Statistical Science
Professor of Statistical Science
I have three primary areas of methodological research:
The first is statistical disclosure limitation, i.e., how statistical agencies and other data producers can disseminate or share data that protect the confidentiality of data subjects’ identities and sensitive attributes. I develop new approaches to confidentiality protection, including frameworks for obtaining valid inferences from data that have been altered to protect privacy. I develop methods that satisfy formal privacy guarantees, such as differential privacy. I also work to translate theory into practice, supervising or consulting on the creation of public use data products for government agencies.
My second main area of methodological research is missing data methods, in particular multiple imputation (MI). In MI, we fill in the missing values with multiple draws from probability distributions. I develop frameworks for handling complicated missing data scenarios, including nonignorable missing data, data fusion (also known as statistical matching), and methods for automatic editing of measurement errors. An integral component of my research is adapting the multiple imputation framework -- originally developed for missing data imputation in large sample surveys -- to handle methodological problems in other contexts.
My third main area of methodological research is combining information from different data sources. This includes methods for propagating uncertainty in downstream inferences when using inexactly linked data files, as well as methods for fusing information in convenience samples with information in traditional surveys.