Searching For Important Attributes and Appropriate Levels

Donald R. Lehmann, Columbia University
[ to cite ]:
Donald R. Lehmann (1979) ,"Searching For Important Attributes and Appropriate Levels", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 280-283.

Advances in Consumer Research Volume 6, 1979      Pages 280-283

SEARCHING FOR IMPORTANT ATTRIBUTES AND APPROPRIATE LEVELS

Donald R. Lehmann, Columbia University

The concept of preference and subsequently choice resulting from how alternatives are perceived on a set of key attributes is well established (Fishbein, l957; Lancaster, 1966; Rosenberg, 1956; Wilkie & Pessemier, 1973). Work on such models requires identification of the key attributes and the desirable levels of these attributes. This paper will examine the recent contributions made by Myers (1978), Olson and Muderrisoglu (1978), and Quelch (1978) in this area, as well as some problems with their results. In addition, some other issues will be discussed, including whether attribute-based decision making is a sound model of actual behavior and if so, what methods might be used to study it.

Assuming we are interested in studying attribute-based decisions; the issue of for what purpose becomes relevant. Some of the major choices which must be made include:

1. Prediction vs. Understanding of Process. Models based on multiattribute models usually work well predictively. When the issue is process understanding or causality, however, much greater care is needed in establishing their usefulness.

2. Group vs. Individual Behavior. For providing guidance to product designers or managers, group level modeling is often sufficient. Studying process, on the other hand, leads to a focus on individual behavior.

3. Subject of Investigation. Studying key attributes has at least three definable categories: determination of what are the key attributes, determination of the relative importances of the key attributes, and determination of optimum levels on the attributes.

Here the papers by Olson and Mudderisoglu and Quelch focus on individual behavior and process while the paper by Myers is oriented toward management decisions and aggregate behavior.

OLSON AND MUDDERISOGLU

Olson and Mudderisoglu study the determination of key attributes by means of free elicitation. Their major concern is in determining how reliable such free elicitation producers are under three different cues: product category, product/purchase situation, and brand.

Strengths

1. Important Topic. The issue of the stability of free elicitation procedures in specific and attribute importance determination procedures in general is crucial. Yet, few authors have explicitly set out to determine how stable such responses are.

2. Execution. The procedures followed seem carefully executed, something which is lacking in many studies.

Problems

1. The Sample. The usual caveats about a sample of 30 students are relevant.

2. Procedure. The procedure itself, though carefully executed, has some problems. First, the one-week time period between measurements may have been too short to avoid carryover effects over time (e.g., either remembering answers given in a previous week or being bored with the procedure the second time). Since the products were in stable categories (and hence few real changes would be expected), a longer period would be somewhat more desirable. Also, by using three cues sequentially, there is a real possibility of carryover across the three cues. An analysis of the similarity of the elicitations generated across the three cues would have been interesting in that regard. Finally, it would have been very interesting to see how the stability of free elicitation responses compared with that of other methods, but obviously this extension would be a major one.

3. Results. There may be more in the data than is discussed. For example, it is interesting that the number of elicitations seems to be in the 7 + 2 range. Similarly, the apparent decrease in the number of elicitations and even more dramatically in the time taken by subjects might be attributable to learning plus a wearout in the novelty of participating in the study. Also, it would be interesting to know what percent of the responses appear to be attributable to advertising playback.

Summary

Olson and Mudderisoglu have presented some interesting data concerning the stability of free elicitation procedures. Assuming this objective, accounting for individual differences (as they suggest) as well as extensions to more complex product categories seems appropriate. More broadly, comparison of both the results and the stability of free elicitation with other procedures seems appropriate future research.

QUELCH

Quelch uses an information display board methodology to uncover the importance of a variety of attributes about cold cereals.

Strengths

1. Sample. Using 250 real shoppers is commendable.

2. The Concept. The concept of relating attribute importance to information acquisition is appealing.

Problems

1. The Procedure. The entire procedure seems artificial. There are only five dimensions: price and two natural pairs; ingredient and nutritional information and physical appearance and physical product. These dimensions are compound, thus muddying the interpretation. Forcing respondents to obtain information about all six brands on the first attribute chosen seems unnatural. One also wonders if the subjects were able to deduce the name of the products (e.g., Kellogg's Corn Flakes) from the appearance.

2. The Repeat Purchase Issue. The paper begins by mentioning trial and repeat choice. The data, however, is a single choice of an unbranded product and hence all the choices are really trials.

3. The Determinance Measure. The determinance measure may be flawed. It is defined as the number of brands selected on a given dimension minus the number selected on the previous dimension. This suggests that people always look for fewer brands as they proceed (which they do not always do unless constrained to). Also the measure is at least potentially order dependent (the first attribute has a potential score of 6, the subsequent ones do not) and a determinance measure based on the percent of the remaining brands eliminated rather than the raw number seems preferable.

4. The Relation of the Measures of Importance. Three measures of importance based on information acquisition are suggested and a separate constant sum scale measure is available, yet the resulting average importances are not presented. Moreover, one has no idea how closely the different measures are related to each other on an individual basis. Also comparison with other results (e.g., Jacoby, Szybillo, and Busato-Schach, 1977) would have been useful.

Summary

This pilot study attempts to deduce attribute importances from information display boards. As such it provides an interesting starting point for research in this area, but one which needs further refinement.

MYERS

Myers does not directly assess the importance of attributes. Rather he attempts to provide guidance to product designers. In doing so he focuses on finding the ideal levels of descriptive attributes which can be used in guiding laboratory personnel in product design.

Strengths

1. The Sample. This paper has a sample size of 621 real people rating products they actually use.

2. Cleverness. The approach is clever, simple, and easy to use. It allows anyone with a knowledge of cross-tabs to indirectly deduce ideal levels on a set of attributes.

3. Usefulness. Assuming the results of this approach were correct, they are directly useful in solving a real problem, product design.

Problems

1. The Attributes Used. The attributes used (by being not evaluative) may also not be relevant to the consumer and hence the resulting product may be of limited appeal. In short, the whole issue of determinant attributes becomes relevant (Myers and Alpert, 1968; Alpert, 1971).

2. The Causality Assumption. The method implicitly assumes the direction of causation is from attribute position to attitude. Yet there is considerable evidence (e.g., Beckwith and Lehmann, 1975) that causation, at least in measured responses, runs both ways. Also measured ratings on any dimension will generally have an evaluative component may be especially large on less important attributes.

3. Collinearity. Examining the attributes one at a time assumes that the attributes are independent, which they almost certainly are not, and hence the estimates are biased. The best solution to this problem is a multivariate procedure which simultaneously estimates the effects of different levels of the attributes via ANOVA or dummy variable multiple regression.

4. Level of Aggregation. As in almost any model, the question of the appropriateness of aggregating people is relevant. Here people were grouped by favorite brand, which is a step in the right direction. Still the question of segment homogeneity could have been more fully explored.

Summary

Myers has presented an interesting approach for deducing the ideal level of attributes. If the procedure is extended to be multivariate (which is easy to do) and the direction of causality can be established more unambiguously (which is hard to do), then the procedure can be useful. Incidentally, the procedure can easily be adapted to estimating attribute importances. The range of the mean scores in the cross-tab approach (or dummy variable coefficients in the preferred multivariate approach) can be treated as a measure of importance as in conjoint analysis (Green and Wind, 1975).

CONSIDERATIONS FOR FUTURE RESEARCH

Are Attributes Relevant

All of the papers here and much of the research in marketing, including that of the author, is based on the assumption that consumers make trade-offs on attributes. Yet this fundamental tenet deserves re-examination.

The view that consumers are actively engaged in evaluating alternatives on attributes is clearly an appealing and flattering view of man. Unfortunately, at least for most decisions it is probably also false. Every day is filled with enough decisions so that only a few can consciously be considered in anything like the fullness that multiattribute models imply. The rest are handled by rules or standard operating procedures which have been developed over time. Hence when consumers give responses about which attributes are relevant, the responses may be as much rationalization as truth. While simplified rules make more sense than complex ones (Wright, 1975), even simple ones may be employed infrequently.

Putting the proposition differently, multiattribute models are probably a good descriptive or predictive model of how consumers behave when they are heavily involved in a decision. (They may or may not accurately reflect the actual process followed by consumers.) Similarly in a market where information is readily available and the marketplace relatively free, multiattribute models are likely to fairly accurately forecast the the equilibrium share of competing brands. For most situations, however, it is inappropriate to assume that individuals follow the process of information acquisition and processing implied by multiattribute models. (The situations under which such models are likely to be used are described in Lehmann, 1978.)

The importance of this is that responses to probes about what attributes are being used may measure the attributes which were used in developing a decision rule. Alternatively the responses may reflect convenient attributes to talk about, attributes which "should" be important (e.g., nutrition), or advertising playback. In summary, then, it is advisable to ask the question, "Are the respondents likely to be consciously using attributes at all?"

Alternative Methodologies

Assuming consumers can reasonably be expected to be using multiattribute models, a large variety of methodologies are available for studying the process (Table 1), all of which measure somewhat different things and which have fairly well known strengths and weaknesses. The major weakness of all of them is that they cannot directly measure what is going on inside a consumer's head. The methodologies can be classified in terms of direct or indirect in their data collection strategies. Methods for asking consumers to directly provide information include the following:

1. Elicitation can be used for determination of key attributes or importance weights, but is not useful for determining optimum levels. This method tends to measure conscious associations as recalled in memory and is affected by quality of memory, verbal ability, etc.

2. Adjective Check Lists are useful for reducing a long list of key attributes to a smaller one, and can be used to determine attribute importances.

3. Paired Comparisons of attributes can be used to determine the relative importance of pre-specified attributes, especially when graded paired judgments are collected.

4. Scaled Importance Ratings are commonly used to rate the importance of a list of attributes on a scale (e.g., 1 to 6).

5. Constant Sum Scales can also be used to get the relative importance of key attributes.

Indirect data collection procedures include:

1. Information Boards measure acquisition and hence can be used to indicate which attributes are most often used (Jacoby et al 1977). Whether this measures importance, curiosity, or failure of memory is unclear, however.

2. Eye Movements measure which attributes are being examined and hence also may indicate importance (Russo, 1977), although some obvious problems exist with this approach as well as other physiological measures.

3. Protocols can be used to elicit both which attributes are being used and how important they are. Recall protocols, however, suffer many of the problems of elicitation procedures.

4. Brand Rating and Preference Data can be used to deduce attribute importance or optimum levels by using regression analysis (Beckwith and Lehmann, 1978) or programming methods (Srinivasan and Shocker, 1973; Pekelman and Sen, 1974).

5. Multidimensional Scaling can be used to deduce both the content and importance of attributes as well as their ideal levels. However, similarity based maps may not reflect preference dimensions and the procedure is limited to a small number of attributes whose identity is always uncertain.

6. Conjoint Analysis can be used with a small number of dimensions to find both desired levels and relative importance.

CONCLUSION

Searching for key attributes and attempting to measure their relative importance and most desired levels is an important area of research. However, in many instances individuals may not be consciously using attributes in making decisions. In such cases, attempts to determine key attributes can be deceptive.

Assuming multiattribute models are appropriate or useful measures of behavior, a variety of methods are available. Which method is used depends to a large extent On the objective of the analysis. However, it seems clear that a research strategy which employs several of the methods stands a better chance of overcoming their individual weaknesses than a monolithic approach to the problem.

TABLE I

METHODOLOGIES AVAILABLE FOR STUDYING KEY ATTRIBUTES

REFERENCES

Mark I. Alpert, "Identification of Determinant Attributes: A Comparison of Methods," Journal of Marketing Research, Vol. 8, May, 1971, pp. 184-191.

Nell E. Beckwith and Donald R. Lehman, "The Importance of Differential Weights in Multiple Attribute Models of Consumer Attitude," Journal of Marketing Research, Vol. 10, May, 1973, pp. 141-145.

Nell E. Beckwith and Donald R. Lehmann, "The Importance of Halo Effects in Multi-Attribute Attitude Models," Journal of Marketing Research, Vol. 12, August, 1975, pp. 265-275.

Fishbein, Martin, "Attitude and the Prediction of Behavior," in Martin Fishbein, ed., Readings in Attitude Theory and Measurement New York, John Wiley & Sons, 1967, pp. 477-492.

Paul E. Green and Yoram Wind, "New Way to Measure Consumers' Judgments," Harvard Business Review, Vol. 53, July-August, 1975, pp. 107-117.

Jacob Jacoby, George J. Szybillo, and Jacqueline Busato-Schach, "Information Acquisition Behavior in Brand Choice Situations," Journal of Consumer Research, Vol. 3, No. 4, March, 1977, pp. 209-216.

Jacob Jacoby, Robert W. Chestnut, Wayne D. Hoyer, David A. Sheluga, Michael J. Donahue, "Psychometric Characteristics of Behavioral Process Data: Preliminary Findings on Validity and Reliability," in J. Keith Hunt, ed., Advances in Consumer Research, Vol. 5, Association for Consumer Research, 1978.

Kelvin J. Lancaster, "A New Approach to Consumer Theory, " Journal of Political Economy, Vol. 74, April, 1966, pp. 132-157.

Donald R. Lehmann, "Determinate of Choice Approach: A Research Paradigm," working paper, Columbia University, Graduate School of Business, 1978.

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

James H. Myers, "A Comparison of Two Methods for Determining Optimum Levels of Product Characteristics," in Advances in Consumer Research, Vol. 6, Urbana, Ill., Association for Consumer Research, 1978,

Jerry C. Olson and Aydin Muderrisoglu, "The Stability of Responses Obtained by Free Elicitation: Implications for Measuring Salience and Memory Structure," in Advances in Consumer Research, Vol. 6, Urbana, Ill., Association for Consumer Research, 1978,

John W. Payne and E. K. Easton Ragsdale, "Verbal Protocols and Direct Observation of Supermarket Shopping Behavior: Some Findings and a Discussion of Methods," in J. Keith Hunt, ed., Advances in Consumer Research, Vol. 5, Association for Consumer Research, 1978.

Dov Pekelman and Subrata K. Sen, "Mathematical Programming Models for the Determination of Attribute of Attribute Weights," Management Science, Vol. 20, April, 1974, pp. 217-29.

John A. Quelch, "Behavioral Measurement of the Relative Importance of Attribute-Related Information Cues: The Case of Cold Breakfast Cereals," in Advances in Consumer Research, Vol. 6, Urbana, Ill., Association for Consumer Research, 1978,

M. J. Rosenberg, "Cognitive Structure and Attitudinal Affect," Journal of Abnormal and Social Psychology, Vol. 53, November 1956, pp. 376-372.

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

V. Srinivasan and A. D. Shocker, "Linear Programming Techniques for Multidimensional Analysis of Preferences," Psychometrika, Vol. 38, No. 3, September, 1973, pp. 337-369.

William L. Wilkie, and Edgar A. Pessemier, "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, Vol. 10, November, 1973, pp. 428-441.

Peter Wright, "Consumer Choice Strategies: Simplifying Vs. Optimizing," Journal of Marketing Research, Vol. 12, February, 1975, pp. 60-67.

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