Many important questions about SSA’s programs could be investigated by the RDRC research community using only county-level statistics generated from SSA’s administrative data. However, such statistics are often unusable by researchers because of cell suppression procedures used to protect individual privacy when disclosing statistics based on small numbers of observations. Recognizing this, federal agencies are increasingly using “differential privacy” noise-infusion methods, which offer better privacy protection, less statistical bias, and minimal loss in statistical precision. This project proposes to assess whether these methods could be used to protect the privacy of individual records when releasing research or other policy-relevant statistics at SSA. Our aims are as follows:
• Estimate county-level DI statistics: We will estimate DI exit rates for successful return to work by county from the Disability Analysis File.
• Apply alternative privacy-protection methods: We will apply traditional cell suppression and the proposed differential privacy noise-infusion method to the county-level DI exit rates.
• Quantify tradeoffs: We will compare privacy-protection methods in terms of privacy loss, statistical
bias, and statistical precision, specific to the SSA data setting.
• Produce manuscript: We will prepare a paper summarizing our findings.