Reporting Overall Scores and Domain Scores of Bifactor Models

• In Event: Marriott Roundtable Session 29
  In Roundtable Session: Estimation and Reporting of Multidimensional Item Response Theory Models

Abstract

This study investigated several methods to generate overall scores and domain scores based on bi-factor model, and to compare them with the MIRT method under different simulation conditions. The compared methods were: original scores from bi-factor model (Bifactor-M1); summed original scores from bi-factor model (Bifactor-M2); weighted sum original scores from bi-factor model based on all the items (Bifactor-M3 and Bifactor-M5); weighted sum original scores from bi-factor model based on items of each dimension (Bifactor-M4 and Bifactor-M6). The factors were sample size, test length and correlation between dimensions. An empirical study for science test of National College Entrance Examination was conducted as well. The results indicated that Bifactor-M4 and Bifactor-M6 generally performed better than the other proposed methods.

Authors

• Yue Liu, Beijing Normal University

• Hongyun Liu, Beijing Normal University