Saturday, June 18, 2011

Workshop, July 20, Cognitive Science Society

I'll be teaching a day-long workshop on doing Bayesian data analysis at the conference of the Cognitive Science Society, July 20, 2011.


For other future and past workshops, see my teaching page.






Workshop, July 11-15, at New Bulgarian University

I'll be teaching a week-long workshop on doing Bayesian data analysis at the New Bulgarian University Cognitive Science Summer School, July 11-15, 2011.

Other past and future workshops at my teaching page.

Workshop, June 20-24, at ICPSR, U. Michigan

I'll be teaching a week-long workshop at the Inter-University Consortium for Political and Social Research (ICPSR), June 20-24, 2011, at the University of Michigan.

Highlights of agenda (much more will be covered):
Day 1: What's wrong with p values. Bayes’ rule, grid approximation. The R programming language.
Day 2: Markov chain Monte Carlo methods, BUGS/BRugs, hierarchical models.
Day 3: Null hypothesis significance testing, Bayesian model comparison, Bayesian assessment of null values. Statistical power.
Day 4: Multiple linear regression, logistic regression, ordinal regression.
Day 5: Hierarchical Bayesian ANOVA and contingency table analysis. How to report a Bayesian analysis.

Additional info at the ICPSR course web page.

Other past and future workshops at my teaching page.

Update from on site, 6/22/2011. There's an entire theater here devoted to statistics! Photographic proof:

Thursday, June 16, 2011

BRugs for Mac users

[You don't have to use BRugs any more; you can use JAGS instead. See this more recent post about JAGS.]

BRugs, as of now, can only be used with 32-bit Windows. Mac users must install and run through Windows (non-)emulators such as Wine. Here are some preliminary notes about how to install and use BRugs on a Mac.

Update: See also the more recent post regarding BRugs for Linux users.

Mac users: Please reply with additional tips, tricks, and info! (I am not a Mac user, so all of my info is 2nd hand.)

Installation:

In principle, the idea is to install 32-bit Windows R through Wine or WineBottler. In practice, I hear that this is not easy, and a search of the web shows some discussion that makes it seem like quite a process. Fortunately, the process has been nicely packaged by Jack Harris and posted by Seth Frey at http://enfascination.com/wiki/Weblog:Bayesian_data_analysis_on_a_Mac

Use:

Some of the Windows-friendly R commands are not interpreted properly on a Mac. Thanks to Chris Street for pointing out the following substitutions:

dev.copy2eps( file="blah" )
can be changed to
savePlot( filename="blah" , type="eps" )

At the command console,
help("command")
fails, but
??command
works.

Again, please reply (comment) with additional discussion.

Monday, June 13, 2011

Better than the t test: Robust Bayesian Comparison of Groups

[An updated post appears HERE.]
It's been said that if all you're doing is a t test, Bayesian methods don't get you anything more. Wrong! Here's a manuscript showing that robust Bayesian estimation (not Bayesian model comparison involving Bayes factors) produces far more information about the difference of means, the difference of standard deviations, the influence of outliers, and power of the test. The Bayesian method can also accept the null, not merely reject it. The software and programs are free and easy to use. It is time for robust Bayesian estimation to supersede traditional methods.

Thursday, June 2, 2011

Two Bayesian ways to assess null values. Which is better?

Psychologists (and any other researchers) have been trained to do data analysis by asking whether null values can be rejected. Is the difference between groups non-zero? Is choice accuracy not at chance level? These questions have been addressed, traditionally, by null hypothesis significance testing (NHST). NHST has deep problems that are solved by Bayesian data analysis. As psychologists transition to Bayesian data analysis, it is natural to ask how Bayesian analysis assesses null values. The article explains and evaluates two different Bayesian approaches. One method involves Bayesian model comparison (and uses “Bayes factors”). The second method involves Bayesian parameter estimation and assesses whether the null value falls among the most credible values. Which method to use depends on the specific question that the analyst wants to answer, but typically the estimation approach (not using Bayes factors) provides richer information than the model comparison approach.
That's the abstract from a recent article*, which you can find >here< in its pre-publication form or >here< in its published form. Note that the published form suffers from a production error in which Equation 1 was omitted from p. 301; see the pre-publication version for the intact equation! The article covers some of the topics in the book (Doing Bayesian Data Analysis) but in a succinct and stand-alone way. The article also has an example (loosely based on results of Bem's recent results regarding "feeling the future") showing how Bayes' factors can be extremely sensitive to the choice of alternative-hypothesis prior.

The article is part of a special section of the journal on Bayesian data analysis. Two other articles in the section can be found at the journal website (click the link to the published form, above).

UPDATE: See this additional example for another illustration of how the estimation approach makes more sense than the model comparison approach for null assessment.

* Kruschke, J. K. (2011). Bayesian assessment of null values via parameter estimation and model comparison. Perspectives on Psychological Science, 6(3), 299-312.