|
Activity Number:
|
244
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 13, 2002 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Survey Research Methods*
|
| Abstract - #300996 |
|
Title:
|
Evaluation of Generalized Variance Function Estimators for the U.S. Current Employment Survey
|
|
Author(s):
|
Moon Cho*+ and John Eltinge and Julie Gershunskaya and Larry Huff
|
|
Affiliation(s):
|
U.S. Bureau of Labor Statistics and U.S. Bureau of Labor Statistics and U.S. Bureau of Labor Statistics and U.S. Bureau of Labor Statistics
|
|
Address:
|
2 Massachusettes Avenue NE, Washington, District of Columbia, 20212, U.S.A.
|
|
Keywords:
|
Variance estimator stability ; Degrees of Freedom ; Design-based inference ; Generalized least squares ; Model-based inference ; Superpopulation model
|
|
Abstract:
|
In applied work with generalized variance function models for sample survey data, one generally seeks to develop and validate a model that is relatively parsimonious and that produces variance estimators that are approximately unbiased and relatively stable. This development and validation work often begins with regression of initial variance estimators (computed through standard design-based methods) on one or more candidate explanatory variables. Evaluation of initial modeling results is often complicated by correlation among the initial variance estimators. This paper considers ways in which to address this problem, with principal emphasis on three issues: 1.) approximate conditional or unconditional independence of subsets of initial variance estimators; 2.) use of 1.) and additional conditions to evaluate the properties of the estimators of the coefficients of a generalized variance function model; and 3.) evaluation of the stability of the resulting variance estimators. Some of the proposed diagnostics are applied to data from the U.S. Current Employment Survey.
|
