|
Activity Number:
|
122
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #301808 |
|
Title:
|
An Extension of Generalized Regression Estimator for Dual-Frame Surveys
|
|
Author(s):
|
Shiying Wu*+ and Avinash C. Singh
|
|
Companies:
|
Research Triangle Institute and RTI International
|
|
Address:
|
3040 Cornwallis Rd., RTP, NC, 27709,
|
|
Keywords:
|
Calibration weights ; combining estimates ; coverage bias ; GREG
|
|
Abstract:
|
An alternative approach to the commonly used generalized regression (GREG) estimator for single frames is first proposed by using linear modeling for coverage bias. The proposed approach provides an extension of GREG to dual frames. The predictor variables for the coverage bias model may include the usual auxiliary information used for poststratification as well as extra information in terms of key study variables from the overlapping sample. The calibration variables for the extra information involve suitable positive coefficients (which sum to one) in order to combine domain estimates for the overlapping part. For a given set of coefficients, the estimator can be expressed as a calibration estimator with a corresponding set of calibration weights. Besides bias and some variance reduction achieved through this calibration, the variance could be further reduced by choosing the coefficients such that the relative variance of calibrated weights is minimized. Such an optimal choice of coefficients has the desirable property of not being dependent on a particular study variable. Monte Carlo simulation results on relative performance of several alternative estimators are presented.
|
