Uses of Multidimensional Scaling

There are many possible uses of this method of scaling; new applications are being developed steadily. Some of the uses that have already been identified are the following:

Market segmentation: If brands are located as points in preference space, as in the example and consumers ideal points are similarly located, market segments may then be viewed as subspaces in which consumers have similar ideal positions and perceive the brands similarly.

Product life cycle: By analyzing respondent perceptions at different times, researchers may be able to relate movement along various dimensions (characteristics) to some measure such as market share, and, thus develop a new concept of product life cycle.

Vendor Evaluations: Industrial purchasing agents must choose among vendors who differ – for example, in price, delivery, reliability, technical service and credit. How purchasing agents summarize the various characteristics to determine a specific vendor from whom to purchase would be information that would help vendors design sales strategies.

Advertising Media Selection: Which magazines should be used for an advertising campaign to reach a given audience? Different possible media could be identified aspects in similarity pace (as were cars in the example) and members of their audiences located as ideal points. This would be similar to the market segmentation process. A similar approach might be taken to locate the best media for specific ads.

Limitations of multinational Scaling:

The above discussion, it has been suggested that many problems exist in the use of multidimensional scaling and must be considered significant limitations in use of the technique. These limitations can be classified in the following three categories:

Conceptual Problems: Definitions of ‘similarity’ and ‘preference’ that are conceptually clear and that can be communicated accurately to respondents have not been developed and may not be achievable. Criteria in which similarities are gauged may vary during an interview with respondents; they may vary by the context in which respondents think such as purchase for themselves or as a gift; and small variations in one criterion may be more important than large variations in another. None of these factors is fully understood or amenable to control at present.

Current studies assume that each stimulus (brand, in the example) is an only choice. If the preferred brand were not available, it is assumed the summer would take the second choice. But what if the consumer purchases two items—for example two cars? After the first choice has been bought, will the order of preference change for the second car?

How do preferences change over time? Do they change frequently or are they relatively stable? The answers in these questions have much to do with the operational use a firm can make of a multidimensional analysis.

Empirical Problems: In the discussion of the multidimensional scaling process, it was pointed out that the labeling of the various dimensions (criteria) of importance to respondents is subjective and, hence, open to question.

The data collection process is as open to bias in multidimensional scaling projects as in any other but the relative impact of such biases is less well known. Procedure for collecting data and the general back ground conditions or ‘scenarios’ in which a project is presented has yet to be standardized.

The example presented here was for one respondent. The problem of aggregating responses over a large group of people has not been worked out.

Computational Problems: All analyses of the type discussed here require computer programs. Several different ones are used; but it is known how, if at all, they vary in results according to variation in such inputs as number of points and experimental error.

Most computer programs assume the measurements between points that is, the distances between items that are seen as different are straight line or linear distances. Computer programs are available that use different distance functions, but researchers do not know what functions are appropriate. As a result the linear assumption is made.