Brand Perceptions: Relative vs. Absolute Ratings
When consumers rate brands on particular product attributes, their ratings can be influenced by the brands they are asked to evaluate.
Background:
Studies are often conducted to determine how brands are perceived on relevant attributes. It is sometimes assumed that these perceptions are given in an absolute sense. That is, the rating for one particular brand is independent of other brands rated with it. Thus, when perceptions of a large set of brands are desired, the assumption is made that questionnaires can be versioned to obtain ratings of subsets of brands, without any confounding effects
This paper reports the results of a survey that investigated if brand perceptions are influenced by the set of brands chosen for evaluation.
Research Method:
Market Facts, Inc. investigated the issue of brand ratings using Consumer Mail Panel and its Data Gage service. Each of four versions of a questionnaire was mailed to a sample of 1,000 households. Each panel was "balanced" to reflect U.S. Census statistics concerning geographic location, age of female head, household income, and population density.
Respondents were asked to rate the Chevrolet Nova on an 8-point agree/disagree scale on its perceived gas mileage. In one version, only the Chevrolet Nova was rated. In the other versions, Chevrolet Nova was rated with other cars. This provided an opportunity to measure rating biases, if any, under a variety of conditions. The brands selected to berated were:
| Version | Condition | Brands Rated |
| 1 | Base case | Only Chevrolet Nova was rated |
| 2 | Expected good mileage cars | Chevrolet Nova, Toyota Corolla, Ford Pinto, Volkswagen Rabbit (Diesel), Honda Civic |
| 3 | Expected poor mileage cars | Chevrolet Nova, Lincoln Continental, Buick Electra, Chrysler New Yorker, and Cadillac Eldorado |
| 4 | Combination of expected good and poor mileage cars | Chevrolet Nova, Ford Pinto, Cadillac Eldorado, Volkswagen Rabbit (Diesel), and Lincoln Continental |
General examples of the specific versions were as follows:
Version 1:
Please indicate the degree to which you agree or disagree with the statement by placing an "X" in one of the eight boxes to the right of the car. "X" only one box keeping in mind that "8" means you "agree strongly" and "1" means you "disagree strongly,"
Version 3:
Below are listed five cars. Please indicate the degree to which you agree or disagree with the statement by placing an "X" in one of the eight boxes to the right of each car. "X" only one box for each car keeping in mind that "8" means you "agree strongly" and "1" means you "disagree strongly." Please note any subtle differences you may have between cars.
The wording for Versions 2 and 4 was the same as Versions 3 with the exception of the cars being rated.
There was approximately an 80% return rate per version. All versions were tabulated and a one-way analysis of variance was performed to determine if there were any significant differences between versions in the Chevrolet Nova ratings.
Results:
The one-way analysis of variance was significant at the 99% confidence interval. Examination of the mean ratings (1 = disagree very much and 8 = agree very much) shown in the following chart reveals the differences in the four versions. When the Chevrolet Nova is rated by itself it obtains a similar rating as it does when it is rated with cars giving good gas mileage (Version 2). However, when the Chevrolet Nova is rated with cars giving relatively poor gas mileage it obtains a much higher rating (Version 3) and a somewhat higher rating when it is rated with a combination of good and poor gas mileage cars (Version 4). All contrasts between the Chevrolet Nova ratings were significant at the 99% confidence intervals using t-tests with the exception of the contrast between Version 1 and Version 2.
Further examination of the results obtained from Version 4 is useful. Both good mileage cars received better ratings (Pinto and Rabbit), and one of the poor mileage cars (Continental) received a worse rating than when those same cars were evaluated in Versions 2 and 3.
Discussion and Conclusions:
It has been shown that a respondent's evaluation of a brand can be influenced by the other brands rated. It is conceivable that a different set of cars with comparable gas mileage in Version 4 could have had the same influence on the ratings of the Chevrolet Nova. But this could lead one to falsely conclude that versioning is possible.
There are two basic problems with versioning. First, perceptions are collected on more than one variable. Creating balanced sets of brands based on their expected values on several variables is very difficult. But, more importantly, the expected values for many brands are not known. Indeed, getting this information is generally the objective of the study.
Extreme caution should be used comparing ratings of particular brands if they are obtained under different conditions. If the respondent's task is to rate several brands, consideration must be given to the brand mix. This is particularly true if the study is to be replicated. Thus, whenever possible, the set of brands should be kept the same from one study to the next.
