A methodological refinement for studying the STEM grade-point penalty

We present a study that explores the grade-point average (GPA) penalties that students face when taking introductory STEM courses. Previous work has found that there is a large and significant grade point penalty for women and minorities in most STEM classes (that is, these students perform worse in these classes than their overall GPA would suggest). We recreated this work using a new, large data set (63,012 students over 10 years) of student performance, and found that the initial results held when using the original approach. We argue that there are methodological shortcomings to the original approach, however, as there is no attempt to control for individual student program difficulty (STEM majors and non-STEM majors share some classes, but have very different overall suites of courses that determine their overall GPA). As the female/male and racial ratios vary across majors it is therefore likely that a division by gender is not comparing equivalent sample populations. By controlling for student test scores or major most of the penalty is removed. The initial findings of large GPA penalties in STEM courses appears to be an example of "Simpson's Paradox".