Identifying Consumer Information Processing Strategies: New Methods of Analyzing Information Display Board Data

Thomas Hofacker, Johann Wolfgang Goethe-Universitat
ABSTRACT - This article presents new ideas on analyzing information acquisition data in brand choice. Firstly, an improved analysis of transitions is introduced that allows for the identification of paired comparisons and for discrimination between the structured and unstructured elements of acquisition sequences. Secondly, a moving average algorithm is put forward so as to identify changes in a subject's information acquisition sequence. Thirdly, a cluster analysis is used to classify natural types of processing strategies. Empirical results evidence a considerable amount of multistage information acquisition as well as six clear-cut search strategies: processing by attributes, two different types of brand processing, paired comparison processing, mixed search and random processing.
[ to cite ]:
Thomas Hofacker (1984) ,"Identifying Consumer Information Processing Strategies: New Methods of Analyzing Information Display Board Data", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 579-584.

Advances in Consumer Research Volume 11, 1984      Pages 579-584

IDENTIFYING CONSUMER INFORMATION PROCESSING STRATEGIES: NEW METHODS OF ANALYZING INFORMATION DISPLAY BOARD DATA

Thomas Hofacker, Johann Wolfgang Goethe-Universitat

[Preparation of this report was facilitated by a grant of the "Deutsche Forschungsgemeinschaft", Bonn-Bad Godesberg, West Germany.]

[Thomas Hofacker is Doctoral Candidate, Johann Wolfgang Goethe-Universitat, Mertonstr. 17, 6000 Frankfurt/M, West Germany.]

ABSTRACT -

This article presents new ideas on analyzing information acquisition data in brand choice. Firstly, an improved analysis of transitions is introduced that allows for the identification of paired comparisons and for discrimination between the structured and unstructured elements of acquisition sequences. Secondly, a moving average algorithm is put forward so as to identify changes in a subject's information acquisition sequence. Thirdly, a cluster analysis is used to classify natural types of processing strategies. Empirical results evidence a considerable amount of multistage information acquisition as well as six clear-cut search strategies: processing by attributes, two different types of brand processing, paired comparison processing, mixed search and random processing.

INTRODUCTION

Over the last decade consumer information seeking and processing has emerged as an important field of inquiry. Bettman (1979) has illustrated that a cognitive approach provides a very promising mode of explaining consumer behavior. Some methods (e.g. information display board, recording of eye fixation) have been developed in order to capture the process of information acquisition in brand choice. Numerous empirical studies have used these process tracing methods to describe and explain consumer information processing (cf. Bettman and Jacoby 1976; Biehal 1980; Capon and Burke 1980; Jacoby et al. 1976; Jacoby, Szybillo and Busato-Schach 1977; Lussier and Olshavsky 1979; Payne 1976; van Raaij 1977).

Despite the theoretical and empirical interest in this area methodological contributions are still rather limited. This applies both to methods of process data collection as well as to the analysis of such data once collected (Bettman 1979). The second of these two points will be discussed here. Problems of data collection are not considered. See Bettman (1979) Jacoby et al. (1976) or Russo (1978) for a description of process tracing techniques. In the course of this article new methods for analyzing information acquisition process data will be presented. The advantages of the methods proposed will be demonstrated using experimental data from the brand choices of 360 consumers. Despite the fact the data was collected by a kind of information display board, the data analysis methods are equally applicable to acquisition process data collected by other techniques.

THEORETICAL BACKGROUND

In consumer information processing theory, multiattribute decision processes are described by means of choice heuristics. Consumers are held to use different kinds of heuristics when comparing and choosing brands. Bettman (1979, pp. 179-185) provides a detailed overview of important heuristics. Two aspects of these heuristics are of particular relevance for the problems discussed here: the form of processing implied and the existence of phased (multistage) strategies. Some heuristics (e.g. linear compensatory, conjunctive or disjunctive) assume processing by brand. That is to say, consumers choose a brand, collect information on several attributes and then evaluate that brand. Then they choose a second brand and so on. Other heuristics (e.g. lexicographic, elimination by aspects) imply processing by attributes. This means that all brands are compared according to a single attribute, followed by a comparison based on a second attribute and so forth. Finally, yet other heuristics (e.g. additive difference, attribute dominance) assume a paired comparison as a special kind of attribute processing. In this form of processing, two brands are directly compared for their different attributes in an XYX-manner (Russo and Rosen 1975). Subsequently, another alternative can be selected and compared with the previous winner.

There is additionally evidence to suggest that some consumers make use of other forms of unstructured (random) or mixed processing (Bettman and Jacoby 1976; Bettman 1979; Capon and Burke 1980). Phased or multistage strategies are an important kind of mixed processing. To simplify complex choices, consumers may use different kinds of heuristics in the course of their decision process. For example, an elimination by aspects rule might be used to eliminate some brands in the first decision phase and a linear compensatory or an additive difference rule applied in the second phase to make comparisons within this smaller set of acceptable brands (Bettman 1979; Bettman and Park 1980; Biehal 1980; Raju and Reilly 1980). The choice heuristics mainly refer to internal processing, but they also imply specific search strategies (Einhorn and Hogart 1981). The form of processing adopted and the single stage or multistage nature thereof are important characteristics of these information acquisition strategies (Bettman 1979; Biehal 7980; Svenson 1979).

METHOD

State of Research

The existing techniques for information acquisition data analysis were developed mainly by Bettman, Jacoby and Payne in the mid-1970s (cf. Bettman and Jacoby 1976; Bettman and Kakkar 1977; Jacoby et al. 1976; Payne 1976). In order to describe the strategies of individual consumers such techniques concentrate on the sequence of information acquisition responses. An adjacent pair of acquisition responses is referred to as a 'transition'. Therefore if "n" pieces of information are required, there are "n-1" transitions. These transitions can be divided into four types. Type 1 (the same attribute and brand), Type 2 (the same brand, different attributes), Type 3 (different brands, the same attribute) and Type 4 (different brands and attributes). Depending on the proportion of these transition types, particularly of Types 2 and 3, in the acquisition sequences, these can be grouped into three or more acquisition patterns (cf. Bettman and Kakkar 1977; Capon and Burke 1980)

With regard to the important theoretical aspects of acquisition strategies, these traditional methods have two significant shortcomings: firstly, the form of processing cannot be described with sufficient accuracy in terms of the proportions of the four transition types. -Transitions of Type 4 are required for shifts to other attributes or brands, even highly structured (e.g. linear compensatory, lexicographic) sequences. The occurrence of Type 4 transitions, caused by the structure of the underlying heuristic. cannot be limited to somewhat infrequent shifts. In paired comparison processing, alternating sequences of transition Types 3 and 4 are almost typical (van Raaij 1977). The normalization proposed by Bettman et al. (Bettman and Jacoby 1976; Bettman and Kakkar 1977) must thus be regarded as an inadequate solution to the problem of structured shifts. Secondly, the above classification being based on full-sequence proportions, the traditional concepts imply that information acquisition is of a single stage nature. Bettman and Jacoby (1976) tried to include multistage processing, but their attempt was limited to a weak identification of paired end comparisons. Changes in search strategies are usually analyzed by breaking individual sequences into two halves and then comparing the first and second halves (Biehal 1980; Svenson 1979; van Raaij 1977).

Analysis of Triple Transitions

To overcome the limitations inherent in the existing techniques, as outlined above, the author has developed a set of improved methods for the analysis of acquisition process data. First of all, the concept of transitions must be extended to a sequence of three acquisition responses. For a detailed description see Hofacker (1983a). Thus, 'n' units of information acquired form 'n-2' triple transitions. The concept of triple transitions requires a larger number of transition types than does the traditional concept of paired transitions. There are 3x3x3=27 possible variations of brands and attributes within each triple transition. These variations are divided into five categories: Categories I and III constitute those variations embracing either one or three brands (attributes). Category II consists of all triads of acquisition responses which embrace two alternatives or attributes. This category can be subdivided into those transitions which embrace

- twice the same and then another brand or attribute (Category IIa),

- first one and then twice another (Category IIb), and

- one, then another, and then the first again (Category IIc) [Jacoby et al. (1976) propose a different concept of triple transitions. They do not divide Category II into subcategories. As a consequence Category II consists of 18 of the 27 possible variations of brands and attributes. The resulting transition types then being too heterogeneous for interpretation.].

25 transition types can be derived from the five transition categories of brands and attributes. The complete classificatory matrix is shown in Figure 1. In each matrix _ell the transition pattern is depicted graphically in addition to an indication of its type.

With regard to the objective of describing different forms of processing in information acquisition, the matrix can De partitioned into two triangular matrices along the main diagonal, which contains unstructured elements. The upper triangle consists mainly of processing by attributes, the lower mainly of processing by brands. The problem involved in categorizing paired comparisons concerns Types 6 and 7. If one of these transition types occurs, the transitions bordering on it have to be examined as to whether a paired comparison sequence exists. If this is indeed the case, type 6 is recoded as Type 26 and Type 7 as Type 27. Otherwise the initial classification remains unaltered. Similar problems can arise when distinguishing between unstructured shifts and those caused by structured processing.

FIGURE 1

CLASSIFICATORY MATRIX OF THE ANALYSIS OF TRIPLE TRANSITIONS - TRANSITION TYPES AND GRAPHIC PATTERNS

Normally, a diagonal shift is regarded as unstructured A transition type is recoded as a structured shift only in the case of the same processing structure (by attributes, alternatives or paired comparison) being valid on both sides of the diagonal shift. Only Types 11 and 12, both clearly being paired comparisons, can be excluded from this category. The precise differentiation of transition Types 6, 7, 9, 10, 18, 19, 20 and 21 is indicated in brackets in the classificatory matrix (cf. Fig. 1).

Five strategy indices can be calculated from the above 33 transition types. These indices refer to the proportion of

- processing by brands,

- processing by attributes,

- paired comparison processing,

- structured shifts, and of

- unstructured (random) processing.

The first two indices differ but slightly from the corresponding indices of the traditional concept of paired transitions. Differences exist merely in the method of calculation, for each triple transition consists of two transition parts that both individually correspond to the types of the paired concept. To calculate the proportion of processing by brands, all transition types which contain parts of a search by brands and which are not paired comparisons are added together Type 25 consists of two vertical parts and thus is entered twice in the sum. The proportion can finally be computed as the ratio of this addition to twice the number of all transitions. The index for processing by attributes is obtained in a similar manner

The proportion of paired comparisons is computed by the sum total of Types 11, 12, 26 and 27 being divided by the total number of transitions for each subject Transition Types 28 to 33 provide the computational basis for the index of structured shifts. These six types are added together and then divided by twice the number of all transitions Its counterpart, namely the index of unstructured processing, is comprised as a residual of all transition parts which can be neither allocated to paired comparisons to processing by brands or attributes, nor to structured shifts. The sum of these parts (diagonal shifts and reaccesses) is divided by twice the number of the transitions of an individual subject.

Identifying Changes in Acquisition Sequences

The second improvement made in comparison to existing techniques concerns the multistage character of information acquisition sequences. With the method of analysis presented in this article, not only the information acquisition process as a whole can be described, but rather in the calculating of the five indices of the triple transition analysis the process can be checked for changes in the acquisition strategy used It is assumed that multistage strategies consist of at least two internally relatively homogeneous substrategies (Bettman 1979; Bettman and Park 1980; Svenson 1979). A break occurs between these substrategies, leading to a change in the transition indices at the level of information acquisition.

An algorithm based on the notion of a moving average (Biehal 1980) was devised to identify this change. A vector of Xm=n-2' types of triple transitions was first constructed from the 'n' acquisition responses of a subject, as described above. A window of width 'k' moves along this transition vector. The five strategy indices of the triple transition analysis can be calculated for each position of the window, i.e using a window width of 6 first of all for the transitions 1 to 6 and then for the transitions 2 to 7 and so forth. For "m" transitions and with a window width of 'k', a result of 'm-k+l' vectors of strategy indices is arrived at. The extent of strategy change is then given by the difference between the index vectors of adjacent window positions. These 'm-k' distances can be quantified according to a Minkowski-supremummetric (e.g. Duran and Odell 1974; Hartigan 1975). That is to say, the distance corresponds to the index with the greatest absolute difference. If the distance is zero, then one can safely say that no strategy change has taken place. If the measure of distance is, however, greater than zero, then the processual structure of information acquisition has changed.

If, however, all distances greater than zero were to be interpreted as the beginning of a new phase, then the actual extent of multistaging would be vastly overestimated. Even abrupt strategy changes are smoothed by the moving average down to continuous changes, whereby the smoothing factor is dependent on the width of the window. Furthermore, it is to be assumed that empirical information acquisition processes do not correspond completely with the theoretical models, but rather that accidental interruptions (short or minor strategy changes), which must be eradicated in the search for theoretically relevant strategy changes, are superimposed over these. With the aid of an analysis of the differences that occur, continuous and oscillating changes are collected into one typical borderline. Minor changes in the information acquisition sequence caused by bias effects can be corrected by means of minimal cut-offs of both the distance between adjacent phases as well as to the relative length of each phase. Any requirements made of the minimum length run the danger of confusing short adjacent phases, that, however, differ radically in their form of processing. Therefore it would seem advisable to make careful use of the minimum length Parameter (e.q. < 20 E).

Clustering Acquisition Strategies

As a result of the moving average algorithm the single stage processes and the individual phases of the multistage processes can be characterized by the five strategy indices of the triple transition analysis. In order to facilitate interpretation the separate phases are grouped by means of a cluster analysis into typical processing strategies in information acquisition. Unlike previous conceptualizations (Bettman and Jacoby 1976; Capon and Burke 1980) classification is not made here by explicitly predetermining threshold values in a discrimination net. Rather, empirical types are identified in the totality of phase-related data.

The classification can be accomplished by means of a two stage cluster analysis. Prior to the actual analysis principle components are calculated from the correlated strategy indices. Following this, the first step of the cluster analysis involves a hierarchical classification using the algorithm proposed by Ward (1963). The next step entails an optimization of the preliminary classification arrived at by the hierarchical procedure via an iterative relocation algorithm (Hartigan 1975; Sneath and Sokal 1973). Both procedures use a quadratic measure of distance to produce spheric clusters of minimal variance. On account of the data processing facilities available, the author made use of subprograms of the CLUSTAN 1C Package (Wishart 1975); other computer programs for cluster analysis should, however, be equally usable A description of the computer programs used in the methods of analysis presented in _his article is to be found in Hofacker and Piesold (1983).

SUBJECTS AND PROCEDURE

Information acquisition data from an empirical study on product decision making is used to demonstrate the results obtainable using the improved methods of data analysis developed. The study was conducted in the Rhein-Main area of West Germany in the summer of 1981 as part of a larger research project. The subjects were 360 female consumers, all mothers of babies. They represented a full sample of the mothers of births registered at the Frankfurt and Offenbach Registry Offices in December 1980 and in May 1981.

The subjects had to choose one brand from amongst several brands of baby diapers. Product information was presented via a brand x attribute matrix in the form of a product test by the "Stiftung Warentest", the German consumer product testing agency. The of information acquisition responses were recorded by means of a process tracing technique specially designed for field studies using questionnaires. This process methodology, a synthesis of information display board and verbal protocol, is discussed in Kaas and Hofacker (1983). The main differences between the method of the present study and those used in previous studies are the larger sample size (360 subjects instead of a few dozen), the use of non-student subjects, the presentation of information by means of a product test and the use of a questionnaire-based technique to collect acquisition data See also Kaas' article (1983) for a more detailed description of the study.

RESULTS

The results of the methods of analysis introduced above concern the similarities and differences of strategy indices in paired and triple transition analyses, the frequency of single and multistage information acquisition processes and the strategy types found by means of cluster analysis. Furthermore, the analytical procedure suggested here enables statements to be made both on the frequency of strategy types in single stage and multistage processes as well as on the combination of certain substrategies. These results cannot be elaborated on in this paper for reasons of space but are discussed in Hofacker (1983b).

New and Traditional Strategy Indices

The strategy indices of the traditional paired concept can be compared at the level of full sequence data with the extended concept of triple transitions. The mean values of the transition indices are displayed in Table 1. [The original proportions of transition Types 2 and 3 in the paired concept are computed without any kind of normalization. Owing to very short acquisition sequences (less than three responses) the sample size in the triple transition analysis is reduced from 360 to 356 subjects.]

TABLE 1

STRATEGY INDICES OF PAIRED AND TRIPLE TRANSITION ANALYSIS

As expected, the proportion of processing by brands, processing by attributes and of unstructured processing is lower in the triple concept because of the two additional indices. However, the extent of the change is interesting Whereas a slight difference exists for the brand index, the search by attributes is lessened considerably and the proportion of random processing is reduced to less than half. T-Tests for dependent samples show significant differences at the level of p < Q.001 for the three indices. In this manner the criticism levelled against the paired transition concept is supported by empirical evidence. Indeed, the consumer's information acquisition processes contain quite a considerable proportion of paired comparisons and structured shifts, which are not treated as separate indices in the paired concept, but, moreover confound the remaining indices. The index of random processing seems to be most effected by this error.

Apart from the difference in mean values, the strategy indices of both concepts are highly correlated. The Pearson coefficients for processing by brands and by attributes are r=0.99 and r=0.95 respectively. The convergence of random processing is somewhat smaller (r-0.77). These results show that, especially in the case of brand and attribute processing, the new triple transition concept is highly comparable with the traditional paired transition concept. See Hofacker (1983a) for a more detailed presentation and discussion.

Single Stage and Multistage Processes

Not only the exact description of the form of processing in information acquisition but also the frequency of strategy change is of interest. Using the data mentioned above the moving average algorithm provided stable solutions as to the number of phases over a wide interval of variable parameters. This holds true for a window width of 4 to 8 transitions, for a minimum distance between adjacent phases of 0.1 to 0.4 and for a relative minimum length of the separate phases of 0 % to 20 8. With regard to the interpretability of the number and content of strategy phases a window width of 6 and a minimum distance of 0.4 without a determined minimum length was chosen. The frequence of single stage and multistage processes in information acquisition is given in Table 2.

TABLE 2

PERCENTAGE OF SINGLE STAGE AND MULTISTAGE PROCESSES

More than half of the consumers retain a strategy once chosen without making any substantial changes to it. [The number of subjects with a transition vector smaller or equal to the chosen window breadth is, at 17.7 %, not exactly small. The methodological artefact thus determined should, however, not be accorded great weighting, for there is only a slight probability of multi-staging in such short sequences.] In the case of the other subjects, the form of processing in information acquisition changes. The two or three stage forms of processing constitute by far the greatest part of these multistage strategies. Four and more stage strategies are extremely rare. The frequency distribution agrees in size with that in previous studies, which found 60-80 E of the information acquisition processes to be single phase and 20-40% to be multiphase (cf. Lussier and Olshavsky. 1979, pp. 161-163; Barbour 1981, pp. 6-8).

Types of Processing Strategies

The single stage processes and the strategy phases of the multistage acquisition processes are characterized by the five indices of the triple transition analysis. On this basis they can be condensed into strategy types by means of the cluster analysis described above. The total 573 phases are classified by the cluster analysis. Applied to this da;a both Ward's (1963) hierarchical algorithm as well as the relocate algorithm show a distinct increase of inner-cluster variance when moving from 6 to 5 clusters. Because of this clear elbow in the increase of error variance, the number of 6 strategy types was used as the basis of all further analysis. The cluster's frequencies and the mean strategy indices' of the clusters are given in Table 3. [The following process of error variance within the cluster was found for the relocate procedure when using a 10 cluster start configuration obtained from the hierarchical algorithm: 9CL=15.22; 8C,=17.43; 7CL=30.34; 6CL=31.29 5CL=70.11; 4CL=112.02; 3CL=214.19; 2CL=260.62. Almost identical results were obtained with a relocate procedure using a random start configuration.]

TABLE 3

PHASES OF INFORMATION ACQUISITION: TYPES OF PROCESSING STRATEGIES

The six strategy classes are significantly different at the level of the five transition indices (MANOVA: Wilks Lamda p < 0 001). The dominance of individual strategy indices in the clusters is a very important finding. The classes clear-cut and compelling structure is particularly noteworthy because this structure is solely determined by the distribution of density in the space of the transition indices. It is not determined by arbitrary models, as was the case in previous analyses (Bettman and Jacoby 1976; Capon and Burke 1980; Jacoby et al 1976).

The first four clusters can be assigned to certain choice heuristics on the basis of the form of processing. In Cluster 1 the proportion of processing by attribute is very high and the proportion of structurally-determined shifts not inconsiderable. This corresponds to processing by lexicographic or elimination by aspects heuristics. Cluster 2 exhibits a very high proportion of brand transitions and a low proportion of structurally-determined shifts, which points to the use of a linear compensatory heuristic. In Cluster 3 processing by brands is predominant, but the proportion of structurally-determined shifts is also considerable. Such short sequences of brand processing are to be expected when conjunctive or disjunctive heuristics are used. Cluster 4 contains paired comparisons in the information acquisition process which corresponds to the additive difference or the attribute dominance rule. The strategy phases collected in Cluster S cannot be allocated to one dominant strategy. In this mixed type there are considerable proportions of processing by brands, by attributes and random processing. A further analysis of Cluster 5 showed in almost all cases frequently changing, short sequences of different forms of processing. Bettman and Jacoby (1976) labelled this phenomenon feedback processing. Finally, in strategy cluster 6 all information acquisition phases are collected in which unstructured (random) processing predominates.

SUMMARY AND DISCUSSION

In consumer information processing theory a row of choice heuristics are discussed. The use of certain heuristics is precipitated in the information acquisition process. which can then be recorded by means of process tracing techniques. Traditional methods for the analysis of information acquisition data only allow a quite crude description of information acquisition strategies to be made. The new methods presented above provide a more detailed picture of the acquisition processes and thus enable a more exact interpretation of the data to be undertaken.

The comparison of the strategy indices of paired and triple transition concepts demonstrated the advantages of the new technique for the description of the form of processing used in information acquisition. Despite the differences in mean values, there was a strong convergence of the proportions of processing both by brand and by attribute, which ensures the comparability of both analysis techniques.

The new methods not only offer a better description of the form of processing but also of the dynamic change of information acquisition processes. The moving average algorithm is an attempt to identify the theoretically expected changes between homogeneous strategy phases in empirical information acquisition data. Problems arise because of both smoothed and minor changes in information acquisition sequences. Precisely these problems were solved by the use of a quite complex computer program (cf. Hofacker and Piesold 1983). The results of the moving average algorithm show a high proportion of multistage information acquisition processes that could not be adequately considered in previous studies (e.g. Biehal 1980).

The strategy types discovered by means of the cluster analysis provide criteria for judging the suitability of the new analytical techniques. The natural strategy clusters are clearly delineated from one another with respect to the form of processing used. Furthermore, four clusters contain information acquisition sequences theoretically to be expected when specific choice heuristics are used. These four clusters contain more than 70 % of all phases.

Despite the promising results obtained with the new methods of analysis a concluding evaluation is not possible as yet. Further knowledge can only be gained by applying the new methods to data from other studies with difference samples, different products and different process tracing techniques.

Two substantial implications present themselves. Both for empirical research into and for the theory of consumer information processing. First of all, greater consideration must be paid to the multistage character of information acquisition and processing. This is particularly the case for theoretical models in which the almost exclusive presence of two stage processes composed of elimination and selection phases has been assumed (Bettman 1979; Lussier and Olshavsky 1979; Raju and Reilly 1982). Secondly, in future empirical investigations other characteristics should be included in the description of the phases of multistage strategies. Such features as depth and content of information acquisition would come into question. At the same time this would provide an opportunity to improve on the methods of analysis presented here.

REFERENCES

Barbour, Frederic L. (1981), "Consumer Decision Processes: An Unobtrusive Measure", Working Paper, University of Tennessee.

Bettman, James R. (1979), An Information Processing Theory of Consumer Choice, Reading, Mass.: Addison-Wesley.

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

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

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

Biehal, Gabriel J. (1980), "An Exploration of Consumer Search Strategies in Internal and External Environments", Working Paper, Center for Consumer Research, No. 23, University of Florida.

Capon, Noel and Burke. Marian (19 30), "Individual, Product Class, and Task-Related Factors in Consumer Information Processing", Journal of Consumer Research, 7, 314-326.

Duran, Benjamin S. and Odell, Patrick L. (1974), Cluster Analysis, A Survey, Berlin: Springer.

Einhorn, Hillel J. and Hogart, Robin M. (1981), "Behavioral Decision Theory: Processes of Judgement and Choice", Annual Review of Psychology, 32, 53-88.

Hartigan, John A. (1975), Clustering Algorithms, New York: Wiley & Sons.

Hofacker, Thomas (1983a), "The Structure of Information Acquisition Processes - An Improved Analysis of Transitions and Empirical Results", Journal of Consumer Research, in Review.

Hofacker, Thomas (1983b), "Entscheidung als InformationsverarbeitungsprozeB - dargestellt anhand der Verarbeitung von Produktinformationen durch Konsumenten", Doctoral Dissertation, Johann Wolfgang Goethe-Universitat Frankfurt am Main (West Germany), in Press.

Hofacker, Thomas and Piesold, Ralf R. (1983), "IASTRA: Structural Analysis of Information Acquisition Processes in Multiattribute Brand Choice", Journal of Marketing Research, 19. in Press.

Jacoby, Jacob, Chestnut, Robert W., Weigl, Karl C. and Fisher, William (1976), "Prepurchase Information Acquisition: Description of a Process Methodology, Research Paradigm, and Pilot Investigation", in Advances in Consumer Research, Vol. 3, ed. Beverlee B. Anderson, Cincinnati: Association for Consumer Research, 306-314.

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

Kaas, Klaus P. and Hofacker, Thomas (1983), "Informationstafeln und Denkprotokolle - Bestandsaufnahme und Entwicklungsmogllchkeiten der ProzeRverfolgungstechniken", in Innovative Marktforschung, ed. Forschungsgruppe Konsum und Verhalten, Wurzburg: Physica.

Lussier, Denis A. and Olshavsky, Richard W. (1979), "Task Complexity and Contingent Processing in Brand Choice", Journal of Consumer Research, 6, 154-165.

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

Raaij, W. Fred van (1977), "ConsUmer Choice Behavior: An Information Processing Approach", Doctoral Dissertation, Tilburg University.

Raju, Puthankurussi S. and Reilly, Michael D. (1980), "Product Familiarity and Information Processing Strategies", Journal of Business Research, 8, 187-212.

Russo, J. Edward (1978), "Eye Fixations Can Save the World: A Critical Evaluation and a Comparison Between Eye Fixations and Other Information Processing Methodologies", in Advances in Consumer Research, Vol. 5, ed H. Keith Hunt, Ann Arbor: Association for Consumer Research, 561-570.

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

Sneath, Peter M. and Sokal, Robert R. (1973), Numerical Taxonomy, 2nd ed., San Francisco: Freeman & Company.

Svenson, Ola (1979), "Process Descriptions of Decision Making", Organizational Behavior and Human Performance, 23. 86-112.

Ward, J.H. (1963), "Hierarchical Grouping to Optimize an Objective Function", Journal of the American Statistical Association, 58, 236-244.

Wishart, David (1975), CLUSTAN 1C User Manual, London (GB): University College London.

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