Like most values, diversity does not come cheap. In an era of increasingly competitive college admissions, constrained resources, and the usual uncertainty about where students of all types will enroll, figuring out how and who to attract can be a challenge. Colleges have turned to predictive analytics—using past enrollment data to make predictions about future enrollment trends—to make tough decisions about who to actively recruit, admit, and support financially. In fact, a recent KPMG survey found that 41 percent of colleges are using their data to make predictions about future conditions and/or events.
Increasingly, colleges have proclaimed diversity as a value, and are committed to ensuring each incoming class is more diverse than the one before it. With underrepresented students disproportionately less likely to enroll in and finish college—and schools’ stated commitment to provide access to quality education for all—predictive analytics could enable colleges to focus their energies on recruiting the diverse students they say they want to attract. However, predictive data may be driving them to do just the opposite.
Predictive models attempt to create a profile of the “ideal student”—generally, a student that has successfully enrolled—and compare it against profiles of prospective students who have yet to be admitted. These models often include demographic data such as what high schools students attended as well as their ages, standardized test scores, race or ethnicities, and socioeconomic status. This information helps determine student profiles, which are assigned a score—1 is a low match to the ideal student, 10 is a high match.
The problem is that colleges have to use their past data to predict future success. And if they haven’t successfully recruited or supported underrepresented students in the past, the model of the “ideal” student will continue to exclude underrepresented students. The very factors that go into making predictions—like SAT scores and demographic information—have undoubtedly made gaining admission an uphill climb for students who are low-income, the first in their families to go to college, immigrants, or a person of color.
Colleges may ask themselves how they can justify using limited resources to recruit students whose odds of enrolling and succeeding are slim to none. For instance, a 2013 Ruffalo Noel Levitz study showed at the median, private colleges spent over $2,000 to recruit a new student and public colleges spent over $400. Schools may also question granting generous need-based aid packages to students who might not even show up for their first day of classes, instead of using those funds as merit aid to lure students who are likely to show up and would otherwise pay the full ride.
These are the tough choices admissions officers are faced with everyday. As a nation, we may not be able to live with the consequences of making the wrong choice. Colleges must ensure that all members of society—including the least privileged of them—have a chance to access a quality education. So how can predictive data be used to increase underrepresented students’ odds of admission rather than reduce them?
One approach would be for colleges to highlight the human side of data; at the end of the day, these are people we are talking about. Predictive data shouldn’t make enrollment processes less holistic, but more. In the overall admissions process, colleges are starting to move away from heavily weighing—or weighing at all—standardized test scores as measures of academic achievement. There is a recognition that standardized test scores tell little, if anything, about a student's academic abilities and motivation to succeed in college; students are far more than their scores. What will it take for colleges to similarly broaden their view of underrepresented students with low predictions, to recognize the limitations of predictive scores just as they have realized the limitations of standardized test scores?
If this were to happen, predictive data would only serve to encourage admissions officers to look beyond a student's predictive score and to uncover how each student might benefit from, succeed at, and contribute to an institution.
I believe colleges would be surprised at what they find. After all, we can’t predict everything.