Analyzing Spatial Responses: A Comparison of IRT-Based Approaches

Abstract

We investigate two approaches for analyzing spatial coordinate responses using models inspired by Item Response Theory (IRT). In the first, we use a two-stage approach to first construct a pseudo-response matrix using the spatial information and then apply standard IRT techniques to estimate proficiency and item parameters. In the second approach, we introduce the Spatial Error Model and use the spatial coordinates directly to infer information about the true locations and participant precision. As a motivating example, we use a study from forensic science designed to measure how fingerprint examiners use minutiae (small details in the fingerprint that form the basis for uniqueness) to come to an identification decision. The study found substantial participant variability, as different participants tend to focus on different areas of the image and some participants mark more minutiae than others. Using simulated data, we illustrate the relative strengths and weaknesses of each modeling approach, and demonstrate the advantages of modeling the spatial coordinates directly in the Spatial Error Model.

Keywords

This work was partially funded by the Center for Statistics and Applications in Forensic Evidence (CSAFE) through Cooperative Agreements 70NANB15H176 and 70NANB20H019 between NIST and Iowa State University, which includes activities carried out at Carnegie Mellon University, Duke University, University of California Irvine, University of Virginia, West Virginia University, University of Pennsylvania, Swarthmore College and University of Nebraska, Lincoln.