SEMINARS
|
|
Fall 2007
STATISTICS
COLLOQUIUM
Wednesday, October 8, 2008
3:30-4:00—Refreshments
4:00-5:00—Talk
Yost Hall, Room 101
Bin Wang, Ph.D.
Department of Mathematics and Statistics University of South Alabama
A Generalized Density Estimation Method for High Quantile Estimation
Accurate estimation of high quantiles depends on the right-tail modeling of the underlying distribution which is closely related to the asymptotic distribution of the sample extreme values. Due to sparseness of data values in the right-tail of heavy tailed distributions, modeling the right-tail behavior can be very technical and difficult. In survival data analysis, lifetimes are usually right-skewed and thus data are sparse in the right tail. In addition, censoring occurs often in survival data, which causes additional uncertainty with regards to right-tail behavior of the underlying distribution. In this study, we propose to make inferences of high quantiles of a distribution by fitting a generalized lambda distribution to a random sample. The estimation results will be demonstrated via simulation results. A transform-retransform density estimation method based on an exponentiated Weibull family of distributions will also be studied to improve the high quantile estimation.
|
