A minimum discrimination information estimation of multiattribute market share models

Multiattribute market share models (MMSM) using supplier, product, and buyer attributes as explanatory variables generally use a logit model structure. The literature indicates two basic types of approaches for estimating coefficients, regression type approaches generally using log-share ratios and a nonregression approach which is based on maximum-likelihood principles. This article discusses a nonregression estimation approach, a minimum discrimination information (MDI), whose formulation differs from MDI procedures previously applied to the marketing literature. The MDI formulation enables researchers to estimate a logit model with proportions as the dependent variable and is computationally easy because it uses the same estimators as an MLE approach which assumes independence of trials. Equivalence between MDI and MLE estimates, under certain conditions, are discussed. The estimation methods are applied to a real world data set. The results indicate the MDI/MLE method does best in predicting the market shares of the suppliers according to some basic error criteria both at disaggregate and aggregate levels.