This is an interesting research to investigate the effects of item parameter estimation error in VL-CAT. It is common to assume the item parameters as true values in CAT, but this study provided evidence that the error for estimating the item parameters should not be ignored, otherwise the SE for ability estimation will be underestimated, and consequently affected the whole procedure of CAT. When the sample size decreased, the effects of item parameter estimation error would become more serious, as the errors became larger. In VL-CAT, given the errors existed for item parameter, the item information would be overestimated. When employing the optimization (maximum Fisher information) as item selection method, tests tend to be spuriously short, especially with the combination of stopping rule as CSE.
Information provided by this study is quite useful when conducting VL-CAT. However, as the phenomenon is realistic, what can researchers and users can do to reduce the effect of item parameter estimation? It seems advisable to use larger sample size, which would result in more precise item parameter. As mention in discussion, the corrected SE (eq(1)) is not feasible in CAT, then what can be done to control the SE in CAT?
Errors for item parameter become larger as sample size decreases, any suggestion for the acceptable level for the SE of item parameter? Or suggested sample size, or the ratio of sample size to item bank?
Various item selection methods, stopping rules, exposure control methods, and content balance control methods could be combined to evaluate the effect of item parameter estimation. Methods that depend less on item parameters would probably be preferred.