During the early stages of CAT, items are selected from strata with lower discrimination indices. These items are less informative, and tend to have low exposure rates. Limiting item selection to these strata increases the exposure rates of these items and reduces the problems of underexposure. The best results with exposure control problems have been obtained when methods for dealing with over exposure were combined with methods that deal with underexposure. Veldkamp, Verschoor e Eggen (2010) therefore proposed a method that combined both approaches.
Item selection in CAT has been an important research topic for many years, also because the adaptive item selection process is what makes CAT different from administering linear test forms. This paragraph only covers a small part of the literature on it. Most of the existing papers are about CAT with dichotomously scored items that are calibrated with a unidimensional IRT model. Nowadays, the focus is shifting more towards developing methods for CAT with more complex item types that are calibrated with more complicated, often multidimensional, IRT models. Even though impressive results have been obtained, there are still many more areas that have to be studied.