Dear members of Psychometric group,

What kind of experience do you have with market segmentations (MS)? Would you consider the process of creating MS to be an art or more of a science? Are you a "lumper" or a "splitter" (i.e., do you prefer to have a large or small number of segments in your solution)? What statistical methods do you employ: classical tandem approach (i.e., variance decomposition and then cluster analysis) or something more advanced like fuzzy clustering, e.g., latent class/profile analysis)? Are your classification models based on some theory or a simple data mining experience?

I definitely consider MS to be more of a science: classification, after all, is another form of scaling, and i'd rather have a scale that is reliable and valid than "beautiful" and "fancy"! I am also more of a "splitter" myself, but often rely on statistics to identify the optimal number of segments (e.g., the Cubic Clustering Criterion). A choice of the method often depends on such factors as a number of variables and purpose of the segmentation: are we trying to segment the general population or identify number of segments that "drive" certain behavior?