Optimizing Item Exposure Control Algorithms for Polytomous Computerized Adaptive Tests With Restricted Item Banks

In the face of modern-day developments in translating assessment instruments into computer-based adaptive measures, more and more concerns regarding test security and test validity have been raised. Further, as non-cognitive measures are being transformed into functioning CATs, item exposure control has become an important component of managing item banks. With the inception of evaluating exposure control mechanisms conditionally, many of the non-high stakes assessments are now seeking solutions to issues associated with constrained item banks and limited item availability. This simulation study compared three item exposure control mechanisms (Sympson-Hetter conditional, progressive restrictive maximum information, and total rate simplified exposure) against two unmanaged CAT administrations (simple maximum information and random item selection) for two restrictive item bank conditions. The competing methods were evaluated in their ability to provide conditional exposure control along with adequate bias, root mean square errors, and deviations of asymptotic valid standard errors from observed conditional standard deviations. Benefits and shortcomings are discussed.