Propensity score based conditional group swapping for disclosure limitation of strata-defining variables

Area of Science:

• Statistics
• Data Privacy
• Statistical Disclosure Limitation

Background:

• Categorical variables are crucial for defining subpopulations in data analysis.
• Traditional data swapping methods can distort relationships within and between subpopulations.
• Preserving analytical characteristics within important groups is essential for valid data analysis.

Purpose of the Study:

• To introduce a novel statistical disclosure limitation method called Conditional Group Swapping.
• To address the distorting effects of traditional data swapping on categorical variables.
• To enhance the utility of anonymized data for subgroup analysis.

Main Methods:

• Proposing Conditional Group Swapping for statistical disclosure limitation.
• Utilizing conditional probabilities based on record characteristics for swapping.
• Leveraging propensity score methodology to compute swapping probabilities.

Main Results:

• Conditional Group Swapping demonstrates good utility properties.
• The method effectively reduces data distortion compared to general swapping.
• Analytical characteristics within important subpopulations are better preserved.

Conclusions:

• Conditional Group Swapping offers an effective approach to statistical disclosure limitation for categorical variables.
• The method balances data privacy with data utility, particularly for subgroup analysis.
• This technique is suitable for protecting design and strata-defining variables.