Bayesian Regression Models with RStanARM

TJ Mahr
Sept. 21, 2016

Madison R Users Group

Github repository
@tjmahr

Overview

Learning more

You want to learn RStanARM

See the vignettes. There is one for each kind of model.

RStanARM vignettes

You want to learn Stan

Go to http://mc-stan.org/documentation/
Look at the example models and read the manual.

The Stan Manual

You are trained in classical regression

Read Statistical Rethinking and watch the lectures.

'Rethinking'

This book is exceptional, stuffed to the brim with trivia, advice, and wisdom.
Not just a book about statistics, but how we use statistical models in scientific practice.
Explains side issues like MCMC, basics of information theory, or why the normal distribution is so prevalent – but just deep enough for the reader to get the intuitions needed for practice.
My sole criticism is that its companion R rethinking package is not on CRAN.

You need puppies on the cover

Kruschke (2015). Doing Bayesian Analysis.

I just got this book, but so far it's approachable and comprehensive, like a statistical cookbook. Many more equations and proofs than in Rethinking.

You study psycholinguistics

Nicenboim and Vasishth (2016). Statistical methods for linguistic research: Foundational Ideas - Part II. https://arxiv.org/abs/1602.00245
Sorensen and Vasishth (2016). Bayesian linear mixed models using Stan: A tutorial for psychologists, linguists, and cognitive scientists. https://arxiv.org/abs/1506.06201

You need a reading list and fast

Etz, et al. (2016). How to become a Bayesian in eight easy steps: An annotated reading list.
Etz has also great tutorials. They focus mostly on Bayes factors, a way to compare models using the denominators from Bayes theorem. Statistical Rethinking mentions these in passing in a single paragraph.

You need to figure out how many socks are in the laundry

See Rasmus B��th's intro to approximate Bayesian inference.
Kind of removed from our topic, but entertaining example that emphasizes how Bayesian models are generative.
If we are exploring all the plausible ways the data could have been generated, we're probably doing a kind of Bayesian analysis.