AIMS EduData gives early-career researchers grants and access to data from widely used math learning platforms—Curriculum Associates, Khan Academy, EdLight, and the University of Florida Lastinger Center for Learning—to study how students and teachers actually learn math. The first cohort of AIMS EduData grantees were the first group to work with this data; this summer, they came together to share what they found.
I attended as an outsider. My team manages these grants, but I hadn’t worked directly on any of the studies, so I sat in the audience without a stake in the results. I left impressed. The questions were sharp, and the analysis behind them was careful, with a few findings particularly worth sharing.
Three teams of researchers looked at how educators move through the Math Matrix Micro-Credential, and their results pointed in the same direction. Researchers from Stanford University found that a teacher’s satisfaction with a course depended far more on whether the teacher believed the course would help them teach than on how easy it was to use. Meanwhile, a joint team from the University of Illinois Urbana-Champaign and the University of Cincinnati found that teachers who felt their practice had improved reported greater gains in their students’ learning and motivation, and teachers who found the course useful were more likely to use it and recommend it to other educators. Finally, a collaborative study by researchers from the Desert Research Institute and University of Nevada, Las Vegas found that experience helped teachers avoid the weakest teaching practices, but it did not necessarily lead to the strongest ones. Their findings suggest that developing high-level teaching practices likely requires targeted professional learning, not simply more time in the classroom.
Students who are behind need a goal they can reach soon. A year-end growth target is easy to agree with and easy to ignore; it’s too far off to shape what a student does on a given Tuesday. A Vanderbilt-led team translated annual i-Ready targets into adaptive weekly goals in a spring pilot. Weekly goal attainment rose from 26% to 40%, and students completed more lessons. A controlled trial is planned for this fall.
At two KIPP middle schools in Camden, New Jersey, students used Khan Academy Reimagined, a new AI-driven version of Khan Academy, in their catch-up block. In the fall, students mostly worked through it independently, on below-grade practice with little connection to their regular lessons. Engagement was low, and Khanmigo, its AI tutor, went largely unused. Partway through the year, the team shifted to on-grade, lesson-aligned content built by the interventionist, but the engagement stayed low, and across the year students learned about as much as peers using the school’s existing software. The KIPP research team’s interpretation is that the tool itself wasn’t the problem—it sat outside the school’s routine and curriculum. This year, they’re giving one person clear ownership of the rollout and tying the practice more directly to what teachers are teaching.
In the Math Matrix Micro-Credential, teachers are the ones being assessed as they work through a professional development course. A researcher from the University of Utah adapted knowledge tracing, the method platforms use to estimate what a student has mastered, to do the same for teachers, pairing it with large language models to identify which specific skills each teacher has and hasn’t picked up, rather than producing a single overall score.
Two teams took on different parts of this problem. The models that predict what a student knows most accurately tend to be “black boxes,” meaning their predictions can be difficult for teachers to understand, and a number a teacher can’t interpret isn’t one they can act on. Researchers from the University of Minnesota built a model that came close to that predictive accuracy, while producing something more legible: an estimate that rises after a correct answer, falls after a wrong one, and fades as a student goes without practice. A group from George Mason University focused on timing, building an early-warning approach that reads a striving learner’s first 12 weeks of activity on i-Ready, including patterns of engagement and persistence, to flag who is likely to pull ahead or fall behind, while there’s still time for a teacher to step in.
The technical work is impressive across all of these projects, but what stood out to me was how deliberately each team kept a person in view, specifically a teacher who has to be convinced a tool is worth her time, or a student who needs a goal within reach this week.
Cohort 2 is now underway with a new set of questions. I’ll be watching for what they find.
To learn more about the AIMS EduData grantees, view their awarded projects.
