Skip Crooker (University of Central Missouri) "Predicting freshmen credit hours earned with statistical learning methods"

Friday, April 25, 2014 - 12:00 pm to 1:20 pm
Event Type: 

Location:  368A Heady Hall

Description:  Skip Crooker (University of Central Missouri) "Predicting freshmen credit hours earned with statistical learning methods"

Abstract: Beginning in academic year 2012, the Missouri Department of Higher Education instituted a performance based funding mechanism for all public 4-year institutions in the State of Missouri. The stated goal of the model is to distribute 100% of all new money earmarked for higher education according to each institution's respective performance. With 5 performance metrics, 20% of an institution's new appropriations is conditional upon success on a particular metric. Among the 5 metrics identified as performance metrics is the proportion of the institutions first-time full-time (freshmen) cohort successively earning 24 credit hours in the academic year. To achieve success in this metric, the institution must not experience a decrease in the 3-year average proportion of students earning 24 credit hours. As this particular metric is not necessarily an attribute of the freshmen cohort that has been historically forecasted, we have a lack of understanding of important characteristics of the freshmen experience that influence success in this metric. Meanwhile, in academic year 2014, the value of achieving this single metric of success eclipsed $400,000 in annual funding. For these reasons, we are interested in quickly learning the signals that may explain success or failure in this metric. Our objective in this paper is to explore statistical learning approaches to explaining factors impacting an individual student's ability to earn 24 credit hours in his or her rst year of study. The predominant approaches considered are decision-tree, bagging, random forests and boosting methods. We compare the statistical learning algorithms against more conventional econometric estimators in a cross-validation study design.