Generating Product Ideas: a Modification of the Dual Questioning Technique

Carl Obermiller, The Ohio State University
ABSTRACT - A brief discussion of marketing research into multi-attribute models, concentrating on the direct dual questioning technique, is followed by a suggested method for determining potential product modification ideas. The technique consists of measures of three dimensions of product attributes: importance, perceived difference, and satisfaction with current alternatives.
[ to cite ]:
Carl Obermiller (1980) ,"Generating Product Ideas: a Modification of the Dual Questioning Technique", in NA - Advances in Consumer Research Volume 07, eds. Jerry C. Olson, Ann Abor, MI : Association for Consumer Research, Pages: 767-771.

Advances in Consumer Research Volume 7, 1980     Pages 767-771

GENERATING PRODUCT IDEAS: A MODIFICATION OF THE DUAL QUESTIONING TECHNIQUE

Carl Obermiller, The Ohio State University

ABSTRACT -

A brief discussion of marketing research into multi-attribute models, concentrating on the direct dual questioning technique, is followed by a suggested method for determining potential product modification ideas. The technique consists of measures of three dimensions of product attributes: importance, perceived difference, and satisfaction with current alternatives.

THE MULTI-ATTRIBUTE MODEL

The purpose of this paper is to briefly review the development of the multi-attribute model, particularly as it relates to the development of the concept of determinant attributes and to gain an understanding of these constructs sufficient to develop a managerially useful approach to product idea generation.

Although much work has been done in the area of multi-attribute marketing, almost any review article will indicate that this work has succeeded in raising as many questions as it has answered. [See, for example: Mahajan, et al (1978), Day (1972), Shocker and Srinivasan (1977), Wilkie and Pessemier (1973), and Myers and Alpert (1976).] In fact, if interest in the topic has waned, it is more likely due to frustration than from a lack of unsolved problems. Efforts to explain have frequently been more complex than the phenomenon to be explained. However, much that we have learned can be made useful by a simple, direct method of application.

Marketers regard products as bundles of attributes. These attributes are defined in terms of the benefits consumers perceive products as offering. However, the product's attributes are not confined to physical characteristics. They include psychological and sociological benefits, such as status and image, as well as such tangible features as price and composition. The difference between the conceptions of the product as the sum of its physical features, which the manager knows well, and the product as the sum of its attributes, which the consumer perceives, can be large. One of the critical problems of the product manager, therefore, is to understand how the product is being perceived by consumers. Only by gaining an understanding of how consumers view product alternatives and use product characteristics in making choices can the product manager expect to achieve either effectiveness or efficiency in shaping marketing strategy. This is especially true for the generation of new product or product modification ideas.

Marketing communication is generally designed to change, form, or reinforce consumers' beliefs about some component of the buying process--their needs, the relative importance of those needs, or the need-satisfying attributes of the product. The individual consumer's overall attitude toward the product can be expressed as a combination of these beliefs and the evaluation the individual makes of them:

EQUATION    (1)

This is the Fishbein belief-value model, [Fishbein (1967).]

"o = the overall attitude toward object o

Bi = the ith belief about the object o

ai = the value that the individual places on the ith belief.

The general multi-attribute model is quite similar to the Fishbein construction. It considers the overall attitude toward an object to consist of the belief that the object possesses the attribute, weighted by the importance attached to that attribute. Thus,

"o = Bi Ii   (2)

where

"o = the overall attitude toward an object

Bi = the belief that the object possesses attribute I

Ii = the importance the individual places on attribute i.

It is implicit in this general model that a product is a bundle of attributes. The number of possible attributes that could be included is infinitely large; however, for most of these, the belief that the product possesses them will be zero. On the other hand, for any given product a very substantial list of attributes might be drawn up with beliefs greater than zero. Yet, for most of these, the importance value is zero or insignificant even though belief is near certainty. (For example, the elemental composition of the ink used to print these words certainly has some characteristics, but the exact nature of the ink's composition is of zero importance for most consumers.) Thus, most attributes have no impact upon overall attitude.

However, every product can be described in terms of a number of attributes that it is perceived to possess and that are important to the consumer. These attributes define a product space of which each attribute is a vector and each importance value a weight for the vector. Within this space the product is located according to its rating on attributes and their importances. Likewise, competing brands, substitute products, and the "ideal" product are also located within this space. A managerial understanding of such a product space is imperative. Existing products can be repositioned by changing their attributes and/or effectively communicating their possession of important attributes. New product opportunities can he conceptualized as gaps in the product space near ideal points. Alternative openings in the product space can be measured in terms of their proximity to existing competitive products which may differ in flexibility or strength. Thus, the usefulness of multi-attribute approaches in generating product ideas lies in its potential for uncovering the structure of the attitudes that comprise the market for a set of substitutable products. It is an understanding of consumer attitude structures that allows management to predict efficacious new product ideas. It is necessary for management to assess both the beliefs about attributes and their importance. [Butler and Shuart (1972).] Since importance measures are a function of consumer values, management must address the process of consumer evaluation. Generally, it is much easier to change beliefs than values. Values are quite basic to the individual and evolve over a long period of complex social and individual interactions with family and reference groups.

Beliefs, on the other hand, are functions of the individual's perceptions. Beliefs can be changed by either product modification or persuasive communication. Thus, of the two components of overall attitude, importance, which is a function of values, is far more difficult to change than beliefs, which are a function of information.

THE DUAL QUESTIONING METHOD

Despite a history rich in attitude-related research, it has only been recently that marketers have discovered it relevant to ask what attitudes mean and precisely how they are related to preference. [Rosenberg and Fishbein are representative of a distinct direction of attitude research in psychology that is less concerned with the precise measurement of the strength of attitude than the discovery of underlying structures. Consequently, both regarded predictive validity as secondary to construct validity. This emphasis has been reversed by most marketing uses of multi-attribute models, though it has been retained in this paper.] By analyzing attitudes to ascertain their underlying structures, we are able to distinguish the critical cause-effect relationships between marketing decision variables and consumer preferences.

One of the first significant steps toward attitude structure analysis was made by Myers and Alpert (1968). They describe overall attitude as some combination of attitudes toward the various components of the object. They point out that these attributes differ in importance to the consumer. Furthermore, some attributes differentiate products from one another, i.e. some products possess more of an attribute than others. Those attributes that are both important and differentiating Myers and Alpert call determinant. These are the attributes that determine preference. Thus, dual questioning improves upon simple direct questioning by distinguishing attributes on the dimension of difference.

Myers and Alpert discuss three general methodologies for identifying determinant attributes:

1.  Direct questioning

2.  Indirect questioning

3.  Observation and experiment.

The most direct method for measuring determinance is to ask people what they think is important in making a purchase decision. Traditionally, marketers obtained this information by asking people why they bought a product. This approach seems intuitively high on construct validity. However, it rests on two questionable assumptions: (1) The respondent knows why he prefers one product to another, and (2) he is willing to tell why. Unfortunately, research has shown quite conclusively that respondents are frequently unable to articulate the real reasons for their choices, are sometimes, in fact, unaware of them, and occasionally unwilling to state them for fear of appearing foolish or irrational. [For a recent elaboration, see Nisbett and Wilson (1977).] These same constraints apply to the use of direct questioning methods to determine consumer ideal points with the added difficulty of conceiving of objects that don't already exist.

In spite of its weaknesses, direct questioning is often the best approach because of the problems associated with alternatives. Indirect questioning produces, at most, results that are very difficult to interpret and, at best, covariation data that indicate relations but not causation. Experimentation, the ultimate arbiter of questions of causality is usually too costly and cumbersome to be practicable.

Alpert and Myers returned to the concept of determinance in later work. Alpert (1971) compared eight methods of measuring determinance. He found dual questioning and regression of attribute rankings to be essentially equivalent and superior as predictors of overall preference with R2's of .729 and .728. Alpert concluded that dual questioning was somewhat superior to regression because of the time and expense savings enjoyed by not requiring an examination and rating of individual brands in terms of attributes. He tempers this conclusion by suggesting that indirect methods may be more useful for products with less objective attributes.

Myers and Alpert (1976) return to the issue of attribute determinance in order to clear up confusion in the research among the measures of attribute importance. Importance had variously been defined in terms of salience, importance, and determinance. Salience is defined by Myers and Alpert as a measure of top-of-mind awareness; importance, as significance to the product; and determinance, as consequence for the decision. A clear distinction implicit in these definitions is between importance and determinance. An attribute is important if it is considered a significant feature of the product; it is determinant only if it is important in preference decisions. This distinction accounts easily for the oft-cited automobile-safety example: Safety is important to the product, but not to the preference decision. Salience causes more of a problem, however. According to Myers and Alpert, it has generally been operationalized as order of elicitation. Unfortunately, there is no single way to characterize the earliest mentioned attributes. They may be the most heavily advertised, they may be the most recently processed thoughts; but neither of these is necessarily related to determinance. Thus, it would seem advisable for researchers to maintain the distinction between important and determinant but to avoid use of salience as a measure of either of the other constructs.

Also in their 1976 article, Myers and Alpert respond to modifications of the dual questioning method. Wilkie and Weinrich (1972) had found that the dual questioning method was no better predictor than a model that included only importance measures. Berkowitz, Ginter, and Talarzyk (1976) reported similar results. However, both used modifications of the dual questioning model that included standardized importance and standard deviations of difference ratings. Myers and Alpert suggest that this manipulation probably gave unequal weight to the importance rating, thus, resulting in little difference from the inclusion of importance alone. Berkowitz, Ginter, and Talarzyk found significant differences in the ranking of attributes between their measure and the use of standardized scores for both importance and difference. The conclusion drawn from these two studies is that when both importance and difference are given equal weights, they yield better results than importance alone, though the different forms of their inclusion will result in small differences in attribute rankings.

Anderson, Cox, and Fulcher (1976) used the dual questioning method to measure determinant attributes in a bank selection decision. Fifteen attributes were developed from previous experience and preliminary interviews. Respondents were asked to rate these on a five item importance scale and a four item difference scale. These scales were used to isolate five determinant attributes. The authors then clustered individual respondents and found they fell into two general groups, one that regarded bank services as convenience goods and one that ascribed more importance to them. These clusters were then correlated with a number of demographic and socioeconomic variables. The results were used to define two market segments--service-oriented and convenience-oriented--and design separate patronage appeals for them. [The two major criticisms of Anderson, Cox, and Fulcher (1976) were (1) an inappropriateness of sample: respondents were not new customers, they already had banks, (2) the single attribute that had previously been found to be determinant, location, was rated as high in importance, low in difference. The first of these flaws is here regarded as an error in research methodology; the second is important, and is discussed in this paper, but can be seen as a direct result of the first error--since most of the sample had already selected a bank, they undoubtedly saw most banks as equally convenient. Most prospective customers would have a different perspective. This may be an instance in which salience of attributes should be investigated, see Myers and Alpert (1976).]

Shimp and Lindgren (1977) raise three methodological considerations they feel may bias the results of the dual questioning method. First, they argue that the number of brands included in the comparison set will have an effect on the results. They point out that respondents cannot be expected to give meaningful difference ratings for as many as ten brands. If as many as ten choices exist the researcher must select some subset to investigate. This choice, of course, will pre-determine results. Secondly, the authors argue that the inclusion of a dissimilar alternative will affect the set of determinant attributes. Finally, they contend that the extent to which the comparison set matches the respondent's evoked set will bias determinance ratings. Their argument is that evoked sets by definition include brands that are similar on important attributes, so unevoked sets would include dissimilar brands.

Shimp and Lindgren present data which support each of their contentions. However, their conclusions require cogent for clarification. First, though it is true that respondents can only deal reasonably with about seven alternatives, there is an undeniable trade-off in reducing the size of the comparison set. If the goal of determinant attribute research is to gain an understanding of the underlying attitude structure of the market, it cannot be attained unless all the relevant market choices are included in the research.

The biasing effects of dissimilar choices and evoked sets are less clearly established. Shimp and Lindgren's contentions are accurate but not unexpected. If one brand is divergent in a large set, the set actually is characterized by less difference than if it is divergent in a small set. And, a non-evoked set can be expected to be characterized as less similar than an evoked set. These "dangers" are similar; each is a function of the inclusion of dissimilar or unfamiliar brands. Whether or not this constitutes a bias is dependent upon the purpose of the research. When positioning a product within an established market, a researcher is interested in evoked sets. Generally, these will be fairly similar choices. However, when describing the attitude structure of a market in order to generate product ideas, the researcher should strive to include all possible choices for the usage situation. In this latter case, limitation to evoked sets or otherwise similar choices would do little to uncover potential new ideas. Thus, the Shimp and Lindgren criticisms should be considered as delineations rather than weaknesses.

A final modification of the dual questioning technique has been offered by Hansen (1978). Hansen perceived two limitations of the method: the problem of non-unique solutions and the loss of information through aggregation of data. The non-unique solution problem results when an attribute with high importance, medium difference yields the same determinance score as an attribute with medium importance and high difference. Hansen argues cogently that these two attributes should not be treated the same. The loss of information from aggregation results from gathering data on an individual level and reporting the mean scores on importance and difference ratings. This aggregation procedure can easily mask important differences. For example, consider an extreme hypothetical situation in which half the respondents rated an attribute highly important and highly differentiating while the other half rated it low on both dimensions. The overall mean would show the attribute as only moderately determinant, thus masking obviously valuable information.

Hansen proposes a modified dual questioning method. Responses to both importance and difference ratings are cross-tabulated in a three by three table for each attribute:

TABLE

The cell scores would consist of the percentages of total respondents who rated the attribute in that cell. Thus, each attribute would result in a table with cells totaling 100% and each cell indicating the percentage that responded with that particular determinance combination.

Using the Myers and Alpert concept of determinance, each attribute's determinance score would be the percentage in cell 9. The attributes with the largest entries in cell 9 would be "most determinant". Hansen, however, notes that the entire bottom row (cells 7, 8, and 9) is critical since it indicates the percentage of the sample that rates the attribute high on importance. This is particularly true when the method is used as a tool for generating product ideas.

While attributes with high scores in cell 9 are determinant, Hansen argues that cell 7 may offer the greatest potential for modification ideas because it indicates the proportion of respondents who regard the attribute as highly important but low on difference. Because it is low on difference it offers the most potential for differentiation. The manager may discover that attributes with high percentages in cell 7 may be features that are capable of a simple physical modification, which would provide the product with a determinant advantage.

Although Hansen limits his analysis to the potentials of cells 7 and 9, it seems clear that cells 8 and 6 may also indicate potentially determinant attributes. Cell 8 represents high importance and medium difference; this could be a critical combination. The medium difference rating may reflect real differences that have not been adequately communicated in promotion materials. Like wise, cell 6, which is high in difference, medium in importance, may be a useful attribute to investigate further. It seems possible that some attributes may be potentially important, i.e. if that attribute were to be made more salient it would become more important. Given that the attribute is high on difference, it may already be fairly salient, however, its potential may warrant further investigation. [For a discussion of strategies for changing either beliefs or importance weightings see Lutz (1978).] Hansen applied his modified dual questioning technique and compared results with the standard method. The product was an industrial hard hat. Hansen used a combination of expert judgment and personal interviews to generate attributes. In addition to attribute ratings he asked respondents to indicate specific changes or modifications they would like to see offered in the product. Hansen's analysis measured and compared methods against two criteria: (1) amount of managerially useful information, and (2) flaws or inconsistencies in the results. Hansen used the results from the requests for product modifications to validate the usefulness of the other two models as generators of product modification ideas. His findings showed high correlation between the dual questioning model and the modified model for distinguishing determinant attributes. However, it was the potentially determinant attribute scores, cell 7, that predicted desired product modification ideas. In this finding lies the value of Hansen's modification to the product manager.

Although Hansen's modification scheme improves upon the dual questioning technique, it, too, requires improvement. Since a determinant attribute is perceived as being offered differentially, consumers are confronted with a range of more or less satisfying alternatives. Yet, even with a wide range there is no guarantee that the best available attribute offering is fully satisfying. Thus, satisfaction should be considered when managers address consumer attitude structures. Moreover, for potentially determinant attributes, which offer little choice in attribute offering, the chances of offering satisfactory levels may be increased. Thus, it is when an attribute is important, not differentiating, and not satisfactorily present that product modification potential surely exists. Hansen's modification requires a measure of this additional dimension to be complete.

An alteration of Hansen's method may result in a more complete measure of managerially useful information. This modification would add a low, medium, high rating of consumer satisfaction, [Myers and Alpert (1976) suggested the possibility of measuring satisfaction.] yielding a three by three by three cube in which the critical side would be those nine cells (the shaded area) that are high in importance:

FIGURE

Cells 27, 18, and 9 indicate high satisfaction; these would be the best descriptors of current purchasing decisions. Cells 24 and 15 indicate high importance, moderate satisfaction and high or moderate difference; these would indicate little potential for product modification since consumers are apparently satisfied. Cell 21, which indicates high importance, high difference, and low satisfaction, is peculiar because it implies that consumers see the products as different on an important attribute, but none of the alternatives is satisfying. A large cell entry would indicate a potential modification idea, provided an improvement on that attribute is feasible. The remaining three cells (3, 6, 12) are probably the best sources of modification ideas since they indicate high importance, low or moderate difference, and moderate or little satisfaction with existing offerings. If cells 3, 6, or 12 contain significantly large entries they would indicate potential opportunities for increasing market share. ("Significantly large" in this case is a managerial measure based on past experience and the particular situation rather than statistical measures. However, simple t tests can be used to determine if cell entries are significantly different from one another and to decide upon a cut-off point after attributes are ranked by cell entries.) A final note: because of the expansion to 27 cells, even though many cell entries can be expected to be small, a larger sample of respondents will be necessary than for the nine cell model. (Hansen's sample of 604 would most likely be adequate.)

CONCLUSIONS

The purpose here has been to develop a technique for analyzing consumer attitude structures as an aid to generating product modification ideas. One of the guiding principles has been to keep it simple. [See Leavitt (1978).] Simplicity is desirable, not because managers are less sophisticated than the rest of us, but merely because simple tools are more apt to be used, more flexible, and usually less expensive than complex ones. [For an example of the difference in data gathering costs and respondent fatigue consider a typical case of ten choice elements and five attributes: A regression analysis would require 60 separate evaluations from each respondent -- for a sample of 500, a total of 30,000 responses. The three-question technique would require 15 evaluations from each respondent for a total of 750 responses. (Even though the three-question technique would supply--more information.) The reason for the large difference is the three-question technique does not require brand evaluations of every attribute. Additional advantages of the method are the reduction of halo effects by eliminating brand overall evaluations and the ability to use the method with various sizes of choice and attribute sets. For example, conjoint measures are difficult if these sets are large. Finally, it can be used on an individual level, whereas regression requires aggregation.] To that end a proposal has been made of a fairly simple tool--a three-question technique, that is straight-forward, intuitively appealing, easy to administer, easy to interpret, inexpensive, and generalizable to a number of uses. The results can be interpreted to describe the underlying attitude structure relating to preferences and to indicate at least three types of potential product ideas: (1) those modifications that will bring the current product into line with the existing determinant attitudes, (2) those modifications that may be possible if product perceptions can be made to reflect possession of important attributes that currently are not differentially available but are desirable, and (3) modifications made possible by altering the values of consumers to coincide with existing product attributes.

REFERENCES

Alpert, Mark (1971), "Identification of Determinant Attributes: A Comparison of Methods," Journal of Marketing Research, Vol. III, May, 184-181.

Anderson, W. Thomas, Cox, Eli and Fulcher, David (1976), "Bank Selection Decisions and Market Segmentation," Journal of Marketing, Vol. 40, January, 40-45.

Butler, Stewart and Shuart, Alan (1972), "On Applying Expectancy-Attitude Models in Marketing Research," Working Series in Marketing Research, No. 14, June, The Pennsylvania State University.

Day, George (1972), "Evaluating Models of Attitude Structure," Journal of Marketing Research, Vol. IX, August, 279-286.

Fishbein, Martin (1976), "A Behavior Theory Approach to the Relations Between Beliefs about an Object and the Attitude Toward the Object," Attitude Theory and Measurement, Martin Fishbein (ed.), New York: John Wiley and Sons, Inc., 389-400.

Hansen, Robert (1978), "Determinant Attribute Analysis and the Search for Product Modification Ideas: Theoretical and Managerial Considerations," unpublished paper.

Hughes, G. David (1978), A Planning Approach to Marketing Management, Reading, Massachusetts: Addison Wesley Publishing Company, Inc., 164.

Leavitt, Theodore (1978), "A Heretical View of Management 'Science'," Fortune, December, 50-52.

Lutz, Richard J. (1978), "How Difficult Is It to Change Consumer Decision Structures?" University of California, Los Angeles Working Paper Series, Center for Marketing Studies, Paper No. 62, February.

Mahajan, Vijay, Jain, Arun, Thangaraj, Arun and Goodwin, Stephen (1978), "Attribute Determinance: An Evaluation of Alternative Measures and Some Potential Applications," Analytical Approaches to Product and Market Planning, Alan Shocker (ed.), Boston: Marketing Science Institute.

Myers, James and Alpert, Mark (1968), "Determinant Buying Attitudes: Meaning and Measurement," Journal of Marketing, Vol. 32, October, 13-20.

Myers, James and Alpert, Mark (1976),"Semantic Confusion in Attitude Research: Salience vs. Importance vs. Determinance," Advances in Consumer Research, Vol. 5, 106-110.

Nisbett, Richard and Wilson, Timothy Decamp (1977), "Telling More Than We Can Know: Verbal Reports on Mental Processes," Psychological Review, Vol. 34, No. 3, May, 231-260.

Rosenberg, Milton (1967), "Cognitive Structure and Attitudinal Affect," Attitude Theory and Measurement, Martin Fishbein (ed.), New York: John Wiley and Sons, Inc., 389-400.

Shimp, Terrence and Lindgren, John (1977), "Some Methodological Considerations and Empirical Findings Concerning the Use of the Dual Questioning Method to Identify Determinant Attributes," Proceedings: American Marketing Association, 327-332.

Shocker, Allan and Srinivasan, V. (1977), "Multi-Attribute Approaches for Product Concept Evaluation and Generation: A Critical Review," Working Paper No. 240, Graduate School of Business, University of Pittsburgh, October.

Wilkie, William and Pessemier, Edgar (1973), "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, Vol. X, November, 428-441.

Wilkie, William and Weinreich, Rolf (1972), "Effects of the Number and Type of Attributes Included in an Attitude Model: More Is Not Better," Association for Consumer Research Proceedings, 325-340.

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