|
Activity Number:
|
250
- Weighting and Variance Estimation in Complex Samples
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 9, 2022 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Survey Research Methods Section
|
|
Abstract #322403
|
|
|
Title:
|
Evaluating the Use of Design Weights in Classification Trees for Modeling Survey Nonresponse
|
|
Author(s):
|
Tien-Huan Lin* and William Cecere and Jennifer Kali and Michael Jones
|
|
Companies:
|
Westat and Westat and Westat and Westat
|
|
Keywords:
|
classification trees;
nonresponse bias;
response propensities;
design weights;
survey weights
|
|
Abstract:
|
Nonresponse adjustments are often performed on survey weights to reduce the bias of estimates when analyzing complex sample survey data. Several algorithms are available when modeling survey nonresponse for these adjustments, many of which include the option to incorporate design weights. The literature reports uncertain findings related to the benefits of weighting in these settings. Lohr et al. (2015) found no benefits in using weights when modeling response propensity; Lin et al. (2019) showed minor improvements with weighted analysis; and Cecere et al. (2020) and Jones et al. (2021) reported mixed results. A shared limitation of these studies is that they were not specifically designed to assess the use of weights when using these methods. In this paper, we investigate the sensitivity of select classification tree-based algorithms when using weights by conducting a simulation study of a stratified sample design with design weights highly correlated to the outcome variable. We compare unweighted and weighted analysis and evaluate the effect of incorporating design weights on estimating response propensity and reducing nonresponse bias.
|
Authors who are presenting talks have a * after their name.
