On this page
Should weaker and stronger students sit in separate classes? Indian parents and teachers often argue about it. A well-known randomised experiment on the question comes from primary schools in western Kenya. Its answer runs against a common fear: when first-grade pupils were split by prior scores, both groups learned more.
The study is "Peer Effects, Teacher Incentives, and the Impact of Tracking: Evidence from a Randomized Evaluation in Kenya" by Esther Duflo, Pascaline Dupas and Michael Kremer, published in the American Economic Review in 2011. The numbers below come from the authors' revised paper hosted by MIT and the earlier NBER working paper.
How was the Kenya tracking study designed?
In 2005, 140 primary schools in Kenya's Western Province received funds to hire an extra contract teacher for first grade, according to the authors' paper. The researchers focused on the 121 schools that had a single first-grade class and split it into two sections.
Tracking schools (60): pupils in the bottom half of the school's first-term exam scores went to one section, the top half to the other.
Non-tracking schools (61): pupils were assigned to the two sections at random.
Teachers: the contract teacher and the regular civil-service teacher were randomly assigned to the sections, so the stronger group did not automatically get the better teacher.
A random number generator decided which schools tracked, so this is a randomised controlled trial. The program ran for 18 months. The sample frame was about 10,000 first-graders, and endline test scores exist for 5,796 of them, per the paper.
Did tracking help or hurt low achievers?
It helped. After 18 months, pupils in tracking schools scored 0.14 standard deviations higher than pupils in non-tracking schools, or 0.18 after controlling for baseline scores and other variables, according to the study.
After controlling for baseline scores, pupils in the top half gained 0.19 standard deviations and pupils in the bottom half gained 0.16 (paper).
One year after tracking ended, when classes were mixed again, tracking schools still scored 0.16 standard deviations higher (paper).
Comparing pupils just below and just above the median, the researchers could not detect a difference in final scores, even though those just above sat with much stronger classmates.
A standard deviation measures spread; 0.14 is a modest but real gain.
Why did both groups gain?
The authors' explanation is about teachers, not peers. In their model, a teacher aims at one level, and pupils far from that level gain little; Kenya's exam system gives teachers reasons to aim at the top. Tracking narrowed each class, so instruction could match the pupils in front of the teacher.
Teacher effort shifted too. Civil-service teachers assigned to the top section in tracking schools were 11 percentage points more likely to be found in class and teaching, which the paper calls a 25 percent increase in teaching time. Those assigned to the bottom section taught about as often as teachers in mixed schools.
What does the study not show?
Young children, one country. These were Class 1 and 2 pupils in Kenyan government schools, not 16-year-olds preparing for JEE Main or NEET-UG.
Equal resources. Class size and teacher quality were held equal across sections. The authors warn that if the best teachers go to the top students, weaker students could be hurt, so caution is needed in generalising.
Wider evidence is mixed. A 2016 review of about 100 years of research in Review of Educational Research by Steenbergen-Hu, Makel and Olszewski-Kubilius found no benefit from between-class grouping (g = 0.04 to 0.06), but benefits from within-class grouping (g = 0.19 to 0.30) and cross-grade subject grouping (g = 0.26).
The common thread with India's Teaching at the Right Level trials and the Balsakhi remedial study is, in our reading, that instruction pitched at the learner's current level seems to matter more than which classmates sit nearby.
What can a Class 9-12 or JEE-NEET student take from it?
This is our inference, not a finding of the study. Its lesson for you is about level, not labels.
Do not read a "B batch" as a verdict. In Kenya, the lower section still gained. What matters is whether teaching meets you where you are.
Find your level in each chapter. Take a short diagnostic test, then sort topics into "can solve alone", "can solve with hints" and "cannot start".
Work at the edge. Spend most of your time on the "with hints" pile. Questions far above your level, like lessons aimed at the top of a class, teach little.
Ask your teacher one direct question: "Which three problems should I be able to solve by next week?" That can bring teaching in a mixed class closer to your level.
If you do not know where to start, Fermi's AI tutor works out what you already know and builds a plan for the session, so you begin at your level rather than the class average. For adaptive practice research from Delhi, see what the Mindspark study found.



