Evaluation of ACT/SAT scores as predictors of student success in general chemistry

Prediction of student success in a single course such as general chemistry has traditionally been the purview of the department which owns the course such as the department of chemistry. In the analysis of data such as ACT or SAT scores, chemists have traditionally used the statistic techniques common to their research pursuits. For many chemists, this is typically been ordinary linear regression where letter grades that students receive are treated as the dependent variable while factors of interest were explored as potential predictive variables.;Evaluation of the underlying mathematical basis of linear regression reveals that its use in success predication violates one of more of the assumptions that must be made about the type and quality of data. The most commonly violated assumption is homoscedacity (or constant variance of the errors). Additionally, the ultimate goal of success prediction is to determine if a student will either pass or fail a course based on the predictor variable---that is---only 2 possible outcomes. With the increasing availability of personal computing, other means of data analysis are now commonly available via statistical software packages such as SPSS. One alternative technique that continues to grow in the interest of chemists' evaluation of student success is logistic regression. Logistic regression is a general linearized model in which the natural logarithm of the odds is linearly related to the predictor variable.;This study investigates the use of the ACT and SAT to predict student success in general chemistry at The University of Mississippi, CHEM 105, over a four year period. Additionally this study seeks to determine a typical student profile in order to determine if the widely accepted notions of the typical CHEM 105 student are reasonably accurate.