A logistic model fitted to UFLA's 2006-2 admission
exam with candidates to Agronomy

Maria de Lourdes Lima BRAGION[1]

Júlio Sílvio de Sousa BUENO FILHO1

§    ABSTRACT: A two parameter Item Response Theory (IRT) model was fitted to analyse the admission exam to a Federal University in Lavras, Minas Gerais, Brazil (UFLA). Exam was held to admission in the second semester of 2006. The parameters in the logistic model are: degree of discrimination (a) and dificulty  level (b), along with the individual hability (theta). Model were fitted within Bayesian Inference framework using Metropolis-Hastings algorithm to get samples from the posterior distribution. Individual hability presented strong correlation with final grades. The most difficult questions, on average, came from Mathematics followed in decreasing order by Physics, Chemistry, Biology, History and Portuguese. Geography and Phylosophy presented a low level of difficulty. Regarded to discrimination, Biology has shown the better results with excelent discrimination. Good discrimination were reached, in decresing order, by Chemistry, Physics and Mathematics exams. The other subjects did not bring any relevant information to rank the candidates. Two parameter logistic IRT models has been shown a powerfull tool to understand and discuss the quality of UFLA admission exams to Agronomy.

§    KEYWORDS: Bayesian inference; item response theory.



[1] Departamento de Ciências Exatas, Universidade Federal de Lavras – UFLA, Caixa Postal 3037, CEP: 37200‑000, Lavras, MG, Brasil. E-mail: lourdinha.bragion@gmail.com / jssbueno@ufla.br