Skip To Content
Rapid Insight: Higher Education case study

Predictive Modeling for Enrollment Management – Dickinson College

A data-driven campus culture

Data analysis and predictive modeling had been a part of the culture at Dickinson College long before Dr. Mike Johnson came on as Director of Institutional Research. Dickinson found that predictive modeling allowed them to plan better by anticipating the future and has embraced the process, especially in enrollment. As the institution found more and more uses for predictive modeling to guide data-driven decisions, they realized that bringing predictive modeling in-house would be the most efficient and cost-effective way to do so.

Bringing predictive modeling in-house

Initially, Dr. Johnson wasn’t sure exactly which tools would be  best to  bring predictive modeling in-house, so he started looking at software packages and asking other people how they were doing it. He wanted a solution that could be used in near “real time” and was flexible enough to allow him to have his hand in the modeling, but didn’t want to spend his time programming. Dr. Johnson also knew he needed to be able to defend the resulting models and predictions to campus leaders and decision makers. This ruled out “black box” software programs which produce models that can be difficult to explain. He compared Rapid Insight’s Predict to a number of sophisticated modeling packages and decided on it for its ease of use,  transparency, and because he could produce results quickly.

Shortly after purchasing the software, he started working on his first model. Whenever he had a question, he’d give the Rapid Insight Support Team a call. Working with the analytic team from Rapid Insight proved to be a good way to jumpstart his modeling efforts.  As Dr. Johnson describes, “That kind of follow through on support saved more than hours, it saved days of sitting there and trying to figure something out. If you struggle, you can talk to someone, get through it, and move on. I had an indication of good customer support before I purchased but didn’t anticipate what I’d actually experience.”

Enrollment cycle modeling
Dr. Johnson focused his initial predictive modeling efforts on the end of the admissions cycle; specifically, he was interested in enrollment modeling for the accepted applicant pool. He built individual models for each of three subpopulations within the applicant pool (early decision, early action, and regular admits) after finding that each subpopulation behaved quite differently.

The models are used to make yield, quality and financial aid projections for the entire accepted applicant pool, or for specific subgroups.  As each mailing date deadline approaches, his team uses the models to walk through simulations based on various scenarios to check things like diversity in the incoming class, projected quality of students, and projected gender mix. This allows for an interactive discussion between admissions, institutional research, and financial aid to make collective decisions about what they’d like the class to look like.  Additionally it enables them to fine tune the prospective class before sending out mailings.

Outcomes

After several years of predictive modeling, Dr. Johnson has been pleased with the results. The models have given them good projections of what their incoming class will look like. This accuracy is important because missing a class size by just a couple of percentage points can have a significant impact on financial aid, housing, and even the number of sections in a first year seminar. As Dr. Johnson observes, “You can never be too accurate.”

Next project

Dickinson has also built a separate model to predict end-of-first-semester GPA for each student. They use this during the enrollment cycle as a litmus test to ensure that the incoming class has the potential to succeed – and again when enrolled students finish their first semester to identify underperformance.   Dr. Johnson’s next project is to implement a new proactive retention strategy using this model. His team will be comparing predicted GPA with actual first semester GPA to find students who have underperformed within their cohort but are still in good academic standing. He’s already found that the bottom 10% of underperforming students in terms of actual vs. predicted GPA are 1.7 to 2.3 times less likely to be retained. By targeting these students with resources and programs designed to improve their second semester performance, they hope to ultimately affect the school’s overall retention rate. Dr. Johnson is excited about the process and thinks it can make an impact. “This is a new area of predictive modeling for us. We’ve been growing into it for two years, but now we’re at the point where we can really act on it.”

About Dickinson College

Dickinson College, founded in 1773, is a highly selective, private residential liberal-arts college known for its innovative curriculum. Its mission is to offer students a useful education in the arts and sciences that will prepare them for lives as engaged citizens and leaders. The 180-acre campus of Dickinson College is located in the heart of historic Carlisle, Pa. The college offers 42 majors with an emphasis on international studies, has more than 40 study-abroad programs in 24 countries on six continents and offers 13 modern languages.