SCS Reads 2016-2017

From Wiki1

Jump to: navigation, search

Contents

Getting started

We chose to read McElreath's Statistical Rethinking.

  • To usefully work along in R with the text, you'll need to install the rethinking package. But this requires RStan, which in turn requires the Rtools development software. You need to do this in the order `Rtools -> RStan -> rethinking`.
    • RStan Getting Started describes the steps to install Rtools first and then RStan
    • McElreath's software page describes installing the rethinking package. It is best to do this from the github source package as he describes.

Meetings

Week 1, Sept. 23

  • McElreath, Chapter 1

Another paper drawing an analogy between statistical modeling and engineering (and art): File:Thissen2001.pdf

Week 2, Oct. 7

  • McElreath, Chapter 2

Answers to some exercises from Ch. 2 (and corresponding RMarkdown file).

Week 3, Oct. 21

  • McElreath, Chapter 3 and Chapter 4 through p. 91

Answers to some exercises from Ch. 3 (and corresponding RMarkdown file).

Week 4, Nov. 4

  • McElreath, remainder of Chapter 4 and all of Chapter 5
  • Regarding better plots for parameter estimates in Bayes: This Paper by Kay, Nelson & Hekler, "Researcher-Centered Design of Statistics: Why Bayesian Statistics Better Fit the Culture and Incentives of HCI" uses what I mentioned as violin plots to show posterior densities of parameters. It is worth reading for the comparison of frequentist and Bayesian approaches.
  • Matthew Kay also has developed a tidybayes R package for composing/extracting tidy data from Bayesian samplers. Only on github, and no real examples, but it seems promising.

Week 5, Nov. 18

  • McElreath, Chapter 6

Week 6, Dec. 2

  • McElreath, Chapter 8

Week 7, January 13

  • McElreath, Chapter 7

Week 8, January 27

  • McElreath, Chapter 9

Week 8.5, February 3

We thought that we'd view and discuss John Oliver's Last Week Tonight Show on 'Scientific Studies', originally aired in May 2016.

Of related interest is a recent (January 16, 2017) article in the "The Upshot": How to Prevent Whiplash From Ever-Changing Medical Advice

Week 9, February 10

  • McElreath, Chapter 10

Week 10, March 3

  • McElreath, Chapter 11

Week 11, March 17

  • McElreath, Chapter 12

Week 12, March 31

  • McElreath, Chapters 13 and 14

Candidates

Andrew Gelman et al. (2014) Bayesian Data Analysis, 3rd edition

  • Amazon link to Gelman et al. BDA3
    The Amazon preview provides access to the table of contents and to many of the earlier pages in the book.
  • The home page for BDA3 has datasets, lecture slides, some solutions, etc.
  • My opinion: Lots of exercises with a range of difficulty. For some, a current introductory 'bible' on applied bayesian data analysis with solid theoretical content. I think that in one year we can get far enough to learn how to use Hamiltonian Monte Carlo with STAN in R.

Richard McElreath (2016) Statistical Rethinking: A Bayesian course with examples in R and STAN.

  • Recommended by Heather. This might be the 'best new book' on Bayesian data analysis. Fewer equations and less theory than BDA3 but, instead of formulas, the book has short snippets in R that you run to illustrate concepts. It doesn't go as far as BDA3, e.g. no non-linear and non-parametric chapters, but has many observations (the ones I have read so far are sound in my opinion) about statistical methodology and connections between bayesian and frequentist concepts -- although the author suggests that he would have liked to have done more but didn't want to crowd out the bayesian material. Here's an excerpt to illustrate the author's point of view:
    As a consequence, this book doesn't argue against p-values and the like. The problem in my opinion isn't so much p-values as the set of odd rituals that have evolved around them, in the wilds of the sciences, as well as the exclusion of so many other useful tools.
  • Lots of R code (pre-Wickham in style), lots of exercises conveniently labelled 'Easy', 'Medium' and 'Hard'.
  • Here's a photograph of a page 2016-09-10 11.49.15.jpg to give an idea of the style of the book.
  • Web sites for the book:
    • McElreath's web site listing software related to the book
    • The GitHub repository for the `rethinking` R package. A quick look here shows that the package includes a large number of data sets and functions for doing the analyses and graphs described in the book.
  • Ordering info:
    • I just ordered a copy from Amazon.ca for CDN $103. The site said shipping in 2-4 weeks.
    • Also available from CRC Press for USD $75. No idea how long shipping will take.
  • Opinions:
I like the overall flavor of what I've seen of this book. (M. F.)
Personal tools