Creating Synthetic Data for Complex Surveys Using the Research and Development Survey: A Comparison Study

Area of Science:

• Statistics
• Data Science
• Survey Methodology

Background:

• Synthetic data generation is increasingly used to balance data utility and confidentiality.
• Creating synthetic data for complex surveys presents significant challenges.

Purpose of the Study:

• To compare methods for incorporating survey design information into synthetic data generation.
• To evaluate the impact on data utility and disclosure risk for complex surveys.

Main Methods:

• Utilized the Research and Development Survey (RANDS) data.
• Compared parametric (logistic/linear regression) and nonparametric (CART) methods.
• Incorporated survey design elements like stratification, clustering, and sampling weights.

Main Results:

• Parametric methods showed improved data utility when survey design information was used as predictors.
• Nonparametric methods (CART) achieved higher data utility but introduced a marginal increase in disclosure risk.
• Evaluation metrics included confidence interval overlap, propensity scores, and re-identification probabilities.

Conclusions:

• Survey design information is crucial for enhancing synthetic data utility in complex surveys.
• Method choice involves a trade-off between data utility and disclosure risk.
• Findings inform best practices for generating privacy-preserving synthetic survey data.