Microdelays disrupt online learner engagement

The advent of online education promised to democratize access to knowledge, but that promise remains unmet. We identify an overlooked barrier to success in online education: load time microdelays that impede learners with slower internet connections. We show that microdelays on the order of 1 to 2 seconds predict impaired learning in two large field studies, spanning over a quarter million observations from 79k learners. U.S. K-12 students (N = 22,355) who faced longer load times when completing math assignments were less likely to complete assignments, needed more sessions to complete a given assignment, were more likely to get distracted mid-task, and earned lower scores. Likewise, aspiring coders from around the globe (N = 56,655) who faced microdelays spent less time on a coding learning platform, visited fewer pages, were less focused, and completed fewer lessons. We confirm the robustness of these results in multiverse analyses across a range of model specifications, and find evidence supporting the causal role of microdelays in analyses that leverage quasi-random within-person fluctuations in load times. These findings challenge the prevailing assumption that the digital access divide is largely resolved; instead, this divide persists through seemingly trivial differences in internet speed.