Application of computerized adaptive classification test approach on English placement test results

In this study, it was aimed to investigate the applicability of the CACT approach when the classification is made in two, three, or four categories on the response patterns of 256 one-dimensional readıng items of the English placement test administered to 919 individuals studying at Bilkent University School of Foreign Languages. In the study, when Maximum Likelihood Estimation (MLE), Weighted Likelihood Estimation (WLE), Expected a Posteriori (EAP) methods were used as ability estimation methods, average test length (ATL), average classification accuracy (ACA), correlation between real and estimated thetas (r), Root Mean Square Error (RMSE) indicating the standard error of estimation, mean absolute error (MAE) and bias values of Sequential Probability Ratio Test (SPRT), Generalized Likelihood Ratio (GLR) and Confidence Interval (CI) classification criteria and how they change in two- three, and four-category classification according to Estimated Ability-based Maximum Fisher Information (MFI-EB) and Kullback-Leibler Information (KLI-EB), cut-point based Maximum Fisher Information (MFI-CB) and Kullback-Leibler Information (KLI-CB) item selection methods were examined. In the study, a total of 108 conditions were created for the simulation based on real data using R software. At the end of the study, it was concluded that the CI classification criterion showed high performance for test effectiveness (ATL, ACA), and the WLE classification criterion showed high performance for measurement precision (correlation, bias, RMSE, MAE) in two, three and four category classifications. The GLR classification criterion was found to be successful for test effectiveness in four-category classification with MLE and EAP ability estimations and MFI item selection methods.