“Higher education today is where banking was in the 1970s,” said Dr. Paul von Hippel, who prior to joining the LBJ School at UT Austin worked as a bank data scientist. “Until the 1970s there was limited use of credit scores, and banks often made lending decisions on the basis of personal relationships. That could work for a small bank, but it limited the number of loans that could be made, loan decisions could be inconsistent or arbitrary, and some banks were discriminating on the basis of race and neighborhood.
The Need for New Technologies and Analytics
Changes between 1968 and 1994 made those practices untenable.”. Mergers created bigger banks, credit cards became widespread, and fair lending laws held banks to new standards of fairness. “Banks needed new technologies to make millions of credit decisions quickly, using objective criteria that could be defended against charges of discrimination. That’s how credit scores came to be so important,” said von Hippel. “Predictive analytics became fundamental to banking.”
Today he sees higher education at a similar crossroads, with an operating system that’s been in place for decades, no longer completely sufficient in answering questions and providing individual solutions. “Universities have made some good use of data,” he said, “especially in admissions. What’s happening in technology is changing both what’s possible and what’s necessary. Just as banks became more accountable for their decisions, so are universities who now must explain to stakeholders why more students aren’t graduating in four years, and what improvements can be expected. It’s hard to give clear answers, and it’s hard to know how much improvement is possible unless universities take a more analytic approach.”
He cites the challenges universities face with data silos, where much information is gathered, but not shared into a cohesive big picture as a student moves from admissions to the registrar, to financial aid, to advising, to the classroom.
Being Effective at Scale
As universities are forced to ask themselves which students will have more trouble graduating, and why, von Hippel says they need to be able to identify interventions that individual students need, rather than subjecting a whole group to an intervention or support service that many may not need. He likens this to the work that banks do to protect customers from fraud. “There’s a simple way to protect customers from fraud,” he said, “and that’s to scrutinize every single transaction over, say, five dollars! But that’s impossibly expensive and it inconveniences a lot of customers who aren’t having a problem. An effective fraud control policy begins with analytics that identify the small number of transactions that are actually at risk, and then intervenes in only those cases.”
Where Universities are Today in Analytics
Which brings us back to higher education in 2013. “Using predictive analytics,” said von Hippel, “universities can identify the students who are more likely to have trouble and provide what they need, when they need it, or ideally, before they need it, without subjecting a whole subset of students who don’t need help to the same intervention.”
Descriptive, Predictive and Prescriptive Analytics
Universities have a long history of data analysis, although not necessarily in predictive analytics, von Hippel explains. “Take descriptive analytics – universities tend to be strong at this.” He cites the example ‘What was the average ACT score of a specific incoming class?’ But predictive analytics is more challenging, it asks questions like ‘What will be the four-year graduation rate of the incoming class, and can we assign each student a probability of graduating? Can we update those probabilities as students move through the curriculum, passing key milestones and obstacles?’ That’s predictive. Then there’s prescriptive analytics which uses data to recommend and evaluate interventions that raise the graduation rate in a cost effective way, focusing only on those students who need the help.”
Von Hippel advocates for a cycle of continuous improvement. “The smartest modern banks don’t go with a single plan of action,” he said, “They run multiple strategies side by side, and regularly correct course based on data and results. Universities need to do this as well.” He points to the strength of predictive modeling. “Once you have a good predictive model in place, you can identify students at risk, and the model continues to improve as you bring in additional data based on the findings from informed interventions. It just gets better.”
Creating a Continuous Cycle of Improvement
He says analytics can change the culture of a university over time. “Universities can develop a couple of different strategies and pit them against each other, evaluate which is more effective, and swap out less effective practices so that strategies evolve to align with goals. Few universities are there today – but progress is being made. It’s an approach that requires continually improving both predictions and policies – those two things have to go hand in hand.”
Dr. Paul von Hippel
A professor in UT Austin's LBJ School of Public Affairs, Dr. von Hippel is an expert on research design, and on statistical methods for missing data. Before his academic career he was a data scientist who developed fraud-detection scores for banks including JP Morgan Chase and the Bank of America. His research interests include educational inequality and the relationship between schooling, health, and obesity.