Combining Probability and Nonprobability Samples by Using Multivariate Mass Imputation Approaches with Application to...

Area of Science:

• Statistics
• Public Health Research
• Survey Methodology

Background:

• Nonprobability samples are widely used but prone to selection bias.
• Mass imputation can enhance the representativeness of these samples.
• Integrating multiple outcome variables requires robust methods.

Purpose of the Study:

• To compare two mass imputation approaches: latent joint multivariate normal model (GERBIL) and Fully Conditional Specification (FCS).
• To assess their effectiveness in integrating multiple outcome variables simultaneously.
• To evaluate these methods using real-world public health data.

Main Methods:

• Latent joint multivariate normal model mass imputation (GERBIL).
• Fully Conditional Specification (FCS) with predictive mean matching.
• Monte Carlo simulation studies.
• Application to Tribal Behavioral Risk Factor Surveillance System and Behavioral Risk Factor Surveillance System data.

Main Results:

• Both GERBIL and FCS with predictive mean matching demonstrated benefits in balancing Monte Carlo bias and variance.
• The methods showed effectiveness in integrating multiple outcome variables.
• Evaluation using combined public health datasets confirmed the utility of the approaches.

Conclusions:

• Mass imputation, particularly GERBIL and FCS, is a valuable technique for improving the representativeness of nonprobability samples in public health.
• These methods offer a way to handle multiple outcome variables effectively.
• The study provides evidence for the practical application of these advanced imputation techniques in survey research.