A Study of the Amount of Halo in the Perceptions of Automobiles

William L. Moore, Columbia University
William L. James, University of Alabama
ABSTRACT - Only a small amount of halo was found in attribute judgments of automobiles using a methodology similar to Beckwith and Lehmann. The amount of halo found in perceptual spaces formed through either attribute or similarity judgments was found to be about equal and low in both cases.
[ to cite ]:
William L. Moore and William L. James (1978) ,"A Study of the Amount of Halo in the Perceptions of Automobiles", in NA - Advances in Consumer Research Volume 05, eds. Kent Hunt, Ann Abor, MI : Association for Consumer Research, Pages: 481-484.

Advances in Consumer Research Volume 5, 1978      Pages 481-484

A STUDY OF THE AMOUNT OF HALO IN THE PERCEPTIONS OF AUTOMOBILES

William L. Moore, Columbia University

William L. James, University of Alabama

ABSTRACT -

Only a small amount of halo was found in attribute judgments of automobiles using a methodology similar to Beckwith and Lehmann. The amount of halo found in perceptual spaces formed through either attribute or similarity judgments was found to be about equal and low in both cases.

INTRODUCTION

Multiattribute models have been used by marketers in an attempt to explain the preferences for objects by the amount of certain attributes that the objects are perceived to possess. While the studies have found that the models generally have good predictive power, a disturbing aspect of these studies has been that respondents have a tendency to rate objects they prefer higher than expected on desirable attributes.

This tendency generally increases the within subject correlation between belief scores on different attributes and it also increases the within subject correlation between the belief scores on attributes and the preference measure. This limits the multiattribute model's value as a diagnostic tool because the dimensionality of the attribute structure of the product class is confounded, as is the analysis of brand strengths and weaknesses. Similarly, if halo effects are present in data used to build perceptual spaces, they will be distorted and attempts to position products in "halo spaces" could be dangerous.

The purposes of the paper are twofold. First, is to study the amount of halo in attribute judgments of automobiles in a manner similar to Beckwith and Lehmann (1975) to determine if their findings are replicated in a very different product category. Second, to compare the amount of halo found in perceptual spaces built with similarity judgments with spaces constructed from attribute ratings.

BACKGROUND

Psychologists have been aware of the presence of halo for over 50 years (Thorndike, 1920). Much of their work has focused on methods of measuring the amount of halo in trait ratings. This has been done by measuring the correlations between trait ratings (Symonds, 1925; Bingham, 1939; and Keaveny and McGann, 1975) and by measuring the variance of ratings across traits for a given object (Brown, 1968, and Wilkie, McCann, and Reibstein, 1973). A review of their studies is in Huber and James (1976) and won't be repeated.

Beckwith and Lehmann (1975) used a series of regression equations to study beliefs about the attitudes toward television shows. In the first equation they modeled an individual's attitude toward a T.V. program as a function of his beliefs about the program and the sample average attitude toward it

EQUATION   (1)

In the remaining equations the individual's beliefs about the attribute levels possessed by a program were modeled as functions of the sample average beliefs about the attribute levels and the individual's attitude toward the program.

EQUATION   (2)

The coefficients of the seven equations (one for attitude and one for each of the six attributes) were estimated using ordinary least squares. These equations were then estimated as a simultaneous system. While the simultaneous system was under identified for some individuals, similar parameter estimates were obtained with the two methods.

Beckwith and Lehmann found that the variation in an individual's beliefs about the attribute levels of a program could be modeled as a function of the sample average beliefs about the attribute levels and the individual's attitudes toward the program. While the regression coefficients associated with the average beliefs were usually larger than the coefficients associated with overall attitudes, they were of roughly the same size. These results indicated that a person's perception of or beliefs about an object can be explained almost as well by knowing what his attitude toward the object is as by knowing what the average perception of the object is.

There has been some criticism of the simultaneous equation methodology and behavioral interpretation (Johansson, MacLachlan, and Yalch, 1976). However, there seems to be little disagreement that there is a systematic relationship between average attribute beliefs, individual attitudes and individual attribute beliefs. This is not to imply that people know what the average beliefs are and that they consciously distort their judgments, but just that there is some systematic variation in individual attribute beliefs.

DATA

Judgments on perceptions and preferences for ten automobiles were collected from forty junior and senior marketing students. Each student rated each automobile on ten attributes using a ten point Likert scale. Similarity judgments were also collected on a ten point scale. The ten automobiles and the ten descriptive attributes are given in Table 1.

TABLE 1

LIST OF BRANDS AND ATTRIBUTES

Preference measures were collected using Pessemier's dollar metric technique (Pessemier, Burger, Teach, and Tigert, 1971). Each student was asked which one of a pair of automobiles would be purchased upon graduation if he had to choose between those two. Then he was asked how much would the price of his choice have to increase before he would change his decision.

Subjects were first screened on their stated knowledge of the cars in the survey. Then they were screened based on a lack of pattern filling out the attribute ratings. Finally, these preference judgments were converted into the best fitting linear preference scale and only those subjects whose preference judgments were significantly different from random (Bechtel, 1967, Pessemier and Teach, 1970) were kept. Six subjects were eliminated through this screening.

AMOUNT OF HALO IN ATTRIBUTE RATINGS

The first objective of this study is to determine if the results of Beckwith and Lehmann (1975) were idiosyncratic to the T.V. data, or if these findings are typical of the amount of the halo to be found in attribute judgments of other product classes. In doing this, the individual's beliefs about attribute levels of a brand were modeled as a linear function of the average beliefs about the attribute levels of the brand and the individual's preference for the brand. This is equation (2) above.

These regressions were estimated separately for each individual using OLS. The individual coefficients were aggregated across all the subjects and these aggregate results are presented in Table 2.

TABLE 2

ESTIMATED COEFFICIENTS OF THE BELIEF EQUATIONS

The average standardized regression coefficients and average absolute values of the t statistics are much larger for the belief variable than the attitude variable for every attribute. In addition to the regression coefficients associated with the average belief measures being much larger, they are also significantly different from zero a much higher proportion of the time in the individual regressions than the coefficients associated with attitude.

In contrast to the Beckwith and Lehmann study, the halo effect appears to be relatively unimportant.

HALO IN PRODUCT SPACES

If there is a halo effect biasing attribute judgments, then product spaces constructed using either discriminant analysis (Johnson, 1971) or factor analysis (Urban, 1975) would also contain halo. The effects of halo on judgments of overall object similarity has not been researched. However, it has been hypothesized (Beckwith and Lehmann, 1975) that product spaces built from similarity judgments would contain less halo than product spaces constructed using attribute judgments. This hypothesis was explored by constructing both aggregate and individual perceptual spaces then modeling the location of an object along a dimension in the individual space as a function of the object's location on the corresponding dimension in the aggregate space and the individual's attitude toward that object. This gave an equation analogous to equation (2) above.

EQUATION   (3)

Similarity Judgments

Similarity judgments for each of the 34 respondents were transformed into individual perceptual spaces using KYST (Kruskal, Young, and Seery, 1973). An aggregate similarity matrix was formed by averaging the pair-wise similarities over all the respondents. This matrix was used to form the aggregate perceptual space. Each of the individual spaces were rotated and reflected to a position of maximum congruence with the average space using a program based on Schonemann's solution to the Orthogonal Procrustes Problem (Schonemann, 1966). Then equation (3) was used to determine the relative importance one individual's preference and average perceptions in modeling individual perceptions.

Attribute Ratings

A reduced space description of the attribute ratings was formed through discriminant analysis. In this application of discriminant analysis, the brands under study were the groups and the discriminant functions were the linear combinations of the attributes that maximally separate the brands in the produce space. The location of the ith brand in the aggregate produce space X*i was found by multiplying the matrix, D, whose columns contain the discriminant vectors, times the vector of average attribute ratings B*i.

EQUATION   (4)

Similarly, the location of a brand in an individual product space, Xi, was found by multiplying the matrix D times the vector, Bi, containing that individual's beliefs about the attribute ratings for the brand

EQUATION   (5)

Equation (3) was used to determine the relative ability of individual preferences and average perceptions to model individual perceptions. These regressions were run for two dimensions in the product spaces for each individual and the results were aggregated across individuals and presented in Table 3.

TABLE 3

ESTIMATED COEFFICIENTS OF THE PRODUCT SPACE COORDINATES

In both methods of modeling perceptions, the average perceptions dominate the effect of preference. Comparing the t statistics, the average beliefs are a stronger influence in the attribute models and the effect of the preference component appears to be about the same in either case. It appears that there is little distortion due to halo in either method of constructing produce spaces and there would be little danger of positioning a product in a "halo space".

CONCLUSIONS

While these results aren't necessarily generalizable to the general public or to other product classes, this paper has shown that the amount of halo in attribute models may be quite low. These results also indicate that there may be only a very small danger of positioning some products in halo spaces.

Previous studies have indicated that certain types of attributes may be more susceptible to halo than other types: those without a clear physical analogue (Huber and James, 1976) vague, ambiguous and less important (Beckwith and Lehmann, 1975). Two more types of attributes that may give excessive halo are affective attributes (style and appearance) and the attributes that people have little knowledge of (dealer support).

It is also possible that some product classes generally have a greater degree of distortion due to halo than others. Frequently purchased goods may be routinely purchased without much thought of the attribute levels that different brands posses. When people are asked to make attribute judgments about these products, their responses may be more distorted by halo than their responses about the attribute levels of products that are purchased with more thought. Halo may be more of a problem in product classes where the differences between brands are fairly small, but much less of a problem in product classes with significant differences.

A useful area for future research is a thorough investigation of the amount of halo in attribute judgments of different produce classes. Also, while this study has shown that there is little difference in the amount of halo in attribute and similarity judgments, this topic deserves further study.

REFERENCES

Gordan G. Bechtel, "The Analysis of Variance and Pair-wise Scaling," Psychometrika, Vol. 32(March 1967), pp. 47-65.

N. E. Beckwith and D. L. Lehmann, "The Importance of Halo Effects in Multi-Attribute Attitude Models," Journal of Marketing Research, Vol. 12(August 1975), pp. 265-275.

W. V. Bingham, "Halo, Invalid and Valid," Journal of Applied Psychology, Vol. 23(1939), pp. 221-228.

Eva Brown, "Influence of Training, Method and Relationship on the Halo Effect," Journal of Applied Psychology, Vol. 52(July 1968), pp. 195-199.

Joel C. Huber and William L. James, "A Theory of Halo," working paper Krannert Graduate School of Industrial Administration, Purdue University, March, 1976.

W. L. James and F. S. Carter, "Halo Effects in Location Preferences," Paper presented at Albert Haring Symposium, April, 1976.

J. K. Johansson, D. L. MacLachlan and R. F. Yalch, "Halo Effects in Multiattribute Models: Some Unresolved Issues," Journal of Marketing Re-Marketing Research, Vol. 13(November, 1976), pp. 414-417.

R. M. Johnson, "Market Segmentation: A Strategic Management Tool," Journal of Marketing Research, Vol. 8 (February 1971), pp. 13-18.

Timothy Keaveny and Anthony McGann, "A Comparison of Behavioral Expectation Scales and Graphic Rating Scales," Journal of Applied Psychology, Vol. 60(December 1975), pp. 695-703.

J. B. Kruskal, F. W. Young and J. B. Seery, "How to Use KYST, A Very Flexible Program To Do Multidimensional Scaling and Unfolding," unpublished manuscript, Bell Telephone Laboratories, 1973.

E. A. Pessemier, P. C. Burger, R. D. Teach and D. J. Tigert, "Using Brand Preference Scales to Predict Brand Choice," Management Science, Vol. 6 (1971), pp. 371-385.

E. A. Pessemier and R. D. Teach, "Disaggregation of Analysis of Variance for Paired Comparisons: An Application to a Marketing Experiment," Institute for Research in the Behavioral, Economic, and Management Sciences, Krannert Graduate School of Industrial Administration, Purdue University, Paper No. 282, August 1970.

P. H. Schonemann, "A Generalized Solution of the Orthogonal Procrustes Problem," Psychometrika, Vol. 31(March 1966), pp. 1-9.

Percival Symonds, "Notes on Ratings," Journal of Applied Psychology, Vol. 9(1925), pp. 188-195.

E. L. Thorndike, "A Consistent Error in Psychological Ratings," Journal of Applied Psychology, Vol. 4(January 1920), pp. 25-29.

G. L. Urban, "Perceptor: A Model for Product Design," Management Science, Vol. 21(April 1975), pp. 858-871.

W. L. Wilkie, J. M. McCann and D. J. Reibstein, "Halo Effects in Brand Belief Measurement: Implications for Attitude Model Development," paper presented at the Annual Association for Consumer Research Conference, October 1973.

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

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