Investigating Rasch and Two-Parameter Logistic (2PL) Short-Form Model Selection by Minimizing Out-Sample Classification Error

• In Event: 68.062 - Poster Session 15
  In Poster Session: 68.062-3 - Poster Session: Innovations on Diverse Methodological Approaches

Abstract

This study explores the usefulness of multiple cross-validation and bootstrap methods as tools to select between various short-forms of binary Rasch and 2PL tests, using classification error to an external criterion variable as the selection criteria. A simulation study is performed in R to investigate how these methods perform across different factors. A total of 12 methods in the Rasch and 9 in the 2PL simulation are investigated. In addition, the best-performing procedures are applied to short-forms of the Narcissistic Personality Inventory (NPI-40). The results indicate that the simplest methods, e.g. resubstitution error and leave-one-out cross-validation, have the best results. Explanations for the results and suggested areas of future research are provided in the discussion section.

Author

• Anthony Raborn, University of Florida