A Method For Predicting Post Secondary Educational Outcomes

Identifying potential engineering students and understanding what affects their choice of college major is critical to engineering educational research. Insufficient numbers of students are majoring in Science, Technology, Engineering, or Mathematics (STEM) topics. Understanding the factors that affect students’ interest in studying STEM, capability of succeeding in STEM, and likelihood of persisting to achieve a STEM degree is of vital concern to educators. This study used an extensive national longitudinal dataset of over 12,000 students to develop a set of logistic regression models for predicting which students ultimately achieve a STEM degree vs. another educational outcome. The potential educational outcomes included no college degree, a less than four year college degree, a Non-STEM college degree, a STEM college degree, and a newly proposed category of STEM-Related college degree. Another model comparing the probability of STEM vs. all the other possible outcomes combined was also constructed. The resulting models demonstrated strong predictive accuracy in discriminating between a STEM degree and an alternative educational outcome. The predictive accuracy of the models was examined with Receiver Operating Characteristic (ROC) Curves. Several measures of student academic capability, prior academic performance, attitudes, experiences, and family influences were consistently found to be statistically significant predictors of STEM.