Dr. Ricardo Moura

Title: Single Imputation Synthetic Data: Using it to protect data and how to analyze it.

Abstract:

Data confidentiality is a hot topic nowadays. The safest way of protecting the data is to not share any of it, but at the same time this path leads to the nullity of the data utility. Several researchers have been addressing the problem of maintaining the equilibrium between the protection of the respondents and the quality of the data when releasing ‘altered’ versions of the original data. The author will present two methods of generating synthetic versions of the data, the Plug-in Sampling (PLS) and the Posterior Predictive Sampling (PPS) methods, based on the premise of multiple imputation. It will be shown that multiple imputation is not necessary and that is possible to perform the analysis of the data when only one synthetic version of the data is released. The focus of the presentation will be regarding data that is fully synthesized and partially synthesized assuming the multivariate normal and the multivariate linear regression models.