Models for Semi-Ordered Data to Address "Not Applicable" Responses in Scale Measurement

• In Event: Measuring and Modeling Test-Taker Behaviors and Experiences

Abstract

“Not Applicable” response options commonly appear in educational measurement scales. Completely ignoring these responses or treating them as missing data may lose information and cause measurement bias, especially when the Not Applicable responses actually correlate to the underlying trait of measurement. To handle this situation, we offer a new model to directly incorporate Not Applicable responses into the estimation of the latent trait. This model was developed based on the nested relationship between the generalized partial credit model and the nominal response model. We use an application to operational data and a small simulation study to evaluate the model's ability to produce valid information about person and item parameters.

Authors

• Sherry Zhou, University of Florida

• Anne Corinne Huggins-Manley, University of Florida

• James Algina, University of Florida