Professors have no control over their students’ age, ethnicity, or income level, yet many student-retention systems use these data points—sometimes described as the demography of destiny—as the foundation of their schools’ predictive algorithms. While the approach has shown success at many schools, it’s ill suited to institutions like Strayer University, a primarily online, open-access institution that caters to working adults. So Strayer developed its own retention strategy based on benchmarks over which it does have some control: student engagement.
“We found that our learning management system attributes—especially when grouped in ways that get at the work effort and patterns of students—are far more predictive than other data sources including income, age, or race,” said Joe Schaefer, Strayer’s chief technology and innovation officer. “If we can keep students working with positive academic habits and build on those, then positive outcomes are much likelier for students.”
Making the Engagement Shift
Strayer’s shift to an engagement-based strategy started three years ago when it partnered with analytics provider Civitas Learning to develop a customized tool based on the company’s Illume analytics platform. While the school had initially expected to use the same demographic data points as other higher education institutions, its open-access model meant that Strayer lacked some data sources, such as test scores, available to other schools. Plus, the makeup of its student body is very different.
“Every institution we work with tells us it’s a snowflake, and in the end you find out it’s right,” said Mark Milliron, co-founder and chief learning officer at Civitas Learning. “Every institution has different policy, practice, and people sets, and each has a different student population.”
While Civitas researchers did find some interesting trend lines in Strayer’s data using demographic variables, they struck predictive pay dirt when they examined derivative variables based on engagement metrics in the school’s LMS. “The metrics might be logins relative to the class average; they might be time spent on content; they might be the number of discussion board posts relative to the class average,” explained Schaefer. “The top predictive factors don’t have to be the same for everybody: For first-term students, maybe the predictive power of LMS logins is number two, but for third-term students in the seventh week of class maybe it’s ninth. The models are constantly updating.”