Browse All

Theses & Dissertations


  • Submissions (Articles, Chapters, and other finished products)

Detecting differential item functioning using the DINA model

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Wenmin Zhang (Creator)
The University of North Carolina at Greensboro (UNCG )
Web Site:
Terry Ackerman

Abstract: "DIF occurs for an item when one group (the focal group) of examinees is more or less likely to give the correct response to that item when compared to another group (the reference group) after controlling for the primary ability measured in a test. Cognitive assessment models generally deal with a more complex goal than linearly ordering examinees in a low-dimensional Euclidean space. In cognitive diagnostic modeling, ability is no longer represented by the overall test scores or a single continuous ability estimate. Instead, each examinee receives a diagnostic profile indicating mastery or non-mastery of the set of skills required for the test, namely the attribute mastery pattern. The purpose of the study had three objectives; first to define DIF from a cognitive diagnostic model perspective; second, to identify possible types of DIF occurring in the cognitive diagnostic context introduced into the data simulation design; finally, this study compared traditional matching criteria for DIF procedures, (e.g., total score) to new conditioning variable for DIF detection, namely the attribute mastery patterns or examinee profile scores derived from the DINA model. Two popular DIF detection procedures were used: Mantel-Haenszel procedure (MH) and the Simultaneous Item Bias Test (SIBTEST) based on total test score and profile score matching. Four variables were manipulated in a simulation study: two sample sizes (400 and 800 examinees in each group), five types of DIF introduced by manipulating the item parameters in the DINA model, two levels of DIF amount on a 25-item test (moderate and large DIF), and three correlations between skill attributes for both groups (no association, medium association and high association). The simulation study and the real data application demonstrated that, assuming cognitive diagnostic model was correct and the Q-matrix was correctly specified, attribute pattern matching appeared to be more effective than the traditional total test score matching observed by lower Type I error rates and higher power rates under comparable test conditions."--Abstract from author supplied metadata.

Additional Information

Language: English
Date: 2006
DIF, Cognitive assessment models, Euclidean space, diagnostic profile, skills
Educational tests and measurements