Comparison of measurement precision values obtained from different conditions in computerized adaptive testing

In this study it is aimed to investigate the change in measurement precision estimation values that has been obtained from two different sizes of item pools in a CAT application under several item pool restrictions such as content balancing and item exposure control according to several item selection and ability estimation methods. For this aim, ability parameters related to a group of 2000 people and two different item pools one of which is 320 and the other is 1000 have been developed simulatively. Later, the conditions of the study in both balanced and unbalanced contents are constructed by developing a CAT application using catR package in RStudio with item selection, ability estimation and item exposure control methods and then these conditions of the CAT application are carried out by using simulations of two different item pools and the same participants. The 48 different conditions developed in the study are compared by making 156 replications. As a result of simulations, measurement precision estimation values (RMSE, correlations and bias) are computed and how these values are changed under 48 conditions are investigated. The findings obtained from the study have showed that the item pool with 1000 items has better results compared to the item pool with 320 items in terms of RMSE, bias and correlation values. Moreover, according to RMSE, bias and correlation values; MFI is seen to be the most effective among item selection methods and among ability estimation methods, EAP is found out to be the most efficient method. In addition; an increase in RMSE value has been observed in balanced content when it is compared with the value in unbalanced content. It is also found out that IEI method is more effective and has higher performance when it is used with MFI and MSI methods together.