Halo Effects and Location Preferences

William L. James, University of Alabama
Forrest S. Carter, Georgia Institute of Technology
ABSTRACT - The two methods of measuring halo that have been used in marketing were compared and were found to be highly correlated. Some evidence was found to support the hypothesis that halo decreases as preference increases. Halo was not found to increase as familiarity decreases. Halo was found to be a more important source of error than familiarity. Attributes with less clearly defined physical correlates were found to have more halo than those attributes with clear physical correlates.
[ to cite ]:
William L. James and Forrest S. Carter (1978) ,"Halo Effects and Location Preferences", in NA - Advances in Consumer Research Volume 05, eds. Kent Hunt, Ann Abor, MI : Association for Consumer Research, Pages: 474-476.

Advances in Consumer Research Volume 5, 1978      Pages 474-476

HALO EFFECTS AND LOCATION PREFERENCES

William L. James, University of Alabama

Forrest S. Carter, Georgia Institute of Technology

ABSTRACT -

The two methods of measuring halo that have been used in marketing were compared and were found to be highly correlated. Some evidence was found to support the hypothesis that halo decreases as preference increases. Halo was not found to increase as familiarity decreases. Halo was found to be a more important source of error than familiarity. Attributes with less clearly defined physical correlates were found to have more halo than those attributes with clear physical correlates.

INTRODUCTION

The focus given to attitude research in marketing and the behavioral sciences has led to a renewed interest in halo effects. The halo effect was noticed as early as 1907, but was named by Thorndike (1920). There have been two slightly different ways in which halo has been conceptualized. Tiffin and McCormick (1965) in their classic textbook view halo as the domination of all traits by a particular trait. A more common view is that halo occurs when an overall impression, such as preference, dominates the traits being rated. This latter view appears to dominate in the marketing literature. The concept of halo is important to marketing researchers due to its effect on the prediction of preference, the development of product spaces, and the identification of marketing opportunities. Any time that attribute ratings are collected the researcher must consider the potential impact of halo on the relationships being investigated. Halo might be beneficial if the objective is to predict preference from attribute ratings since it will increase the correlation between the attributes and preference. Halo is, however, detrimental when attribute scores are used to develop product spaces or identify marketing opportunities. It would be a mistake to develop or reposition a product to fit some set of consumer ratings that are heavily influenced by halo. Thus, it is necessary to measure the degree to which attribute ratings are influenced by halo in a given situation. If marketers could identify those consumers who exhibit high levels of halo in attribute judgments it might be possible to determine their susceptibility to advertising and other forms of marketing activity.

BACKGROUND AND PERSPECTIVE

The first empirical research on measuring halo appears to have been done by Symonds (1925). He used partial correlations between traits to demonstrate that halo raised the correlations of trait ratings by .245 on the average. He speculated that large halo in a trait occurs if a trait is not easily observed or if it is not clearly defined. This line of research led to the view of halo as the excessive correlation between attributes. Unfortunately, as Bingham (1939) points out, some correlation between attributes is to be expected, and there seems to be no clear cut method to determine when the correlation is excessive. This approach usually led to factor analytical solutions where the first factor extracted was assumed to be a measure of overall attitude. The trait intercorrelations were then studied after this factor had been removed. Later research using this definition, such as Keaveny and McGann (1975), focused on average correlations between attributes and attempted to lower these intercorrelations through instructions and training of raters.

A second approach to the measurement of halo is due to Guilford (1954) who viewed halo as an interaction between rater and object. Guilford (1954) defined several different types of errors which can occur when human raters are used. These errors are halo, leniency error, logical error and rater-trait error. Leniency error is defined as the tendency of a subject to overvalue or undervalue objects in general. Logical error occurs when subjects give similar ratings to objects on traits that appear to the subject to be closely related. Rater-trait error is defined as the tendency of a subject to overvalue or undervalue an object on a particular trait due to a favorable or unfavorable attitude towards the object. An analysis of variance design is then used to test for the rater by object interaction. This definition will overestimate halo to the extent that there are factors other than preference contaminating attribute judgments. An advantage of this approach is that a lack of a significant interaction term indicates the absence of halo.

A third approach to the measurement of halo, used by Brown (1968), focuses on the variance of ratings of an object across attributes. The higher the variance in ratings for an object, the less halo possessed by the object. It is important to note that this approach assumes that all attributes for a given subject are scored so that the subject views more as better. Thus, this approach requires a judicious choice of attributes or the measurement of preference for each subject so that it can be determined whether any recoding is in order for the halo measure. Unfortunately this measure does not indicate how little variance is needed for halo to be present. Wilkie and McCann (1972), and Wilkie, McCann, and Reibstein (1973) were the first to apply this approach in a marketing context. They found, among other things, that instructions could increase variance, that higher variance was found for the most preferred brand than for the other brands, and suggested that low brand familiarity contributes to low variance in ratings. Koltuv (1962) found that as familiarity decreases, halo increases, and that as halo increases the relevance of the trait increases. This is the only approach which has been used by both psychologists and by marketers.

A fourth approach to the study of halo is a regression model by Beckwith and Lehmann (1975) which has only appeared in the marketing literature. Their general findings agree with those from the industrial psychology literature. The most important contribution of this model is that it allows the assessment of the degree of halo exhibited by individual attributes for each subject. This approach views beliefs as being composed of a true belief rating, estimated from average belief scores, and halo error, estimated using preference. The model for an individual subject is given below:

Bij = b1B-ij + b2Pi + uij   (1)

where:

Bij is the belief rating for object i on attribute j

B-ij = is the average belief rating for object i on attribute j

Pi is the preference for object i

b1 is the importance of the true value of the object on attribute j

b2 is the measure of degree of halo for the object on attribute j

uij is random error

All variables are standardized.

PURPOSE OF THE STUDY

The present study follows up on the Beckwith and Lehmann (1975) and the variance approaches. Specifically this paper investigates four areas:

(i) The relation of the variance measure to both familiarity and preference within a subject.

(ii) The convergence of the variance approach and the Beckwith and Lehmann (1975) approach.

(iii) Whether halo is greater for attributes with less well defined physical correlates.

(iv) Whether halo effects can be separated from misperceptions due to a lack of familiarity.

DATA BASE

Fourteen masters students rated seventeen cities on nine attributes plus familiarity. Then preference for the cities as places to work was measured using a technique described by Huber and James (1976), resulting in dollarmetric scale values of preference. The attributes were measured on 10 point Likert scales. The cities were selected from a list of 40 used in a pretest The attributes and the dollarmetric scale were also pretested. In the variance analyses all attribute ratings were rescaled whenever necessary so that for each subject all attributes had positive correlations with preference. The nine attributes rated were: population; spectator sports; opportunities for indoor recreation; opportunities for outdoor recreation; cultural activities; warmth of the climate; pleasantness of the summer; pleasantness of the winter; and snowfall.

RESULTS

(i) Within subject analyses using the variance measure are presented in Table 1:

TABLE 1

CORRELATIONS BETWEEN PREFERENCE, FAMILIARITY, AND VARIANCE

Since there are 14 subjects and only a few have significant correlations a reasonable way to test whether relationships exist is to perform a binomial test. Wilkie, McCann, and Reibstein (1973) found that the lowest preference had the lowest variance, therefore there are 3 correlations out of 14 that are significant at the .05 level for a one tailed test. The probability of 3 or more successes out of 14, given the probability of an individual success is .05, is .0300. Thus, the results are highly unlikely to be random perturbations. Rather, it is reasonable to conclude that a positive relationship exists between preference and variance. To test whether preference and familiarity are positively correlated we find that there are 4 significant positive correlations at the .05 level for a one tailed test. The probability of 4 successes out of 14 is only .0041. Thus it is unlikely that these came about by chance, and it can be concluded that preference and familiarity are positively correlated. To test whether halo is due to misperceptions of the product due to a lack of familiarity we note that there are 2 significant positive correlations at the .05 level for a one tailed test. The probability of 2 successes out of 14 is .1529, and is thus, likely to have occurred on a chance basis alone. Consequently the table does not offer any evidence that for this product there is a relationship between halo and familiarity.

(ii) The degree of convergence between the variance approach and the Beckwith and Lehmann (1975) approach was assessed by converting the halo measures, b2 in equation (1), into an overall measure for each subject. The overall halo measure was simply the sum of the absolute values of the individual b2. This measure of overall subject halo was then correlated with both the mean variance of a subject's ratings, and with the variance of the variance measure. The correlation of halo with the mean variance was -.46 (p = .095) and is in the expected direction. The variance of the variance measure was expected to be a better measure since it measured the dispersion rather than the absolute level of variance in product ratings. This variance of the variance measure was correlated -.65 (p =.011) with halo, and is in the expected direction. Thus, the two different halo measures agree and provide convergent measures of halo.

(iii) In order to test whether those variables with less clearly defined physical correlates exhibited greater halo, the Beckwith and Lehmann (1975) measure was used. Table 2 shows the mean and variance for halo scores by attribute and by grouping:

TABLE 2

HALO BY ATTRIBUTE

Population, spectator sports, warmth of climate, and the amount of snowfall have clear physical correlates and were pooled for this analysis. Indoor recreation, outdoor recreation, cultural activities, and pleasantness of the summer were pooled as attributes with less clearly defined physical correlates. When these two groups are compared, the less physical attributes had significantly higher halo scores (t = 5.03) at the .005 level. If pleasantness of the winter is pooled with the less physical attributes, the difference (t = 4.26) is still significant at the .005 level.

(iv) In an attempt to separate halo from perceptual error, familiarity was added to equation (1). It is assumed that familiarity reflects possible misperceptions due to random nonrepresentative experiences with the object. Table 3 shows the beta weights for a typical subject.

TABLE 3

HALO VERSUS RANDOM EXPERIENCES EXPLANATION

The analysis indicates that familiarity is usually less important than halo as a contributor of error, and that for this product it is not tenable to attribute most of the halo to a lack of familiarity with the cities in the questionnaire.

CONCLUSION

The most important conclusion is the finding that both the variance measure, used fairly extensively in psychology, and the Beckwith and Lehmann (1975) measure correlate highly. The finding that halo was higher for variables with less clearly defined physical correlates implies that in order to change attitudes prior to trial of a product it may be necessary to focus on those attributes with clearly physical correlates. The low amount of halo found in this study indicates that there is wide variation in the amount of halo in different types of products, and that until more is known about which types of products are heavily influenced by halo each product will have to be investigated on its own.

REFERENCES

Neil Beckwith and Donald Lehmann, "The Importance of Halo Effects in Multi-Attribute Attitude Models," Journal of Marketing Research, 12 (August, 1975), 265-275.

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

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

J. P. Guilford, Psychometric Methods. New York, McGraw-Hill, 2nd edition, 1954.

Joel Huber and Bill James, "The Monetary Worth of Physical Attributes: A Dollarmetric Approach," paper presented at the Seventh Annual Attitude Research Conference, Hilton Head, South Carolina, on February 12th, 1976.

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

Barbara Koltuv, "Some Characteristics of Intrajudge Trait Intercorrelations," Psychological Monographs, 76, Whole No. 552, (no. 33, 1962).

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

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

Joseph Tiffin and Ernest J. McCormick, Industrial Psychology, Prentice-Hall, Englewood Cliffs, New Jersey, 5th edition, 1965.

William Wilkie and John McCann, "The Halo Effect and Related Issues in Multi-Attribute Attitude Models B An Experiment," Paper No. 377, Institute for Research in the Behavioral, Economic and Management Sciences, Krannert Graduate School of Industrial Administration, Purdue University, October, 1972.

William Wilkie, John McCann and David Reibstein, "Halo Effects in Brand Belief Measurement: Implications for Attitude Model Development," Proceedings, Fourth Annual Conference, Association for Consumer Research, November, 1973.

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