The research team, led by Asaf Mazar of the Wharton School at the University of Pennsylvania and including Angela Duckworth, Geoff Tomaino, Neil Heffernan, Ziv Carmon and colleagues, set out to examine a feature of modern education that most people experience but few have studied systematically. As an ever-growing share of learning migrates to screens, students’ outcomes increasingly depend not only on curriculum quality and motivation but on the technical performance of the platforms they use. While the digital divide has traditionally been framed as a question of access—whether people have internet connections at all—the new study suggests that the quality and speed of those connections may matter just as much for learning.

The researchers define microdelays as brief loading lags that occur when a page, exercise or video takes a moment longer than expected to appear. These delays are far too short to be perceived as a technical failure, yet long enough to break the flow of a learning session. To investigate their consequences, the team conducted two longitudinal quasi-experimental field studies. The first followed 26,707 primary and secondary school students in the United States who used an online mathematics platform. The second tracked 55,184 aspiring coders, primarily in Latin America, working through an online programming course. In both settings, the researchers measured how naturally occurring variation in load times related to learning behavior and outcomes over time.

The results were strikingly consistent across both populations. Among the US school students, microdelays as short as one to two seconds predicted that learners completed fewer online math assignments, took longer to finish the assignments they did attempt, were more likely to become distracted mid-task, and ultimately earned lower scores. The aspiring coders showed a parallel pattern: when they experienced microdelays, they spent less time actively engaged with the material, multitasked more often, completed fewer lessons, and were less likely to reach a key course milestone. In other words, the same brief friction produced the same behavioral signature in two very different learning contexts, age groups and regions of the world.

What makes these findings particularly compelling is the study’s causal design. Because the researchers were not simply comparing fast students with slow students—a comparison that would be hopelessly confounded by socioeconomic differences—they instead leveraged within-person variation in load times. The same student sometimes experienced quick loads and sometimes experienced slower ones, depending on network conditions at a given moment. By comparing each learner to themselves across these fluctuating conditions, the analysis effectively holds constant stable differences between students, such as family background, prior achievement and motivation. This within-person approach supports a causal interpretation: the microdelays themselves, rather than the characteristics of students who happen to have slower connections, drive the declines in engagement and achievement.

The robustness of the results was tested extensively. The team reports that the findings held across a range of model specifications, an approach consistent with modern best practices in psychology and education research, where analysts examine whether conclusions survive many reasonable analytical choices. The researchers also conducted a preliminary representative survey, a randomized experiment and analyses of media coverage, all of which are included in the publicly available data and code hosted on the Open Science Framework. This methodological transparency matters, because the effect sizes involved—while meaningful in educational terms—are subtle enough that sloppy analysis could easily have produced misleading conclusions.

The psychological mechanism behind these effects connects to a well-established literature on interruption and attention. Previous research has shown that even brief interruptions fragment work, increase error rates and impose costs on task performance, and that people frequently switch between media contents on a single device in ways that disrupt concentration. Microdelays appear to function as miniature interruptions: in the seconds while a screen loads, attention is released, and learners may reach for a phone, open another tab or otherwise disengage. Once attention has wandered, returning to a demanding task such as algebra or coding becomes harder, and each delay compounds the risk of abandoning the session altogether. Over weeks and months, these small losses accumulate into fewer completed assignments, fewer finished lessons and lower attainment.

The equity implications are significant. The digital divide literature has long distinguished between initial access to technology and the quality of that access, noting that low-income households and communities in developing regions are disproportionately likely to rely on slower or less reliable connections. The study’s second cohort, aspiring coders primarily in Latin America, illustrates the global dimension of the problem: learners in regions where broadband infrastructure lags behind are systematically more exposed to microdelays, and therefore more likely to experience the disengagement and attrition the study documents. A student on a slow connection does not merely wait longer; the evidence suggests they learn less, persist less and are more likely to fall behind peers on faster networks. In this way, microdelays represent a third-level digital divide—not a gap in whether people are online, but a gap in who actually benefits from being online.

For platform designers and education technology companies, the findings offer a clear and actionable lesson. Investments in faster load times, lighter page weights, offline caching and adaptive content delivery may yield learning benefits that rival far more expensive interventions. The study’s authors note that education interventions often produce small effects, and the effect sizes associated with reducing microdelays should be evaluated against realistic benchmarks in the field. But unlike curriculum reform or tutoring programs, reducing loading lags is a comparatively inexpensive technical fix—one that could be implemented at scale by any organization operating an online learning platform. The researchers also point to prior work showing that people systematically discount the effect of friction on behavior, suggesting that decision-makers may underestimate how much these tiny delays matter.

As online education continues to expand worldwide, the study reframes a mundane technical detail as a first-order educational concern. Brief loading lags of one to two seconds, experienced repeatedly across thousands of learning sessions, quietly erode the engagement and achievement of tens of thousands of learners. The research suggests that ensuring fast, reliable connections is not merely an infrastructure issue but a pedagogical one, and that the fight for educational equity in the digital age will be won or lost not only in classrooms and curricula, but in the milliseconds between a click and a load.

Subject of Research: The effect of brief loading delays on engagement and achievement in online learning

Article Title: Microdelays disrupt online learning

Article References: Mazar, A., Tomaino, G., Siedahmed, A., Abdolsaheb, A., Heffernan, N. T., Carmon, Z., & Duckworth, A. L. (2026). Microdelays disrupt online learning. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02598-y

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02598-y

Keywords: microdelays, online learning, digital divide, educational technology, engagement, achievement, internet speed, quasi-experimental study, attention, equity, Nature Human Behaviour, learning platforms