Design and evaluation of an adaptive matrices test

In order to take advantage of the benefits of computerized adaptive testing , a sufficiently large set of content and psychometricaly homogenous items is needed. The goal of this study is to create such an item bank. Matrix it ems measuring non-verbal reasoning ability are designed using explicit item construction rationales. 270 items are evaluated and calibrated with large samples in Katovice, Moscow and Vienna. All item parameters are estimated using a one parameter linear logistic tes t (Rasch-, 1PL-) model and are compared across samples. The item parameters of the corresponding items in the different samples, as well as the parall el items used here, show considerable similarity. A comparison of item diff iculties with parallel items from previous studies yields r=.77 to r=84. It em design rules account for about 60% of all item difficulties.