Multiple Imputation with Massive Data: An Application to the Panel Study of Income Dynamics

Area of Science:

• Statistics
• Econometrics
• Survey Methodology

Background:

• Multiple imputation (MI) is a robust statistical technique for addressing missing data.
• Its application to large, complex datasets like the Panel Study of Income Dynamics (PSID) presents practical challenges.
• Current hot deck methods in PSID, while simple, introduce significant random fluctuations in wealth data.

Purpose of the Study:

• To evaluate the practicality and effectiveness of multiple imputation (MI) for handling missing wealth data in the 2013 PSID.
• To compare MI's performance against the existing univariate hot deck approach.
• To address and overcome operational challenges associated with MI in complex survey data.

Main Methods:

• Utilized a sequential regression/chained-equation approach with IVEware software for multiple imputation.
• Applied MI to cross-sectional wealth data from the 2013 PSID.
• Compared analyses from MI-imputed data with those from the hot deck method.

Main Results:

• Multiple imputation (MI) demonstrated superior performance by preserving correlation structures, including associations between wealth components and relationships with sociodemographic factors.
• MI facilitated complete data analyses and increased analytical efficiency.
• The study successfully navigated practical difficulties such as non-normal variables, skip patterns, and multicollinearity.

Conclusions:

• Multiple imputation (MI) offers significant improvements over traditional hot deck methods for analyzing complex survey data, particularly for wealth variables.
• The sequential regression/chained-equation approach is a practical and effective strategy for implementing MI in large-scale surveys.
• MI is recommended for its ability to enhance data quality, preserve relationships, and increase efficiency in statistical analyses, especially when missing data is substantial.