Investigating Different Treatments of Rapid Guessing Through Concepts and Models Grounded in Missing Data Theory

• In Event: Novel Approaches for Model Fit

Abstract

Increased availability of time information through computer-based assessment enabled new ways to measure test-taking engagement. One way is through the distinction of solution- and rapid guessing behavior. To deal with rapid guessing prior research recommended response-level filtering. Filtering can lead to biased parameters if rapid guessing depends on the measured trait or (un-)observed covariates. Therefore a model based on Mislevy and Wu (1996) was applied, enabling to investigate nonignorability and compare different approaches in terms of sensitivity and model fit through model parameterization. Lower ability test-takers rapid guessed more frequently, indicating a violation of the ignorability assumption by filtering. Model-based approaches showed better fit and concurrent validity compared to response-level filtering, illustrating the need for methods able to handle nonignorability.

Authors

• Tobias Deribo, DIPF | Leibniz Institute for Research and Information in Education

• Ulf Kroehne, German Institute for International Educational Research

• Frank Goldhammer, DIPF | Leibniz Institute for Research and Information in Education