Neural networks and the multinomial logit for brand choice modelling: a hybrid approach

The study of brand choice decisions with multiple alternatives has been successfully modelled for more than a decade using the Multinomial Logit model. Recently, neural network modelling has received increasing attention and has been applied to an array of marketing problems such as market response or segmentation. We show that a Feedforward Neural Network with Softmax output units and shared weights can be viewed as a generalization of the Multinomial Logit model. The main difference between the two approaches lies in the ability of neural networks to model non-linear preferences with few (if any) a priori assumptions about the nature of the underlying utility function, while the Multinomial Logit can suffer from a specification bias. Being complementary, these approaches are combined into a single framework. The neural network is used as a diagnostic and specification tool for the Logit model, which will provide interpretable coefficients and significance statistics. The method is illustrated on an artificial dataset where the market is heterogeneous. We then apply the approach to panel scanner data of purchase records, using the Logit to analyse the non-linearities detected by the neural network. Copyright © 2000 John Wiley & Sons, Ltd.

[1]  Wilfred W. Recker,et al.  The Multinomial, Multiattribute Logit Choice Model , 1979 .

[2]  R. Palmer,et al.  Introduction to the theory of neural computation , 1994, The advanced book program.

[3]  Donald G. Morrison,et al.  Making the Cut: Modeling and Analyzing Choice Set Restriction in Scanner Panel Data , 1995 .

[4]  J J Hopfield,et al.  Learning algorithms and probability distributions in feed-forward and feed-back networks. , 1987, Proceedings of the National Academy of Sciences of the United States of America.

[5]  J. Lattin.  A Model of Balanced Choice Behavior , 1987 .

[6]  Harald Hruschka,et al.  Determining market response functions by neural network modeling: A comparison to econometric techniques , 1993 .

[7]  Allan D. Shocker,et al.  Consideration set influences on consumer decision-making and choice: Issues, models, and suggestions , 1991 .

[8]  Rick L. Andrews,et al.  Studying Consideration Effects in Empirical Choice Models Using Scanner Panel Data , 1995 .

[9]  G. Tellis.  Advertising Exposure, Loyalty, and Brand Purchase: A Two-Stage Model of Choice , 1988 .

[10]  Naresh K. Malhotra,et al.  The Use of Linear Logit Models in Marketing Research , 1984 .

[11]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[12]  Lawrence D. Jackel,et al.  Backpropagation Applied to Handwritten Zip Code Recognition , 1989, Neural Computation.

[13]  Peter S. Fader,et al.  Estimating Nonlinear Parameters in the Multinomial Logit Model , 1992 .

[14]  John A. Stankovic,et al.  Distributed Processing , 1978, Computer.

[15]  J. Friedman.  Multivariate adaptive regression splines , 1990 .

[16]  John R. Hauser,et al.  Testing the Accuracy, Usefulness, and Significance of Probabilistic Choice Models: An Information-Theoretic Approach , 1978, Oper. Res..

[17]  R. Tibshirani,et al.  Flexible Discriminant Analysis by Optimal Scoring , 1994 .

[18]  D. McFadden.  Conditional logit analysis of qualitative choice behavior , 1972 .

[19]  Christopher M. Bishop,et al.  Neural networks for pattern recognition , 1995 .

[20]  Akhil Kumar,et al.  An empirical comparison of neural network and logistic regression models , 1995 .

[21]  Kurt Hornik,et al.  Neural networks and principal component analysis: Learning from examples without local minima , 1989, Neural Networks.

[22]  R. Bucklin,et al.  Reference Effects of Price and Promotion on Brand Choice Behavior , 1989 .

[23]  R. Tibshirani,et al.  Penalized Discriminant Analysis , 1995 .