|
Activity Number:
|
362
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics & the Environment
|
| Abstract - #300188 |
|
Title:
|
Bootstrap-Monte Carlo Hybrid Upper Confidence Bounds for Right-Skewed Data
|
|
Author(s):
|
Michael E. Ginevan*+
|
|
Companies:
|
BBLSciences
|
|
Address:
|
307 Hamilton Ave., Silver Spring, MD, 20901-3336,
|
|
Keywords:
|
bootstrap ; Monte Carlo ; sensitivity ; log-normal ; upper bound
|
|
Abstract:
|
Environmental contamination data often follow an extreme right-skewed distribution. In risk assessments it is of interest to use the sample data to calculate an upper bound on the population arithmetic mean. Recent work has shown that upper bounds based on log-normal approximations can severely overestimate the correct upper bound for the arithmetic mean and that bootstrap methods give more reliable results. A remaining concern is that bootstrap bounds may underestimate the correct upper bound. This paper shows how one can perform sensitivity analyses for bootstrap bounds for the arithmetic mean using bootstrap-Monte Carlo hybrid methods. In these, the data are resampled as in the bootstrap, but Monte Carlo methods are used to assess the influence of the unobserved upper tail of the right-skewed contamination distribution. We demonstrate methods for defining the unobserved right tail of the population distribution and combining bootstrap and Monte Carlo methods to perform sensitivity analyses. The resulting sensitivity analyses demonstrate that the influence of the unobserved right tail is generally modest and that simple bootstrap bounds are usually adequate.
|
