[Ml-stat-talks] [stat seminar reminder]: Friday March 25, 12:30pm by Prof. Zhigen Zhao

xtong at princeton.edu xtong at princeton.edu
Thu Mar 24 18:35:27 EDT 2011


  Empirical Bayes Confidence Intervals for Selected Parameters with Unknown
and Unequal Variances
 Prof. Zhigen Zhao, Temple University
 Friday, March 25th, 12:30pm-1:30pm
 ORFE, statistics lab 

 Abstract:

 In modern application, such as microarray experiments, scientists
 are especially interested in the inference of parameters which have been
 selected. Scientists often constructed the usual interval, ignoring the
 fact that these genes have been selected. However, this interval suffers
 low coverage probability. The other commonly used method is to construct
 the interval with Bonferronis correction which guarantees good
 coverage probability but has a long length.
 In this talk, we assume that the observation are normally distributed
 with unequal and unknown variance. The parameters have
 been selected according to the magnitude of the t-values. We then
 construct the empirical Bayes confidence intervals for these selected
 parameters. We compare this new approach with Bonferronis Correction.
 Our interval has good empirical Bayes coverage probability, and
 has much shorter average length. Our interval is also applied to the
 Spike-in data.
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20110324/11503590/attachment.htm>


More information about the Ml-stat-talks mailing list