Distributions & Uncertainty
Materials for class on Monday, September 23, 2019
Contents
Annoucements
DataCamp 2 is due by 11:59pm Tuesday, Sept 24th.
Reminder that Problem Set 2 is due by 11:59pm Monday, Oct 7th (during Fall Break).
Charlotte R Meetup Group. Possibly present on Tuesday, 12-1pm in 12/17 meeting. See Slack channel post.
Tweet of the Day:
My ??? talk is now in video form: A biased tour of the uncertainty visualization zoo. If you liked the version with just slides, you'll love (?) the version with me talking over slides :) https://t.co/hSdczZlgKW
— Matthew Kay (@mjskay) December 7, 2018
Matthew Kay is an assistant professor of Information at the University of Michigan School of Information. His research focuses on communicating uncertainty, especially from a Bayesian perspective including authoring the tidybayes package. If you’re new to Bayesian statistics, he has an excellent paper on why Bayesian statistics is appropriate for human-centered (HCI) research. I also highly recommend his research appearance on the DataStories podcast with his colleague, Jessica Hullman.
Slides:
Lab 5:
For the lab session, we’ll use the same gganimate project from last week that’s on our RStudio.cloud workspace.
If we have time, I also included Matthew Kay’s Uncertainty visualization examples in the RStudio.cloud project (see uncertainty-examples folder).Be sure to check out Claus Wilke’s ungeviz package.
Links:
Kristoffer Magnusson’s visualization demos:
Multiple Views Blog: They Draw it (eliciting users’ prior beliefs) Try out TheyDrawit
StackOverflow: Difference between Bayesian and BoostrappingOr if you want even more technical, see the Wikipedia page on Expected Loss. This gets at the core difference between Bayesian and Frequentist schools of thought.
Pierre Dragicevic’s “Fair statistical communication in HCI” paper.This paper is an excellent guide for ways to appropriately communicate statistical models and uncertainty. This provides a good background on why overemphasis on p-values and dichotomous testing can go wrong and miss important perspectives when doing hypothesis testing.