In designing any educational intervention one ofen needs to determine what factors are related to success and failure in a course, identify students at risk, evaluate the impact of any new program on student pe~ormance. In this paper we will discuss the use of discriminant analysis as a technique for addressing all of these issues.Discriminant analysis is a statistical technique designed to investigate the differences between two or more groups of people with respect to several underlying variables. This technique is more appropriate than commonly used educational measures (correlations, regression weights, etc.) because the variable being predicted is categorical. Moreover, this approach results in a unit of analysis, predicted category membership, that is more useful in evaluating instructional interventions.We have used discriminant analysis to predict student pe~ormance in an introductory electromagnetism course at Georgia Tech. In this course there is a high failure rate (greater than 30% make a grade of D or F ) which results in a great cost to the institute and to students as success in the course is a prerequisite for all engineering majors. A technique that could identify the factors that are predictive of course pelformunce and identify students who are at risk would be of great benefit to the design, implementation, and evaluation of any educational intervention.We will present a case study with over 1600 engineering majors where discriminant analysis was used successfully to determine the predictors of course performance. We looked at over 15 possible predictor variables (SAT, GPA, etc.) and determined which variables would be most useful in identihing students at risk. Using information available from student's records we were able to successfully predict 50% of the students who eventually failed the class.We will also discuss how discriminant analysis can be used as an evaluation technique. We show that this approach provides results that are both more intelpretable and statistically sound than traditional measures.

No abstract

No abstract

No abstract