Tutoring has the strongest evidence base of any large-scale learning intervention: a meta-analysis of 96 randomized controlled trials by researchers at Harvard, the University of Toronto and Northwestern (published as an NBER working paper in 2020 and in a peer-reviewed journal in 2024) found an average effect equivalent to moving a student from the 50th to roughly the 65th percentile — larger than most interventions education has ever tested at scale. But the research is equally clear about what separates tutoring that works from tutoring that is just extra hours: frequency, group size, and who the tutor is. "Tutoring" on a brochure promises little; the research describes the version that delivers.
A scope note: this is a summary of research evidence, not a recommendation for any particular child. Students behind in school need assessment and a plan matched to what they are missing.
What did the meta-analysis actually find?
The authors pooled the results of 96 randomized trials spanning decades, most conducted in schools, and the pattern was unusually consistent. Tutored students gained substantially more than comparable untutored students, with the largest effects in early-grade reading and in math. Effects were bigger for younger students and for one-on-one or small-group formats. Notably, trained non-teacher tutors — paraprofessionals, college students, volunteers — produced gains on par with certified teachers when programs were structured, a finding with big cost implications: schools can staff effective tutoring without hiring a fleet of new teachers. The study became the intellectual foundation for the high-dosage tutoring wave that followed pandemic learning loss, including federal support for tutoring programs coordinated through the U.S. Department of Education beginning in 2021-22.
What is "high-dosage" tutoring?
The term of art in the research is specific: at least three sessions per week, each 30 minutes or more, one-on-one or in groups no larger than three, embedded in the school day, delivered by a trained tutor who stays with the same students over time, and aligned with what the student is learning in class. Programs matching that definition reliably show the large effects the meta-analysis reports. Programs that dilute it — weekly drop-in help, large-group homework clubs, untrained volunteers with no curriculum — show far less. Dosage is not a detail; it is the finding.
What matters more, the tutor or the structure?
Structure, with one caveat. The trials found the format and frequency mattered more than the tutor's credentials — a trained paraprofessional following a structured program outperformed an unstructured arrangement with a more impressive resume attached. The caveat is relationship: consistent tutors who see the same students repeatedly build the accountability and trust that make students actually do the work. The common thread in failed programs is not bad tutors; it is programs designed so no tutor could succeed — no curriculum, no training, no schedule, no continuity.
Does online tutoring work?
The evidence is thinner and more mixed. Most of the strongest trials in the meta-analysis were in-person; evaluations of large online tutoring deployments during and after the pandemic often found smaller effects, particularly when tutoring was optional drop-in or scheduled after school, when attendance collapsed. Online tutoring is not disqualified — early studies of well-structured programs with consistent tutors and required attendance show real gains — but a family evaluating an online option should apply the same tests as for in-person: how often, how small a group, same tutor each time, and is attendance required. Convenience that lowers attendance lowers the effect.
What about the tutoring app my child's school uses?
Different tool, different evidence. Human tutoring and software-based practice both have research behind them, but they are not the same intervention, and the 96-trial finding does not transfer automatically to an app. Some digital tools have strong randomized evaluations of their own; many do not. The question for any product — human or software — is whether the vendor can point to a randomized or rigorous matched-comparison evaluation of that product, and whether the usage it studied matches the usage your school is buying (minutes per week, sustained for months).
How should a family or school choose?
Ask five questions, in this order. How many sessions per week, and are they required? What is the group size? Will the same tutor return each session? What curriculum or skill is being targeted — and was the student assessed so the tutoring aims at the actual gap? And is it scheduled during the school day, when attendance is near-certain, rather than before or after? A program answering those five well is consistent with what the research literature supports; the What Works Clearinghouse publishes intervention reports for families and districts that want the underlying evaluation evidence.
How long should tutoring last?
Longer than most programs run it. The trials behind the large average effects typically delivered tutoring for a full school year, at the three-times-weekly dosage, and several evaluations found effects still growing when programs continued into a second year. Programs that run a single semester — often for budget reasons — reliably show smaller effects, and summer-only versions smaller still, partly because gains need sustained contact to consolidate and partly because attendance decays when the novelty fades. This matters for families reading school announcements: a district advertising a ten-week tutoring push is not describing what the research tested, whatever the brand on it. The realistic commitment for a student a year behind is a school year of tutoring aimed at the specific gap, then a reassessment before anyone declares victory or defeat. The question worth asking at the start is the one programs dodge: what happens at week twelve, when the initial enthusiasm is gone and the student is tired?
What tutoring cannot do
It cannot substitute for attendance, sleep, or a matched difficulty level in the regular classroom, and the research says effects fade when tutoring stops mid-skill. Tutoring accelerates students who are present and taught at the edge of what they can already do. The realistic goal, per the trials, is meaningful catch-up over a semester of consistent sessions — not a transformation in six weeks, and not a replacement for the everyday instruction that carries most of a child's year.
For more context, read How Universal Pre-K Programs Are Scaling Across States.
For more context, read science of reading.
For more context, read career and technical education.