A new algorithm allows organizations that care about diversity to do so while sacrificing very little talent and performance, said Kadeem Noray ’15, an assistant professor at Harvard Business School in a lecture Sept. 24. 

Praxis, the political economy club at Hillsdale College, hosted Noray’s presentation of his working paper on selecting diverse talent. Noray said he first became interested in this topic after the Supreme Court’s decision in Students for Fair Admissions v. President and Fellows of Harvard College striking down the use of affirmative action in college admissions.

“Even though affirmative action had been struck down, organizations were still interested in something that they thought of as diversity, and there were lots of different dimensions that, say, universities were interested in: whether or not you have different countries represented, different states represented at the university,” Noray said. “Despite like a lot of the attacks on, say, demographic diversification — like a lot of the attacks on DEI — there are a lot of private organizations that are still interested in something that they call diversity.”

For Noray, this turned what initially seems to be a social problem into a mathematical one, which he called the Diverse Talent Selection problem. This problem centers on how to find the groups that achieve the most diversity at any given performance level.

“You know what they can do in these two dimensions,” Noray explained. “They’re always going to prefer to select better groups on one characteristic without reducing the other characteristic.”

The issue is that the sheer number of possible groups in any reasonable scenario — such as the international scholarship program he worked with in the paper — makes finding the frontier of these best possible groups inefficient, Noray said.

“Your effect on diversity into a group is going to depend on the characteristics of every single other person in that group,” Noray said. “You would actually need to evaluate each potential cohort, and it turns out that that’s just a very difficult thing to do.”

To avoid this issue, Noray decided to use a so-called “greedy” algorithm to estimate the best possible groups. After selecting several possible “production functions” to weight diversity and performance against each other, a computer adds students one by one to get example groups for each possible weighting.

“You begin with an empty cohort, and you sequentially add applicants who increase this production function the most at each step,” Noray explained. “Then you do the whole thing again, but you adjust the weight that you’re putting on diversity versus performance.”

If the organization wants better choices, they simply need to invest more time trying more possible weightings. By checking smaller cases — and making some mathematical arguments — Noray was able to show that the estimates from his procedure were close to optimal.

“You can actually brute force these problems if you’re willing to pay enough money for certain solvers. It takes a while,” Noray said. “I decided to invest in this just so I knew exactly how far away from the truth this was, and it turned out to be shockingly close — better than the error bound that I’m able to prove.”

When applied to the international scholarship program he worked with, he was able to vastly increase both diversity and group performance compared to prior application cycles.

“They could have been roughly 15% more diverse without sacrificing any performance, or they could have been just under 60% you know, higher performing without sacrificing any diversity,” Noray said. “In cycle three, I intervened with the program and I allowed them to use the estimation technology we had designed.”

Another advantage of his technology, Noray noted, is that his algorithm doesn’t impose a particular tradeoff between performance and diversity, allowing organizations to evaluate their best options based on the example groups from various levels of the tradeoff.

“We actually are agnostic about how you weight diversity versus performance,” Noray explained. “There’s 20 cohorts that we gave them access to as part of their process.”

Senior economics major Luke Hill, who is considering graduate school, found Noray’s mathematical approach to model what he hopes to do in the future.

“There are definitely complicated components to the math and the computer science for the average person. But the thing that Dr. Noray did really well was take a somewhat complex problem and translating it in an economic way,” Hill said. “Here’s a very practical application of the skills I’m learning at school to a very specific problem.”

Senior Brian Shia, who is president of Praxis, thought the presentation provided a valuable perspective on how diversity can be used without needing to engage fully in radical diversity, equity, and inclusion policies.

“DEI is something that’s happening in a lot of institutions, whether we like it or not. Although we’ve seen a lot of pushback, these entities are still engaging in it,” Shia said. “I think it’s very important to think about how we can make that less bad, rather than just saying, ‘It’s bad.’ What Dr. Noray showed us was that it’s possible to come up with solutions that help us optimize based on both.”