Estimation of IRT Difficulties in Adaptive Testing: An LCT Assisted Estimation Method
collaborative work testing
learning check testing
item response theory
dually adaptive IRT
It seems difficult to estimate the difficulties of the problem items in the adaptive testing even if we use the EM-type IRT (item response theory) method because the response matrix made by the adaptive testing becomes sparse which is not enough to estimate the item difficulties and examinees’ abilities. In this paper, we propose to use the known ability values from other testing results if the examinees are the same. In the follow-up program systems, we have been using three kinds of testing, the LCT (learning check testing), the CWT (collaborative work testing), and the FPT (follow-up program testing), in which the LCT provide accurate estimates for ability values. We use the LCT’s ability values in the CWT estimation. For stable estimation, we have used one-parameter estimation method (Rasch model) instead of two-parameter estimation method which is used in common. The estimation is performed well. In addition, comparing the method of 2017 CWT difficulty estimation with that of the 2016, we have found that in estimating the difficulty parameters using artificial response matrix where 0 values are imposed in the vacant elements the estimates of difficulty parameters are heavily biased to difficult problem side.