|
Activity Number:
|
75
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 6, 2006 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #305541 |
|
Title:
|
Parametric Distance Estimators versus Maximum Likelihood Estimators in Estimating Quantiles with Misclassified Data
|
|
Author(s):
|
Elliott Nebenzahl*+ and Dean Fearn
|
|
Companies:
|
California State University, East Bay and California State University, East Bay
|
|
Address:
|
1196 Stafford Drive, Cupertino, CA, 95014,
|
|
Keywords:
|
quantile distance estimators ; minimum distance estimators ; survival distributions ; quantile functions
|
|
Abstract:
|
We create parametric quantile estimators by minimizing distances chosen from a family of distances between a parametric quantile function and the sample quantile function. We compare these estimators to each other and the standard maximum likelihood (ML) estimators in terms of "closeness" of the fitted quantile function to the true quantile function. We are particularly interested in robustness, where the true population for the data does not agree with the parametric family we are considering. For example, we consider fitting a Weibull parametric model to generalized exponential data. For various data and different sample sizes and parametric models, we also compare the distance estimators and ML estimators in terms of unbiasedness and mean-squared error of the respective estimated pth (0< p< 1) quantiles.
|
