1. Goodbye and good luck!
Congratulations on finishing all the exercises in this course. Well done! I'd like to end the course with a quick goodbye and good luck.
2. Choices in building models
I'd also like to share a few final thoughts.
When you are building a choice model, there are lots of decisions to make. Which attributes will you include? Will you treat numeric attributes as factors? Will you include interactions between attributes or between attributes and decision-maker characteristics? Should you use a hierarchical model? Should it have correlations between coefficients?
3. Other choice model features
And there are other modeling choices that I haven't covered like different distributions for random coefficients, probit models, nested logit models or Bayesian choice models (which you can fit with the bayesm package or with Stan).
There are so many choices it can be a little daunting to get started. Don't let that stop you! Now that you've worked through my sports car and chocolate data sets, you should try fitting choice models with your own choice data.
4. Advice for building models
Let me give you a few words of advice as you start building your own models.
First, always start by inspecting the data. Look at a few individual choices to make sure you understand the data structure. Then compute choice counts to summarize the data. Many people jump into fitting models without looking at the data first, which means they must blindly trust their models.
Second, you should always start with simple models first and build up to more complex models. This makes it easier to diagnose problems that crop up. If your estimated parameters have very large standard errors, then you've probably added too much model complexity for the data you have. Back up to a simpler model.
Finally, for models describing human behavior, heterogeneity is usually a good idea. In nearly all domains, different people have different preferences. I'd encourage you to use those models whenever you observe more than one choice for each decision maker.
5. Go fit some choice models!
So go out there and find or collect your own choice data and start fitting choice models. That's when this material will really come to life for you!