A0390
Title: Efficient calibration of items in mixed format achievement tests using optimal design methodology
Authors: Ellinor Fackle-Fornius - Department of Statistics (Sweden) [presenting]
Frank Miller - Stockholm University (Sweden)
Abstract: For large achievement tests, like national tests in school, item calibration is used to determine characteristics of an item, such as difficulty and discrimination. It is important to estimate the item characteristics with as good precision as possible, before the item can be administered in an operational test. We propose an ability-matched item allocation method based on optimal design theory. The method is adapted to handle test items of varying formats analysed using different IRT models, such as the 2-parameter and 3-parameter logistic, as well as graded response models. We demonstrate that the proposed optimal design method leads to increased efficiency and illustrate for which item types it performs best. We also present the results of a real calibration study conducted for the national test in mathematics in Sweden, where the method was evaluated.
