Seminar Details

Seminar Details

Title: Confidence Intervals and the Stability of Standard Errors
Speaker: Bradley Efron, Ph.D.
Speaker Link: http://statweb.stanford.edu/~ckirby/brad/
Affiliation: Stanford University, Department of Statistics
Host: Robert Gibbons
Contact Info: brad@stat.standford.edu
Location: **Location change** Dora De Lee, L168
Start: 4/5/2017 3:30:00 PM
Url: http://health.bsd.uchicago.edu/

Abstract:

Approximate 95% confidence intervals for an unknown parameter are typically of the form (t - 1.96*se, t + 1.96*se), where t is the point estimate and "se" is its estimated standard error. Here a tacit assumption is being made: that the true standard error of t is constant, or nearly so, across the interval. Bootstrap methods now allow us to assess not only se but also its rate of change. The talk describes such methods and how they can be incorporated into more accurate approximate confidence intervals.