Predicting Success in Programming I
Britton Wolfe, Eli Lowry · 2024
We investigate the degree to which student scores on a diagnostic quiz can predict success in a first programming course.The diagnostic questions do not require programming knowledge, nor are they programming questions, to facilitate early feedback and avoid measuring student confusion about programming syntax.Instead, the questions attempt to assess fundamental reasoning skills that we hypothesize are important for success as a programmer.A multiple linear regression finds a significant correlation between scores on some of the diagnostic questions and exam scores in the course, but there remains substantial variance in exam scores unexplained by the model, even after accounting for students' major and different instructors.Nonetheless, students with low scores on one particular diagnostic question have a much higher risk of earning a D, F, or W in the course (44%) compared to the other students in the course (13%), as computed on a validation data set.