1. Why choice modeling?
Hi, I'm Elea McDonnell Feit and I'm a marketing professor at Drexel University. Before I became a professor, I worked at General Motors where I used choice models to help design new cars. I found my work at GM so interesting that I decided to get a PhD focusing on choice modeling and after I finished that, I had a job as a methodologist at a company that specialized in using choice models to help Intel, Warner Brothers, Dole, and other big companies make important product design and pricing decisions. Choice modeling and its close cousin conjoint analysis are one of the most popular modeling tools used in marketing. They are also used in other fields like political science and transportation.
2. Regression modeling relates predictors to numeric outcomes
In linear regression, which you should already be familiar with, we predict a number. For instance, we could predict the sales at a store based on the features of that store like its size or the population near the store. In that case, the thing we are predicting - sales - is a number.
3. Many events we want to understand and predict are **choices**
But many events that we are interested in as data scientists and marketers are choices from a set of things.
When a customer goes to an online retailer, she selects one dress from many that are available.
When you decide what to watch on Netflix tonight, you will choose one show from a menu of available content.
When a customer buys a car, she chooses a model from those that are available in her region. As a marketer, we want to know how features of cars relate to which car a customer will choose and we can use that information to predict what will happen to market share if we change our product.
4. Choices require their own special type of regression
Data on choices doesn't fit well within the linear regression framework.
Instead, we use another type of model called a multinomial logistic regression or the multinomial logit model. Multinomial logit models are used to predict a choice from a set of two or more options. The prediction is based on the features of each alternative. For instance, we can predict the likelihood of choosing a particular car based on the features of the available cars.
You may have heard about logistic regression before. Logistic regression is a special case of multinomial logistic regression that we use with data on binary "yes/no" choices like consumers deciding whether to redeem an offer. We are going to focus on multinomial logistic regression which is more general and can be used to predict choices from among two, three or more alternatives.
5. Marketing applications for choice models
Choice models have a lot of applications within marketing.
When I worked at GM, we used choice models to understand which features would make our cars more desirable.
You can also use choice modeling to determine how to price a product based on how customers trade-off price against other product attributes.
An online retailer can use choice models to understand how features of the website - like a "customer favorite" flag that you put on a product preview - affect what customers ultimately buy. All of these are great examples of how choice models are used in marketing.
6. What choices are *you* interested in analyzing?
In this chapter, I'm going to focus on a data set that describes customers choosing sports cars and in later chapters, we will analyze chocolate choices. These are two of my favorite product categories. But these are just examples and I hope you will start thinking about all the different types of choices you might want to study.