Choice Modeling

Abstract

Much of the data we observe in marketing describes customers purchasing products. For example, as we discussed in Chap. 12, retailers now regularly record the transactions of their customers. In that chapter, we discussed analyzing retail transaction records to determine which products tend to occur together in the same shopping basket. In this chapter we discuss how to analyze customers' product choices within a category to understand how features and price affect which product a customer will choose. For example, if a customer comes into the store and purchases a 30 oz. jar of Hellman's brand canola mayonnaise for $3.98, we can conceptualize this as the customer choosing that particular type of mayonnaise among all the other mayonnaise available at that store. This data on customers' choices can be analyzed to determine which features of a product (e.g., package size, brand, or flavor) are most attractive to customers and how they trade off desirable features against price. On the surface, this may sound quite similar to what we discussed in Chap. 7, where we cover how to use linear models to identify drivers of outcomes. It is similar, except that product choice data doesn't fit well into the linear modeling framework, because the outcome we observe is not a number or a rating for each product. Instead , we observe that the customer makes a choice among several options, each of which has its own set of attributes. To accommodate this unique data structure, marketers have adopted choice models, which are well suited to understanding the relationship between the attributes of products and customers' choices among sets of products. In this chapter, we focus on the multinomial logit model, the most frequently used choice model in marketing. While choice models are often used to analyze retail purchase data, there are some settings where it is more difficult to collect data on customers' product choices. For example, when people shop for a car, they typically gather information from many sources over several months, so it is more difficult to reconstruct the set of products that they considered and the features and prices of those products. In these settings, marketers turn to choice-based conjoint analysis, which is a survey method where