245 – Working with Missing Data: Nonresponse, Imputation, and Suppressed Data
Reducing the Infeasibility and Oversuppression for M-LP Cell Suppression Process
Bei Wang
U.S. Census Bureau
The 2012 Economic Census uses cell suppression to protect sensitive information. We general use 1-LP sequential approach for small table and m-LP partial simultaneous approach for large table. This research is about limiting oversuppression and infeasibility caused by grouping particular cells together in m-LP.
- We have some examples illustrating why and how m-lP causes infeasibility and oversuppression.
- We establish a baseline to evaluate oversuppression. The baseline uses 1-LP, but given a particular n, there n! outcomes needed to run 1-LP multiple times to get an average and variance. We use a 3-d table from the 2012 Economic Census.
- We develop some algorithms to reduce infeasibilities. The general idea is to set m cells wide apart in terms of relationships such that each targeted cell finds its own protection without interacting each other. We identify a class of cells which should be done first and fit this into our algorithm for forming m-groups.
This research should benefit the 2017 Economic Census.