A Nested Logit Model of Coffee Purchase

On a shopping trip to a supermarket a customer may purchase a product in a given category and, if so, buys a particular brand. A previous paper by the authors models the brand choice part of this process. A multinomial logit model describes the selection of brand and size given that the customer makes a category purchase. Explanatory variables include store actions, such as price and promotion, and customer characteristics, such as brand and size loyalty. We now extend the formulation to include the decision to make a purchase in the category on a shopping trip. This additional step not only provides a more complete description of the buying process but also makes possible a better calculation of sales response by including the effect of marketing actions on category sales as well as brand share. The methodology employed is the generalization of the multinomial logit known as the nested logit. The shopper's decision includes two components: the selection of the category and the choice of the brand-size combination. The model of brand-size choice is essentially that of our earlier paper. The category choice introduces new variables including household inventory, category price, and the attractiveness of purchasing a product now as opposed to later. Calibration of the nested logit is done by sequential estimation. The model, applied to regular ground coffee data from a UPC scanner panel, tracks sales well in a holdout sample, both at the aggregate and individual levels. Use of the model to calculate short term market response to promotion demonstrates the phenomenon that brand sales can increase because of the expansion of the category as well as share.