As the debate continues over gathering data to advance teaching and learning in a way that lets students have a say over how their information is used, a University of Michigan study shows that giving them the opportunity to opt out may skew the data, biasing it against the people institutions most want to help.

The study by researchers at the U-M School of Information shows that students identifying as Black were significantly less likely to respond in the first place, and when doing so often self-reported lower levels of trust with how institutions would handle their data. Female students, overall, were concerned about personal data collection but were more comfortable having instructors use their data but less so with institutions.

“If we engage students simply by giving them a choice to share data or not, then we should expect that our learning interventions will be biased,” said Warren Li, doctoral student at the School of Information and lead author. “The right to choose whether your data is shared or not sounds good from a personal liberty perspective, but we must understand the impact of that choice.”

The researchers started with a sample of 4,000 students—272 of whom interacted with an email prompt and 119 completed a follow-up survey. The email was a one-question preference prompt that asked if they hypothetically would agree to have their data used in learning analytics systems. Women were significantly more inclined to respond to it than men, and white students far more than Black students. They were then directed to the survey.

Contrary to other recent research, the study found less trust in institutions to handle the data properly.

“The opt-in or opt-out model needs to be replaced with education, consultation and deep consideration,” said Kaiwen Sun, a doctoral student at the School of Information. “Importantly, we don’t make a call for ignoring student agency, but instead see this as a demonstration of the impact of non-thoughtful implementations of student choice in data sharing.

“With this work we have set a baseline which will help researchers, educators, technologists and policymakers to think more deliberately about how to engage with students around the issues of data use in education.”

Data gathered in classrooms has been used to help enrolled students through various interventions, including early warning systems and targeted support when they are struggling. The data also are used to inform predictive models of learners to address trends that can determine performance and engagement in the overall.

At pioneering institutions like U-M, such programs developed early on as ways to prevent the melt of STEM majors, those who struggle in an early course and end up deciding the field is not for them—a particular concern with women and underrepresented groups, the researchers said.

As these technologies have grown rapidly and been adopted widely across higher education, so have questions about student privacy and agency over their own data. Opt-out has been one solution proposed for this concern.

“The difference in opt-out preference rates in our study among different student groups show that there’s a risk that student autonomy in the use of learning analytics technologies can lead to biased models and systems. However, I don’t think that means we shouldn’t give students choice, said co-author Florian Schaub, assistant professor of information.

“Just giving students an opt-out is not enough, the models need to be adjusted carefully. In addition, there’s also a need to set clear and transparent institutional policy on how and for what AI is used in education, ideally informed by multistakeholder processes that involve students from different groups and backgrounds in the conversation and decision making.”

Christopher Brooks, assistant professor of information, added that in written comments many students expressed understanding that the overall goal of teaching and learning data is bigger than the individual.

“Several students who were willing to provide their data noted specifically how they did so with the hope that it would help empower technologies to help others, and this may be a powerful way to communicate the opportunity of learning analytics and data consent to students in the future,” he said.

“The key opportunity we face with this study is how to really understand and empower student agency in a way which will provide positive meaningful change in our educational technology landscape.”

More information:

• Study: Disparities in Students’ Propensity to Consent to Learning Analytics