column | Postsecondary Learning

As the University of South Africa Considers Predictive Analytics, Ethical Hoops Emerge

By Manuela Ekowo (Columnist)     Sep 27, 2017

As the University of South Africa Considers Predictive Analytics, Ethical Hoops Emerge

At the University of South Africa (Unisa) in Pretoria, the typical student experience could take place anywhere. As an “open distance learning” institution, students attending Unisa receive a course textbook and a study guide to direct them through course materials. If learning takes place online, students may use a learning management system (LMS) to log into their courses, and even watch live video lectures.

Unisa claims to serve nearly one third of South African students alone. But enrollment isn’t the school's biggest challenge; rather, it’s figuring out how to best ensure students are engaged in their learning and reach their academic goals. The university has started using predictive analytics as a solution to that, but, like many colleges and universities throughout the world, is grappling with how to do so without discriminating against students based on assessments of their needs and potential, and making the correct inferences from quality data.

With students hailing from 130 countries around the globe (91 percent of students are from South Africa), part of the reason for Unisa’s astronomical student enrollment figure is due to the university’s distance education mode of instruction. Unisa is also a relatively affordable institution. Tuition and fees for the first year of a bachelor of arts degree in 2015 was R13,600 (a little over $1,000 U.S. dollars), according to Africa Check, a nonprofit that tracks information on African crime, health, education, and more.

Unisa’s teaching model has served many students well. Nelson Mandela received his bachelor of laws (LLB) from the college in 1989 while imprisoned. But, like most primarily distance education institutions, Unisa faces challenges that it must tackle to ensure its diverse student population graduates. One way to assess students’ progress and intervention needs is to use predictive analytics—decoding patterns in historical data to predict events that can impact a student’s academic journey and supporting them through in the best way possible.

In order for Unisa to use predictive analytics to improve these outcomes though, the school may first have to jump through some ethical and technical hoops. For example, how can institutions avoid mischaracterizing students with labels such as “at-risk”, and instead situate and serve them within the most appropriate context? And, how does an institution ensure that data are inputted and processed accurately, and especially if it has to be done manually?

These were some of the key issues raised by Unisa’s Sydney Louw Butler, instructional technologist in the academy for applied technology and innovation, and Dr. Angelo Fynn, manager of student success projects. (Interestingly, their challenges are almost identical to some presently faced by institutions in the U.S., especially those serving low-income and other disadvantaged students.)

The potential to misunderstand students’ abilities as well as their areas for improvement is of primary concern for Fynn.“I don’t share the common view that students are inherently at risk but rather that risk is contextually and historically located”, he said. Fynn went on to explain:

Contextual in that curricula and assessment practices assume certain normative conditions that not all of our students are able or desire to comply with (such as instruction in English in a predominantly non-English country like South Africa). And, historical because there are structural elements in our society that facilitate or inhibit access to quality education.

Fynn is referring to the enduring legacies of apartheid—a legal system that racially segregated South Africans between 1948 and 1994.

These realities are something Fynn hopes Unisa and other institutions can navigate with care. “There’s a balancing act that we need to play when developing analytics,” Fynn said. If the age-old markers are used to determine risk, like income, parent’s educational level, and high school GPA, Fynn worries institutions may not be exercising enough caution to ensure that nuances can be captured and understood. According to Fynn, “given the structural inequalities in our education system, we may just be measuring the extent of the disadvantage.”

In addition, institutions hoping to help students reach success despite these circumstances, may need to balance appropriately who exactly is responsible for student success. “Are we predicting the likelihood of a student's success, or are we predicting the likelihood that they could be successful with us (the institution)?” asked Fynn. The difference is important to Fynn because “where there is a poor fit between the institution and student, then the blame or categorization of risk should be shared and not placed solely on the high-risk student.”

The other hurdle is ensuring that Unisa has quality data in order to be able to intervene on students’ behalf. “With an organization at Unisa’s scale, there’s just no way to ensure high levels of data quality with analog, manual data collection methods,” professed Louw Butler. Luckily, Unisa is already making headway on this front. “This problem will largely be resolved as the university becomes fully digitized and automated, with instant validation and intelligent data management”, reported Louw Butler.

A related challenge is being able to verify that data collected tells you what you think it’s telling you. “Often in predictive analytics, we are reliant on data that were collected with other uses in mind and we therefore need to pay attention to the data ecology of the institution”, said Fynn. In other words, the origins of institutional data and whether meaning is lost or changed as data are carried from one context to another. Fynn believes institutions may be wise to “consider the implications of repurposing data and our assumptions of what the variables represent.”

Unisa graduates close to 40,000 students a year (the total student body of some large public universities in the U.S.), but they would like to see these numbers increase, and be able to retain more students semester-to-semester. And although Unisa is committed to offering a flexible schedule and most students attend part-time, problems with retention may be attributable to Unisa's commitment to opening access to high-quality, low-cost education. Recall that Mandela completed his degree from a prison cell. While an extraordinary example of a Unisa student, it also shows the unique circumstances of Unisa students that could inhibit retention and graduation without the right support.

Manuela Ekowo (@ekowohighered) is a policy analyst with the Education Policy program at New America.

column | Postsecondary Learning

As the University of South Africa Considers Predictive Analytics, Ethical Hoops Emerge

By Manuela Ekowo (Columnist)     Sep 27, 2017

As the University of South Africa Considers Predictive Analytics, Ethical Hoops Emerge

At the University of South Africa (Unisa) in Pretoria, the typical student experience could take place anywhere. As an “open distance learning” institution, students attending Unisa receive a course textbook and a study guide to direct them through course materials. If learning takes place online, students may use a learning management system (LMS) to log into their courses, and even watch live video lectures.

Unisa claims to serve nearly one third of South African students alone. But enrollment isn’t the school's biggest challenge; rather, it’s figuring out how to best ensure students are engaged in their learning and reach their academic goals. The university has started using predictive analytics as a solution to that, but, like many colleges and universities throughout the world, is grappling with how to do so without discriminating against students based on assessments of their needs and potential, and making the correct inferences from quality data.

With students hailing from 130 countries around the globe (91 percent of students are from South Africa), part of the reason for Unisa’s astronomical student enrollment figure is due to the university’s distance education mode of instruction. Unisa is also a relatively affordable institution. Tuition and fees for the first year of a bachelor of arts degree in 2015 was R13,600 (a little over $1,000 U.S. dollars), according to Africa Check, a nonprofit that tracks information on African crime, health, education, and more.

Unisa’s teaching model has served many students well. Nelson Mandela received his bachelor of laws (LLB) from the college in 1989 while imprisoned. But, like most primarily distance education institutions, Unisa faces challenges that it must tackle to ensure its diverse student population graduates. One way to assess students’ progress and intervention needs is to use predictive analytics—decoding patterns in historical data to predict events that can impact a student’s academic journey and supporting them through in the best way possible.

In order for Unisa to use predictive analytics to improve these outcomes though, the school may first have to jump through some ethical and technical hoops. For example, how can institutions avoid mischaracterizing students with labels such as “at-risk”, and instead situate and serve them within the most appropriate context? And, how does an institution ensure that data are inputted and processed accurately, and especially if it has to be done manually?

These were some of the key issues raised by Unisa’s Sydney Louw Butler, instructional technologist in the academy for applied technology and innovation, and Dr. Angelo Fynn, manager of student success projects. (Interestingly, their challenges are almost identical to some presently faced by institutions in the U.S., especially those serving low-income and other disadvantaged students.)

The potential to misunderstand students’ abilities as well as their areas for improvement is of primary concern for Fynn.“I don’t share the common view that students are inherently at risk but rather that risk is contextually and historically located”, he said. Fynn went on to explain:

Contextual in that curricula and assessment practices assume certain normative conditions that not all of our students are able or desire to comply with (such as instruction in English in a predominantly non-English country like South Africa). And, historical because there are structural elements in our society that facilitate or inhibit access to quality education.

Fynn is referring to the enduring legacies of apartheid—a legal system that racially segregated South Africans between 1948 and 1994.

These realities are something Fynn hopes Unisa and other institutions can navigate with care. “There’s a balancing act that we need to play when developing analytics,” Fynn said. If the age-old markers are used to determine risk, like income, parent’s educational level, and high school GPA, Fynn worries institutions may not be exercising enough caution to ensure that nuances can be captured and understood. According to Fynn, “given the structural inequalities in our education system, we may just be measuring the extent of the disadvantage.”

In addition, institutions hoping to help students reach success despite these circumstances, may need to balance appropriately who exactly is responsible for student success. “Are we predicting the likelihood of a student's success, or are we predicting the likelihood that they could be successful with us (the institution)?” asked Fynn. The difference is important to Fynn because “where there is a poor fit between the institution and student, then the blame or categorization of risk should be shared and not placed solely on the high-risk student.”

The other hurdle is ensuring that Unisa has quality data in order to be able to intervene on students’ behalf. “With an organization at Unisa’s scale, there’s just no way to ensure high levels of data quality with analog, manual data collection methods,” professed Louw Butler. Luckily, Unisa is already making headway on this front. “This problem will largely be resolved as the university becomes fully digitized and automated, with instant validation and intelligent data management”, reported Louw Butler.

A related challenge is being able to verify that data collected tells you what you think it’s telling you. “Often in predictive analytics, we are reliant on data that were collected with other uses in mind and we therefore need to pay attention to the data ecology of the institution”, said Fynn. In other words, the origins of institutional data and whether meaning is lost or changed as data are carried from one context to another. Fynn believes institutions may be wise to “consider the implications of repurposing data and our assumptions of what the variables represent.”

Unisa graduates close to 40,000 students a year (the total student body of some large public universities in the U.S.), but they would like to see these numbers increase, and be able to retain more students semester-to-semester. And although Unisa is committed to offering a flexible schedule and most students attend part-time, problems with retention may be attributable to Unisa's commitment to opening access to high-quality, low-cost education. Recall that Mandela completed his degree from a prison cell. While an extraordinary example of a Unisa student, it also shows the unique circumstances of Unisa students that could inhibit retention and graduation without the right support.

Manuela Ekowo (@ekowohighered) is a policy analyst with the Education Policy program at New America.

From our Guide

further reading

GET THE LATEST HIGHER ED NEWS
Be the first to know, with our weekly newsletter.

GET THE LATEST HIGHER ED NEWS
Be the first to know, with our weekly newsletter.