Chris Saunders
South Dakota State University

Using Subsampling to Investigate the Dependency of Match Probabilities on the Size of Writing Samples

A random match probability (RMP) of interest in handwriting analysis is the chance of randomly selecting two different individuals from some relevant population and then randomly selecting a writing sample from each individual that are declared to “match” by a given comparison procedure.  A complementary probability, the random nonmatch probability (RNMP), is the chance of randomly selecting a single individual and then randomly selecting two writing samples from the selected individual’s body of handwriting that fail to “match.”  In handwriting analysis, the RMP and the RNMP are standard measures of a comparison procedure’s ability to discriminate among writers; both depend upon the comparison procedure used and the sizes of the writing samples being compared, as well as the relevant population of individuals to which the comparison procedure is being applied. In this presentation, I will discuss strategies for estimating the RMP with a specialized U-statistic (and corresponding process), the asymptotic properties of the estimators, and subsampling methods that are useful for developing computationally efficient approximations of these U-statistics. For this presentation, I will focus on a research data set of handwriting samples collected by the FBI Laboratory Division that has been processed by the FLASH ID system.

Refreshments at 3:45pm in Snedecor 2101.