Student Success Beyond the First Year
Often, student success efforts are focused primarily on retaining first-year students, but fail to continue supporting students throughout their college careers. At the University of North Carolina-Greensboro, the institution’s leadership wanted to take a broader approach to student success by developing a predictive model that would include upperclassmen.
In this web seminar, the vice chancellor for enrollment management and the data manager from UNC-Greensboro discussed the collaborative approach the institution took in developing a full retention predictive model, which now provides the probability of retention for freshmen, sophomores and juniors as well as incoming freshmen, and some best practices for using predictive analytics to support student success at any institution.
Vice Chancellor for Enrollment Management
University of North Carolina-Greensboro
University of North Carolina-Greensboro
Senior Data Analyst
Rapid Insight Inc.
Dr. Bryan Terry: Enrollment management is about paying attention to details. When new students come in, we know some are sure to succeed. They’re asking the right questions, they want to be Spartans, they want to graduate as Spartans. But we know we have others who are going to struggle. So one of the things we wanted to do—because we don’t have the budget to buy a $250,000 software package that will solve all of our problems, or to hire a counselor for each student—was to take our limited resources and pair them against those students who need the academic advisement the most.
That’s what Rapid Insight did for us. It was a very low-cost option that allowed us to be a data-driven institution. So far it’s been very successful.
Jon MacMillan: Rapid Insight is a predictive analytics software company founded in 2002. We are in a few different markets, but our main focus has been higher education, serving over 200 institutions nationwide, from small to large, private and public, community colleges, law schools, four-year institutions and so on. Our whole goal is to make data analytics predictive modeling very easy and intuitive for users of any ability.
One way we do that is we create software that fits all customers’ needs. So we have powerful data analytics software for data scientists, but if you lack coding skills and statistical knowledge, the products are still intuitive, allowing users of all abilities the option to perform these different analyses. We also have an open services support model—if you have a question, we’re always here to answer.
Then on top of that we have QuickStart templates. We partnered with UNC-Greensboro to develop a template for actually creating this data to build these predictive models.
We provide software that allows you to integrate data from any source. That may be pulling directly from your student information system or other databases, and then cleaning that up, performing statistical functions, creating new metrics based on the information you have, and then automatically identifying the key indicators of whatever outcome you’re interested in.
Jeffrey Collis: We have two different models that were our first models to be established. The first one “likelihood to enroll,” and then we also wanted “likelihood to retain.” With incoming classes of 2,500 to 3,000 first-time, first-degree undergrads, we needed to, as soon as possible, identify what their needs are and target that outreach. We need to figure out what their strengths are and what their weaknesses are and how we can intervene to get them moved along the cycle.
With our model for likelihood to enroll, we discovered that the key piece is the distance from the student’s primary home to campus. We were having significant issues with people who were outside of a certain mileage from our campus. Typically, if they were about 150 to 200 miles away, they just didn’t retain as well. We are certain that it’s likely associated with having a large first-gen population, having a large Pell Grant population, having families that are having difficulty trying to discover what to do.
So we discovered that out-of-state recruitment probably wasn’t optimal for us, and we moved into only our most lucrative markets. We were able to increase our enrollment 6.2 percent from fall 2013 to fall 2014. That was a wonderful gain. How did we do this? We used Analytics to target our messages. We couldn’t have done that if we didn’t have VeeraTM and Analytics—if we weren’t working with Rapid Insight to help us identify who was going to stay, who wasn’t going to stay, who was going to enroll, who was not going to enroll.
So what about retention? With using Analytics and Veera, we’re able to bin students on their likelihood to retain. Whatever the numbers might be, we’re able to concentrate our efforts. Predictive modeling gives us a chart. We’re able to look and see by decile who’s leaving, who’s not leaving. This was based off of looking at a three-year trend of students and all the various pieces that led to them being in this decile. It might be gender, it might be race and ethnicity, or it might be Pell and high school GPA and SAT.
We are not trying to pigeonhole somebody and say, “This is your destiny.” We are trying to identify where we can improve our enrollment management efforts. How can we intervene now to make sure this student doesn’t act like someone who looked like them in the past? We want to intervene to change the plan for them—to make it better so that they are successful.
We were able to see a retention result improvement of 4.5 percent. We’ll take that gain. That is wonderful. Having that many more students stay in keeps us all employed, for one thing, but it also gets the students to remain on track. They are that much closer to meeting their goal, their family’s goals, being productive members of society with college degrees, contributing to the economy.
There are lots of other variant pieces we’re looking at trying, more and more data that will help us predict so that we can intervene and help students be successful.
Jon MacMillan: The QuickStart templates are an automated data preparation process. They are a visual workflow of pulling data directly from the Banner student information system, identifying all of those tables that you need to access, as well as all the information, all of the columns, the variables/metrics that we use to predict student retention. The templates pull all of that together to prepare a nice, clean data set.
If you’re interested in seeing more about this process, you can always check out our website, www.rapidinsightinc.com.
To watch this web seminar in its entirety, visit www.universitybusiness.com/ws030917