Investigating attribute risk associated with noise-multiplied microdata and noise-infused tabular data, and constructing linkage error model for probabilistically-linked data

The thesis presents our research developments throughout the course of my Ph.D study. My research focuses on two aspects. The first aspect is using data perturbation to achieve data confidentiality. We identified and discussed value disclosure risk issues associated with noise-multiplication masking scheme. The main achievement is that we developed measures which could help a data provider with the process of noise generating variable selection in practice, so that the data provider could produce noise-multiplied data with a desired utility-risk tradeoff. We also studied how output perturbation can be used to protect data privacy in a query system, especially the effect of a differencing attack. We developed a perturbation algorithm which effectively protect against the differencing attack. The second research aspect is analysing probabilistically-linked data. Because data linked by a computerised linkage algorithm contains linkage errors, various approaches have been proposed in the literature for analysing linked data so that unbiased estimates could be obtained. Our research is based on using linkage error model to correct the estimation bias. Our research achievement is that we developed a new linkage error model which is more efficient than other models, and unbiased estimates could be obtained while other models fail to do so.