Resumo

ROSAS, M et al. Construction of linear regression models with quantitative and qualitative variables in order to inference about student accomplishment. Rev. Fac. Agron. [online]. 2006, vol.23, n.2, pp.197-214. ISSN 0378-7818.

The aim of this research was to construct multiple regression models in presence of qualitative and quantitative variables that would allow predicting the student's accomplishment and thus suggesting the student an alternative to fulfill success in his/her studies at the University Institute of Technology of Yaracuy. There were 28 explicative variables, and the index of the academic proficiency was used as a dependent variable once the student graduates. Qualitative variables were included, that mentioned the necessity of the use of some dummy variables, and the study of the models' behavior was done under these conditions. For each of four specialties, was obtained the complete model as well as models reduced as a cause of the methods of all possible regressions and also done step by step. T tests were done in the complete model and selected variables were compared with those included in the selected models by both methods. Regression coefficients of the selected variables were studied in order to detect their stability. A model was selected for each of the four specialties: Agriculture, Conservation of Renewable Natural Resources (C.R.N.R), Food and Livestock. These models explained the 56.41%; 89.66%, 69.33% and 73.10% respectively of the total variability of achievement and chosen variables differed according to their specialties.

Palavras-chave : Dummy variables; all possible regressions; stepwise; academic yield; qualitative and quantitative variables; model comparisons; R²; Press; Cp of Mallows.