Teoria da resposta ao item : o uso do modelo de Samejima como proposta de correção para itens discursivos

In large-scale educational evaluation, there is a need for a data system whose objective is to provide instruments of learning management, aiming for the implementation or maintenance of educational policies, either public or private. This data system is necessary, since it promotes a continuous monitoring of the adopted strategies, aiming to detect its fragilities and improvements. Therefore, an evaluation system must obtain and organize periodical and comparable information about the di erent aspects of the educational system. In this sense, for large-scale educational evaluation, several countries make use of the Item Response Theory (IRT), which, in principle, has come to complement some limitations of the Classic Testing Theory. In Brazil, IRT was used for the rst time in 1995, to analyze data from the Sistema Nacional de Ensino Básico SAEB. Among other large-scale evaluations, it is also used in the Exame Nacional do Ensino Médio ENEM. However, even though the IRT is used in the ENEM knowledge tests of Mathematics, Human Sciences, Nature Sciences and Languages and Codes, grading of the ENEM essay is made under aspects of the Classic Testing Theory. The purpose of this work is to present, among non-dichotomous items models, the Samejima model, which permits the creation of a grading scale for the ENEM essay based on the IRT. 1: National Basic Education System 2: National High School Examination Key-Words: Large-scale evaluluation, IRT, Samejima, ENEM, essay.