Ten introductory courses. Four STEM disciplines. One extra person in the room.
That person is a learning assistant, or LA: an undergraduate who has already taken the course and comes back to work alongside the instructor, circulating among students during class, asking questions, listening to how groups reason through problems. LAs are not teaching assistants in the traditional graduate-student sense, and they are not tutors sitting in an office waiting for someone to show up. They are peers, embedded in the class, usually paired with a weekly seminar on how people learn.
Cassandra Paul and David Webb wanted to know whether that arrangement shows up in the numbers. In a preprint posted to arXiv on 29 July 2026, they report on a multilevel statistical analysis of student records from those 10 introductory courses, comparing sections that had an LA against sections of the same courses that did not.
Multilevel modeling is worth a moment of explanation, because it is doing real work here. Students are not scattered at random across a university; they sit inside class sections, which sit inside courses, which sit inside departments. A model that ignores that nesting can mistake a quirk of one popular professor for a general effect. A multilevel model keeps the layers separate, so that variation between individual students and variation between whole sections do not get blended into a single misleading average.
What the authors found, stated plainly: having an LA in a class was associated with improvement in student-level outcomes, pass rates and retention. Not for one subgroup at the expense of another. For all students.
The differences inside the average
The more interesting result is what happens when you stop looking at the average and start looking at who moved most.
Paul and Webb report noticeable demographic differences in the size of the improvement. LAs were associated with larger pass-rate gains for students from historically marginalized groups than for students overall. And graduation rates rose most for women who identify as belonging to historically marginalized races and ethnicities, measured against students from that same group who happened to land in sections without an LA.
That last comparison is the one that gives the finding its weight. The authors are not holding these students up against the student body at large, a comparison that would tell you mostly about the gaps a university already has. They are comparing like with like: similar students, same courses, differing in whether an undergraduate peer was walking the aisles during class.
A gap that narrows because everyone improves and some improve more is a genuinely unusual shape for an educational result. Interventions often lift the students who were already doing reasonably well, because those students are best positioned to take advantage of one more resource. Here the pattern runs the other way.
Some care is owed to what this study can and cannot claim. The authors use the language of association throughout, and so should any reader. They did not randomly assign students to LA and non-LA sections. Departments decide where to place LAs, students choose their schedules, and both of those choices can quietly correlate with the outcomes being measured. The paper's own conclusion is hedged accordingly: LA programs "may be an excellent investment for diverse institutions." May. This is also a preprint, a six-page conference submission, and it has not yet been through peer review.
Why it matters
Universities spend a great deal of effort trying to close outcome gaps in introductory STEM courses, which function as a filter: fail organic chemistry or introductory physics and a whole career path quietly closes. Most of the proposed fixes are expensive, slow, or both. Smaller classes cost money that does not exist. Curriculum overhauls take years and a faculty willing to sit through the meetings.
LAs are comparatively cheap. They are undergraduates, usually paid a modest stipend or given course credit, drawn from a pool of students who are already enrolled and often glad to be asked. A department can start a program in a semester. That is why a result like this one lands differently from a result about, say, an elaborate new curriculum: the thing being studied is something a mid-sized public university could plausibly do next fall.
The equity finding sharpens that further. If LAs helped everyone by roughly the same amount, they would be a nice quality improvement and nothing more. Because the gains appear to be larger for the students institutions have historically served worst, the same money buys something a budget office and an equity office can both defend.
What the study does not establish is why. The design measures outcomes, not mechanisms. Whether it is the reduced social distance of asking a question of someone two years older, or the smaller effective group size, or the way instructors change their own teaching once they have help in the room, this analysis cannot say. Those are questions for the next study, ideally one that assigns sections rather than observing them.