In this afternoon workshop, participants will be given an introduction to the Stan modeling language. Please register your attendance via the link below.
Stan is a flexible modeling language capable of performing efficient Bayesian inference on any model with a continuous parameter space for which we can evaluate a (log) likelihood. It implements a cutting-edge variety of Hamiltonian Monte Carlo, which will happily work with tens of thousands of parameters, and often produces reliable estimates with only a few hundred iterations. It is currently possible to call Stan from within R, Python, Julia, Mathematica, MATLAB, Stata, or at the command line.
13-13:45: A brief introduction to Stan.
What it is, how programs are set up, how to call them.
Exploring model fits in Shinystan
14 - 15: A modern statistical workflow.
This workflow helps researchers iterate towards richer, higher quality models with less pain. We’ll use a simple time-varying-parameters AR model as the example.
15:15 - 17:00: Working through a more complex model.
We’ll deploy the workflow on a model that is known to be fairly hard to fit: aggregate random coefficient logit (AKA BLP).
Participants should have R, R Studio and Stan installed (we will only use R to run and evaluate models; participants needn’t be proficient in R).
R may be installed from cran.r-project.org
R Studio from https://www.rstudio.com/products/rstudio/download/
Stan from Cran. You will need to follow the installation instructions here, depending on your system:
About the instructor:
Jim Savage is an applied modeler and Data Science Lead at frontier markets lender Lendable in New York City. Previously he was at the Grattan Institute, La Trobe University, and the Australian Treasury. With Andrew Gelman, Shoshana Vasserman and David Stephan, he is currently writing a book on Bayesian Econometrics in Stan.