May Austin UserRs Group Meetup: Dr. James G. Scott presentation

Cherry picking: a Bayesian view

Assistant Professor of Statistics at UT-Austin, Dr. James G. Scott

http://www2.mccombs.utexas.edu/faculty/james.scott/home/Home.html

An unavoidable challenge in dealing with large data sets is the need to draw sensible conclusions from data points that have been cherry-picked:

• A biologist runs a genome-wide association study and flags 10 regions of the human genome that were most highly associated with schizophrenia. Having read the study, what should other scientists think about the true probability of association?

• The top 10 mutual fund managers of 2011 averaged a 40% return. What should we expect from them in 2012?

• A large web retailer flags the 50 ZIP codes most strongly associated with probability of purchase. How likely are these associations to persist?

In each case, the data have been subjected to some prior filtering mechanism before they come to your attention. I will walk through this problem from a statistical perspective, and show how Bayesian models can be used to correct for selection bias in a principled way.

BEER and PIZZA will be provided so come and get to know your fellow Austin UseRs and let's get this group off to a great start.

Thanks to Revolution Analytics ( http://www.revolutionanalytics.com/ ) and Bazaarvoice ( http://www.bazaarvoice.com/ ) for their sponsorship!

Thanks to Infochimps ( http://www.infochimps.com/ ) for hosting us!