Clusterwise regression and market segmentation : developments and applications

The present work consists of two major parts. In the first part the literature on market segmentation is reviewed, in the second part a set of new methods for market segmentation are developed and applied. Part 1 starts with a discussion of the segmentation concept, and proceeds with a discussion on marketing strategies for segmented markets. A number of criteria for effective segmentation are summarized. Next, two major streams of segmentation research are identified on the basis of their theoretical foundation, which is either of a microeconomic or of a behavioral science nature. These two streams differ according to both the bases and the methods used for segmenting markets. After a discussion of the segmentation bases that have been put forward as the normative ideal but have been applied in practice very little, different bases are classified into four categories, according to their being observable or unobservable, and general or product- specific. The bases in each of the four categories are reviewed and discussed in terms of the criteria for effective segmentation. Product benefits are identified as one of the most effective bases by these criteria. Subsequently, the statistical methods available for segmentation are discussed, according to a classification into four categories, being either a priori or post hoc, and either descriptive or predictive. Post hoc (clustering) methods are appealing because they deal adequately with the complexity of markets, while the predictive methods within this class (AID, clusterwise regression) combine this advantage with prediction of purchase (predisposition). Within the two major segmentation streams, segmentation methods have been developed that are specifically tailored to the segmentation problems at hand. These are discussed. For the microeconomic school focus is upon recently developed latent class approaches that simultaneously estimate consumer segments and market characteristics (market shares, switching, elasticities) within these segments. For the behavioral science school focus is on benefit segmentation. Disadvantages of the traditional two-stage approach, in which consumers are clustered into segments on the basis of benefit importances estimated at the individual level, are revealed and procedures that have been addressed to one or more of these problems are reviewed. In Part 2, three new methods for benefit segmentation are developed: clusterwise regression, fuzzy clusterwise regression and generalized fuzzy clusterwise regression. The first method is a clustering method that simultaneously groups consumers in a number of nonoverlapping segments, and estimates the benefit importances within segments. The performance of the algorithm on synthetic data is investigated in a Monte Carlo study. Empirically, the method is shown to outperform the two-stage procedure. Special attention is paid to significance testing with Monte Carlo test procedures, and convergence to local optima. An application to segmentation of the meat-market in the Netherlands on the basis of data on elderly peoples preferences for meat products is given. Three segments are identified. The first segment weights sensory quality against exclusiveness (price), in the second segment quality is traded off against fatness. This segment, comprising predominantly of females, had the best knowledge of nutrition. In the third segment preference is based on quality only. Regional differences were identified among segments. Fuzzy clusterwise regression extends clusterwise regression in that it allows consumers to be a member of more than one segment. It simultaneously estimates the preference functions within segments, as well as the degree of membership of consumers in those segments. Using synthetic data, the performance of the method is evaluated. Empirical comparisons with two other methods are provided, and the cross-validity of the method with respect to classification and prediction is assessed. Attention is given in particular to the selection of the appropriate number of segments, the setting of the user defined fuzzy weight parameter, and Monte Carlo significance test procedures. An application to data on preferences for meatproducts used on bread in the Netherlands revealed three segments. In the first segment, taste and fitness for common use are important. In the second segment, taste overridingly determines preference, but products that are considered more exclusive and natural and less fat and salt are also preferred. In segment three the health related product benefits are even more important. The importance of taste decreases from segment one to three, while the importance of health-related aspects increases in that direction. The health oriented segments comprised more females, older people and people who attributed causality of their behavior more to themselves. The method was also applied to data on consumers image for stores that sell meat. Again three segments were revealed. The value shoppers, trade off quality and price. They come from smaller families and spend less on meat. In the largest segment store image is based upon product quality. Females have higher membership in this segment, that is more involved with the store where they buy meat. For service shoppers, both service and atmosphere are important. This segment tends to be more store-loyal. Next, a generalization of fuzzy clusterwise regression is proposed, which incorporates both benefit segmentation and market structuring within the framework of preference analysis. The method simultaneously estimates the preference functions within each of a number of clusters, and the parameters indicating the degree of membership of both subjects and products in these clusters. The performance of this method is assessed in a Monte Carlo study on synthetic data. The method is compared empirically with clusterwise regression and fuzzy clusterwise regression. The significance testing with Monte Carlo test procedures, and the selection of the fuzzy weight parameters is treated in detail. Two segments were revealed in an analysis of consumer preferences of butter and margarine brands. The segments differed mainly in the importance attached to exclusiveness and fitness for multiple purposes. The brands competing within these segments were revealed. Females and consumers with a higher socioeconomic status had higher memberships in the segments in which exclusiveness was important. Finally, the clusterwise regression methods developed in this work are compared with other recently developed procedures in terms of the assumptions involved. The substantive results obtained in the empirical studies concerning foods are summarized and their implications for future research are given. The implications and the contribution of the methods to the development of marketing strategies for segmented markets are discussed.