A theoretical and empirical investigation of factor analytically-based matching criteria in differential item functioning

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Robert Lewis Johnson (Creator)
Institution
The University of North Carolina at Greensboro (UNCG )
Web Site: http://library.uncg.edu/
Advisor
Lloyd Bond

Abstract: Modern investigative procedures to detect differential item functioning (DIF) match examinee groups on ability before comparison. The validity of DIF procedures depends, in part, on the unidimensionality of the matching criterion; however, the most popular proxy for ability, examinees' raw scores on the test, consists of items with varying levels of multidimensionality. This study explored the efficacy of two matching criteria—total score and factor score—as tests become increasingly multidimensional. The investigation used empirical data to determine the consistency of flagging items as displaying DIF when matching with total and factor scores in tests that varied in factorial complexity. In addition, in a series of simulations, increasingly complex factor structures were created. In one variation referred to as Factor Structure 1, items loaded either on a first factor (the target factor) or a second factor (the nuisance factor)—but not both. Factor Structure 2 was composed of items that loaded primarily on a first factor and secondarily on a second (or nuisance) factor. Bias was simulated in items associated with the nuisance factor.

Additional Information

Publication
Dissertation
Language: English
Date: 1995
Subjects
Factor analysis
Educational tests and measurements
Examinations $x Design and construction

Email this document to