Conversations
Study of more than 80000 online learners: 1- to 2-second delays may affect learning performance
Two longitudinal quasi-experimental field studies covered 81,891 online learners. The studies found that brief loading delays of 1 to 2 seconds were associated with U.S. primary and secondary school students completing less math homework, spending more time, becoming distracted, and scoring lower. Programming learners, mainly from Latin America, showed reduced active engagement, increased multitasking, lower course completion, and lower rates of reaching key milestones.
The research focuses on often-overlooked “micro-delays” in online education, meaning brief loading lags for pages or content. The research team conducted two longitudinal quasi-experimental field studies with a total sample of 26,707 U.S. K-12 students and 55,184 programming learners mainly from Latin America, totaling 81,891 people.
In the U.S. K-12 student sample, when 1- to 2-second micro-delays occurred in online math learning, students completed less online math homework, took longer to complete it, were more likely to become distracted midway through tasks, and scored lower.
In the sample of aspiring programming learners, mainly from Latin America, micro-delays were likewise associated with poorer learning performance: learners spent less active time engaged, multitasked more frequently, completed fewer courses, and were less likely to reach key course milestones.
The paper's abstract says these results remained consistent across multiple model specifications and used within-person variation in loading times to support a causal interpretation. The researchers therefore believe micro-delays are a previously underappreciated barrier in online learning and may put learners with slower network conditions at a disadvantage in engagement and grades.
The material also explains that relevant data and materials are available through OSF. For the K-12 student analysis data, noise was added to school-level demographic information to prevent schools from being identified; all analyses reported in the paper used the original, non-noised variables. Data for the second group of programming learners can be obtained by contacting the authors with a reasonable request.