Dr Robin Mitra is a visiting academic within Mathematical Sciences at the University of Southampton.
2008 - 2017 Lecturer in Statistics – University of Southampton, UK
2008 PhD in Statistics - Duke University, Durham NC
2006 M.S. in Statistics - Duke University, Durham NC
2004 BSc in Mathematics, Operations Research, Statistics and Economics - University of Warwick, Coventry UK
My main areas of research are in multiple imputation approaches to problems in missing data and data confidentiality.
Missing data is a common problem that arises in many fields. If dealt with carelessly it can result in a loss of efficiency and cause analyses to be distorted. Multiple imputation is the process by which one repeatedly draws values for the missing data from statistical models that characterize relationships between variables in the data. In this way the data set can be completed with plausible imputed values and statistical analysis can be conveniently performed on the imputed data sets. A key requirement for this approach is to ensure that imputation models correctly capture the relationships between variables; this becomes difficult in problems of higher dimensions. My research work has focused on designing imputation strategies to be able to generate plausible missing values in problems involving many variables, where relationships may not be so easy to identify.
Another important application of multiple imputation is in the area of data confidentiality. Agencies (such as US Census Bureau) collect data from individuals on a range of issues which researchers are keen to analyze, however agencies are also required to protect the privacy of participating individuals. In order to achieve this often the data is distorted somewhat, perhaps by adding random noise or swapping records, but this damages statistical properties of the data and thus reduces its utility for users. If we can design good statistical models that reflect properties among variables in the data then we could draw synthetic values to replace some or even all of the values in the data set. If draws are from their posterior predictive distributions then we can obtain inferences from the synthetic data sets that are similar to inferences obtained from the original sensitive data. However as some values in the synthetic data set are drawn from statistical distributions, confidentiality can be preserved. Designing good imputation models as well as evaluating risks with this approach is an active area of my research.
Southampton Statistical Sciences Research Seminar Organiser
Mathematics with Statistics Programme Co-ordinator
Website Co-ordinator for the Statistics Group