Will the Real Steve Fienberg Please Stand Up? Getting to Know a Population from Multiple Incomplete Files

• Activity Number: 572
• Type: Invited
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Rob C. Hall and Rebecca C. Steorts*+ and Stephen E. Fienberg
• Companies: Carnegie Mellon University and Carnegie Mellon University and Department of Statistics, Carnegie Mellon University
• Keywords: multiple record linkage ; Bayesian methods ; parametric methods ; MCMC ; categorical data analysis ; Census

Abstract

Record linkage is an important statistical problem arising when data about some population of individuals is spread over several files. As kids, we grew up with the game ``Where in the World Is Carmen Sandiego"? Nowadays, the name of the game for the U.S. Census Bureau and other organizations is who's the real Steve Fienberg, where they are dealing with deciding if someone named Steve Fienberg is the same person across multiple lists. This problem has been looked at in the past in the two-file setting and more recently for multiple files. We propose Bayesian parametric methodology for multiples files in which the fields are regarded as independent. A linkage structure represents whether or not individuals i and j in particular lists are matches. We estimate the posterior distribution of this linkage structure via a hybrid Markov chain Monte Carlo (MCMC) algorithm. Our model is tested using the National Long-Term Care Survey (NLTCS) dataset, where ground truth is known since patients are tracked by a unique patient identifier. Challenges regarding convergence and the high-dimensional parameter space are discussed, as well as ongoing extensions towards linkage of Census records.