Segmentation and the Research Process

Segmentation studies are among the most difficult ones to undertake regardless of which segmentation model is used. There are difficulties in terms of the sample design, the reliability of the data collected, data collection methods, and data analysis as well as in the interpretation of the results.

Segmentation studies seek not only to segment a sample on the basis of “differences” but also to project the findings to the appropriate universe. The latter often poses a difficult problem in that the unit of analysis is not always clearly specified. In studies involving consumer goods, respondents are typically housewives despite the fact that much household purchasing involves other members of the family to varying degrees. This unit-of-analysis problem is also a problem in the study of purchase behavior relating to industrial goods where multiple purchase influences abound.

In addition to the above problem, there is a usually a sampling problem because often quota samples are used despite their obvious limitations. Most segmentation studies also ignore the non-response problem although there is a possibility that there are significant differences between respondents and non-respondents. The increasing costs of doing research encourage the use of quota samples and the ignoring of the non-response problem. Under such conditions it can be argued that the researcher would be well advised to use small probability sample that is well controlled with respect to callbacks and includes a study of non-respondents.

Every segmentation study has the problem of defining, in an operational sense, the dependent and independent variables. In product attribute studies there is the question of how to define or describe product characteristics in a way that will mean the same to all respondents. Should these be presented verbally, pictorially, or how? How should importance be measured? If a scale is used, how many points should be used and how should the scale be anchored? Should variables be ranked using verbal descriptions (very important) or rank ordered in sequence of importance? These kinds of decisions will impact on the number and type of segments that emerge from such studies, and yet there no ready answers to these questions.

Few segmentation studies have concerned themselves with the question of data reliability. Much the same can be said to apply to segment stability. How stable will the segment be over time? The latter is a more difficult subject to come to grips with since stability is a function of changes in the marketplace as well as in the consumer’s perception of the relevant products. Because of these problems, it is difficult to know the validity of a segmentation study.

Segmentation studies have used a variety of data collection methods, including personal interviews, mail, and telephone. Which one is best in terms of a balancing of reliability and costs is a moot question. Much, of course, depends on the situation and what kinds of data are sought. If pictures are used to portray a new product concept, then the telephone method is ruled out. If it is thought necessary to obtain the uncontaminated views of a respondent, then the use of a mail survey is precluded. Computer based telephone interviewing systems have increased the efficiency of telephone studies so that the telephone may be used even when it may not provide the greatest data reliability.

An ever increasing number of models are available to analyze data derived from market segmentation studies. Typically, on has to understand a variety of techniques in order to decide which are most appropriate to classify respondents into segments (techniques ranging from simple cross tabulations to multidimensional scaling) and then to discriminate among segments on the basis of respondent profiles.

Unique to segmentation studies is the need to apply a variety of analytical procedures in tandem. Most segmentation studies involve complex designs revolving around several hybrid bases for segmentation. However because one cannot know in advance which basis for segmentation will lead to the identification of meaningful segments, segmentation studies should be flexible, allowing diverse analyses aimed at the identification of relevant segments. This need creates special demands for researchers with knowledge of a large number of analytical procedures, good conceptual understanding of alternative segmentation models, and a high level of research creativity. —