Estimating choice models in data-sparse environments: Taking advantage of perceived similarity

Research suggests that choice models conditioned on correctly identified consideration sets outperform choice models conditioned on the awareness set (Hauser, 1978; Roberts and Lattin, 1991). However, in data-sparse environments, where purchase history information is not available or not relevant, choice models conditioned on the consideration set often yield nonunique or nonsignificant solutions. In these environments, we propose the use of similarity information to improve the performance of choice models. Support for this position is found in an empirical application involving automobiles.