Using the asymmetry of item characteristic curves (ICCS) to learn about underlying item response processes

Abstract

In this chapter, we examine how the nature and number of underlying response subprocesses for a dichotomously scored item may manifest in the form of asymmetric item characteristic curves. In a simulation study, binary item response datasets based on four different item types were generated. The item types vary according to the nature (conjunctively versus disjunctively interacting) and number (1–5) of subprocesses. Molenaar’s (2014) heteroscedastic latent trait model for dichotomously scored items was fit to the data. A separate set of simulation analyses considers also items generated with non-zero lower asymptotes. The simulation results illustrate that form of asymmetry has a meaningful relationship with the item response subprocesses. The relationship demonstrates how asymmetric models may provide a tool for learning more about the underlying response processes of test items.