|
Activity Number:
|
154
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #305867 |
|
Title:
|
Estimation Using Gaussian Replicates of the Pivotal Based on Weighted Quasi-Score Vector
|
|
Author(s):
|
Avi Singh*+ and Claude Nadeau
|
|
Companies:
|
NORC at the University of Chicago and Statistics Canada
|
|
Address:
|
55 E.Monroe St., 30th Floor, Chicago, 60603,
|
|
Keywords:
|
Taylor variance estimate ; Wald Interval Estimate ; Estimating Function based MCMC
|
|
Abstract:
|
The method of weighted quasi-likelihood (wql) is commonly used for point estimation in survey data analysis along with the Taylor method for variance estimation (VE) and Wald for interval estimation (IE). However, for finite samples, it is known that the above VE may be unstable and IE may have poor coverage properties. We consider ways to improve standard VE and IE by using Gaussian replicates of the pivotal estimating function (EF) derived from the wql-score vector. The basic idea is that the (nonstudentized) pivotal EF is closer to normal than the wql-estimator. The replicate estimators are obtained by setting the pivotal EF equal to random draws from the standard multivariate normal distribution. The computational problem of solving a multivariate EF for each replicate may be quite difficult . We propose an EF-based MCMC under a frequentist framework for this purpose.
|
