Modeling Online Consumer Product Search∗

Using publicly available aggregate data at an online retailer, we study how consumers search for durable goods. Specifically, we investigate the search patterns when, as is common with durable products, the total set of options consumers can choose from is very large. To study search patterns, we propose to analyze our aggregate data by defining a product network, consisting of the products and links that designate if a product is searched conditional on searching other products. We model these links using a flexible linear mixed model for binary data. Our model combines fixed effects of brands, product attributes, and prices on product search patterns with random effects capturing dependencies among the searched products. The latter can be displayed and interpreted as a perceptual map of “search proximity” or as a “product search map.” The proposed Bayesian estimation approach helps a practitioner understand which products are searched and compared in the same online session, and importantly analyzes how observed information acquisition is organized in brand, attribute, and price related search strategies. Using a data set covering consumer search of digital camcorders at Amazon.com, we infer that the product search is strongly attribute based.