Análise comparativa do Exame Nacional do Ensino Médio (Enem) via Teoria Clássica dos Testes e Teoria de Resposta ao Item

• Ano de defesa: 2019
• Autor(a) principal: Sousa, Leandro Araujo de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Avaliação em larga escala; Psicometria clássica; Modelo de traço latente
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/48275

Large-scale educational assessments, including the National Exam of Upper Secondary Education (Enem), are important measures of school performance in Brazil today. For a long time the Classical Test Theory (CTT) has been predominantly used. In this context, Enem since its creation in 1998 until 2008 has used this analysis paradigm. However, since the 1990s, part of the CTT procedures have been replaced by Item Response Theory (IRT). Thus, in 2009 this exam started to use the IRT. Thus, we question whether there are differences in measurements from CTT and IRT, therefore, whether there is comparability between item parameters and participants' scores. Thus, this research aimed to evaluate the results of the 2017 Enem from CTT and IRT. It is a research with quantitative approach and exploratory objective. A sample of 10.000 participants selected by simple random sampling participated in the research. The analysis of the exams dimensionality was performed with the Parallel Analysis and Full Information Factor Analysis test. Subsequently, the parameters of the exam items and participants were estimated from CTT and IRT (1, 2 and 3 parameters) and compared using Pearson's correlation coefficient (r) and simple linear regression. Software R statistical packages were used for all analyzes. The results from the Parallel Analysis indicated evidence of a dominant dimension in each exam. Full Information Factor Analysis indicated that most items had high factor loadings in a single factor. But some items had low factor loadings (<0.30). After adjusting a one-dimensional model and estimating item and participant parameters by CTT and IRT, a strong correlation was found between the difficulty and item discrimination parameters by CTT and 1 and 2-parameter IRT models. Correlation values ​​decrease with the 3-parameter model. The same occurs when comparing the participants' scores. In both cases the adjusted simple regression model was significant. It is concluded that there is high comparability between the parameters of the items and the participants' scores by CTT and IRT models. However, comparability becomes weaker with the 3-parameter IRT model. Thus, the need for the IRT model for the purposes of this evaluation is questioned.