A Simulation Method to Estimate Nonparametric Distribution of Heterogeneous Consumer Preference From Market-Level Choice Data

In recent decision-based design trends, product design is optimized for maximizing utility to consumers. A discrete-choice analysis (DCA) model is a widely utilized tool for quantitatively assessing how consumers evaluate utility of a product. Ordinary DCA models specify utility as linear combination of attribute values of a product and coefficients that represent preference of consumers. Assuming that the coefficient value is heterogenous between individual consumers, this study proposes a method to estimate its nonparametric distribution using market-level data, which is the market share of existing products. Where consumers consider k attributes of a product, his/her preference is represented by a k-dimensional vector of coefficient values. This method simulates an empirical distribution of the vectors in k-dimensional space. The whole space is first fragmented by disjoint regions, vectors in which prefer a specific product than others, and then, random points are sampled in each region as much as market share of the corresponding product. In a sense that more points are sampled for a more popular product, the empirical distribution is population of preference vectors. This method is practically useful since it utilizes only market-level data, which are relatively easy to gather than individual-level choice instances. In addition, the simulation procedure is intuitive and easy to implement.