Scholar advocates regulating the use of enrollment management algorithms in higher education.
The average yield rate among first-time freshmen enrolling in U.S. institutions of higher learning decreased by over 10 percentage points between 2007 and 2017. These decreases prompted public universities to compete with highly selective institutions to enroll students.
Simultaneously, state and local funding for higher education continually decreased. As a result, many public universities altered their admissions practices by implementing new recruitment strategies and tuition variation to entice students to enroll.
One admissions innovation implemented by universities in recent years entails the use of enrollment management algorithms. The use of these algorithms may reduce the amount of scholarship aid granted to students who need it most, argues data science expert Alex Engler in a recent Brookings Institution report.
Schools have increasingly used algorithms during admissions cycles to boost yield and meet tuition-based revenue goals. In fact, the majority of universities participating in a 2015 survey reported using artificial intelligence (AI) to optimize enrollment management. These schools hire independent vendors to develop predictive models to help them determine the likelihood that a student will enroll if offered admission.
According to Engler, the use of these algorithms should be scrutinized because they threaten to compound pre-existing financial crises in the higher education sector. Universities without large endowments face enormous pressure to yield enough students to cover institutional expenses without awarding too much scholarship aid, often while weighing other goals, such as enticing diverse and well-rounded students. Recent data reveal that the use of enrollment algorithms may cause schools to prioritize yield and scholarship optimization over their other goals.
Scholarship optimization has traditionally been conducted manually, allowing financial aid officers to combine the use of predictive algorithms with human selection mechanisms to account for factors such as student diversity. Growing trends, however, signal that more schools are moving toward pure algorithmic optimization, which eliminates human selection from the process.
Sole reliance on algorithmic enrollment management could result in an overemphasizing metrics such as test scores and engagement in pre-college interactions with a given school. As a result, algorithmic biases may discriminate against students of color and those who come from low-income families by decreasing the amount of aid awarded to them.
The use of algorithms in enrollment management is not inherently problematic, notes Engler. Such algorithms do provide meaningful benefits to the universities that use them.
For instance, the use of predictive modeling in the admissions process decreases the workload of financial aid offices, allows schools to prepare sufficient student housing, and assists administrators in ensuring course availability. In addition, some vendors advertise that their algorithms will help institutions significantly increase student enrollment and drive up revenue through scholarship optimization, which could translate into millions of additional dollars in tuition revenue for a school in a given year.
The benefits of pure algorithmic optimization are numerous, observes Engler. A recent study found that reliance on algorithmic scholarship optimization significantly increased out-of-state applicant matriculation at a large public university. Another study simulated that this approach would significantly increase yield. Furthermore, at least one vendor has implemented this kind of algorithm, allowing its clients to drive up enrollment while minimizing the amount of scholarship aid awarded to students. With such clear benefits to universities, the market will likely see a proliferation of AI in enrollment management in coming years, Engler suggests.
Engler argues that policymakers should demand transparency from both the vendors providing enrollment management algorithms and the universities employing them. The federal government could accomplish this, Engler suggests, by establishing an independent commission to study universities’ use of AI in the admissions and enrollment processes. He also recommends that the U.S. Congress encourage data sharing among both universities and vendors as a means of informing future policymaking.
Engler also argues that the U.S. Department of Education should use its regulatory authority in alignment with new Office of Management and Budget guidance to federal agencies on how and when to regulate the use of AI in the private sector. The Education Department should seek to issue corresponding guidance to institutions on best practices and pursue enforcement actions against those using enrollment management systems irresponsibly, urges Engler.
Ultimately, Engler recommends that policymakers reflect on the reasons why so many institutions of higher education have increasingly turned to enrollment management algorithms in the first place. He suggests that government needs to improve college access and affordability by providing better funding for higher education and by mandating that schools lower tuition prices. Then universities will feel less pressure to engage AI to optimize scholarship offerings as a means of yield protection to the detriment of students with the greatest financial need.
The European Commission has already recognized the high-risk nature of enrollment management algorithms and subsequently issued proposed AI regulations constricting their use. Advocates and scholars who agree with Engler urge the United States to follow suit.
