Bayesian Data Analysis with BRMS (Bayesian Regression Models Using Stan)

Intro

Mitzi Morris, a Stan developer, shows how you can quickly build robust models for data analysis and prediction using BRMS (Bayesian Regression Models Using Stan). After a brief overview of the the advantages and limitations of BRMS and a quick review of multi-level regression. We will work through an R-markdown notebook together, to see how to fit, visualize, and test the goodness of the model and resulting estimates.

Video

Resources

Section Timestamps of Video

00:00:00 R-Ladies NYC Intro
00:04:55 Data Umbrella Intro
00:08:25 Speaker Introduction - Mitzi Morris
00:10:15 What is BRMS? (Bayesian Regression Models Using Stan)
00:11:15 Three reasons to use BRMS
00:13:51 Bayesian Workflow Overview
00:15:25 Modeling Terminology and Notation
00:17:54 Multilevel Regression
00:21:30 Regression Models in R & brief recent history of Bayesian programming languages
00:27:22 Linear Regression
00:28:52 Generalized Linear Regression
00:31:05 Regression Formula Syntax in BRMS
00:34:33 BRMS Processing Steps
00:37:13 Notebook - link to online notebook and data
00:37:38 Demo - in Markdown (.rmd)
00:38:18 Load packages (readr, ggplot2, brms, bayesplot, loo, projprod, cmdstanr)
00:38:38 Book - ARM
00:39:07 Example - Multilevel hierarchical model (with EPA radon dataset)
00:40:32 Further description of radon
00:41:37 Regression model
00:42:02 Demo - data example
00:42:26 3 Modeling Choices
00:44:31 Choice 1 - Complete Pooling Model (simple linear regression formula)
00:48:22 Choice 2 - No Pooling Model (not ideal)
00:50:17 Choice 3 - Partial Pooling Model
00:56:26 Q&A - How to compare the different models? (run loo)
01:00:00 Q&A - Does BRMS have options for checking model assumptions?
01:01:00 Q&A What were the default priors? (student T-distribution with 3 degrees of freedom)
01:05:27 References

About the Speaker

Bio

Mitzi Morris is a member of the Stan Development Team and serves on the Stan Governing Body. Since 2017 she has been a full-time Stan developer, working for Professor Andrew Gelman at Columbia University, where she has contributed to the core Stan C++ platform and developed CmdStanPy, a modern Python interface for Stan. She is also as an active Stan user, developing, publishing, and presenting on Bayesian models for disease mapping. Prior to that she has worked as a software engineer in both academia and industry, working on natural language processing and search applications as well as data analysis pipelines for genomics and bioinformatics.