Explaining consumer choice through neural networks: The stacked generalization approach

Abstract

This paper uses neural network models and the ensemble technique of stacked generalization to investigate the relative importance of situational and demographic factors on consumer choice. We discuss the theoretical justification for this approach and develop the level-0 and level-1 models which we estimate using a consumer choice data set from AT&T. The findings confirm the superiority of situational factors in explaining the consumers’ choice of communication modes and support the strength of stacked generalization. Important clues that emerge from the analysis regarding the factors to be accounted for when developing efficient neural network structures are also discussed.