Using Multivariate Adaptive Regression Splines (MARS) to Find Interactions of Socio-Demographics That Model Individual Differences in Australian Farmers Purchase Behavior

• Activity Number: 467
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Marketing
• Author(s): Ingo Bentrott*+
• Companies: University of Technology Sydney
• Keywords: Data Mining ; MARS ; Discrete Choice ; Interactions ; Heterogeneity ; Marketing

Abstract

Socio-demographics play a major role in accounting for preference heterogeneity and market segmentation in discrete choice models. The use of demographic segments to account for heterogeneity in choice models has been proposed by Ben-Akiva & Lerman (1985) and complex models such as random coefficients logit have been used to account for unobserved differences in preferences. To enter demographics into a repeated choice stated preference model, they must be interacted but, due to the complexity of finding and modelling socio-demographic interactions (McLelland & Judd 1993), the interactions are often restricted to simple terms that act global over the data space. We use MARS to overcome the difficulties associated with detecting and integrating socio-demographic interactions in localized areas of the data space. In our study, heterogeneity that exists amongst farmers can be accounted for by localized interactions of the observable demographics with the experimentally designed choice attributes using basis function found by MARS. The MARS basis functions are hybrid into a conditional logit model that outperforms a hybrid of the MARS basis functions in a random coefficients logit.