Activity Number:
|
139
- Improving Population Inference Using Statistical Data Integration
|
Type:
|
Topic Contributed
|
Date/Time:
|
Monday, August 8, 2022 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Survey Research Methods Section
|
Abstract #322955
|
|
Title:
|
Embedded Multilevel Regression and Poststratification: Model-Based Inference with Incomplete Auxiliary Information
|
Author(s):
|
Katherine Li* and Yajuan Si
|
Companies:
|
University of Michigan School of Public Health and University of Michigan
|
Keywords:
|
incomplete poststratifiers;
synthetic population;
Bayesian bootstrap;
sequential imputations
|
Abstract:
|
Multilevel regression and poststratification (MRP) has become a popular approach for selection bias adjustment and subgroup estimation. MRP stabilizes subgroup estimation by fitting multilevel models and balances the sample decomposition by poststratification to match the population distribution of the auxiliary variables. The key to success is the availability of a rich set of predictive auxiliary variables for the survey outcomes, but their joint distribution in the population is often unknown. Embedding the estimation of population cell counts during poststratification into MRP, we develop an integrative inference framework: embedded MRP (EMRP). We first generate synthetic populations of the auxiliary variables and then implement MRP in propagation of all sources of estimation uncertainty. Via simulation studies we compare different methods and demonstrate the improvements of EMRP on the tradeoff between bias and variance and the capability to yield valid subpopulation inferences of interest. We apply the EMRP to the New York Longitudinal Survey of Wellbeing and find that the improvement is primarily on subgroup estimation with efficiency gains.
|
Authors who are presenting talks have a * after their name.