Analysis of Success in General Chemistry Based on Diagnostic Testing Using Logistic Regression

• Published in: Journal of chemical education Vol. 78; no. 8; p. 1117
• Main Authors: Legg, Jason C, Greenbowe, Thomas J, Legg, Margaret J
• Format: Journal Article
• Language: English
• Published: Easton Division of Chemical Education 01.08.2001; American Chemical Society
• Subjects: Achievement tests; Chemistry; Diagnostic systems; Diagnostic tests; Logistics; Organic chemistry; Placement tests; Regression analysis; Science education; Statistical analysis; Textbooks / Reference Books; First-Year Undergraduate / General; Chemical Education Research; Chemometrics
• ISSN: 0021-9584; 1938-1328
• DOI: 10.1021/ed078p1117

Several chemistry diagnostic and placement exams are used to help place chemistry students in an appropriate course or to determine strengths and weaknesses for specific topics in chemistry or math. The purpose of obtaining pre-course measurements is to increase students' academic success. Often these tests are used to predict the chance a student has in passing a course. This paper discusses the statistical methods of logistic regression applied to predicting the probability of passing a course, based on the scores on the California Chemistry Diagnostic Test at two different institutions with two different instructors over multiple years. This technique describes the relation of a test score (a continuous variable) to the probability of passing the class (a binary variable). Many papers in the Journal of Chemical Education have used a simple linear regression technique to correlate placement test scores with the proportion of students passing a course. The model assumptions are difficult to satisfy when using simple linear regression. Simple linear regression is useful when continuous predictor variables predict a continuous response, whereas logistic regression is useful when continuous predictor variables predict a binary response. Differences between simple linear regression and logistic regression and methods for evaluating linear regression model assumptions are discussed in detail. The fundamental concepts behind regression are described, with the caveats of using regression equations for predictions. By using logistic regression, instructors will be able to provide students with an estimate of their probability of passing the course.