A1848
Title: Strategies for testing the equality of properties of two distributions of scores
Authors: Christopher Saunders - South Dakota State University (United States) [presenting]
Clarissa Giefer - South Dakota State Univerisity (United States)
Janean Hanka - South Dakota State University (United States)
Semhar Michael - South Dakota State University (United States)
Abstract: A common question when deciding whether to consider a type of forensic evidence is the degree to which trace objects from different sources can be distinguished; this degree is commonly referred to as the discriminating power of the comparison process. Regardless of the quality of the two trace objects being compared, if the comparison process does not give rise to an exclusionary difference between the two sources, then they are said to have indistinguishable profiles with respect to the comparison process. A new level-alpha test for the equality of scores is defined, based on all pairwise comparisons using both within-source and between-source comparisons. The applicability of this method is demonstrated using a handwriting dataset; this test compares average within-writer scores with average between-writer scores. The approximate distribution of the test statistic is discussed and the corresponding p-values from all tests of equality between all pairs of sources in a mixture model are used to make a statement about the proportion of p-values that arose from two writers with indistinguishable profiles, which in turn leads to an indirect inference of the discriminating power.
