|
Activity Number:
|
229
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #301343 |
|
Title:
|
BRR versus Inclusion-Probability Formulas for Variances of Nonresponse Adjusted Survey Estimates
|
|
Author(s):
|
Eric V. Slud*+ and Yves Thibaudeau
|
|
Companies:
|
U.S. Census Bureau and U.S. Census Bureau
|
|
Address:
|
Statistical Research Division, Washington, DC, 20233-9100,
|
|
Keywords:
|
Variance estimation ; Replication methods ; Weight Adjustment ; Misspecified model ; Quasi-randomization ; SIPP
|
|
Abstract:
|
We compare two methods, employing possibly misspecified parametric models for nonresponse, of estimating theoretical variances for nonresponse adjusted estimates of survey totals. The first method is based on a formula of Sarndal and Lundstrom (2005) requiring known joint inclusion probabilities and a calibration model for response probabilities. The second method is based on balanced repeated replicates (BRR). Both methods are compared with the correct design-based variance in a superpopulation framework with independent random response indicators. Numerical calculations and simulation results are given for a split-PSU design, with simple random sampling within half-PSU's and with weight factors for nonresponse constant over adjustment cells. The accuracy of variance estimators is related to the population balance across half-PSU's of intersections of true and working adjustment cells.
|
