Online Program

Return to main conference page

All Times ET

Program is Subject to Change

Tuesday, June 15
Tue, Jun 15, 11:30 AM - 1:00 PM
TBD
Treatments for Nonresponse: Imputation, Weighting, and Other Models

Weighting for Nonresponding Institutions in a Two-Stage Probability Proportional to Size Sampling Design when Measure of Size Is Estimated (308136)

Wan-Ying Chang, National Science Foundation 
Peter Einaudi, RTI International 
*Amang S. Sukasih, RTI International 
Sara Wheeless, RTI International 

Keywords: Early Career Doctorates Survey, institution sampling, population estimation

In a two-stage sampling design, it is common to apply probability proportional to size (PPS) sampling in the first stage where the true measure of size (MOS) for the primary sampling unit (PSU) is unknown and must be estimated. The Early Career Doctorates Survey, sponsored by the National Center for Science and Engineering Statistics within the National Science Foundation and by the National Institutes of Health, has such a two-stage design. In the first stage, a stratified PPS sample of academic institutions was selected using estimated MOSs that are a composite measure of the numbers of early career doctorates across key domains. In the second stage a sample of early career doctorates was selected from each of the responding institutions.

After obtaining the frame list of early career doctorates from each responding institution (PSU), the second-stage sample allocation can be adjusted according to the actual institution MOS to improve the coverage of key domains. However, the unknown size of nonresponding institutions poses a challenge to weighting adjustments, especially when a sizeable subset of large institutions selected with certainty do not respond. To solve this problem, the initial estimate of MOS for nonresponding institutions may be updated or re-estimated using the true sizes of responding institutions.

In this presentation, we demonstrate and compare different weighting adjustment approaches, including using institution matching and multiple imputation to account for uncertainty in re-estimating the MOS of nonresponding institutions. Our results show that when estimating totals of ultimate units in the population is one of the main survey goals, attention must be given to accounting for the uncertainty of estimating the MOS.