript.

Schank, R. (1981b) "Failure Driven Memory." Cognition and Brain Theory, 1, 41-60.

Schank, R. (1980) "Language and Memory." Cognitive Science, 4, 209-241.

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

Schank, R., R. Wilensky, J.G. Carhouell, Jr., J.L. Kolodner and J.A. Hendler (1978) "Representing Attitudes: Some Primitive States." Research Report #128, Computer Science Department, Yale University, New Haven, CN.

Simon, H.A. (1967) "Motivational and Emotional Controls of Cognition," Psychological Review, 74, 29-39.

Srull, T.K. (1981) "Person Memory: Some Tests of Associative Storage and Retrieval Models," Journal of ExPerimental Psychology: Human Learning and Memory, 7, 440-463.

Taylor, S.E. and S.C. Thompson (1982) "Stalking the Elusive 'Vividness' Effect," Psychological Review, 89. 155-181.

Tversky, A. (1973) "Availability: A Heuristic for Judging Frequency and Probability," Cognitive Psychology, 5, 207-232.

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

Tybout, A., B.J. Calder and B. Sternthal (1981) "Using Information Processing Theory to Design Marketing Strategies," Journal of Marketing Research, 18. 73-79.

Wyer, R.S. (1975) Cognitive Organization and Change: An Information Processing Approach, Hillsdale, NJ: Lawrence Erlbaum Associates.

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