PREDICTING STUDENT PERFORMANCE IN ENTRY LEVEL ELECTRICAL ENGINEERING TECHNOLOGY AND MATHEMATICS COURSES USING PRECOLLEGE DATA

The major problems of this study were to identify the relationships between precollege variables and performance in entry level courses in technical curricula through the formation and testing of equations which predict final course grades, and to compare two algorithms for selecting subsets of regression variables. The study examined the problems through efforts to update, improve, extend, and validate methods used at Purdue University Calumet for predicting grades in entry level collegiate courses while describing the improved procedures in a way which would assist officials at other institutions with similar interests.;Multiple regression analysis followed by cross validation was the principal technique. Stepwise analysis and optimal subset selection were investigated as methods for selecting the best subset of the independent variables.;The variables found to be the best predictors of performance in entry level mathematics courses are: mathematics SAT scores, high school grade average, the product of high school mathematics semesters and high school average grades in mathematics, high school science grade average, high school mathematics semesters, and age. The variables which were found to be the best predictors of performance in EET 102 are: high school science grade average, age, the number of semesters of high school science, and performance in prerequisite and corequisite mathematics courses.;It was found that the accuracy of predictors increased with increasing course rigor. Predictors were least successful for courses whose students were primarily majors in technical curricula.;The population consisted of students who enrolled in certain entry level courses in the departments of Mathematical Sciences and Electrical Engineering Technology at Purdue University Calumet between June, 1979 and June, 1982. Precollege data formed the independent variables. The dependent variable was the final grade in the courses studied.;The optimal and stepwise regression methods led to essentially identical results with negligible computational cost differences.