This study was based on a four-cluster rotation design of 13 linked test booklets from PISA 2006 science data. It investigated effects of item positions on their difficulty and discrimination parameter estimates obtained from one and two parameter IRT Partial Credit models. The analyses were done separately for 57 test language groups from 53 countries with a total of about 340,000 students.
The results revealed that for all of the test language groups the items tended to become more difficult when they were located later in the test. However, a high linear relationship between the item difficulty estimates by the four cluster locations was found. Moreover, open-ended items seemed to show more change than items in other formats. There were small variations in the cluster locations for the item point-biserial discrimination and the item discrimination parameter from the two-parameter Partial Credit model across the test language groups.
Le, L.T. (2007, July). Effects of item positions on their difficulty and discrimination : A study in PISA Science data across test language and countries. Paper presented at the 72nd Annual Meeting of the Psychometric Society, Tokyo.