|
Activity Number:
|
335
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #305886 |
|
Title:
|
Bootstrap Investigation of the Median Curve of a Functional Dataset
|
|
Author(s):
|
David B. Hitchcock*+
|
|
Companies:
|
University of South Carolina
|
|
Address:
|
Department of Statistics, Columbia, SC, 29208,
|
|
Keywords:
|
functional data ; bootstrap ; data depth ; median curve
|
|
Abstract:
|
A definition of the median of a set of functional data is given. The sample median curve (or "most representative" curve) is defined as the data curve with the minimum average pairwise L1 distance to all other curves, and is a better representative of a "typical" curve than the pointwise mean or pointwise median curve. The median curve has connections to the sample median of a set of univariate or multivariate observations. An analogous definition of a population median curve is presented, along with methods for approximate inference about the population median curve. These methods---in particular, approximate confidence bands for the median curve---are based on a bootstrap approximation of the sampling distribution of the sample median curve.
|
