No Walk in the Park: Predictive Analytics in Higher Education

A number of industries like healthcare, retail, and banking, use their historic data to predict future events to meet business goals or improve customer satisfaction. Predictive analytics, as it’s called, is also a growing practice in higher education. Like businesses, colleges and universities have operational goals they hope to meet and customers they wish to satisfy. In this case, the “customers” are students, faculty, and other institutional stakeholders.

For colleges and universities, this typically means using scarce institutional resources as strategically as possible to move as many students through to graduation. Predictive analytics allow institutions to make decisions in a streamlined and more data-informed way about how students will enter, experience, and exit their institutions.

Despite its potential to make college decision-making processes more data-informed, predictive analytics shouldn’t be a standalone tool. That’s because predictive tools can help institutions discriminate against certain students, make institutional practices less transparent, and make vulnerable individuals’ data privacy and security.

In a new paper, The Promise and Peril of Predictive Analytics in Higher Education: A Landscape Analysis, we outline how predictive analytics are used in higher education to provide personalized student advising through early-alert and major recommender systems, personalize learning through adaptive technologies, and manage enrollment decisions like which students to recruit and offer generous financial aid awards to.

We researched predictive analytics practice in higher education, the ethical concerns involved in using data to make predictions, and the implications for underrepresented students in particular by reviewing existing literature, interviewing a number college administrators, experts, and vendors who supply predictive tools and support to colleges and universities, and visiting Georgia State University. Georgia State is praised for using predictive analytics to help eliminate the achievement gap for its first-generation, Pell-eligible, black, and Latino students.

The Promise and Peril of Predictive Analytics in Higher Education features a number of examples from institutions about their predictive tools and successes to date. Consider for example, Temple University’s homegrown early-alert tool that analyzes the institutions’ data to identify students on the verge of dropping out of school and allows Temple to provide robust advising to ensure students remain in college. We learned from Peter Jones, Senior Vice Provost for Undergraduate Studies, professor of Criminal Justice, and mastermind behind Temple’s early-alert tool that the actual rates of dropping out for students identified as at-risk are generally a third to one half the predicted rates. In other words, students are dropping out at a lower rate than predicted. This may be because students are identified early and quickly given extra support.

Despite Temple’s and many other colleges’ successes, using student and institutional data to make predictions about what may happen in the future is no walk in the park. Early-alert tools like the one used at Temple and at many other institutions, can be discriminatory because they have to rely on past data, which often includes demographic data like a student’s race, financial aid status, and gender to make predictions about which students are not likely to succeed.

If an institution’s past data show that students of color and low-income students disproportionately struggle at the institution, predictive tools could disproportionately flag these students as at-risk, even when they may not be. This could lead to profiling students in ways that have demonstrably negative effects that can last throughout their academic careers.

The potential to label and stigmatize students isn’t the only issue with haphazardly using predictive analytics. As predictive tools make their way onto college campuses, students, faculty, or staff may not be told why the institution is using the tools; how the tools were built (with vendors often leading the way); what data was used to build them; and the limitations of the tool. For example, does it always accurately predict who will fail? With a powerful tool like predictive analytics, transparency should be heightened, not reduced.

Lastly, big data (a ton of data stored in one place) in higher education makes ensuring data privacy and security harder, not easier. As colleges collect and analyze data on how students engage with the institution over time, it may not always be clear whether students should be informed about the amount of their data being collected and mined for new insights, even if the intent is to help students succeed. And, as the rate at which student and institutional data is generated and stored accelerates, how to keep all of this data secure becomes increasingly challenging, yet all the more important.