Imputation of confidential datasets with spatial locations using point process models
by
Thais Paiva
Duke University

We suggest a method to generate synthetic data sets with imputed spatial locations, respecting individuals' confidentiality without losing statistical utility. Our primary interest is to impute the spatial coordinates conditional on the response and explanatory variables. We generate the imputed data sets using spatial point models. The underlying spatial intensities are modeled allowing flexible relationships among the variables and the spatial locations. Using a Bayesian framework, we obtain posterior samples of the intensities, and use them to generate imputed data sets for public release. We verify the quality of the synthetic data, along with the level of confidentiality.

Date received: June 7, 2012