Research Studies

Stanford’s National Student Support Accelerator ran the largest randomized trial to date. Researchers embedded tutor co-pilot systems within 900 tutors serving 1,800 students. Overall mastery rose four percentage points over control groups. Moreover, students paired with lower-rated tutors gained nine points. World Bank teams replicated positive effects in Nigerian secondary English classes. The AI assistant there delivered 0.31 standard deviation growth within six weeks. Consequently, analysts equated the short program to almost two years of schooling.

A new study by Stanford researchers Hsiaolin Hsieh, David Gormley, Carly D. Robinson, and Susanna Loeb suggests why one-on-one tutoring has been found to produce double the gains in student learning than two-on-one tutoring.

Analyzing 16,629 transcripts from 2022-23 school year tutoring sessions from an earlier study that established the greater gains under one-on-one tutoring, the researchers examined how tutors allocated their time and attention across both one-on-one and two-on-one formats. The tutoring sessions focused on early literacy and served kindergarten through second grade students, with 510 students receiving one-on-one tutoring and 570 students receiving two-on-one tutoring. All students met with their tutor online for 20 minutes during the school day, four times per week.

The most recent of these, from researchers at Stanford University’s SCALE Initiative, examined math and reading tutoring programs in a large, urban district during the 2022-23 and 2023-24 school years. Neither led to overall gains in academic achievement.

But when researchers dug deeper into the data, they identified implementation problems that could be driving these null effects.

The findings add to a growing body of evidence around tutoring in a post-COVID landscape that suggests the effectiveness of a program hinges on the nitty-gritty details of how it is run—how often students meet with their tutors, for instance, or whether lessons are tailored to their specific needs.

Studying these implementation details could help school systems build more effective tutoring initiatives in the long run, said Elizabeth Huffaker, an assistant professor of educational leadership and policy at the University of Florida, and the lead author on the SCALE paper.

Experts view the findings as a cautionary tale of how tutoring can go wrong.

The district had to wait on background checks for tutors, many students were still chronically absent and the tutoring sessions often conflicted with other lessons or special events. As a result, students didn’t receive the 30 hours or more required under a state law mandating tutoring for those who failed the annual state test. Instead of five days a week as planned, 81% of the students attended tutoring three or fewer days, and most students worked with a different tutor every time they attended a session.

The findings reinforce the importance of protecting the time students are supposed to receive tutoring, said Elizabeth Huffaker, an assistant professor of education at the University of Florida and the lead author of the study.

High-dosage models — featuring individualized sessions held at least three times a week with the same, well-trained tutor — can still “drive really significant learning gains,” she said, “but in the field, things are always a little bit more complicated.”

Two studies from Stanford University’s National Student Support Accelerator released Wednesday used natural language processing technologies to review transcripts from tens of thousands of hours of virtual tutoring sessions. Their goal: to better understand exactly what happens between tutors and students in these sessions.

One study examines the impact of disruptions as revealed through tutor comments, such as “You can’t see me? I’m not sure why you can’t see me” or “Sorry. Did you say something? It was hard to hear.”

Researchers found that 19% of available time was lost to disruptions, whether from technological issues, distracted students, or background noise. Time lost to disruptions was even greater when tutors were working with more than one student, especially if one of the students entered the session late.

A new collaboration between Stanford’s SCALE and OpenAI, the creator of ChatGPT, strives to better understand how students and teachers use the popular AI platform and how it impacts learning

Education is one of the fastest-growing use cases of AI products. Students log on for writing assistance, brainstorming, image creation, and more. Teachers tap into tools like attendance trackers, get curriculum support to design learning materials, and much more.

Yet despite the rapid growth – and potential – a substantial gap remains in knowledge about the efficacy of these tools to support learning. 

A new research project from the Generative AI for Education Hub at SCALE, an initiative of the Stanford Accelerator for Learning, aims to help fill that gap by studying how ChatGPT is used in K-12 education. In particular, the research will examine how secondary level teachers and students use ChatGPT.