Estimating missing values in China's official socioeconomic statistics using progressive spatiotemporal Bayesian...

Area of Science:

• Socioeconomic statistics
• Spatial statistics
• Bayesian modeling

Background:

• Official socioeconomic statistics in China are incomplete at the county level due to significant spatial and temporal missing data.
• Existing imputation methods struggle with large, complex spatiotemporal datasets.

Purpose of the Study:

• To develop and validate a novel imputation method for creating a complete socioeconomic dataset for China.
• To address challenges posed by missing socioeconomic data at the county level.

Main Methods:

• Developed a progressive spatiotemporal (PST) imputation method using Bayesian hierarchical modeling.
• Incorporated spatial autocorrelations, temporal trends, and covariate information from complete variables.
• Compared PST with k-nearest neighbors (kNN), expectation-maximization (EM), singular value decomposition (SVD), and random forest (RF) methods.

Main Results:

• The PST method significantly outperformed kNN, EM, SVD, and RF in imputing missing socioeconomic data.
• Demonstrated the effectiveness of integrating spatial, temporal, and covariate information for improved imputation accuracy.
• Successfully generated a complete socioeconomic dataset for China (2002-2011).

Conclusions:

• The PST method provides a robust solution for imputing missing values in large spatiotemporal socioeconomic datasets.
• This methodology enables the construction of comprehensive national statistics and can be broadly applied to similar data challenges.
• Highlights the importance of advanced statistical techniques for accurate socioeconomic data analysis.