ERIC Number: ED442830
Record Type: Non-Journal
Publication Date: 2000-Apr-25
Reference Count: N/A
Evaluation of the IRT Parameter Invariance Property for the MCAT.
Kelkar, Vinaya; Wightman, Linda F.; Luecht, Richard M.
The purpose of this study was to investigate the viability of the property of parameter invariance for the one-parameter (1P), two-parameter (2P), and three-parameter (3P) item response theory (IRT) models for the Medical College Admissions Tests (MCAT). Invariance of item parameters across different gender, ethnic, and language groups and the invariance of ability parameters with respect to test difficulty were assessed. The study also sought to test the stability of ability estimates obtained for random sample "X" using calibrations from different groups and to identify the most efficient IRT model for the MCAT data. All data were from the fall 1994 administration of the MCAT, with 9 random samples of 1,100 drawn from each of 3 test sections (out of 16,520 eligible test takers). The assumption of unidimensionality was first tested, and evidence was found of two or more underlying dimensions affecting test performance. The three IRT models were fit to the dichotomous response data from every sample, and all three showed adequate fit for the MCAT data, with the 1P model item estimates having the smallest estimation error, although the differences among 1P, 2P, and 3P models were very small in magnitude. Evidence was found to support the conclusion that item and ability parameters are stable/invariant with respect to gender and racial/ethnic and language groups for all models. Ability estimates also appear to be invariant with respect to test difficulty for all models. Nine appendixes contain data tables that supplement the discussion. (Contains 28 tables, 59 figures, and 11 references.) (SLD)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Authoring Institution: N/A
Identifiers - Assessments and Surveys: Medical College Admission Test