Combining information from multiple complex surveys

Area of Science:

• Statistics
• Survey Methodology
• Data Science

Background:

• Combining data from multiple surveys is crucial for robust population analysis.
• Existing methods for data integration often struggle with complex sample designs.
• The National Health Interview Survey (NHIS) and Medical Expenditure Panel Survey (MEPS) are key national health surveys.

Purpose of the Study:

• To present a novel nonparametric method for generating synthetic populations.
• To enable the combination of information from multiple complex-sample surveys.
• To facilitate valid statistical inference from integrated survey data.

Main Methods:

• Utilizes a finite population Bayesian bootstrap for nonparametric synthetic population generation.
• Accounts for complex sample designs automatically during synthetic data creation.
• Employs extensions of combining rules for synthetic data to merge estimates.

Main Results:

• Successfully generated synthetic populations that reflect complex survey designs.
• Demonstrated valid inference by combining point and variance estimates.
• Illustrated the method's application using the 2006 NHIS and MEPS datasets.

Conclusions:

• The developed method offers a powerful tool for integrating data from multiple complex-sample surveys.
• This approach allows for the use of standard complete-data analysis software.
• Valid statistical inference can be achieved when combining survey data using synthetic populations.