1. Using choice models to make decisions
The real power of a choice model is that it allows us to make predictions about what people will choose when presented with a new set of options.
2. Predicting choice shares
Market share is the percentage of people who choose a particular product from the set that is available in the market. We can predict market shares for a set of products using the choice model. For example, suppose you are designing a sports car and you know that you will have two major competitors. We can use our choice model to make a prediction about how likely people would be to choose our car versus the competitors.
This output here shows the share predictions from our model, which are in the first column labeled share. These shares are computed from the model based on the product features which are described in the other columns. Suppose you were considering developing the 2-seater manual transmission car in the first row. That's the car I would choose. However, you can see that the predicted share for that car is only 17-point-7%. If you are planning to make a car like that, this is not great news and based on what we saw before in the coefficients, you might want to think about lowering your price or switching to an automatic transmission.
And we could make those predictions with the model. When we change the design of the first sports car, it gets 65-point-3% share. That's a lot better!
These market share predictions are incredibly useful to product designers. The product designer can use the model to understand how her design choices will affect share. This share can be compared to the costs of producing different products to figure out which design is most profitable.
3. predict_mnl() function
You may have used the predict() function to make predictions from a linear regression model before. There is a similar function for mlogit models, also called predict(), but it has a few features I don't like, so I wrote my own function for prediction from a multinomial logit model. It is called predict_mnl().
The first input to predict_mnl() is the mlogit model object that we get when we call mlogit(). The second input is a data frame that describes the new choice alternatives that you want to predict share for. You can think of this products data frame as describing the sports cars that are available in some hypothetical country somewhere.
In chapter 3, we'll look at how this prediction function works internally and this will give you a better understanding of how a multinomial logit model works. But for right now, let's just use the function and get a feel for what it does.
4. Let's predict some shares!
Let's predict some shares for sports cars!