Two Methodologies: How Well Can Universities Predict Retention

Student retention has been a long standing focus in higher education research with one of the earliest work dating back to 1937. Many researchers have proposed factors that affect a student's decision to depart from the university without successfully completing a degree. It is important to not only research different attributes and characteristics that affect student departure but it is also important to study different statistical methodologies. With the advancement in technology, new methodologies such as the Classification and Regression Tree (CART) have proven to yield significant results in a variety of research fields. As these new statistical methodologies emerge, it is always worthwhile to compare the modern approaches with the longstanding classical statistical approaches. The present study utilized historical archived data in order to compare the performance of the Logistic Regression (LR) methodology with the CART methodology in predicting first-year retention for new freshmen at the University of Mississippi. It was found that the logistic regression method was more accurate than the CART methodology, with the overall accuracy of 83.3% and 82.6% respectively. However, the CART methodology was more specific than the logistic methodology, meaning that the CART model correctly predicted more students to not be retained. The logistic regression model failed to identify at-risk students. Note that 98% of the time the CART model and the logistic regression model yielded the same classification result. Among those 2% that the classification decision differed, the CART model was more accurate than the logistic model to predict non-retained students. Thus using the prediction outcomes of the two methodologies in tandem of each other leads to more accurate results overall.

Name thesis:degree_name

M.S. in Mathematics

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Mathematics

Year dc:date.available

2013

Author dc:creator

• Gregory, Tiffany Lynette

Contributors dc:contributor

• Xin Dang
• Martial Longla
• Hailin Sang

• Classification And Regression Tree
• Logistic Regression
• Retention
• Statistics and Probability

Repository record dc:identifier

https://egrove.olemiss.edu/etd/671

OAI identifier oai:identifier

oai:egrove.olemiss.edu:etd-1670