A Domain-Based Estimation Framework for Measuring Risk and Utility for Both Input and Output De-Identified Data

• Activity Number: 346
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Avinash Singh and Joshua Borton*+ and Yongheng Lin
• Companies: NORC at the University of Chicago and NORC at the University of Chicago and NORC at the University of Chicago
• Keywords: Disclosure limitation ; de-identification ; disclosure risk ; data utility

Abstract

NORC has recently developed two innovative statistical disclosure limitation (SDL) methods under the X-ID system. AL-PUF (input de-identification) is not based on the traditional identifying/sensitive variable (IV/SV) framework. This method uses summary stats for various analytic domains at aggregate, or micro-group, level and introduces uncertainty via subsampling and calibration. Disclosure risk is the probability of 'too precisely' estimating values for very small (~unique) domains. Data utility is the probability of adequately estimating values for analytically valid domains. Q-PUF (output de-id) is a non-traditional approach where only pre-specified analytic domains are allowed for queries. This prevents differencing attacks, which cause trouble for traditional approaches. Log-linear estimates of sensitive cells are used to judge how well they are being protected. Disclosure risk is the probability of too precisely estimating the values of sensitive cells and data utility is the probability of estimating non-sensitive cells (including complementary ones) with adequate precision. Replication methods are used to obtain empirical measures of risk and utility for the two methods.