The scaling methods discussed above all enable the researchers to measure to some degree consumer attitudes toward products and brands. In general, they permit one to determine such things as which brand is perceived by consumers to be more economical than another, less glamorous, more old fashioned and so on. Two or more brands can thus be compared on as many different characteristics as the researchers consider significant or is imaginative enough it bring into consideration. But the various brands are measured against each characteristic one at a time. The measurement process tells little about the relative importance of the different characteristics or how the characteristics relate to each other in reference to the various brands.
Multidimensional scaling is a term used to describe a group of analytical techniques used to study consumer attitudes particularly relating to perceptions and preferences. These techniques attempt it identify the product attributes that are important to consumer and to measure their relative importance. They are useful in studying questions of the following types:
(1) What are the major attributes of a given class of products (e.g. soft drinks, fabric softeners, modes of transportation) that consumers perceive in considering the product and by which they compare different brands of the product?
(2) Which brands compete most directly with each other? Which the least directly?
(3) Would consumers like a new brand with a combination of characteristics not found in the market?
(4) What would be the consumer’s ideal combination of product attributes?
(5) What sales and advertising messages are compatible with consumers brand perceptions?
An example may clarity the idea. suppose consumers are asked it compare each of a group of cars with each of the others and to specify the two they perceive as being the most similar, the pair that is next most similar, and so on. Respondents are to use any criteria they choose. For the following 11 cars this would mean ranking on similarity all 55 possible pairs.
1. Ford Escort 7. Jaguar Sedan
2. Ford Taurus 8. Honda Accord
3. Lincoln Town Car 9. Chevrolet Monte Carlo
4. Buick Riviera 10. Oldsmobile Cutlass
5. Volkswagen 11. Dodge Aries
This could be done by putting each of the pairs on one of 55 cards. The respondents could separate these into two groups – those that have pairs that tend to be similar and those with pairs that tend to be different. He could then take the ‘similar’ pile and separate it into those that are very similar and those not so similar. Within one of these groups, the respondent would then choose the most similar the next most, and so on. In this step-by-step procedure a complete ranking on similarity would eventually be obtained.
Analysis of these results by multidimensional methods is at a level of sophisticated beyond this article. In general, the procedure requires the power of a computer to be practical. A number of computer programs are available for the purpose. Three basic questions are typically considered:
(1) How many dimensions (product attributes) underlie the consumer’s perception of the cars?
(2) What is the actual configuration of the consumer’s perceptions of the brands – which cars are the most alike and which the least?
(3) What are the actual attributes underlying the configuration?
The general analysts can be seen by following the car example in an overly simplified form.
The computer programs proceeds with the analysis of the first questions by testing the data to see if there is a configuration of points on one dimension that fits the rank order data satisfactorily. Dimensions as referred to here relate to product attributes, such as price, fuel economy, and performance but it is important to note that which attributes are actually involved is not known. If one dimension does not provide a fit, two dimensions are tried; if these do not give a satisfactory configurations, three dimensions are tried, and so on. As the fitting of a configuration to the data is a mater of degree, the researchers must trade off goodness of fit against a large number of dimensions. a lack of fit index called stress is calculated by the computer program. Rule of rule policies or acceptable levels of this index have been established, but the general goals are to find a reasonably small number of dimensions that will eliminate most of the stress.