Adaptive multi-level model for mathematical ability estimation - Experimentation on large-scale surveys

This research illustrates the construction of a computer-based multilevel adaptive test (MST) to estimate mathematics skills of students in Grade 10. The research shows that the test overcomes some of the limitations of linear and classical adaptive tests. It proves to be more informative than a linear test built on the same item bank. Consequently, it allows us to estimate with greater reliability the ability of students within a wide interval of the continuum, improving evaluation for all students and especially for those at the extremes of the continuum. The test succeeds either in providing low ability students with an appropriate number of items that they are actually able to answer or high ability students with items that are still challenging. In the construction phase of the test, it was possible to respect constraints of construct coverage that ensured the test was balanced with respect to it. The analysis of the results obtained from the field MST administration also provided useful insights for further research.