Abstract: In this presentation, I will draw on two recent studies that address challenges in Multi-Stage Testing
(MST). The first study of Ricarte, Curi and von Davier (in progress) investigates the issue of misrouting
when smart test takers make mistakes on easy items in the routing module. We investigated a Raschversion
of the logistic positive exponent model (Samejima, 2000) with a fixed value (larger than 1) of the
exponent parameter so that the model penalizes less dramatically than the traditional IRT models for
mistakes on easy items. A simulation study was conducted, in which the sample ability is drawn from a
mixture distribution: a part of the sample had the response patterns simulated according to the 3PL IRT
model, and the other part had the response patterns simulated according to the 4PL IRT model, where a
slip parameter leads to a lower probability of success for high-ability subjects. The second study
investigates the issue of small samples for an MST administration and continues the research of non-IRT
methods, such as regression tree, to routing and scoring in MST (Reshetnyak, von Davier, Lewis, & Yan, in
progress). In particular, in this study, a comprehensive comparison of the IRT and CART methods is
presented.

Keywords: IRT models, Multi-Stage Testing (MST), simulation study