Multiple imputation for missing income data in population-based health surveillance

Area of Science:

• Health Surveillance
• Biostatistics
• Data Science

Background:

• Missing data is a significant challenge in population-based health surveillance.
• Advanced multiple imputation (MI) methods are increasingly used but underreported in health surveillance contexts.
• This study specifically addresses the application of MI for incomplete income data.

Purpose of the Study:

• To examine the application and effectiveness of multiple imputation (MI) for incomplete income data within population-based health surveillance.
• To assess the validity and consistency of MI-imputed income data compared to original data.
• To evaluate the statistical inference capabilities of MI under varying proportions of missing data.

Main Methods:

• Utilized data from the 2002-2003 Los Angeles County Health Survey (N = 8,167).
• Applied MI to impute 1,381 (16.9%) missing household income cases (Federal Poverty Levels - FPLs).
• Assessed validity via random masking and consistency using Z tests; examined statistical inference with Pearson correlation coefficients.

Main Results:

• Imputed and original FPLs showed high consistency (96.3%) for demographics but lower consistency (19.4%) for outcome variables.
• MI demonstrated powerful statistical inference with up to 15% missing data when using well-established covariates.
• Inference power diminished as the missing proportion exceeded 15%.

Conclusions:

• Multiple imputation offers a viable approach for addressing incomplete income data in health surveillance.
• MI's performance is superior for demographic variables compared to outcome variables.
• Reliable statistical inference using MI is achievable for missing data proportions up to 15% with appropriate covariates.