Creating Software to Enhance Admissions Diversity

Clemson U.

Juan Gilbert
April 29, 2013

Admissions deans have long described their work as a blend of art and science. Juan E. Gilbert has designed a tool to enhance the latter. Call it the diversity algorithm.

The story began 10 years ago, when the U.S. Supreme Court ruled on the landmark admissions cases at the University of Michigan. After the decisions came down, Mr. Gilbert turned on CNN. Two commentators with opposing views of affirmative action were claiming victory. Both had it wrong, he thought.

"They were saying race, gender, and national origin was the issue," says Mr. Gilbert, chairman of Clemson University's human-centered computing division. "But the issue has to do with capacity—there are more qualified applicants than available slots, so you're going to turn away someone who's qualified."

THE INNOVATOR: Juan E. Gilbert, Clemson U.

THE BIG IDEA: Admissions offices can build more-diverse student populations by using sophisticated software.

That someone might be white or black, a legacy or a first-generation student, a Spanish major or an engineering major. In a realm of scarce seats, the system is never going to be entirely fair to everyone. Still, Mr. Gilbert, 44, believes the selection process could become more transparent—and consistent.

How might colleges enroll diverse classes without giving preferential treatment to any students? How might admissions officers minimize subjectivity, and, in turn, guard against charges of unfairness? The answers, he argues, lie in a more sophisticated means of comparing applicants.

Not long after the Michigan decisions, Mr. Gilbert wrote data-mining software called Applications Quest, which automates the nuanced evaluation of applications known as "holistic review," a fixture at selective colleges. The program allows users to assign equal weight to various attributes, such as an applicant's race, gender, geographic location, and intended major.

The software was designed to remove the variability in outcomes (if asked to repeat its process, an admissions committee wouldn't necessarily choose all the same applicants again). "This program would give you a holistic review that's 100-percent reproducible, with no bias," Mr. Gilbert says. And the why behind a particular acceptance, he says, would be measurable.

Applications Quest compares each applicant with every other applicant in the pool. This is done by measuring the similarities—and differences—among all applications on a 100-point percentage scale (two identical applications would be 100 percent the same, 0 percent different). These quantitative measurements produce clusters of similar applications. "These clusters represent holistic, diverse applicant pools and can facilitate holistic review," Mr. Gilbert wrote in a 2008 article published in the Journal of College Admission. "By selecting applications from each cluster, holistic diversity can be optimized."

Mr. Gilbert's definition of "holistic diversity" goes beyond race and ethnicity. In a forthcoming journal article he co-wrote, he describes holistic diversity as "multifaceted variation among applicants, where the goal is to increase minority representation across a number of attributes, where 'minority' refers to the values within an attribute." Men, prospective physics majors, low-income students, and first-generation applicants all might be underrepresented in a given applicant pool.

The article, now under peer review, summarizes the experimental use of Applications Quest at an unnamed major research university. The institution, identified as "Experiment University," ran a batch of freshman applications through the software. After the applications were grouped into clusters, the program recommended the application that was most unique within each cluster. "The application still exemplifies the characteristics of its particular clusters," the authors explain, "but what makes the application different is the variation of all the application's attributes based on the holistic comparison of all other applications."

In the end, the program recommended a class that was more diverse, broadly speaking, than the admissions committee had selected, with similar academic credentials. The committee took about five weeks to reach its decisions; the software delivered results in 10 minutes. By using the program, the paper concludes, colleges "can increase holistic diversity without disenfranchising any specific group."

Susan Allen has seen the program firsthand. She is director of operations in the enrollment-services division at Auburn University, where Mr. Gilbert previously taught. He helped her experiment with the software. One year, the program recommended 276 academically qualified applicants, about two-thirds of whom her staff had already decided to admit; the other third was to be denied.

When Ms. Allen took a closer look at the 90 or so students the computer had picked but Auburn had not, she said, "Wow, we missed those." That group included many minority students, legacies, out-of-state applicants, and applicants interested in underenrolled programs.

"We had probably eliminated them based on academics, but they were not bad students," she says.

After considering those students' other qualities, Ms. Allen says, she and her colleagues decided to admit about a quarter of them. Auburn isn't using the software now, however. Ms. Allen says she wouldn't feel comfortable running the complex program without Mr. Gilbert around to guide her.

Elsewhere, some deans who have read about the software are skeptical. After all, many colleges already have sophisticated databases. "It's difficult to see if, operationally, this will be a revolutionary change," says Robert Springall, dean of admissions at Bucknell University. And not everyone buys the idea that colleges could—or should—squeeze subjectivity out of evaluations.

About 20 colleges have invited Mr. Gilbert to demonstrate his software. At each campus, he says, admissions officers were impressed. (A patent is pending on the program.)

So far, he says, only Clemson's school of nursing uses it to evaluate applicants.

Some college officials may be awaiting the outcome of the latest Supreme Court cases on the issue of race in admissions. A decision is expected this spring or summer in a legal challenge to a race-conscious admissions policy at the University of Texas at Austin. The court recently also agreed to weigh the constitutionality of a voter-passed ban in Michigan on the use of racial or ethnic preferences in admissions at public colleges.

Mr. Gilbert argues that his program would help insulate colleges from legal challenges: "I keep telling them, the institution that does this broadly will be seen as an innovator, and will attract a very diverse group to their school."