All these item selection rules
have in common that they try to maximize information obtained about the candidate in order to minimize the error of estimation. Chang e Ying (1999),on the other hand, observed that during the early stages of CAT administration, the ability estimate is not very precise yet. They reasoned that selection of very informative items at an uncertain ability estimates might not be optimal in practice. As an alternative, they proposed alpha-stratified CAT, where the item bank is stratified with respect to the discrimination parameter.