Tutorial on Using Stan in Marketing Research

2017 Advanced Research Techniques (ART) Forum
June 25-28, 2017

Using Stan to Estimate Hierarchical Bayes Models
Elea McDonnell Feit, Assistant Professor, Drexel University
Kevin Van Horn, Senior Data Science Engineer, Adobe

Stan (http://mc-stan.org/) is a new open-source tool for estimating complex statistical models using MCMC. Similar to JAGS and BUGS, Stan allows you to specify complex models using a modeling language. Once the model is specified, Stan automatically generates a routine to sample from the posterior, allowing the user to focus on the model and posterior estimates, rather than on details of the MCMC routine. In this hands-on tutorial, we will show how to use Stan to fit popular models in marketing including the hierarchical choice models typically used for conjoint, a nested logit model for the no-choice option and data fusion models. Users will be encouraged to run code on their own laptops during the workshop. Stan can be accessed through R, Python or Matlab and several other statistical programming tools; our focus will be on using the RStan interface. Users who are already familiar with Bayesian inference and basic R syntax will get the most out of this tutorial. Instructions for installing Stan and example R code will be provided. Users are encouraged to install Stan before the tutorial.


Thanks. And in case anyone else is wondering, we love getting talk announcements.

@betanalpha — would you mind creating a tag for “talks” or “tutorials” or “classes” or “events” or “meetups” or something we can put this under? Publicity doesn’t seem quite right. And I’d think “teaching” might be good for discussions of using Stan for teaching.

a) Tags are descriptive comments that augment the categories and anyone can create a tag. You’re proposing a new category proper which is limited to admins.

b) Those suggestions are too specific and would require multiple new categories to fill out all of the complementary possibilities. “Publicity” was chosen as all of those fall under people wanting to publicize their work with Stan, something that we want to encourage.

I can see an argument for separating between static publicity (papers, case studies, etc) and time-dependent publicity (upcoming events) but otherwise I am hesitant to make any changes.

For anyone who’s interested, here’s a link to the git repo containing slides and example code from the tutorial. If you’re not familiar with git, you can download the ZIP file accessible from this page.