Improving items that do not fit the Rasch model

Development of measurement scales is a time- and resource-consuming process with many stages. This chapter introduces a systematic approach to improving poor performing items based on Nielsen and Kreiner's strategy for scale improvement. This strategy combines the results from graphical log-linear Rasch modeling with the results of a subject matter analysis of items within the particular theoretical framework to facilitate systematic decisions about which items should be modified and which items should be exchanged improve the scale psychometrically, that is toward functioning as a pure Rasch model (RM). The chapter briefly covers the assumptions and requirements of the graphical log-linear Rasch model (GLLRM) and the effects of local response dependence (LD) and differential item functioning (DIF) in such a model on the reliability and validity of the scale in question. Next, it describes the strategy for scale improvement in detail, and presents some results of applications of the strategy.