Why Student Data Sharing Could Hurt Online Education

Student data sharing can widen exclusion
Shared information can weaken stronger firms
Policy must protect second chances

There are 37.6 million working-age adults in the United States with some college but no credential. That fact underscores why sharing student data has emerged as the dominant solution. Moving records faster, making credits more legible, and providing colleges with an early warning about potential completers could re-engage many of these learners. The incentive is plain. The policy story is over, simplified. In online education, sharing student data is about both access and competition. According to the Brookings Institution, while the promise of increased data sharing is that it removes barriers and helps smaller schools compete, simply having more information does not always improve the market and can sometimes make it less welcoming. Sometimes it does. But it can also transfer value from colleges that have studied students for years to rivals that have not, and it can cause the hardest-to-serve learners to be deemed risky everywhere at once.

Sharing student data is also a competition rule

This debate is relevant now because the online world is no longer a side lane. It is a main lane. In Fall 2024, NC SARA logged 1,661,689 students enrolled out of state via online courses at SARA-participating colleges. As of November 1, 2024, the Common App had processed 4,017,250 applications from 904,860 first-year applicants and 863 returning members. Similar efforts are underway in Europe. The European Commission is designing digital credentials and crafting new rules enabling easier data exchange across institutions and borders. The point is not that these projects are unwarranted. The point is that data sharing is becoming a core architecture decision in higher education. The critical question is no longer whether data can flow but who benefits from that flow, who gains or loses power, and which students become simple to identify.

The case for sharing student data in public is compelling. Presumably, a clearer transcript reduces credit loss. Presumably, a common barometer of learning accelerates adult students' reentry. A common format prevents fraud and saves students from reissuing the same forms. These are powerful incentives. But they do not settle the more consequential issue. Higher education is not a pure search market where greater visibility is automatically advantageous. It is also a selection market where colleges not only find a match between applicants and programs. They also estimate which students will persist, how much support they will need, the associated costs, and whether they will graduate. In online environments where advising is scalable, ads are targeted, and attracting new students can be expensive, these predictions are exceedingly influential. Consequently, sharing student data must be considered a market rule. It alters what students carry from college to college and which students seem desirable.

Sharing student data can operate as a consumption subsidy

That is a dimension overlooked by numerous reform proposals. Large universities, major platforms, and well-established online schools do not merely possess data. They dedicate years to developing methodologies to transform imperfect student histories into actionable intelligence. They interlink admissions information with clickstream activity, funding histories, coaching notes, transfer trends, and stop-out risk estimates. They develop teams and products to optimize messaging strategies for persistence. Some of those innovations enhance instruction. Some boost retention. Some improve student recruitment. All are expensive. According to the Brookings Institution, when policies enable widespread sharing of student data without clear limits, it can give certain institutions an advantage by providing them with information they did not earn themselves, effectively acting as a subsidy that shifts value from the college that originally developed and supported the student to others that benefit from that data.

This danger is most pronounced in the adult stop-out market. Over 1 million re-enrolled in 2023–2024, while the universe of working-age adults with some college but no degree reached 37.6 million. That represents a considerable opportunity for online recruiting. According to a report from AERA Open, when institutions share student data that includes predictive algorithms meant to identify who is most likely to graduate, there is a risk that these tools may be racially biased against Black and Hispanic students. This could allow new supporters to focus their efforts on students already predicted to succeed, potentially overlooking and leaving more costly or challenging cases to the original institution, which raises concerns about fairness and equity in educational opportunities.

If institutions are unable to earn an adequate return from the taxonomies they develop, they will have less incentive to produce this data in the future. They may continue to satisfy common standards while withholding their most refined prediction engines and support plans. They may edit down rich student histories into anemic datasets that travel well but prove less effective in providing true academic guidance. Alternatively, they might prioritize intensive advisement and reallocate resources to brand development; since a brand is less easily duplicated than data insights, it represents a more durable competitive advantage. This preference would be detrimental to students, as it would lead to more transparent data sharing but less tailored support. Moreover, with the rising ubiquity of generative AI93% of North American college administrators and faculty expect increased AI utilization in the next two years; the importance of robust data assets becomes even more critical. This intensifies the need for thoughtful design in data sharing mechanisms.

Sharing student data can encourage widespread exclusion

This is the second challenge, and it is even more profound. In economic terms, the Hirshleifer effect warns us that readily available perfect information can eliminate the utility of car insurance. In the higher education context, if all decision-makers perceive identical indicators of risk, dropout propensities, or other early warning signals, then data sharing removes the relative advantage of insuring against these risks by making that information uniform across institutions. No school has to deny a student a position based on a costly-to-support profile; they need only to average across every other school and reach the same conclusion. This mentality allows institutions to inform one another about targeted students without any formal agreement and can foster uniform exclusions without explicit collusion.

This adverse outcome is not purely hypothetical. Educational institutions already use predictive models for admissions, institutional planning, and resource allocation, yet considerable bias persists. A 2024 study published in AERA Open revealed that the average campus.

Success models tend to yield weaker results for Black and Hispanic students than White and Asian students, often misdiagnosing their likelihood to succeed. The false negative rate for Black students was 19%, and for Hispanic students it was 21%, compared to 12% for White students and 6% for Asian students. A 2025 analysis by the same authors at the Brookings Institution warned that omitted variables, such as campus climate and familial support, can further distort these predictions. According to a 2024 report from the Open University Institute of Educational Technology, efforts are underway to develop AI-based strategies for detecting bias in educational content, highlighting concerns that predictive models might introduce bias related to age, disability status, and gender when used in student engagement monitoring. The report suggests that when such models are widely shared across colleges, their errors may become more broadly accepted and influential perceptions. Consequently, students perceived as risky in one environment are likely to appear so everywhere else they venture.

This results in a new version of sorting. A student deemed risky at one college can automatically become risky elsewhere. This is particularly concerning for groups that policies aim to support, such as older learners, transfer students, part-time students, those taking breaks, multilingual individuals, and anyone with an uneven record from an irregular life. These are the groups online education purports to serve uniquely well. Yet they are also the demographic more susceptible to blunders based on outdated data. Predictive models may interpret interruption, weak performance in early courses, or sporadic study logs as indications of poor learning potential. They may miss underlying factors like supportive household networks, lack of reliable internet access, or prior institutional mismatches, which may predict far better long-term achievement in flexible modalities. Data sharing can illuminate these students' attributes while obscuring their true futures.

Data sharing demands constraints and discretionary judgment

The correction to this problem is not a call to stop data sharing. It is a call to control it through utilization rules. Data confirming course completions or digital badges should be inherently portable. Risk estimates or retention markers created by third-party companies should not be. Public institutions will require clear boundaries separating standardized student performance records from subjective risk profiles. These regulations should involve more nuanced distinctions between concrete milestones and predictive inferences. The former should be easily transferable; the latter should be restricted from. Inspired by European systems, rules can require that each piece of evidence be verifiable and reliable before it is permitted to cross institutional boundaries.

Sharing access must also carry responsibilities. Policymakers should prohibit institutions from using shared data solely to lure optimistic students with little intention of remaining long-term. Transparency in admissions, retention, credit recognition, and deferral rates, as well as public accountability measures, should be implemented. High-stakes decisions should be scrutinized for group error rates. College personnel ought to document whether particular data drove outreach, rejection, or redirection, and whether someone actually examined the recommendation. Any institution that gains an advantage by sharing data must contribute to resources for the students they most want to retain and ultimately graduate.

We should keep the original focus on the scale of the existing challenge. A population of 37.6 million adults in America with some college but no degree does not suffer from a deficit of information; rather, it suffers from a paucity of willing and capable institutions to make strategic risks based on imperfect data. Sharing data more effectively can boost efficiency, streamline transfers, and facilitate credit transfers. However, if those projects are perceived solely as toolkits for enhancing competition, they will do more than optimize information distribution; they will enable selective bias. An ideal policy framework should promote data sharing that fosters greater access, encourages robust investment in student success, and facilitates thoughtful judgment and second chances.

The views expressed in this article are those of the author(s) and do not necessarily reflect the official position of The Economy or its affiliates.

References

Banerjee, A.N. and Seccia, G. (2002) On the “Hirshleifer effect” of unscheduled monetary policy announcements. Discussion Papers in Economics and Econometrics, No. 0213. Southampton: University of Southampton.
Common App (2024) Deadline updates, 2024–2025: First-year application trends through November 1. Arlington, VA: Common App.
Cosconati, M., Xin, Y. and Wu, F. (2026) ‘Information equalisation and competition in selection markets: Evidence from auto insurance’, VoxEU.
Ellucian (2024) Ellucian’s AI Survey of Higher Education Professionals Reveals Surge in AI Adoption Despite Concerns Around Privacy and Bias. Reston, VA: Ellucian, 22 October.
Gándara, D., Anahideh, H., Ison, M.P. and Picchiarini, L. (2024) ‘Inside the Black Box: Detecting and Mitigating Algorithmic Bias Across Racialized Groups in College Student-Success Prediction’, AERA Open, 10. doi:10.1177/23328584241258741.
Note: this is the published journal version; it is stronger and more accurate to cite than the earlier arXiv preprint.
National Council for State Authorization Reciprocity Agreements (NC-SARA) (2025) Annual Data Report 2024: Fall 2024 Exclusively Distance Education Enrollment and 2024 Out-of-State Learning Placements. Boulder, CO: NC-SARA.
Note: your draft item used the executive-summary wording; this is the fuller report title used by NC-SARA.
National Student Clearinghouse (2025) ‘Record Number of Students Re-Enroll in College’, NSC Blog, 4 June.
OECD (2005) Catastrophic Risks and Insurance. Policy Issues in Insurance, No. 8. Paris: OECD Publishing.
Note: “Policy Issues in Insurance” is the series title; the actual volume title is Catastrophic Risks and Insurance.
Perry, J.T., Pett, T.L. and Ring, J.K. (2012) ‘Comparison of the information-sharing benefit of the internet for family and non-family firms’, International Journal of Information Technology and Management, 11(3), pp. 186–200.
Spitalniak, L. (2025) ‘Working-age adults with some college but no credential reaches 37.6M, report finds’, Higher Ed Dive, 4 June.
user200947 (2019) ‘Insurance and Hirshleifer effect’, Economics Stack Exchange, 25 February.
Vincent-Lancrin, S. and González-Sancho, C. (2023) ‘Interoperability: unifying and maximising data reuse within digital education ecosystems’, in OECD Digital Education Outlook 2023: Towards an Effective Digital Education Ecosystem. Paris: OECD Publishing.