The Design of p-Optimal Item Pools for Computerized Adaptive Tests

CAT procedures do not function as expected unless they have an item pool that provides appropriate items for the item selection criteria. This paper describes a methodology for describing the characteristics of an item pool that is needed to support a CAT procedure. The methodology is called poptimal because it does not require the specific item that matches a current ability estimate, but only one that provides at least p-proportion of the maximum information from the item. Optimal item pools are very large in size. The use of p-optimality allows the item pools to be reasonable in size while still providing an item pool that allows the CAT to function as planned. An example of the item pool design process is given using a simple example based on the Rasch model.