Abstract:

It has become commonplace to note the growing use of sophisticated techniques in processing market research: cluster analysis and segmentation analysis, as well as factor analysis, discriminant analysis, canonical analysis, multivariate analysis of variance and, more recently, multidimensional scaling of perception and preference. Naturally, the results of these different techniques are of more or less satisfactory practical use depending on the degree of skill and acumen of the researcher who uses them. We feel however that disillusions are particularly high in the field which interests us, that which covers the uses of cluster analysis and segment analysis programs. We will attempt here to discern the cause of these disillusions and, if possible, to define a more judicious use of these methods until such time as the statistician perfects them, making them easier for the practician to employ. The analysis of the practical problems encountered in using cluster analysis and segment analysis programs in our company prompted us to sort these difficulties into three classes, which form the structure of my paper. Class 1 comprises the difficulties involved in choosing a method in consideration of the objectives of the research effort. Class II is connected with the methods themselves, and more particularly with the invariance of the results obtained. Class III is connected, we find with a certain ambiguity of the notion of segment as understood by the marketing man on the one hand and the statistician on the other. After dealing with these three classes, I shall conclude by examining a few means now available to get around these difficulties.