Priori Predictive Model

In the class of a priori and descriptive, the methods such as cross tabulation, more specifically contingency table and log-linear models included, while the second class of priori and predictive includes regression, discriminant analysis and limited dependent variable models. In the last two classes mostly more elaborated models such as clustering, non-overlapping and AID, mixture models are included. So, let us go deeper into the four classes and their distinctions with details.

A Priori descriptive methods

As mentioned before according to this type of methods, the type and number of segments are determined before data analysis. Cross tabulation approach has been a popular technique for describing the segments in the earlier years of

In a-prior predictive category, a-priori descriptive segments based on one set of criteria, and the subsequent use of predictive models to describe the relation between segment membership and a set of independent variables. (Wilkie & Cohen , 1977) developed two approaches: (1) forward approach, in which background characteristics such as demographics and psychographics are firstly used to determine a-priori segments and then further analysis can be conducted on the segments to reveal purchase behavior and (2) backward approach, on contrary the segments firstly formed based on the purchase related variables then the demographics or psychographics are applied to capture general consumers’ characteristics. The most popular methods in a priori predictive category are discriminant analysis and limited dependent variable models, including logit models. The logit models have been increasingly used in market segmentation with the growing availability scanner panel data. (Imran, 1981) developed a hybrid approach, in which clustering and multinomial logit models are combined, more specifically he firstly identified the segments by employing clustering, once the segments identified from each basis logit model is applied explaining the customer choices as functions of their

The post-hoc predictive methods are based on estimated relationship between a dependent variable and a set of predictors. The traditional method in this field is Automatic Interaction Detection (AID) model, which identifies interactive effects of categorical segmentation bases on a dependent variable, such as a measure of purchase behavior, in other words, it divides the market into groups along according to the dependent variable, which is actually characterized by independent variables, normally socioeconomic and demographic variables. Another interesting model, increasingly used in market segmentation practice is Artificial Neural Network (ANN) analysis. The attempts to imitate neural net’s function and use them in marketing, more specifically in market segmentation practices have been attractive up until now. Conjoint analysis has been the center of the category due to its ability in grouping of costumers according to how they respond to product features (Wedel & Kamakura, 2000). An extension of conjoint analysis, componential segmentation has been considered recently, in the segmentation both of product and respondent profiles are analyzed. Although the approach successfully explains the customer’s response to product features, it lacks in validity and reliability of the idiosyncratic important values,