Item Parameter Recovery of the Compensatory 2PL MIRT Model by MH-RM and MML (Poster 33)

• In Event: Friday Poster Session 9:35 am
  In Poster Session: Division D Poster Session 2

Abstract

This study assessed the item parameter recovery accuracy by MH-RM and MML under the compensatory 2PL MIRT for binary data. A simulation was designed as a 3x3x3x2 mixed factorial study with number of latent dimensions, model structure, and latent distribution as three between-subjects factors, and estimation method as the within-subjects factor. Results indicated that MML and MH-RM had comparable performances under two-dimensional models, but MH-RM was superior when model dimensionality increased. Nonnormal traits and nonsimple structures were associated with higher estimation errors, but the impacts of these factors were less pronounced than dimensionality. Item intercepts were more accurately estimated than item discrimination in general. Discussions and recommendations based on the study findings were presented.

Authors

• Hoan Do, Clinical Outcomes Solutions

• Gordon P. Brooks, Ohio University