Facing federal and state pressure to raise retention and graduation rates, dozens of colleges and universities are developing analytics tools to help students make better decisions about everything from courses to social activities.
Purdue University has notified its 7,300 incoming freshman about a new web application that could help them better acclimate to life on campus. Administrators at the West Lafayette, Ind., school view the software as a critical tool for an institution whose graduation rate hovers at around 50 percent.
Called Forecast, the software is designed to anticipate the danger of students performing poorly by analyzing the time they spend in class and on campus, as well as how often they access coursework. “We’re looking at how student engagement is impacting the overall prediction of success on campus,” says Brent Drake, Purdue’s chief data officer, who’s leading the initiative.
Analytics as a salve
Purdue is one of dozens of institutions of higher-learning, from private colleges to large public universities, that are turning to data to improve their student retention and graduation rates. The national average six-year graduation rate for students attending four-year schools is about 54 percent, according to the National Student Clearinghouse Research Center. Poorly performing schools post percentages in the low 30s, and the average of higher-performing institutions is around 70 percent. At 51.5 percent, Purdue is in the middle of the pack.
“There’s a real issue in retention and graduation rates,” says Glenda Morgan, a Gartner analyst who tracks the use of analytics and other technologies in education. “Data is seen as one of the solutions to that, and that’s one of the big drivers for adopting learning analytics.”
Purdue sees a potential salve in Forecast, which prompts students to take corrective action before they fall behind. Parsing aggregate and anonymized data generated by the behavior of former Purdue students, the software can, among other things, generate bar charts that illustrate the dangers of waiting too long to register for classes and the benefits of getting involved in activities early.
For example, Purdue found that students who register for a class before it starts post an average GPA of 2.95, with significant drop-offs to 2.65 for those who register one to seven days after class starts and an average of 2.52 for those who wait three weeks to join a class. “Past the two-week mark, GPA falls off a cliff,” Drake says. “You’re behind on everything.”
Of course, common sense could lead to the same conclusions: Students who join classes late have to scramble to make up assignments, end up joining study groups late and just generally don’t have enough time to acclimate to their environments.
But Forecast also uncovers less-obvious insights, including the finding that those who take classes with friends or make friends in their classes tend to earn higher grades.
“What we are trying to do with the tool is to show students data and how it relates to their success on campus so they can make better-informed decisions,” Drake says.
Over the course of a semester, the software will parse the copious amounts of data students generate by registering for courses in the school’s financial system and accessing coursework and completing assignments in the learning management system. It also keeps a record of how often students log on to the campus Wi-Fi network or swipe their ID badges to enter dining halls. It will make inferences about success by analyzing whether students are spending time with their peers and faculty members.
Forecast will use that data to generate new modules and graphs for students. “We will push data back in graphic form to show students the general relationship” between their behaviors and their results, Drake says. When viewed in aggregate, the various pieces of data paint a picture of student engagement, which can be a harbinger of success, Drake says.
A self-inflicted wound
The country’s mediocre graduation rate is a self-inflicted wound. For decades, schools measured success by their enrollment numbers and rarely focused on making sure students graduated, says Morgan. Rather, the emphasis was on attrition and academic success was a matter of survival of the fittest. Schools were bent on weeding out the weak. Many university presidents would welcome new students with some variation of, “Look to your left, look to your right. One of you won’t be here next year.”
That attitude is a nonstarter in today’s higher education market. Dropouts count as lost revenue for schools, and the ex-students find themselves degreeless and in debt. High dropout rates are especially painful for institutions whose state funding is based partially on performance, which takes retention and graduation rates into account, according to Morgan.
Indiana is one of the 30 or so states that have performance-based funding models, according to the National Conference of State Legislatures. Therefore, the state funding Purdue receives is determined in part by its overall graduation rate, its on-time (four-year) graduation rate, the number of degrees it awards, the number of degrees it awards in high-demand majors and its transfer metrics. In the last two-year budget cycle, 6.5 percent of Purdue’s base allocation was based on performance, Drake says.
The upshot of all this is that if Forecast is successful at getting students more engaged in their schoolwork, it could bolster the school’s funding prospects.
Learning analytics has also become a priority for smaller schools, such as Marist College. A private liberal arts school in Poughkeepsie, N.Y., with 6,500 students, 4,800 of them undergraduates, Marist also feels pressure from the government to do a better job at keeping students on track for graduation, says CIO Bill Thirsk.
In 2013, Thirsk began piloting the Open Academic Analytics Initiative, an analytics application that aggregates student GPAs, SAT scores and demographic data, and correlates it with information about how often students submit assignments and engage with instructors online. Like Purdue’s Forecast, Marist’s application is designed to analyze the click impressions students create as they navigate the school’s various software systems.
The software takes stock of such details as how fast students click on assignments to review them, how fast they complete their homework and whether they chat with peers to complete team assignments. “We can see very early on which students are lagging,” Thirsk says. However, the data is anonymized until a faculty member decides that a problem warrants some attention. At that time, school officials have the option of staging an intervention for the student.
Thirsk says his algorithms can anticipate whether students are going to earn a minimum of a C or lower in the first two weeks of a class with an 85 percent certainty.
Marist’s program has been adopted by North Carolina State University, which has about 40,000 students. Lou Harrison, NC State’s director of educational technology services, said Marist’s model proved 80 percent accurate despite some false positives. “We didn’t tweak their model too much at all, and it turns out it was pretty predictive,” Harrison says.
Here are some other examples of schools that are finding success with academic analytics:
- Austin Peay State University in Clarksville, Tenn., in 2011 built an automated engine to inform students of their likely success in a given class based on their past performances, and on the performances of those who took the class in the past.
- Georgia State University used predictive models to boost its six-year graduation rate from 48 percent to 51 percent from 2012 through 2014.
- Baltimore’s Johns Hopkins University and several schools in the University System of Maryland are using similar programs.
Morgan, the Gartner analyst, says that 30 percent of higher education institutions worldwide will have adopted analytics strategies by 2018.
Collecting reams of data is something that public- and private-sector enterprises do on a daily basis to glean insights that, for example, help hospitals improve healthcare, enable municipalities to build smart cities and help companies develop highly targeted marketing campaigns.
But as useful as those initiatives may be, the efforts to collect the data that make them possible raise privacy concerns, and analytics programs in higher education are giving rise to similar concerns, according to Lee Tien, a senior staff attorney at the Electronic Frontier Foundation. Merely possessing data creates potential ethical challenges for an organization. One of the biggest issues is that, while the current use of the data may be legitimate, in the future it could be used for purposes other than what was originally intended.
“There is this inexorable drive to collect more data, and to view everything as a data collection opportunity,” Tien says. “Big data begets big data, so will we see mission creep? Do we see that because there is this belief that the data helps retention that they see everything as opportunity to get more data?”
Yet those who oversee the analytics initiatives at Purdue, Marist and NC State insist that they’re laser-focused on improving student outcomes.
At Purdue, Forecast constitutes the second phase of the school’s learning analytics journey. CIO Gerry McCartney says an earlier version known as Signals, which was launched in 2009, depended heavily on the input of faculty members, who set up the scoring model and warned students when they were at risk. Signals provided a template for Forecast, and for the initiatives at Marist and other schools.
With analytics efforts growing more sophisticated, McCartney says, “we know a lot more about students, and we can produce a much richer set of guides.”
Because it’s possible to learn so much about students, Purdue is exercising caution with Forecast. Participation is voluntary and students must log in to access the tool, says Drake.
Over the course of the year, Purdue will email students a heads up when it launches new software modules.
Depending on how the tool fares, Purdue may begin pushing alerts to students to warn them when they are in danger of falling off. It will also develop a mobile version of Forecast, which students will be able to download to their smartphones.
However promising such tools seem to be, the science of learning analytics remains imperfect. Academic success hinges on several factors, from student aptitude and interest in a particular course to socioeconomic and cultural situations. Higher education is still figuring out the right formula.
One thing Marist’s Thirsk says he’s sure of is that schools that don’t embrace some form of analytics will fall behind the curve. “Schools that aren’t doing this kind of work in some form are not providing great value,” Thirsk says. “If you don’t know what’s going on course by course, then how can you say you have a great degree program?”