Um modelo inteligente para seleção de itens em testes adaptativos computadorizados

Abstract

Computerized Adaptive Tests (CAT) are a type of assessment tests applied through computers which main feature is the adequacy of the test questions to the performance of each examinee. The two main elements of a CAT are: (i) the item pool, which is the set of available questions for testing; (ii) the selection model, which pick out the questions, named items, applied to the examinees. The item selection model is the core of CAT, and its main task is to identify examinees knowledge level as the items are applied and to adapt the test, selecting the most proper items to produce an accurate measure. This thesis proposes a model for item selection based on goals for the test precision using the estimation of the proficiency standard error. For that, an specific control of the goals for each step of the test is developed. Using simulated tests, the results are compared to two traditional item selection models, evaluating the performance of the proposed model in terms of measure accuracy and the level of exposure of the items. Finally, a specific analysis is performed on the accomplishment of goals over the tests and the possible influence on the final result, in addition to considerations on the behavior of the model in relation to the characteristics of the item pool.