Item response theory : the application of both dichotomous and polytomous item response models to a provincial exam data set
Date
1998
Authors
Boughton, Keith Andrew
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This study was conducted to investigate the feasibility of using binary logistic models with the dichotomously scored items (multiple choice) and partial credit models with the polytomously scored items ( open-ended) from the January administration of the 1996 British Columbia grade 12 Provincial Math Examination. The findings in regard to feasibility show a need for better ways of testing for unidimensionality. The item-fit statistics showed the 2-parameter (2PL) and 3-parameter logistic (3PL) models as better fitting models than the I-parameter (IPL) model with the multiple choice section of the exam. The I-parameter partial credit model (lPCM) from the RUMM software program (Sheridan, Andrich, & Luo, 1997) showed 1 of 8 items as fitting the model in the openĀended section of the exam. Invariance of parameter estimates was supported for the 1 PL, 2PL, and 3PL models. The consistency of student classification results suggested that the I PL/I PCM model combination had closer estimates to the Provincial Exam raw scores than the 2PL/2PCM model combination. The item characteristic response functions (ICRFs) from the IPCM were found to be very meaningful for the interpretation of how the item categories were operating within the open-ended section of the exam. Lastly, it should be noted that the different software programs housing the IRT models did not produce equivalent estimates and therefore, the same software should be used in order to have comparable results. The above suggests that it may be feasible to use IRT models with the Provincial Math Examinations.