Evaluating Perturbation Impact on Key Travel Models
August 19, 2013
- Zador, Paul and D. Levinson (2013) ￼Evaluating Perturbation Impact on Key Travel Models: Census Statistical Disclosure Control Research Project 1. Prepared for US Census Bureau by Westat.
Introduction: Census Transportation Planning Products (CTPP) are a set of tabulations of the American Community Survey (ACS) designed for transportation planners to estimate, calibrate, and validate transportation forecasting models. Statistical Disclosure Control (SDC) treatments were used for perturbing the ACS 2006-2010 microdata prior to generating the CTPP. The SDC approach maintains ACS respondent confidentiality when releasing aggregate work trip records, since perturbing selected data items reduced the disclosure risk that singly, or in combination, could be used to identify respondents. The table generator for the CTPP processed the SDC-treated microdata for a subset of the CTPP pre-specified tables. The SDC treatment process was developed from an extensive research study, National Cooperative Highway Research Program (NCHRP) 08- 79, which was undertaken in 2010-2011 to develop random perturbation procedures that would produce small area data (e.g., residence to workplace flows for areas approximately the size of Block Groups) that would not violate the Census Bureau’s Title 13 of the U.S. Code, under which the Census Bureau collects its data. Details about the research can be found in NCHRP (2011). During the NCHRP 08-79 research, Westat, under contract to the National Academy of Sciences, worked closely with the transportation planners as represented by the Transportation Research Board represented by the Transportation Research Board and Vanasse Hangen Brustlin (VHB), Inc.
The CTPP tables have been divided into two sets: Set A and Set B. The “Set A” tables were produced from un-perturbed data and “Set B” tables, were produced from perturbed data. The guidelines for generating the Set A and Set B tables were provided to the ACS operations (ACSO) staff. This general approach used perturbed data where tables would have been subjected to DRB disclosure rules, and used the original ACS five-year data for tables where there were no disclosure thresholds. It was designed to retain as much observed ACS data as possible.