Model Identifiability for MCCFA Model and Its Applications to Differential Item Functioning Detection

• Activity Number: 131
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: International Chinese Statistical Association
• Author(s): Yu-Wei Chang* and Rung-Ching Tsai and Nan-Jung Hsu
• Companies: University of Illinois at Urbana-Champaign and National Taiwan Normal University and National Tsing Hua University
• Keywords: Differential item functioning ; Identifiability

Abstract

Differential item functioning (DIF) studies aim to understand whether an item behaves the same, for examinees with the same ability, among groups. Multiple-group categorical confirmatory factor analysis (MCCFA) is a common practice for DIF detection on polytomous items. However, traditional MCCFA practices usually set too many constraints on MCCFA, and may further cause model mis-specification. In the present study, we illustrate and justify the identifiability constraints for MCCFA, so that we can avoid the model mis-specification in utilizing MCCFA for DIF detection. As a consequence, the size for DIF detection can be well-controlled, and the test results would be more accurate than mis-specified cases.