A Multivariate Analysis of Perceived Risk Self-Confidence and Information Sources


William G. Zikmund and Jerome E. Scott (1974) ,"A Multivariate Analysis of Perceived Risk Self-Confidence and Information Sources", in NA - Advances in Consumer Research Volume 01, eds. Scott Ward and Peter Wright, Ann Abor, MI : Association for Consumer Research, Pages: 406-416.

Advances in Consumer Research Volume 1, 1974    Pages 406-416


William G. Zikmund, Oklahoma State University

Jerome E. Scott, University of Colorado

[William G. Zikmund is Assistant Professor of Marketing at Oklahoma State University.]

[Jerome E. Scott is Assistant Professor of Marketing at the University of Colorado.]

Few addresses to the American Marketing Association have stimulated as much empirical research as Bauer's (1960) paper on perceived risk. Although Bauer's initial statement mentions multiple consequences of purchasing decisions, most of the empirical studies have utilized unidimensional measures of overall risk associated with certain products or brands (e.g., Arndt 1967, Cunningham 1965). In some instances, products were simply inferred or assumed a priori to be more risky than others. Later studies have recognized the need to identify components of general risk and have focused in varying degrees on such dimensions as social, performance, and financial risk (e.g., Perry and Hamm 1969, Roselius 1971). The separation of overall risk into these components was motivated in part by the belief that the type of risk perceived would influence brand choice, information handling, and shopping behavior. Further, that differences in the nature of the risk perceived would interact with certain personality variables in moderating purchase related behavior.

This research extends the inquiry into the multidimensional nature of perceived risk and the relationship of these risk dimensions to overall risk perception, information handling and generalized self confidence. Specifically, we investigate the nature and dimensionality of risk for three product classes using an empirical rather than a normative approach. Second, we relate these empirically derived risk components to overall risk perception to demonstrate a method by which marketers can assess which types of risk are determinant in ordering overall risk for a product class. Third, we investigate the relationship of the risk components derived to the importance of various risk reducing information sources. This permits the marketer to associate type of product-specific risk with the source of information identified as important in making a brand choice decision. Further, we study whether generalized self-confidence influences product-specific risk and utilization of information.


While previous research has identified different types of risk, the categories used have resulted from normative, a priori reasoning and are not expressed in terms of the attributes of the product and the purchase situation. The familiar dichotomy of performance and psychosocial risk is measured and studied across all product classes by direct one sentence questions. Moreover, the questions are the same for all product classes. Several interesting questions arise, however, which are not answered by this approach. Is the dimensionality of the risk space for any product class the same and is it too dimensional? Is it not cognitively more direct to scale risk components in terms of attributes of the product and aspects of the purchase environment rather than by interpretive questions on "performance", or "social evaluations" risk. It would seem that assessing the risk domain in this manner would be a more reliable approach because it elicits responses from bits of information as they are maintained by the individual. Furthermore, measurement in terms of the attributes is more meaningful to the marketer because this relates risk to product-specific information.

This approach requires identification of principal risk dimensions within an appropriate theoretical framework from a wide variety of empirically derived product related attributes. We have chosen to treat overall risk and its components in a fashion similar to Fishbein's (1967) attitude paradigm in cognitive theory. That is, if a person perceives a great deal of overall risk in a particular product, then this belief should be associated with a set of beliefs that the product is risky on one or more of its important attributes or on other purchase related factors. It would be inconsistent to hold an overall general belief that a product is a risky purchase if one did not perceive certain aspects of the purchase as having some degree of risk. Further that the risk belief elements relating to product attributes can be treated as having both an uncertainty dimension and a consequence dimension. The uncertainty dimension is analogous to Fishbein's likelihood of association component and measures a consumer's subjective probability that the purchase may result in undesirable consequences. The consequence dimension measures the extent to which a consumer seeks to avoid certain possible outcomes of the purchase.

This framework is consistent with Bauer's contention that perceived risk is composed of the subjective uncertainty involved in a situation and the possible consequences of a given action. We, therefore, measure both the extent to which an individual perceives that the purchase of a product is likely to eventuate in a set of undesirable consequences and the seriousness of the consequences. This framework presents investigation into the riskiness of a product not previously possible. Following the disaggregated regression approaches now in use with attitude models (Scott 1972), we can isolate which aspects of the product or purchasing circumstances are ordering overall risk perception. In effect we are recovering the determinant risk components which distinguish between those who perceive risk in a given purchase from those who do not.

It has been argued that "the amount and nature of perceived risk will define consumers information needs, and consumers will seek out sources, types, and amounts of information that seem most likely to satisfy their particular information needs," (Cox 1967). While many studies provide partial evidence supporting this hypothesis, (See Zikmund 1973), most researchers have dealt exclusively with personal information sources. Some problems associated with past research are illustrated in Perry and Hamm's study (1969). While their results show a positive relationship between "social" and "economic risk" and personal information sources, these variables were, apparently, determined without regard for the uncertainty and consequence aspects of risk. In addition, only the influence of personal sources were reported.

Partially due to the use of unidimensional measures of risk, more researchers have analyzed differences in search behavior between high and low-risk perceivers. Some evidence supports the information seeking hypothesis: however, there have been negative findings (See Zikmund 1973). We investigate this hypothesis based on the multidimensional risk description previously discussed. It may be that the relationship between risk perception and preferences for certain information sources is obscured by using the overall risk measure as opposed to its associated cognitive components. The multidimensional approach will permit a simultaneous investigation of which information sources are relied on to reduce specific types of risk.

Several researchers studying the consumer's decision making process have concerned themselves with the relationship between the personality variable, generalized self-confidence, and utilization of information. After the Harvard Group's research, Cox (1967) restated the conceptual scheme of perceived risk and argued that risk ranges from specific risk (related to a brand or product) to the general. Brody and Cunningham (1968) argued that perceived performance risk may interact with specific self confidence. Postulating that generalized self-confidence is inversely related to perceived risk, Hisrich, Dornoff and Kernan (1972) provide evidence for weak to moderate relationships between risk and both specific and generalized self-confidence. This research provides further evidence on the proposition that generalized self-confidence influences product specific risk and that generalized self-confidence influences utilization of information.


The research was conducted in two stages to avoid respondent fatigue. The first stage was designed to determine the multidimensional risk components for each of eight product classes. To accomplish this, a large number of possible consequences associated with the purchase of each type of product was generated. The list was derived from informal interviews with consumers, a content analysis relevant promotional materials, and previous research with the product classes chosen. Personal interviews were held with a convenience sample of 57 housewives in Boulder, Colorado. Each respondent was asked to rate each product class on an overall risk measure and to rate the likelihood and seriousness of each potential consequence. The individual risk variables were then factor analyzed to define the principal risk dimensions for each product class. After obtaining the factor structure, key definers of the factors were chosen based on highest factor loadings. These key definers were then used in the second stage of the research to scale risk perceptions for each component.

The second stage of the research was designed to investigate the relationships among overall risk, risk components, information handling, and self confidence. Subjects for Stage II were obtained by soliciting the cooperation of five women's clubs in Boulder, Denver, and Longmont, Colorado. A donation of $1.50 per subject was made and 181 usable responses were obtained.

Product Classes

In Stage I, information was collected on eight product classes. The products, as well as average overall risk scores and standard deviations are shown in Table 1. Three of these classes were chosen for inclusion in the second stage of the research. They are portable color televisions (high risk), metal lawn furniture (medium risk), and personal stationery (low risk). These choices were made to permit investigation of the relationships discussed over a wide range of risk perception and to make the results more general. To insure respondents' familiarity with all products investigated, an extremely high risk product, microwave ovens, was eliminated from consideration. Because of questionnaire length, only one product class, the medium risk product, was included in those areas of the research relating to information utilization.



Measurement of Perceived Risk

The nature of the research required two types of risk measures for both phases of the study. First is an overall measurement of risk for each product category. To obtain this we used Spence, Engle, and Blackwell's (1970) unidimensional risk measure. The second type of risk measure was designed to scale the uncertainty and consequence components of each risk dimension. Operational definitions of these dimensions were based on the conceptual work of the Harvard group and were organized within the risk framework presented earlier. Cox's definition of consequences "the amount that would be lost (i.e., that which is at stake) if the consequences of the act were not favorabLe" (1967, p. 5-6) were used as Cunningham's definition of uncertainty, "the subjective probability that an event will occur" (See Cox 1967).

Cunningham's rationale (12) for weighing uncertainty and consequence dimensions equally was assumed to be valid, and the uncertainty and consequence scores on each variable were multiplied times each other to form the risk perception for each variable. This procedure was followed prior to the factor analysis of Stage I and also in the various analyses of the Second Stage.

Information Sources and Self-Confidence

Included in the study were both personal and non-personal sources of information. The sources included were:

News and editorial materials, manufacturer's brochure, expert opinion, visit to store, magazine advertisements, TV commercials, friends opinion, radio commercials, discussion with spouse, consumers group, store clerk, product display, newspaper advertisements, and previous experience with product.

Respondents were asked to indicate how important each is in choosing a brand from the-product class. Measurement was performed using seven-point scales. The ten items Bell (1967) used to measure generalized self-confidence were used in this research.


Perceived Risk Components

The data obtained in the first stage of the research on risk variables were factor analyzed to obtain the principal risk dimension for each product class. An orthogonal rotation to simple structure was employed and plots of eigenvalues as well as percent of common variance were used as an aid in determining the number of risk factors to extract. The resulting factor structures were well defined with four variables loading on almost every factor. [A variable was considered to be loaded on a factor if its loading was at least .5 on the factor and no more than .3 on any other factor.] Six factors were extracted for color T.V., seven for lawn furniture, and three for personal stationery. The relatively few factors for personal stationery is probably due to the fact that stationery is much less complex having fewer evaluative dimensions than the other two products. For this reason it also has fewer areas of concern (risk) for the prospective buyer.

Interpretation of the risk dimensions recovered are listed on Table 2. All three products seemed to have dimensions relating to quality or reliability and the reaction of significant others who might judge a purchase. An interesting finding is the factor "future opportunity lost" which is associated with both T.V. sets and lawn furniture. Apparently, an important source of risk for these products is the expectation that an improved or lower cost product may be available at a future time which would be precluded by a current purchase. While other similarities can be observed there are also differences. A shopping frustration factor seems to be associated with lawn furniture and stationery but not color televisions. Additionally, lawn furniture is viewed as having potential adjustment or maintenance problems which does not appear with the other products.



This illustrates an importance reason for investigating perceived risk in a multidimensional fashion. Consumers evaluate products on the basis of a few principal attributes and each represents a potential source of risk. Further, these attributes vary across product classes. Disaggregating perceived risk into product-specific components in the fashion provides much more information about why a consumer perceives risk than overall measures such as social or performance risk.

Determinant Risk Components

Measurement of risk perceptions on various components as suggested above provides information on which attributes of a product are considered to be most risky across a group of potential consumers. It may also be of interest to know which risk components distinguish between those consumers who perceive high general risk and those who perceive low general risk. These components should be the most fruitful in attempting to reduce overall risk perception. This is analogous to the relationship between brand appeal and product attributes in attitude research where some product attributes may be considered very important, yet not be determinant in ordering appeal because they are either taken for granted or considered unavailable.

To recover the determinant risk components requires investigating the relationship between overall risk perception and the multidimensional risk components. If a relationship of sufficient strength can be demonstrated, then some evidence regarding which attributes are determinant in ordering overall risk perception may be revealed. We investigated this relationship by regressing general perceived risk on the multidimensional risk components considered as separate regressors in a multiple regression. The results of the regression are shown in Table 3.



A statistically significant relationship (p < .05) was found between general perceived risk and the risk components associated with portable color television and metal lawn furniture. For the low risk product class, stationery, a positive relationship at the .10 level was found when regressing general risk on all three risk components. [Including only low risk components in the model, the relationship is significant at the .05 level.] Although the strength of these relationships was not great, evidence of a relationship is provided.

Risk components which are most determinant in ordering overall perceived risk can be identified by standardizing the raw score regression coefficients (which obtains the so-called beta coefficients). The component with the largest standardized regression coefficient can be considered as most operative in distinguishing between those who perceive a high level of general risk and those who do not. For example, performance reliability and unanticipated problem were clearly the most determine risk components influencing overall television risk. Social reaction and adjustment or maintenance problem were the determinant risk components influencing overall lawn furniture risk.

It is interesting to note that the percentage of variance explained (by the regressors) for the high-medium risk product (R2 = .198) was larger than the variance explained for the medium-risk product (R2 = .085) which is, in turn, larger than the variance explained by the low risk product (R2 = .034). This finding suggests that the consumer is possibly aware of the cognitive dimensions of risk as the product becomes increasingly more risky, and may indicate that perceived risk is useful only in studying products where risk exceeds some threshold level.

Risk Perception and Information Search

Past research has provided some evidence that high risk perceivers engage in more risk reducing activities, such as information search, than low risk perceivers. Two procedures were used to further investigate this relationship.

In the first procedure, subjects were classified into two groups (high risk and low risk) based on their scores on the unidimensional risk measure. Pattern differences between the two groups on the importance of fourteen information sources were then identified using multiple discriminant analysis. The second procedure treats risk perception according to the multivariate perspective by assessing the relationship between the seven lawn furniture risk components and the importance of information sources. Canonical correlation was used for this analysis.

To conduct the discriminant analysis each respondent was categorized into a high risk group (N = 92) or a low risk group (N = 89) based on the median score for the entire sample. Although a significant discriminant function was obtained using a stepwise procedure (Wilks' Lambda = .96, F2 178 = 3-74) with 2 variables entering, the ability of the function to classify the subjects was poor. Only 62% of the high risk perceivers and 588 of the low risk perceivers could be properly classified. These results have a high probability of occurrence by chance alone, particularly when one recognizes that the classification analysis was conducted with the same subjects as were used to construct the function.

These findings suggest that the rating of information source importance does not provide a very good basis for separating high risk perceivers from low risk perceivers when an overall risk measure is used to group subjects. Or alternatively, that low risk perceivers consider the same sources of information as important as high risk perceivers.

It may be, however, that the use of an overall risk measure obscures the relationship between risk perception and information source importance. Furthermore, use of an overall risk measure does not connect risk to product-specific risk dimensions. The second procedure was designed to over come these difficulties by investigating the relationship between the product-specific components of risk and the sources of information with canonical analysis.

The results of this analysis suggest a much stronger relationship between risk and information source importance. The canonical correlation for the first canonical variate was .50 and was statistically significant beyond the .001 level (using Bartlett's chi-square test). The second canonical variate (R = .412) was significant at the .10 level. None of the remaining variates were statistically significant at reasonable levels. Interpretation of the sign and relative size of the coefficients for these three functions provides simultaneous estimates of the relationship between risk components and sources of information.

Variables with the largest canonical weights for the first variate are shown in Figure l. These indicate that respondents perceiving lawn furniture as high in risk related to self-expression, future opportunity (lost), and adjustment or maintenance problems tend to rate information from consumer groups, salesmen or clerks, product displays, and previous experience as important in the purchase of a lawn furniture set.



Respondents with high negative scores on a risk component perceive low risk associated with this component. A high negative score on an information source means the source was not believed to be important. Thus, a respondent believing lawn furniture was low in performance/quality risk and perceiving little risk of unrealized expectations rate visits to stores, television commercials and expert opinion as unimportant in making the purchasing decision. Conversely, consumers high in performance/quality risk and the risk of unrealized expectation rate information from visits to stores, television commercials, and experts as important in making the purchasing decision. Interpretation of these results suggest the media to be used for reducing specific types of risk. Although interpretations of the second and third canonical variates were equally as interesting, space limitations do not allow for presentation of these findings. Additional findings may be found in Zikmund (1973).

Effects of Generalized Self Confidence

Much has been written on the potential influence of self-confidence on risk perception and information seeking. Cunningham, for example, contends that there must be a certain amount of self-confidence to initiate word-of-mouth conversation. The influence of the trait on overall risk perception and the importance of various sources of information was also tested in this research to add to the evidence on this topic. The results obtained were negative in both cases. We were unable to find significant differences in risk perception among groups described as high, medium, and low in generalized self-confidence for any of the eight product classes tested in Table 1. Further, there did not appear to be differences in the importance of various sources of lawn furniture information among groups described as low, medium, or high in generalized self-confidence.


The results of the study provide considerable evidence that perceived risk should be treated multidimensionally and with regard to the specific product class under study. The empirical approach to recovering the principal risk dimensions of several product classes indicates that both the dimensionality and nature of risk components vary by product as the evaluative dimensions change. This implies that the marketer would gain more useful information on why a product is perceived to be risky and, therefore, be in a better position to reduce consumers' risk perception.

The moderate to weak positive relationships between perceived risk (general) and the risk components provide additional evidence that risk can be studied as a multidimensional phenomena. Based on a similar approach, Jacoby and Kaplan (1972) found a strong positive relationship between overall risk and certain components for students who were not naive, i.e., "just completed a two-hour section on the topic of perceived risk." While the relationships reported here are not as strong as those found by Jacoby and Kaplan the difference in research setting, such as use of housewives versus "informed" students, could account for some of the discrepancy. The important finding supported in both studies was that multidimensional measures of risk had some predictive validity. This approach also permits the researcher to recover which risk components distinguish between those who perceive much risk in a purchase and those who do not. These determinant risk components are of primary importance in risk reducing communications.

Evidence of a relationship between risk perception and the importance of various risk reducing sources of information was found. The relationship, however, was found to be much stronger when the multidimensional product-specific measures of risk were used than when the unidimensional risk measure was used. Use of an overall risk measure appears to obscure the relationship between risk and information search. Perhaps consumers perceive certain salient consequences of their action and attempt to reduce this special risk through appropriate sources, whereas a summary of perception of overall risk may not be related to any particular source. This is a possible explanation for the variety of findings relating risk to interpersonal communications and other means of risk reduction.

No evidence was found to support the hypothesis that self-confidence was a key moderating variable. Thus, the personality variable self-confidence does not appear to influence specific product risk or search behavior, at least for the product classes investigated in this research.


Arndt, Johan. "Role of Product-Related Conversations in the Diffusion of New Products." Journal of Marketing Research, Vol. IV, (August, 1967), pp. 291-295.

Arndt, Johan. "Perceived Risk, Sociometric Integration, and Word of Mouth m the Adoption of a New Food Product," In Cox, Donald F. (ed.). Risk Taking and Information Handling in Consumer Behavior. Boston: Harvard University.

Arndt, Johan. "Word-of-Mouth Advertising and Informal Communication." In Cox, Donald F. (ed.). Risk Taking and Information Handling in Consumer Behavior. Boston: Division of Research, Harvard Business School, 1967.

Bauer, Raymond A. "Consumer Behavior as Risk Taking." Proceedings American Marketing Association, (December, 1960). In Cox, Donald F. (ed.). Risk Taking and Information Handling in Consumer Behavior. Boston: Harvard University, 1967, pp. 389-398.

Bauer, Raymond A. "Self-Confidence and Persuasibility: One More Time." Journal of Marketing Research, Vol. VII, (May, 1970), pp. 256-258.

Bell, Gerald D. "Self-Confidence, Persuasibility, and Cognitive Dissonance Among Automobile Buyers." In Cox, Op. cit.

Brody, Robert P. and Scott M. Cunningham. "Personality Variables and the Consumer Decision Process." Journal of Marketing Research, Vol. V, (February. 1968), pp . 50-57

Cox, Donald F. (ed.). Risk Taking and Information Handling in Consumer Behavior. Boston: Harvard University, 1967.

Cox, Donald F. and Stuart U. Rich. "Perceived Risk and Consumer Decision Making--The Case of Telephone Shoppers." Journal of Marketing Research, Vol. l, (November, 1964), pp. 32-39.

Cunningham, Scott M. "The Major Dimensions of Perceived Risk." In Cox, Donald F. (ed.). Risk Taking and Information Handling in Consumer Behavior. Boston: Harvard University.

Cunningham, Scott M. "Perceived Risk as a Factor in Product-Oriented Word-of-Mouth Behavior: A First Step." In Bennett, Peter (ed.). Proceedings American Marketing Association, 1965, pp. 229-238.

Deering, Barbara J. and Jacob Jacoby. "Risk Enhancement and Risk Reduction as Strategies for Handling Perceived Risk." A paper presented to the Association for Consumer Research, (November, 1972).

Fishbein, Martin. Readings in Attitude Theory and Measurement. New York: John Wiley and Sons, 1967.



William G. Zikmund, Oklahoma State University
Jerome E. Scott, University of Colorado


NA - Advances in Consumer Research Volume 01 | 1974

Share Proceeding

Featured papers

See More


A Computational Social Science Framework for Visualizing the Possibility Space of Consumer-Object Assemblages from IoT Interaction Data

Donna Hoffman, George Washington University, USA
Thomas Novak, George Washington University, USA

Read More


How Matte Product Surface Enhances Perceived Durability

Taehoon Park, University of South Carolina, USA
Junghan Kim, Singapore Management University, Singapore

Read More


L1. The Effects of Cultural Syndromes on Customers’ Responses to Service Failures: A Perspective-Flexibility-Based Mechanism

Vincent Chi Wong, Lingnan University
Robert Wyer Jr., University of Cincinnati, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.