Pre-Purchase Information Acquisition: Description of a Process Methodology, Research Paradigm, and Pilot Investigation

Jacob Jacoby, Purdue University (student), Purdue University (student), Purdue University (student), Purdue University
Robert W. Chestnut,
Karl C. Weigl,
William Fisher,
ABSTRACT - Using a methodology which captures consumer information acquisition as a dynamic behavioral process, four relatively clear-cut search strategies were identified: brand name reliance; predominantly within-brand search; predominantly within-dimension search; and complex search. [A variety of descriptive labels have been employed in recent attempts to categorize information acquisition strategies. Most often used are the terms "within-brand" and "within-dimension." To reduce confusion, the following can be equated. Interdimensional search (Payne, 1975), choice by processing brands (Bettman and Jacoby, 1975), type "2" search (this paper), and within-brand search all denote the successive acquisition of some set of information dimensions on a single brand or alternative. In contrast, intradimensional search (Payne, 1975), choice by processing attributes (Bettman and Jacoby, 1975), type "3" search (this paper), and within-dimension search all denote the successive comparison of some set of brands or alternatives on a single dimension of information.] Path analysis suggests search strategy to be partially determined by brand loyalty and consumption frequency. These findings are considered in terms of a general "coding" explanation and placed within a four-stage paradigm developed for guiding, integrating and interpreting consumer information processing research.
[ to cite ]:
Jacob Jacoby, Robert W. Chestnut, Karl C. Weigl, and William Fisher (1976) ,"Pre-Purchase Information Acquisition: Description of a Process Methodology, Research Paradigm, and Pilot Investigation", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 306-314.

Advances in Consumer Research Volume 3, 1976      Pages 306-314

PRE-PURCHASE INFORMATION ACQUISITION: DESCRIPTION OF A PROCESS METHODOLOGY, RESEARCH PARADIGM, AND PILOT INVESTIGATION

Jacob Jacoby, Purdue University

Robert W. Chestnut (student), Purdue University

Karl C. Weigl (student), Purdue University

William Fisher (student), Purdue University

[Preparation of this report was facilitated, in part, by a grant from the National Science Foundation (GI-43687) to the senior author.]

ABSTRACT -

Using a methodology which captures consumer information acquisition as a dynamic behavioral process, four relatively clear-cut search strategies were identified: brand name reliance; predominantly within-brand search; predominantly within-dimension search; and complex search. [A variety of descriptive labels have been employed in recent attempts to categorize information acquisition strategies. Most often used are the terms "within-brand" and "within-dimension." To reduce confusion, the following can be equated. Interdimensional search (Payne, 1975), choice by processing brands (Bettman and Jacoby, 1975), type "2" search (this paper), and within-brand search all denote the successive acquisition of some set of information dimensions on a single brand or alternative. In contrast, intradimensional search (Payne, 1975), choice by processing attributes (Bettman and Jacoby, 1975), type "3" search (this paper), and within-dimension search all denote the successive comparison of some set of brands or alternatives on a single dimension of information.] Path analysis suggests search strategy to be partially determined by brand loyalty and consumption frequency. These findings are considered in terms of a general "coding" explanation and placed within a four-stage paradigm developed for guiding, integrating and interpreting consumer information processing research.

INTRODUCTION

The three-fold purpose of this paper is to describe: (1) a methodology developed for studying consumer information acquisition as a dynamic process; (2) a paradigm for guiding research and integrating findings from consumer information processing investigations, and (3) a pilot study which employed this methodology and paradigm. Our objective in so structuring this paper is to have it serve as a focal statement around which a series of completed, on-going, and planned studies can be positioned and interpreted.

A PROCESS METHODOLOGY FOR STUDYING CONSUMER INFORMATION ACQUISITION

Background

The present methodology evolved from a programmatic series of investigations which, at the outset, examined the implications of information load in the consumer context. By 1974, we had conducted more than 20 investigations directed at different aspects of consumer information processing. As detailed elsewhere (Jacoby, 1975), despite our extensions and improvements, we became increasingly aware of limitations in our approach. Three things in particular disturbed us. First, most of our early studies employed hypothetical rather than actual brands and product information. Second, operating within the tradition of manipulating levels of variables in tight experimental designs, subjects were provided with fixed amounts of product information (e.g., 12 items of information for each of 8 hypothetical brands) and required to arrive at purchase decisions based only upon the information provided -- no more, no less. In contrast, subjects in actual shopping situations are free to select as much or as little information as they wish prior to making purchase decisions. Finally, the designs employed were basically static in nature, focusing on the outcome of consumer decision processes rather than upon the dynamic nature and content of the process itself.

By late 1972, while our main research thrust continued to reflect these problems, we began developing more satisfactory process oriented approaches. Two such investigations were conducted during the 1972-73 academic year (Berning and Jacoby, 1974; Jacoby, Szybillo, and Busato-Schach, 1974). Further methodological developments ensued and culminated in a proposal submitted to the National Science Foundation in mid-1973.

This proposal was funded effective June 1974 -- with a novel twist. In an attempt to maximize external validity and practical utility, a special Research Advisory Board was established in conjunction with the grant and was composed so that various interests were represented thereon. James Turner (of Swankin & Turner), a prominent public interest attorney in Washington, represented the public sector. The business sector was represented by Norman Pugh of Sears, Roebuck & Co. Government interests were represented by Richard Herzog of the Federal Trade Commission, Evelyn Gordon of the Food and Drug Administration, and Alden Manchester of the U.S. Department of Agriculture. William Wilkie, a marketing professor who had worked at the FTC and was familiar with public policy issues relating to consumer information processing (cf. Wilkie, 1974; Wilkie and Gardner, 1974), and Howard Fromkin, a social psychologist known for his expertise in behavioral science research methodology (cf. Fromkin and Streufert, 1976), represented the academic perspective. George Brosseau, of the National Science Foundation, participated as an ex officio member.

The Research Advisory Board convened on July 11, 1974 and devoted most of the day to discussing the methodology described in the proposal as well as several alternative procedures. The most desirable features of each were fused into a single approach which, in general, represented an integration of the Berning and Jacoby (1974) and Jacoby, Szybillo, and Busato-Schach (1974) methodologies.

Methodology

Within the context of our program, several different process methodologies have been developed and used to examine a variety of consumer information processing issues (cf. Jacoby, 1975). In all instances, we have focused on pre-purchase information acquisition which, it should be noted, can and often does differ in nature and content from pre-consumption and predisposition information acquisition (cf. Jacoby, Berning, and Dietvorst, 1975). For example, price information may be used prior to making a purchase, but be ignored prior to usage of the product. Suggested oven temperature may be ignored during pre-purchase deliberations, but be of vital importance when the consumer begins to cook. Whether an empty aerosol container can be safely placed in an incinerator might be irrelevant during pre-purchase and pre-usage information acquisition but become an important item of information prior to disposal.

On a conceptual level, the most basic form of the methodology involves presenting the subject with an array of product information and permitting him to acquire as much or as little of this information as he wishes prior to arriving at a purchase decision. The actual form that this array may take can and does vary. In the investigation to be described below, as well as in several other investigations now in progress, the format adopted consists of presenting information via a specially designed information display board (IDB) which presents information arranged in the form of a p x q matrix. The various brands (i.e., purchase alternatives) appear listed across the top of the matrix while the types of information available (i.e., information dimensions, e.g., price, net weight, ingredients, etc.) are listed down the left- and right-hand sides of the matrix. Thus, each column of information in the matrix represents a single brand while each row represents a dimension of information (e.g., price) across all the brands presented. Each cell in the matrix contains the actual "value" of the information (cf. Jacoby, 1974, p. 117-18 and Jacoby, 1975, p. 206, for a more detailed discussion of this nomenclature) for the particular brand x information dimension crossing which forms that cell. For example, crossing "Crest" with "net weight" would yield the value "S ounces," assuming this was the size tube being studied. In the most frequently employed variant of the methodology, the specific brand names and information dimensions being used are plainly visible at the outset, while the actual values which result from crossing each brand with each information dimension are concealed.

Subjects are placed into a simulated shopping situation and instructed to purchase one brand from among the set of brands provided in the information display. They are informed that, using the IDB, they are to "shop" for the product as they usually do, taking as much or as little time and acquiring as much or as little information from the matrix as they desire in order to arrive at their purchase decision. Moreover, they can start at and move to any point they wish in the matrix. The instructions also indicate that they need not acquire any information from the matrix and can "purchase" a brand simply on the basis of brand name, if this is how they usually purchase this product. When information values are selected, a process record is formed of the individual's information acquisition behavior which can then be analyzed in terms of the depth (i.e., extent), sequence, and content of the information acquired.

In order to insure that a degree of realistic motivation is present, subjects are told in advance that they should make their decision count since they will receive five "304-ofF' coupons valid only for the specific brand which they select and which can be used in their next five purchases of this product. This cents-off procedure is used in preference to providing subjects with the actual product, because it reduces the likelihood of individuals selecting the highest priced brand even when this is not what they normally would do.

The basic approach is quite flexible and capable of being modified in a variety of ways in order to address numerous theoretical and applied questions. Two modifications are of particular interest here. First, brand names for established products often serve as information "chunks" (cf. Jacoby, Szybillo, and Busato-Schach, 1974; Simon, 1974), thereby conveying information regarding other attributes of the brand in question. Thus, given the availability of brand name (e.g., "Pledge" furniture polish), subjects often need not behaviorally access certain information values because they already have these in memory.(e.g., type of scent: lemon). Accordingly, while the pilot investigation described below makes brand names available for all subjects, some of our earlier and current investigations also use numbers or letters in lieu of actual brand names.

Second, the pilot investigation contains a-modification which attempts to approximate reality by introducing a memory load factor in information acquisition. In the basic procedure described above, once information is accessed, it remains visible and available throughout the remainder of the pre-purchase information acquisition decision task. Adding "depth" to each cell on the matrix permits examination of the re-accessing of information values which have not entered long term memory, but have been erased from short term memory. Specifically, this involves constructing the IDB so that each cell in the matrix consists of a three-dimensional pocket. A set of ten identical information value card is placed into each pocket and arranged so that the backs (i.e., non-information bearing side) of the cards face the subject. A subject interested in acquiring a specific piece of information (e.g., price for Crest) goes to the cell in the matrix containing these values, removes one of the cards, and looks at the information value appearing on the reverse side (e.g., S34). Given that each pocket in the IDB contains ten identical cards, the individual may behaviorally re-access information up to ten times, should he so desire. Once a card is acquired and the information looked at, the subject must place it face downward in a collection tray (thus yielding a behavioral process record of information acquisition) before he may acquire another card. Since the subject is always confronted by a display in which all the specific information values are concealed, a memory load factor is imposed.

Five Characteristics of the Methodology

Consumer information processing is, by definition, a dynamic phenomenon. Yet virtually all attempts to measure information processing in the consumer context have employed static, cross-section approaches, usually involving verbal reports collected from respondents at a time far removed from the actual information processing and decision-making itself. In contrast, our methodology permits us to capture the depth, content, and particularly the sequence of information acquisition, as it occurs, so that it can later be examined in terms of the dynamic process that it is.

A related characteristic is that it focuses on behavior, not verbal reports. In other words, it focuses on what people actually do, not what they say they do. As many a social scientist will readily testify, behavior and verbal reports of behavior can be (and often are) completely unrelated. In the several studies we have conducted thus far, we generally find only a +.4 (ca.) correlation between the information dimensions that a respondent tells us he refers to and the information values that he actually selects. In our opinion, verbal reports -- particularly those involving recall of long forgotten and/or relatively trivial events (e.g., "Tell me what kinds of information you looked at when you bought your last package of laundry detergent."), and those requesting a description of likely behavior in the future (e.g., "How likely are you to use nutritional information when you buy frankfurters?") -- are replete with biasing characteristics and are less valid predictors of actual in-vivo behavior than are data derived from behavior manifested in the simulated shopping situation just described.

It should also be noted that our methodology is descriptive, not prescriptive. It makes no value judgments; it says nothing about what kind of information should be provided. The procedure simply addresses the issue of just which information is used. In a very real sense, the concern is not so much with the nature of the information provided by the source, as it is with the impact that this information has on the receiver (see Jacoby, 1974, p. 103-106 for an extended discussion of this issue, and Jacoby and Small, 1975, for one concrete application of this approach).

Another aspect of our approach is its attempt to inject a degree of realistic motivation in the decision-making task. While the task is only a simulation, the consequences are real. Something of value to the subject is riding on his decision. If he doesn't like Post Toasties, he had better not select that brand because part of his payment for participating in the study will come in the form of $1.50 in coupons (i.e., 5 x 50c-off) good only toward the purchase of that brand. While our subjects are also paid $3.50 in cash, this payment functions primarily to motivate them to participate in the study. The additional $1.50 in cents-off coupons is designed to motivate them to take the decision task seriously, that is, to behave realistically while actually engaging in the decision-making task itself.

Finally, the methodology is highly flexible. Two aspects of this flexibility were described above. Several other modifications have been utilized; yet others are currently being developed and evaluated. These will be described in forthcoming papers (e.g., Chestnut and Jacoby, 1975b).

A RESEARCH PARADIGM FOR THE ANALYSIS AND INTERPRETATION OF PROCESS DATA

The analysis of process data is appealing from the standpoint of theory. As Simon (1957) notes, only through a complete understanding of the events leading up to a decision can we hope to explain the "basis" for choice. Such understanding, however, is not easily obtained (cf. Lanzetta, 1963). As methodological complexity increases (i.e., in going from static to process measurement), the quantity and nature of the resultant data can begin to exceed the investigator's limited statistical and conceptual models. It becomes necessary to limit research aims and proceed via a logical schema of analysis.

Fortunately, the data analysis problem has, to some extent, already been dealt with in the information processing literature. In the mathematical modeling of decision policies, investigators have typically engaged in a three-step procedure. As reviewed by Dudycha (1970) and Slovic and Lichtenstein (1971), research has attempted: first, to "capture" the decision strategy (i.e., to isolate the salient characteristics of an individual's information processing); then, to "cluster" decision-makers (i.e., grouping individuals according to the homogeneity of their strategies); and, finally, to assess how the clusters or decision strategies relate to some criterion of decision quality (e.g., the "achievement index" of the Brunswik Lens Model). The present paper proposes an extended, four stage version of this paradigm for the study of consumer information acquisition (cf. Figure 1). Each stage focuses on a different set of variables (represented in rectangles). Attempting to relate one set to another -- making transitions between stages -- highlights important research issues (represented in circles).

FIGURE 1

FOUR STAGE PARADIGM FOR THE ANALYSIS AND INTERPRETATION OF PROCESS DATA

Stage I encompasses individual differences (e.g., motivation, uncertainty, perceived risk) and environmental variables (e.g., time pressure, amount of information, distraction) which arouse and direct search behavior. The research issue involved in the transition from Stage I to Stage II is one of establishing causal relationships between specific Stage I variables and the salient characteristics of search strategy. Studies seeking to find such relationships have focused almost exclusively on environmental variables. In the Newell and Simon tradition (cf. Newell and Simon, 1972), the quality of the "task environment" has been viewed as a major determinant of problem-solving or search activities. Although the environment contains a seemingly infinite number of characteristics, investigators (e.g., Lussier and Olshavsky, 1974, p. 3) have typically maintained that only those characteristics directly relevant to acquisition need be explored. In the realm of consumer behavior, this has led to a recent emphasis upon the environmental phenomena of "information overload" (e.g., Jacoby, Szybillo, and Busato-Schach, 1974; Lussier and Olshavsky, 1974; Payne, 1975, 1976) and "information costs" (e.g., Swan, 1972; Wright, 1974; Winter, 1975; Chestnut and Jacoby, 1975b). Findings almost universally support the impact of such variables upon the nature of consumer information processing. Given the importance of the "task environment" in directing information acquisition (search) behavior, an emerging trend has been to suggest redesigning the consumer's environment so as to optimize the ease and quality of information acquisition (cf. Jacoby, 1974; Bettman, 1975; Russo, 1975).

A factor overlooked in a majority of these studies concerns the potential influence of individual difference variables -- what Newell and Simon would refer to as the problem-solver's "inner-environment." How might motivation, personality, attitudes, or experience affect information acquisition processes? Although investigators have recognized the importance of the individual in terms of "processing" limitations (e.g., Bettman, 1975), few have gone on to study the individual's role in determining the nature of acquisition. Notable exceptions are Swan's (1972) exploration of the effect of product expectations and Weigl's (1975) findings with regard to perceived risk. The pilot investigation described below studied a number of such individual difference variables. Due to space limitations, these are considered in a companion paper (Jacoby, Chestnut, Fisher, and Weigl, 1975).

Stage II involves a consideration of those variables which "capture" (i.e., in some way describe) the salient characteristics of search. Three categories of such variables are depth (i.e., amount of information acquired), sequence (i.e., temporal pattern of information acquired), and content (i.e., specific nature of the information acquired, e.g., price vs. nutrient information).

Given the data obtained from using a "process" methodology, numerous variables and combinative statistics can be generated. Table 1 organizes a variety of such measures in terms of depth, sequence, and content. (N.B. Brief labels are provided in Table 1; more detailed descriptions of selected measures appear in the .Appendix.) Some measures are undoubtedly highly correlated with others and could be dispensed with. All are included to emphasize the variety of measurement possibilities. Many have been developed in the course of the present investigation; others are derived from related research independent of our own.

TABLE 1

INFORMATION ACQUISITION VARIABLES. (STAGE II)

The major research opportunity provided at this juncture suspends interest in explaining search until decision-makers have been satisfactorily "clustered" into homogeneous groups according to their information acquisition strategies. This involves moving to Stage III and raises an important issue. What exactly constitutes a "strategy" of information acquisition? According to our paradigm, strategies reflect some combination of Stage II variables. The problem thus becomes: which combination?

In one of several companion pieces to this paper, Bettman and Jacoby (1975) analyze the pilot data described below by clustering individuals via a sequence definition of strategy. /heir primary grouping statistics are the Same Brand Index (SBI) and Same Attribute Index (SAI; see their Figure 1). Russo and Rosen's (1975) XYX brand choice pattern [i.e., Bettman and Jacoby's "End Comparison Phase") is entered at selected nodes to achieve a finer gradation of pattern. Once having clustered or defined these strategies, they then proceed backward to Stage I and demonstrate a number of viable relationships (e.g., brand loyal consumers engage in more "Choice by Brand Processing").

Payne (1975) clusters decision-making strategies along somewhat different criteria. Although he too employs the within-brand/within-dimension distinction (i.e., inter- vs. intra-dimensional search), a new factor is added: variance in the depth of search per alternative. This provides a slightly different classification rationale and, consequently, new strategies are isolated (e.g., Tversky's "elimination by aspects" and Wright's "conjunctive processing").

Are these the only clustering algorithms of value? Probably not. Table 1 lists a variety of clustering criteria yet to be exploited. Since the actual quality of a decision is more likely to be a function of selected depth and content features, definitions of "strategy" should go beyond mere patterning if the fertility of process data is not to be ignored. Even if sequence were the only salient characteristic of search, indices such as SAI and SBI "capture" a minimal amount of sequence information. Both statistics are defined on an ordering of two consecutively selected information values and would seem inadequate for reflecting the process of any search of substantial duration or complexity. Clearly, the conceptualization of "strategy'' needs to be expanded and at least some of the many other criteria for clustering evaluated. Alternative models for defining strategies need to be developed and the interrelationships of these clustering algorithms examined. This should be done with an eye toward "converging operationalism" (cf. Garner, Hake, and Eriksen, 1956). Otherwise, the proliferation of parallel operational definitions will severely impede the accumulation and integration of knowledge. The greatest negative impact would be on any attempt to make progress in terms of Stage IV in our paradigm.

Stage IV consists of variables which act as standards for evaluating the "quality" of information acquisition in terms of the "accuracy" of the resultant decision. As noted in Figure 1, either "optimizing" or "satisfying" standards could be used in such an evaluation (cf. March and Simon, 1958, p. 141). The large scale research program presently underway collects data using both these standards.

The pilot investigation described below has furnished the basis for considerable exploration of the paradigm just described. One thrust has focused on considering the interrelationships between and among stages. For example, Jacoby, Chestnut, Fisher, and Weigl (1975) examined the relatively neglected issue of how individual difference variables (Stage I) relate to the nature of information search (Stage II). Greater effort, however, has been devoted to exploring appropriate indices for clustering search strategies (Stage III). In developing our process methodologies, it quickly became apparent that the statistics typically employed in consumer research were inadequate for describing and analyzing the data we would be collecting. To exert better leverage on this issue, two independent efforts were mounted to develop suitable process statistics. In one effort, because of his directly relevant earlier work (cf. Bettman, 1970, 1971, 1974a, 1974b), James Bettman was retained as a consultant and given data from the pilot investigation to analyze. The analytical and interpretive work appearing in Bettman and Jacoby (1975) is the output from this effort. Because of space limitations, only one portion of the second effort is described here. Companion pieces (Chestnut and Jacoby, 1975a; Jacoby, Chestnut, Fisher, and Weigl, 1975) provide a more complete accounting.

A PILOT INVESTIGATION

Subjects and Test Product

The study was conducted during the fall of 1974 as a pilot for a larger investigation now in progress. Subjects were 60 Purdue University undergraduates (52 females, 8 males) whose participation met a course requirement. These particular subjects received no other incentives (e.g., cents-off coupons). Cold breakfast cereal, a frequently purchased non-durable, was the test product.

Procedure and Apparatus

Subjects, who were interviewed individually, first responded to items assessing product importance, frequency of purchase and consumption, "evoked set," perceived quality differences among brands, perceived healthfulness of cold breakfast cereals, and several other product related indices. Following this, they were introduced to the IDB which was divided into 16 columns (purchase alternatives, i.e., brands) and 35 rows (information dimensions). Purchase alternatives were the 16 brands of cold breakfast cereal having the largest national market shares. Each brand was randomly assigned to a purchase alternative column; each brand name appeared at the top of the column to which it had been assigned.

Information dimensions rows consisted of 35 types of information found on most cold cereal package panels (e.g., price, calories per serving, net weight, etc. These information dimensions appeared as three alphabetically ordered columns down the center and along the left- and right-hand sides of the IDB. There were thus (35 x 16) 560 cells, each with 10 identical cards specifying an information x alternative crossing. In instances where no information existed regarding a particular information dimension for a given brand, the words "No information" were printed on the reverse side of the cards in that cell.

Subjects were instructed to shop for the product, as described above. The cards each subject deposited in the collection tray provided a behavioral process record (in terms of the depth, sequence, and content) of his pre-purchase information acquisition behavior. Response latency (from the time the first card was touched to the moment a subject announced his purchase decision) was also recorded. Upon completion of the IDB choice task, subjects responded to additional items assessing choice certainty and satisfaction, perceived risk, subjective states experienced during the task (e.g., confusion), recall of information dimensions actually used in information search, rank ordering of the various brands and information dimensions in terms of importance and preferences, and other items relating to the task, the decision making process, and cold breakfast cereal. Finally, a series of open ended debriefing questions sought to determine if demand characteristics were present, and how well subjects comprehended the interviewer's instructions.

Analysis

The sequence of acquiring information values can be characterized in a number of ways (cf. Table 1). However, in the relatively few attempts to quantify information acquisition patterns (Bettman and Jacoby, 1975; Payne, 1976), one system of nominal classification has predominated. This approach might be labeled "analysis of transitions" and rests on a consideration of the change in brand and dimension from one acquired information value to the next (i.e., from n to n+1). Focusing upon this relatively simple two item sequence, four possible types of transitions emerge and are distinguished by varying combinations of brand/dimension similarity (see Table 1, IIB, 1-4). Given these basic definitions, a process record of n values can be expressed as a vector of n-1 transitions with each element in the vector taking on the value 1, 2, 5, or 4, depending upon the nature of the transition. If vectors vary greatly in length (as in the pilot study) and cannot be adequately compared in terms of overall structure, it becomes necessary to aggregate and describe the pattern of search in terms of the proportion of the vector devoted to each type of transition. Such proportions are then the "observed" probabilities of a given brand/dimension transition in the course of search. Although "analysis of transitions" can be made considerably more complex by accommodating sequences of more than two cards (cf. Table 1, IIc, 5-9), the present investigation restricts itself to the simplest case in order to enhance the clarity of our initial presentation. A more complete analysis is provided in Chestnut and Jacoby (1975a).

Bettman and Jacoby (197S) use a normalized probability measure to cluster subjects into nominal classifications of "strategy." In contrast, the present approach utilizes an interval-like description of search pattern. To achieve this description, five post-hoc models of expected transition probabilities were hypothesized (cf. Table 2). The rationale behind each is as follows. Type 1 transition probabilities were set to zero in all models, since they failed to appear in any great quantity in the observed data (i.e., less than 1% of all transitions). Type 2 and 3 transition probabilities were arbitrarily and systematically varied in order to reflect increasing dominance of either within-brand or within-dimension search (cf. Models I and II, respectively). Type 4 probabilities were increased in two models to indicate search being expanded to include a larger set of brands or dimensions (cf. Models III and IV, respectively). Model V reflects an equal proportion of within-brand (Type 2) and within-dimension (type S) search. Given such expected probabilities, Chi-square statistics could be computed to describe the degree to which each model "fits" the observed data.

TABLE 2

DEFINITION OF MODELS BY EXPECTED TRANSITION PROBABILITIES

Results

Twelve subjects made a purchase decision based only on the available 16 brand names; another subject acquired only one information value. Analysis of the "goodness of fit" between "expected" transition models and "observed" transition probabilities generated five Chi-square statistics for the remaining 47 subjects. Forty-six subjects were categorized under the model for which they had the lowest Chi-square value (cf. Table 3); one subject, whose search consisted of a single type 4 transition (i.e., two values) was not adequately modeled. Mean Chi-square values within each classification were low, indicating the appropriateness of the intuitively chosen probability levels. As search length increased, the model's approximation of the observed probabilities generally improved.

Although not completely similar (for a detailed comparison see Chestnut and Jacoby, 1975a), the clustering results are comparable to Bettman and Jacoby's (1975) discrimination net findings. Combining models I with III and II with IV, we find equal numbers of subjects (17) engaging in either predominantly within-brand (CPB) search or predominantly between-brand (CPA) search. The remaining 12 subjects appear to be just as likely to acquire information through a type 2 transition as they are a type 3 transition (Bettman and Jacoby's CFP strategy).

TABLE 3

MEAN CHI-SQUARE VALUES AND NUMBER OF SUBJECTS CLASSIFIED WITHIN EACH MODEL

Given the rather clear-cut search patterns that emerge, it seems reasonable to expect causal factors to be operating. Jacoby, Chestnut, Fisher, and Weigl (1975) present data bearing on a number of potentially influential individual difference variables. Two, in particular, seem promising: brand loyalty and consumption frequency. Table 4 presents a path analysis (cf. Van de Geer, 1974) of the impact of these variables upon search patterning. Five multiple regressions were computed utilizing the Chi-square index of model fit as the criterion, with consumption frequency (operationally defined by a 7-interval scale ranging from "more than once a day" to "less often than once every two weeks") and brand loyalty (defined in terms of the percent-of-purchase devoted to the leading brand) as the predictors. Since the correlation between predictors was -.06 (n=48), they were assumed to act independently. Path coefficients could therefore be interpreted in terms of a causal model. Because subjects in the "light search" category were inadequately modeled by the Chi-square statistic (cf. Table 3) and could obscure the results, searches of less than seven values were excluded from the analysis. This reduced sample size to 32.

TABLE 4

MULTIPLE CORRELATIONS AND PATH COEFFICIENTS FOR A PREDICTION OF MODEL FIT

The resultant multiple correlations were moderate, ranging in significance from .01 to .06. Path coefficients indicated three types of predictor influence. Within-brand (i.e., type 2 search, models I and III) was best predicted by consumption frequency; specifically, the more often the product was consumed, the greater the tendency to engage in type 2 search. The remaining models were determined by a combination of consumption frequency and brand loyalty. Between-brand comparison (i.e., type 3 search, models II and IV) was more likely with non-loyal consumers having a low frequency of consumption. Complex strategies (model V), on the other hand, were employed by nonloyal consumers having---a high frequency of consumption. Although these relationships are significant, the unexplained variance attributable to latent constructs not made explicit in the path analysis should be considered. Approximately 70 to 80 percent of the variance in model fit remains unexplained. However, despite sample limitations and self-report scaling of the predictor variables, the results seem promising and warrant further examination (see Jacoby, Chestnut, Fisher, and Weigl, 1975).

DISCUSSION AND CONCLUSIONS

Quite clearly, there appear to be distinct differences in the types of information acquisition strategies consumers employ. Different consumers use different information acquisition strategies. Whether these same patterns emerge with other kinds of products and consumers, as well as the causal factors underlying these patterns, is currently being examined. Data regarding the relationship of these patterns to criteria of decision quality (cf. Stage IV in our paradigm) are also being collected.

Another apparently inviolate generalization emerging from this and other investigations (e.g., Jacoby, Szybillo, and Busato-Schach, 1974; Olson and Jacoby, 1972; Payne, 1975) is that consumers typically acquire only a small proportion of the available information. Utilizing a 3S x 16 matrix display of package information relevant to breakfast cereal, we find a median search length of less than seven information values. Of the 560 values available, half the sample found it necessary to acquire less than 2% of the information available in the display prior to making their purchase decision. Even if we correct for the fact that brand names were exposed (i.e., could be treated as an acquired information dimension) and that some of the dimensions were more relevant to pre-usage and predisposition decision making (thereby reducing the matrix to a conservative estimate of 16 pre-purchase dimensions such as price, net weight, etc.), a selective exposure of less than 9% can be estimated. Given that the methodology employed, if anything, should have increased search length (i.e., through potential demand characteristics and the minimization of acquisition costs relative to such costs in an actual supermarket setting), this degree of selectivity is striking. It is of course possible that using brand name as an information chunk (cf. Jacoby, Szybillo and Busato-Schach, 1974; Bettman and Jacoby, 1975) may be partially responsible for some of this selectivity in this investigation, and ongoing research is exploring this question.

When consumers do acquire package information, however, an interesting relationship appears through an analysis of Stage I individual difference variables and their effect upon selected sequence statistics at Stage II. Specifically, consumption frequency and brand loyalty are seen to determine the relative proportions of type 2 and type 3 transitions. One explanation for this relationship might be framed in terms of a "stimulus re-coding" interpretation of the information acquisition process. If we assume that the type of stimulus recoding employed in the search task is a function of the nature of the information stored in long-term memory (cf. Massaro, 1975, p. 249), it would seem only reasonable to expect a relationship between selected purchasing characteristics (i.e., as correlates of knowledge) and the pattern of information acquisition. That is, a person who consumes the product quite frequently is more likely to have stored information about the properties of the product and, thus, employ a within-brand recoding which would input values of information in terms of well-known properties related to a single brand. In contrast, a person who is non-loyal could be expected to be more aware of the many different brands present in the product class (cf. Jacoby and Hillemeyer, as cited on p. 206 of Jacoby, 1975) and, thus, resort to a within-dimension recoding which would input values of information in terms of well-known brands that emerge successful from a consideration of a single dimension. Much as in the paired-associate analysis of mnemonic techniques (cf. Paivio, 1971), the subject is utilizing a conceptual peg of personal salience on which to hang (i.e., remember) a specific item of information. Such an interpretation fits the empirical findings of the path analysis and would seem to represent a worthwhile topic for further research.

With regard to our analytic method, the Chi-square modeling approach to clustering subjects is but one of many ways to examine information acquisition patterns. Bettman and Jacoby's decision net approach is another. Payne's reliance on the variance in the depth of search per alternative is yet another. Probably the most important distinction between these approaches is that the Chi-square method generates interval-like data, thereby providing a more satisfying means for relating transition probabilities to other variables. Yet other clustering approaches are described and employed in Chestnut and Jacoby (1975a).

Regardless of which clustering algorithm is used, arbitrary assumptions are usually involved (cf. Cormack, 1971). Although these assumptions may describe the data at hand, they need not reflect the true nature of reality. The stability of derived clusters and validity of inferences drawn from their analysis need to be further established. As indicated, ongoing investigations are focusing on cross-validation and on the extension of algorithms to include depth and content criteria. If the same relatively distinct clusters emerge, empirical description can give way to a more theoretical analysis since, as Hartigan (1975) notes, "clear-cut and compelling clusters...require an explanation of their existence and so promote the development of theories...'' (p. 7).

A limitation common to both the Chi-square modeling and decision net approaches is that they reduce the dynamic information acquisition process to a static representation. Now that consumer research is beginning to develop methodologies which capture the dynamic character of the information acquisition process, a set of analytical statistics which preserve this character is needed. One such approach is described in Chestnut and Jacoby (1975a).

A final point should be noted in regard to the place of verbal protocols in our methodology. As in many of the quantitative modeling investigations (cf. Wright, 1972), our subjects are questioned about their search strategy only after the completion of their decision. This is in marked contrast to related work by others (e.g., Bettman, 1971; Payne, 1975; and Lussier and Olshavsky, 1974) in which verbal protocols of search are collected concurrent with behavioral simulation. The reasons we adopted our post hoc procedure are threefold. First, our intent has been to develop a minimally disruptive and reactive approach to measuring the actual acquisition process. By not forcing the subject to verbalize or provide rationale behind his acquisition behavior, we lessen the possibility that his behavior and protocols will reflect biasing demand characteristics and leave him free to interact with the information display in a manner more akin to actual package information search. Second, because of the unwieldy volume of data that is generated per subject, methodologies employing verbal protocol tracing procedures have typically been forced to consider very small numbers of subjects (e.g., n=6 and 12 in Payne, 1975; n=2 in Bettman, 1970). In contrast, the present study utilized 60 subjects; our ongoing investigation involves a sample size in excess of 750. Finally, it is an open question as to exactly how much of what a subject verbalizes about the purchase of a relatively minor, frequently purchased non-durable such as breakfast cereal is actually clear and meaningful to the consumer himself. In contrast, decision criteria and information acquisition and processing strategies with respect to infrequently purchased and relatively important purchase decisions (e.g., purchase of a home, renting an apartment) are likely to be more cognitively focused and clear to the decision maker. Verbal protocols should be much more meaningful and revealing in these instances. It should be obvious, however, that studies are needed which comparatively test the impact of collecting verbal reports concurrently vs. post hoc.

Above and beyond the few concrete "findings" presented, we view this paper as making several contributions. In particular, we point to the description of our process methodology, the presentation of our four stage research paradigm, and the development of the Chi-square modeling approach. Hopefully, another contribution will be the heuristic influence this paper will have on future research regarding consumer information processing.

APPENDIX

DEFINITIONS OF SELECTED STAGE II VARIABLES

I.E. Mean rated importance of all values chosen: Given numerical importance ratings (typically obtained via a standard Likert scaling) for all dimensions and brands, each information value choice can be represented by the product of the two ratings. An average of these products over some duration of search is then computed.

I.F. Mean rank importance of brands consulted: If brands are rank ordered, an average rank of brands consulted in search can be computed.

I.H. Brand redundancy: The number of times a subject returns to a brand previously examined.

I.J. Brand Run index: Defining a "brand run" as any sequence of two or more values on the same brand, this index can be defined as the number of values chosen minus the number of brand runs divided by the number of values chosen minus the number of brands chosen.

I.L. Proportion of acquired values devoted to brand chosen: Given that the subject is involved in a decision-making task resulting in the purchase of a single brand, the proportion of search devoted to the brand eventually purchased can be calculated.

I.M. Discrepancy between actual and "optimal" search: Given ratings of the number of brands thought to be "acceptable'' and the number of dimensions thought to be "important for purchase", an estimate of "optimal" search is simply the product of the two ratings. The above index calculates the discrepancy by subtracting the optimal search estimate from the actual number of information values chosen.

II.C.1. Proportion of transition changes: The frequency of change in type ot transition divided by the number of transitions in the transition vector.

II.C.2. Analysis by segments: Search length can be standardized by dividing the transition vector into a unit length (halves, thirds, quarters, etc.). Given this partitioning, a predominant transition per unit length can be designated and an analysis of the resulting equal-length vectors attempted.

II.C.3. XYX brand selection: This is simply the frequency of any XYX pattern in the vector of brand choice.

II.D.3. Vector of brand choice: This is a 1 x q vector where the column-order is the number of values chosen and each element is a digit representing a specific brand.

II.D.4. Vector of dimension choice: This is a 1 x q vector where the column-order is again the number of values chosen and each element is a digit representing a specific dimension of information.

II.D.5. Raw data matrix: This a 2 x q matrix where the column-order is the number of information values chosen and the two row vectors are the vectors of brand and dimension choice.

III.D. Behaviorally derived importance weight: The sum of the position ranks for choices relating to a given dimension divided by the number of times the dimension was chosen. This can then be standardized across different search lengths by dividing by the number of information values chosen. For example, in a 10 value search, the 1st and 5th value might relate to price. An index of price information importance would then be calculated as follows: [(1+5) ) 2] )10 = .3

III.E. Percent of intrinsic information in search: Intrinsic information is defined as any information dimension for which a change in value would result in a direct change in the product's physical composition (cf. Szybillo and Jacoby, 1974). Calories, therefore, would be classified as intrinsic whereas price might be thought of as extrinsic information. Given a classification of each dimension in terms of these two categories, the relative proportion of search devoted to each category can be calculated.

III.G. Clustering based upon intercorrelation patterns among dimensions acquired: Given a p x q matrix of subjects by dimensions, where individual entries represent the number of times a given subject consulted a given information dimension, a Q-factor analysis can be used to cluster subjects by general type of information chosen.

REFERENCES

Carol A. Berning and Jacob Jacoby, "Patterns of Information Acquisition in New Product Purchases," Journal of Consumer Research, 1(September, 1974), 18-22.

James R. Bettman, "Information Processing Models of Consumer Behavior," Journal of Marketing Research, 7(August, 1970), 370-76.

James R. Bettman, "The Structure of Consumer Choice Processes,'' Journal of Marketing Research, 8(November, 1971), 465-71.

James R. Bettman, "Toward a Statistics for Consumer Decision Net Models," Journal of Consumer Research, 1(1974), 71-78.

James R. Bettman, "Decision Net Models of Buyer Information Processing and Choice: Findings, Problems, and Prospects," in G. D. Hughes and M. L. Ray, eds., Buyer/ Consumer Information Processing (Chapel Hill, North Carolina: University of North Carolina Press, 1974, 59-74).

James R. Bettman, "Issues in Designing Consumer Information Environments," Working Paper Series, Center for Marketing Studies, No. 32 (UCLA, July, 1975).

James R. Bettman and Jacob Jacoby, "Patterns of Processing in Consumer Information Acquisition," in B. B. Anderson (ed)., Advances in Consumer Research,Vol.3 (Proceedings of the Sixth Annual Conference of the Association for Consumer Research, Cincinnati, Ohio, October 50-November 2, t975).

Robert W. Chestnut and Jacob Jacoby, "Analytical Techniques for Examining Process Data," Purdue Papers in Consumer Psychology, No. 154, 1975a.

Robert W. Chestnut and Jacob Jacoby, "The Impact of 'Time Costs' upon the Acquisition of Package Information,'' Purdue Papers in Consumer Psychology, No. 155, 1975b.

R. M. Cormack, "A Review of Classification," Journal of the Royal Statistical Society, A134 (1971), 321-67.

Arthur L. Dudycha, "A Monte Carlo Evaluation of JAN: A Technique for Capturing and Clustering Raters Policies," Organizational Behavior and Human Performance, 5(5, 1970), 501-516.

Howard L. Fromkin and Siegfried Streufert, "Laboratory Experimentation," in M.D. Dunnette, ed., The Handbook of Industrial and Organizational Psychology (Chicago: Rand-McNally, 1976).

Wendell R. Garner, Harold W. Hake, and Charles W. Ericksen, "Operationism and the Concept of Perception," Psychological Review, 65(May, 1956), 149-59.

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

Jacob Jacoby, "Consumer Reaction to Information Displays: Packaging and Advertising," in S.F. Divita, ed., Advertising and the Public Interest (Chicago: American Marketing Association, 1974, 101-118).

Jacob Jacoby, "Perspectives on a Consumer Information Processing Research Program," Communication Research, 2(3, 1975), 203-15.

Jacob Jacoby, Carol Kohn Berning, and Thomas F. Dietvorst, "What about Disposition?," Purdue Papers in Consumer Psychology, No. 152, 1975.

Jacob Jacoby, Robert W. Chestnut, William Fisher, and Karl C. Weigl, "Individual Difference Factors in Consumer Information Acquisition," Purdue Papers in Consumer Psychology, No. 155, 1975.

Jacob Jacoby and Constance B. Small, "The FDA's Approach to Defining Misleading Advertising," Journal of Marketing, 39(October, 1975), 65-68.

Jacob Jacoby, George J. Szybillo, and Jacqueline Busato-Schach, "Information Acquisition Behavior in Brand Choice Situations," Purdue Papers in Consumer Psychology, No. 140, 1974. Provisionally accepted for publication in the Journal of Consumer Research.

John T. Lanzetta, "Information Acquisition in Decision Making," in O. J. Harvey, ed., Motivation and Social Interaction: Cognitive Determinants (New York: Ronald Press, 1963, 239-65.

Denis A Lussier and Richard W. Olshavsky, "An Information Processing Approach to Individual Brand Choice Behavior,'' paper presented at the ORSA/TIMS Joint National Meeting, San Juan, Puerto Rico (October, 1974).

James G. March and Herbert A. Simon, Organizations (New York: Wiley, 1958).

Dominic W. Massaro, Experimental Psychology and Information Processing (Chicago: Rand McNally, 1975).

Allen Newell and Herbert A. Simon, Human Problem Solving (Englewood Cliffs, N.J.: Prentice-Hall, 1973).

Jerry C. Olson and Jacob Jacoby, "Cue Utilization in the Quality Perception Process," in M. Venkatesan, ed., Proceedings, Third Annual Conference of the Association for Consumer Research (1972), 167-79.

Allan Paivio, Imagery and Verbal Processes (New York: Holt, Rinehart & Winston, 1971).

John W. Payne, "Heuristic Search Processes in Decision Making," in B. B. Anderson, ed., Advances in Consumer Research, Vol. 3 (Proceedings of the Sixth Annual Conference of the Association for Consumer Research, Cincinnati, Ohio, October 30-November 2, 1975).

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

J. Edward Russo, "An Information Processing Analysis of Point-of-Purchase Decisions," Proceedings, American Marketing Association Fall Conference (1975), forthcoming.

J. Edward Russo and Larry D. Rosen, "An Eye Fixation Analysis of Multi-Alternative Choice," Memory and Cognition, 3(May, 1975), 267-76.

Herbert A. Simon, Models of Man (New York: Wiley, 1957).

Herbert A. Simon, "How Big is a Chunk?", Science, 183 (February 8, 1974), 482-88.

Paul Slovic and Sara Lichtenstein, "Comparison of Bayesian and Regression Approaches to the Study of Information Processing Judgment," Organizational Behavior and Human Performance, 6(6, 1971), 649-744.

John E. Swan, "Search Behavior Related to Expectations Concerning Brand Performance," Journal of Applied Psychology, 56(August, 1972) 532-35.

George J. Szybillo and Jacob Jacoby, "Intrinsic vs. Extrinsic Cues as Determinants of Perceived Product Quality,'' Journal of Applied Psychology, 59(1, 1974), 74-78.

John P. Van de Geer, Introduction to Multivariate Analysis for the Social Sciences (San Francisco: W. H. Freeman and Co., 1971).

Karl C. Weigl, "Perceived Risk and Information Search in a Gift Buying Situation," unpublished M.S. thesis, (Purdue University, 1975).

William L. Wilkie, "Assessment of Consumer Information Processing Research in Relation to Public Policy Needs," draft report for the Office of Exploratory Research and Problem Assessment, National Science Foundation Grant GI-42057), June, 1974

William L. Wilkie and David M. Gardner, "The Role of Marketing Research in Public Policy Decision Making," Journal of Marketing, 38(17, 1974), 58-47.

Frederick W. Winter, "Laboratory Measurement of Response to Consumer Information," Faculty Working Papers, School of Commerce, No. 227(University of Illinois,Jan.10,1975).

Peter L. Wright, "Consumer Judgment Strategies: Beyond the Compensatory Assumption," in M. Venkatesan, ed., Proceedings, Third Annual Conference of the Association for Consumer Research, 1972, 316-24.

Peter L. Wright, "The Harassed Decision Maker: Time Pressures, Distractions, and the Use of Evidence," Journal of Applied Psychology, 59(5, 1974), 555-61.

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