Parametric and nonparametric IRT models for assessing differential item functioning
Karen S. Lee · Human Biology · 2003
The results of Michigan's K--12, statewide assessment program are currently being used for accreditation and accountability purposes. When the results of such tests are used to make such high-stakes decision, the need to insure test fairness becomes of paramount importance. This study sought to provide an analysis of the psychometric properties of the 2002 8 th grade mathematics results of the Michigan Educational Assessment Program (MEAP) test. Of particular concern was the identification of any possible differential item functioning (DIF) that could result in overall test bias. Comparative groups included in the study were White vs. African-American examinees and "low income" vs. "not low income". MEAP results were downloaded from the Michigan Department of Education's public domain website. An assessment of item level bias was conducted using the nonparametric model based procedure, SIBTEST (Stout, 1996), to identify possible suspect items. A comparative analysis was later conducted by employing a parametric procedure for assessing DIF (BILOG-MG). While both programs identified large numbers of items measuring significant beta estimates, neither program flagged significant numbers of items exhibiting DIF levels that could be considered to have practical significance. In addition, the number of items tending to favor the reference group over the focal group were almost equally matched by those favoring the focal group over the reference group.