MULTIPLE IMPUTATION FOR SHARING PRECISE GEOGRAPHIES IN PUBLIC USE DATA.

Area of Science:

• Data privacy and statistical disclosure control
• Geospatial data analysis
• Public health informatics

Background:

• Data stewards must balance data utility with subject confidentiality.
• Releasing detailed geographic information poses significant privacy challenges.
• Existing methods may not adequately protect sensitive location data.

Purpose of the Study:

• To develop and illustrate a novel method for protecting geographic confidentiality in public datasets.
• To enable the release of informative yet privacy-preserving highly resolved geographical data.
• To provide tools for generating and assessing simulated geographic data.

Main Methods:

• Geographic data converted to latitude and longitude coordinates.
• Bivariate response models estimated conditional on sensitive attributes.
• New latitude and longitude values simulated using multiple imputation.
• Regression trees used for generating simulated geographies and attributes.

Main Results:

• Demonstrated a practical approach for anonymizing geographic data.
• Illustrated the method with cause-of-death data from Durham, North Carolina.
• Developed tools for simulating data and assessing disclosure risks.

Conclusions:

• Multiple imputation offers a viable strategy for enhancing geographic data privacy.
• The proposed methods effectively balance data utility and confidentiality.
• This approach supports the ethical and legal release of sensitive geographic information.