Register Now for our Webinar: Maximize Enterprise Productivity with SharePoint and E-Signatures

Focus on KM in higher education:
Learning analytics efforts apply business intelligence to student retention.

This article appears in the issue May 2012, [Volume 21, Issue 5]
Page 1 of 2 next >>


   Bookmark and Share

Across American colleges and universities, only about two-thirds of freshman students stick around for sophomore year. New students often feel overwhelmed and unprepared for the academic and social challenges of college life.

Schools offering online degree programs have an even higher rate of attrition, perhaps because students can easily feel isolated. Determining which factors have the biggest impact on student success and retention has become a very hot topic in higher education. The term "learning analytics" seems to have a number of definitions, but it basically involves using predictive analytic tools to improve learning and education.

Disenrollment rates

Here is one of the KM challenges: How do you identify which variables are causing students to fail academically or feel disconnected from the university? To help with pattern recognition and to craft intervention strategies, several universities have recently turned to the use of predictive analytics. Many of those efforts are still in the pilot phase, but administrators are encouraged that their projects will bear fruit.

One of the leaders in the effort to apply business intelligence techniques to higher education is Phil Ice, VP of research and development at American Public University System (APUS) in Manassas, Va., an online university serving the military and those interested in public service programs, such as criminal justice, public safety and national security.

The disenrollment rates can be six to seven times higher for online schools than in face-to-face settings, according to Ice. "But online schools have access to a wealth of data to evaluate why students are disenrolling and make changes to reduce that rate," Ice says. To make sense of the data, APUS first had to create a common repository and federate the data. "It is analogous to the business intelligence efforts in the corporate world," he explains. "We are using the same underlying principles. We had to tie together data from seven different databases, including learning management systems and student information systems. That was one of the primary challenges."

Predictors

APUS has identified several predictors that students may drop out. One is no previous experience of higher education. "Our average age of student is 31," Ice says. "If they have no previous degree work, that is a red flag." The researchers also realized that it is not total time spent online but rather frequency of visits in the learning environment that tends to predict completion of courses. It is also apparent, Ice says, that students must embrace the online platform for social interactions in order to succeed. In assessing end-of-course surveys, APUS found that 20 percent of the students who say that they do not believe that online is a good platform for social interaction disenroll before or during the next course.

When students are identified as at risk of dropping out, the information is sent to deans, program directors and advisers, but because the system is relatively new, APUS doesn't have any data yet on whether increased contact and interventions with students has been effective.

Course signals at Purdue

A real pioneer in the use of predictive analytics for student retention is Indiana's Purdue University whose Signals software was first developed in 2005 to detect early warning signs and provide intervention to students who might not be performing to the best of their abilities. I recently interviewed the application's chief architect, John Campbell, associate VP for academic technologies at Purdue, for an article in Campus Technology magazine. He explained that to identify students at risk academically, Signals combines predictive modeling with data mining from the Blackboard Vista learning management system. Each student is assigned a "risk group" determined by a predictive student success algorithm. One of three stoplight ratings, which correspond to the risk group, can be released on students' Blackboard homepage. Signals suggests to students that they use available resources on campus—such as office hours and study materials—to increase their academic success.

Campbell also says that although he and his team were content with the progress they were making with the project on the Purdue campus, publicity about Signals soon made it a much larger endeavor. After NBC News did a story about the application, many other schools contacted Purdue, asking for help developing something similar for their campus. Because it did not have the resources to support all those other schools, Purdue chose to work with SunGard Higher Education to commercialize the application.

The application needed to be made generic to work with several student information systems, says Tom Wagner, SunGard Higher Education's product manager for retention and student success. Purdue continues to provide input on the product, and SunGard gets feedback from other schools adopting the software, Wagner adds.

Keeping up the PACE

Over the years, Rio Salado College, a two-year college in Arizona with a strong online presence, has tried many things, including student surveys, but they were all retrospective, according to Michael Cottam, associate dean of instruction. "We want data to use for interventions with current students," Cottam says. "We are looking at data from the learning management system to identify what makes a student successful. We can compare that to people who have completed courses successfully and give people feedback on whether they are on track to be successful or not."

Eventually, administrators chose to formalize the process of looking for predictors with the development of a homegrown system called PACE (Progress and Course Engagement). The predictions are based on student habits such as log-ins, site engagement and pace, says Shannon Corona, Rio Salado physical science faculty chair and lead for the predictive analytics pilot program.

Perhaps surprisingly, they found that early grades on assignments are not as predictive as other data elements. Log-in behavior, site engagement and pace through the course material in the first eight days are the strongest predictors of success, the researchers found.

Page 1 of 2 next >>

Search KMWorld

Connect