A nonparametric method to generate synthetic populations to adjust for complex sampling design features.

Area of Science:

• Statistics
• Survey Methodology
• Computational Statistics

Background:

• Standard statistical methods often assume simple random sampling (SRS), yielding independent and identically distributed (IID) data.
• Applying IID-based methods to complex survey data without accounting for design features can result in incorrect inferences.
• Accurate analysis of complex survey data is crucial for applications like synthetic population generation in missing data or disclosure risk analyses.

Purpose of the Study:

• To propose a method for generating synthetic populations that inverts complex sampling design features.
• To enable the analysis of complex survey data as simple random samples by adjusting for design complexities.
• To facilitate the use of finite population Bayesian inference for complex survey data.

Main Methods:

• Extension of the finite population Bayesian bootstrap literature.
• Development of a nonparametric method to generate synthetic populations from a posterior predictive distribution.
• The method adjusts complex survey data to simulate simple random samples from a superpopulation perspective.

Main Results:

• A simulation study demonstrated the method's effectiveness with a stratified, clustered, unequal-probability of selection sample design.
• The proposed method was successfully applied to generate synthetic populations for the 2006 National Health Interview Survey (NHIS) and the Medical Expenditure Panel Survey (MEPS).
• Generated synthetic populations allow for analysis as simple random samples, mitigating issues arising from complex survey designs.

Conclusions:

• The proposed method provides a robust approach to generating synthetic populations from complex survey data.
• This technique enhances the applicability of standard statistical methods to data from complex sample surveys.
• The approach is valuable for settings requiring synthetic data, such as missing data imputation and disclosure risk assessment.