Effects of Low Category Response Rates on Common Polytomous Item Response Theory Item Calibrations

• In Event: Item Response Theory With Complex Data

Abstract

When an ordered polytomous response scale is used, proportions of responses in the end categories tend to have low response rates. The purpose of this simulation study was to investigate and compare the impacts of low-response categories on item calibrations with the graded response model. When items with a more extreme true location have a low proportion of responses, i.e., 1% or 2% in this study, RMSE and relative bias of item discrimination tended to be higher, and to a larger degree when the GPC model was used as compared to the GR model. In this case, researchers are encouraged to use the GR model over the GPC model.

Authors

• Ki Cole, Oklahoma State University

• Insu Paek, Florida State University

• MOSTAFA HOSSEINZADEH, Oklahoma State University