2011/11-08

From SIPB Cluedumps

[edit] Statistics and the Non-Conflict between Bayesians and Frequentists

Date: November 8, 2011, at 3:00 PM

Presenters: Keith Winstein (keithw)

Location: 3-133

Abstract: In 2007, an academic cardiologist downloaded 42 medical studies from the Web site of drug giant GlaxoSmithKline, combined them, and found that the world's best-selling diabetes drug caused heart attacks. GSK lost about $12 billion in sales and market value, and last Thursday agreed to pay a $3 billion fine over its marketing of the drug. But a different way to analyze the same data, called Bayesian analysis, finds with conservative assumptions that the drug actually reduces heart attacks. Or does it? I'll present a new take on the difference between classical statistical methods and Bayesian techniques, showing how the opposing schools are two sides of the same coin. Viewed formally in the context of confidence and credibility intervals, criticisms of each practice have a tight symmetry and can be transformed into each other. I'll also show results from a new algorithm that calculates the performance of contemporary "exact" hypothesis tests that hadn't previously been characterized.

Bio: Keith Winstein is a graduate student in Course 6. From 2007 to 2010, he covered science, technology and medicine as a staff reporter for The Wall Street Journal.