Summary: | The purpose of this study is to compare the performance of Between-item dimensionality-based
Multidimensional CAT designs and to examine the effect of content balancing on different MCAT designs. For
this purpose, real data set obtained from English Proficiency Test (EPT), which was administered by Hacettepe
University between 2009 and 2013 academic years, was used. The three dimensional item pool consisted of
real items measuring students’ listening, grammar and reading abilities. Item pool consisted of 555 items which was calibrated with 2PL between-item MIRT model. In this study, two different theta estimation
(Fisher scoring and Bayesian MAP) methods, three different fisher information based item selection methods
(A-optimality, D-optimality and Random) and three different precision based termination methods were used
in order to determine the best MCAT design. In addition, results of MCAT algorithms with content
distribution and without content distribution were compared so as to examine the effect of content balancing in
the context of MCAT. The results of each MCAT condition were compared with respect to, reliability index,
SEM, RMSD values associated with each dimension. According to results, both Bayesian MAP and Fisher’s
scoring methods yielded similar results when A-Optimality item selection method was used. However, Fisher’s
scoring method appeared to be affected from item selection methods and content balancing. Moreover, average
number of items tended to increase and reliability coefficients tended to decrease somewhat, while standard
error and RMSD values tended to decrease when content balancing was applied in MCAT.
|