Evaluation of Person Fit Under the Least Squares Distance Model of Cognitive Diagnosis

• In Event: Roundtable Session 117
  In Roundtable Session: Item Response Theory

Abstract

This study addresses issues of aberrant processing of latent attributes that underlie the responses on binary test items in the framework of the least-squares distance method for cognitive diagnosis (LSDM). Proposed is an approach to investigating such aberrancy by adapting the dependable person-fit statistic HT (Sijtsma & Meijer, 1992) for use with generated binary responses on attributes across locations of examinees on the IRT logit scale. Discrepancies between the item characteristic curve (ICC) of an item and the LSDM recovery of the ICC are also examined for detection of aberrancy in responses to items and/or processing of underlying attributes across locations on the logit scale. Extensions of these approaches and illustrations with real and simulated data are also considered.

Authors

• Dimiter M. Dimitrov, George Mason University

• Dimitar V. Atanasov, New Bulgarian University, Bulgaria