Identifying Factors Contributing to University Dropout with Sparse Logistic Regression

Prediction and prevention of dropout are main challenges of institutional research (IR) in universities nowadays. If we can identify factors significantly contributing to dropout by statistical inference, then it is helpful for developing strategies for dropout prevention. The main problem in identifying such significant factors is that available data contain much more candidate factors than the students. Most of conventional statistical methods do not work well unless the number of data is much more than the number of parameters to be estimated, which means that we cannot apply such methods in identifying factors contributing to dropout. To circumvent the situation, we propose to use sparse logistic regression for identifying factors contributing to dropout based on data with a large number of candidate factors. Sparse logistic regression is a method that can analyze such data reliably by pruning factors that do not contribute to the analysis. To demonstrate how sparse logistic regression identifies factors contributing to dropout, we applied the method to actual university credit data of 410 students for 302 courses and identified 18 courses that significantly contribute to dropout. The contributions of the identified courses to dropout are interpreted.