Disclosure Risk and Data Quality
Lawrence Cox
National Institute of Statistical Sciences, Ellicott City, MD, United States

National Statistical Institutes have the dual responsibility to release high quality data while limiting the risk of disclosure of confidential respondent data. Traditionally, this has been done mostly in an ad hoc manner and only partially. Namely, some form of disclosure limitation, possibly heuristic or ad hoc, is applied to the point where the NSI is comfortable that disclosure risk is acceptable. In some cases, the data are then released without any quality review. In other cases, the data are examined to assure conformance of key estimates between original and disclosure limited (masked) data. If conformal, the masked data are released. If not, some form of direct intervention (tweaking) of the masked data is performed to the point where the NSI is comfortable with both the disclosure protection and the data quality. To move this process from art to science and to ensure greater reliability, reproducibility, data security and data quality, what is needed are (1) quantitative measures of data risk and data quality, (2) functional forms measuring the balance between these two types of measures, and (3) disclosure limitation methods that are sensitive and responsive to both the risk and quality measures. We discuss these issues and present examples of such methods.

Keywords: Statistical disclosure limitation; Balancing data confidentiality and data quality; Rounding; Controlled tabular adjustment

Biography: Larry Cox is Assistant Director for Official Statistics at the National Institute of Statistical Sciences. Previous positions include Associate Director for Research and Methodology, National Center for Health Statistics, Senior Mathematical Statistician for the US Environmental Protection Agency and for the US Census Bureau, and Director of the Board on Mathematical Sciences, US National Academy of Sciences. Dr. Cox is well published and has lectured widely in survey methodology, particularly in statistical disclosure limitation. He is a Fellow of the American Statistical Association, an Elected Member of ISI, and a member of the ISI Council.