Wim van der Linden

How to Make Adaptive Testing More Efficient?

Wim J. van der Linden


As a rule of thumb, adaptive testing requires some 50% of the test length to achieve the same efficiency of ability estimation as a linear test. In this presentation, I will use hierarchical IRT modeling to show that further increase of efficiency is quite possible. In one application, a hierarchical framework with an IRT model for the responses and an additional model for the response times on the test items is used to improve ability estimation and item selection in the adaptive test. In another application, a hierarchical IRT approach is used to sequence a test battery, allowing the use of information from an earlier subtest to adaptively pick the next subtest. Both applications are illustrated for an adaptive version of the Law School Admission Test (LSAT).